As was documented in a seminal paper by Pernanen (1974), survey research has a history of failures in estimating national alcohol consumption. When individual responses to interview questions about the drinking of alcoholic beverages are aggregated to national levels established by sales and tax statistics, there is an underreporting not by a few percentage points but by the lion's share. Later reviews (Lemmens, Knibbe & Tan 1987, Kühlhorn 1998) have not changed this judgement.

The availability of gross sales statistics and the dearth of reliable survey statistics has put us in a position where we know much about the total level of alcohol consumption in a country or region but relatively little about the consumption in the various subgroups of its population. For better or worse, the World Health Organization has rationalized this situation in its general but blunt recommendation to member states to pursue policies that reduce the total national level of alcohol intake, i.e. what at present is subject to acceptable statistics (WHO 1980).

In a Swedish project named KALK we have explored the frontiers of survey research in the measurement of alcohol consumption. This task has forced us to take a fresh look at several traditional sources of error in survey research such as the effect of question wordings and the effect of non-responses.

The sponsors of KALK and Survey A are the major Swedish private and public stakeholders in alcoholic beverages: The National Alcohol Board, The National Institute of Public Health, Systembolaget, Vin & Sprit Group, and, The Swedish Brewers Association. The Swedish Council of Crime Prevention sponsored Surveys B and C. The authors of this paper are the members of the KALK research team, which is headed by Eckhart Kühlhorn.

Secondary analysis

All consumption of alcoholic beverages is not recorded in the sales and tax statistics (Kühlhorn, 1994). Liquor, wine and beer may be homemade, emanate from moonshine factories, be smuggled from countries with less or no tax, or be legally imported as "tax-free". With varying reliability such sources of alcohol can be ascertained by survey research and/or projections from health statistics. In Figure 2 we have added them to the recorded statistics for Sweden by using data from Norström (1997) and a secondary analysis of seven surveys from the period 1980-1996.

Our new surveys

Our new data are mainly found in three Swedish national samples:

10.073 telephone interviews in a sample based on random digit dialing (RDD). Comparable daily samples were interviewed every day for a whole year from March 1996 through February 1997. GfK Sweden did the fieldwork, which after six recalls had a completion rate of 50.4 per cent. There was no advance notice by mail about these interviews; you cannot write letters to phone numbers.

2.303 telephone interviews in a high precision unrestricted random probability sample from the population registry. Conducted in April through August by Statistics Sweden which after 15 recalls had a completion rate of 74.7 per cent. Advance notice by mail were sent to all chosen for interviews.

1.460 telephone interviews with repeat drunken driving offenders from the official crime registry. Fieldwork in the spring and summer by Statistics Sweden which after 15 recalls had a completion rate of 33.6 per cent. Advance notices by mail were sent to all chosen for interviews. The letter indicated that the researchers were aware of the court records of the respondents. There was therefore less incentive among the latter to totally deny their alcohol problem.

Full description of Sample A is found in Kühlhorn & Ramstedt (1998) and of Samples B and C in Kühlhorn, Leifman, & Borschos (1998).

Accuracy of recall during a week

We asked everybody " Did you drink any hard liquor /wine/ beer/ yesterday, that is, x-day?" and "Did you drink any hard liquor /wine/ beer/ the day before yesterday, that is, x-day?" The same type of question was asked for each day during the week preceding the interview in a part of the A-survey comprising 1.267 interviews split in equivalent daily samples with respondents who had reported that their alcohol consumption had been normal during the previous week. If the answers were entirely accurate the difference between the days after recall would be no larger than random sampling variations. However, the workings of memory in answering this type of question causes the quantity of alcohol consumed to decrease from day to day, counting from the day of the interview. The results after recalculation in pure alcohol intake are found in Table 1.

The decline in recall of alcohol consumption already starts after one day. There is a rapid decline to almost half the level reported for the day before an interview. This fact raised our hopes that one could formulate questions in survey research on alcohol to avoid or reduce the underreporting. Encouragement comes from studies of memory and mismemory in surveys of health events (hospitalizations, miscarriages, taking blood pressures) and of political events (registration, voting); some of them are summarized in Tanur 1992, part 3).

One can, of course, discuss the accuracy of our base line of 100 per cent after one day. In reality, the figure for one day after recall may also have distortions, particularly among the heavy consumers who may suffer from hangovers. But this does not affect our general conclusion about a rapid deterioration of recalls of events involving alcohol.

Daily rhythms

N=10013

The number of drinking men and women doubles on Fridays and Saturdays. More than half (58%) of the weekly alcohol intake occurs on Friday and Saturday. Also, using an indicator of heavier drinking — at least one bottle of wine or its equivalent in beer or spirits, i.e. 8.4 centiliter pure alcohol —the same pattern appears. There are three times as many heavy drinkers on Fridays and five times as many on Saturdays than on the other days of the week.

This pattern of alcohol consumption does not interfere with the normal workweek in industrial society. It was established during the early part of the twentieth century as part of a process of disciplining a work force with a predominantly rural background to a new labor market (Ambjörnson 1988).

We will discuss three common types of question wordings used in surveys on alcohol against the background of the patterns of weekly distribution of intake and declining recall. The three types of questions differ in the type and complexity of calculations imposed on the respondents. We refer to them as "frequency & quantity", "normal week", and "last seven days".

For wine and beer the response alternatives referred to glasses, cans, or bottles.

A problem with these questions is found in the fact that most people (in Sweden) drink a great deal more on Fridays and Saturdays and therefore have difficulty in giving an average figure for each occasion which they have consumed alcohol. You do get answers, but they may not be accurate.

A Normal Week

The difficulties are considerable for the respondents to estimate their weekly consumption over a highly varied weekly rhythm. Our questioning about alcoholic beverages during a normal week, therefore, divided the week into homogenous periods. Keeping in mind the Swedish weekly rhythm of drinking we asked:

Do you ever drink hard liquor/wine/ beer in a normal week?

Yes
No

If "Yes":

About how much hard liquor/wine/ beer/ do you drink during a normal week? This may of course vary in the course of the year, but try to give an average amount. You may give your answer in number of /centiliters, or, shots, or, drinks/glasses, or, bottles/glasses, or, bottles, or, cans.

How many drinks of hard liquor /glasses of beer/ wine/ do you normally drink on Sundays?

How many drinks of hard liquor /glasses of beer/ wine/do you usually drink on a weekday from Monday to Thursday?

How many drinks of hard liquor /glasses of beer/ wine/ do you normally drink on Fridays?

How many drinks of hard liquor /glasses of beer, wine/ do you normally drink on Saturdays?

If "No": Ask the Frequency and Quantity question. (This was done in Sample B, not in the other samples)

A normal week would exclude the great festivities and carnival weeks, New Year celebrations, et cetera. This might save the respondent from being embarrassed by having to consider intoxicating amounts, and thus underestimate the total consumption. At any rate, we found that answers are quickly and readily given to this type of questioning.

There are, however, particular difficulties in estimating the "normal" if consumption is irregular. As we know, some people have very high consumption periods followed by periods of low or no consumption.

The Last Seven Days

To reduce respondent calculations some researchers have asked about alcohol intake during the last seven days rather than during a normal week. A difficulty with this approach lies in the fact that the researcher does not know how he shall calculate the weights of the answers unless he interviews during every week of the year, or unless he has prior knowledge of the seasonal variations in alcohol consumption. We resolved this problem by a filter question:

About one-third in our A-sample reported an atypical situation during the last seven days.

Discussion : Who shall do the calculations: the researcher or the respondents?

There are at least two schools of thought about questionnaire construction aimed at gross statistics:

In a survey interview the respondents give the facts, and the researcher calculates the distributions of these facts.

This is the rule of thumb in a "positivist tradition" of social research. The survey interview is here usually seen as a series of stimuli and responses. Survey respondents should only be asked to provide raw material for the researchers’ calculations. If the respondents are asked to do calculations there will be errors and lack of uniformity.

The errors occurring in this approach lie mainly with the respondents whose memory and perception thus should be checked and double-checked. The few conceivable errors of the researcher are only arithmetic and can be corrected without new fieldwork. The questionnaires we have called "Frequency and Quantity" and "Last seven days" are written in this tradition.

The second view:

A survey interview is a structured conversation in which the respondents give their considered judgements, and the researcher shall understand and summarize these judgements.

The researchers in this more "cognitive tradition" accept that respondents are ready, willing and able to enter a conversation with the interviewer and give his or her estimates also of complex situations. It is understood that such estimates are formulated in terms of the respondents' personal values and by the opinion climate in the circles where they move. The survey interview in this approach is often seen as a process of mutual symbolic interaction between interviewer and interviewee, not as a stimulus-response situation.

The questionnaire we have called "The normal week" comes nearer this approach than the others. Every respondent may here define what is normal for him or her, given his or her values and circles, and the interviewer and interviewee together fill this self-chosen conception of a normal week with counts of shots of liquor, glasses of wine, or cans of beer.

Sample surveys consist of at least two strata: respondents and non-respondents. Non-response introduces distortions if a variable measured has a different distribution in the non-response stratum than in the responding stratum.

The non-response stratum can be divided into at least two sub-strata: the accessible and the non-accessible for an interview. When we sample the stratum of non-respondents for a follow-up study, the new respondents from the follow-up belong to the accessible sub-stratum. In Sample A we made a follow-up of the non-respondents with 264 interviews. Their answers were weighted into the respondents' answers in all tabulations. The differences between the accessible non-respondents and the respondents were very small, so small that they could have been neglected without any change in our conclusions.

Sample A is a typical market research sample, with a completion rate of 50.4 per cent. Sample B is a high quality government survey. The B sample included many more from the accessible non-respondent stratum in A. The B-Sample had advance notices by mail sent to all chosen for interviews, and 15 recalls were made before the interviewers were allowed to give up. The completion rate is 74.7 per cent.

Our estimates of alcohol consumption were, however, only marginally affected by reducing the non-response from some 50 percent to 25 per cent. Sample A and Sample B show only small differences. In other words, the accessible non-respondents in the market research sample do not differ from the respondents in the high-quality government survey.

In Sample C of the repeatedly drunken drivers, we have a concentration of alcoholics, near-alcoholics, and persons with alcohol problems and/or alcohol irresponsibility. The response rate of Sample C was 33.6 percent using the same thorough methods and interviewers as in Sample B. Persons with alcohol problems are thus grossly underrepresented with normal survey procedures. They are over-represented among the non-accessible non-respondents. Since they do drink copiously — among male alcoholics about a bottle of hard liquor a day and much more beer than the average person drinks —– they account for the most significant amount of the underreporting of alcohol consumption in surveys.

Smuggling of alcoholic beverages, home production, travelers' private imports can be estimated from interviews. We start the estimation with the official tax-registered alcohol sales, and then add from the interview data the homemade, privately imported, and illegally obtained shares. In estimating total consumption we thus restrict the use of survey data to provide estimates supplementary to the known sales of alcohol.

In this process we have considered three types of question wordings:

We are now in position to see how well the various interview questions are able to reproduce the total alcohol consumption in the country. Table 3 tells the story.

All question-types show gross underestimation. Survey research is still short of success in measuring total alcohol consumption in a society. The consumption a "Normal week", in spite of the several problems we noted, gives the best estimate, at least with regard to national consumption: it records 58 per cent of the total consumption. The "Quantity-Frequency" questions record only 37 per cent.

We conclude that we can reduce the underreporting by avoiding the "positivist" questions in "Frequency & Quantity" and using the more "cognitive" questions of "Normal week".

Alcoholic Beverages and European Society. (1993). Amsterdam: The Amsterdam Group.

Ambjörnsson, R. (1988), Den skötsamme arbetaren: idéer och ideal i ett norrländskt sågverkssamhälle 1880-1930, Stockholm: Carlsson.

Björkman, N-M. (1979), Social önskvärdhet som felkälla i frågeundersökningar. En jämförelse mellan två datainsamlingsmetoder, Stockholm.

Brazeau, R., & Burr, N. (1992). International survey: Alcoholic beverage taxation and control policies (8th ed.). Ottawa: Brewers Association of Canada.

Cutler, S. F., Wallace, P. G. & Haines, A. P. (1988), "Assessing alcohol consumption in general practice patients – a comparison between questionnaire and interview." (Findings of the Medical Research Council’s general practice research frame work study on lifestyle and health), Alcohol and Alcoholism, vol. 23, No 6, pp. 441-450.

Davis, P., & Walsh, D. (1983). Alcohol problems and alcohol control in Europe, New York: Gardner Press.

Hukens, C.L.H., Knibbe, R.A., & Drop, M.J. (1992). "Alcohol consumption in the European community: Uniformity and diversity in national drinking patterns." Paper presented at the 18th Annual Alcohol Epidemiology Symposium, Toronto, Canada.

Jansson, C-G. (1990), "Retrospective data, undesirable behaviour, and the logitudinal perspective" in Magnusson, D. & Bergman, L. D.(ed), Data quality in longitudinal Research, Cambridge, Cambridge University Press.

Kühlhorn, E. (1994), Statistikförd och icke-statistikförd alkohol. Bilaga 8 i SOU 1994:25, Socialdepartementet.

Kühlhorn, E (1998), "Problem och nya grepp i kartläggningar av alkoholkonsumtionen och alkoholskadorna" in Kühlhorn, E. & Björ, J., Svenska alkoholvanor i förändring. Sober förlag.

Kühlhorn, E. & Leifman, H. (1993), "Alcohol Surveys with High and Low Coverage Rate: A comparative analysis of survey strategies in the alcohol field", Journal of Studies on Alcohol, vol. 54, pp. 542-554.

Kühlhorn, E & Ramstedt, M. (1998). Svenskarnas konsumtion av alkohol från legala och illegala källor 1996, (forthcoming).

Kühlhorn, E., Leifman, H. & Borschos, B. (1998) Kartläggning av sociala förhållanden i socialt marginaliserade grupper, (forthcoming).

Lemmens, P., Knibbe, R.A. & Tan, F. (1987), "Weekly recall and diary estimates of alcohol consumption in a general population survey", Journal of Studies on Alcohol, vol 49, pp.131-135.

Marshall, M. (Ed.). (1979). Beliefs, behaviors and alcoholic beverages: A cross-cultural survey, Ann Arbor, MI: University of Michigan Press.

Midanik, L. (1982), "The validity of self-reported alcohol consumption and alcohol problems: A literature review", British Journal of Addiction, vol 77, pp. 357-382.

Norström, T. (1997), "Alkoholkonsumtionens mörkertal i Sverige 1960-1994", Nordisk Alkohol- och narkotikatidskrift, vol. 14, nr. 2, pp 65-73.

Pernanen K. (1974) "Validity of survey data on alcohol use" in: Gibbins, R.J., Israel, Y., Kalant, H., Popham, R.E., Schmidt, W and Smart, R. G.. (Eds) Research Advances in Alcohol and Drug Problems, Vol. 1, New York: John Wiley & Sons, Inc., 1974, pp 355-374.

Pittman, D.I., & White, H. R. (Eds.). (1991). Society, culture and drinking patterns revisited, New Brunswick, NJ: Rutgers University.

Sabroe, K.E. (1994). Alcohol in society, Aarhus: Aarhus University Press.

Saunders, J.B., Aasland, O.G., Babor, T.F., De La Fuente, J. & Grant, M. (1993) Development of the Alcohol Use Disorders Identification Test (AUDIT). WHO Collaborative project on early detection of persons with harmful alcohol consumption - ll, Addiction, 88, 791-804.

Smith, D E. & Solgaard, H.S. (1998). "Global trends in European Alcoholic drinks consumption", Marketing and Research Today, vol 27, no 2, pp. 80-85.

Spak, F (1996) Women and alcohol in Göteborg: an epidemiological study of female alcoholism and alcohol related problems. The Department of Social Medicine, Göteborg University, Vasa Hospital and the department of Psychiatry, Karolinska Institutet, Huddinge University Hospital, Göteborg

Sudman, S. & Bradburn, N.M. (1982) Asking questions. Jossey-Bass Publishers, San Francisco, Washington, London.

Tanur, Judith M. editor (1992) Questions about questions. Russel Sage Foundation, New York.

WHO (1980), Problems related to alcohol consumption. Report of a World health organization Expert Committee. Technical Report Series 650. Geneva, WHO, 1980.

Wines and Spirits in the European Community. (1993). Brussels: Club de Bruxelles.

World Drink Trends. (1995). Oxford, England: NTC Publications Ltd.