Missing value imputation in longitudinal measures of alcohol consumption