Objective: Estimated test-retest reliabilities of self-re- ported drinking measures are affected by the extent to which respondents provide consistent reports of their own behaviors (reliability) and the extent to which the behaviors reported are stable over time (stability). Unstable behaviors may be reliably reported but correlate poorly over time. This study tests whether an estimate of the stability of drinking patterns is related to test-retest reliabilities of drinking measures.
Method: Data were from a general population telephone survey given twice, 1 month apart, to 307 adult drinkers. Drinking measures included age of onset, and graduated frequency measures used to estimate drinking frequencies, average quantities, and total alcohol consumption. Measures of drinking stability were estimated using a well-tested model for the analysis of drinking patterns (i.e., variances in drinking quantities and frequencies). Heteroscedastic regression models were used to partition covariances in self-reports between Times 1 and 2 into components related to stability and reliability, providing a more accurate picture of the reliability of drinking measures for different drinking groups.
Results: Overall test-retest reliabilities were good, ranging from a low of .65 for drinking quantities to a high of .85 for drinking frequencies. The stability of quantity measures had a large impact on estimated test- retest reliabilities. Stable drinking patterns were associated with much greater test-retest reliabilities.
Conclusions: Data on alcohol use from general population telephone surveys are generally reliable. However, observed reliability is a function of the stability of drinking patterns. Ostensibly unreliable self-reports may be highly reliable but may reflect unstable drinking patterns.