Conferences

In recent years, it has become more common to ask survey respondents questions about characteristics of their peer group. The use of these so-called proxy questions increased due to the progress in ego-centric network analysis where the influence of a respondent’s peer group is investigated. However, validation studies have shown that answers to proxy questions are very often incorrect. Even though the use of proxy question has become a standard approach, the reasons for respondents’ mistakes have hardly been understood.

In this study, we develop hypotheses about the conditions under which incorrect answers to proxy questions are given based on Krosnick’s theory of survey satisficing (Krosnick 1991, 1999). According to this theory, respondents can derive answers in different modes of information processing, between which the quality of answers varies considerably. The response mode is determined by the respondents’ motivation, their (cognitive) abilities and the task difficulty. We empirically show that those determinants indeed explain which respondents give incorrect proxy answers. Furthermore, in contradiction to a common hypothesis, we show that the determinants do not moderate each other’s effects but rather are independent. This can have important consequences for research with proxy questions since it shows that it is not necessary for researchers to improve all three determinants in order to receive correct answers from respondents.

We test our hypotheses in a unique setting of an ego-centric network study in which proxy answers of ego about the alters’ characteristics have been validated by interviewing these alters. Assuming that self-reports of the peer group represent the correct answers to the proxy questions, we show that satisficing theory correctly predicts the likelihood of correct survey responses. Respondents that are highly motivated (operationalized as personal relevance of the survey topic and importance of accurate answers), those with high abilities (closeness to alter, duration of relationship with alter, and discussion of the survey topic with alter), and those with high cognitive skills (highly educated and had to spend less effort to answer a survey) are more likely to give correct answers, while an increase in the task difficulty (questions about alters’ socio-demographic characteristics vs. questions about their attitudes) decreases this likelihood. Furthermore, insignificant interaction effects suggest that the determinants do not moderate each other’s effect. Instead, the determinants seem to be additively connected and can substitute each other. An additive index of the determinants is indeed positively related to the likelihood of correct proxy answers and negatively related to the use of the “don’t know” option. We discuss to what degree the results can be translated to response generating processes in general.