A statistical enquiry into the effect of test length on a math gender gap
This study researches whether girls are advantaged (vis-à-vis boys) as math test length increases. The meta-analysis on trends in gender and mathematics performance done by Lindberg, Hyde, Petersen & Linn (2010), was extended with variables on test length to obtain the final dataset for analysis. This contains 435 studies, in which 1,277,598 subjects are included. Proxies for test length are the number of questions and maximum time allowed in the studies. Small negative coefficients for both length measures are found when regressing test lengths on the standardized score difference, in the range of -.0007 to -.0022. The full regression model including test and student background controls and weighted observations estimates a negative coefficient of -.002 (p<0.05) for the number of questions and -.002 (p=0.15) for the maximum amount of minutes on a standardized difference. These findings indicate performance bias may be a threat for tests as an evaluation mechanism. However, further sensitivity analyses yields mixed results. Combined with the explorative character of this thesis the need for further research to test the hypothesis of gender bias with increasing test length is evident.