To determine relationships between measures both correlational and principal components analysis were performed. Results of the correlational analysis are represented in Table 7.
Note. Measures represented in upper case represent purported measures of implicit knowledge, and those in lower case represent measures of explicit beliefs. Measures represented in italics are competence-related measures; P(S) = rejection of influence (behaviour); own = own ability relative to self (expectations). OVM = OthsVsM; OVF = OthVsF; CNW = Stereotype-consistent response latencies in the WrdTask; INCW = Stereotype-inconsistent response latencies in the WrdTask; COR = Correct response rate in the WrdTask; yvm =YrVwsM; yvf = YrVwsF.
* p < 0.05. ** p < 0.01.
Assessing these results (Table 7) it is interesting to note that the SentComp, a proposed measure of implicit knowledge, significantly correlated with almost all of the explicit beliefs measures. It was also interesting to note that the rejection of influence measure did not significantly correlate with any of the stereotyping measures. This will be further examined in the analysis of the second set of hypotheses. Overall, these results suggest that while measures of explicit beliefs were largely intercorrelated, implicit knowledge measures were correlated to a much lower degree.
Competence-related measures, while it interesting to note their lack of correlation with stereotyping measure, will be discussed later in this paper.
Test of hypotheses. To test the predictions that the set of implicit knowledge measures would not be related to the set of explicit beliefs measures, and that measures would be related within each set, a principal components analysis was conducted. In doing this it was recognised that the solution may have been inaccurate given the small number of cases included.
Prior to the analysis, the presence of multivariate outliers and the factorability of R was assessed. On the basis of Mahalanobis distance criterion, no outliers were detected, and further, R was deemed factorable as several significant correlations were detected among the set of measures (Table 7). Measures that were found to be highly correlated in the correlational analysis (Table 7) were collapsed to form one variable. On this basis the decision to collapse the ASI and the WSQ was made (r = 0.58, p< 0.01). This variable was labelled ASIWSQ.
As the main objective was to determine whether the nine measures of stereotyping would reduce to two constructs, a principal components extraction with varimax rotation was employed (Tabachnick & Fidell, 1996). In specifying a two-factor solution, the resultant eigenvalues were 2.53 for the first factor (accounting for 28% of the variance), and for the second factor 1.73 (accounting for 19% of the variance). Hence a total of 47% of the variance was explained by the two factors. Refer to Table 8 for the factor loadings of each index of stereotyping.
|Measures of Stereotyping||Factor One||Factor Two|
|Consistent pairs (WrdTask)||0.43||0.10|
|Inconsistent pairs (WrdTask)||0.84||0.17|
|Correct response rate (WrdTask)||0.31||0.58|
Note. Measures loading onto Factor One are represented above the line, measures loading onto Factor Two are below the line.
The measures loading onto Factor One (Table 8) did not have uniformly high factor loadings, however, they were well defined by the factor: all measures were those purporting to tap implicit knowledge. The high factor loadings in Factor Two suggest that it was internally consistent and well defined by the measures. However, the measures were not well defined by the factor. This factor comprised two proposed measures of implicit knowledge (correct response rate in the WrdTask and SentComp) together with all measures of explicit beliefs.
These findings can thus be seen to lend partial support for the predictions. As all measures of explicit beliefs loaded onto the same factor, it supports the prediction of relatedness within this set. While not all implicit knowledge measures formed their own factor, those that did formed a separate factor from the explicit beliefs measures. As such, it partially supports the predictions that implicit knowledge measures would be interrelated, and that implicit knowledge measures would not be related to those measuring explicit beliefs. The implicit knowledge measures that loaded onto the explicit beliefs' factor suggests that these measures may have, in fact, been tapping explicit beliefs.