aalr - virginia tech
TRANSCRIPT
Observation, Description, and Prediction of Long-Term Learning on a Keyboarding Task
by
Mark L. McMulkin
Thesis Submitted to the Faculty of the Virginia Polytechnic Institute and State University
in Partial Fulfillment of the Requirements for the Degree of
Masters of Science
in
Industrial and Systems Engineering
APPROVED:
——
AAlr / CHE. Kroginer, Chairman
FT - NY
\ i. % \ fo \ ‘
\ Hl ‘ \ oN vs | \ \ a ‘
“ye ' ( }\ wy pM?
\ Te . SN SS \ Ry ¥ \ NX , MS
“TRC. Wiliges |
.C. Woldstad
Blacksburg, Virginia
January, 1992
Lv
SoS >
VAST
J 912
[226 cy
Observation, Description, and Prediction of Long-Term
Learning on a Keyboarding Task
by
Mark Lee McMulkin
Committee Chairman: Karl H. E. Kroemer
Industrial and Systems Engineering
(ABSTRACT)
Three major principles of learning a chord keyboarding task were
investigated. Five subjects were taught 18 characters on a chord keyboard,
then practiced improving their keying speed for about 60 hours.
The first objective of the study was to observe long-term learning on a
keyboarding task. The performance, in characters typed per minute, was
recorded over the entire range of the experiment. Typing skill improved quickly
in the beginning and then slowed, but performance had not reached a stable
peak by the end of the experiment.
The second objective of this study was to determine a function that
describes performance progress from initial training to a high keying speed.
Five functions were evaluated; a function which predicts the logarithm of the
dependent variable (characters per minute) from the logarithm of the regressor
variable provided a good fit to the actual data. The final form of the equation
was CPM; = ePo7 Br where CPM; = performance in characters per minute on
the i-th interval, T; = the i-th interval of practice, and By and B, are fitted
coefficients.
The second objective also considered the form that T; (from the above
equation) should take. Performance can be predicted from number of
repetitions such as trials, or from amount of practice such as hours. Both trials
and time were used as predictor variables and both provided equally accurate
predictions of typing speed. Both also provided excellent fits in conjunction with
the Log-Log equation. Thus, it appears the Log-Log function is fairly robust in
predicting performance from different variables.
The third objective was to investigate how many trials of performance are
needed before the entire learning function can be reasonably determined. In
this experiment, subjects practiced for an extended period of time (about 60
hours) so a fairly complete progression of performance could be gathered. Yet,
it would be more convenient to collect data for only a few hours and deduce the
ensuing performance of the subject. The coefficients of the Log-Log function
were determined using only the first 25, 50, 100, 150, and 200 of the initial
performance points (out of about 550 total actual data points). The mean
squared error (MSE) was calculated for each of these fits and compared to the
MSE of the fit using all points. It appears that at least 50 performance data
points are required to reduce the error to a reasonably acceptable level.
ACKNOWLEDGEMENTS
This work was possible due to the funds and facilities provided by the
Center for Innovative Technology (CIT) of Virginia, the Department of Industrial
and Systems Engineering, in particular, the Industrial Ergonomics Laboratory,
and Vatell Corporation.
| would like to extend my appreciation and thanks to my committee chair,
Dr. Karl Kroemer, for his encouragement, assistance, time, and support
throughout this project and my entire Masters Degree education. | would also
like to thank my other committee members, Dr. Jeff Woldstad and Dr. Robert
Williges, for their advice and guidance on this project which were instrumental
to the completion of this thesis.
Finally, | would like to thank my fellow graduate students for their help
during my thesis, particularly Chris Rockwell for his assistance in
troubleshooting the experimental set-up.
TABLE OF CONTENTS
ADSHACE cee eeessssecssssccsesnescsccvsnsecesssscsensecsssascesssesesnensucensesensnaseesnansetsoeeseseasenenanaeesenes il
ACKNOWlECQMENMS ..........:cccesecessceceteesseceeseesesecesseeestececseccssneeseaeeeeseecueneessaeeesseesseneceenens iv
List Of ADPONdICES ou... eee eesseseeesssecceecncessceeesetseseessensnsesneecaaeseseseeesesesessteneessneeeness vill
List Of T€ADIOS oo... cceccecssccessensessceessseeceeeseeesecesessseaecesaaeesneeesessaeesesaeesenenseesnnseenseenes ix
List Of FIQUIES oo... cetecseeeeseessecseesseeeessecsseeeseeeeesacseecseceeseeceseeaeeeaeeeetuaesnaesnaseagens X
Chapter Page
1. INTRODUCTION .............:cccccscsecsssssceseseacecensneeeeetsnsecesssnseceesaseeseseaeeeeseseeeeessaeees 1
1.1 The One-Hand Ternary Chord Keyboard ............cccccsscseessescesseesseeens 1
1.2 Rationale .0........eeeeeccccscssscssececeeecesssnonsccecceseessssneeescesseeseesaaenenscessessanaasenees 3
1.3 Experimental Purpose and ODjeCtiIVES .............eeeeeseseceeecssesssenenaceneeeees 5
2. LITERATURE REVIEW . 0.0.0... cccccccccecsssseeeecessnaceecesssseeeeeeeeesscaeeecessesaneeees 6
2.1 Overview of Keyboard DeSIQN ...........cssssssssecsscesneecesseessecessseesseeeceeees 6
2.2 The QWERTY Keyboard... ecscessscsssscssscsssccesesecsaeeesssseaseseesaseaeesasenses 7 2.2.1 CHAPMAN (1919) oo... ccesseecesssecesessceecesseecesssseesessesecesesseeeeess 8 2.2.2 THUPStONE (1919) eee eccctecssssseesecseesssesseeseecseesseeeeesnsesneeeeees 10 2.2.3 GreON (1940) cee ceecssccseececessessseeseeseeeseecseeeseeeaecaeteeeeeessneensees 13
2.3 Alternative Layouts of the Standard Keyboard ..............ccccsssssssseeeeees 14 2.3.1 StrONG(1956) ee ceccsecseeesssseseesseerseseseeceeesseeessuesseesseeessnees 16
2.4 Binary Chord Keyboard ou... ccecsesssesssesscssecsesenseseessesseeeecessecsaeceeeceaseas 16 2.4.1 Kl@MMEr (1958) oo. ccceesessseecscseeccsseceessesessesseeetsneeesseeseseeens 17 2.4.2 Lockhead and KlemmMer (1959) ou... ccc cesssssceeeseessseeseeeseeeeees 19 2.4.3 Ratz and Ritchie(1961) and Seibel (1962) 0... 20 2.4.4 Cornog, Hockman, and Craig (1963) ..........c ce ececesseseeeeeeseees 20 2.4.5 Conrad and Longman (1965) .0.........cccceceeseseeeeeesesssneecereeseees 22 2.4.6 Bowen and Guinness (1965) .........cccesssscccesssseecessssseeeeeeseees 24 2.4.7 SidOrsky (1974) ooo... ccceccecscsccesssecssseeesseeeessceceeeeeeeeneeseceneesssneenss 25 2.4.8 Bequaert and Rochester (1977) and Rochester,
Bequaert, and Sharp (1978) ...........cccececsessennececceeeeeceeeeeeseeees 26 2.4.9 Gopher and Eilam (1979) ..........ccccecssseseceeeeseneseteessssneeeeeesees 28 2.4.10 Gopher, Hilsernath, and Raij (1985); Gopher (1986);
and Gopher and Raij (1988) ....... cc eeeeeeceeeseeseessesneeees 29
2.4.11 Richardson, Telson, Koch, and Chrysler (1987) ..............
2.5 Ternary Chord Keyboard Experiments .........eeeceeeecesceeceseeeeeceeeeeaeens 2.5.1 Fathallah (1988) - TOK #1... cccccccsccesseccesssseeeeesseesesssneeeees 2.5.2 Kroemer (1990) - TOK #2 oe cecccscsscceceeenseeseceesneceeeenersennees 2.5.3 Kroemer (1990) - TCK #3 uc ccecssessesecesseeseesssessesaneseeeseens 2.5.4 McMulkin and Kroemer (1991) and
Lee and Kroemer (1991) oe eeeseescessessssssesseesseeseeseeeeenens
2.6 Length of Training S@SSION ........ eee eeecesecececeeeteeeeeeecesaeeeeseeeneseneeaneernans
2.7 Format of Text for TYPiIng ooo... cee ccsccecssecsseeeesteeseeeeeseeecssesessaeeseeenesseseeas
2.8 Sequences of Finger Movements that Yield Fast Typing Rates ....
2.9 Which Hand to Use on a One-Hand Keyboard... eeeecessesereeecees
2.10 Literature SUMMALY ........cccccesscccsssssseeeesesssceeeeeesssceeesesscseneetesesssseneass
3. METHOD 1... 0... cee cccccccccssscessssesecesseeeeeesssseeeesseeeenesssaeesensauseeseesnsuueccasseseeceesensseneans
3.1 SUDJOCES ooo. ceecsecsseessssesseeesseeeesecsensecseneeensecsnsecsaeeseneetsessecseseseseesesuessaes
3.2 APPArAatUS oo. ceecesccseccesscesececnseseecseesssescseeeeecesseeesaeesaessatessteeseesaeeees 3.2.1 Dimensions of the One-Hand TCK uuu... ccc cccseenecssesseseeenees 3.2.2 CHOP COIN ......eccecceccceseesseesseeseecnesseeeseceseeesessecseessaneanecseeses
3.3 PLOCECUIES o.oo... eecestcecceesssccesseecssecesseneesacecsancessesessecesseecesesessanesuaeeneneeteagens 3.3.1 Experimental De@SIQN ...........ccccsssccsssssseecsssseceecesssnseneessesseesenes 3.3.2 Schedule Of Practice 0... ee eeecssscsseeseeseeesetsceesnsssreeeecsseeeeeeas
3.3.2.1 Breakdown Of SESSIONS ou... eesessesecsseeseeeesenens 3.3.2 2 Training Program .........ccssesceessteessseeeesseetsneesseeeenees 3.3.2.3 Makeup Of Trials oo... cece eeseeeseeeeeneceeseeeteeeeenes
3.3.3 Determining Peak Performance ...........:ccccccccsssssssssseneneeeeeeeees 3.3.4 SUBJECTS’ OPINION uo... eee cccessseessecesssecsecessneesseecseseecseseeneecases
3.4 Experimental Protocolo... cece ceseceessecseneecsecesssecseceesacesenseesseesenesetaaes
4. RESULTS. 0000... ccccceccsesccessececssnceseesaneesensneevsneeeesesueeseesseesessaeeesesseseueessaneeenasseees
4.1 Learning ReESUItS oo... sesseesensesesesssecesssnessssecseneseneeeesnssesaeeseeneeseaeenans
4.2 Speed - Accuracy Performance .0.......ececeecceseeeeeseceeeesseceaecseseeseaeeneeaes
4.3 Using Trials to Predict Performance oo... eee eeeseeecseeececeeeeeereeeeatecens
4.4 Using Time to Predict Performance ou... eee ceeeeeesceeeseseeeesteeeeenees
vi
4.5 Subjective Ratings of Performance... ccccccccecsecstectseertasenren 83
4.6 Fastest Speed Attained by the Subjects 2000.00... ce ccc cece eee 86
4.7 Prediction of Learning with Less Than All data Points «0.20.0... 86
4.8 Chord AnalyS@s oo... cccccccccceccesceeecceececseeessseneeeeecvstecsectiteneeecerseeess 102
4.9 Analysis of Average Performance of All Subjects ....0000...00 ee. 111 4.9.1 Prediction of All Subjects’ Performance oo... eee 111 4.9.2 Prediction of Average Performance... cecceeeeceeeees 114
Loam Bho} Os USS) | @] ee 116
5.1 Speed - Accuracy Performance 20.0.0... cece cece ecce sete seeceensetteeeneeas 116
5.2 Using Trials to Predict Performance 0.0... ccccccccccceeseeeetssseeeeeeees 117 5.2.1 Coefficients of the Log-Log Equation ............ cece 118 5.2.2 Prediction of Peak Keying Speed oo... cece ccc cece sees 119 5.2.3 Combined and Average Performance of All the
SUDJOCHS. 200. ee cece ecececesscceeecetseeesceceeessueseesenerseuereeeestnnaees 124
5.3 Predicting Performance from TIME o....... ccc ccc ccccceceeessceecessteentseeeesees 124
5.4 Subjective Ratings 0... cec cece cesseeessescesseceeeesseeecseesessaeeesseeneasees 125
5.5 Evaluation of Criteria to Predict Performance. .....0......0.cccceecc cece 126
5.6 The Concept of Peak Performance. ........000 occ cece cee cc ceee cee e ese ecenees 127
5.7 Prediction of Entire Curve from the Initial Data Points ...0.00000... 128
5.8 Chord AnalySiS oo... ccc cccccecesseseecesecensectseseseseeeeecssseseeseeesesenteensaess 129
5.9 Further RESCAIrCH o.oo. ec ccccceccccesececeeesstsaeecesersaseeseeeesaeeceessteeeesetstseeeceeeas 130
6. REFERENCES ooiniiicccccccccccccecsccceessetecssecesscccseesetseecesseecesessssseeeiteeeesees 132
ADDON GICES ooo cecccccceccccccceccccceccesenesesaeeeececesessssseeeceeeesenessaseeesscceseeetssssssseeseeneaas 137
MV ccc een e nea E Eee cece en DETER EAE EEE A UEEEAEE ESSE EcELLGAEGEEEEEEEEEEAEEe ES et ts bs adEaEOEEEEE Sets 191
vii
List of Appendices
Appendix Page
A Example of a Test Trial oo... escecsessseeseeesesrecceseseeeesseeseseeeeesneseeeees 137
B Experimental INStructions ...........ccccsssssecssscsseesescessseeseessseesasesseessseessreeees 139
C Informed Consent Form 2.0... ecsececeeseeceseesseceesececscseecsacecsceeesaeenseeeeeeees 141
D Derivation of CPMj = CT Bt i eeccssssssssssssssseesseessccesessssssssssnsnnesessseete 144
E Graphs of All Time Increments and Subjects for CPM; = Bo + ByP;
+ Bo/P; + Ba/Pi2 and CPM, = CBP Bt oc ceccsssssneeseeen 146
F Average Response Times for Each Chord by Trial .........eeceeeeeeee: 172
viii
List of Tables
Table Number Page
Table 1. Summary of Important Variables in Keyboarding Studies. .......0..... 48
Table 2. Average Accuracy Across Trials for Each Subject... 64
Table 3. Number of Trials Completed by Each Subject... eee 65
Table 4. Summary of R-Squared Values for All Functions Considered When Trial is the Regressor Variable. 0... ccc eeceeeeeeeeeeeereen 68
Table 5. The Fitted Log-Log Equation for Each Subject. oo... ee 78
Table 6. Statistical Comparison of Coefficients Bo and B1. 0.0... 80
Table 7. Average Length (and Standard Deviation) of 30, 40, 50, and 60 Minute Time Increments Used in the Regressions. ..............0..ccee 82
Table 8. Summary of R-Squared Values for All Functions Considered When Time Increment is the Regressor Variable. 30, 40, 50, and 60 Minute Increments and Sessions are Used. oo... 84
Table 9. Ratings Ranging From 0 to 9 Given by the Subjects Each Week Assessing Their Progress and Corresponding Performance. ......... 87
Table 10. Fastest Speed Attained on Randomized and Repetitive Trials. ..... 88
Table 11. Mean Squared Error for Log-Log Functions Fit Using Only Some of the Initial Performance Data Points. ........... eee 100
Table 12. Mean Squared Error for Four Equations Fit Using Only Some of the Initial Performance Data Points. For Subject 2 Only. ....0...... 109
Table 13. Chord Response Times (in ms) Averaged Over Trials 468 to 487, the Last 20 Trials Performed by All Subjects. «0.0.0... 110
Table 14. Summary of R-Squared Values for All Functions Considered for Average of the Subjects and All Subjects’ Performance. .......... 112
Table 15. Values of R@ and Peak Performance when Fitting the Equation
CPNj = Bo + B;/T; + Bo/T2 Without the First 100 Performance POINS. occ ccccccccccececcesseeeee sees e na nsseeeeetesecetsaeeeeeseeeesessieeeeseseeenesteiieeeeeas 123
List of Figures
Figure Number Page
Figure 1. View of Binary and Ternary KeyS.. ou... eseesceceesseeeceseeseeeeetneeeeeseeeneees 2
Figure 2. Learning Curve fora QWERTY Keyboard Found by Chapman (1919). eee eeecsesescseeeeecseesneeeeeeaecnaecsaesnseeseeseceseceeeseaecuseneesaseneeseeseseeanees 9
Figure 3. Learning Curve fora QWERTY Keyboard Found by Thurstone (1919) Predicted by Number of Pages Typed. ........cseeseneeeeeee 11
Figure 4. Learning Curve fora QWERTY Keyboard Found by Thurstone (1919) Predicted by Number of Weeks of Practice. oe 12
Figure 5. Median Speed for Four Semesters of Typewriting on a QWERTY Keyboard. Adapted from Green (1940). wu. 15
Figure 6. Typing Speed on a 10-key Chord Keyboard as a Function of Practice. Taken from Klemmer (1958). ou... ccsccssssecsssseesseeereesseeseens 18
Figure 7. Comparison of Typing Speeds for Chord and QWERTY Keyboards. Taken from Conrad and Longman (1965)... 23
Figure 8. View of a Chord Keyboard Developed by Rochester et al. (1978). Taken from Rochester et al. (1978). oo. eeeeseeeseeeseeneee 27
Figure 9. Typing Speed for Three Keyboards as a Function of Practice. Taken from Gopher and Raj (1988). oo. ceeeeeceteeeeeeeseeeeeeees 31
Figure 10. Ternary Chord Keyboard Model #1 oo... ccc eeceeeceseeeecereeeceaeeeeeeetenes 34
Figure 11. Ternary Chord Keyboard Model #2. oo. ee eeseseeceseeeesseeeeseaeees 36
Figure 12. Learning Curve for the Ternary Chord Keyboard #2. owe 38
Figure 13. Ternary Chord Keyboard Model #3. oo. eccceesesessesseneesereeennenes 39
Figure 14. Learning Curve for the Ternary Chord Keyboard #3. owe 40
Figure 15. Learning Curve for the One-Hand Ternary Chord Keyboard. ....... 43
Figure 16. Effect of Training Schedule on Typing Performance. Taken from Baddeley and Longman (1978). ...........cccssssscceeeesseesesnsnsesereeerees 45
Figure 17. Dimensions of the One-Hand Ternary Chord Keyboard. ............... 53
Figure Number
Figure 18.
Figure 19.
Figure 20.
Figure 21.
Figure 22.
Figure 23.
Figure 24.
Figure 25.
Figure 26.
Figure 27.
Figure 28.
Figure 29.
Figure 30.
Figure 31.
Figure 32.
Figure 33.
Figure 34.
Page
Chord Assignments for the Numeric One-Hand Ternary Chord KOYDOAIC. oo... ee ecesecsrseseeecnscereneeesenecsaaesenceesneetsvsessasesneeeesenecsaesenanersaeeesess 55
Scale Used by Subjects to Rate Their Performance Each WEEK. once. cccsccccsssecesscecssscecssscecescaeceesaeseesensecseacecsssaeeeeseeseesesseceesesaeessseesenees 61
Subject 2, Prediction of CPMi = Bo + B1*Trial and Actual PErfOrMaNnce. ou... ceecesscsssccesecescceesecssecsceesaeesessseeessensrseeaeecssssasecsasenseenaes 69
Subject 2, Prediction of CPMi = Bo + B1/Trial + B2/Trial*2 and Actual Performance. .0......cccccesssecsssssecceeceecceneessessssseeaeeeeaeseceeeeseeeceseesseees 70
Subject 2, Prediction of CPMi = Bo + B1i*Trial + B2/Trial + B3/Trial*2 and Actual Performance. oe. eecseessecesecseeeeceessesseees 71
Subject 2, Prediction of CPMi = Bo + B1*Ln(Trial) and Actual POrfOrMAN Ce. ce cecssccesssccsssseceesseeecesseecesseeeseeaeeeceneeeceeatecsusueesenueesseaeeeenees 72
Log-Log Prediction and Actual Performance, Subject 1. .............. 73
Log-Log Prediction and Actual Performance, Subject 2. .............. 74
Log-Log Prediction and Actual Performance, Subject 3. .............. 75
Log-Log Prediction and Actual Performance, Subject 4. ............... 76
Log-Log Prediction and Actual Performance, Subject 5. .......0....... 77
Log-Log Prediction for Each Subject. 0... eecssscecsssssecesessseeeceeseees 79
Log-Log Regressions Using Only the First 25, 50, 100, 150, and 200 Points, SUDJOCt 1. ou... eee esececesseceeseeseceeeseneeneecesseeeeeeseseeeees 89
MSE for Log-Log Regressions Using Only a Few Initial Points, SUDJOCE 1. oo. eeecessesssccvsseecseeceseesssecesceecssceecsneesseeecssesecceeseeeeessstecserseesees 91
Log-Log Regressions Using Only the First 25, 50, 100, 150, and 200 Points, SuDjeCt 2. oo... eee eceecceeeeeceeeeeeeeteeteeeeaeeneeesnaeaaaeeneess 92
MSE for Log-Log Regressions Using Only a Few Initial Points, SUDJOCH 2. cee cccccecscssscssescesscesecseecsaecseeenseseseeeesseceasensecseesseseesseesensseetseeeees 93
Log-Log Regressions Using Only the First 25, 50, 100, 150, ANd 200 POINTS, SUDJECt 3... eee scccsseeeeeseceessneeseseetenseeseseseeeeetsaneessnes 94
xi
Figure Number
Figure 35.
Figure 36.
Figure 37.
Figure 38.
Figure 39.
Figure 40.
Figure 41.
Figure 42.
Figure 43.
Figure 44.
Figure 45.
Figure 46.
Figure 47.
Figure 48.
Figure 49.
Page
MSE for Log-Log Regressions Using Only a Few Initial Points, SUDJSCE Oo icc cccccecesececenssececsseceseseeeesseeeecsateserseseseeseseseaesssieaeeeeses 95
Log-Log Regressions Using Only the First 25, 50, 100, 150, and 200 Points, Subject 4. occ rere teneteteeettnereetteeeees 96
MSE for Log-Log Regressions Using Only a Few Initial Points, SUDECE 4. cece ccc cecseccssccessecsesseecsseeecsesenseeeseseeecsseeeesesrtseeetseseneeees Q/
Log-Log Regressions Using Only the First 25, 50, 100, 150, and 200 Points, SUDjECt DO. ieee ce ceeceeeeeeeeeetsseettttnteeeseaeees 98
MSE for Log-Log Regressions Using Only a Few Initial Points, SUDJSCH ccc ccc ccccsecccnteceeecsaeceeaeseeseseesessseeeesegessssesesseeeesiiseeeecsas 99
Mean Squared Error as a Function of Number of Points Used in Regression for All SUBJECTS. ooo ce eect cee ce ects cece eeeereeeeseeneey 101
Subject 2, Bo+B1/Ti+B2/Ti*2 Regressions Using Only the First 25, 50, 100, 150, and 200 PointS. 2.0... cece eeeeeeeeeeetttneeaes 103
Subject 2, MSE for Bo+B1/Ti+B2/Ti*2 Regressions Using Only @ Few Initial PONS. ooo eeccccccsssseccessecesaseeesseeeeesssseesesseecesseseeees 104
Subject 2, Bo+B1*Ti+B2/Ti+B3/Ti42 Regressions Using Only the First 25, 50, 100, 150, and 200 Points. oer creee 105
Subject 2, MSE for Bo+B1*Ti+B2/Ti+B3/Ti*2 Regressions Using Only a Few Initial POInts. oo. ccecsecccceeeeeesseeeeeseseeeaees 106
Subject 2, Bo+B1*Ln(Ti) Regressions Using Only the First 25, 50, 100, 150, and 200 POINtS. ooo. cc ccccccccsteeeeesestseeeesentsteseeeeeas 107
Subject 2, MSE for Bo+B1*Ln(Ti) Regressions Using Only a Few Initial POINS. oo... ccc cc cceeceeeeeeeeceeceeeeceseeseeeeeeestttsssssseeeaeaees 108
Log-Log Prediction and Actual Performance, All Subjects’ Data. ioc ccc eccceececsceccssesenesececcssecesaseseessseseesesecsseseesseeensiveserteasereranens 113
Log-Log Prediction, Actual Average Performance, and + and - 1 Standard Deviation of the Average of All Subjects. 0... 115
Estimate of Peak Speed Using the Eqn CPM = B+B1/Ti+B2/TiA2, Without the First 100 Trials, Sub 3. ow... 121
xii
1. INTRODUCTION
1.1 The One-Hand Ternary Chord Keyboard.
The keyboard is currently the primary device which people use for data
input to, and interaction with, a computer. The accepted keyboard standard is
the QWERTY keyboard (named for the first six keys on the left side of the
second row). Chord keyboards have been proposed as an alternative device to
the QWERTY keyboard. A chord is defined as the simultaneous activation of
two or more keys to produce one input. Although many differences exist
between various chord and sequential keyboards, almost all use binary keys.
The binary key has only two states: on and off. An example of a binary key is
the type of key used on the standard QWERTY keyboard.
A new type of chord keyboard has been design by VATELL Corporation,
Christiansburg, Virginia (U.S. Patent 4,775,255 of October, 1988). This
keyboard, called the Ternary Chord Keyboard (TCK), uses keys which have
three states: two on and one off. Figure 1 shows a comparison of the binary and
ternary keys. The TCK uses the three-position ternary keys, whereas all
previous keyboards have used the two-position binary keys. The TCK,
discussed in detail in section 2.5, is intended to be used by two hands (eight
keys one for each finger) for data entry of the alphabet, numbers, punctuation,
symbols, and cursor keys. Each of 64 "simple" chords is generated by
activating one key with the left hand and one key with the right hand. Several
authors have analyzed the TCK for performance compared to the QWERTY
keyboard, learning times, and performance capabilities (Fathallah, 1988;
Key Activation (finger pushes downward)
-— 7 49
LL ~ ul
(State 1, no input) (State 2, input 1)
Neutral Downward
(a) Binary Key
Key Activation
(finger pushes forward or pulls backward)
7 Lt Vi\ &
| |
(State 1, no input) (State 2, input 1) (State 3, input 2)
Neutral e.g., backward e.g., Forward
(b) Ternary key
Figure 1. View of Binary and Ternary Keys
Callaghan, 1989; Kroemer, 1990). All of these experiments investigated the
use of a two-hand TCK as an alternative to the QWERTY keyboard.
Many data computation and entry tasks only require the use of a
numerical keypad. Numerical keypads require only 18 distinct characters (i.e.,
the numerals 0 to 9, "plus", "minus", "divide", "multiply", "decimal point",
"backspace", "subtotal", and "total"). A chord on the one-hand TCK is defined as
the activation of two keys by two different fingers on the same hand. (This is
different from the two-hand TCK in which one finger is used from both hands to
generate a chord.)
A total of 24 simple chords (those generated by two fingers) are possible
on the one-hand TCK. McMulkin and Kroemer (1991) and Lee and Kroemer
(1991) determined subjective ratings for ease of activation, response times, and
fastest possible keying capabilities for all 24 chords on the one-hand TCK. The
characters (numbers and functions) used for a numeric keypad were then
assigned to the 18 "best" chords of the one-hand TCK.
1.2 Rationale.
When considering how people learn to use keyboards, several questions
arise. First, what is the “learning curve” to reach peak performance? In other
words, how does performance progress, how long does it take to reach
maximum keying rate, and how fast does a person type after a certain number
of hours using a keyboard? Second, what is the maximum speed a person can
type using a particular keyboard (maximum speed being the point where the
rate of typing stops to increase significantly)? Finally, how can it be judged
when peak performance has been reached, and what criteria should be used to
assess performance/skill? Prior research on keyboard use has concentrated
primarily on finding the maximal performance and ignored the rate of progress
to reach final speed and the criteria on which to predict speed.
The term “learning curve" is often used when discussing keyboarding
tasks. For the purposes of this study, learning curve is defined as the way in
which a subject's performance progresses or improves over a period of time or
number of trials. The term learning curve is not intended to mean that there is
one equation that describes the way in which people learn all skills.
The QWERTY keyboard has been in use for over 100 years, yet
surprisingly few studies have been conducted to establish learning curves.
Naturally, with the number of people that use the QWERTY keyboard, maximal
typing speeds have been reported, but the criteria (time or number of characters
typed) used to indicate performance progress to top speed have not been
studied.
As alternative keyboard designs are investigated, the main thrust of
experimentation is to compare the performance of the new design with the
QWERTY design. In most reported experiments, subjects using the new
keyboard did not achieve peak performance (peak performance being the
fastest attainable keying speed); however, the experimenters compared
performance between the new keyboard and QWERTY keyboard at equivalent
points in learning. For example, Bowen and Guinness (1965), in evaluating a
mail sorting task, stated that after 25 hours of training, people using a chord
keyboard sorted 30 letters per minute while people using a QWERTY typewriter
sorted 33 letters per minute. There was no discussion of how performance
progressed during those 25 hours. No indication was given that “hours of
training” was the best criterion to evaluate speed; perhaps, "number of
characters typed" would be a better measure.
This study examined the performance progress of subjects on the one-
hand TCK. The subjects were taught 18 chords. After learning the chords, the
subjects practiced using the one-hand TCK for an extended period of time, 60
hours, so that they approached maximal keying speed.
1.3 Experimental Purpose and Objectives.
This study had the following primary experimental objectives:
1. Investigate the performance progress of people on the one-hand Ternary
Chord Keyboard as they attain peak keying speed.
2. Find a function which accurately predicts the learning or progress of the
subjects. Evaluate a set of criteria to find which can be used to best
predict a subject's performance in keyboarding. The criteria sets are:
1) The number of hours a person spends practicing.
2) The number of trials or characters typed.
3) Subjects’ opinion of when they have reached peak performance.
3. Determine the number of initial performance data points that are needed to
predict the entire learning function.
2. LITERATURE REVIEW
2.1 Overview of Keyboard Design.
The first keyboard that used the QWERTY layout was designed and
produced in 1874. Several attempts have been made to improve this design.
Many of the alternative designs have been concerned with rearranging the
keys. The most famous of these are the Dvorak keyboard (named for the
inventor August Dvorak, not for the arrangement of the keys) and the alphabetic
layout (Noyes, 1983b).
Other alternative designs have employed chords (simultaneous
activation of two or more keys to produce one input) to enter data. Colombo
was the first to attempt to develop a mechanical chord keyboard in 1942 when
he requested a patent (Noyes, 1983a). Since that time, many other chord
keyboards have been proposed and tested (Gopher and Raij, 1988;
Richardson, Telson, Koch, and Chrysler, 1987; Gopher, 1986; Kirschenbaum,
Friedman, and Melnik, 1986; Gopher, Hilsernath, and Raij, 1985; Bartram and
Feggou, 1985; Mussin, 1980; Gopher and Eilam, 1979; Rochester, Bequaert,
and Sharp, 1978; Sidorsky, 1974; Bowen and Guinness, 1965; Conrad and
Longman, 1965; Cornog, Hockman, and Craig, 1963; Seibel, 1962; Ratz and
Ritchie, 1961; Lockhead and Klemmer, 1959; Klemmer, 1958). These
keyboards involved different numbers of keys, required the use of one hand or
both hands, and used different numbers of fingers to activate a chord.
The literature on various keyboards will be reviewed for information
relevant to the objectives of this study. Emphasis will be given to studies
evaluating performance progress during a keyboarding task and the criteria
used to evaluate learning. Several studies have been conducted in which only
final performance on the keyboard is listed with little regard for the way it was
attained (e.g. performance at hourly increments). These will be reviewed
because they provide some insight to the objectives of this experiment. The
important variables that needed to be considered for this study will be
discussed.
2.2 The QWERTY Keyboard.
The typewriter has changed significantly from its first design in 1874. It
began as a manual entry device in which the human fingers generated the force
to move the typebars which struck the paper. The next major evolutionary
advancement in typing was the advent of the electric typewriter. The fingers still
activated the keys, but an electric motor generated the force to actuate the type
bars (or some other device to print characters on paper). Now, computer word
processing is replacing the conventional typewriter; keys are still activated by
the fingers, but electrical impulses generate the characters on a CRT to be
printed later on paper. Through all these mechanical and electrical changes
one element of data entry has remained essentially the same - the layout of the
characters on the keyboard.
Since the layout of the keys has not changed much since its invention,
the ability of people to learn the location of the characters and attain peak
performance should not have changed much either. However learning curves
for the QWERTY keyboard are not well documented. In the early 1900's,
several schools existed that taught students to type. Three of these schools
documented speed progress to improve teaching methods. These three studies
will be reviewed (Chapman, 1919: Thurstone, 1919; and Green, 1940).
2.2.1 Chapman (1919).
The purpose of Chapman's (1919) study was to observe the rate of
typing improvement of high school students. The number of characters taught
to the subjects was approximately 65. Although this was not explicitly stated,
subjects were taught to use the typewriter; the early typewriters probably
contained between 60 and 70 characters. He administered a five-minute typing
test to 20 students once a week; the tests were not begun until the students had
completed 20 hours of practice time to learn the keyboard. Every week, for
each student, the author recorded hours practiced (both in and out of school),
gross number of words typed during the five-minute test, and number of errors.
The author then plotted average net words per minute typed (gross words per
minute minus the number of words with errors) versus practice time for 20 to
180 hours. Chapman (1919) found the following results: 1) an initial linear
learning relationship existed for the first 75 hours, 2) the slope of the learning
curve abruptly changed at 90 hours of practice; the rate of learning was six
times as large between 20 to 90 hours as it was between 90 to 180 hours, 3) at
180 hours of practice the curve was nearly flat which indicates the students
were at nearly peak performance (48 words per minute or 240 characters per
minute were actually reached). Figure 2 shows the learning curve that
Chapman (1919) obtained. Note that the curve shown is the average over all
20 students; it is not for a single subject.
(6161) uwewdeuy
Aq punoy pueoghdy ALHAMDO
B& 40}
ening Bulusea
“Zz aunbi4
SINOH
O8l O91
Ov l Ocl
OO| O8
09 Ov
0¢ 0
0 OV
08
Ocl
ys
Sa]NuIW eAl4
x
ul padA| Spon,
Ps 091
00¢
al
_-
:- Ove
08¢
One problem with this study is that the practice time between tests for
each student was not kept constant. For example, subject A may have practiced
eight hours in week five, while subject B practiced only three hours during that
week; yet, the same test was administered to both subjects at the same time.
This problem left unequal sample sizes at each interval of time, as well as
different subjects’ data at each time.
An interesting result of this study was that subjects reached peak
performance in about 180 hours (although practice on each character is not
equivalent). Another important note is that the author related performance to
hours of practice. No other practice criterion was measured, such as pages
typed.
2.2.2 Thurstone (1919).
Fifty-one students learning to type at a business school were tested for
typing speed once a week for 28 consecutive weeks. Students practiced 10
hours a week (2 hours daily). At the end of each week, Thurstone (1919) gave a
test to measure the number of words typed over four minutes. Words typed per
four minutes were plotted as a function of total practice pages typed. The author
fit a curve to the data which yielded the following equation: y = 216(x +19)/(x +
148), where y was average words typed in four minutes and x was number of
pages of practice. The above equation predicted a maximum typing speed of
54 words per minute (WPM) or 216 words in four minutes (as shown in Figures
3 and 4) after over 1000 pages of practice. Subjects actually reached an
average of 41 WPM (205 characters per minute, CPM) after about 250 pages of
10
OO€ O8c
09¢ Ove
Occ O0Z
O8l O9I
OVI Ol
OO!
padA| sebey
j0 waqUuINN
Aq payaipaid
(6161) suojsuinyy,
Aq punoy
pueoghkay ALHAMD
B® 10}
BAIND Bulwee]
‘¢ eunbi4
pedd| seabed
jo vaquiny
jejo,
08 09
Ov O02
O
/
Jf
oT
ar
a"
_
a?”
.
—
at een
nt
—orre
Ole
yur BOB
Pslpipsaird
lenjoy .
paypal ww]
0 0c
Ov
09
08
001
Ocl
Or |
091
08 |
00¢
Occ 4
11
seinul~— sno4
ul padA|
SPJOAA
DONOCI
JO SHBOAA
JO OQUUNN
Aq pajyoIpald
(6161) euojsunyy
Aq punoy
pueoghsy ALHAMDO
& 10}
avINy Buiwea]
“py aunbi4
BOIIDBId JO
SYB9OM
ce O€
82 92
ve 22
OF BI
9 VE
Zt OL
8 9
vw @
O
nt
UGHBIO|BISY- OANSOd —
eo
we
a
A
uolqiajaooy enefon
a
> a
S$
ae “
¢
ow wg
¢
9L¢ ‘ytht]
Ss1e1_
PSssipaig
jemoy }
payipald yur)
0 02
OV
09
08
OOl
Oct
Ov |
091
081
00¢
Occ
Ove
seynuiw ino4
ul pedA|
SPlO/AA
12
practice. The entire learning curve for performance versus pages typed is
shown in Figure 3. This curve is the average over all subjects.
Thurstone (1919) then plotted words typed in four minutes versus weeks
of practice (about 10 hours of practice per week for 28 weeks), see Figure 4.
This plot yielded a curve which was treated as a third order polynomial,
whereas the plot of WPM versus pages of practice was described as a second
order polynomial by Thurstone (1919).
This study suffers from the same constraints as Chapman's (1919). The
tests were administered to all subjects at the same time, but the subjects were at
different levels of practice (number of pages typed). It is interesting to note that
the shape of the learning curve changes from second to third order when the
criterion to judge performance was changed from pages typed to weeks (hours)
of practice. This might be due to the fact that the pages typed per hour
increased over the 28 weeks.
2.2.3 Green (1940).
Green (1940) studied the typing progress of students over one semester
of classes. There were four groups of students: those in first semester typing,
those in second semester typing (who had already taken typing in the first
semester), those in third semester typing (who had already taken typing in the
first and second semesters), and those in fourth semester typing (who had
already taken typing in the first, second, and third semesters). Progress was
only recorded for one semester; therefore, for example, the learning curve for
the first three semesters of the fourth semester students was not known. Speed
tests were administered to the subjects once a week for 18 weeks.
13
Students in the fourth semester reached a peak speed of about 47 WPM
at week 18. This was presumably after a total of about 360 hours of practice (18
weeks X 5 hours/week X 4 semesters). The rate of learning (slope of the
learning curve) became flatter from the first to the fourth semester, see Figure 5.
A concern with this study is that four different groups of people are used
for different sections of the learning curve. This made it difficult to track
performance progress from start to finish. Also, WPM was plotted as a function
of weeks of practice; this was a fairly coarse measure, considering how much
practice time could vary between subjects in a week's time.
2.3 Alternative Layouts of the Standard Keyboard.
The most famous attempts to change the QWERTY layout have been the
Dvorak keyboard named for its inventor, August Dvorak, and the alphabetic
layout (Noyes, 1983b); the most common alphabetic layout had the letters
arranged with ABCDEFGHIJ on the second row, KLMNOPQRST on the third
row, and UVWXYZ on the bottom row, although other alphabetic layouts existed
(Norman and Fisher, 1982). While neither of these designs has experienced
commercial success, studies have been conducted to compare the performance
of alternative designs to the QWERTY layout. The majority of the papers
published on alternative layouts present theoretical arguments for the
advantages of one layout over the other. The one paper found that employed
experimentation to assess performance of subjects using an alternative
keyboard will be discussed.
14
50
4th Semester
Progress 46
44
42
40
38
36
34
32
30
28
26
24
22
20
18
16
14
12
10
3rd Semester
Progress
2nd Semester
Progress
1st Semester
< Progress
Spee
d (in
Words
per
Minu
te)
mo NN
ff
DMD
®
012 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
Weeks
Figure 5. Median Speed for Four Semesters of Typewriting on a QWERTY Keyboard. Adapted from Green (1940).
15
2.3.1 Strong (1956).
In an experiment to evaluate the Dvorak layout, Strong (1956) retrained
ten typists to use the Dvorak keyboard. They were trained on the new keyboard
until they had reached the original speed they had possessed on the QWERTY
keyboard. The time to retrain the subjects was recorded in number of hours and
was largely a function of original speed; the faster the original speed the longer
the retraining. Once the subjects had been retrained on the Dvorak keyboard,
ten additional subjects using the QWERTY keyboard were added to the
experiment who had roughly the same typing speed as the Dvorak group. Both
groups were then given an additional 100 hours of training. Over the final 100
hours of training the QWERTY group improved from an average of 61.7 WPM to
81.7 WPM (on a five minute typing test), a 20 WPM improvement. The Dvorak
group improved from an average of 54.9 WPM to 66.2 WPM (on a five minute
typing test), an 11.3 WPM improvement. Overall the QWERTY group improved
8.7 WPM more (20 WPM improvement minus 11.3 WPM improvement) than the
Dvorak group. No further conclusions could be drawn from this experiment.
Although the Dvorak keyboard has been a popular alternative to the QWERTY
keyboard, no other experiments were found which compared it to the QWERTY
keyboard. Therefore, no further conclusions can be drawn for this study.
2.4 Binary Chord Keyboards. |
A variety of chord keyboards have been investigated which differ in
aspects such as number of keys used, number of fingers used to generate a
chord, and character set size. For each of the following studies, how the
16
keyboard worked will be briefly described, as well as performance results
obtained.
2.4.1 Klemmer (1958).
One of the first documented experiments on a chord keyboard was
performed by Klemmer (1958) on a ten-key chord keyboard (one key for each
digit). Two subjects were taught a set of 30 characters (26 letters, "space",
"return", "period", and "comma"). Ten characters were generated with one
finger, nine with two fingers of the same hand, and eleven with one finger from
each hand. Klemmer (1958) had the subjects repeatedly type 17 different text
sets which consisted of 60 words drawn from a list of the 1000 most frequent
words in English. The subjects typed the word sets for approximately 35 hours
(50 minutes a day, 3 times a week for 12 weeks); one subject improved from 20
WPM to 60 WPM (went 19 times through the group of 1020 words), while the
other went from 17 WPM to 37 WPM (12 times through the 1020 words).
Learning curves, Figure 6, were generated as WPM versus number of times
through the group of 1020 words. Trends in the curves are difficult to extract.
For an additional six hours, subjects typed samples of text containing about 670
characters. The first subject consistently averaged about 47 WPM while the
second averaged about 30 WPM.
A relation between size of character set and amount of practice time to
high performance can be inferred by comparing this study to the studies by
Chapman (1919), Thurstone (1919), and Green (1940). Klemmer (1958) taught
30 characters and reached around 45 WPM (225 CPM) in approximately 35
17
(9961) JowWal|y
WO
UeYe] ‘aOOBI
JO UONOUNY
e Se
Preoghay ProYyD
Aey-01 B
UO paads
BudAy “9g
aunbl4
SVVINL
SAISSZIONS
LX31
430 SQHOM
Gt6
AO
SdNOHD
SQYOM
WOONVY
OzZOi 4O
SdNodYo
be
zdgii 6}
8! 2)
Of Sl
bl fi
fio 01
6 82
9 GC
¢ ¢
2 !
O
TT
TT
1 ITT
TT TTT
1 T7712?!
(if)
LO3BrAaNsS o
(gy) LOB¢aNsS
VY
SNOYUM S3MOYULS
4O %=
LNIOd TWINL
HOVZ 3A08V
S3YNdIZ
vor
20 >-
roo.
20
OS
Q3dAL BLNNIW U3d ,SOYOM, Y3LIVYVHO-S SSOYD
09
18
hours. Chapman (1919), Thurstone (1919), and Green (1940) taught about 65
characters and did not reach 45 WPM (225 CPM) until 160 to 300 hours.
2.4.2 Lockhead and Klemmer (1959).
A chord keyboard using eight keys, one for each finger, was developed
by Lockhead and Klemmer (1959). Chords were generated by simultaneous
activation of one to eight keys. Lockhead and Klemmer began by teaching four
subjects the chord patterns for 100 words. On the average, subjects took 20
hours to learn the 100 word patterns. Subjects then learned the 26 letters of the
alphabet and the 10 numerals; it took an additional six hours to learn the 36
new characters and integrate them with the original 100 word patterns. The
subjects then typed text (for about two hours) which contained approximately an
equal number of words which were in the 100 word pattern chords and words
which had to be typed one letter at a time. Subjects averaged about 15 WPM
(or 45 chords entered per minute, [15/2]X2 + 15/2, as estimated by the author of
this paper).
This study used a considerably larger chord/character set than the other
experiments that are reviewed. Long training times and low chords per minute
are expected with such a large set of chords. The chords entered per minute
are difficult to interpret because all the data are reported as WPM. Lockhead
and Klemmer (1959) guessed that the upper limit of speed for chords using
three fingers or more is 120 chords per minute. Single finger key activations
were estimated to have an upper speed of 300 key entries per minute.
However, they gave no indication as to the length of training time required to
reach these limits.
19
2.4.3 Z Ritchie (1961) an ibel (1962
Ratz and Ritchie (1961) and Seibel (1962) used the same one-hand five-
key chord keyboard. A total of 31 chords were possible using 1, 2, 3, 4, or 5-
finger combinations. The chord to be activated was presented on a display of
five lights that corresponded to the five keys. Both studies measured the
response time of all 31 chords (time from onset of light pattern to activation of all
of the corresponding keys). Ratz and Ritchie obtained an average response
time of 1.2 seconds after 2,660 measured response times. Seibel found an
average response time of about 0.32 seconds after 11,000 response times. The
large difference in response times was assumably due to more practice in
Seibel's experiment. In fact, Seibel (1962) performed the study to show that
Ratz and Ritchie (1961) had not attained the shortest response times possible.
These response times do not indicate training periods to reach maximum
performance on a typing task, but they indicate the approximately fastest keying
rate attainable when there are 31 choices.
2.4.4 Cornog, Hockman, and Craig (1963),
Cornog et al. (1963) experimented with a chord keyboard for sorting mail
that had 12 keys (2 rows of 6) for each hand, two keys for each finger and four
keys for the thumb. Each hand, on its own, could generate all 26 letters and 10
numbers (each hand operated an independent keyboard). Twelve characters
could be entered by activating one key while the remaining 24 required the
depression of two keys (from the same hand) simultaneously. The authors
trained subjects for one-and-a-half hours a day for 11 weeks, a total of about 80
20
hours until they reached "moderate" proficiency; the authors do not elaborate on
how proficiency was determined or what "moderate" means. For the next 15
weeks, the subjects performed sorting on an outgoing mail task. No learning
curves of progress performance were reported; however average performance
at the end of 15 weeks was 232 characters per minute (CPM). For weeks 16 to
30 the subjects performed an incoming mail sorting task. Typing speed at the
end of week 30 was an average of 180 CPM. Finally, during weeks 31 to 36,
subjects performed an originating mail sorting task. At the completion of week
36, average typing speed was 170 CPM.
Subjects received a considerable amount of practice using the chord
keyboard (estimated at least 300 hours). However, they were learning a very
difficult task. First, they had to learn 36 character chords for each hand.
Second, they had to learn the coding schemes to sort mail, i.e., what letters and
numbers to enter for each address. The task of the subjects was to read the
address on the letter, determine the code that needed to be entered on the
keyboard, recall the chords needed to produce the code, and then enter those
chords. Therefore, it is difficult to separate the portion of practice attributable to
learning the keyboard from the portion of time attributed to learning to read and
code addresses. Also, the average CPM at the end of each section of practice
was confounded with the type of task - outgoing, incoming, or originating mail.
However, it appears that highest performance, 232 CPM, was found before the
task was changed at the end of 11 weeks of training and an additional 15
weeks of practice - a total of about 190 hours.
21
2.4.5 Conrad and Longman (1965),
An experiment to compare the standard typewriter with a chord keyboard
was conducted by Conrad and Longman (1965) for the British post office. The
chord keyboard was comprised of 12 keys, six for each hand. Letters and
numbers, a total of 36 characters, were generated by pressing two keys, one by
each hand. An additional five keys were used for the functions of enter, cancel,
and coding of local addresses. Subjects participated in the experiment for 3 1/2
hours a day, 5 days a week for 7 weeks. One group of 24 subjects learned the
36 characters on the standard typewriter, while another group of 22 subjects
was taught the chord keyboard.
The intent of the study was to compare relative performance of the two
keyboards at stages along the learning curve, not to find the maximum
performance possible in seven weeks of practice. However, the learning curves
and method are pertinent to the objectives of this study.
The group trained on the chord keyboard became "operational" in an
average of 12 days, while the typewriter group needed an average time of 21
days; "operational" was defined as the subjects being able to type all 36
characters accurately as determined by the instructor. The typewriter group
became "operational" an average of 9 days slower than the chord group. After
learning the characters, the two groups practiced to increase speed for the
remainder of the 35 days. Figure 7 shows the learning curves for both groups
as a function of days. Performance on each day was an average of four speed
tests across all subjects that were considered operational by that day. It is
important to note that the performance curves begin earlier for the chord group
because its members were operational earlier. When the typewriter group was
22
Os
O8
06
ool
(G96) uewHu07
pue peiu0y
wo
uaye] ‘spueogheay
ALHAMO
puke puoyy
410} speeds
BHuldA| jo
uosuedwoy
‘7 ainbi4
BINIWDIg pO
SADG 3A1199)I9
ee ce
oe sz
9e ve
zz 0%
gi 91
vi zl
Ol 6
a
GG
Mi i
YILIUMSdAL
{
r AVO
SIHL NO
SLD3PENS JO
oN ‘
t ‘
‘ t
t '
§ ‘
. ‘
‘ '
‘
61— O2-
O2— 12-2
zz ~-61—
6i— 8I—
81 — SI
—C1— a
— ZI
— zi—2r— zl
OS
O9
oe
O06
OOo!
ww/do 32g s3902)sh7y
23
added to the graph, it was consistently slower than the chord group. It should
also be pointed out that at the end of training, after 120 total hours, typing speed
had reached almost 100 CPM for the chord group and was still increasing
approximately linearly. However, each day represented three hours of training
So it is difficult to interpret the utility of each hour.
An editorial note: the training schedules in the studies differed
significantly. For example, Klemmer (1958) trained subjects for one hour a day
while Conrad and Longman (1965) had subjects practice nearly four hours a
day. Massed or distributed training can have profoundly different effects on
performance (Goldstein, 1986); this aspect will be covered in greater detail later
(section 2.6). In reviewing each of the studies, caution should be used in
relating total hours of practice to final performance because the method to reach
that total number of hours may drastically affect performance.
2.4.6 Bowen and Guinness (1965),
Bowen and Guinness (1965) designed another experiment investigating
the differences between chord and standard keyboards in a mail sorting task.
Three keyboards were investigated: a standard typewriter, a large chord
keyboard (12 keys for each hand), and a small chord keyboard (6 keys for each
hand). Subjects learned 11 keying patterns corresponding to 11 3-digit
numbers displayed. For the typewriter, two keys were depressed sequentially
for each 3-digit pattern; for the large chord keyboard, 1, 2, or 3 keys were
depressed simultaneously for one 3-digit pattern, while for the small chord
keyboard 1, 2, 3, or 4 keys were activated at the same time.
24
After practicing 20 hours, six subjects on each keyboard performed 800
experimental encodings. The typewriter group performed an average of 40
sorts per minute (this is actually 80 key presses per minute), while the large
chord keyboard group and small chord keyboard groups averaged 49 and 55
sorts (or chords) per minute, respectively. No learning curves over the 20 hours
of practice were provided. The training schedule, e.g. hours per day, number of
days, etc., was not indicated.
2.4.7 Sidorsky (1974),
Sidorsky (1974) experimented with a one hand combined chord and
sequential keyboard for use in battlefield data entry. The keyboard consisted of
five keys, one for each digit of the right hand. Single characters were generated
by two consecutive inputs of single keys or chords. Each input was either the
thumb, little finger, or combinations (chords) of the index, middle, and ring
fingers. For example, the letter F was produced by simultaneously pressing the
keys under the index, middle, and ring fingers followed by depression of the
index finger key. Ten subjects were taught about 40 characters (letters,
numbers, and punctuation). Subjects were then given 20 text tests (each 255
characters long, consisting of typical sentences) each day to type. The training
and testing process was conducted in 10 sessions of 1/2 hour over 10 to 15
days, a total of 5 hours. Considering only each subject's fastest trial, the
average speed achieved was 63 CPM.
Sidorsky (1974) provided no information on performance progress from
day to day. He only counted the fastest trial a subject obtained on any day.
Considering the size of the character set, 40, and the length of practice, the
25
subjects attained fast data entry, because 63 CPM actually corresponded to
twice as many chord activations.
2.4.8 Bequaert and Rochester (1977) and Rochester, Bequaert, and Sharp
(1978),
A chord keyboard consisting of two rows of five keys was developed by
Rochester et al. (1978) for use with one-hand. Figure 8 shows the layout of the
keys and the patterns of keys required to generate characters; the bottom part of
the figure shows the symbols activated by the shift function. Four subjects
learned about 54 characters (letters, punctuation, and symbols, but no
numbers).
Bequaert and Rochester (1977) and Rochester et al. (1978) appear to
have reported on the same experiments. Subjects were trained for one 50-
minute session a day, 5 days a week. After an average of 49 sessions or 37
hours, subjects could touch-type all 54 characters. Their average speed at this
time was 18 WPM (90 CPM), however, this was achieved during repeated
practice of one text file. The subjects then strived to increase their tying speed.
They did this by repeatedly practicing a short text passage, alternated about half
of the time with new typing material. The results are difficult to understand. For
example, Rochester et al. (1978) stated that three subjects achieved typing
speeds of over 40 WPM (200 CPM); the fourth gained a speed of 25 WPM (125
CPM) after a total of about 90 hours of practice. These speeds were reported as
achieved on previously unpracticed material. Whereas, Bequaert and
Rochester (1977) reported typing speeds of 46, 47, 32, and 45 WPM for
previously unpracticed material. They also provided performance progress
26
DIMPLES PRESSED WITH
| INDEX | MIDDLE | RING | | FINGER | FINGER | FINGER |
| | | | D 8 F Ss N K ! J U A
| | | ~<| T Ww P Vv H x Zz
{ i
| G c M | L R a | Y | O |
J J | !
SPECIAL
NUMBER
UPPER—LOW
SPACE
SPACE— UP
PER
SPACE—
REVERSE
REVERSE
I ' ‘ ! | INDEX | MIDDLE | RING | | FINGER | FINGER | FINGER |
4 | T O “Vel i(Vr “x¥s]i;¥N =VK/ t:¥s «Vu /-VYa @
T 7Y¥w P vi l/V¥rH - ¥x V7 é E { ‘ '
VA >Ycl],¥M SVL] I<V¥R }YVal]]- Ik aViil |7Vo | |
| l |
Figure 8. View of Chord Keyboard Developed by Rochester et al. (1978). Taken from Rochester et al. (1978).
2/
curves as a function of one-hour sessions. It appears that the maximum typing
performance occurred at the final session. It is assumed the tests, given (once
every four sessions) to judge typing speed, were new (unpracticed) material,
but it was not explicitly stated.
It is difficult to determine any trends in performance progress from the
learning curves. First, the tests were only given every four sessions. Second, it
is not known if new or old material was used, how long the tests were, or what
the distribution of characters in the test was. Therefore, it is hard to determine if
a “maximum” performance was approached.
2.4.9 Gopher and Eilam (1979),
A one-hand chord keyboard similar to that used by Sidorsky (1974) was
tested by Gopher and Eilam (1979). The keyboard consisted of only four keys,
one for each digit of the right hand except the little finger. Two consecutive
chords by combinations of the three fingers were required to create one
(Hebrew) character. The thumb was pressed sequentially with one of the three-
finger chords to activate an editing function.
In the first experiment by Gopher and Eilam (1979), subjects were taught
30 letters and functions. The training occurred over seven one-hour sessions.
Three one-hour sessions separated by two hours were administered on day
one, two one-hour sessions on day two, and two one-hour sessions six weeks
later (to test retention). At the end of the seventh session subjects typed an
average of 65 CPM (130 strokes). There was linear increase in speed over the
seven sessions, even with six weeks off after five sessions.
28
In the second experiment, Gopher and Eilam (1979) taught subjects 40
letters and functions in six one-hour sessions spread over three to five weeks.
For half of each session, subjects practiced text from a newspaper while the
other half they practiced nonsensical material. Subjects reached an average
speed of 56 CPM on text and 41 CPM on nonsense trials. Their performance
progress on both sets of material was close to linear.
For the same amounts of practice, the subjects in experiment #1
consistently typed 10 CPM faster then subjects in experiment #2. One
explanation of this phenomenon may be the larger character set in the second
experiment. The increase in number of choices of chords may have caused
reaction times to be longer and typing speed to be lower. This effect may mean
that peak performance is attained faster with smaller character sets. It is difficult
to interpret if maximal speed might be different or if time to reach top speed
varies by character set size.
2.4.10 Gopher, Hilsernath, and Raij (1985): Gopher (1986): and Gopher and
Raij (1
In a set of related studies, Gopher et al. (1985), Gopher (1986), and
Gopher and Raij (1988) performed experiments on a one-hand five-key chord
keyboard, one key for each digit. A total of 31 chords were possible considering
1, 2, 3, 4, or 5 finger combinations. The most recent and comprehensive of
these studies, reported by Gopher and Raij (1988), will be discussed in detail.
Three groups of subjects were assigned to one of three keyboards: the
QWERTY type keyboard, a one-hand chord keyboard, and a two-hand chord
keyboard (each hand could produce all characters). Gopher and Raij (1988)
29
taught all subjects 22 Hebrew letters and the space character. All subjects in
the three groups were able to memorize the patterns for the characters in 30-45
minutes. The subjects then participated in 35 one-hour sessions; three to four
sessions were held a week. Each session consisted of 15 three-minute typing
trials: each trial was followed by a one-minute rest break. The average CPM for
each session (mean of 15 trials) was then plotted versus session to obtain a
learning curve (Figure 9). The characters per minute plotted on the chart is an
average over all subjects.
At the end of the thirty-fifth session the performance of the one-hand
chord keyboard group was 178 CPM; for the two-hand chord keyboard group
performance was 208 CPM, while the QWERTY group had a speed of 118
CPM. The results of the one-hand chord keyboard are most applicable to this
study. The rate of learning between sessions 20-24 was 4.36 letters per
session for the one-hand chord keyboard; between sessions 26-35, the rate of
learning had dropped to 1.86 letters per session. This indicates a flattening of
the learning curve beyond 30 hours of training. While apparently peak
performance had not been obtained, progress was clearly slowing for the one-
hand chord keyboard group after 30 sessions.
2.4.11 Richardson, Telson, Koch
Richardson et al. (1987) compared three types of keyboards for a mail
encoding task: a one-hand conventional calculator layout, a two-hand 10-key
sequential keyboard like the top row of a QWERTY layout, and a two-hand 10-
key chord keyboard with 5 keys for each hand. Richardson et al. were
interested in data entry of ZIP codes so the subjects were taught only the ten
30
220
200 4
180
160 4
140 4
120 4
100 4
80 -
Choracters
/ Minute
60 -
+>
40 + °
Two hands chord
Single hand chord
Standard QWERTY
0 i T TT T
0 10 20
Session Number
Figure 9. Typing Speed for Three Keyboards as a Function of Practice. Taken from Gopher and Raij (1988).
31
30
numerals. On the chord keyboard, each of the hands could input all ten
numbers. Subjects were presented single numbers and required to press the
correct key or chord. They had to encode an average of 100 numbers per
minute with 95% accuracy for three consecutive tests (300 numbers per test).
Next, two digit strings were presented with a passing criterion of 70 strings per
minute with 95% accuracy. Finally, training was completed with subjects
entering three digit strings at a rate of 65 strings per minute with 95% accuracy.
Training was conducted in sessions that were three hours and ten
minutes long. Within each session, subjects practiced for 30 minutes, followed
by a 10 minute break, followed by 30 minutes of practice, and so on. The mean
time for training of the calculator and sequential keyboards to the criterion of two
digit strings (70 strings per minute or 140 CPM) was estimated at 22 hours; for
the chord keyboard, it was 97 hours. Therefore, the time to reach 140 CPM fora
character set size of 10 was 22 hours for the two sequential keyboards and 97
hours for the chord keyboard. However, the training methods used to teach the
subjects were peculiar. First, the subjects were not given any feedback on their
accuracy during a trial. It is also not known if the subjects knew the final
criterion needed to advance to the next stage. Subjects may have had to
randomly set their speed and accuracy levels until they hit the correct criterion.
Second, although the chord keyboard group could enter the strings
independently with each hand they were not given specific instructions such as
enter the first number with the left hand and the second with the right hand.
32
2.5 Ternary Chord Keyboard Experiments.
As stated earlier, the ternary chord keyboard differs from all chord
keyboards discussed so far in that its keys have three states instead of just two
(Figure 1). Therefore, previous studies on the ternary chord keyboard (TCK)
provided much of the foundation for this study .
2.5.1 Fathall - TCK #1
A study to compare performance on the TCK versus the QWERTY
keyboard was conducted by Fathallah (1988). Figure 10 shows the Ternary
Chord Keyboard Model #1. Four subjects were trained to use the TCK, and
another four learned the QWERTY keyboard. Both groups of subjects were
taught a total of 17 characters composed of 15 letters, space, and help. All
subjects were taught the characters in the same manner. For the TCK group,
chords were generated by using only two fingers, one finger from each hand.
Practice sessions (and al! other sessions) lasted about 1 1/2 hours broken into
15 minute sessions of typing followed by 5 minute breaks. Five characters were
learned per session and then reviewed on the fourth and final practice session.
Then, in what was called training sessions, the subjects were given text files to
type which were made up of words which used only the 15 letters. The files
were approximately 880 characters long. Subjects were trained until they met
the following criteria: 97% accuracy per character, and a mean response time of
800 msec per character. The two keyboards were evaluated in terms of number
of 15-minute sessions needed to meet the criteria: the average number of
sessions were 10 for the QWERTY group and 12 for the TCK group (these were
not significantly different).
33
"(afeos 0}
Jou ‘WO
U! SUO!SUBUUIP)
L# [EPO
PAeOghey ProYyD
AWeUsa! ‘O}
euNbI4
A
< 0'Sz
. Ws
G6
34
After training was completed, subjects participated in what was called
test sessions. Four 15-minute test sessions were given per day for five days.
On the fifth day average speed was 545 msec/character (110 CPM) for the
QWERTY group and 571 msec/character (105 CPM) for the TCK group; the
difference was not significant. In summary, subjects attained an approximate
typing speed of 107 CPM with a character set size of 16 after an average of 20
total hours of practice on either keyboard.
2.5.2 Kroemer (1990) - TCK #2.
Kroemer (1990) reported on the first training of subjects on TCK #2
(called #2 because it was modified from the one used by Fathallah) with a full
character set. Figure 11 shows the TCK #2. A total of 64 characters consisting
of 26 letters, 10 numbers, 4 cursor functions, and 24 punctuation marks,
symbols, and editing features were taught. Twelve subjects were trained in two
one-hour sessions per day. Characters were presented in "blocks" of six to ten.
Subjects were taught one block of characters and tested. Then another block of
characters was added until all 64 characters were learned. Subjects were
considered to have learned the keyboard once they could pass a test including
all the characters (presented four times each in random order) with 97%
accuracy two times consecutively. On the average, subjects learned all
characters in approximately four hours.
The next stage in the experiment was aimed at increasing the subjects’
speed. They were first given 20 text files that were 406 characters long, each
called a "trial". These were called "equal distribution trials" because each of 58
characters (the 4 cursor functions, shift, and backspace were not incorporated in
35
Figure 11. Ternary Chord Keyboard Model #2.
36
these trials) occurred 7 times, though randomly presented. By the twentieth
trial, subjects were averaging 54 CPM. Next, 20 text trials, 390 to 411
characters long, were typed by the subjects; the text, consisted of meaningful
sentences which attempted to capture the majority of the characters. At the end
of the fortieth trial, average speed was 78 CPM. The performance progress
over all 40 trials is shown in Figure 12.
In this experiment, subjects completed four trials in each one-hour
session. Since some subjects were faster than others, not all needed a full hour
to type the trials. Therefore, it is difficult to relate performance to hours of
practice. However, the total time spent using the keyboard averaged
approximately 12 hours.
2.5.3 Kroemer (1990) - TCK #3.
Kroemer (1990) reported on experiments with a modified keyboard, TCK
#3, as shown in Figure 13. The training procedure for TCK #3 was essentially
the same as TCK #2, except each subject was trained in one two-hour session
per day. Ten subjects learned all 64 characters in an average of about three
hours.
After training, subjects worked on improving speed, continuing the same
schedule as in the learning phase, i.e. one session per day lasting two hours.
The subject typed 40 text "trials" of meaningful sentences; each trial was 391 to
513 characters long. (No “equal distribution" trials were presented.) On the
average, subjects began trial number one keying 36 CPM; by trial 40 they
reached an average of 70 CPM. The performance progress for the 40 trials is
shown in Figure 14. Total time spent on the keyboard can not be directly
37
‘ZH pueoghsy
psouy Aveusa]
uy} JO}
BAUND Burwee
“z} aunbig
JEquNN jeu
Ov 8&
9E VE
cE OF
BC 92
He Ceo
OS BSL
OL FL
Gh OL
8 9
F @
as | -
——_o—__
SAV ——™——_
as |
+ —_-)——
Ol
02
0€
Or
0S
09
OZ
08
06
001
alnul—; sed syajoeseyuy
38
‘C# lepO-
Pueoghey
Puoyo Aueuwsal
“E4 eunbi4
39
"e# preoghey
puoyyD Ayeusay
ay} 10}
BAIND Buiwsee7
“p} eunBi4
JOQUINN Jeu
Op 8&
9E VE
ce OF
B82 92
He ce
OC BL
OE FPL
2h OL
8 9
FY @
O
foo }
| {
{ {
{|
0 FE
] 1
I tT]
I J
Tt |
] Pt
as | - —
—o——
SAV ——-—
+
0S
as |
+ —_-)—_
— OO0L
40
elnull sed suajoeieyuD
expressed by the number of trials since time to type each trial was not constant.
However, the total hours spent practicing on the keyboard was approximately
10 hours.
Two subjects continued practicing past the first 40 trials; each subject
typed an additional 120 trials, for a total of 15 to 17 hours. Their highest inputs
per trial were 134 and 183 CPM, respectively. Yet, until the end both subjects’
performance still increased in a seemingly linear fashion.
2.5.4 McMulkin and Kroemer (1991) and L nd Kroemer (1
A pilot study was conducted on a one-hand TCK by McMulkin and
Kroemer (1991) and Lee and Kroemer (1991). The apparatus used for the
experiment was the right-hand side of the keyboard used by Kroemer (1990),
TCK#2 shown in Figure 11. A total of 24 two-finger chords can be produced on
the one-hand TCK. The method in which chords were generated on the one-
hand TCK is different from the method used with the two-hand TCK. For the
two-hand TCK, chords were always produced by using one finger from each
hand; on the one-hand TCK, chords were produced by using two fingers from
the right hand.
Twelve subjects were taught 24 letters of the alphabet (V and Q not
included). Chords were grouped into blocks of eight letters. In teaching a block
of eight finger patterns, care was taken to make sure each chord was practiced
an equal number of times, which is different from the training program used in
the earlier experiments (Kroemer, 1990). The subjects learned all 24 chords in
an average of one hour.
41
The subjects returned on four consecutive days for one-hour testing
sessions. In each session, they completed five "trials". Each trial consisted of
the 24 letters randomly presented 10 times for a total of 240 characters. The
average typing speed began at 33 CPM and reached 63 CPM by the twentieth
trial. The performance progress is shown in Figure 15. The main purpose was
to collect response times for all 24 chords, so that the 18 "fastest" could be
selected.
2.6 Length of Training Session.
As alluded to earlier, the schedule of training can have a profound effect
on progress of speed acquisition. Glencross and Bluhm (1986) trained 242
persons on the QWERTY keyboard. Two hundred thirty-three people were
trained using ten 45-minute sessions spread over two weeks, one lesson per
weekday. Nine subjects received the same lessons, except that two sessions
were given per day for five days. At the end of the training, the two-week group
had an average keying speed of about 14 WPM as compared to the one-week
group's speed of about 11.2 WPM, a significant difference (p=0.012).
Baddeley and Longman (1978) conducted the most comprehensive
experiment on the effect of training session length on acquisition of typing
speed. Four levels of training schedule were established: one one-hour
session per day (1 X 1h); two one-hour sessions per day (2 X 1h); one two-hour
session per day (1 X 2h); and two two-hour sessions per day (2 X 2h). Multiple
sessions were separated by at least two hours of break. All groups were trained
for at least 60 hours; this means the 1 X 1h group had a total training period of
12 weeks, the 1 X 2h and 2 X 1h groups trained for 6 weeks, and the 2 X 2h
42
‘pueoghey puoyy
Areusla] PUBH-8UO
Oy} JO}
BAIND Buiusee]
“S} eunbi4
JOQUINN [UL
Oc 6|
St ZL
OF StL
VI EF
cl LL
OO 6
8
Jo {of
| |
| |
{of
| —|
FY
gs |
- ——o——_
SAV ——a-—_
as |
+ ——_)»—_
Ol
0c
0€
Ov
OS
09
OZ
08
06
anu Jed suajpoeieyD
43
group trained for 3 weeks. The training time until the groups became
operational was listed as follows: 1 X 1h - 35 hours, 2 X 1h - 42.6 hours, 1 X 2h-
43.2 hours, and 2 X 2h - 50 hours. This means the latter groups were operating
less efficiently per hour, but completed training much faster in practical terms of
weeks.
After becoming operational, the remaining portion of the 60 hours was
spent increasing speed. Typing tests were given once an hour. At the end of
60 hours each of the groups obtained the following speeds: 1 X 1h - 79 CPM, 2
X 1h- 73 CPM, 1 X 2h- 71 CPM, and 2 X 2h-65 CPM. The speed of the 1 X th
group was significantly faster than the 1 X 2h and 2 X 2h groups. The 2 X th
group was significantly faster than the 2 X 2h group. The performance progress
by hour is shown in Figure 16. An important finding was that at every hour of
practice, the rank order of groups from fastest to slowest was 1 X 1h, 2 X 1h, 1 X
2h, then 2 X 2h.
Baddeley and Longman's (1978) results indicated where the optimum
efficiency for training period length lies. Although less efficient, longer sessions
decreased the total number of weeks needed to perform an experiment. To
further illustrate this point, the 2 X 1h and 1 X 2h groups reached 80 CPM (the
60 hour level of the 1 X 1h group) in approximately 75 hours or 7.5 weeks. At
that time, the 1 X 1 group had barely become operational.
2.7 Format of Text for Testing.
The format of the text used to test keying performance can affect the
measures of speed. The main variable in text is whether it is nonsensical, such
as in a random presentation of letters, or whether it makes sense such as by
44
z Ss %F n w x
2 80F
g - 707 UO
Ecol YU q
50 ‘ . . . 40 50 60 70 BO
HOURS OF PRACTICE
Figure 16. Effect of Training Schedule on Typing Performance. Taken from Baddeley and Longman (1978).
45
having random words, phrases, or sentences. Gopher and Eilam (1979),
Sidorsky (1974), Alden, Daniels, and Kanarick (1972), and Seibel (1972) all
indicated that faster performance will be obtained if the text material makes
some sense. However, in this study, data were entered in the form of numbers
and mathematical operators; there seems to be little redundancy or predictable
order in numerical data entry (Seibel, 1972). Because of the lack of
redundancy in numerical data, this present study probably did not find keying
speeds comparable to a study which might have used sentences.
2.8 Sequences of Finger Movements that Yield Fast Typing Rates.
The time between keystrokes on the QWERTY keyboard has been
measured by several investigators (Salthouse, 1986; Salthouse, 1984; Gentner,
1983; Evey, 1980; and Kinkead, 1975). There has been general agreement
that the categories of consecutive keystrokes that rank from fastest to slowest
are as follows: 1) successive strokes from different hands, e.g. "th" , 2)
successive strokes from different fingers of the same hand, e.g. “er" , 3) same
key struck twice by the same finger, e.g. "ss" , and 4) different keys struck by the
same finger, e.g. "ed". Theses rules hold for skilled typists, not beginners.
Again, these rules do not seem to be helpful for numerical entry. Since
numerical entry has no order or redundancy, it is difficult to develop design
guidelines for assigning the numbers to particular chords based on the four
rules above. In addition, since the experiment conducted was constrained to
one-hand use of a keyboard, the fastest variety of successive keystrokes
(between hands) could not be implemented.
46
2.9 Which Hand to Use on a One-Hand Keyboard.
There is some disagreement on the issue of which hand should be used
in a one-hand keyboard task. Morgan, Drake, Heck, and Long (1981) trained
two groups of people on a 10-key calculator, a left-handed group and a right-
handed group. They found the left hand group to be significantly faster.
However, Dvorak (1943) and Dvorak, Merrick, Dealey, and Ford (1936) argue
that the right hand should be used more frequently in keyboard tasks. Since
there is no clear evidence of an advantage for either hand, the right hand was
assumed to be appropriate for the completed experiment. Also, using two
fingers of the same hand to generate a chord requires greater dexterity than a
Strict tapping task. The preferred hand probably possesses greater dexterity.
2.10 Literature Summary.
The ability of people to learn keyboard operation depends (among other
variables) on the number of characters taught, type of keyboard used
(sequential or chord), and criterion employed to establish proficiency. Total
amount of training time, number of trials or sessions, and total inputs have been
used to predict and analyze performance progress. It does not appear that any
of these criteria have been investigated to determine the most effective one.
Table 1 provides a summary of the findings, as they pertain to the objectives of
the present study.
The first experimental objective is to investigate the performance
progress of subjects on a keyboard. Few studies have evaluated keyboard
performance from initial learning to peak performance; thus, performance
progress on keyboards is not well Known. The training time to maximal
47
Table 1. Summary of Important Variables in Keyboarding Studies
Total Provide
Type of | Character| Training Total Char- CPM Learning
Author | Keyboard| Set Size | Schedule Hours acters at end Curve?
Chapman QWERTY 65 n/a 180 n/a 240 yes (1919)
Thurstone QWERTY 65 2 hrs/day 280 250 pages 205 yes
(1919) 5 days/wk
Green QWERTY 65 1 hr/day 360 n/a 235 yes (1940) 5 days/wk
Klemmer Chord 30 1 hr/day 35 15,000 225 yes
(1958) 3 days/wk words
Lockhead Chord 136 1 hr/day 45 na 40 no
& Klemmer 50 days
(1959)
Cornog Chord 36 1.5 hr/day 190 n/a 232 no
et al. (1963) 11 weeks
Conrad & QWERTY 36 3.5 hrs/day 120 n/a 92 yes Longman Chord 36 5 days/wk 120 n/a 100 yes
(1965)
Bowen& QWERTY 11 n/a 20 n/a 80 no
Guinness Chord 11 n/a 20 n/a 50 no (1965)
Sidorsky Chord 40 0.5 hr/day 5 n/a 63 no
(1974) 10 days
Rochester Chord 54 50 min/day 90 n/a 200 yes
et al. (1978) 5 days/wk
Gopher & Chord 30 7, 1hr sess 7 n/a 65 yes
Ejlam Chord 40 6, 1hr sess 6 n/a 56 yes
(1979)
Gopher et Chord 23 1 hr/day 35 na 208 (2-hand) yes
al. 85, 86, Chord 23 3-4 day/wk 35 n/a 178 (1-hand) yes
and 88 QWERTY 23 35 sess 35 n/a 118 yes
Richardson Chord 10 3 hr/day 97 n/a 140 no
et al. 30 days (1987)
Fathallah Chord 17 1.5 hrs/day 20 n/a 105 yes
(1988) QWERTY 17 12 days 20 na 110 yes
TCK #2 Chord 64 2, 1hr sess 12 16,000 78 yes (1990) per day
TCK #3 Chord 64 2 hrs/day 10 18,000 70 yes
(1990)
One-Hand Chord 24 45 min/day 3 2,400 63
TCK (1991) 5 days
48
performance seems to be largely dependent on character set size, see Table 1.
Since only 18 chords are used for this study, the training time is expected to be
considerably shorter then for a set size of 64. The training schedule has a large
effect on the learning curve; this study uses a schedule that is as efficient as
possible to obtain results rapidly.
The second experimental objective is to find a function which predicts
learning and evaluate three different criteria to assess which criterion is most
effective at predicting performance progress. Only three of the reviewed studies
attempted to find a function which predicted learning, Thurstone (1919),
Kroemer (1990), and McMulkin and Kroemer (1991). The conclusions that can
be drawn about each criterion from the literature may be summarized as
follows:
1) Number of hours of practice
Fourteen of the seventeen studies summarized in Table 1 used practice
time to describe performance progress. In none of these studies was it
investigated whether this was the best predictor. The assumption that hours of
practice was indeed the best criterion should be tested against other predictors,
which is one goal of this study.
The amount of practice to reach maximal performance with an 18-chord
set was estimated from the studies at 50 hours. This was based on the
assumption that the subjects would reach a peak speed of close to 200 CPM
(this is an interpolation of the results listed in Table 1). With character set sizes
of 50 or more, 200 CPM was reached after about 200 hours (Chapman, 1919;
Thurstone, 1919; and Green, 1940). Therefore, with a character set size of only
49
18 the time was expected to be about a fourth as long. However, due to the
differences in training schedules and method of testing, this remains a guess.
2) Number of trials or characters
Only five of the studies reviewed in Table 1 recorded the number of
characters practiced as a function of performance. In none of these studies
were subjects able to attain peak performance, so an estimate of characters
practiced until peak performance was almost impossible. There was no
discussion that number-of-characters-typed was the best predictor in any of the
five studies. Only Thurstone (1919) provided learning curves for both hours of
practice and pages typed. However, he did no analysis to show which criterion
provided a better prediction.
3) Subjects’ Opinion
None of the studies considered the opinion of the subject as a method to
assess how close or far the subject was from maximal performance.
50
3. METHOD
3.1 Subjects.
Five subjects participated in this experiment. The only criteria used to
select subjects was that they write with their right hand and have no history of
wrist problems (such as carpal tunnel syndrome). Four of the subjects were
female, and one was male. The subjects’ ages ranged from 20 to 29 years.
The total time commitment of each subject was 60 hours. Subjects were
compensated at a rate of $5.00 per hour, plus a $50 bonus for participating for
all 60 hours, for a total of $350. None of the subjects dropped out of the
experiment.
3.2 Apparatus.
The one-hand Ternary Chord Keyboard, shown previously in Figure 11,
was used for this study. Activation of a key was detected by microswitches. The
microswitches were connected to an Input/Output carrier on an IBM (AT)
computer via a bus cable. A software program (written by VATELL Corporation)
interpreted and manipulated inputs from the TCK.
For each trial the following data were recorded: subject number, trial
number, total time to complete the trial, and the number of errors committed by
the subject. Also, the following data were collected on every chord within each
trial: subject number, trial number, chord number, and response time for the
chord. If the chord was not keyed correctly, the incorrectly entered (actual)
chord was recorded.
51
3.2.1 Dimensions of the One-Hand TCK.
The physical dimensions and positions of the keys for the one-hand TCK
are shown in Figure 17. A palm rest was provided for the subjects to use if they
wished. The wooden palm rest was shaped approximately like a half cone,
11.5 cm long, sloped from a radius of 2.7 cm at the thumb side of the palm toa
radius of 1.6 cm at the little finger side of the palm. The palm rest was
adjustable longitudinally (4.5 to 11 cm from the center of the keys to the center
of the palm rest) and laterally (1 cm in either direction from the center). It was
fastened to the base of the keyboard by velcro strips. The entire keyboard base
could be rotated in roll, pitch or yaw. The keyboard (rotated as preferred by the
subject, no specific pattern) and chair height were adjusted so that the subject
felt comfortable, the wrist was straight, and the elbow formed about a 90 degree
angle.
3.2.2 Chord Coding.
The method to assign the numbers and functions to the chords was
determined largely from the experiments conducted by McMulkin and Kroemer
(1991) and Lee and Kroemer (1991). Of the 24 possible chords, 6 were
eliminated, based on user preference and response times. The remaining 18
chords were assigned by VATELL Corporation to the numbers and functions.
On an adding machine, it was assumed the ten numbers, “decimal point", and
"plus", "minus", "multiply", "divide", "subtotal", and "total" functions are all used
with about equal frequency. No literature could be found which ranked the
frequency of these 18 numbers and functions. Therefore, some of the
assignments were chosen based on ease of learning. All the "odd" numbers
52
“othe
2)
D a
® a
>
OE
O'@ oo
$s
were assigned to chords in which both the fingers moved backward. All the
“even” numbers were assigned to chords in which the fingers moved forward.
The "backspace" chord was assigned to the chord in which the index finger
moves backward and middle finger forward; this is like turning a wheel where
the top of the wheel moves to the left. No learning principle existed for the
remaining seven functions, so they were matched with the remaining seven
chords. Because each of the chords was to appear equally often in the text
trials, the chord assignments should not have had an effect on performance.
Figure 18 shows how the chords were assigned to the characters. The letter S
was used to represent "Subtotal", the letter T represented "Total", and an arrow,
© , represented "Backspace".
3.3 Procedures.
3.3.1 Experimental Design.
One of the objectives of the experiment was to determine a function
which effectively predicts learning and evaluate a set of criteria for predicting
performance progress and peak keying speed. Therefore, the regressor
variables used to predict performance were number of hours spent practicing,
and number of repetitions (trials). The dependent measures recorded were
time to complete a trial/test, number of errors during a trial, and opinion of the
subjects about their performance at the end of each week. From the first two
dependent measures, characters per minute were computed for each trial and
for each time increment of practice.
54
Finger: Index
Middle
Ring
Little
(The chord numbers describe which fingers were used to generate a chord.
The first number of the chord represents the first finger's direction while the
second number represents the second finger and its movement direction.
For example, the chord number 36 represents the middle finger moving
backward and the ring finger moving forward.)
Character
i,
+o OOn oar
|] DY
*
/
. (Decimal Point)
S (Subtotal)
T (Total)
€ (Backspace)
Chord
15
26
17
28
35
46
37
48
57
68
24
23
45
36
13
25
16
14
Figure 18. Chord Assignments for the Numeric One-Hand Ternary
Chord Keyboard.
55
3.3.2 Schedule of Practice.
Subjects were scheduled for one one-and-a-half hour session per day,
five days a week. Subjects participated for eight weeks for a total of 60 hours.
(As pointed out in the literature review this may not have led to optimum
efficiency for each hour of practice, but the study was to be completed in
approximately eight weeks.) Only one keyboard was available for
experimentation; this meant one-and-a-half hour sessions required a chunk of
7.5 hours per day be allotted to testing (this was close to the upper limit of the
time available by the experimenter). Subjects were scheduled for the same
times each day. Having the sessions the same time each day was the most
convenient for the subjects and greatly reduced the probability of them dropping
out. Daily sessions also allowed the subjects to participate in the experiment
before, after, or between classes, and avoided conflicts with other activities.
Overall, 1.5-hour sessions offered the best compromise between utility of each
hour of practice and convenience for the subjects.
Due to a variety of reasons, each subject missed some of the sessions.
Any session that was missed was rescheduled for another day either at the end
of the eight weeks or on a weekend. Subjects never participated in two
sessions in one day.
3.3.2.1 Breakdown of Sessions.
The first session was dedicated to teaching the subjects the 18
characters (training program described in detail in the next section). The
second through the fortieth sessions had the same format. The subjects were
given text consisting of 504 characters to input. Typing each such text was
96
called a “trial” (the terms text and trial are used interchangeable from this point
on). During each session the subject was presented a trial to type. The subject
was instructed to input the text as rapidly as possible maintaining 97%
accuracy. The subjects took approximately 15 minutes to type the first trial.
After the end of the trial, subjects were given a break of two to three minutes and
then started another trial. As the subjects typed the trials faster, they were given
shorter breaks, of about one minute to maintain a fairly constant ratio between
practice and rest. The experimenter determined how long the rest breaks would
be so that the ratio of practice to rest was approximately constant (about two to
three minutes practice to every one minute of rest). The session continued in
this way until the end, after one-and-a-half hours.
3.3 2.2 Training Program.
During the first session the subjects were taught the 18 chords. The
characters were taught in two “blocks”: first, the numbers 1 to 9, and 0; second,
the functions "plus", "minus", "multiply", "divide", "decimal point", "backspace",
"subtotal", and "total". Within each block, the first character was presented
along with the finger pattern corresponding to that character. The subject then
practiced the character five times. The second through fifth characters were
presented in the same manner. A "mini-quiz" was then given over the first five
characters in the number block, each character presented in random order four
times. During the "mini-quiz", no finger patterns were displayed for the
characters unless the subject entered a chord incorrectly. The sixth through
tenth characters were then presented five times each. A "mini-quiz" was then
given over the final five characters in the number block. After the first learning
a7
block, a quiz was given over all the characters in that block; each was
presented in random order a total of three times. The subject had to pass the
quiz two consecutive times with 97% accuracy; there was no time limit. An
accuracy of 97% was selected so that subjects could not incorrectly enter one
chord all three times it was presented and still pass. If the subject did not pass
the quizzes, the learning block was repeated and the quizzes repeated at the
end of that learning block.
After passing the quizzes, the next block of characters was presented in
the same manner. Quizzes were given over the second block to ensure the
subject had learned the eight symbols. Upon completion of the second learning
block quizzes, a cumulative quiz over all characters was given. Each character
was presented randomly three times. The subject had to pass the quiz two
consecutive times with 97% accuracy (again subjects could not miss 1 of the 18
chords all three times and still pass). If the quiz was failed, the subject had the
option of reviewing a block or retaking the final quiz. After passing two
cumulative quizzes, the subject was considered to have learned the 18
characters.
The number of keystrokes performed during training as well as the time
to complete the training session was recorded for all subjects.
It might be argued that this training program had not been experimentally
verified as being the most effective method of teaching the chords. This is true,
but the same training program has been used successfully in the experiments
conducted by Kroemer (1990), and McMulkin and Kroemer (1991). There a
total of 64 chords were taught in three to four hours, and 24 chords were taught
08
in an hour. These results indicated that the training program brought about a
short training time.
3.3.2.3 Mak { Trial
Each trial contained 504 characters; therefore, each of the characters
was presented 28 times. The trials were generated using a PASCAL program
which randomly selected the order of the characters. Each trial presented to the
subjects was a different trial created by the program. However, all the subjects
were presented with the same trials in the same order. As subjects’ keying
speed became faster it took the subjects less time to complete the trials. Due to
the variable keying speed of the subjects, some typed trials that were never
presented to others. Appendix A shows an example of a trial.
3.3.3 Determining Peak Performance.
A problem in analyzing the experimental results is to establish a subject’s
peak performance, i.e. the fastest keying speed possible. Bittner (personal
communication, April 11, 1991) suggested having the subjects perform a series
of characters over and over; he assumed that this allows them to develop a fast
motor skill pattern, thus give a measure of peak performance.
Following Bittner’s suggestions, at the end of the final session, subjects
continuously typed a single series of characters. Five different character strings
were used: 6.7-3S, 4/0<2T, 584+1*9, 142-3. , and 123456. Each trial was 504
characters long, consisting of 84 repetitions of one of these five character
strings.
99
3.3.4 Subjects' Opinion.
Every subject knew that an objective of the study was to reach maximal
input speed. It was hypothesized that the subjects’ opinion of when they were
typing as fast as they could may be a good measure of when peak performance
was obtained. Once a week the subjects were asked to rate their performance
on a0 to 9 scale, 9 representing the anchor for peak performance; Figure 19
shows the scale that was used.
3.4 Experimental Protocol.
The experiment was conducted in 40 one-and-a-half hour sessions, 5
days a week, for 8 weeks. In the first session, the subject was given the written
experimental instructions to read (Appendix B). The experimenter then
reviewed the instructions and answered any questions. The subject was given
the informed consent form to sign (Appendix C). The subject was then seated in
front of the keyboard; the seat height, the keyboard, and palm rest were
adjusted until the subject was comfortable. The subject was not required to use
the palm rest. Two subjects used the palm rest throughout the entire
experiment. Two subjects never used the palm rest, and one subject started
using it about halfway through the experiment. The subject was presented
introductory information on the one-hand TCK and started with the training
program. After learning the 18 chords, the subject was dismissed for the day.
At the start of the second session the subject was again required to take
a cumulative quiz over all 18 characters until it was passed. The subject was
then presented with the first trial to type. Additional trials were presented until
the end of the session. As the session approached the end of one-and-a-half
60
1 __ Beginning Keying Speed
2 __ Very Slow Keying Speed
3 __ Slow Keying Speed
4 _ Below Moderate Keying Speed
5 _ Moderate Keying Speed
6 __ Above Moderate Keying Speed
7 _ Fast Keying Speed
8 __ Very Fast Keying Speed
9 _ Fastest Possible Keying Speed
Figure 19. Scale Used by Subjects to Rate Their Performance Each Week.
61
hours, the experimenter decided if enough time remained for an additional trial.
The third through fifth sessions paralleled the second; the subject was required
to pass a quiz and then practiced keying text trials. After the fifth session (after
about 7.5 hours of practice), the subject was no longer required to take the
cumulative quiz. During the sixth through the fortieth sessions, the subjects
practiced keying text. At the end of the fortieth session, the subject was paid,
thanked, and dismissed.
For all trials, the subject was instructed to increase input speed while
maintaining at least a 97% accuracy level; 97% was selected to match the
accuracy stressed in the training program. To help control the speed versus
accuracy tradeoff, feedback was given to the subjects at the end of every trial.
As stated before, for each trial the following data were recorded: subject
number, trial number, total time to complete the trial, and the number of errors
committed by the subject. Immediately after the completion of each trial, the
characters per minute and accuracy were calculated from these data and
displayed on the screen along with total time to complete each trial and number
of errors. The subjects received immediate feedback on their speed and
accuracy. They could then make adjustments in their performance to increase
input speed, but maintain their accuracy levels. No specific rewards were given
to the subjects to maintain 97% accuracy or increase their speed; presenting
feedback of their performance to them was considered an effective way to
control the variability due to speed versus accuracy.
62
4. RESULTS
4.1 Learning Results.
All five subjects passed the final training test within the first one-and-a-
half hour session; the average learning time was 35 minutes (including quizzes)
with a standard deviation of 8.3 minutes to achieve the required minimum of
97% accuracy on all 18 chords. The number of times a subject had to go
through each training block ranged from two to four times and the number of
quizzes per block taken to pass ranged from two to six.
4.2 Speed - Accuracy Performance.
The subjects were instructed to increase speed while maintaining a 97%
accuracy level. Table 2 shows, for each subject, the average accuracy and
standard deviation of the trials typed over the entire experiment. All the subjects
had an average accuracy greater than 97%.
4.3 Using Number of Trials to Predict Performance.
The five subjects each completed different numbers of trials over the 60-
hour experimental! period, see Table 3. Analyses were done using individual
subjects’ data, except for chord and average performance analyses. This
means that regressions for individual subjects include different numbers of data
points, e.g. 487 points for subject 1 and 636 points for subject 3.
One of the objectives of this experiment was to find an equation that
accurately describes the performance progress of each subject. The dependent
variable being predicted is characters per minute. The keying speed of the
63
Table 2. Average Accuracy Across Trials for Each Subject.
Average Standard Subject Accuracy Deviation
(in %) (in %)
1 97.39 0.76
2 98.51 0.61
3 98.20 0.80
4 97.75 0.85
5 97.19 0.94
64
Table 3. Number of Trials Completed by Each Subject
Number of Trials Completed
Subject 1 487
Subject 2 622
Subject 3 636
Subject 4 509
Subject 5 516
65
subjects in characters per minute was calculated by dividing the correct entries
(total number of characters minus number of errors) typed during a trial by the
total time needed to complete the trial.
The major criterion used to determine which equations should be used
was that they predict performance in a monotonically increasing fashion.
Equations which would predict increasing and then decreasing performance
were not considered, such as polynomial equations (e.g. By + Bi*X + Bo*X2).
Five candidate functions were considered to fit the data. First, a simple
linear relationship between characters per minute (CPM) and trials was used:
CPM; = Bo + By*T; (Equation 1)
where CPM; is the characters per minute typed on the i-th trial
(dependent variable), Tj is the i-th trial, and Bp and B; are fitted
coefficients.
Second, a relationship suggested by Bittner (1991) was used:
CPM; = Bo + By/T; + Bo/Ti2 (Equation 2)
The first two equations were combined to generate the third function:
CPM; = Bo + By*T; + Bo/T; + Ba/Ti2 (Equation 3)
Fourth, a logarithmic function was tried of the form:
CPN; = Bo + By*Ln(T}) (Equation 4)
The final attempt to fit the data was by a Log-Log function:
Ln(CPM;) = Bo+ By*Ln(T;), this reduces to
CPM; = C*T;81 (Equation 5)
66
where C = eBo: see Appendix D for the derivation.
The subjects’ performance was analyzed individually, not collectively.
Each of the five functions above was fit to each of the subject’s actual data. The
statistical package SYSTAT was used to calculate the coefficients using the
method of least squares. The criterion used to judge goodness of fit of the
functions is the coefficient of determination, R2, also calculated by SYSTAT.
Table 4 summarizes the R2 values obtained for each formula for each subject.
The regressions of the data of subject 2 are selected as a typical
example of each of the five function’s predictions plotted with the actual
performance in Figures 20 to 23 and 25. Comparing the Figures 20 to 23 and
25, it becomes apparent that the Log-Log function most accurately follows the
actual performance progress. This impression is supported by the statistical
findings that the Log-Log function accounts for the largest portion of variation,
i.e. has the highest R2.
The plot of actual performance and predicted performance from the Log-
Log equation (in Characters Per Minute typed) for each subject is shown in
Figures 24 to 28. The Log-Log equation which best fit actual performance of
each subject is listed in Table 5. Figure 29 shows the predictions for all
individual subjects.
The Log-Log equation which was used to fit the actual performance had
the form: Ln(CPMj) = Bo + By*Ln(Tj). The coefficients Bp and B; which have a
different value for each subject are listed in Table 6. A t-test can be used to
indicate if any of the Bo or By; coefficients for each subject are statistically
different from one another (Neter and Wasserman, 1974). All Bo coefficients
67
Table 4. Summary of R-Squared Values for All Functions
Considered When Trial is the Regressor Variable.
Equation
(487 to 636 observations for each fit)
CPMi = Bo+|CPMi = Bo+| CPMi=Bo+ |CPMi =Bo+| CPMi =
B1*Ti Bi*(1/Ti) + | B1*Ti+B2*(1/Ti)} Bi*Ln(Ti) | C*(Ti*B1)
B2*(1/Ti*2) | +B3*(1/Ti*2) {log-log}
Subject 1 0.899 0.361 0.927 0.896 0.961
Subject 2 0.825 0.430 0.903 0.921 0.953
Subject 3 0.817 0.427 0.897 0.924 0.955
Subject 4 0.819 0.498 0.920 0.953 0.973
Subject 5 0.802 0.440 0.879 0.898 0.939
Average 0.832 0.431 0.905 0.918 0.956
CPMi = Characters per Minute on the i-th trial
Bo, B1, B2, and B3 are fitted coefficients
Ti = Trial
C = Exp(B)
68
OS9 009
“QOURWUOLO ENO
pue feUL,+g
+ Og = INO
JO UOHIpald ‘Z alang
“og eunbiy
JaquNny jeuL
OSS 00S
OSb O00F
OSE O0€
OS¢ 002
i
W
| |
J. {
| |
| a
{ i
I I
l |
ai
it
\
HNN
if nae
i ‘i
|
ID a a
air Ny
[PHL.SELO +
SV O0L = Wdd
OSt OOF
0S
| L
| |
I t
+H fo
ay
Wl
Oc
Ov
09
08
OO!
Ocl
Ori
O9L
O8l
00¢
eynull sed suajpoeeyyD
69
“BOUBLUIOPSd jenjoy
pue Zyjeul/Zgt+
|EUL/LG +
OG =
INWdd JO
UOHdIPald ‘Z
y~alqns "1Z
eunbi4
JOquNnN jeu
0S9 009
OSS 00S
OSr O0v
OSE O0€
OS2 002
OSt OO!
0S
CvIEHI/8°9EL +
[EL L/9'B6Z - ES
6HL =IWdd
OV
08
Ocl
O91
00¢
anu sed suajoeseyD
70
‘OURO
emoy puke
Zyfeu L/eg + IEUL/ZA
+ IEUL.La + 0g
= INdd JO UONAIpeld
‘Z yalqns ‘zz
eunbi4 JOqUNN
jeul
0S9 O09
OSS 00S
OSb O0v
ose O00&
O0S¢ 00¢
£OS!I
| |
| |
| |
| |
| !
| I
| 1
I {
{ 1
I ]
| |
tea he
bi, A
Healt
\ \
| fl
i
hu af
\ Wl
Ae aN
A A
ny ry
ovIeH L/P ole
+ IEUL/C
C9E - IELL.VLL'O
+ 80°LOL
= Ndd
OOl OS
0c
OV
09
08
0O0-
Oct
Ort
O91
081
00¢
eInulpy Jed suapeseyy
71
0S9
“QOUBWUOHA jenjoy
pue (jeu
)UT.1@ +
OG = INdd
JO UOHIpPald
‘2 jalqns ‘Ez
eunbl4
JOQUNN
[BULL
009 OSS
00S OSr
OO0v OSE
O00€ ose
002 oS!
ool os
oO
! !
oS yy
| an
i My - }
mag I"
| yn
| art
M hhh
|
| Kal
Al i
fh
Vi ha
At hat h
y
A :
|
cH Te
(jeu )U7,91°9¢
+ 961°0
=INdOD
0d
OV
09
08
OOL
Oct
Ovl
O91
08}
00¢
eInuly Jed syaj~eieyD
72
", palqns
‘eoueWOLaY
jenjoy puke
uO!oIpal 4 607-607]
‘pz eunbi4
JOQUUNN Jeu
069 009
OSS 00S
OSr OOv
OSE O0€
OS¢ 000
OSl OO!
0S 0
[+ +}
| :
: I—
+ |
if yr
y
yr pe
| Pe
mT iV
T
r ot
NA} a
1 ML
deat Mr
1 di my
Rv aul
qi’ i
4 aL
(pSEOvieUL). LSB
= NdO
Oc
Ov
09
08
001
Ocl
Ort
091
08 |
00¢
e]null Jed srajoeueyy
73
0S9
‘2 yalqns ‘aoueBWOLaY
jenjoy pue
uoNoIpesg 607-607
“Sg aunbi4
JOQUINN Jeu,
009 OSS
006 OS
O0v OSE
oO0& OS¢
002 £OS/I
OOl 0S
0
+ +
/
) per?
|
wath ;
ul | fc
| {al
tk!
iy An
PG Ih
"a
NOH
Hl I
MW |
i ft
HR
i |
lh le ea
TTP (222 OvieUL).L2Z
kv = NdO
Oe
Ov
09
08
001
Oct
Ort
O9l
O8l
00¢
ENUF, Jed suajoeseyy
74
0S9 T
‘€ joalqns
‘eoueWOLad
jenjoy pue
uoloipesg 607-607
‘92g eunbi4
JEquINN |euL
009. OSS
00S OSh
O00b oOSt
O00€ OSe¢
000 OSI
O00! 0S
| {
—
| |
I |
I |
{
nah ft
Mi
!\ A
iin 4
iW IW
yen
Hd
nm Vi
|
(661 OvIGHL),
Ea 8h =
NWdO
r Oe
r OV
- 09
- 08
r OOL
r OSh
- Ol
r O9-
- O8t
L 00¢
elnull Jed suejpeseuyD
75
0S9 009
‘p yoalqns
‘eoueWOLad
jenjoy pue
uoloipald 607-607
“72 eunbi4
JOQUINN Jeu
OSS 00S
OSh O0v
OSE O0€&
O0S¢ 000
OS! OO!
0S
| |
| _t
| |
| |
| |
I I
I t
J t-
{ I
1 1
(Zya OvieUL),ZEL' CE = NdO
0d
OV
09
08
001
Ocl
a
O91
O8l
00¢
elnull—y Jed suajoeeyyD
76
0S9
‘g oalqns
‘eoueWWOLeY jenjoy
pue UO!OIpaig
607-607 “gz
aunbi
JEquUNN jeu
009 OSS
00S OSr
O0v OSE
O0€ OG2
002 OSt
OOl OS
0
t+— {
+ —t
4
J+ Vr
De +
Len] er
M up
\ hs Wh
i iy a7
| y en
hy |
(ZL 7 OvleuL),6Z6°9€ = WdO
0c
OV
09
08
OOl
Oc |
Sinull Jed suajoeeyD
OVl
091
081
002
v7
Table 5. The Fitted Log-Log Equation for Each Subject.
Subject Log-Log Equation Re
1 CPM; = 18.541*T;0.394 0.961
2 CPM; = 41.721*T,0-222 0.953
3 CPM; = 48.231*T,9-199 0.955
4 CPM; = 32.137°T;0-242 0.973
5 CPN; = 36.379°T;0-217 0.939
where CPM j is the characters per minute typed on the i-th trial
(dependent variable) and Tjis the i-th trial of practice
78
‘palqns yoeg
10} uonoipeig
607-607 “62
aunbi4
Jaquunu jeuy
0S9 009
0SS 00S
OSr O00v
OSE O0€
OSe 002
OS! 00}
0S
| t—
: |
. |
: [+
§ ang yang
Lang
LaviPHL.O = Nd
0¢
OV
09
08
OOl
Och
OVl
091
081
00¢
elnulp Jed syajoeieyy
79
Table 6. Statistical Comparison of Coefficients Bo and B1.
Comparison Between Bo Coefficients
Subject 1 2 3 4 Bo Value 2.9197 3.7310 3.8761 3.4698
1 2.9197 Difference Between Coefficients
2 3.7310 0.8113 *
3 3.8761 0.9564 * | 0.1451 *
4 3.4698 0.5501 * | 0.2612 * | 0.4063 *
5 3.5942 0.6745 * | 0.1368 * | 0.2819* | 0.1244 *
Comparison Between B1 Coefficients
Subject 1 2 3 4 B1 Value 0.3545 0.2218 0.1995 0.2422
1 0.3545 Difference Between Coefficients
2 0.2218 0.1327 *
3 0.1995 0.1550 * | 0.0223 *
4 0.2422 0.1123 * | 0.0204* | 0.0427 *
5 0.2166 0.1378* | 0.0051 0.0172 * | 0.0256 * |
* indicates a significant difference, at p < 0.001
80
were found to be significantly different from each other (p < 0.001), see Table 6.
The By, coefficients for subjects 2 and 5 were not found to be significantly
different each other (p = 0.10), but all other comparisons between B,
coefficients were found to be significantly different (p < 0.001), see Table 6.
4.4 Using Time to Predict Performance.
Another method of regressing performance is to predict the dependent
variable, characters per minute, from practice time. For example, performance
could be predicted from the equation CPM; = Bog + By*Log(Pj) where P;
represents the i-th time increment of practice, for example 30 minute increments
(Pj = 1, 2, 3, ..... ; not 30, 60, 90, .... ). During a given time increment of practice,
each subject typed a different number of trials.
All the data were collected in terms of time needed to complete a trial. To
create 30-minute time periods, the times to complete trials were added until the
total reached at least 30 minutes. Therefore, none of the periods were exactly
30 minutes, but none were less than 30 minutes. The time periods also
overlapped the sessions of practice. Session one might include the first 30-
minute period and half of the second 30-minute period; session two might cover
the other haif of the second period, all of the third, and then part of the fourth.
The overlap of the periods is not constant across subjects either. The
procedure to create periods was repeated for 40, 50, and 60-minute time
periods. The average length of the 30, 40, 50, and 60-minute time increments
for each subject is shown in Table 7, along with the standard deviation. The
characters per minute typed for a period was calculated by averaging the
characters per minute of the trials that constituted a given time period.
81
Table 7. Average Length (and Standard Deviation) of 30, 40, 50,
and 60 Minute Time Increments Used in the Regressions.
30 Minute Period Subject Average (in Min) | Std Dev (in Min) N
1 32.7 2.01 67 2 32.2 1.43 70
3 31.9 1.22 70 4 32.4 1.99 70
5 32.2 1.52 71
40 Minute Period Subject Average (in Min) | Std Dev (in Min) N
1 42.5 2.11 5 1
2 42.0 1.47 54
3 41.4 0.93 54
4 42.5 1.54 53
5 42.3 1.48 54
50 Minute Period Subject Average (in Min) | Std Dev (in Min) N
1 52.6 2.44 41
2 51.9 1.20 43
3 51.6 1.11 43
4 52.4 1.61 43
5 52.3 1.33 43
60 Minute Period Subject Average (in Min) | Std Dev (in Min) N
1 62.8 2.13 35
2 62.1 1.24 36
3 62.0 1.01 36
4 62.5 1.79 36
5 62.2 1.55 36
82
The average characters per minute typed was also calculated by
session. Although the sessions were one-and-a-half hours long, not all of that
time was spent typing because of rest breaks after each trial. Calculating
characters per minute by session does not indicate any specific time period, but
maintains the continuity of an entire day of practice.
The same five equations used to fit trials were tried with time periods:
CPM; = Bo + ByP; (Equation 1a)
CPM; = Bo + Bi/P; + Bo/P;2 (Equation 2a)
CPM; = Bo + ByP; + Bo/P; + Ba/Pi2 (Equation 3a)
CPMj = Bo + ByLn(P)) (Equation 4a)
CPM; = eBo p;B1 (Equation 5a)
where CPM; is the characters per minute typed on the i-th time
period (dependent variable), Pj is the i-th time period (30, 40,
50 or 60 minute periods or days) and B's are fitted coefficients.
Table 8 lists the R@ values found for all equations for each time period by
subject. The equations CPM; = Bo + ByP; + Bo/P; + B3/Pi2 and CPM; = eBo p;Bt
(Log-Log) provided the best fit for every time increment considered. Their R2
values are equal to at least 0.96 in every case. Appendix E contains the graphs
of actual performance and predictions using both CPM; = Bo + By;P; + Bo/P; +
Ba/Pj2 and CPM; = eo P;®1 for all time increments and subjects.
4.5 Subjective Ratings of Performance.
At the end of the every fifth session the subjects were asked to rate their
performance on a 0 to 9 scale (shown previously in Figure 19). The ratings for
83
Table 8. Summary of R-Squared Values for All Functions Considered
30 Minute Period
(approximately 70 observations
When Time Increment is the Regressor Variable. 30, 40, 50, and
60 Minute Increments and Sessions are Used.
CPMi = Bo + | CPMi= Bo+}| CPMi=Bo+ | CPMi=Bo+ CPMi =
B1*Pi B1/Pi + Bo*Pi+B1/Pi+ | B1*Ln(Pi) | e*Bo*(Pi*B1)
Ba/Pi*2 Ba/Pi*2 {log-log}
Subject 1 0.978 0.624 0.984 0.868 0.965
Subject 2 0.893 0.802 0.982 0.966 0.987
Subject 3 0.896 0.778 0.978 0.966 0.989
Subject 4 0.880 0.812 0.979 0.973 0.989
Subject 5 0.892 0.780 0.973 0.960 0.986
Average 0.908 0.759 0.979 0.947 0.983
40 Minute Period
(approximately 54 observations
CPMi = Bo +; CPMi=Bo+] CPMi=Bo+ | CPMi=Bo+ CPMi =
B1*Pi B1/Pi + Bo*Pi+B1/Pi+ | B1*Ln(Pi) | e*Bo*(Pi*B1)
Ba/Pi*2 B2/Pi*2 {log-log}
Subject 1 0.980 0.672 0.985 0.871 0.964
Subject 2 0.896 0.842 0.986 0.971 0.988
Subject 3 0.894 0.824 0.981 0.973 0.991
Subject 4 0.878 0.859 0.983 0.979 0.987
Subject 5 0.893 0.823 0.975 0.962 0.986
Average 0.908 0.804 0.982 0.951 0.983
50 Minute Period
(approximately 43 observations
CPMi = Bo + | CPMi = Bo+] CPMi=Bo+ | CPMi=Bo+ CPMi =
B1*Pi B1/Pi + Bo*Pi+B1/Pi+ | B1*Ln(Pi) | e*Bo*(Pi*B1)
B2/Pik2 Ba/Pi*2 {log-log}
Subject 1 0.981 0.716 0.986 0.878 0.968
Subject 2 0.898 0.873 0.989 0.975 0.989
Subject 3 0.896 0.850 0.981 0.975 0.989
Subject 4 0.882 0.883 0.986 0.982 0.986
Subject 5 0.894 0.861 0.979 0.986 0.987
Average 0.910 0.837 0.984 0.959 0.984
84
Table 8. (continued)
60 Minute Period
(approximately 36 observations
CPMi = Bo +] CPMi=Bo+] CPMi=Bo+ | CPMi=Bo+ CPMi =
B1*Pi B1/Pi + Bo*Pi+B1/Pi+ | Bi*Ln(Pi) | e*Bo*(Pi*B1)
B2/Pi*2 B2/Pi*2 {log-log}
Subject 1 0.982 0.749 0.987 0.885 0.973
Subject 2 0.897 0.896 0.991 0.980 0.988
Subject 3 0.897 0.874 0.984 0.979 0.990
Subject 4 0.884 0.900 0.987 0.984 0.984
Subject 5 0.897 0.885 0.983 0.974 0.989
Average 0.911 0.861 0.986 0.960 0.985
By Day
(39 observations)
CPMi = Bo +} CPMi=Bo+ | CPMi=Bo+ | CPMi=Bo+ CPMi =
B1*day Bi/day+ |B1*day+B2/day| Bi*Ln(day) |e*Bo*(day*B1)
Ba/day*2 +B3/day*2 {log-log}
Subject 1 0.954 0.765 0.983 0.931 0.988
Subject 2 0.817 0.892 0.986 0.979 0.976
Subject 3 0.778 0.910 0.984 0.974 0.975
Subject 4 0.821 0.918 0.990 0.993 0.964
Subject 5 0.814 0.889 0.982 0.977 0.980
Average 0.837 0.875 0.985 0.971 0.977
Average over Subjects
CPMi = Bo + | CPMi = Bo+|] CPMi=Bo+ | CPMi=Bo+ CPMi =
Average R*2 B1*Pi B1/Pi + Bo*Pi+B1/Pi+ | B1*Ln(Pi) | e*Bo*(Pi*B1)
Be/Pi*2 B2/Pi*2 {log-log}
30 Min 0.908 0.759 0.979 0.947 0.983
40 Min 0.908 0.804 0.982 0.951 0.983
50 Min 0.910 0.837 0.984 0.959 0.984
60 Min 0.911 0.861 0.986 0.960 0.985
By Day 0.837 0.875 0.985 0.971 0.977
Avg of Avg 0.895 0.827 0.983 0.958 0.982
CPMi = Characters per Minute on the i-th time period
Bo, B1, B2, and B3 are fitted coefficients
Pi = Time Period
85
all the subjects are shown in Table 9. The average characters per minute typed
on every fifth day is matched with the subjective ratings. No further analysis
was performed on the subjective ratings.
4.6 Fastest Speed Attained by the Subjects.
For most of the experiment, the subjects were presented trials which
were composed of randomly ordered characters. However, during the final
session, subjects were presented trials which were composed of a string of six
characters repeated 84 times (see Section 3.3.3). In Table 10, the fastest typing
speed on the repetition trials is compared with the fastest speed attained on the
randomized trials. The mean of the fastest speed attained on the random trials
(170.1 CPM) was compared with the mean of the fastest speed on the repeated
character string trial (175.3) using a t-test; the difference between the two
means was not significant (p > 0.20).
4.7 Prediction of Learning with Less Than all Data Points.
The analysis to this point has focused on the prediction of learning for
individuals where all actual performance points were used in the regression. lt
is desirable to be able to predict an accurate performance curve of long term
learning using only a few initial performance points.
The coefficients of the equation CPM; = eBoT,B1 (1; = trials} were
determined for each subject when only the performance in either the first 5, 10,
15, 20, 25, 50, 75, 100, 125, 150, 175, and 200 trials was used in the
regression. Figure 30 shows the actual performance by trial and the predictions
when either 25, 50, 100, 150, 200, or all points were used in the regression for
86
Table 9. Ratings Ranging From 0 to 9 Given by the Subjects Each Week
Assessing Their Progress and Corresponding Performance.
Subject 1 | Subject2 | Subject3 | Subject4 | Subject 5 Avg. Avg. Avg. Avg. Avg.
Day |Rating) CPM |Rating] CPM |Rating] CPM |Rating! CPM |Rating; CPM
5 4 53.3 4 84.7| 4.5 | 96.8) 5.5 | 67.6 4 71.4
10 5 | 74.2} 5 |118.8! 6 |122.4 89.0) 5 |101.4
15 5.5 |89.5]} 5 |124.8] 6.5 |127.8 106.2} 4 | 96.9
20 6.75 ]113.8] 6 |136.3] 6.5 |143.8 118.3} 5 |112.1
25 8 |133.0) 6 |147.2] 7 |156.8] 8.3 |128.1] 5.5 [126.6
30 8.5 |144.9) 7 |162.9) 7.5 |167.5| 9 |130.6 119.9
35 8.8 1156.0} 7 |165.8] 7.5 |170.4 136.0 136.8
40 8.9 {170.9} 8 |174.8] 8 |173.1 9 |144.2; 7.5 1147.3
Rating ranges from 0 to 9
Avg. CPM = The average characters per minute typed over the
entire day
87
Table 10. Fastest Speed Attained on Randomized and Repetitive Trials.
Peak Keying Speed
(Characters per Minute)
Fastest attained in Fastest attained in
Randomized Trials Repetitive Trials
Subject 1 177.9 198.6
Subject 2 186.3 173.0
Subject 3 180.6 173.0
Subject 4 150.6 166.0
Subject 5 155.2 166.0
88
Sld S¢
Sid OS
Sid 00}
Sid OS|
Sld 00¢
Sid IIV
| joalqns ‘sjulod
OO Pur
‘OSI ‘OO!
‘OG ‘Gz
Isuy 84}
AjUO Bulsq
SuoIssesHay 607-607]
‘og asnbi4
009 00S
OOP
JOQUINN [eu L
00€ | |
00¢ 001
0c
OV
09
08
00L
Ocl
Ovl
091
O8l
00¢
alnul sed suejoeseyyD
89
subject 1. (Regressions with 5, 10, 15, 20, 75, 125, and 175 points were also
calculated, but are not shown due to space limitations.) To determine how
accurately each equation fits the actual data, the mean squared error (MSE)
was calculated for all regressions considered. The squared error (predicted
CPM minus actual CPM, squared) was calculated for all the trials that each
subject completed. All squared errors (for example 487 squared errors for
subject 1) were averaged to obtain the MSE. The smaller the MSE, the more
closely the equation predicts the actual performance over the entire range
considered. Figure 31 shows the mean squared error as a function of the
number of initial points used to calculate the regression for Subject 1. Figures
32, 34, 36, and 38 show the prediction equations for subjects 2, 3, 4, and 5
respectively. Figures 33, 35, 37, and 39 plot the mean squared error as a
function of initial points used to calculate the regression for subjects 2, 3, 4, and
5 respectively.
The mean squared errors are shown in Table 11 for all subjects and the
number of initial points used to fit the Log-Log equation. The data from Table
11 (MSE'’s for all subjects) are graphed in Figure 40.
It is possible that one of the other original functions considered (e.g.,
CPM; = Bo + ByLn[Tj]) would be effective in predicting final performance from
only a few initial points even though it did not provide good fits using all the data
points. For subject 2, the following three equations were also used to predict
performance when only the first 25, 50, 100, 150, and 200 data points were
included to calculate the coefficients.
CPM; = Bo + B4/T; + Bo/T 2
CPM; = Bo + ByT; + Bo/T; + Ba/Ti2
90
Mean Sq
uare
d Error
(for
all points)
1800
1600
1400
1200
1000
800
600
400
200 0 50 100 150 200 250 300 350 400 450 500
Number of Points Used in Regression
Figure 31. MSE for Log-Log Regressions Using Only a Few Initial
Points, Subject 1.
91
oOo wh
a & oO ”
2 8 a AN ~ oF 4
100
Pts
40 +
enul—; sed suajoeieyy
92
20
+ 10
0 15
0 20
0 25
0 30
0 35
0 40
0 45
0 50
0 55
0 600
650
50
Tria
l Number
Figure
32.
Log-
Log
Regressions
Usin
g only th
e first
25,
50,
100,
15
0, an
d 20
0 Points,
Subj
ect
2.
Mean Squared
Erro
r (for
all po
ints
)
1800
1600
1400
1200
oma
©
© ©
800
600
400
200
| | | | | | | | | I | | y l T T I I I } I } T t l
O 50 100 150 200 250 300 350 400 450 500 550 600 650
Number of Points Used in Regression
Figure 33. MSE for Log-Log Regressions Using Only a Few Initial Points, Subject 2
93
25
Pts
SS S— °
SS =<
| | | | | { | | I | T T 1 J I ] i
oO © oO Oo © oO oS oO oO eo ite) wt © © © wT NI Tr r r r” 20
0
ejNul ed SyajoOBIeUyD
94
600
500
400
300
200
100
Tria
l Number
Figure
34.
Log-Log
Regressions
Usin
g on
ly the
firs
t 25,
50,
100,
150, and
200
Poin
ts,
Subj
ect
3
Mean
Squared
Error
(for
all po
ints
)
1800
1600
1400
1200
1000
800
600
400
200
O 50 100 150 200 250 300 350 400 450 500 550 600 650
Number of Points Used in Regression
Figure 35. MSE for Log-Log Regressions Using Only a Few Initial
Points, Subject 3
95
py yalgngs ‘siUulod
OOZ PUB
‘OSI ‘ODL
‘OS ‘GZ
ISU aU
AjUO Bulsq
suoIssesBay 607-607
‘9¢ eunbi4
009 00¢
0Ol
Sid IIV
sJaulo Sld
SZ Sid
0S
00S oor
nos
|
; -
l
|
|
nl
Pr if
[ht
‘| (4 NM
Ae gr
0c
Ov
09
08
001
Ocl
Ovi
09}
08 |
00¢
a]null\ Jed suajoeeyy
96
Mean
Sq
uare
d Er
ror
(for
all points)
1800
1600
1400
1200
1000
800
600
400
200
|
0 50 100 150 200 250 300 350 400 450 500 550 600 650
Number of Points Used in Regression
Figure 37. MSE for Log-Log Regressions Using Only a Few Initial Points, Subject 4
97
Sld 0S
Sid S¢
SIBUIO
G algns ‘s}UlIOg
0OZ PU
‘OSI ‘OO!
‘OSG ‘Gz
Isuy @Y}
AjuUO Hulsq
SuoIssesBey 607-607
‘ge eun6i4
009 00S
007
JEquINN jeu L
00€ 002
001
0d
Ov
09
08
001
Ocl
Ovl
091
08 |
00¢
ainuly Jad syajoeieuy
98
Mean
Sq
uare
d Er
ror
(for
all points)
1800
1600
1400
1200
1000
800
600
400
200 h woo 0 jl Fl l l (°
T T T “T t i qT i I i T T 1
0 50 100 150 200 250 300 350 400 450 500 550 600 650
Number of Points Used in Regression
Figure 39. MSE for Log-Log Regressions Using Only a Few Initial
Points, Subject 5
99
Table 11. Mean Squared Error for Log-Log Functions Fit Using Only Some
of the Initial Performance Data Points.
Number of
Initial
Points Used Mean Squared Error
in Regression
Subject 1| Subject 2} Subject 3} Subject 4) Subject 5] Average
25 1471.4 857.3 125.9 88.7 66.1 521.9
50 1130.6 43.1 36.3 184.9 118.9 302.8
75 1147.2 49.4 34.8 105.0 38.1 274.9
100 639.9 80.3 34.8 41.7 38.4 167.0
125 515.4 75.2 37.0 35.3 38.2 140.2
150 337.4 57.1 35.1 42.8 40.0 102.5
175 224.9 42.9 35.6 42.8 38.3 76.9
200 133.0 49.3 36.6 47.1 38.5 60.9
487 43.9
509 19.6
N
516 38.2] “ 35.5
622 40.9
636 34.8
100
"soalqns ji'y 40}
uoissesHay ul
pesn S]UIOd
JO JOQUINN
JO UO}OUNY
eB Se 10
Pauenbs ues
‘Or aunbl4
uoissalBay ul
pasn siulod
jo Jaquinyy
0S9 009
OSS 00S
OSh OOF
OSE ODE
OSG2e 002
OSGt OOl
OS O
re | eT
SBE
=F
0 TW NY
\ +
+ 002
g alqng —_»
7 00%
p palqnsg
—_4
+ 009
€ pelqng —_»+—
7 008
Z palqns —___»
T OOOL
| jelqnsg
—__
+ O01
7 00}
1 0091
101
JO pesenbs ueayy
CPM; = Bo + ByLn(T})
Figure 41 shows the regression equations obtained for the function CPM; = Bo +
B1/Tj + Bo/Tj2; the mean squared error for each equation is graphed in Figure
42. The predictions from CPM; = Bo + By Tj + Bo/T; + B3/Ti2 are shown in Figure
43 and the mean squared error for each fit in Figure 44. Finally, the fit of CPM; =
Bo + ByLn(Tj) using only some of the initial data points is graphed in Figure 45
and the mean squared error in Figure 46. For subject 2, the mean squared
errors for the four equations considered are listed in Table 12 by number of
points used to fit the coefficients. It is clear from Table 12 and Figures 42, 44,
and 46 that only the Log-Log function can accurately predict the entire learning
curve and final performance from a few initial data points.
4.8 Chord Analyses.
Response times for each of the 18 chords were recorded for all trials for
every subject. The average response times for each chord by trial were plotted
for all chords for the first 487 trials -- see Appendix F. The first 487 trials are
used because these were completed by all five subjects.
An analysis of variance was performed on the chord data for the last 20
trials, i.e. the trials 468 to 487. Response time was modeled on a within-subject
design of trial and chord. The chord effect and trial X chord effect were both
significant at p < 0.0001. A post-hoc analysis (Student-Newman-Keuls test)
was performed on the chord effect to determine which chords were significantly
fastest. Table 13 lists each chord, its average response time, and its relative
statistical significance. No post-hoc analysis was performed on the trial X chord
data, owing to the large number of comparisons (360).
102
‘SJUlOd 00Z
PUL ‘OSI
‘OOL‘OS ‘Sz
IS4 Oy
AlUO Bus;
suo|ssesBey ZvLL/ZatiL/La+og
‘z yelqns “Lp
aunbiy
JOQUINN [eu
0S9 009
OSG 00S
OSr O0h
OSE O0€
OS¢ 006
OSt OO
0S 0
L |
| t
4 —F
| |
| |
0
| 02
| | OV
09 j
} Sld
SZ fl y
+
08
“™ y
| OO}
Sid 001
pe
Sid OSI
Choc
Sid 002
| i
(Md if +
Od
Sid IIV po
Ay |
IN +
Ol
cal
i" im +
O8t
— 00¢
ejnuly sed suygjoeieyy
103
Mean
Sq
uare
d Er
ror
(for
all
points)
5000
4500
4000
3500
3000
2500
2000
1500
1000
500 | | | | | | | | | | | | | y LJ T 1 i T q 7 i r I 1
50 100 150 200 250 300 350 400 450 500 550 600 650
Number of Points Used to Predict
Figure 42. Subject 2, MSE for Bo+B1/Ti+B2/Ti*2 Regressions
Using Only a Few Initial Points.
104
"SIUIOd OO
PUB ‘OSL
‘OO! ‘OS
‘Gz ISI4
OY} AjUG
Buisp suoissalBey
?viL/eg+l/2gtiL, Ldt+og
‘z yelqns ‘ep
eunbiy JOquINN
jeuL
0S9 009
OSS 00S
OSt 00h
OSE O0€
OS2 002
OSI O01
OS 0
|
l |
l I
—T 0 08
001
Oc |
e]null Jed suejOeBeyy
Ovt raat
IK +
O91
7 O8t
Wf
Std IV
Sld 002
Sid 0S1 Sld 001
“Sid Sz Sid 0
~~ 002
105
5000 +
4000 +
® oO Oo
© | !
(for
all points)
Mena
Sq
uare
d Er
ror
N
© © Oo
1000 + 0 { + {— t——1—
O 50 100 150 200 250 300 350 400 450 500 550 600 650
Number of Points Used to Predict
Figure 44. Subject 2, MSE for Bo+B1*Ti+B2/Ti+B3/Ti*2 Using
Only a Few Initial Points.
106
SJUIOd OOZ
PUL ‘OG|
‘OOL ‘OG
‘Gz ISUIY
AY} AjUD
Buisq suoissasBay (11)u7,
La+og ‘2 oalqns
“Gp eunbi4
JOqUINN [eu
0S9 009
OSS 00S
OSr O0v
OSE OOF
OG2 O00¢
OSt OO
OS 0
I- |
4 |
7
| M iy
i
a
T
a
Sid SZ
Ls
i py
Sld 0S ant
me
¥ in
L Sid
0S4‘00L weer
——
gn
Sld 002
7
Sid IV \ My i pica
i. Hi
Oc
OV
09
08
001
Oc |
Ori
091
O81
00¢
alnull\ Jed suajoeIeyD
107
Mean
Sq
uare
d Er
ror
2000
1800
1600
1400
1200
800 (for
all
points)
° ©
©
600
400
200 0 50 100 150 200 250 300 350 400 450 500 550 600 650
Number of Points Used to Predict
Figure 46. Subject 2, MSE for Bo+B1*Ln(Ti) Regressions Using
Only a Few Initial Points.
108
Table 12. Mean Squared Error for Four Equations Fit Using Only Some of the Initial Performance Data Points. For Subject 2 Only.
Mean Squared Error for
Number of Points Used in the Regression
25 50 100 150 200 All
CPMi=Bo+B1*Ti+ 21860.1} 71309.7| 14519.7) 2904.1) 1284.4 68.8 B2/Ti+B3/Ti*2
CPMi=Bo+B1/Ti+ 4815.6] 3056.1) 1580.5] 1376.6] 1063.7) 405.5 B2/Ti*2
CPMi=Bo+B1*In(Ti) 1989.3 547.8 170.9) 174.1] 117.8 44.1
CPMi=(e*Bo)*(Ti*B1) 875.3 43.1 80.3 57.3 49.3 40.9 CPMi = Characters per Minute
Bo, B1, B2, and B3 are fitted coeffiecients
Ti = Trials
109
Table 13. Chord Response Times (in ms) Averaged Over Trials 468 to 487,
the Last 20 Trials Performed by All Subjects. (Means with the same letters are not significantly different)
Chord Mean Std. Dev. SNK Grouping
13 325.0 139.7 A
24 329.0 125.7 A B
14 350.5 144.5 A B C
36 359.2 138.0 A B C
35 361.8 148.4 A B C
28 364.6 150.3 A B C
26 368.3 148.3 A B CC OD
45 368.9 145.8 A B C OD
46 371.0 160.6 A B Cc OD
68 371.4 176.2 A B CC OD
23 372.9 140.7 B cC OD
15 375.5 168.9 B Cc OD
17 378.5 160.0 Cc OD
57 380.2 171.4 Cc oD
16 381.0 168.1 C D
25 399.4 177.7 Cc D
37 413.4 167.2 D
48 414.3 175.3 D
110
4.9 Analysis of Average Performance of All Subjects.
To this point the analysis was focused on predicting performance of
individual subjects. It is also of interest to evaluate the ability of the functions to
predict the combined subjects’ data as well as the average typing speed.
4.9.1 Prediction of All Subjects Performance.
Using regressions to predict combined data of the subjects differs from
predicting average performance; each trial has five values (one from each
subject) whereas an average prediction has only one point for each trial. The
five functions considered for individual data were also used to predict all
subjects’ data:
CPM; = Bo + By Tj (Equation 1)
CPM; = Bo + By/T; + Bo/T)2 (Equation 2)
CPM; = Bo + Bi7; + Bo/T; + Ba/Ti2 (Equation 3)
CPM; = Bo + ByLn(T}) (Equation 4)
CPM; = eBo 7;B: (Equation 5)
These equations were only fit on the first 487 trials - those that included all
subjects. The R2 values for each function are listed in Table 14. The following
Log-Log equation provided the best fit (R2 = 0.843):
CPM; = 33.115*T,9-291
The actual data points of all subjects and the plotted equation are shown in
Figure 47.
Table 14. Summary of R-Squared Values for All Functions Considered for Average of the Subjects and All Subjects’ Performance
R-Squared Value
Equation All Subjects |Avg. of Subjects
CPMi=Bo+B1*Ti 0.747 0.883
CPMi=Bo+B1/Ti+B2*(1/Ti*2) 0.351 0.457
CPMi=Bo+B1*Ti+B2*(1/Ti)+B3*(1/Ti*2) 0.799 0.951
CPMi=Bo+B1*Ln(Ti) 0.791 0.956
CPMi=Exp(Bo)*(Ti*B1) 0.843 0.991
CPMi = Characters per Minute on the i-th Trial
Ti = Trial
Bo, B1, B2, and B3 are fitted coefficients
112
"eleq ,sjoalans
yy ‘souewJOLeg
jenjoy pue
uoioipasg 607-607
‘zp aunbi4
JOQqUNN JeUL
0S9 009
OSG 00S
OSb O00v
OSE O0€&
O0S¢ 000
OSt O01
| |
| |
| __|
| |
| |
| I
po 1
1 1
| |
1 I
\ |
| J
001
Ocl
Ov}
091
08}
- 002
(GZ OvIeWL).GLEES
=Wdd
e]nuIW sed srejoeseyy
113
4.9.2 Prediction of Average Performance.
The average typing speed and its standard deviation were calculated for
the trials completed by all five subjects (up to trial 487). The same five
equations used to predict individual performance were considered to fit average
performance. Table 14 lists the R@ values for each function. The following Log-
Log equation provided the best fit of the data (R2 = 0.991):
CPM; = 34.467+T,0-244
Figure 48 shows the average characters per minute typed, the prediction of the
average using the above equation, and the average plus and minus one
standard deviations.
114
“sjoalqns jy
jo abeJaay
aul jo
uOIeIANG
DIJEPURIS |
- pue
+ puke
‘BOUBWUOLAY abeieAy
jenjoy ‘uoHoIpaig
607-607 ‘gp
ainbi4
0S9 009
OSS 00S
A8q PIS
I -
ABQ PIS
I+
JOQUINN [eu L
OSp O0b
OSE O0€&
OSe O02
OSt OOF
OS
| |
| J
| |
| i
| |
I T
t I
I t
t t
(ppz OvieUl),Z9r
ve = NdO
0g
OV
09
08
00}
Oct
eynui Jed suajoBseYD
Ovl
091
08}
00¢
115
5. DISCUSSION
The goals of the experiment were to find answers to questions that have
not been fully addressed in the literature. This study provided information on
what function best predicts learning, the best predictors of performance, and
criterion of peak keying speed. The study did not address all the questions
raised in the literature review; for example, the relationship of character set size
versus time to maximal speed is important, but was not investigated.
5.1 Speed - Accuracy Performance.
Providing feedback to the subjects at the end of each trial seemed to be
very effective in controlling speed and accuracy tradeoffs. All the subjects had
an average accuracy greater than 97% over the entire experiment. In addition,
the subjects generally set goals for themselves such as finish a trial in less than
five minutes or reach 100 characters per minute. By providing feedback of
performance, there seemed to be natural performance goals throughout the
experiment for which the subjects aimed.
The subjects probably did not make adjustments in their speed and
accuracy criteria after every trial, but more likely made adjustments after
perceiving a trend in either speed or accuracy performance. For example, if a
subject's accuracy level for a particular trial fell below 97%, that person
probably did not immediately slow down on the next trial. However, if accuracy
performance on many consecutive trials was below 97%, the subject probably
maintained speed but tried to increase accuracy. Overall, the method of
providing immediate feedback to the subjects on speed and accuracy appeared
116
effective in motivating the subjects and controlling some of the error associated
with speed accuracy tradeoffs.
5.2 Using Trials to Predict Performance.
One goal of this study was to find an equation which described the
learning or performance progress on a keyboarding task. The following five
candidate functions were fit to individual performance curves.
CPM; = Bo + ByT; (Equation 1)
CPM, = Bo + By/T; + Bo/T 2 (Equation 2)
CPM, = Bg + ByT; + Bo/T; + B3/T;2 (Equation 3)
CPM; = Bo + ByLn(T}) (Equation 4)
CPM; = eBo T;81 (Equation 5)
where CPM, is the characters per minute typed on the i-th trial
(dependent variable), Tj is the i-th trial, and Bo and B, are fitted
coefficients.
Only equations that predicted performance in a monotonically increasing
fashion were considered. Actual performance on a keyboarding task is not
always increasing with each additional trial of practice; however, there is a
general trend of improvement with practice in learning a skill. A Log-Log
equation of the form CPN; = eBo TB 1 (where CPM, is performance on the i-th
trial or hour, Tj is the i-th trial or hour of practice, and Bo and By, are fitted
coefficients) described each subject’s performance most accurately. The Log-
Log equation had an average R2 of 0.956 when using trials as the regressor
variable. An important characteristic of this type of equation is that it does not
117
reach an asymptote, and therefore does not predict a maximal keying speed.
The only equation considered which predicts an asymptote (CPM; = Bo + Bi/Tj +
B2/T;2) did not even account for 50% of the variance (average R@ = 0.431).
Overall, the Log-Log function provided the best description of subject
performance on the keyboarding task up to the first 60 hours of practice when
trials are the predictor variables. The ability of the equation to predict beyond
that point is unknown because there are no actual! performance data for
comparison with the predictions. However, from Figures 24 to 28 it appears that
the subjects did not reach peak performance (i.e., a stable, non-increasing
value) in this experiment.
5.2.1 Coefficients of the Log-Log Equation.
It appears that the shapes of the Log-Log curves (performance
predictions) for subjects 2 to 5 are very similar, as shown previously in Figure
29. However, the performance prediction of subject 1 varies significantly from
the predictions for the other four subjects. The predictions for subjects 2 and 3
are nearly identical, as are the predictions for subjects 4 and 5. Subjects’ 2 and
3 performance predictions seem to differ from subjects 4 and 5 by only a
constant. It was hypothesized that one or both of the coefficients for subject 1
were significantly different from the other subjects causing the different shape of
the prediction curve. However, after using a t-test to compare the values of the
coefficients, it was found that all the Bo coefficients were significantly different
from each other, and all B; comparisons except one were significantly different,
as shown previously in Table 6. The reason for so many significant differences
118
in the coefficients is that the t-tests were so powerful because of the large
number of observations for each subject.
5.2.2 Prediction of Peak Keying Speed.
While the Log-Log function predicts the actual progress of the subjects
very well, by its nature it does not approach a peak value asymptote. Four
techniques can be used to estimate each subject’s peak performance. First, a
large number of trials chosen to represent the upper limit of practice can be
entered into the Log-Log equation. For example, 5,000 trials might be assumed
to be a long enough training period for a person to reach a peak keying speed.
For subject 3, this leads to a speed of 265 characters per minute (CPMsooo =
48.231+50009-199 = 265).
The second method uses the slope of the Log-Log function to determine
peak keying speed. Although the Log-Log function never approaches an
asymptote, its slope decreases as trial numbers increase. Slope is equal to the
first derivative of the performance function, CPM; = 48.231*T,9-199. The first
derivative of this equation is as follows:
d(CPM) / dT; [slope] = (0.199)(48.231)*T;0-801 = 9.6 «7,;0-801
Thus, a given slope could be associated with top speed, the trial number
determined when this slope occurs, and finally that trial number could be used
to calculate maximal performance. For example, it may be assumed that when
the slope decreases to 0.025 CPM/Trial, a person has effectively reached peak
performance. For subject 3, a slope of 0.025 CPM/Trial occurs at trial 1684
(solving for T; when 0.025 = 9.6 «7, 9-891) at trial 1684, the Log-Log equation
119
predicts a speed of 212 characters per minute (CPMy¢94 = 48.231*16849-199 —
212). |
The third technique to predict top speed involves using the function CPM;
= Bo + B,/T; + Bo/T|2 as suggested by Bittner (1991). He assumed that the
equation Y; = Bo + By/T; + B2/T;2 can predict peak performance, where Y; is the i-
th trial performance, Tj is the i-th trial, and Bo, By, and Bo are fitted coefficients.
As the number of trials increases the magnitude of the last two terms decreases
very rapidly ( B,/T; and B2/T;2) and Bg estimates peak performance; it is the only
function of all functions considered that will reach an asymptote. However, this
formula does not predict the performance progress of the subjects very well (all
R2 are less than 0.5 as shown in Table 4); thus, Bp can not be considered an
accurate prediction of the fastest keying speed.
The function appears not to predict final performance well because of the
steepness of the initial learning curve; thus, it underestimates peak
performance, see Figure 21 for subject 2. If several of the early points are
deleted from the regression, this function might predict a more accurate
asymptote. As an example, the formula CPM; = Bo + B;/T; + Bo/T;2 was fit to the
data of subject 3 ignoring the first 100 data points. It appears that most of the
steep portion of the learning curve is eliminated by ignoring the first 100 points,
see Figure 49. The fit of the remaining 536 points yields the equation CPM; =
195.7 - 16,199(1/T;) + 919,625(1/T;2) which has an R2 of 0.863 and predicts a
peak performance of 195.7 characters per minute.
The fourth technique used to predict peak performance employs only the
final trials. These trials were composed of a string of six characters repeated 84
times (see Section 3.3.3). It was felt that if keying the string became a highly
120
'E ANS ‘SIEUL
OOL ISUl4
SUL INOYWM
‘ZvIL/ZAtLL/ +a
= Wd
uby ay}
Guisq peeds
yead jo
ayewisy “Gy
aunb4 JOQUINN
jeu
069 009
OSS 00S
OSr O0r
ose o00€
ose 002
Ost ool
og 0
| I
} |
{ |
| |
| |
| |
| I
I {
| {
t 1
1 I
] I
| i
0
+ OV
I 09
+ O01
+ Oct
ainulyy Jad suajoeseuy
ry +
OVI Ph
TY
Nf yr
oF Wi
+
O91
+ O8t
— 00¢
ovil/ed+ W/lG@+@
--------- enjoy
121
automated motor program (with minimal response times), the performance on
these trials would represent peak keying speed. However, two of the subjects
typed faster in the random trials than in the repeated character string trial, as
shown previously in Table 10. In addition, the means of the fastest speeds
attained on randomized and repetitive trials were not significantly different
which indicates that the repetitive trials, in fact, were not a good indication of
peak keying performance.
Of the four techniques, using the equation CPM; = Bo + By/T; + Bo/Ti2
(without initial trials) might provide the best method for predicting peak
performance. This equation provides a much better predictor of actual
performance if initial performance points are ignored; the R2 values and
predicted peak performance for each subject are shown in Table 15. This
equation, without the 100 initial trials, predicts peak (asymptotic) performance
which was not attained by any of the subjects on the randomized trials
(comparing Table 15 with previously shown Table 10). The major difficulty with
this approach is to determine how many of the original points should be left out
of the regression.
Assuming a large amount of practice time, assuming a minimal slope, or
using the repeated string trials do not seem to be suitable methods to use to
predict peak performance. By what criteria could the number of trials be chosen
and considered the upper limit of practice; should it be 2000, 5000, 10000, or
more trials? The same dilemma arises when trying to assume a minimal slope
as the stopping point of improvement.
122
Table 15. Values of R2 and Peak Performance when Fitting the Equation CPM;
= Bo + By/T; + Bo/Ti2 Without the First 100 Performance Points.
Subject R2 Predicted Peak (in CPM)
1 0.940 196.8
2 0.854 197.7
3 0.863 195.7
4 0.870 158.8
5 0.759 161.1
where CPM j is the characters per minute typed on the i-th trial
(dependent variable) and Tjis the i-th trial of practice
123
5.2.3 Combin nd Aver Performan f All th
The Log-Log equation fit the combined and average performance of the
subjects best. It was expected to fit best because it was the best at predicting
each individual’s performance. The equation fit to the combined subjects’ data
probably had a high R@ (0.843) because there were not more than five subjects’
data. Kroemer (1990) and McMulkin and Kroemer (1991) did not find an
equation which accurately fit combined subjects’ data. This could be due either
to the larger number of subjects (10 to 12) or the shorter training periods (3 to
12 hours).
5.3 Predicting Performance from Time
The average characters per minute entered by the subjects were
calculated for 30, 40, 50, and 60 minute time periods and for sessions.
Because the time periods were composed of entire trials, the average lengths of
the 30, 40, 50, and 60 minute time periods were closer to 32, 42, 52, and 62
minutes respectively, shown previously in Table 7. It is not so important that the
length of the time periods are factors of 10, but that the average length of the
time increments is close to equal for all subjects. Inspection of Table 7 reveals
that for all four time periods each subject had approximately the same average
amount of practice.
After establishing the performance of the subjects over the time intervals,
the same five equations used to predict performance from trials were fit to the
time period data. The equations CPM; = Bo + ByP; + Bo/P; + B3/Pi2 and CPM; =
eBo p,B1 (a Log-Log equation) provided the best fits. P; is equal to 1, 2, 3,.... not
124
30 minutes, 60 minutes, and so on. The general form of the Log-Log equation
(CPM; = eBo P;B1) provided good fits for both repetition and time data.
It does not appear that changing the length of time increments has much
effect on the ability to predict performance. Five different time increments were
used as regressor variables: 30 minutes, 40 minutes, 50 minutes, 60 minutes,
and total sessions (1.5 hours, not all this time was spent practicing). For the
Log-Log function, the largest difference in R2 values was only 0.008 (Table 8).
5.4 Subjective Ratings.
The subjective ratings could not be used in any analysis to predict
performance of the subjects. The subjects probably found it difficult to rate their
performance early in the experiment because they did not know what their final
performance would be. At the beginning of the experiment, subjects might have
simply chosen a middle rating and then gradually increased it until the end of
the experiment. For this reason, the ratings that the subjects provided are not of
an interval nature, but are instead at best an ordinal measure. The data of
subjects four and five illustrate some of the problems with the subjective rating
procedure, shown previously in Table 9. Subject four rated performance at a
nine by day 30 but was clearly still improving over the following two weeks.
Subject five rated performance at a four on day 5, then a five on day 10, and
then back to a four on day 15. This subject probably felt the ratings of the initial
two weeks were too high and then readjusted the scale for week three and
started the rating at four again. Overall, little information could be extracted from
the results of the subjective ratings.
125
5.5 Evaluation of Criteria to Predict Performance.
Part of the second goal of the experiment was to determine which
criterion of the following three criteria could be used to predict keying speed:
number of trials/repetitions, hours (or other time increments) of practice, and
subjective opinion. The Log-Log equation, CPMj = eBo T;B1, provided the most
accurate prediction of performance when considering two regressor variables,
time and repetition. The average R2 values using the Log-Log function for
these two criteria were as follows:
Trials (504 repetitions per trial) R2 = 0.956
Time of Practice (average for all increments) R2 = 0.982
This gives further credence to the assertion that the Log-Log function describes
the underlying progress of performance. It does not appear to matter much
whether repetition or time of practice is used to predict performance. Although
the equation CPM; = Bo + ByPj + Bo/P; + B3/P2 is effective when time increments
constitute the regressor variable, it did not predict performance as well as the
Log-Log equation when trials constitute the predictor variable. This indicates
that the equation to describe performance remains the same when the predictor
variable changes from trials to time. These findings contrast those of Thurstone
(1919); he asserted that the function changed when time instead of repetition
was used to describe performance progress. The Log-Log function continued
to show flexibility with different levels of the time predictor. The fits of the Log-
Log equation had fairly constant R@ values across the five time increments
considered.
126
Subjective opinion data, as elicited in this study, could not be used to
predict performance due to the ordinal nature of the ratings. See Section 5.4 for
further discussion.
5.6 The Concept of Peak Performance.
It is apparent from Figures 24 to 28 that the actual performance of the
subjects did not approach an asymptote which could be considered the peak
performance. The Log-Log equation, which provided a very good description of
the actual performance does not predict an asymptote; it presumes continual
improvement in performance which, however, becomes increasingly slow and
gradual.
Both the actual performance observed and the prediction of the Log-Log
equation seem to contradict the traditional presumption of learning - a slow
increase in performance until gradually a upper limit is reached that can not be
exceeded. It might be that there is no limit to performance improvement.
The response times of the keyboarding task of this study can be broken
down into their components. First, the stimulus must be sensed by the subject,
for example the number 8 is to be typed. Second, the sensed information must
be transmitted through the (afferent) peripheral nervous system to the brain.
Third, a decision must be made in the brain as to what response to make to the
incoming stimulus. Fourth, the response signal must be transmitted along the
(efferent) peripheral nervous system to muscles. Fifth, the ensuing movement of
the fingers determines the "response time" and a character is typed. The third
and fifth steps of the response are probably becoming faster with practice with
most of the improvement likely to occur in the third step, the decision making
127
process. There may be a limitation to how quickly response decisions can
become. However, that minimal decision time may never actually be reached.
5.7 Prediction of an Entire Curve from the Initial Data Points.
From a practical standpoint, to predict the performance of subjects, it is
costly to gather over 500 data points. It is, therefore, of great interest to find out
how many initial performance points are needed before an accurate learning
curve can be predicted.
For subject 2, four different equations were fit to the actual data when
only 25, 50, 100, 150, and 200 of the initial performance points were used in the
regression. The mean squared error between the actual and predicted
performance for all 20 cases (4 equations and 5 quantities of initial number of
points used in the regression) was shown previously in Table 12. By perusing
the information in Table 12, it becomes clear that the only function capable of
accurately predicting the entire learning curve from the initial data points is a
Log-Log function.
The remaining functions do not closely approximate learning or final
performance. The Log-Log equation has a mean squared error which is at least
an order of magnitude smaller that the other three functions when only 25, 50,
100, 150, or 200 initial data points are used in the regressions. Inspection of
previously shown Figures 32, 41, 43 and 45 clearly indicates that only the Log-
Log function predicts final performance effectively from regressions including
only a few initial performance points. The other functions either predict too high
a final performance (CPMj = Bo + By,Tj + Bo/T; + B3/Ti2) or too low a final
performance (CPM, = Bo + By/T; + Bo/Tj2 and CPM; = Bo + ByLn[TjJ).
128
It is difficult from a small sample size of five subjects to draw a definite
conclusion about how many points are sufficient to accurately predict the entire
performance curve using the Log-Log equation. The regression using all points
is assumed to have the lowest mean squared error. Thus, the MSE of
regressions using less points should be compared to the MSE of the all points
regressions to determine its relative accuracy. For four of the subjects (2
through 5), the MSE is drastically reduced when the number of points used
reaches 50, see Figure 40 and Table 11 shown previously. The MSE resulting
from including more than 50 points is a fairly flat function for these four subjects.
In contrast, subject 1 exhibited a much more modest decrease in MSE as the
number of points included in the regression increased.
Overall, the Log-Log function proved to be the only equation able to
predict the entire learning curve from a limited number of initial data points.
Due to the small number of subjects, it is difficult to make assertions about the
number of points needed with the Log-Log equation to capture the entire
performance progress with a reasonably small mean squared error: as few as
50 initial points were needed for four of the five subjects (out of about 550 total
points). However, when applying these findings, the amount of uncertainty that
is acceptable needs to be determined which then points to the number of initial
performance points needed.
5.8 Chord Analyses.
The average response times for each chord were calculated for the last
20 trials including all subjects, i.e., trials 468 to 487. Table 13, shown
previously, lists the chords’ average response times and relative ranking. The
129
major result of the chord analysis is that after over 400 trials of practice there are
still chords that are significantly faster than others. However, significant
differences found could be due to the large power of the tests. Each chord
average response time consisted of 2800 observations (28 chords per trial X 20
trials X 5 subjects).
The chord analysis was based on practice due to repetition not amount of
time because response times from trials 468 to 487 were used. For subject 1,
these trials were the last ones completed, while for subjects 2 to 5 they were
completed before the end of the experiment. The number of final trials to
include in the analysis was limited to 20 for two reasons. First, chord
performance was of interest after a large amount of practice. It was of interest to
find if differential chord performance existed after a large amount of practice.
Second, the last 20 trials include enough observations to obtain a
representative average and standard deviation. It is possible that fewer number
of final trials could be used for the analysis.
5.9 Further Research.
The results of this study point to several areas for future research and
analysis. First, the trials were held constant at 504 characters. It would be
interesting to investigate the effect of different trial lengths on the ability of the
Log-Log function to predict performance. In the same way that variations of time
increments were used with the Log-Log equation, number of repetitions per trial
could be varied and accuracy of the Log-Log function analyzed.
Second, the performance points seem to have some autocorrelation.
Performances of the subjects “bounced” about the Log-Log curve. Apparently,
130
series of points alternated between being higher and lower than the previous
point. It might be possible to include an autocorrelation factor in the model of
performance and then account for more of the variation.
Third, as mentioned in the literature review, there appears to be a
relationship between number of characters that are learned and the steepness
of performance progress. Research is needed on how the values of the
coefficients in the Log-Log equation change with character set size. For
example, faster keying speeds might have been obtained more quickly if this
study had used 10 characters instead of 18.
Fourth, the subjective opinions were that were obtained in this study
could not be correlated with performance. Perhaps other techniques of rating
could be used to find a relationship between performance and subjective
opinion about that performance.
131
6. REFERENCES
Alden, D.G., Daniels, R.W., and Kanarick, A.F. (1972). Keyboard design and operation: a review of the major issues. Human Factors, 14(4), 275-293.
Baddeley, A.D, and Longman, D.J.A. (1978). The influence of length and frequency of training session on the rate of learning to type. Ergonomics, 21(8), 627-635.
Bartram, D., and Feggou, O. (1985). An evaluation of four different keyboard designs for foreign destination coding desks. Report ERG/Z6545/85/4c. Hull: University of Hull, Ergonomics Research Group.
Bequaert, F.C., and Rochester, N. (1977). Teaching typing on a chord keyboard. IBM Report TR 00.2918. Cambridge, MA: SPD - Cambridge Systems Group.
Bittner, A.C. (1991). Analysis of learning-curve asymptotes: ergonomic applications. In W. Karwowski and J.W. Yates (eds.), Advances in industrial ergonomics and safety Ill. London, England: Taylor and Francis Ltd.
Bowen, H.M., and Guinness, G.V. (1965). Preliminary experiments on keyboard design for semiautomatic mail sorting. Journal of Applied Psychology, 49(3), 194-198.
Callaghan, T.F. (1989). The utility of a technique for testing the difference in ease of chords on the ternary chord keyboard. Unpublished masters thesis, Virginia Polytechnic Institute and State University, Blacksburg, VA.
Chapman, J.C. (1919). The learning curve in typewriting. Journal of Applied Psychology, 3, 252-268.
Conrad R., and Longman, D.J.A. (1965). Standard typewriter versus chord keyboard - an experimental comparison. Ergonomics, 8, 77-88.
Cornog, J.R., Hockman, J.F., and Craig, J.C. (1963). Address encoding - a study of the double-binary keyboard as a link in the machine-sorting of mail (ASME Report Number 63-WA-338). New York, NY: American Society of Mechanical Engineers.
Dvorak, A. (1943). There is a better typewriter keyboard. National Business Education Quarterly, 12(2), 51-66.
Dvorak, A., Merrick, N.L., Dealey, W.L., and Ford, G.C. (1936). Typewriting behavior. New York, NY: American Book Company.
132
Evey, R.J. (1980). How typists type. IBM Report Number HFC-35. San Jose, CA: IBM General Products Division.
Fathallah, F.A. (1988). An experimental comparison of a ternary chord keyboard with the QWERTY keyboard. Unpublished masters thesis, Virginia
Polytechnic Institute and State University, Blacksburg, VA.
Gentner, D.R. (1983). The acquisition of typewriting skill. Acta Psychologica, 54, 233-248.
Glencross, D., and Bluhm, N. (1986). Intensive computer keyboard training programmes. Applied Ergonomics, 17(3), 191-194.
Goldstein, |.L. (1986). Training in organizations: needs assessment, development, and evaluation. Pacific Grove, CA: Brooks/Cole.
Gopher, D. (1986). Experiments with a two hand chord keyboard - the structure and acquisition process of a complex transcription skill. Report Number HEIS-86-5. Arlington, VA: Office of Naval Research, Personne! and Training Research Programs.
Gopher, D. and Eilam, Z. (1979). Development of the letter-shape keyboard: a new approach to the design of data entry devices. /n Proceedings of the Human Factors Society 23rd Annual Meeting (pp.40-44). Santa Monica, CA: Human Factors Society.
Gopher, D., Hilsernath, H., and Raij, D. (1985). Steps in the development of a new data entry device based upon two hand chord keyboard. /n Proceedings of the Human Factors Society 29th Annual Meeting (pp.132-136). Santa Monica, CA: Human Factors Society.
Gopher, D. and Raij, D. (1988). Typing with a two-hand chord keyboard: will the QWERTY become obsolete. /EEE Transactions on Systems, man, and cybernetics, 18(4), 601-609.
Green, L. (1940). A study of typewriting achievement in three high schools. Journal of Educational Research, 34(3), 209-217.
Lee, S.S., and Kroemer, K.H.E. (1991). Report on subjectively preferred one- hand chords on TCK #2. Blacksburg, VA: Industrial Ergonomics Laboratory, Virginia Polytechnic Institute and State University.
Leonard, J.A., and Newman, R.C. (1965). On the acquisition and maintenance of high speed and high accuracy in a keyboard task. Ergonomics, 8, 281- 304.
133
Lockhead, G.R., and Klemmer, E.T. (1959). An evaluation of an 8-key word- writing typewriter. IBM Report Number RC-180. Yorktown Heights, NY: IBM Research Center.
Killeen, P.R. (1978). Stability Criteria. Journal of the Experimental Analysis of Behavior, 19, 17-25.
Kinkead, R. (1975). Typing speed, keying rated, and optimal keyboard layouts. In Proceedings of the Human Factors Society 19th Annual Meeting (pp.159-161). Santa Monica, CA: Human Factors Society.
Kirschenbaum, A., Friedman, Z., and Melnik, A. (1986). Performance of disabled persons on a chordic keyboard. Human Factors, 28(2), 187-194.
Klemmer, E.T. (1958). A ten-key typewriter. IBM Report Number RC-65. Yorktown Heights, NY: IBM Research Center.
Kroemer, K.H.E. (1990). Final report on the experiments performed on VATELL Corporation's ternary chord keyboard research project 43716 (230-1 1- 110F-106-801189-1). Blacksburg, VA: Industrial Ergonomics Laboratory, Virginia Polytechnic Institute and State University.
McMulkin, M.L., and Kroemer, K.H.E. (1991). Report on the experiments performed on one-hand TCK #2. Blacksburg, VA: Industrial Ergonomics Laboratory, Virginia Polytechnic Institute and State University.
Morgan, A.S., Drake, J.B., Heck, S.K., and Long, C.H. (1981). Experimental study of effects on speed and accuracy of teaching alternate-hand method on a ten-key electronic calculator. Perceptual and Motor Skills, 52, 695-700.
Mussin, E.H. (1980). The microwriter. In Excerpt from JERE Electronic Office Conference Proceeding, (129-134). London, UK: Institute of Electronic and Radio Engineers.
Neter, J., and Wasserman, W. (1974). Applied Linear Statistical Models. Homewood, IL: Irwin Inc.
Norman, D.A., and Fisher, D. (1982). Why alphabetic keyboards are not easy to use: keyboard layout doesn't much matter. Human Factors, 24(5), 509- 519.
Noyes, J. (1983a). Chord keyboards. Applied Ergonomics, 14(1), 55-59.
Noyes, J. (1983b). The QWERTY keyboard: a review. International Journal of Man-Machine Studies, 18, 265-281.
134
Ratz, H.C., and Ritchie, D.L. (1961). Operator performance on a chord keyboard. Journal of Applied Psychology, 45(5), 303-308.
Richardson, R.M.M., Telson, R. U., Koch, C.G., and Chrysler, $.T. (1987). Evaluation of conventional, serial, and chord keyboard options for mail encoding. In Proceedings of the Human Factors Society 31st Annual Meeting (pp.911-915). Santa Monica, CA: Human Factors Society.
Rochester, N., Bequaert, F.C., and Sharp, E.M. (1978). The chord keyboard. Computer, 11(12), 57-63.
Salthouse, T.A. (1984). The skill of typing. Scientific American, 250(2), 128-135.
Salthouse, T.A. (1986). Effects of practice on a typing-like keying task. Acta Psychologica, 62, 189-198.
SAS Institute Inc. (1985). SAS user's guide: statistics. Cary, NC: SAS Institute Inc.
Seibel, R. (1962). Performance on a five-finger chord keyboard. Journal of Applied Psychology, 46(3), 165-169.
Seibel, R. (1964). Levels of practice, learning curves, and system values for human performance on complex tasks. Human Factors, 6, 293-298.
Seibel, R. (1972). Data entry devices and procedures. In Van Cott, H., and
Kinkade, R. (Eds.), Human engineering guide to equipment design (pp.311-345). Washington D.C.: American Institute for Research.
Showel, M. (1974). A comparison of alternative media for teaching beginning typists. The Journal of Educational Research, 67(6), 279-285.
Sidorsky, R.C. (1974). ALPHA-DOT: a new approach to direct computer entry of battlefield data (Technical Report Number 249). Arlington, VA: U.S. Army Research Institute for Behavioral and Social Sciences.
Strong, E.P. (1956) A comparative experiment in simplified keyboard retraining and standard keyboard supplementary training. Washington D.C.: General Services Administration.
Thurstone, L.L. (1919). The learning equation. Psychological Monographs, 26, 1-52.
Towill, D.R. (1976). Transfer functions and learning curves. Ergonomics, 19(5), 623-638.
135
Towill, D.R. (1982). How complex a learning curve model need we use? The Radio and Electronic Engineer, 52(7), 331-338.
West. L.J. (1969). Acquisition of typewriting skills. Belmont, CA: Pitman Publishing.
136
Appendix A
Example of a Test Trial
137
Q/728+*0474*-S/S9 . 88¢*08 ¢.¢ -876/8148T 3T118S¢7244S04+467-T7742
1609/T931677*+/29SS .12T+88S¢286877¢++9T/-7T+. 69366831929610/5
*7926480 .€€9072¢21S04/1T 7801/26. -**42/42+108555 ¢. 5312S00+78*
203 -*5S%¢€6S*348-962--T+OT *8-S/-@. 8101/1268 -7T*71T .4 T-*/1*8
711+/-694+4/8*/9947/S¢ -84935¢45T/8/S0- . 5¢1T+060-8*%++-2T+698/20
T3/TOS¢+S3T2*+-52S893+4+714245S. *, &. 79S83S¢32*. +6TO450ST--/41T6
34. -23-T*301¢ -4/T2T48.1/¢ 83ST9226S/7544. /-62469++1+7-*-S914.
4S6. *€17-5644 -134919*/*765937/36967 . +5530. G@9910+*9364+5 +343
5**5*55333..3..5..555555
S represents "Subtotal"
T represents "Total"
represents "decimal point"
€ represents "backspace"
138
Appendix B
Experimental Instructions
139
EXPERIMENTAL INSTRUCTIONS
The purpose of the study is to gather information about a newly designed keyboard, called the One-Hand Ternary Chord Keyboard (TCK). You are expected to participate for 40 one-and-a-half hour sessions. There will be one session a day, five days a week for eight weeks; sessions will be scheduled for the same time each day.
The first session will be the "learning phase" and the following 39 will be the "performance phase”.
Learning Phase. In the first session, you will learn general facts about the One-Hand TCK and how to type characters with it. You should proceed through this section at your own pace, follow directions on the computer screen carefully, and consult the experimenter when suggested by the computer screen and also if you have any questions or problems.
You will learn to type the 10 numbers and 8 mathematical symbols during this session. To help you learn these characters, quizzes will be given at times to test your knowledge. Please, keep in mind that you are expected to memorize how to type these characters, and to type them with not more than 3% errors.
Performance Phase. In each one-and-a-half hour sessions, you will be given text files to type which will increase your speed. Follow the instructions displayed on the computer screen. Please keep errors down at 3%, but - - go for speed!
Keep in mind that the purpose of this experiment is to determine the highest typing speed that you can attain on the One-Hand TCK. During the sessions, please work to increase your speed and give the fastest performance you can. It is expected that you will have a general trend to increase your keying speed over the 40 sessions.
Once a week you will be asked to rate your performance on a 0 to 9 scale. You will still finish the full 40 sessions regardless of your ratings, but it is important to the experimenter how you feel you are progressing.
The computer program will explain a great deal about the TCK, so wait until you have proceeded though the INTRODUCTION to the learning section before asking any detailed questions you might have. Consult the experimenter at any time if there are any questions or problems.
Initials
140
Appendix C
Informed Consent Form
141
CONSENT FORM
l, am participating in this research study because | want to participate. The decision to participate is completely voluntarily on my part. No one has coerced or intimidated me to participate.
The experimenters have adequately answered any and all questions | have asked about this study, my participation, and the procedures involved, which are described in the attached "EXPERIMENT INSTRUCTIONS," which | have initialed.
| recognize the research team as Mark McMulkin, Graduate Research Assistant (231-5359) and Dr. K.H.E. Kroemer, Principal Investigator (231-5677).
| understand that they will be available to answer any questions concerning procedures throughout this study. | understand that if significant new findings develop during the course of this research which may relate to my decision to continue participation, | will be informed. | further understand that | may withdraw this consent at any time and discontinue further participation in this study without prejudice to my entitlements. | also understand that the Principal Investigator, his assistants, or medical consultants for this study may terminate my participation in this study if he or she feels this to be in my best interest. | may be required to undergo certain further examinations, if they are necessary for my health or well being.
| do not have any disorders of my cardiovascular system, of my spinal column (particularly in the lower back), or any other disorders or deficiencies, which make it unadvisable for me to participate in this experiment.
| understand that in the case of physical injury, no medical treatment or compensation are offered under the research program, of by VA Tech-VPI.
| understand that | shall receive payment in the amount of $5 per hour plus a $50 bonus for showing a general trend of improvement in keying speed over all 40 sessions. However, | further understand that if | withdraw from the experiment before it is completed, | will be paid only for the time | actually spent performing in the experiment at $5 per hour. If for some reason the equipment used for the experiment malfunctions before the completion of the experiment, | understand that | will only be paid for the time | actually spent performing in the experiment at $5 per hour.
| understand that the results of my efforts will be recorded and that | may be photographed, filmed, or audio/videotaped. | consent to the use of this information for scientific or training purposes, and | understand that any records of my participation in this study may be disclosed only according to federal law, including the Federal Privacy Act, and its implementing regulations. This
142
means that personal information will not be released to an unauthorized third party without my permission. |
| understand that if | have any further questions about my rights as a participant, | may contact Dr. Ernest R. Stoudt, Chairman of the Institutional Review Board at VPI&SU, at 231-5281.
| FULLY UNDERSTAND THAT | AM MAKING A DECISION WHETHER OR NOT TO PARTICIPATE. MY SIGNATURE INDICATES THAT | HAVE DECIDED TO PARTICIPATE UNDER THE CONDITIONS DESCRIBED ABOVE.
Signature Date
Printed Name SS Number
Address Phone Number
Industrial Ergonomics Lab ISE Department VPI&SU Blacksburg, VA 24061
Protocol ERGLAB 1987 Date: Jan. 1987, renewed February 12, 1989 for two
143
Appendix D
Derivation of CPM; = C+T|B1
144
Derivation of the Log-Log function
Log(CPM)) = Bo + ByLn(T})
CPM, = eBo + BiLn(Ti)
CPM; = eBo eBiLn{T))
CPM; = eBo eb n(TiP1)
CPM; = eBo 7,81
Since eBo is constant let C = eBo
CPM; = C+T;1
145
Appendix E
Graphs of All Time Increments and Subjects for CPM; = Bp + ByP; + Bo/Pj +
Bs/P2 and CPM; = eBo P;B
146
LavieQll.dve
ovd/edtd/ldtd.d
lenpy
POLO, INU]
O€ 40}
uoIssasHay ‘|
yoalqns *1°9
eunBbi4
Poa
aINUIW OF
02 S9
09 SG
OS Sr
Ov Se
OF Sz
OZ SI
OL
|
| _ {
{ |__|
| |
| |
{ |
| I
t 1
1 I
| }
1 i
] !
0c
OV
09
08
OOL
Ocl
OVI
O91
O8l
002
elnuily Jed suajoeeyy
147
Lavibul,. ave
ovd/cdtd/ldtd.d
lenjpy
GS
POUedY a}NUI
Op 40}
UoIssesBayY ‘|
Joalqns °2z°9
aunbi4
OS
Poudd
aINUIW OF
Sb Ob
GSE O&
Ge O06
Gt _
Ol S
0
0¢
Ov
09
08
001
0c}
OVI
091
O8l
002
o]nulyy sed syajoeseyyg
148
POS
S}NuUIPy OS
40} uoISsaiHay
‘| yoolqns
‘Ey eunbi
POHSd aiNuUIl|
OS
GY Ov
GE O€
Ge 0¢
Sl Ol
S 0
| |
| |
| ]
l |—
| +
0 Ov
09
LavibQll,dyve ———-—~—
08
ovd/cdtd/l dtd.
--------- 001
jenjoy Ocl
SINUIW Jed syajoeeYyD
Or |
O91
+ O81
—— 002
149
bavielt .ave
ovd/cdtd/tdtd.d
Jenjov
[EAUalU] AINUI|
O9 40)
SuoisseuBay ‘|
yalqns ‘py
aunbi4
POHad aINUIW
O9
OV Ge
0€ Se
0¢ Sl
Ol S
| |
—__| |
L {
| |
I 1
| {
|
+
0c
OV
09
08
00}
0c |
e]nul Jed suajoeieyD
Ovl
O91
O81
00¢
150
jeAsaju| Aeq
10} uoissaiBHay
‘1 yoelaqns
*s°y eunbi4
Keq
OV Ge
0€ Ge
02 Gl
Ol G
0
| |
| |
| |
{ |
I t
I |
i |
| |
0
LavAed,dve ——-———
cevAeq/zqtAeq/LatAed,d ---------
+ O01
leny +
Ob
alnulj Jad syejoeieyy
7 OVt
+ O91
7 O8t
—- 00¢
151
bavibl1, ave
ovd/edtd/l dtd.
jenpy
OZ
POU,
SINUIW] OE
40} uoISsesBeyY
‘Zz yoalqns
-9°9 aunbi4
POuad SyNuUI
OF
S9 09
GS OS
Sh OF
GE OF
Ge 02
StL OF
GS O
| |
| {st
| {
| |
| |
{
| i
! ]
J |
I |
{ I
| 1
I
0c
Ov
09
08
OOt
Oct
OVI
O9t
O8l
00¢
e}Null\ 49d sugjOeBeYyy
152
bavieHl.dve
ovd/edtd/|atd.d
lenjoy
GS
POudY s}NUI-
OP 10}
uoIsseuBay ‘2
yalqns “7°
eunbi4
POUdd aINUIN
OF
0S Sb
Ov Sse
O0& Ge
O06 Gt
Ol S
| |
| |
jt |
| |
| {
I |
I |
| |
T |
1 1
0¢
OV
09
08
00}
Ocl
Ovi
O91
081
00¢
anu Jad syajoeseuy
153
POMS
a}NUIW OG
40} uoIssasHayY
‘Zz elgns
-g°y eunbi4
poued aINUIW!
OS
Sy CO Oh:Ssi‘<i«sSC“‘<«té‘i:SC“‘wSSC‘iSCC(‘éiR(SSCSG
0 |
| l
a |
| |
i 0 09
LdvieWil.dve8 —————
08
ovd/cdtd/ldtd.d ---------
OO}
jenjoy Oc}
Ori
O91
081
— 00¢
154
ejnull sed syajoeeuy
bavieut.dve
ovd/cdtd/ldtd.d
jenjpoy
[EAIBIU] S}NUII|
OQ 410}
SUOISSasHay ‘zg
yoalqns ‘6°9
eunbi4
POUSd 9}NUI
09
OV GE
O& Se
0c Sl
Ol G
0
0c
OV
09
08
00l
Ocl
OV lL
091
O8t
00¢
eINUIW sed SuBjORIeUD
155
LavAeqd, dave
ovAeq/za+Aeq/1g+Aed,
fenjoy
jeAraju Aeg
40} uoissaiBay
‘g yelqns
‘o1°y sunbi4
Keq
Ov GE
O¢ Ge
0d Gl
Ol G
| |
| |
| {_
{ |
| }
| i
I |
0¢
Ov
09
08
001
Ocl
e]nuIw 16d suajoeeyyD
Ovl
O91
O8l
00¢
156
LaviGlt.dva
ovd/edt+d/tdatd.d
Jenjoy
DOO
SINUI
OF 10}
UoIssalBay ‘¢E
yalqns “1
1°9 eunbi4
poua aynuiW
O€
0Z S9
09 SG
0G Sb
Ov SE
O€ GZ
O02 St
OL ¢
O |
| |
| |
{|
| |
| |
i |
| |
1 |
j |
1 |
1 I
t |
| 1
1
Od
OV
09
08
001
0c}
Ovi
091
O8l
00¢
alnui sed syajoeueyy
157
POWdd aINUIW
OF 40}
UoIssesHay ‘E
yoelqns *z1°Q
eunbi4
Poued aINUIW
OF
S¢ 0S
Sr Or
Se of
Sze of
St OL
¢ oO
|
| |
| |
I 1
I IF
0 09
havieQHl.dve —————
08
évd/cdtd/ldtd.d
--------- 0Ol
yenjoy Oct
OF |
091
O8l
— 00¢
158
e]nul~ 4ad syejoeseyD
POUSAY aINUIP]
OG 404
UOIssalHaY ‘€
yelqns ‘e1°4
aunbig
pouad ainui~w
Og
Sp Ohsti‘<iésESC“<«ié‘iOE:C“‘sSSSC‘i
SC C:t‘éisék
S 0
|
| |
| |
—
| |
) 0 09
bdvieHl,dye —-—-———
08
Cvd/cdtd/ldtd.d ---------
00}
jenjoy Och
Ort
O91
O8 |
— 00¢
159
enul, 18d suapeueyyD
baviebll.dve
ovd/cdtd/|dtd.d
renyoy
jeAlaju] aINUIWY
OO 40}
SuOIssasBayY ‘E
yoaelqns ‘p1"g
eunbi
poued eINuIW
09
Op Ge
O€ GZ
02 Sh
Ol S
0¢
Ov
09
08
001
Ocl
orl
091
O81
00¢
einul~— Jed suajoeieuy
160
jeAlaju| Aeg
10} uoissaiHay
‘¢ yoalqns
“¢4°y aunbi4
Keg
OV Ge
O€ Se
02 Gt
Ol G
0
| —t
| |
| |
| 0
| |
| 4
| i
] |
T Od
09
LavAed,a@ye ————~—
08
evAeq/edtAeq/Lg+Aed,g ---------
001
jenpy
OZL
a
091
08}
—- 00¢
161
ainul Jed syajoeeuy
Laviel1l.dye
ovd/cdtd/tatd.d
lenjoy
POHdd ajNuIWy
O€ 10}
UoISSauBayY ‘p
yoalqns ‘91°9
aunbi4
pouad eInuIW
OF
02 G9
09 SG
0G Sr
Or SE
Of Gz
Of Gt
OL
0
02
OV
09
08
001
Ocl
Ovl
091
O8l
00¢
Slnull\ Jed syajpeweyy
162
POW,
BINUIW OF
10} UoISsalBayY
‘p yoelqns
"74°94 aunbi4
pouad aInuIW
OF
cS 0S
Gr Orv
se oe
G2 of
St oO
¢
O |
| 4
| |
| |
| ]
| |
0
OV
09
LdvieHl,dve —————
08
Cvd/cdtd/ld+d.d ---------
00}
jenjoy Och
alnuljy Jad srajoeseuy
OVI
+ O91
+ O8t
—- 002
163
baviell. ave
ovd/cdtd/l dtd.
lenjpy
Sv OV
POS
SINUIW] OG
JO) UdISsaiBay
‘p joalqns
-g1°5 eunbi4
pouad anu]
OS
sé o€
G2 02
st ors
0
0c
Ov
09
08
001
Ocl
Or |
O91
081
00¢
SINuIW 16d syejoeieUy
164
havieul.dve
ovd/cdtd/ld*d.d
lenjoy
|EAJ}U] SINUIW
OQ 40}
SUOIssaJBayY ‘p
yoelans “61°9
aunbi
POUdd a}NUlW
O9
OV Se
O€ Go
0¢ Sl
OL S
| {.
| I
1 i
1 \
| |
|
Oc
Ov
09
08
001
Ocl
Ovl
091
08 |
00¢
anu Jad sr9joeieyyD
165
LavAeqd, ave
evAeq/zatheq/1g+Aeq,g
enjoy
OV
GE O€
Sd
Aeq
0c Gl ! ft
jealajuy Aeg
40} uoissaiBay
‘p yoalqns
‘ogy eunbi4
Oe
OV
09
08
00!
Ocl
OV l
O9l
O8l
00¢
alnulw sad suajoeieyyD
166
bavi. ave
ovd/cdtd/ld+d.d
fenjpy
POUdd aINuUIP
OE 10}
UOISsalHayY ‘s
oalqns *12°9
eunbi4
POLS, 9INUI|
OF
OL S9
09 GS
OS Sb
OF SE
OF G2
O02 Gt
OL G
O
| |
| I
{ |
| |
| |
| {
{ {
I t
I {
1 {
i |
I j
| |
I i
0c
OV
09
08
00!
Oc |
Ovl
091
08}
002
eynully Jed suajoeueyy
167
bavibl lave
ovd/cdtd/ldtd.d
enjoy
POS
SINUIW Ov
40} uoIssalBaY
‘sg yalqns
-zg"yQ eunbi4
OS Gv
OV
POWdd 9INUIW
OV
Ge SE
0€ 0c | __
St | {——
Ol
0¢
OV
09
08
OOL
Ocl
Ovi
O91
O8l
00¢
enully 18d susjoeeYyyD
168
baviGW1.dve
ovd/cdt+d/ldtd.d
lenjoy
POUAd SINUIW]
OS 40}
UdISsalBaYy ‘G
yoalqns ‘Eez’y
eunbi4
SY OV
+
GE
pousd einuiy
OS
06 Gz
oc St
or gs
0
| |
| |
1 |
\ I
J I
I
0c
OV
09
08
O0Ol
Ocl
Ort
O91
081
002
elnul~-, Jad syajoeseyy
169
[eAJa}U] SINUI,
OY 40};
SuoIssaiBayY ‘sg
yoalqns “pz°y
eunbi4
pousd elnuiyy
09
OP ce
o¢€ Sz
02 SI
Ol c
0
, |
| |
l l
L |
| 0
LavilGWl,dva
—_-——
—
—
Cvd/cdtd/ldtd.d
--------- 7
OO-
lenjoy +
OZt
+ OvVl
+ O91
+ O8t
— 00¢
170
ejnul sed susjoeeYyD
jeAraju Aeg
10} uoissaiBay
‘g yoalqns
°sz°y aunbi4
Keg
OV Ge
O€ Ge
0¢ GI
Ol S
0
LavAed,@ve —————
ovAeq/zat+Aeqg/lgt+Aed,g ---------
+ OOL
Gans
+ Oct
+ Ov
7 O9t
+ O81
— 002
171
einuljy Jed syajoeeyy
Appendix F
Average Response Times of Each Chord by Trial
172
EL puoyy
seul Aq
ow) asuodsay ebesaay
“1°44 ounbi4
JaquUNN jeu
00S OST
OO0r OSE
O00€ 0S¢
00¢ OSI
001 OS
0
[| |
| |
| |
| |
| )
t |
| |
| I
I |
I
+ O01
+ Q0¢
7 OO€&
+ O00
+ 00S
+ 009
+ O0OZL
+ 008
+ 006
- OOOL
- OOLL
- O0cl
+- OO€1
+ OOF1
+ OOSI
+ 0091
+ OOZI
7 0081
+ OO6L
+ 000¢
+ OOLe
— 00¢¢
173
(DeSswW ul) awl] esuodsey abeusay
00S OSL
OOP OSE
OO0€
Pl Puoyy
‘yeu Aq
euly asuodsay ebelaay
‘2°'4 eunbi4
JEQUINN jeuL
0S¢ |
00¢ | I
OSt { i
001 | I
0S | l
001
00¢
OO0€
OOF
00S
009
002
008
006
0001
OO-L
00¢1
OO€L
0O0r |
00S |
0091
OOZ 1
0081
0061
000¢
0012
00¢¢
(Oesw ul) aul, asuOodsey sbeusay
174
00S OSP
007 OSE
00€ | |
G| puoyy
:yeu Aq
eu, asuodsay
ebesany ‘¢'4
ounbi4
JEqUINN Jeu,
0S¢ | |
00¢ OS|
0Ol 0S
OOL
00¢
OO0€&
00v
00S
009
002
008
006
0001
OOLL
00c1
OOE |
00V |
00S1
0091
0021
0081
0061
000¢
00Le
00c¢
(O@sw ul) euly esuodsey obesoAy
175
00S OSt
00r OSE
O00€
Q| puoyy
seus AQ
auly asuodsay
ebeleny “p’4
aunbi4
JEquUNN jeuL
0S¢ | I
00¢ OSI | |
001 { I
OS {| I
OOl
00¢
OO&
00V
00S
009
002
008
006
0001
OOLt
001
00€ |
007+ |
00SI
0091
0021
008 |
0061
000¢
0Ole
00¢e¢
(Oesw ul) sully esuodsey abeisay
176
00S OST
0O0V OSE
00€
Z| puoyy
‘yeuy Aq
awl, asuodsay ebeaay
*s’4 aunbi4
JEQuINN jeu lL
0S¢ | i
00¢ | T
OS 1 | |
OOL 0S
001
00¢
00€
00V
00S
009
002
008
006
0001
OOL |
00c1
OO€ 1
007 |
OOS 1
009 |
OOZ1
0081
0061
0002
0Ol¢e
00¢¢
(Oesw ul) euly asuOdsay sbeusay
177
00S OSt
00v OSE
O0€ | i
Eg puoyy
‘jeu Aq
awl, asuodsay
ebessay -9°+4
aunbi4
JEqUINN jeu
0S¢ 00¢
0S |
001 0S | |
001
00¢
00€
007
00S
009
002
008
006
0001
0011
00¢ |
00€ |
00r |
00S 1
0091
OOZ1
OO08t
0061
000¢
001d
00c¢e
(osu ul) Swi, BsuOdseY eHbeuaay
178
00S OSP +
00v OSE
O0€
pz psouy
:jeuL Aq
euiy asuodsay
sbeyany °7°4
aunbiy
JOQUINN jeuL
0S¢ | |
00¢ OSI
O0Ol OS
OOl
00¢
O0O0€
007
00S
009
002
008
006
0001
001}
00¢ |
OOE |
0O0r |
00S |
0091
OO |
008 |
0061
0002
0012
00¢¢
(O@sw ul) swly asuodsay aBeusay
179
00S OST
OOP OGE
00€
Gz puoyuy
yeu
Aq awiy
asuodsay ebesaay
“g°+4 aunbi-4
JOQUNN jeu]
0S¢ | i
00¢ OS| | ]
001 | I
OS
001
00¢
00€
00V
00S
009
002
008
006
0001
0011
00c1
OOEL
OOVl
00SI
0091
OOZI
0081
0061
000¢
0Ole
00c¢
(O@sw ul) euuly eBsuOdsey sHbeieAy
180
00S OS?
OOP OSE
OO0€
{ |
QZ ps0yy
‘yeu Aq
owl asuodsay ebeuaay
*6°4 aunbi4
JOquny seul
0S¢ 00¢ | |
OSI | |
OOL { |
0S
001
00¢
O0€
00V
00S
009
002
008
006
0001
OOLL
00cl
OOEL
Or
00S |
0091
OOZ1
0081
0061
000¢
0OLe
00é¢
(Oasw ul) dw] ssuOdseYy ebessaay
181
00S OS?
00P OSE
00€ | |
8g PuoUy
:jeuy Aq
aul) asuodsay sbeisay
‘O|'+ aunbi4
JOquNN |e
0S¢ | I
00¢ |
OSI OO}
OS
OOL
00¢
O0O0€
00V
00S
009
002
008
006
0001
OOL |
00cl
OO0€ L
0Ovl
00S1
0091
0021
0081
0061
000¢
0O0Le2
00¢c2
(OesSw ul) eluly eBSUOdseY BbeisAy
182
GE puoyy
:jeuL Aq
oul asuodsey sHesany
*11°4 aunbis
Jaquinn jeu
00S OST
00” OSE
OO0€ 0S¢
00¢ OSI
O0Ol OS
0
| |
| |
| |
| 4
| 0
+ 001
+ 002
+ 00€
+ 00¢
+ 00S
+ 009
+ 004
+ 008
- 006 OOO}
i OO}
+ 001
+ O0€1
+ 00v1
+ 00S}
+ 0091
+ OOLI
+ 0081
+ 0061
+ 0002
+ 0012
+ 00z2
183
(Oesw ul) sul) asuodsey sbeisAy
00S OSV
00r OSE
00€
9€ POY
jeu, AQ
owt, esuodsey ebeyaAy
‘Z1°4 eunbi4
JOqUINN [eu
0S¢ 00¢ | |
OSI 001
00}
00¢
00€
007
00S
009
002
008
006
OO00L
OOl |
00cl
00€ |
007 |
00S1
0091
OOZ1
0081
0061
000¢
00l¢e
00¢¢
(Desw ul) ewuly eSuodseyY ebeieay
184
00S + OSP
OO0V OSE
00€ { |
Z€ puouy
yeuy Aq
ewiy asuodsey sbeseny
“eE1°4 eunbi4
JOQUINN JeuL
0S¢e
| I
00¢ OS |
OOL OS
OOl
00¢
00€
OOP
00S
009
002
008
006
0001
OOl}
00ct
OOEl
OOrl
00S |
0091
OOZL
0081
OO6l
000¢
001¢e
00¢¢d
(Oesw ul) awl, esuodsey sBbeusay
185
00S OSt
0O0r OSE
O00€
Gp Puouy
:jeuy Aq
euly esuodsay
ebeieay ‘p1'4
aunbi
JOQuUNN jeu,
OSe {| l
00¢ OS}
001 | —|—
OS | t—
O01
00¢
OO0&
00v
00S
009
002
008
006
000t
OOl tL
00cl
OOEL
OOF |
00S I
0091
OOZ1
008 |
0061
000¢
O0l¢e
00e¢
(DeSuw ul) eu] easuodsey eBbeieay
186
00S OST
OOV OSE
OO0&
9p puoyy
‘jeu, Aq
ewly asuodsay ebeiaay
“s1°4 aunbi4
JOQUINN jeu
0S¢ 00¢
OS | 001
OS | | : i
001
00¢
O00€
OOP
00S
009
002
008
006
0001
OO} |
00c |
O00 |
OOr
00S |
0091
OOZI
0081
0061
000¢
0OLe
00c¢
(Dasw ul) sul] esuodsey ebeusay
187
00S OS
OOr OSE
O0€ | |
8p Psoyy
:jeu, Aq
aul asuodsay
ebeuany ‘91°44
eunbi4
JEQuNN [el
0Ge 002 |
OS | | {
OOL OS
001
00¢
OO0€
007
00S
009
002
008
006
000l
OOl |
00d}
O0€ |
00r I
00S |
0091
OOZI
008 t
0061
000¢
001d
00¢¢
(oasw ul) eu, esuodsey sBessAy
188
00S OSV
00V OSE
00€ | |
ZG psouy
:jeuy Aq
au) asuodsay
ebesny
*/}°4 eunbi
JequnN jeuL
OS¢ { i
00¢ OS 1
OOL
002
00&
007
00S
009
002
008
006
0001
OOLL
00cl
OOE L
ooVl
OOSt
0091
OOZI
O08 t
OO6I
000¢
00l¢e
00c¢
(Oaswi ul) sul, esuodsaey aBbeisay
189
89 POU
‘yeu Aq
euly asuodsay sbeieay
“g}°4 eunbi-y
Jaqunn
seul
00S OSY
OOP OSE
OO0€ 0S¢
00¢ OSI
00} OS
0
| |
| |
| |
| 0
| i
| |
i |
| ]
| I
OOL
7 00¢
+ OO0€
+ OOD
+ 00S
+ 009
+ 002
+ 008
+ 006
- O00}
r OOLL
- O0Zc!
7 OOEl
qt O01
y OOS1
y~ OO9L
+ OOZt
+ 0081
+ OO6I
+ 000¢
+ O0l2
—— 00¢¢
190
(DeSwW ul) su] esuodsey eHbeusay
VITA
Mark Lee McMulkin was born on May 17, 1966 in Great Bend, Kansas.
He received a B.S. degree in Mechanical Engineering from the University of
Idaho in May 1989. While pursuing his Master Degree at Virginia Polytechnic
Institute and State University (Virginia Tech), he worked as a Graduate
Research Assistant for two years performing research on the Ternary Chord
Keyboard and finger speed. He is a member of Tau Beta Pi and the Human
Factors Society. He will be working on his Doctoral Degree in Human Factors
at Virginia Tech.
Muck. Mb Mark L. McMulkin
191