Measuring alcohol consumption: A comparison of graduated frequency, quantity frequency, and weekly recall diary methods in a general population survey

Abstract

Objective: The objective of this study is to compare self-reports of alcohol consumption obtained by graduated frequency (GF), quantity frequency (QF), and prospective weekly diary methods in a Swiss general population survey using a within-subject design. Another objective is to examine the consumption means, ranking order, and drinking classifications and to relate differences in consumption means and classification inconsistencies to social and drinking characteristics. Method: The data came from the first wave of a longitudinal study on changes in alcohol consumption. A subsample of 767 respondents who completed the three measures was examined. The sample design was accounted for. Results: Weekly drinking diary (WDD) yielded over 50% more alcohol consumption than the QF and GF measures did. Measures on quantity and graduated frequencies did not differ significantly. There were more light drinkers found on QF and more harmful drinkers on the WDD. Classification inconsistencies between GF and the weekly drinking diary were significantly related to gender and problematic drinking.

Introduction

Self-reports are widely used to assess consumption in alcohol research because, in contrast to aggregated level sales data, they permit the analysis of different consumer groups Lemmens et al., 1988, Midanik, 1982, Rehm, 1998b. Sound research results depend on the reliability and validity of measurement. Self-reports of alcohol consumption have been seen to be generally reliable (for an overview, see Harris et al., 1994, Simpura & Poikolainen, 1983, Sobell & Sobell, 1990). As regards validity, the absence of a “gold standard” makes investigations of the criterion validity of self-reports difficult Midanik, 1982, Rehm, 1998b. According to the literature on survey research, systematic variations in measurements may be related to (a) the construct validity of the assessment instrument, that is, the degree the aspects of the construct are addressed by measurement, and (b) the assessment mode Aday, 1996, Groves, 1989. The current study examines alcohol consumption measures stemming from a telephone administered quantity–frequency instrument (QF), a telephone administered graduated–frequency instrument (GF), and a self-administered weekly drinking diary (WDD). With respect to construct validity, QF does not capture variability of drinking, while GF and WDD do. In relation to assessment modes, both QF and GF are retrospective measures, while WDD is prospective and may therefore be less prone to recall errors. Thus, the comparison of QF and GF should concentrate mainly on the effects of variability of drinking, whereas the comparison of GF and WDD should account for differences in recall errors. Previous studies on the comparison of measurement instruments were mainly conducted in Northern Europe, North America, Australia, and New Zealand Feunekes et al., 1999, Rehm, 1998b. When compared with these regions, drinking can be said to be more regular in Switzerland Allamani et al., 2000, Lemmens, 2001. Thus, because variability of drinking may be a minor explanatory factor for the differences in measurement instruments in Switzerland, potential differences between consumption values on WDD and GF should be indicative of memory errors.

Concurrent validity addresses the degree of the variations in the assessment of the same construct by different measures Aday, 1996, Groves, 1989. It has been examined for a variety of measures, including QF and GF, but rarely for prospective WDD. Generally, higher consumption values were found for GF than for QF Midanik, 1994, Rehm, 1998b, Room, 1990. So far, WDD has been used in clinical settings Fiellin, 2000, McManus et al., 1999 and nutritional epidemiology Franson & Rossner, 2000, Lambe et al., 2000, Mathews et al., 2000. Applications of WDD in the field of alcohol research are sparse Corti et al., 1990, Hilton, 1989, Lemmens et al., 1992, as is literature on its reliability and validity (Rehm, 1998b). Existing research shows that WDD generally yields higher consumption than do other measures (for an overview, see Leigh, 2000). To our knowledge, the concurrent validity of GF and WDD has not yet been investigated in a general population, and comparisons between WDD and GF were limited to a volunteer group of 83 relatively heavy drinkers (Hilton, 1989) and a volunteer group of 52, mainly, female drinkers (Poikolainen, Podkletnova, & Alho, 2002).

Assessment of alcohol consumption usually includes the three dimensions of quantity, frequency, and variability Alanko, 1984, Room, 1990. Differences in consumption may be related to a, more or less, accurate coverage of these dimensions by the assessment instrument. Consistent is the finding that the more detailed the questions on alcohol consumption, the more consumption shows up Dawson, 1998, Knibbe & Bloomfield, 2001, McCann et al., 1999. For instance, it has been proposed that unequal assessment of variability explains higher drinking volume on GF than on QF Hilton, 1989, Midanik, 1994. Also, recalls may be plagued with errors due to the failure of retrieval, such as forgetting, interfering, or confusing drinking events in retrospective measures. Thus, prospective measures are generally assumed to be more accurate than retrospective ones are, and prospective drinking diaries have been used to validate retrospective measures Armstrong et al., 1992, Lemmens et al., 1988, Poikolainen et al., 2002.

One drawback in comparative studies is that QF and GF often cover longer reference periods, whereas prospective diaries are usually used for short-term investigations because of the response burden on the individuals and the potential lack of compliance for longer periods (Leigh, 2000). Short reference periods may result in insufficient assessment of exceptional drinking occasions (Dawson, 1998) or overestimation of abstention (Rehm, Walsh, Xie, Greenfield, & Single, 1997). In the present study, this drawback is partly overcome by using a weekly reference period for GF and WDD. QF had a reference period of 6 months. However, QF should be more or less stable, independent of the length of the reference period, because it is expected to measure modal consumption Duffy & Alanko, 1992, Kühlhorn & Leifman, 1993. Despite the lack of empirical evidence in general populations, WDD can be hypothesized to yield the highest consumption, as it avoids averaging processes and because consumption is rapidly recalled.

At a first stage, the present study investigates the differences of overall self-reported alcohol consumption on QF, GF, and WDD. At a second stage, the stability of the rank order of the individuals classified across measures is addressed because validity issues, especially in epidemiological studies, should not be restricted to an overall consumption mean (Rehm, 1998b). Common values for rank order correlations are comprised between 0.4 and 0.7 Feunekes et al., 1999, Lemmens et al., 1992, Rehm et al., 1997. At a third stage, the present study compares the classification of the respondents according to their drinking status. Comparisons between QF and GF showed that GF results in higher proportions of heavy drinkers and QF in higher proportions of light drinkers Midanik, 1994, Rehm, 1998b. Similar differences in the proportions were obtained when comparing diaries with QF (Redman, Sanson-Fisher, Wilkinson, Fahey, & Gibberd, 1987). Finally, at a fourth stage, differences in consumption means and classification inconsistencies were regressed on the demographic and drinking characteristics of the respondents.