selectivity in the german mobility panel tobias kuhnimhof institute for transport studies,...
TRANSCRIPT
Selectivity in the German Mobility Panel
Tobias Kuhnimhof Institute for Transport Studies, University of Karlsruhe
Paris, May 20th, 2005
Institute for Transport Studies, University of KarlsruheMay 20th 2005 - 2
Overview
The German Mobility Panel
MOP-recruitment, non-response and assumptions about selectivity
Additional sources of information to analyse selectivity in the MOP
Some possibilities to analyse selectivity issues
Some findings
– Trustworthiness of CATI-Data
– Who drops-out and who takes part
– The advantages of recruiting households
Conclusions
Institute for Transport Studies, University of KarlsruheMay 20th 2005 - 3
The MOP-survey: 7-day mobility diary in 3 consecutive years since 1994
7 days Longitudinal survey of individual mobility behavior
The German Mobility Panel
Year 1
3 years Survey of individual developments, transitions, causes and effects
Year 2
Year 3
Institute for Transport Studies, University of KarlsruheMay 20th 2005 - 4
The multi-stage recruitment process:
Commercial market research CATI
MOP-recruitment CATI
Written declaration of participation
Participation: 1-Week-mobility-diary report, mailing of documents
recruitment (mailing of documents etc.)
+ 1-week-report
+ 3-year-survey
= high participant burden
There is plenty possibility & it is understandable
not to participate in the MOP
Recruitment, Nonresponse and Assumptions about Selectivity
The MOP recruitment: motivating relialable participants
Institute for Transport Studies, University of KarlsruheMay 20th 2005 - 5
Recruitment, Nonresponse and Assumptions about Selectivity
4.6
6.5
12.9
22.1
44.9
52.5
100
0 10 20 30 40 50 60 70 80 90 100
% Remaining Phonenumbers, Adresses, Participants
MOP-Participation
Returning of documents
Recruitment-CATI
Stratified sampling
Agreement to 2. contact
Initial CATI
Tel.-Nrs. after neutral drop-out
Drop-out in the different stages in the 2002 MOP-recruitment
High respondent burden
High drop-out
Strong selective bias?
Institute for Transport Studies, University of KarlsruheMay 20th 2005 - 6
Mobility in Population
P
Business trips?
High incomes ?
Stress due to job, education, children?
Large leisure activity spectrum?
Not mobile due to incapacity (e.g. sick)?
Not interested in mobility issues
Common Assumptions about selective error in mobility surveys
MOP-Recruitment, Non-response and Assumptions about Selectivity
Institute for Transport Studies, University of KarlsruheMay 20th 2005 - 7
MOP-Participation
Returning of documents
Recruitment-CATI
Stratified sampling
Agreement to 2. contact
Initial CATI
Tel.-Nrs. after neutral drop-out
OBJECTIVE: Get to know the drop-outs
BUT: preserve the survey’s continuity
Mobility interview
Enquiry of reasons for declining
Documenta
tion o
f entir
e dro
p-out
Documentation of calling attempts (max. 12)
Additional sources of information to analyse selectivity in the MOP
Institute for Transport Studies, University of KarlsruheMay 20th 2005 - 8
Core-element of selectivity analysis: Mobility interview in commercial market research CATI
- individual and household socio-economics
- role of the person in the household
- interviewee’s daily obligations (childcare, work etc.)
- individual mobility on a test day (“yesterday”)
- characterizing information about the person’s general mobility behavior
- proxy-data on other household members
Data available about all CATI-interviewees (MOP-participants & drop-outs):
Additional sources of information to analyse selectivity in the MOP
Institute for Transport Studies, University of KarlsruheMay 20th 2005 - 9
MOP-Participation
Returning of documents
Recruitment-CATI
Stratified sampling
Agreement to 2. contact
Initial CATI
Some possibilities to analyse the selectivity issues
Analysees to understand selective bias
Modeling participation: P(participation)=f(mobility, socio-demografics)
Logit 4
(Sampling Drop-out)
Global drop out logit
Logit 3
Logit 2
Logit 1
Institute for Transport Studies, University of KarlsruheMay 20th 2005 - 10
All MOP-Participants (1.-,+ 2.-,+ 3.-time participants)
Respondent groups and data availability:
CATI-Interviewees
CATI and recruitment:
First-time MOP-Participants
(CATI-Interviewees + other household members)
CATI-Interviewees, who participate in the MOP
Possibilities to compare mobility data of - different sources (survey methods) - and different samples
Some possibilities to analyse the selectivity issues
Institute for Transport Studies, University of KarlsruheMay 20th 2005 - 11
Findings: Trustworthiness of CATI-data
Are CATI-responses useful to characterize mobility behavior?
Intrapersonal comparison of general mobility information (CATI) and reported mobility (MOP)
Institute for Transport Studies, University of KarlsruheMay 20th 2005 - 12
Are CATI-responses useful to characterize mobility behavior?
- in most cases: YES
- the stronger the questions relate to routines, the more reliable is the answer
- active persons tend to overestimate themselves
- less active persons tend to underestimate themselves
- “Extrapolation of normal weekday” – Example: use of travel modes
Daily use of car = car use on 5,4 days per week Daily use of PT = PT use on 4,5 days per week
Findings: Trustworthiness of CATI-data
Institute for Transport Studies, University of KarlsruheMay 20th 2005 - 13
Differences in cross-sectional mobility figures in CATI and mobility diary data
Comparision of test-day cross-sectional data of CATI vs. MOP
Findings: Trustworthiness of CATI-data
Institute for Transport Studies, University of KarlsruheMay 20th 2005 - 14
Methodological Differences: The MOP-Report is more exact than the CATI
CATI MOP P > |t|
Estimate of methodological
difference
Share of mobile persons - total 87.50% 91.90% 0.008 ~4.3%
Share of mobile persons – under 60 years 89.9 93.50% 0.06 ~3.6%
Share of mobile persons – over 60 years 82.7 88.60% 0.06 ~6.1%
Trips per mobile person and day 3.07 3.57 < 0.0001 ~0.5
Travel time per mobile person and day [Min] 87.6 85.5 0.68 ~0
Test day data of persons who participated in both surveys*:
*No Fridays, Saturdays, exclusion of daily trips that exceed 6
Findings: Trustworthiness of CATI-data
Institute for Transport Studies, University of KarlsruheMay 20th 2005 - 15
What is the impact of the sample differences due to selectivity on mobility figures?
Comparison of cross-sectional data CATI – MOP: After accounting for methodological differences the selective error can be estimated
Findings: Who drops out and who takes part
Institute for Transport Studies, University of KarlsruheMay 20th 2005 - 16
CATI test day data vs. MOP test day data of all persons in both surveys*:
*No Fridays, Saturdays, exclusion of daily trips that exceed 6, appropriate weighting procedures applied in order to account for socio-demographic differences
CATI MOP
Estimate of methodological
difference
Estimate of difference due
to selectivity
Share of mobile persons – total 83.90% 91.70% ~4.3% ~3.5%
Share of mobile persons – test persons under 60 years 88.10% 93.20% ~3.6% ~1.5%
Share of mobile persons – test persons over 60 years 73.60% 88.00% ~6.1% ~8.3%
Trips per mobile person and day 3.09 3.53 ~0.5 0
Travel time per mobile person and day [Min] 80.83 81.58 ~0 0
Drop-Out of Non-trippers
Findings: Who drops out and who takes part
Institute for Transport Studies, University of KarlsruheMay 20th 2005 - 17
Who drops out and who takes part
Socio-economic aspects of Selectivity
Monthly Household Income
0 0.5 1 1.5 2
<750 Euro (Ref.)
750 - 1500 Euro
1500 - 2000 Euro
2000 - 3000 Euro
3000 - 4000 Euro
4000 - 5000 Euro
> 5000 Euro
Age of Interviewee
0 0.5 1 1.5 2
14-17
18-25
26-35
36-50 (Ref.)
51-60
61-70
>=71
Significant, P<0.1 P<0.2 Not significantP<0.3
Example: Odds ratios of 2 out of 15 variables in participation logit model
“Middle class bias”
Institute for Transport Studies, University of KarlsruheMay 20th 2005 - 18
Who drops out and who takes part
Explanations for increasing share of mobile persons
Example: Odds ratios of 2 out of 15 variables in participation logit model
“Mobility interest bias”
Distance to work place, place of education etc.
0 0.5 1 1.5 2
0-5 KM
5-20 KM (Ref.)
20-40 KM
>40 KM
Weekday Evening Leisure Activities
0 0.5 1 1.5 2
active(daily)
normal(Ref.)
less active(<2x /
Woche)
Significant, P<0.1 P<0.2 Not significantP<0.3
Institute for Transport Studies, University of KarlsruheMay 20th 2005 - 19
The advantages of surveying households
CATI-Interviewees
CATI and recruitment:
First-time MOP-Participants
(CATI-Interviewees + other household members)
CATI-Interviewees, who participate in the MOP
Unemployment rate
Distance to work place
4.2% 14.7 Km
2.7% 17.1 Km
3.7 % 14.8 Km
Participation of other household member can counter-balance selectivity impacts
Institute for Transport Studies, University of KarlsruheMay 20th 2005 - 20
Conclusions and Recommendations
• Socio-economic selective effects dominate drop-out• “Middle Class Bias” good education, good income, middle aged• “Mobility Interest Bias” drop out of non-trippers (particularly elderly,
permanent incapacity?)
Findings on Selectivity
Other Findings
• CATI-data is useful but has to be interpreted• Surveying households counter-balances selectivity
Recommendations
• Balanced recruitment of different mobility styles is vital• Survey households not individuals• Don’t trade data quality for a high response rate