Calibrating the Graduated Frequencies Summary Measure: Preliminary Sensitivity Findings From the Year 2000 U.S. National Alcohol Survey Informed by Drinking Diaries, Protocol Analysis, and Beverage-Type and Context Data

Abstract

Objective: Graduated Frequencies (GF) alcohol consumption measures efficiently summarize drinking pattern and permit volume calculations, both vital for epidemiology and policy studies. However, few studies have assessed accuracy of standard algorithms. We conducted sensitivity and simulation exercises to gain insight into potential effects of calibrating the measure to improve assessment of volume. Method: We used drinking diaries (n=99) followed by a summary GF scale to recalibrate multi-level F and Q constants associated with volume summations. A separate protocol analysis asked respondents (n=58) how many drinks they usually had within each GF range. Third, the Year 2000 National Alcohol Survey (N10) includes a novel technique for improving recall of maximum drinks/day using contextual cues, developed to correct volume/pattern estimates. Proportions of drinking in contexts where hypothetically larger pours of wine and liquor (home, parties and “street” drinking), plus beverage potency (malt liquor beer and spirits), were used to estimate impacts of drink size and strength on volume and consumption distributions. Results: Diary results showed that the mean quantity of a GF quantity range is typically below that range’s midpoint. In a sensitivity result based on preliminary N10 national telephone data (n = 2,447 drinkers) this bias was largely offset by underestimation of GF mid-level frequencies, yielding little volume change (increases and decreases of less than 5%, depending on assumptions). The protocol analysis confirmed the quantity adjustment. In the N10 sample of drinkers, repeating the GF with contextual cues elicited a higher maximum in 15% drinkers; 9% had a large volume increase (+83%) but change was small (+3.5%) averaged over all drinkers. Lastly, pour-size/strength simulations in the same national data showed that such adjustments promise the greatest aggregate effect on population estimates. Various plausible assumptions yielded up to a 37% overall increase in mean volume. Conclusions: Results imply that GF algorithms are robust to typical recall distortions; conversely, size and type of a subjective “drink” (dose) produce the dominating measurement biases sufficient to account for much of the “dark matter”, the amount undercovered by standard self-report survey measures. Improving survey measurement is critical for making sound, evidence-based policy recommendations.