Telling mistakes and striking inconsistencies: Numeracy Measures for Risk and Time Elicitation

Abstract of the Paper

The role of numeracy (statistical literacy) in risk and time elicitation is likely to be significant, as ranking options often requires numerical comparisons. Moreover, as numeracy and preference regarding risk and time are latent variables, studying their interplay is challenging. We address one of them, the numeracy measure. Numeracy measures tend to rely on an aggregate score (typically the sum of correct responses). We develop more refined scores. These take into account that wrong responses vary in their gravity and that inconsistencies across responses may occur. Because these scores rely on more information than the mere sum of correct responses, they may serve as better proxies for the latent numeracy level. We use a large data set from rural Malawi as an illustration. This data set contains a (diagnostic) numeracy test, as well as a wide range of time and risk experiments. We find that more refined numeracy measures tend to yield lower p-values for estimates of risk aversion and impatience. These results point towards a potential benefit of more refined numeracy measures.