Gender differences in research grant allocation -- a mixed picture

TL;DR

This study investigates gender bias in research grant allocation by including merit variables and panel-level analysis, revealing complex patterns of bias that vary across panels and decision stages.

Contribution

It introduces merit variables into gender bias analysis and examines bias at the panel decision level, providing a nuanced understanding of the process.

Findings

Women receive lower scores than men after controlling for merit.

Bias exists against women in initial rejection decisions and varies in final decisions.

Panel-level bias patterns are mixed, influenced by panel composition and stereotyping.

Abstract

Gender bias in grant allocation is a deviation from the principle that scientific merit should guide grant decisions. However, most studies on gender bias in grant allocation focus on gender differences in success rates, without including variables that measure merit. This study has two main contributions. Firstly, it includes several merit variables in the analysis. Secondly, it includes an analysis at the panel level where the selection process takes place, and this enables to study bias more in-depth at the process level. The findings are: (i) After controlling for merit, a consistent pattern of gender bias was found in the scores: women receive significant lower grades than men do. (ii) The scores are an input into the two-step decision-making process, and this study shows bias against women in the first selection decision where 75% of the applications are rejected, and bias in favor of women in the second (final) selection decision. (iii) At the level of individual panels, the analysis shows a mixed pattern of bias: in some panels the odds for women to receive a grant are lower than for men, whereas in other panels we find the opposite, next to panels with gender-neutral decision making. (iv) In the case under study, at an aggregated level the allocation of grants seems balanced. (v) The mixed pattern at panel level seems to relate characteristics such as the panel composition, and the level of gender stereotyping.