Attrition and Selection of alteri Respondents in the pairfam panel

Attrition and Selection of alteri
Respondents in the pairfam panel
Ulrich Krieger, SHARE MEA University of Mannheim
Mannheim Research Institute for the Economics of Aging
www.mea.uni-mannheim.de
Outline
 Motivation
 The pairfam Panel
 Overview of cooperation and response in partner
survey
 Model of cooperation in wave 2
2
Motivation
 Attempt to include alteri respondents in surveys like
partners, parents or siblings raises questions on the
selection process of these additional survey
members.
 Reviewers continue to criticize the selectivity. The
processes need to be better understood.
 Weighting
 Limitations of research
 Recent analysis with the Netherlands Kindship Panel
Study show no bias in models using data of children
(Kalmijn & Liefbroer 2011). But how to model the
error in the sample?
3
The pairfam Panel
 First two waves of the German family panel study
pairfam.
 Design:
 Register based sample of resident population.
 Three age cohorts: 14-17, 24-27, 34-37
 CAPI study administered by Infratest
 Partner survey
 All respondents in a partnership are asked for consent to
approach their partners for an interview.
 within or outside of the household
 PAPI questionnaire handed out, left behind with main
respondent or mailed to partner.
 collected by the interviewer or returned by mail
4
The Role of the Interviewer
 Interviewers are encouraged
 to administer the partner survey the way preferred by
respondents
 to occupy the partner with the partner survey during the
main respondents CAPI questionnaire.
 to collect the questionnaire in person
 Request the partners to participate in person whenever
possible
5
the pairfam panel - wave two
 Anchor: monotonous design, only respondents reapproached.
 Partners: Same design as in wave one.
 Consent request to Anchor, then partners are contacted.
 Partners can be the same as in wave one if the relationship
did hold.
 Partners who refused to answer in wave one are contacted
again.
6
Cooperation and Response - Wave one
N
Anchor
%
12402
with partners
7234
100
consent to partner interview
5281
73
returned partner quest.
3743
52
877
12
lost for W2 due to anchor NR
7
Cooperation and Response - Wave one
Wave 2
N
%
Anchor interview
9069
with partners
5408
100
same partner as in W1
4273
95
Consent to partner interview
3882
73
Consent in both waves
3009
56
returned partner questionnaire
2688
50
returned partner quest. in both waves 2081
38
8
 Two error sources:
 Selection of main respondent and
 Nonresponse of Partner
 Both processes are probably not independent of each
other
 Unsure about how to model the Data
 Here: logistic Regression on wave 2 cooperation,
concentrating on main respondents keeping the same
partner over the 2 waves.
9
Descriptive Statistic
Variables
mean
W2 Part. cooperation
min
max
.58
0
1
W1 Part. cooperation
.64
0
1
W2 Main Resp: Consent
.20
0
1
W2 Survey handed out
.38
0
1
Part. fulltime empl.
.55
0
1
Part. parttime empl.
.09
0
1
Part. self empl.
.06
0
1
Part. Years of educ.
12.92
7
20
Partner female?
.42
0
1
Partner born in Germany?
.85
0
1
Main: Satisf. Relationship
7.99
2.15
0
10
W2 Main: Satisf. Relationship
6.80
2.50
0
10
N: 3748
sd
3.08
10
10
Log. Regression on W2 Partner Coop (only ‘stable’ Rel.)
Variables
OR (se)
W1 Part cooperation
11.26*** (1.28)
W2 Main Resp: Consent
0.95
W2 Survey handed out
4.85*** (0.46)
Part. fulltime empl.
1.01
(0.11)
Part. parttime empl.
1.42*
(0.25)
Part. self empl.
0.64*
(0.13)
Part. Years of educ.
1.07*** (0.02)
Partner female?
1.18
(0.12)
Partner born in Germany?
1.40**
(0.17)
Main: Satisf. Relationship
1.06**
(0.02)
W2 Main: Satisf. Relationship
0.94**
(0.02)
(0.13)
N:3748, Pseudo-R²:0.32, BIC: 3538
11
Conclusion
 Cooperation and Consent in Wave 1 do influence
cooperation in wave 2
 How to better describe the selection process?
 Here I focused on stable relationships. How to
include all relationships, new partners?
12