Sex Differences in Episodic Memory Variance

Abstract

Men as a group have been shown to have larger variances than women in several areas pertaining to both biological and psychological traits, but no investigation has been performed in regard to episodic memory. We conducted an analysis on sex differences in episodic memory variance on 535 studies, representing 962,946 individuals, conducted between 1973 and 2013. Results showed that men had larger variances than women in verbal episodic memory tasks as well as episodic memory tasks having to do with spatial locations. Women, on the other hand, had larger variance than men for tasks involving remembering routes. These effects were for the most part small, and exploratory analyses suggest that they might come about, at least in part, because of measures not sufficiently controlled for ceiling effects. This means that the effects should be interpreted with caution and that further research on sex differences in episodic memory variance is needed.

Keywords: ceiling effect; episodic memory; mean difference; meta-analysis; sex differences; variance.

Figure 1

A flowchart depicting the data collection phase.

Figure 2

Forest plot of estimated effect sizes from a meta-analysis with lnVR as the dependent variable and material category as a moderator. Each row indicates whether the effect size of that specific level of the material category moderator is reliably different from 0. Estimate for Total is based on a meta-analysis using no moderators. lnVR = lnVR; 95% CI = the 95% confidence interval of the estimate; k = number of studies; p = the p-value; I² = statistics denoting the percentage of variation across studies that is due to heterogeneity rather than due to chance.

Figure 3

Scatterplot of lnVR and Hedge's g for the full dataset. The diameter of each data point is equal to the inverse of its squared variance. The line shows the best-fitting regression (see Table 5).

Figure 4

Assumed distributions of male and female performance for Verbal, Locations, and Routes. Distributions are based on the estimates of Hedge's g (see Table 1) and lnVR (see Figure 2). Only individual material categories where both estimates were significant different from zero in their respective main analyses have been plotted.

Figure 5

Funnel plot for the full dataset with lnVR on the x-axis and standard error on the y-axis. A meta-regression with lnVR as the dependent variable and sample size as a moderator showed no relationship between the two (Intercept: 0.017 [0.002, 0.032], p=.02; Sample size: 0.000 [0.000, 0.000], p=.30). A meta-regression with lnVR as the dependent variable and logged (with the natural logarithm) sample size as a moderator showed no relationship between the two either (Intercept: 0.026 [−0.017, 0.060], p=.28; Logged sample size: −0.001 [−0.008, 0.006], p=.78).