A statistical approach to an alcoholic drinking history

The drinking history of a middle-aged male was analyzed statistically on the basis of eight and a half years of notes on the number of drinks consumed per day. During the period his average number of drinks per drinking day increased from about 7 in 1974 to a peak of about 16 in 1980 while the number of abstinent days varied between 23% and 54% with no clear trend. These figures are of the same magnitude as published reports on drinking among alcoholics. Time-series models of intake or drinking frequency could not describe adequately the time-structure of annual or monthly consumption. Occurrence of drinking was analyzed as a random series of events. The time-structure of the series was highly irregular and deviated greatly from the Poisson hypothesis which assumes that each day has an equal probability of becoming a drinking day independently of previous days. Instead, drinking days were clustered into sprees with an average length of 7 days, high variance and a very skew distribution, separated by abstinence periods with an average length of 4 days and a similarly shaped distribution. The entire history could be partitioned into 286 alternating drinking and abstinence intervals, one day intervals included. The drinking rhythm was very stable: no significant trends in the lengths of either type of interval could be found. The main findings are the surprising stability of the drinking rhythm, its independence of the growing amounts consumed, and the independently varying abstinence interval lengths. Even in the absence of reporting and memory bias, such a pattern of drinking may produce very inaccurate recall of the actual long-term alcohol intake, if the recall period is short. The results suggest that periods shorter than one month should be avoided when asking questions about alcohol intake, for example, in research on the effects of treatment on alcoholism or alcohol intake on health.