Assessing and relaxing assumptions in quasi-simplex models

Abstract

Panel data (repeated measures of the same individuals) has become more and more popular in research as it has a number of unique advantages such as enabling researchers to answer questions about individual change and help deal (partially) with the issues linked to causality. But this type of data has some special limitations as well, such as the training effect of respondents and gradual drop-out from the survey (i.e., attrition). In this context an approach that evaluates data quality using reliability (the amount of the true value as opposed to random noise) in panel data has been proposed in previous research. This approach, named the quasi-simplex model, brings a number of innovations but also makes a number of strong assumptions about the data, such as: the absence of memory effects of respondents or equal error over time. This paper aims to assess the validity and impact of these assumptions as these have largely not previously been examined Our research shows that most of the previously made assumptions hold and more often than not the model can be even more restrictive. But, even if this is true, four out of the 22 circumstances analysed here presented violations of an assumption that lead to different results. Our research shows that when processes such as the respondent memory effect are present in the data it can lead to overestimation of reliability and underestimation of stability in time.