Who are the Nonresondents? An Analysis Based on a New Subsample of the German Socio-Economic Panel (SOEP) including Microgeographic Characteristics and Survey-Based Interviewer Characteristics Peter Schräpler 12 , Jürgen Schupp 23 and Gert G. Wagner 24 r-University Bochum, LDS NRW 2 DIW Berlin , 3 FU Berlin, 4 Berlin University of Technology Q 2008 Conference, Rome, 8th.-11th. July 2008
32
Embed
Who are the Nonresondents? An Analysis Based on a New Subsample of the German Socio-Economic Panel (SOEP) including Microgeographic Characteristics and.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Who are the Nonresondents?
An Analysis Based on a New Subsample of the German Socio-Economic Panel (SOEP)
including Microgeographic Characteristics and Survey-Based Interviewer Characteristics
Jörg-Peter Schräpler12, Jürgen Schupp23 and Gert G. Wagner24
1 Ruhr-University Bochum, LDS NRW 2 DIW Berlin, 3 FU Berlin, 4 Berlin University of Technology
Q 2008 Conference, Rome, 8th.-11th. July 2008
Schräpler
2
Outline
Introduction Reasons for Unit Nonresponse Nonresponse in Sample H
Descriptive Analysis Microgeographic data Interviewer data
Multilevel Analysis Consequences Summary and Conclusion
3
Introduction
Unit nonresponse is one of the most important issues in the empirical social science
Danger of selectivity: leads to biased samples, samples are not random it is important to investigate in which manner the realized
sample differ from the intended sample and to look at the consequences
Main reasons for Nonresponse: Problem of Non-Accessibility Problem of Non-Ability Problem of Refusals
4
Reasons for Nonresponse1. level: Accessibility
Result of impossibility to contact household members. It can be seen as (Groves/Couper 1998):
a function of the physical reachability of the household the circadian rhythm of the household members contact strategies of the interviewers Problem: Causes often can‘t be measured directly
Some empirical findings:
socio-economic status, household size, vocational status and age are important for mobility (cf. Goyder 1987, Schneekloth/Leven 2003, Koch 1997, Schräpler 2000)
Interviewers with higher workload have less nonresponse due to non-reachability (cf. Schräpler 2000)
5
Reasons for Nonresponse2. level: Ability
Unit Nonresponse depends on the ability of the household member to participate
Individuals are ill and can‘t participate. Assumption: health problems increase with the age of the respondent (c.f. Schneekloth/Leven 2003)
Assumption: sometimes an alibi and a „soft refusal“
6
Reasons for Nonresponse 3. level: Motivation/Cooperation
depends on respondents’ assessment of the interview situation and evaluation of the consequences of possible actions (RC theory)
Opportunity costs an interview takes time, survey has to serve a meaningful purpose
Privacy and confidentially concerns invasion of privacy (cf. Singer et al 1993) critical distance and possible mistrust of surveys in more
intellectual environments in Germany (Schneekloth/Leven 2003)
Fear of crime high population density areas, anonymous residential zones (cf.
Interviewer interviewer’s age, gender, motivation, attitudes and experience (cf.
Esser 1986, Loosveldt et al. 1998, Schräpler 2006, 2004, 2000)
7
SOEP - Sample H - Fieldwork
Subsample H of SOEP started in year 2006
From 6,000 household addresses (4 per sample point) overall 3,931 household addresses were recorded by random walk
The process of address recording is separated from the interviewing process: the interviewer receives fixed addresses from the fieldwork
organization
The first wave was launched by 234 interviewers. Of these, 143 were already members of the SOEP staff.
All interviews were carried out by CAPI
8
Nonresponse in Sample H
Reasons for Nonresponse in Sample H N % Gross Sample 3931 100 ./. Non-systematic Drop-Outs Household not detectable 169 4.30 At the moment not feasible 12 0.31 Adjusted Gross Sample 3750 100 ./. Systematic Drop-Outs Not accessible 485 12.93 Refusal 1487 39.65 Not able to participate (c.f. nursing case) 172 4.59 Whole Sample Point lost 15 0.40 Individual household without treatment 82 2.19
Serious problem for nonresponse analysis: Information gap on respondents and nonrespondents
to fill the gap we use• commercial microgeographic data on the households‘
immediate neighbourhood• demographic variables of the interviewers• results of an interviewer questionnaire
10
Microgeographic Information
Use of additional commercial microgeograhic data on the households’ immediate neighbourhoods from the MOSAIC Data system
contains more than 75 individual characteristics used to analyse and describe customer databases or markets for instance Sinus Milieus®, Status, removal volume etc.
information is available at the address level and contains 17.8 million buildings in Germany the building level contains seven or eight households on average (at
least five households)
Important: linked information is not necessary in line with the reality of the particular household (only an approximation for the neighbourhood)
11
Interviewer data
Use of interviewer data from the SOEP interviewer data set mainly demographic variables like gender, age,
education, family status etc. Use of a dataset based on a interviewer
questionnaire mainly personality variables and self assessments filled out by 165 of the 234 SOEP interviewers in
sample H
12
Respondents by Sinus Milieus (N=1,449)
rel. Bias in %Ref.: Milieu distr. for addresses
< -50
> -50 till -30
> -30 till -10
> -10 till +10
> +10 till +30
> +30 till +50
> +50
13
Refusals by Sinus Milieus (N=1,435)
rel. Bias in %Ref.: Milieu distr. for addresses
< -50
> -50 till -30
> -30 till -10
> -10 till +10
> +10 till +30
> +30 till +50
> +50
14
Noncontact by Sinus Milieus (N=470)
rel. Bias in %Ref.: Milieu distr. for addresses
< -50
> -50 till -30
> -30 till -10
> -10 till +10
> +10 till +30
> +30 till +50
> +50
15
„Not Able to Participate“ by Sinus Milieus (N=167)
rel. Bias in %Ref.: Milieu distr. for addresses
< -50
> -50 till -30
> -30 till -10
> -10 till +10
> +10 till +30
> +30 till +50
> +50
16
Four Multilevel Logit Models Model 1 – probability for response variable „interview“
(participation) vs. non-response Model 2 – probability for response variable „refuse to participate“
vs. „participate“ Model 3 – probability for response variable „household not
reachable“ vs. „participate“ Model 4 – probability for response variable „household not able to
participate“ vs. „participate“
Two sets of Predictors:1. Model version A with demographic and household variables for the
potential respondents, microgeographic variables and demographic variables for the interviewer
2. Model version B with additional interviewer variables from the interviewer questionnaire
17
Two-level Logit Models
* participation 1, if 0,
0, otherwiseij
ij
yy
* unit-nonresponse (refuse, nocontact, not able)1, if 0,
0, otherwiseij
ij
yy
ij ij ijy u
1
0 , , 01
1 exp( ( ))H
ij j h ij h ij jh
x v
Random-Intercept Model:
Level 1: respondents, Level 2: interviewers
18
Version A: Multilevel logit estimates – age of thepotential respondents
Version A: Multilevel logit estimates – familystructure in the neigbourhood
Variable Coeff. Coeff. Coeff. Coeff.
... ... ... ... ...mainly single household (Ref.)far above average share of single HH -0,005 -0,02 0,574 2,31 * -0,597 -1,90 + -0,319 -0,58above average share of single HH -0,130 -0,60 0,686 2,70 ** -0,765 -2,28 * 0,255 0,47light above average share of single HH 0,104 0,47 0,398 1,54 -0,778 -2,25 * 0,078 0,14mixed family structure -0,087 -0,39 0,704 2,71 ** -0,674 -1,92 + -0,052 -0,09light above average share of family with children0,169 0,73 0,434 1,63 -1,062 -2,78 ** 0,082 0,14above average share of family with children 0,164 0,69 0,460 1,69 + -1,211 -3,08 ** -0,226 -0,37far above average share of family with children0,135 0,55 0,581 2,09 * -1,605 -3,70 *** -1,052 -1,56almost only families with children 0,278 1,08 0,451 1,55 -1,536 -3,13 ** -1,793 -2,06 *... ... ... ... ...
to continue
Participation vs. Refused Nocontact vs.
Nonparticipation vs. Participation Participation
Not Able vs.
Participation
z-value z-value z-value z-value
24
Version A: Multilevel logit estimates –Random effects
Refusals, noncontact and “unable to participate” relate to different respondent, area and interviewer characteristics:
Respondent is easy to persuade: well-established Sinus Milieu age <= 35 years high income families, new private owned houses, old families
in outskirts interviewer with high workload, with experience,
with self assessment: amicable, satisfied with own life, not easy flustered
29
Summary (2) Respondent refuse more likely:
Sinus Milieu: new middle class, experimentalists, modern performer
age > 45 – 50 years families with children cities, simple urban estate interviewer with
low workload, with less experience, high level education, age < 40 & male with self assessment: not amicable, unsatisfied with own life, patient,
not reserved
30
Summary (3)
Respondent is difficult to contact:
Sinus Milieu: experimentalists, modern performer age > 45 – 50 & age > 55 – 60 years single household cities, simple urban estate, areas with high freq. of moves interviewer with
high level education, with self assessment: not creative, not reserved
31
Summary (4) Respondent use “not able to participate” :
Sinus Milieu: upper conservative, traditionalists, new middle class, modern performer
smaller than cities areas with higher frequency of moves interviewer with
male low workload, with self assessment: not communicative, unsatisfied with own life,
sluggish, not inquisitive, easy flustered, patient, not reserved with higher need of social approval
Result does not indicate illness of respondents as expected, but that it may be an alibi used by respondents to avoid participation
32
Conclusion
Microgeographic data, interviewer data as well as interviewer questionnaires are an important source to fill the information gap on respondents and nonrespondents.
Next step of analyses: interaction terms between respondent, interviewer and
area Multilevel Poisson Regressions for the number of