date 17-05-2011 |1 the potential statistics for event ... · the potential statistics for event...
TRANSCRIPT
|Date 17-05-2011 1
The potential statistics for Event Related Potentials (ERPs)
Laura S. Bos
Center for Language and Cognition, Groningen (CLCG)
|Date 17-05-2011 2
Overview
› Extremely quick ERP-course
› The Time Reference ERP study
� Most frequent way of ERP analysis:Repeated measures ANOVA
› When ANOVA might not be right
› An alternative: non-parametric testing
|Date 17-05-2011 3
Extremely quick ERP course
|Date 17-05-2011 4
Extremely quick ERP course
› ERP effectsDe ober die nu de peper maalt krijgt geen fooi.The waiter who now the pepper grinds gets no tip.
* De ober die zonet de peper maalt krijgt geen fooi.The waiter who just before the pepper grinds gets no tip.
P600 effectPz electrode
|Date 17-05-2011
Dragoy, Stowe, Bos & Bastiaanse (subm.)From time to time: Processing Time Reference Violations in Dutch
› Time Reference
� Agrammatic aphasia: reference to the past particularly problematic (cf. present)
� Specific for aphasia? Theory indicates no:
- Reference to past: discourse linking Zagona, 2003
- Present tense: locally bound
- Parallels with e.g. pronouns vs. reflexives
5
|Date 17-05-2011 6
Introduction: discourse linking
Time line› Present: evaluation time = event time (Zagona, 2003)
De man drinkt koffie (onvoltooid tegenwoordige tijd)The man drinks coffee (simple present)
__________________________________________________
› But:De man dronk koffie (onvoltooid verleden tijd)The man drank coffee (simple past)
___________________________________________________
sentence
sentence
|Date 17-05-2011
Method ERP experiment:
Materials
7
|Date 17-05-2011 8
Method ERP experiment:
Materials and procedure› 380 sentences
� 80 (20 x 4) Time Reference (TR)� 300 other sentences- control, filler, other experiment
› 4 lists with 8 subjects per list › word-by-word reading
� 500 ms fixation point *� 240 ms word, 240 ms blank� 1750 ms blinking interval *****
|Date 17-05-2011 9
Method ERP experiment:
Procedure› 3 time windows:
� 300-500ms; 500-700ms; 700-1000ms� Comparisons: target V the same- PresPres vs. *PstPres and PstPst vs. *PresPst
Pz electrode
PresPres
*PstPres
|Date 17-05-2011
Data analysis› Repeated measures ANOVA (N-way/factorial)
� Every subject saw sentences in 4 conditions. - one of every quadruplet- Ho: ERPs violation sentence = ERPs correct sentence- H1: ERPs violation sentence ≠ ERPs correct sentence
� Factors:- TR (present, past)- Violation (correct, violated)- Midline: posteriority (3)- Lateral: posteriority (6) and laterality (2)
10
|Date 17-05-2011
In SPSS› Organised by ROI per condition (left-right) › Every subject one line› Time windows analysed seperately � filter data
11
|Date 17-05-2011
In SPSS
12
|Date 17-05-2011
In SPSS
13
Use syntax function (paste instead of ‘ok’)
|Date 17-05-2011
In SPSS
14
|Date 17-05-2011
Results: overall
15
Time window
Midline ROIs Lateral ROIs
300-500 500-700 700-1000 300-500 500-700 700-1000
Violation n.s. F(1,31)=6.55
p<.02
F(1,31)=4.84
p<.05
n.s. F(1,31)=6.17
p<.02
F(1,31)=1.31
p<.02
Violation *targetV
F(1, 31) = 6.34 p < .02
F(1, 31) = 7.74, p < .01
F(1, 31) = 12.33, p <
.001
(F(1, 31) = 7.35, p < .01
F(1, 31) = 6.36, p <
.02
F(1, 31) = 13.52, p < .001)
› Interpretation:
� 500ms after TR violation by V an effect occurs
� 300ms after TR violation by V, present and past tense verbs are processed differently
|Date 17-05-2011
Results: present reference V
16
Time window
Midline ROIs Lateral ROIs
300-500 500-700 700-1000 300-500 500-700 700-1000
Violation F(1, 31) = 7.97, p < .01
F(1, 31) = 16.46,
p < .001
F(1, 31) = 17.19, p < .001
F(1, 31) = 8.19,
p < .01
F(1, 31) = 12.25, p < .001
F(1, 31) = 17.83, p < .001
Violation* posteriority
n.s. F(2, 62) = 4.83, p < .02
F(2, 62) = 9.62, p < .001
n.s. F(5, 155) = 2.87, p < .08
F(5, 155) = 7.29, p < .01
› Interpretation: see waveforms
� 300ms after Vpresent a violation of time reference elicits a positive wave, peaking at 600-700ms.
� larger effect at central and posterior ROIs
|Date 17-05-2011
Results: past V
17
Time window
Midline ROIs Lateral ROIs
300-500 500-700 700-1000 300-500 500-700 700-1000
Violation n.s. n.s. n.s. n.s. n.s. n.s.
Violation* posteriority
n.s. n.s. n.s. n.s. n.s. n.s.
› Interpretation:
� no effect for a violation of time reference by Vpasttime-locked to the verb
|Date 17-05-2011
Results
18
› Interpretation:
� violations of the temporal context by past tense verbs are processed differently from violations by present tense verbs.
|Date 17-05-2011
When ANOVA might not be right
› McCarthy and Wood (1985)
› ERP is multiplicative:� twice ERP source strength
� twice voltage at ROIs
› ANOVA (+mixed model) is additive:� twice ERP source strength
� plus X μV at every ROI� y = I + ax1 + bx2 + …
� Assumes Stotal = Serror + Smanipulation
� Possible: sign. interaction violation*location even though the distribution is the same (but different source strength).
19
|Date 17-05-2011
When ANOVA might not be right
› Maris (2004) and Maris & Oostenveld (2006)� If locus & time of the effect are not predicted?
� running several tests on the data causesproblems; the family-wise error rate increases
› Nonparametric testing
20
|Date 17-05-2011
When ANOVA might not be right
› Maris & Oostenveld (2006): Nonparametric testing� permutation tests
1.put all trials (correct + violation) into 1 set � trial: time bin of e.g. 100 ms
2. draw as many samples as you had items (so: 20)�this is the ‘random partition’
3. calculate a test statistic4. repeat 2 and 3 many times (e.g. 1000x)5. calculate the proportion of random partitions with a
more extreme value � the ‘p-value’ (Monte Carlo)6. if p < α then the distributions are different
21
|Date 17-05-2011
When ANOVA might not be right
› Maris & Oostenveld (2006)
� Visual inspection:
- Cluster the significant trials bytemporal adjacency
- calculate sum of t-statistics per cluster
- take the cluster with the largestt-statistic
� Uncorrected not conservativeenough
� Bonferroni correction (α/samples) more conservative
22
|Date 17-05-2011 23
Baggio (2008)› Similar study analysed with Maris & Oostenveld’ method� also a P600 effect for PsPr violations (present verb)
� more precise indication when and where
|Date 17-05-2011
Results and conclusions
� Time reference violations by a past tense verb are processed differently from by a present tense verb
� In ERPs, effects are not additive over the eletrodesites, as ANOVA assumes for interactions
� Nonparametrical testing of ERP data: especially iflocation and time point of effect are not predicted
� Different results? � to be tested
24
|Date 17-05-2011 25
Questions?› Thank you for your attention!
› References› Baggio, G. (2008) Processing temporal constraints: An ERP study. Language
Learning, 58, 35-55
› Maris, E. (2004). Randomization tests for ERP topographies and whole spatiotemporal data matrices. Psychophysiology, 41, 142–151.
› Maris, E. & Oostenveld, R. (2007). Nonparametric statistical testing of EEG- and MEG-data. Journal of Neuroscience Methods 164, 177-190.
› McCarthy, G. & Wood, C.C. (1985). Scalp distributions of event-related potentials: an ambiguity associated with analysis of variance data. Electroencephalography and clinical neurophysiology 62, 203-208.
› Zagona, K. (2008) Tense and anaphora: Is there a tense-specific theory of co-reference? In Barrs, A. (ed.) Anaphora: A reference guide. (pp140-171). Oxford: Blackwell Publishing.
|Date 17-05-2011
Results: present reference V
26