A test of expectancy-value theory in predicting alcohol consumption*

Abstract

Background: Research on alcohol-related outcome expectancies has primarily focused on the likelihood of the anticipated effects, while comparatively little attention has been paid to their subjective evaluation. However, according to expectancy-value theory, the expectation that alcohol use will produce certain consequences and the evaluation of those consequences jointly and interactively determine an individual's decision to consume alcohol. Previous research on this issue was hampered by multiple regression strategies that are plagued by measurement error and low statistical power.

Method: To overcome this limitation, we investigated expectancy-value interactions in predicting drinking variables by drawing on latent variable methodology using the five expectancy-value dimensions from the Comprehensive Alcohol Expectancy Questionnaire. Expectancy-value models were tested in a sample of college students (N = 1053) and a sample of alcohol-dependent inpatients (N = 699).

Results: Significant expectancy-value interactions emerged concerning social assertiveness among students as well as for aggression and tension reduction among alcohol-dependent inpatients. The relationship between expectancy and drinking was strongest for pronounced (either positive or negative) valuations of the effect. Effect sizes were small, however.

Conclusions: The results are in partial agreement with basic premises of expectancy-value theory. However, this study also identifies limits to the universal validity of expectancy-value theory, given that prediction of alcohol use depends on the effect domains, alcohol outcome measures, and study populations.

Keywords:

• Alcohol expectancies
• outcome valuations
• evaluations of drinking outcomes
• expectancy-value theory
• CAEQ
• alcohol use

Disclosure statement

The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the article.

Notes

1 Two-way imputation handles missing data by estimating the missing scores and then imputing these estimates, which is an advantage over other approaches (such as Full Information Maximum Likelihood) in the context of the product-indicator approach used in this study. Two-way imputation improves upon simple mean or person-mean imputation by correcting for score differences between respondents and for score differences between items and has been shown to perform on par with other approaches for multidimensional Likert-type data (Bernaards & Sijtsma, 2000).