Affective asynchrony and the measurement of the affective attitude component

Abstract

How should the affective component of attitudes be conceptualised and measured? Three studies compared measures based on different conceptualisations. Affective attitudes can be: (1) holistic reactions to objects or responses derived from spontaneous images of the objects; (2) bipolar or unipolar in structure; and finally (3) discrete emotional evaluations (e.g., angry, happy) or more general valenced evaluations (e.g., good, bad). It is recommended that further research with self-reports of the affective component include the holistic, unipolar, discrete emotion (HUE) evaluative measures developed in this paper in combination with a holistic bipolar valenced evaluation measure. Our results also supported a hypothesised affective asynchrony effect. Specifically, an affective measure that required more deliberative thought reduced the correspondence between the affective component and intended behaviours/attitudes. Implications for how the affective component may be represented in memory and the impact of that representation on constructed attitudes are discussed briefly.

This research was supported in part by a National Institute of Mental Health Emotion Research Training grant (No. MH18935), by a contract to Decision Research from the Nevada Nuclear Waste Projects Office, and by the National Science Foundation (SES-0111941, SES-0241313, SES-0112158, SES-0339204).

This research was supported in part by a National Institute of Mental Health Emotion Research Training grant (No. MH18935), by a contract to Decision Research from the Nevada Nuclear Waste Projects Office, and by the National Science Foundation (SES-0111941, SES-0241313, SES-0112158, SES-0339204).

Acknowledgments

The authors would like to thank Sarah Lichtenstein, Robert Mauro, Lew Goldberg, Craig Smith, James Flynn, C. K. Mertz, Mona Bronson, and two anonymous reviewers for their comments and assistance on earlier drafts of this paper.

Notes

This research was supported in part by a National Institute of Mental Health Emotion Research Training grant (No. MH18935), by a contract to Decision Research from the Nevada Nuclear Waste Projects Office, and by the National Science Foundation (SES-0111941, SES-0241313, SES-0112158, SES-0339204).

¹Breckler and Wiggins (1989, 1991) have drawn a theoretical and empirical distinction between the use of the evaluative terms such as good and bad for evaluation (e.g., “Legalised abortion is … good/bad”) versus the use of these same terms to assess the affective component only (e.g., “Legalised abortion makes me feel … good/bad”).

²A complete assortment of measures would total eight; we tested only six measures because time constraints would not allow us to include unipolar or bipolar discrete emotion versions of the imagery measure.

³Factor analyses were conducted with the 39 discrete emotions terms for each of the five stimuli and then in a general analysis combining all data across the five stimuli. Two similar factors emerged in each analysis and were named positive and negative affect. Convergent and discriminant validity were acceptable. Details are available from the first author.

⁴ HUE measure correlations and systematic response error. The oblique (promax) rotation of the general factor analysis from which the HUE scales emerged indicated that the two factors were mostly independent (r=−.29). However, Russell and Carroll (1999) suggested that systematic and random error in participants’ ratings may cause measures such as this one to appear falsely unipolar, rather than bipolar.

Systematic and random error can be controlled statistically. As did Russell (1979), we focused on statistical control of acquiescent response style (systematic error) and random error. First, acquiescent response bias (i.e., an individual difference in the tendency to agree or disagree consistently with items regardless of content) was calculated by summing responses to all affective evaluation items except the 39 unipolar discrete emotion terms. This summation thus included items for which agreement indicated positive affect as well as items for which agreement indicated negative affect. For the general analysis, this summed response included all stimuli; otherwise, only responses for the relevant stimulus were included. Partial correlations of the positive and negative HUE measures, controlling for acquiescence, were calculated and then were disattenuated for the unreliability of the affect measures.

The final results shifted the correlation between the positive and negative HUE measures in the expected negative direction, but the shift was minor for most stimuli. Prior to correcting for error, the correlation of the positive and negative HUE measures was −.37 for the general solution and ranged from −.06 to −.43 (average r=−.24) for each stimulus solution. After partialling out acquiescence and disattenuating the results, the same correlation was −.44 for the general solution and ranged from −.09 to −.56 (average r=−.27).

Russell and Carroll (1999) provided a convincing argument that participants’ varying interpretations of the response scales would influence the expected correlation for a bipolar relation between positive and negative affect. They estimated that the expected correlation for bipolarity would lie in the range between r=−.467 and −1.00. Correlations for four of the five stimuli and the general solution lie outside this range, ruling out strict bipolarity. The correlation for church (r=−.56) indicated the possibility of a strict bipolar structure.

⁵Increases in deliberative thought, however, may impair affective access only when faced with relatively unfamiliar objects (Hodges & Wilson, 1993) or when appropriate dimensions are difficult to identify (Kmett, Arkes, & Jones, 1999).