![Sample our Communication Studies journals, sign in here to start your access, 2013 & 2014 volumes FREE to you for 14 days]({"type":"image" , "src" : "/sda/5927/TOC-Subject-Sample-2021-Information-Science2.jpg"})
Abstract
The construct validity of Infante and Wigley's verbal aggressiveness scale and Infante and Rancer's argumentativeness scale are assessed with confirmatory factor analysis and multitrait–multimethod analysis. The factor analytic data replicate previous findings that the verbal aggressiveness scale measures two constructs, verbal aggressiveness and verbal benevolence communication style, and that the argumentativeness scale is unidimensional with some poor items. The multimethod data, however, show near zero correlations between self-reports and observed behavior and evidence of method variance. These data indicate a discrepancy between conceptual definitions and behaviors. Rather than measuring behavioral dispositions to communicate argumentatively or aggressively, the scales may function as attitude or self-concept scales.
Notes
1. Hunter and Gerbing's (1982) centroid method does not calculate GFI. Therefore, AMOS was used in a parallel manner to obtain GFI for the CFA. It is worth noting that the Hunter and Gerbing CFA estimates were highly consistent with the AMOS CFA estimates.
2. Whereas the internal consistency of a set of indicators purported to measure the same construct is the extent to which each indicator correlates predictably with other indicators in that set, the parallelism between two or more sets of indicators is the extent to which indicators purported to measure the same construct correlate predictably with indicators purported to measure a different construct.
3. None of the skewed distributions showed evidence of range restriction. Furthermore, Havlicek and Peterson's (Citation1977) Monte Carlo study found that Pearson's r is robust to considerable violations of the normality assumption. Therefore, although the MTMM matrix relied on zero-order correlations, some of which contained variables with skewed distributions, the matrix can be interpreted meaningfully.
4. The only correlation pair that had a difference greater than what could be attributable to sampling error was the message-generation correlation between verbal benevolence and argumentativeness. The correlation between the measures as designed (r=−.93) was outside of the 95% confidence interval for the correlation between the optimized measures, r=−.86, P(−.91≤ρ≤−.81)=.95.