Exploring Attrition Bias: The Case of the Khayelitsha Panel Study (2000‐2004)

Abstract

Attrition bias is a problem for users of panel data. Researchers need to know what socio-economic factors are associated with attrition, and whether this is of relevance for the kind of analysis they want to conduct. This paper discusses attrition bias in the 2000/2004 Khayelitsha panel study. It shows that women, shack-dwellers and people living in smaller households are more likely to attrit, but that the impact of these variables on the probability of attrition is relatively small. The implications for labour-market analysis are then explored using Mincerian earnings functions and a probit regression on whether respondents are wage-employed or not. The coefficients generated using a restricted sample of non-attritors do not differ significantly from those generated by the entire sample. This suggests that attrition bias in this particular data set is not a problem for this kind of labour market analysis. JEL Classifiction: O12, O55

How long does victimization foster fear of crime? A longitudinal study

“…Indeed, to participate in their research, people had to answer questionnaires analyzing victimization and fear of crime each week for 30 months. Based on the methodological literature (e.g., Behr, Bellgard, & Rendtel, 2005;Magruder & Nattrass, 2006), it is plausible that their repeat victims who agreed to answer such a large number of questionnaires about victimization were people who could effectively cope with it. Thus, Winkel and colleagues may have underestimated the effects of repeat victimization on fear of crime.…”

An Evaluation of the Determinants and Implications of Panel Attrition in the National Income Dynamics Survey (2008‐2010)

“…Comparisons of attrition rates between panel surveys are often tricky: the figures above are at the individual level, but the definitions for household level non-response are often inconsistent between surveys (Lee, 2003). Magruder and Nattrass (2006) provide a labour market-orientated analysis of attrition in the KPS, finding that socio-economically upward migration is less evident than other Sub-Saharan African studies predict, and main predictors of attrition as follows: lower household and individual income, living in a shack, recent arrival in the household and being single. Despite evidence of selective attrition, very little evidence of bias inherent in labour market analysis is found.…”

“…However, the threat of attrition to panel samples, defined as a failure to successfully reinterview targeted households or individuals (Kasirye and Ssewanyana, 2010) can reduce the value of panel data. Because of migration, mortality or non-response between waves, attrition can critically undermine the validity of research undertaken using panel data (Magruder and Nattrass, 2006).…”

“…Testing includes bivariate and multivariate examinations of the selective attrition dynamics in the NIDS. The structural equations considered include a canonical Mincerian earnings function, following Magruder and Nattrass' (2006) investigation into attrition in the Khayelitsha Panel Study (KPS), and the adaptation of Wittenberg's (2011) model estimating the correlates of body mass index (BMI) 2 in the NIDS Wave 1.…”

Economists Versus the Street: Comparative Viewpoints on Barriers to Self-Employment in Khayelitsha, South Africa

“…Brief description of Khayelitsha (Magruder and nattrass (2005) After giving their best point estimate for the percent of unemployed residents that would find a particular hindrance to be a large or very large hindrance, individuals were then asked to pick a band saying they were confident the true population value would be within X percentage points of their estimate. This served multiple purposes.…”