Compensatory health beliefs: scale development and psychometric properties

Abstract

Compensatory Health Beliefs (CHBs) are beliefs that the negative effects of an unhealthy behavior can be compensated for, or “neutralised,” by engaging in a healthy behavior. “I can eat this piece of cake now because I will exercise this evening” is an example of such beliefs. The present research describes a psychometric scale to measure CHBs (Study 1) and provides data on its reliability and validity (Studies 2 and 3). The results show that scores on the scale are uniquely associated with health-related risk behaviors and symptom reports and can be differentiated from a number of related constructs, including irrational health beliefs. Holding CHBs may hinder individuals from acquiring healthier lifestyles, for example lose weight or exercise.

Keywords:

• Compensatory health beliefs
• Irrational health beliefs
• Anticipated pleasure
• Self-efficacy
• Dissonance
• Health behavior

Acknowledgements

The reported research was funded, in part, by a New Opportunities Fund (grant 4015) from the Canadian Foundation for Innovation to Bärbel Knäuper. Special thanks are expressed to David D’Alessandro and Kent Harber for their constructive feedback, and to Melanie Yugo for her help with the data collection.

Notes

Concerns have been brought forward in the past that Internet users do not present a representative sample of the general population (see Couper, 2000). This was less of an issue here, though, because the goal was not to reach a sample in which all socio-demographic groups are proportionally represented. It was sufficient to reach some members of all groups, which is realistic given that a certain proportion of members from all socio-demographic groups have Internet access. We received more than 500 entries from people varying in gender, age and country of origin. In terms of the major socio-demographic variables, all groups were represented in the sample, though the recruitment strategy certainly restricted the sample to English-speaking respondents. A large number of the submissions were highly redundant, suggesting that the existing pool of CHBs has been exhausted. To further rule out the possibility that important domains of CHBs were missed, we asked the health psychology experts who reviewed the item pool whether they could contribute additional item ideas. No additional items were suggested by the experts beyond those already in the pool.

A computer-based approach was chosen as a cost-efficient method for collecting the retest data. A large amount of research has demonstrated measurement equivalency between paper–pencil and web- or computer-administered questionnaires. Specifically, measurement equivalency has been found regarding variance, factor structures and factors loadings, covariance structures, internal consistency, and test–retest reliability (e.g., King and Miles, 1995; Stanton, 1998; Finger and Ones, 1999; Donovan et al., 2000; Miller et al., 2002). For the present data, the variance, factor structure, factor loadings, and internal consistency values were comparable for the time 1 and time 2 assessments, supporting the notion of measurement equivalency of the paper–pencil and computer-based version of the CHB scale.

PFA is generally recommended over principal components analysis (PCA) when the goal is to find a parsimonious representation of the relationships between assessed variables (Fabrigar et al., 1999). PFA more realistically estimates factor loadings and factor correlations than PCA because it recognizes the existence of random error in the measured variables and therefore less likely results in inflated factor loadings and an underestimation of factor correlations (Fabrigar et al., 1999; Russell, 2002). When the number of variables and the communalities are sufficiently high, PFA and PCA often result in comparable factor solutions, and this is the case here as well: rerunning the analyses using principal components analysis resulted in the same number of factors with the same variables loading on each of the four factors. The only emerging difference were higher factor loadings when PCA was used.

Table II shows the uncorrected as well as disattenuated correlations. Because disattenuation does not change the overall pattern of correlations substantially, we discuss the uncorrected correlations in the text.