The present research further examines the psychometric properties of the Motivation to Control Prejudiced Reactions Scale (MCPRS). Particular attention is paid to the replicability of its factor structure and its factorial equivalence across samples of university students from Western Canada (n = 235), Eastern Canada (n = 556) and the mid-Western United States of America (n = 404). Confirmatory factor analysis and invariance analysis were carried out using the Analysis of Moment Structures (AMOS) 7.0. Results showed that while the two-factor structure of the MCPRS was replicated across samples, the original model required refinement to produce acceptable model fit (i.e., each sample had a slightly different model). Partial measurement invariance also was demonstrated for a subset of items on the MCPRS. The implications of the results, in terms of future use of the MCPRS are discussed, and limitations of the current study are outlined.
The importance of motivation in the expression of prejudice has long been established, with an acknowledgment that feelings and beliefs in contemporary modern prejudice may be rooted in motivational biases [
Shelton’s [
The publication of a dual-process theory of behaviour, which proposes the involvement of motivation and opportunity as determinants of prejudice (MODE), suggests that negative attitudes can be controlled if an individual has the opportunity and motivation to do so [
Dunton and Fazio [
The convergent validity of the MCPRS was assessed through its association with scores on the Modern Racism Scale (MRS; [
Dunton and Fazio [
Scores on the CAP subscale were found to be strongly correlated with egalitarianism while scores on the RAD subscale were not [
It should be noted that many studies have not investigated the component or factor structure of the MCPRS (e.g., [
The purpose of the current study is to use confirmatory factor analysis (CFA) to determine whether the predefined factor models fit an observed set of data. This is an appropriate means of assessment given Dunton and Fazio’s (1997) contention that motivation to control prejudiced reactions may be partitioned into concern and restraint dimensions. Additionally, although the scale has been used outside the cultural context in which it was initially validated (e.g., Sweden: [
In CFA, a hypothesised model may be tested for fit and compared with competing models which, in turn, allow for revision and refinement of an instrument [
Participants consisted of 1195 undergraduate university students enrolled in psychology courses in Western Canada (n = 235), Eastern Canada (n = 556) and the mid-Western United States of America (n = 404) who participated in a mass survey. The age range of participants for the Eastern Canadian and American samples was 17 to 48 years (M = 20.06, SD = 2.89) and 17 to 46 years (M = 19.60, SD = 2.42), respectively. No demographic information was collected for the Western Canadian sample.
Motivation to Control Prejudiced Reactions Scale (MCPRS; [
The Analysis of Moment Structures Version 7.0 (AMOS; [
Assuming tests of model fit to be close to the hypothesised model for all groups, multi-group factorial invari- ance will be conducted. Tests for equivalence of item measurements (metric invariance) and theoretical struc- tures (structural invariance) across American and Canadian samples2 will then be carried out [
Cronbach’s α for the MCPRS was in excess of 0.70 for all samples, which suggests acceptable scale score reliability. Alpha values, 95% confidence intervals for alpha, means, and standard deviations for total MCPRS scores are provided in
A one-way analysis of variance revealed a statistically significant difference in MCPRS scores across the three groups, F(2, 1163) = 10.47, p < 0.001. Tukey post hoc tests revealed significant differences between the two Canadian samples (p < 0.001), and the Western Canadian and mid-Western USA samples (p < 0.001). Specifically, the Western Canadian sample evidenced a significantly stronger motivation to control prejudice reactions than the other two groups.
The original two-factor model hypothesised by Dunton and Fazio [
Attempts were made to refine the model. Using estimates of regression weights, items and cross-loadings were deleted if non-significant (p > 0.05). For the Western Canada sample, Item 7 was restricted to loading on the RAD factor, and Item 8 was restricted to loading on the CAP factor, as loadings of both items on the respec- tive alternate factor proved non-significant (p > 0.05). Items 5 and 17 were deleted as these items loaded weakly on both factors (maximal loading < 0.29). The modification indices suggested that the error covariances of Items 3 and 11 should be permitted to correlate. This change was deemed permissible as inspection of these items
Western Canada | Eastern Canada | Mid-Western USA | |
---|---|---|---|
Mean | 53.39 | 50.64 | 50.86 |
SD | 8.43 | 7.64 | 8.02 |
α (95% CI) | 0.78 (0.74 - 0.82) | 0.72 (0.69 - 0.76) | 0.76 (0.72 - 0.79) |
Original Model | |||
---|---|---|---|
Measures of Fit | Western Canada | Eastern Canada | Mid-Western USA |
χ 2 | 242.71* | 395.59* | 307.82* |
Df | 114.00 | 114.00 | 114.00 |
Q | 2.13 | 3.47 | 2.70 |
GFI | 0.89 | 0.92 | 0.91 |
AGFI | 0.85 | 0.89 | 0.88 |
CFI | 0.82 | 0.78 | 0.83 |
RMSEA | 0.07 | 0.07 | 0.07 |
90% CI | 0.058, 0.082 | 0.061, 0.075 | 0.057, 0.074 |
AIC | 320.71 | 473.59 | 385.82 |
Note.*p < 0.001.
revealed that both reflect affective reactions that may ensue when one evidences prejudice against a visible minority.
For the Eastern Canadian sample, Item 7 was restricted to loading on the CAP factor due to its non-significant loading on the RAD (p > 0.05). Items 8 and 17 were deleted due to non-significant loadings on either factor (p > 0.05), and Item 5 was removed due to weak loadings on both factors (0.32, RAD; 0.21, CAP). Item 12 was removed due to modification indices suggesting that its error covariance should be permitted to correlate with several other error terms. Such output suggests item redundancy.
For the mid-Western USA sample, Item 7 was restricted to loading on the RAD factor due to its non-signifi- cant loading on the CAP (p > 0.05). Item 8 was deleted due to not loading significantly on either factor (p > 0.05). Also, due to loading weakly on both factors, Items 5 and 17 were removed (maximal loading < 0.28). Item 6 was removed due to potential item redundancy (i.e., modification indices suggest that its error term should be permitted to correlate with several other error covariances). Given similar item content, correlated error terms were permitted between Items 9 and 13.
The refined two-factor models then were tested. Inspection of
The validity of the refined two-factor model was tested for all three samples simultaneously. A test of partial measurement invariance was conducted (i.e., only elements common to all three models were constrained to be equal). Using a macro level strategy, which requests AMOS to allow different path diagrams for each group, baseline models were tailored to fit each sample. A simultaneous multi-group analysis was conducted, where all parameters were freely estimated. Following this, analysis of a restricted model occurred. Factor covariance, factor variances, and factor loadings common to all three groups were constrained to be equal.
Fit statistics for the unconstrained model (i.e., baseline) revealed the model possessed excellent fit (i.e., GFI, AGFI, and CFI values > 0.90, RMSEA values were < 0.05 and Q values were < 3). The difference between the baseline and restricted models was not statistically significant, Δχ2(27) = 35.63, p = ns (see
Therefore, the items common to all three samples (CAP: 1, 3, 10, 11, 13, 14 and 15; RAD: 2, 4, 9, and 16) evidenced invariant factor variances, covariances, and loadings (see Appendix A). The alpha values, 95% confidence intervals for alpha, means, and standard deviations for the 11-item MCPRS scores (and subscales) are provided in
Refined Model | |||
---|---|---|---|
Measures of Fit | Western Canada | Eastern Canada | Mid-Western USA |
χ 2 | 123.17* | 188.55* | 118.29* |
Df | 88 | 64 | 63 |
Q | 1.40 | 2.95 | 1.88 |
GFI | 0.94 | 0.95 | 0.96 |
AGFI | 0.91 | 0.93 | 0.94 |
CFI | 0.94 | 0.87 | 0.93 |
RMSEA | 0.04 | 0.06 | 0.05 |
90% CI | 0.022, 0.058 | 0.050, 0.070 | 0.034, 0.060 |
AIC Refined Model | 187.17 | 242.55 | 174.29 |
AIC Original Model | 320.71 | 473.59 | 385.82 |
Note.*p < 0.001.
Model | ||
---|---|---|
Measures of Fit | Baseline | Restricted Model |
χ2 | 430.08* | 465.72* |
Df | 215 | 242 |
Q | 2.00 | 1.92 |
χ2 Difference | 35.64 | |
GFI | 0.95 | 0.95 |
AGFI | 0.93 | 0.93 |
CFI | 0.91 | 0.91 |
RMSEA | 0.03 | 0.03 |
90% CI | 0.025, 0.033 | 0.024, 0.032 |
AIC | 604.08 | 585.73 |
Note. *p < 0.001.
MCPRS | CAP | RAD | ||||
---|---|---|---|---|---|---|
M(SD) | α(95% CI) | M(SD) | α(95% CI) | M(SD) | α(95% CI) | |
Western Canada | 33.25 (6.12) | 0.73 (0.68 - 0.78) | 23.07 (4.84) | 0.75 (0.70 - 0.80) | 10.18 (2.79) | 0.60 (0.51 - 0.68) |
Eastern Canada | 34.05 (5.56) | 0.68 (0.63 - 0.72) | 19.55 (4.36) | 0.69 (0.65 - 0.73) | 14.49 (2.59) | 0.50 (0.42 - 0.56) |
Mid-Western USA | 34.10 (5.85) | 0.72 (0.68 - 0.76) | 19.53 (4.6) | 0.74 (0.70 - 0.78) | 14.57 (2.73) | 0.61 (0.54 - 0.67) |
Note: M = mean, SD = standard deviation, CI = confidence intervals, CAP = Concern with Acting Prejudiced, RAD = Restraint to Avoid Dispute.
Findings from the current study suggest that the MCPRS possesses satisfactory psychometric properties. Across three samples, scale score reliability coefficients were adequate (>0.70) and similar to those originally reported by Dunton and Fazio [
It should be noted that certain items caused difficulties across all samples. Items 5 and 17, which were permitted to cross-load initially, never loaded strongly enough on either factor to merit their continued inclusion, and continued to cross-load weakly notwithstanding iterative model refinement. As their removal improved the model, it is recommended that these items be deleted. Item 8, which in Dunton and Fazio’s [
There are a number of limitations that warrant mention. First, the data used in the current study were all collected from university students, and previous research conducted with the MCPRS has employed this same group. Second, most participants in this study were psychology students; thus, the generalisability of these findings―even when considered vis-à-vis university students―is further called into question. It is important that reliability, dimensionality, and invariance analyses be conducted with groups other than psychology students. Considering the number of differences that arose across the three samples which, given their student status, would appear to be fairly homogenous, respondents with more diverse demographic profiles may reveal further differences in factorial structure. Third, gender differences in the dimensionality and factorial invariance of the MCPRS were not examined. One would anticipate that, if pronounced differences between male and female students existed with respect to interpretation of items on the MCPRS, the indicants of model fit would have been suboptimal across samples―a result that did not occur. However, the possibility that slightly different models exist for men and women should be investigated, as should the possibility that validation coefficients may differ between these groups. Fourth, although commonly used, some psychometrists have expressed concern about the use of Likert scales in attitudinal measurement. Hodge and Gillespie [
It should be noted that, to date, when employing the MCPRS, researchers have used factor scores and permitted items to load differently on the two dimensions, as dictated by the sample employed [
Despite the aforementioned caveats, the present study provides additional strands of statistical and practical support for the use of the MCPRS as a reliable and valid measure of an individual’s motivation to control prejudiced reactions. Research should continue to examine this construct as it has important implications for inter-group interactions (e.g., [
Todd G.Morrison,Melanie A.Morrison,LorraineMcDonagh,DanielRegan,Sarah-JaneMcHugh, (2014) Confirmatory Factor and Invariance Analyses of the Motivation to Control Prejudiced Reactions Scale. Open Journal of Statistics,04,446-455. doi: 10.4236/ojs.2014.46043
Note. R = reverse-scored.