Gross Income Non-Response in the German Socio-Economic Panel – Refusal or Don't Know?

Schräpler, Jörg-Peter Article Gross Income Non-Response in the German SocioEconomic Panel – Refusal or Don't Know? Schmollers Jahrbuch – Zeitschrift für Wirtschaftsund Sozialwissenschaften. Journal of Applied Social Science Studies Provided in Cooperation with: Duncker & Humblot, Berlin Suggested Citation: Schräpler, Jörg-Peter (2003) : Gross Income Non-Response in the German Socio-Economic Panel – Refusal or Don't Know?, Schmollers Jahrbuch – Zeitschrift für Wirtschaftsund Sozialwissenschaften. Journal of Applied Social Science Studies, ISSN 1865-5742, Duncker & Humblot, Berlin, Vol. 123, Iss. 1, pp. 109-124, https://doi.org/10.3790/schm.123.1.109 This Version is available at: https://hdl.handle.net/10419/292042 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by/4.0/ Gross Income Non-Response in the German Socio-Economic Panel – Refusal or Don’t Know?* By Jo ¨rg-Peter Schra ¨pler Abstract Many validation studies deal with item non-response and measurement error in earnings data. In this paper we explore motives of respondents who fail to reveal earnings using the German Socio-Economic Panel (GSOEP). The GSOEP collects socio-economic information from private households in the Federal Republic of Germany. We explain the evolution of income non-response in the GSOEP and demonstrate the importance of distinguishing between a refusal to respond and a response of \don’t know”. JEL Classification: C 81 1. Introduction Since the survey interview is a major source of research data in the social sciences, it is not surprising that there is a large literature on the quality of survey data focused specifically on respondents and interviewer effects. This paper focuses on item non-response and unit non-response in the German Socio-Economic Panel (GSOEP). We describe the existing literature and try to explain on the basis of a cognitive and rational choice theory why respondents fail to state their income. The GSOEP is a longitudinal representative survey containing socio-economic information on private households in the Federal Republic of Germany (Wagner /Burkhauser/Behringer 1993; Dunn/Frick/ Witte 1998). 1 Schmollers Jahrbuch 123 (2003) 1 Schmollers Jahrbuch 123 (2003), 109–124 Duncker & Humblot, Berlin * In particular I would like to thank Martin David (JPSM) for assistance and valuable suggestions as well as Gert G. Wagner (DIW Berlin, Berlin University of Technology, TUB) and Martin Spiess (DIW Berlin) and Cheti Nicoletti (Tor Vergata University, Rome, Italy) for helpful comments. The usual disclaimer applies. 1The GSOEP provides in the public use version also a wealth of methodological information about the survey methods utilized and the characteristics of the interviewers (c.f. Schrapler/Wagner 2001). OPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.123.1.109 | Generated on 2023-04-04 12:31:05 110 Jo ¨rg-Peter Schra ¨pler 2. Item Non-Response – explanations We have learned much about earnings and their determinants from data collected in income and labor market surveys such as the German Socio-Economic Panel (GSOEP), the Current Population Survey (CPS) and the Panel Study on Income Dynamics (PSID). Unfortunately, the quality of data in these surveys is undermined by the failure of some participants to report their wages and salaries. Several reasons for item non-response are mentioned in the literature. Loosveldt/Pickery /Billet (1999) assume that item non-response occurs when the answering process fails to proceed smoothly because the respondent lacks motivation or ability. Item non-response also depends on the evaluation by the respondents of the questions asked: the questions are too dificult, not interesting, too embarrassing or too threatening (Loosveldt et al. 1999). Furthermore, Burton/Laurie/Moon (1999) express the idea that one can place all potential respondents to a survey on a cooperation continuum. At one end are those who will always take part and will answer any question, on the other end are those respondents who are hard to persuade and will tend to refuse often. In rational choice theory the respondent is a decision maker who maximizes her utility. The respondent faces four alternatives which are related to the cooperation continuum: she participates in the survey and answers the income question truthfully; she participates, but reports false, more socially desirable facts; she participates and refuses to answer or has retrieval problems and does not know the answer; she refuses to participate. Although false reporting can not be ruled out, detecting false reporting requires a true reference value (see Bollinger/David 2001), which is usually not available. We assume serious false reporting is dificult to carry out, because in economic-oriented panel studies like the GSOEP, many variables are related to income and therefore have also to be adjusted by the respondent. Moreover, for consistency, in each wave the false values have to be remembered by the respondent for subsequent waves. Hence, we restrict our analysis to alternatives one, two, four and five. Table 1 relates to the four alternatives to underlying cognitive problems and respondents assessment and evaluation. The table distinguishes two origins for non-response. Cognitive Limitations: Unit or item non-response may occur because a respondent cannot participate in the survey, or finds a question too dificult to answer. There are several distinct stages in the cognitive work necessary to answer an income question, like understanding, retrieval, and response production. Response errors can occur at any stage (c.f. Moore/Stinson/Welniak 2000). Some respondents cannot recall their gross income. This is clearly a Schmollers Jahrbuch 123 (2003) 1 OPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.123.1.109 | Generated on 2023-04-04 12:31:05 Gross Income Non-Response in the German Socio-Economic Panel 111 case of the “don’t know” response category. Our first hypothesis is that respondents who don’t answer a particular question because of cognitive limitations differ from respondents who are not willing to cooperate. Table 1 Respondent behavior alternatives relating to cognitive or other limitations and respondent’s assessment and evaluation 1. Cognitive and other limitations 2. Respondent’s assessment and evaluation benefits costs participation  not possible (moved, sick, dead)  survey serves a meaning ful purpose  to endorse a scientific or public sponsor  interesting entertainment  topic too sensitive  topic out of interest  survey not confidential  opportunity costs response/participation  cognitive limitations, “don’t know” (understanding, retrieval, response production)  doesn’t apply  question serves a meaningful purpose  to endorse a scientific or public sponsor  interesting entertainment  loss due to disclosure to third parties  loss due to social (un)desirability  loss due to invade of privacy item-nonresponse/ participation  no fear of social disapproval  keep particular information private  keep particular information confidential  justification costs (depends on data collection method) unit-nonresponse  time for other opportunities  keep privacy  keep confidentiality  justification costs (depends on data collection method) Source: own arrangement Rather than being unable to respond, however, respondents may choose not to cooperate. This decision will be based on their own assessment and evaluation of the interview situation. Income questions belong to the category of sensitive questions. Their content pertains to personal and intimate information. Sensitive questions have several aspects. Tourangeau et al. (2000) distinguish Schmollers Jahrbuch 123 (2003) 1 OPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.123.1.109 | Generated on 2023-04-04 12:31:05 112 Jo ¨rg-Peter Schra ¨pler three of them: risk of disclosure of answers to third parties; social (un)desirability of the answers; and invasion of privacy. Disclosure to third parties and confidence building: In rational choice theory we can interpret concerns about confidentiality as fear of loss through disclosure to third parties. Individuals appear to vary in the degree of their concern about confidentiality, and this could have an effect on their willingness to cooperate (Singer/Mathiowetz/Couper 1993). As the GSOEP is a panel study, the respondent meets the interviewer several times. When they first meet, the tendency to refuse may be stronger than on later occasions, because successful contacts build confidence 2 . Therefore our second hypothesis is that the refusal rate is highest in the first wave and decreases in subsequent waves. Social desirability: Social desirability has two aspects. First, the respondent may wish to have the interviewer’s approval. Approval depends on the presence of an interviewer, the topic of the question and the facts about the respondent’s conduct or attitudes. Second, personality traits may cause respondents to distort their answers because of underlying needs, such as the need for social approval or the need to confirm to social standards (c.f. Tourangeau et al. 2000, p. 257–258). Fear of disapproval of low earnings by the more accomplished interviewer may create an incentive for low-wage respondents in need of social approval to refuse to give their earnings (c.f. Smith 1991; Wagner/Motel 1996). Our third hypothesis is that lowearning respondents have a higher refusal rate than high-wage respondents due to the influence of social desirability. Breach of Privacy: Some non-respondents may have a general aversion to answering intimate questions and think income questions are a breach of privacy. These respondents are identified by less cooperation and have often several refusals in their questionnaire. We can assume that some members of this group are not whole-heartedly cooperative, and have misgivings about the whole process. If their fears are confirmed, they cooperate less. They drift to the end of the cooperation continuum and are harder to persuade in the following wave (Burton et al. 1999; Bollinger/David 2001). Our fourth hypothesis is that respondents who refuse to divulge their income are more likely to drop out of the survey than others. We expect a negative correlation between refusing the income question and survey participation in the following wave. Justification costs: The mode of the interview may also contribute to the difficulty respondents and interviewers experience when talking about income. Groves (1989) and Jordan et al. (1980) find more missing values for income in telephone surveys than in face-to-face interviews. Moore et al. (2000) argue that the telephone may simply lower some of the social barriers Schmollers Jahrbuch 123 (2003) 1 2A change of interviewer is the exception in the GSOEP. Rendtel (1995) has shown that a change of interviewer is a strong predictor for unit non-response. OPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.123.1.109 | Generated on 2023-04-04 12:31:05 Gross Income Non-Response in the German Socio-Economic Panel 113 against expressing the discomfort about discussing income overtly. Therefore, we suspect the greater the social barriers the higher the justification costs for refusals. These costs may vary with the interview situation. In mail or selfcompleted questionnaires or in telephone interviews it is much easier for the respondent to refuse an answer than in face-to-face interviews where the interviewer asks the respondent directly. We hypothesize that uncooperative respondents choose self-completion or mail modes of response with higher probability then cooperative respondents. Hence, our fifth hypothesis is that respondents who choose self completion modes have higher refusal rates than in face-to-face modes. 3. Itemand unit non-response in the GSOEP As a first step, we take a look at the history of the item non-response rate for the gross income question among employed persons in the GSOEP in Sample A (West German residents who are not members of one of the main immigrant groups) 3 . The rates are a result of blank, don’t know and refused entries. Special problems occur in the case of the income question for self-employed respondents. The income question is not designed to elicit estimates of business profits for the monthly reference period. Due to these problems, self-employed respondents, trainees and proxies are excluded from our empirical analysis. Table 2 shows a decrease in item non-response fromabout 10 percent in the first two waves to approximately 6 percent after 12 years. The gross earnings item non-response rate is quite low in comparison to other international and national surveys (Madow et al. 1983, p. 24). Schupp/Wagner (1996) show that item-nonresponse for gross income depends on the mode of data collection method for GSOEP. The GSOEP represents a “method-mix”. The preferred method for the survey is face-toface interviews (face). Respondents also may complete the questionnaire by themselves and receive help from the interviewer on demand (self). Sometimes both methods are combined in an individual interview (not shown in the figure). A few responses are elicited by mail when respondents would otherwise fail to cooperate (mail). Figure 1 shows that face-to-face interviews have the lowest income-nonresponse, and mail questionnaires the highest. A similar pattern can be observed for respondents with unitnonresponse in the following wave in figure 2. The mode “mail” can be interpreted as a strong indicator for poor cooperation. Schmollers Jahrbuch 123 (2003) 1 3The annual question for all employed respondents is: “How high were your earnings last month? If you received any additional payments last month, e.g. holiday money or back-pay please do not include these. Also allowance for children do not include. However, do include money earned for overtime.” OPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.123.1.109 | Generated on 2023-04-04 12:31:05 114 Jo ¨rg-Peter Schra ¨pler Table 2 Item Non-Response-rate for the gross income question from employed persons in the GSOEP, Sample A (in per cent) including self-employed and trainees excluding self-employed, trainees and proxies employed respondents missing* % selected respondents missing* % wave 1 5017 640 12.8 4135 437 10.5 wave 2 4709 622 13.2 3845 394 10.2 wave 3 4520 526 11.6 3654 314 8.6 wave 4 4525 499 11.0 3677 294 8.0 wave 5 4309 469 10.9 3512 275 7.8 wave 6 4213 427 10.1 3478 252 7.3 wave 7 4128 446 10.8 3420 281 8.2 wave 8 4160 494 11.9 3462 327 9.4 wave 9 4013 433 10.8 3379 286 8.5 wave 10 3952 371 9.4 3339 227 6.8 wave 11 3796 306 8.1 3225 198 6.1 wave 12 3780 304 8.0 3222 204 6.3 total 51122 5537 10.8 42348 3489 8.2 Source: GSOEP 1984–1995 Sample A, employed respondents (own calculation). * result of blank, “don’t know” or refused entries. Schmollers Jahrbuch 123 (2003) 1 Figure 1: Share of incomenonresponse by data collection methods from 1984–1994, Sample A, employed persons. Figure 2: Share of unitnonresponse by data collection methods in the last wave from 1985–1995, Sample A, employed persons. We have to use another measurement to distinguish refusals from don’t know. An additional item about respondent’s net income follows the gross income question. It is reasonable to assume that respondents who state their net income but not gross income simply do not know their gross income. ResponOPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.123.1.109 | Generated on 2023-04-04 12:31:05 Gross Income Non-Response in the German Socio-Economic Panel 115 dents who state their net income reveal that they are not concerned about the sensitivity of the income question. 46% of all missing values for gross income fall in this category. We assume that employed respondents with missing values for both items have refused to answer (54%) 4 . To show the influence of income on income non-response we classify occupation into three skill groups (table 3). The groups are based on type of position (wage, salary, or civil service) and occupational skills. Figure 3 and 4 display the distributions by skill group. The share for the “don’t know” category is shown in figure 3, while figure 4 shows the refusal rate. The high skilled have the highest refusal rates, followed by the medium and low skilled. The refusal rate declines for all occupational states. A stable high proportion of low skilled don’t know their gross income but report valid net incomes. This finding is evidence against our third hypothesis: the low skilled do not fail to report their income due to social (un)desirability, but rather due to cognitive limitations. It might be that their monthly gross income varies and it is easier for them to recall their take-away income (net income). Table 3 Classification of Skill Groups vocational position occupation LOW hourly-paid worker unskilled worker, semiskilled worker MEDIUM hourly-paid worker skilled worker, foreman, master, salaried employee industryand works foreman, employee with simple activity, skilled activity civil servant minor and lower-grade civil service HIGH salaried employee highly skilled activity, executive function civil servant high and senior service To control for further respondent and interviewer variables, we estimate multilevel regression models. The survey data have a hierarchical structure: the respondents are nested within interviewers, where the respondent level is said to be the lower and the interviewer level the higher level. The appropriate method of analysis is the use of hierarchical or multilevel models that estimate Schmollers Jahrbuch 123 (2003) 1 4This method of distinguishing the two types of response behavior is supported by another survey, the British Household Panel Study (BHPS). The BHPS distinguish between ‘don’t know’ and ‘refusals’. Approx. 80% of the ‘don’t knows’ for gross-income are connected with valid net-income in the BHPS (Schr ¨apler 2002b). OPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.123.1.109 | Generated on 2023-04-04 12:31:05 116 Jo ¨rg-Peter Schra ¨pler Schmollers Jahrbuch 123 (2003) 1 Figure 3: Share of missing grossand valid net income respond by occup. status, 1984–1994, Sample A, empl. respondents. Figure 4: Share of missing grossand net income by occupational status, 1984– 1994, Sample A, employed respondents. both the interviewer variance and the effects of explanatory variables measured at the interviewer and the respondent level (c.f. Hox 1994; Pannekoek 1991; Hill 1991). We first estimate a simple univariate probit model with the dependent variable equal to one if the individual provides no income data, and zero otherwise. Next, using the same sample, we estimate a multivariate probit model: three probits for the presumed refusal to give income information, the presumed response ’don’t know’ and unit response (participation) in the next wave, respectively with correlated error structures. The errors are correlated at the period, respondent and interviewer level, reflecting the hierarchical nesting of the data. Level 1 represents the different response variables of the component probits of the multivariate model. We define i= 1 (refuse), i= 2 (don’t know) and i= 3 (unit-response). Level 2 represents the different times of measurement j. Level 3 consists of k respondents and level 4 represents the aggregate level, which is formed by linterviewers. For person kand interviewer lthree dichotomous variables yijkl are observed at time j. y1jkl  1 ; if y 3 1jkl > 0 ; refuse 0 ; otherwise   1  y1jkl  1 ; if y 3 2jkl > 0 ; don’t know 0 ; otherwise   2  y1jkl  1 ; if y 3 3jkl > 0 ; unit-response (next wave) 0 ; otherwise   3  We use a multivariate probit model with four levels: OPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.123.1.109 | Generated on 2023-04-04 12:31:05 Gross Income Non-Response in the German Socio-Economic Panel 123 Groves, R. M. (1989), Survey Errors and Survey Costs, New York. Hill, D. H. (1991), Interviewer, Respondent, and Regional Office Effects on Response variance: A Statistical Decomposition, in: P. Biemer/R. Groves/L. Lyberg/N. Nathiowtz/S. Sudman (eds.), Measurement Errors in Surveys, New York, 463–483. Hox, J. J. (1994), Hierarchical Regression Models for Interviewer and Re-spondent Effects, Sociological Methods & Research 22 (3), 300–318. Jordan, L. A./Marcus, A. C./Reeder, L. G. (1980), Response Styles in Telephone and Household Interviewering: A Field Experiment, Public Opinion Quarterly 44, 210– 222. Loosveldt, G./Carton, A./Billet, J. (1999), Evaluation of the relationship between data quality and interviewer behavior during doorstep contact i n face-to-face interviews, Paper presented at the International Conference on Survey Non-response, Portland, Oregon, USA. Loosveldt, G./Pickery, J./Billet, J. (1999), Item non-response as a pre-dictor of unit non-response in a panel survey, Paper presented at the International Conference on Survey Non-response, Portland, Oregon, USA. Madow, W. G./Nisselson, H./Olkin, I. (eds.) (1983), Incomplete Data in Sample Surveys, Volume I-III, New York/London. Moore, J. C./Stinson, L. L./Edward J. Welniak, J. (2000), Income Mea-surement Error in Surveys: A Review, Journal of Official Statistics 16 (4), 331–362. Pannekoek, J. (1991), A Mixed Model for Analyzing Measurement Errors for Dichotomous Variables, in: P. Biemer/R. Groves/L. Lyberg/N. Nathiowtz /S. Sudman (eds.), Measurement Errors in Surveys, New York, 517–531. Rasbash, J./Browne, W. / Goldstein, H./Yang, M./Draper, D./Plewis, I./Healy, M./ Woodhouse, G. (1999), MLwiN User’s guide, Institute of Education, London. Rendtel, U. (1995), Panelausfa ¨lle und Panelrepra ¨sentativita ¨t, Frankfurt am Main/New York. Schra ¨pler, J.-P. (2002), Respondent behavior in Panel Studies – A Case Study for Income-Nonresponse by means of the German Socio-Economic Panel (GSOEP), Discussion Paper No. 299. Deutsches Instiut fu ¨r Wirtschaftsforschung (DIW Berlin). – (2002), Respondent behavior in Panel Studies – A Case Study for Income-Nonresponse by means of the British Household Panel Study (BHPS), unpublished manuscript. Schra ¨pler, J.-P./Wagner, G. G. (2001), Das “Interviewer-Panel” des Sozio-oekonomischen Panels – Darstellung und ausgewa ¨hlte Analysen, Allgemeines Statistisches Archiv 85 (1). – (1996), Maintenance of Long-On-Going Panel Stud-ies – The Case of the German Socio-Economic Panel Study (GSOEP), Paper presented at the Essex ’96 Fourth International Conference on Social Science Methodology, University of Essex, July 1–5. Schmollers Jahrbuch 123 (2003) 1 OPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.123.1.109 | Generated on 2023-04-04 12:31:05 124 Jo ¨rg-Peter Schra ¨pler Singer, E./Mathiowetz, N./Couper, M. (1993), The impact of privacy and con dentiality concers on survey participation: The case of the 1990 U.S. census, Public Opinion Quarterly 57, 465–482. Tourangeau, R. (1984), Cognitive Science and Survey Methods, In: T. Jabine/M. Straf/J. Tanur /R. Tourangeau (eds.), Cognitive Aspects of Survey Methodology: Building a Bridge Between Disciplines, Washington, DC, 73–100. Tourangeau, R./Rips, L. J./Rasinski, K. (2000), The Psychology of Survey Response, Cambridge/New York. Wagner, G. G./Burkhauser, R. V./Behringer, F. (1993), The English Language Public Use File of the German Socio-Economic Panel, Journal of Human Resources 28 (2), 413–415. Wagner, M./Motel, A. (1996), Die Qualita ¨t der Einkommensmessung bei alten Menschen, Ko ¨lner Zeitschrift f ¨ur Soziologie und Sozialpsychologie 48 (3), 493–512. Schmollers Jahrbuch 123 (2003) 1 OPEN ACCESS | Licensed under CC BY 4.0 | https://creativecommons.org/about/cclicenses/ DOI https://doi.org/10.3790/schm.123.1.109 | Generated on 2023-04-04 12:31:05