Measurement of risk perceptions in social research: a comparative analysis of ordinary least squares, ordinal and multinomial logistic regression models

Drawing on data gathered in the 2006 Monitoring the Future study of American youth (n = 2489), this investigation offers a comparative analysis of ordinary least squares (OLS), ordinal and multinomial logistic regression models in examining the effects of multiple factors on perceptions of alcohol risk. The article addresses limitations of OLS models in risk analyses and demonstrates how scholars can avoid making statistical errors when positioning vague quantifiers as ordinal dependent measures. Substantively, the article finds differential effects for (1) sex, (2) perceived attitudes of peers toward alcohol consumption, (3) frequency of intoxication, (4) teacher efforts toward alcohol education, (5) frequency of communicating with friends, and (6) newspaper exposure, as determinants of alcohol risk perceptions. Through statistical results and visual displays, the article reveals how inferences made about these effects stand to vary depending on the regression method chosen.