Memory Gaps in the American Time Use Survey. Are Respondents Forgetful or is There More to it?

Abstract

Unaccounted respondent memory gaps – i.e., those activity gaps that are attributed by inter- viewers to respondents’ memory failure – have serious implications for data quality. We con- tribute to the existing literature by investigating interviewing dynamics using paradata, distin- guishing temporary memory gaps, which can be resolved during the interview, from enduring memory gaps, which cannot be resolved. We investigate factors that are associated with both kinds of memory gaps and how different response strategies are associated with data quality. We investigate two hypotheses that are associated with temporary and enduring memory gaps. The motivated cuing hypothesis posits that respondents who display more behaviors related to the presence and use of retrieval cues throughout the survey will resolve temporary memory gaps more successfully compared to respondents displaying fewer such behaviors. This should result in overall lower levels of enduring memory gaps. The lack of effort hypothesis suggests that respondent who are less eager to participate in the survey will expend less cognitive effort to resolve temporary memory gaps compared to more motivated respondents. This should then result in a positive association with enduring memory gaps and no association with temporary memory gaps. Using survey and paradata from the 2010 ATUS, our analyses indicate that, as hypothesized, behaviors indicating the use of retrieval cues are positively associated with temporary memory gaps and negatively associated with enduring memory gaps. Motivated respondents experiencing memory difficulties overcome what otherwise would result in en- during memory gaps more successfully compared to other respondents. Indicators of lack of effort, such as whether or not the respondent initially refused to participate in the survey, are positively associated with enduring memory gaps suggesting that reluctant respondents do not resolve memory gaps. The paper concludes with a discussion of implications for survey research.