Filtern
Anzeige der Dokumente 1-4 von 4
Questions on earnings are counted among sensitive topics that often produce high rates of item nonresponse or measurement error. Both types of bias are well documented in the literature and are found to concentrate in the tails of the earnings distribution. In this paper, we explore whether measurement error on earnings could be explained by socially desirable
reporting and whether the error is impacted by interviewer characteristics. Using the linked dataset NEPS-SC6-ADIAB, which contains survey data from the German National Educational Panel Study, Starting Cohort "Adults", linked to administrative earnings records from the German Federal Employment Agency, we analyze the extents of over- and underreporting and the influence of respondent and interviewer characteristics on these behaviors for different quartiles of the earnings distribution. Our results show that the average level of misreporting is relatively low (approximately 6% of median earnings). Our main logistic model reveals that female and more highly educated respondents report significantly more accurately while those with higher earnings misreport to a significantly greater extent. Regarding the impact of personality traits on reporting accuracy, we find significant positive effects for more agreeable respondents and significant negative effects for extraverted respondents. When differentiating by the direction of misreporting, we find, for instance, that women are less likely to overreport across all earnings quartiles. However, the influence of interviewer characteristics is negligible.When it comes to sensitive questions, data is often affected by bias due to non-response or effects of social desirability. Several methods have been introduced to eliminate answer bias by using randomization processes and probabilistic theory to obscure the respondent’s answer and create anonymity, thus facilitating honest answers. The probably most traditional method is the Randomized Response Technique by Warner (1965). However, this method is loaded with certain disadvantages. Therefore, in the last decade, newer methods were introduced that aim at balancing the disadvantages and weaknesses of previous methods, for instance, the non-randomized models Crosswise Model and Triangular Model (Yu et al. 2008) as well as the Parallel Model (Tian 2014). Although especially the Triangular Model is easy to implement in a study, there is only little empirical evidence on its application in different survey modes and populations. Further, it is to assume that certain questions are not equally sensitive for everybody due to specific personal characteristics. Thus, indirect questioning might not be effective in general but only for certain populations. The present study extends prior work on the Triangular Model by evaluating it for different subgroups. The conducted experiment asks for sensitive characteristics in the context of mental stress among students. The Triangular Model achieves significantly higher percentages than conventional direct questioning for illegal drug use among persons that answer socially desirable according to the characteristic of Self-Deception. For the other analyzed subgroups (Impression Management, gender, and depressiveness), the Triangular Model could not achieve higher prevalence rates compared to direct questioning on a sufficient probability level. But still, hard evidence on the effectiveness of indirect questioning models is thin and further critical discussion is needed.
Several repeated cross-national surveys include measurements of attitudes toward gender roles to investigate individuals’ beliefs regarding the appropriateness of men and women’s roles in a particular context. When used to compare attitudes across countries, these measurements reveal critical factors that could cause a lack of equivalence between different cultural contexts, and that could therefore produce misleading results. Nevertheless, the use of such measures to compare country means without assessing measurement equivalence is common. It should also be considered that the assessment of equivalence within a large-scale sample from cross-sectional surveys through multigroup confirmatory factor analysis (MGCFA) often fails because of the strict requirements necessary. The current article is used to assess the measurement equivalence of the gender role attitudes scale included in the last wave of the World Values Survey in 59 countries, with the main goal of identifying the most invariant model for the largest number of groups. The study involved comparing two methods belonging to the frequentist approach: MGCFA and the frequentist alignment procedure, a highly novel and promising method that is still rarely used. Using the first technique, partial scalar invariance was achieved for 27 countries. By employing the frequentist alignment optimization, an acceptable degree of noninvariance was achieved for 35 countries. Thus, the study confirmed the frequentist alignment procedure as a viable alternative to the MGCFA."This article investigates the impact of several sources of method bias on the cross-cultural comparison of attitudes towards gender roles and family ties among non-Western minority ethnic groups. In particular, it investigates how interviewer effects, the use of an interviewer with a shared ethnic background, interview language, interviewer gender, gender matching, the presence of others during the interview and differences in socio-demographic sample composition of non-Western minority ethnic groups affect the cross-cultural comparison of attitudes towards gender roles and family ties between these groups. The data used in this study come from a large scale face-to face survey conducted among the four largest non-Western minority ethnic groups in The Netherlands for which Statistics Netherlands drew a random sample of named individuals from each of the four largest non-Western minority populations living in The Netherlands. Furthermore, methods are introduced to estimate the potential impact of method bias on cross cultural comparisons. The results show that measurement of both gender roles and family ties constructs are full scalar invariant across the different ethnic groups, but that observed differences in attitudes between ethnic groups especially towards gender roles are influenced by method bias. This in turn leads to biased comparisons between ethnic groups because of differences in the size of the various sources of method bias, the differential impact of the same method bias between ethnic groups and the combination thereof." (author's abstract)