Analysis of response latencies. In order to obtain indices that indicate response facilitation, differences in the latency of response as a function of the individual items were eliminated. In this analysis correct responses only were assessed. Employing a procedure used by Wittenbrink et al. (1997), the response latency for a particular trait, following one or the other gendered category labels (e.g., man-aggressive) was subtracted from the response latency for the same item, following the neutral category label (e.g., cloud-aggressive). In this the response latency of a trait following the neutral category was treated as a standard response latency. Using this standard response latency, this procedure was repeated for each trait in each of the four variables that involved the pairing of gendered category labels (man, woman) with stereotype-related traits (male-, female-related). In these four variables greater response facilitation was indicated by the magnitude of positive differences.
In order to determine whether response latencies were facilitated by the stereotype consistent relationship between category-trait pairings, and thereby to ascertain whether implicit stereotypic knowledge was available, the items in the two variables representing consistent pairings (Man-male trait, Woman-female trait) were first averaged, and the items in the two variables representing inconsistent pairings (Man-Female trait, Woman-male trait) were also averaged. The response latencies in the consistent pairings were then compared with those obtained in inconsistent pairings using a 2 x 2 repeated measures ANOVA, with pairing type (consistent, inconsistent) and type of category label (man, woman) the two repeated measures.
After the homogeneity of variance assumption was met, it was evident that while there was no difference in response to gendered primes, stereotype consistent pairings did facilitate shorter response latencies than inconsistent pairings (F(1,40) = 36.33, p < 0.001). As both of these variables (consistent and inconsistent pairings) contained only correct responses to stereotypic category-trait pairs, each was included as a measure of implicit knowledge in subsequent analyses.
Analysis of Error Rates. In order to determine whether stereotype consistent pairings produced fewer errors than inconsistent pairings, error rates across the two stereotype consistent pairing conditions were averaged, and were also averaged across the two inconsistent pairing conditions. A paired samples t-test was then conducted between these two variables after the equality of variances' assumption was checked. While no statistical difference between these two variables was found, there was a trend in the data indicating that stereotype consistent pairings led to fewer errors than inconsistent pairings. For the purposes of subsequent analyses, the error rates were then averaged across these two variables. In order to obtain an index in line with other measures (i.e., the higher the level, the greater the stereotyping), these error rates were then inverted to represent correct response rates, an index indicating participants' agreement with the stereotype. Hereafter this measure of implicit knowledge is referred to as correct response rates.