The power of stereotyping and confirmation bias to overwhelm accurate assessment: the case of economics, gender, and risk aversion

Abstract

Behavioral research has revealed how normal human cognitive processes can tend to lead us astray. But do these affect economic researchers, ourselves? This article explores the consequences of stereotyping and confirmation bias using a sample of published articles from the economics literature on gender and risk aversion. The results demonstrate that the supposedly ‘robust’ claim that ‘women are more risk averse than men’ is far less empirically supported than has been claimed. The questions of how these cognitive biases arise and why they have such power are discussed, and methodological practices that may help to attenuate these biases are outlined.

Keywords::

• stereotyping
• confirmation bias
• gender
• risk aversion
• effect size
• index of similarity

Acknowledgement

This work was supported by the Institute for New Economic Thinking [grant number INO1200012]. Matthew P. H. Taylor, Sophia Makemson, and Mayara Fontes supplied expert research assistance.

Notes

^1. While inclusion of intersex, transgendered, or transsexual subjects would clearly complicate – and enrich – this analysis, this study focuses on the economics literature, in which a sex binary is assumed.

^2. The scholarly literature supporting these claims is discussed in the section on ‘How Biases Arise and Persist,’ below.

^3. Nelson (2013) discusses an additional notable case in which such testing is neglected.

^4. In the literature reviewed here, Eckel and Grossman (2008, p. 15) and Dohmen et al. (2011, pp. 530, 540) are notable for providing any extended discussion of substantive economic significance.

^5. This is most often estimated as:

where s_m, s_f, n_m, and n_f are the standard deviations and sample sizes for the male and female samples.

^6. Hyde found that 78% of reported empirical sex differences were smaller than this value.

^7. A 10% level was chosen, rather than 5% or 1%, to give the existence of ‘difference’ the maximum benefit of doubt. Numeric values for d (or IS) were not calculated when differences were not statistically significant, because of the rather wild values that occurred in some of the small samples.

^8. Because IS is non-directional, it is worth noting that one instance of IS = 0.67 (Beckmann & Menkhoff, 2008) is for a case where fewer men than women chose a risky option.

^9. Eriksson has since acknowledged the error in the statement made in his co-authored 2010 article (2012).

^10. The formula for the standard error of Cohen's d is from Cooper and Hedges (1994).

^11. The risk-taking measure is created by examining the proportion of men and women whose portfolios were mostly stock. One measure is statistically significant at the 5% level, the other is so close to zero (d = 0.02) that it is not.

^12. It may be objected that risks of pregnancy and childbirth are not relevant for comparisons between the sexes, since men do not participate in them. Anecdotally, at least, however, it seems that there is little reticence about making inferences about risk taking from men-only activities. Flying fighter planes, for example, or engaging in combat is commonly taken as indicating masculine bravery, even though institutional constraints have largely prevented women from engaging in these activities.

^13. A study of preferred sex-stereotyped activities (e.g., watching boxing, scrapbooking), also showed a taxonic structure (p. 391), but this was in good part by construction (p. 393). All subjects in this study were self-identified heterosexual men and women.