Who Responds to Voluntary Cognitive Tests in Household Surveys? The Role of Labour Market Status, Respondent Confidence, Motivation and a Culture of Learning in South Africa

Both South Africa’s labour market and education system were directly influenced by the separate development policies of the apartheid regime. To this day, great inequalities persist in both domains. South Africa’s performance in standardized international test scores (such as TIMMS) is poor even relative to most developing countries. Furthermore, the better quality of outcomes in former white schools still leaves learners from former black schools at a disadvantage that feeds through to severe labour market inequalities. This study is the first in a series of papers that attempts to understand the role of school quality on labour market outcomes. Here we scrutinize the measurement of numeracy test scores in the National Income Dynamics Survey (NIDS) of 2008, particularly in light of potential sample selection issues. While this survey measures standard welfare and labour market indicators, it is one of the first in South Africa to also ask respondents to complete a concurrent numeracy test. Response rates on this module were particularly low, given that the test was taken on a voluntary basis. We develop a basic empirical model to understand who is likely to take the test. We postulate that discouraged workers’ low propensity to take the test is correlated with their reduced motivation to undertake job search, that the searching unemployed are highly motivated to take the test (as they wish to gauge their ability or practice assessments while embarking on the job search process), the poorest among the self-employed face severe time opportunity costs (as their low incomes are less secure than those of salaried workers) and the richest amongst the employed exhibit an income effect (in that the time opportunity costs of their high incomes reduce their willingness to respond to the numeracy test). Furthermore, locational effects suggest that those residing in geographical “points of entry” into the labour market are also more likely to take the test. The young (who are still in education) and the most educated (in the whole population) also tend to answer the test more readily. The latter observations indicate that some form of confidence in respondents’ own abilities drives their response patterns. To explain these observed features, we construct composite indices of motivation/emotional well-being and individuals’ confidence in their writing abilities using multiple correspondence analysis. While each of these psychological and behavioural factors is a strong predictor of test response, they do not entirely eliminate the independent contributions of each of the observed influences mentioned above. Coefficient magnitudes of each of the sociodemographic variables are, however, reduced, indicating that the particular behavioural influences introduced in later models tell some of the story. Additional uncaptured behavioural and motivational factors are therefore investigated. Firstly, we investigate the role of survey fatigue (by controlling for the time it took to complete the survey before the test was administered), which plays an important role in the black and coloured subpopulations. It furthermore explains why the wealthiest amongst the formally employed are less likely to complete the numeracy test. However, surprisingly, “pseudoaltruistic” effects appear amongst the (wealthier) white population, in that the longer the duration of the preceding questions, the more likely they are to care about answering the test. However, this result cannot be generalized to the whole white population, as response rates were very low among this group. Secondly, (household) peer effects are strong throughout the population, suggesting that a culture of learning is pivotal in understanding response patterns. The results of this paper suggest that broad sociodemographic and labour market features remain important determinants of test response, even after controlling for behavioural features. This suggests that subsequent labour market work must take these drivers into account to avoid the risk of sample selection bias.