Item functioning of the alcohol dependence scale in a high-risk sample

We conducted in-depth analyses of the functioning of items from the alcohol dependence scale (ADS) in a sample of high-risk alcohol drinkers, specifically 101 men and 93 women mandated to a domestic violence intervention program. We first conducted a maximum likelihood common factors analysis on the ADS, which indicated a primarily unidimensional factor structure. We then used a nonparametric kernel smoothing method to create item characteristic curves (ICC) and option characteristic curves (OCC) for each ADS item. Based on these curves, we identified nine of the 25 ADS items as reliably discriminating between those with no or minimal alcohol problems and those with symptoms of excessive or abusive drinking. Dichotomous scoring appeared most appropriate for these items. No differential item functioning (DIF) by gender was detected, indicating that these items assess alcohol problems similarly in both men and women. This nine-item empirically-derived abbreviation of the ADS appeared to be an efficient and effective measure in this sample; it was highly correlated with the original scale (r(s) = 0.96) yet had superior distributional properties. Retained items reflected primarily excessive or hazardous drinking rather than alcohol dependence per se, suggesting that items targeting these types of symptoms may be most useful in high-risk samples. Combined with previous work with the ADS in treatment-seeking alcoholics, mapping of ADS item severities suggests a continuum of alcohol problem severity from heavy drinking to severe withdrawal that may be reliably tapped with dichotomous items. (C) 2003 Elsevier Ireland Ltd. All rights reserved.