German Journal of Psychiatry
From the Department of Psychology, University of Stockholm, Sweden. Corresponding author: Peter Wennberg, Department of Psychology, University of Stockholm, 106 91 Stockholm, Sweden. e-mail: firstname.lastname@example.org
The study investigated the reliability of a Timeline Follow Back technique for measuring alcohol consumption and the subjects' ability to effectively recall consumption in the recent past. A follow-back period of 28 days was considered. As part of a series of health interviews, evenly distributed over the year, conducted within the confines of a wider-ranging longitudinal project, a sample of Swedish men (n=105) estimated, at the age of 36, their alcohol consumption over a period of 28 days with the aid of a daily calendar. The results showed that a 28-day timeline period is probably too long for a fully effective recall. An estimation based on a linear regression showed a drop in reported consumption of 1.5 percent per day back in time. However, a non-parametric regression implied that the drop in reported consumption is possibly not distributed in a linear way. A 14-day Timeline or shorter is likely to be preferred, mainly on the grounds that it shows a less severe drop in time-related reported consumption.
(German J Psychiatry 1998; 2: 62-68)
Keywords: Timeline Follow Back, alcohol consumption
Received:October 10, 1998
Revised version: November 30, 1998
Published: December 10, 1998
Acknowledgement: This study was supported by grants from the Swedish Council for Research in the Humanities and Social Sciences, the Swedish Alcohol Research Fund, the Bank of Sweden Tercentenary Foundation, the Swedish Council for Social Research, and Sweden's National Council for Planning and Co-ordination of Research.
Several methods are available for assessing alcohol consumption.
They can be classified into biological measures, self-reports,
clinical diagnoses (e.g. by physicians), and indirect measures
(usually vulnerability indicators). Self-report procedures have
been shown to have advantages in terms of simplicity and cost
(for reviews, see Babor et al., 1989; Babor & Del Boca, 1992).
Probably the most common form of self-reporting is based on the
so-called QF method (see Knupfer, 1966 for a discussion), in which
respondents are asked about the typical Quantity (Q) and typical
Frequency (F) of their personal consumption of specified "types"
of alcoholic beverages (usually beer, wine and spirits). The Quantity
(i.e. how much?) and Frequency (i.e. how often) is thereafter
multiplied to form a total measure of volume alcohol consumed
per week or month. Another method utilises what is called the
"Period-Specific Normal Week Scale", for which respondents
are asked about their consumption of certain types of alcohol
(beer, wine and spirits) either during a "normal week"
or subdivisions of such a week (for instance Mon-Thu, Fri, Sat,
Sun). Kühlhorn and Leifman (1993) have shown how the coverage
rate (i.e. the reported consumption's proportion of the estimated
true consumption in a sample) ranges between 28 and 75 percent
when this kind of method is employed. The rate varies with how
the week is divided up, and is higher when weeks are divided into
a greater number of reporting units.
A further method is called "Last Seven Days Recall".
Respondents are asked about their consumption over the seven days
prior to day of interview (Kühlhorn, in prep.). The method
has the advantage that participants are confronted with a concrete
situation rather than presented with a "normal" or "typical"
week. According to Kühlhorn, both "Last Seven Days Recall"
and the "Period Specific Normal Week Scale" can be supplemented
by the QF method so as better to capture the intake of low consumers.
In fact, "Last Seven Days Recall" is a special (short) version of what Sobell and Sobell (1992) have called the "Timeline Follow Back" (TLFB) technique. Respondents are provided with a daily calendar, and instructed to estimate their consumption retrospectively on a day-by-day basis. According to the authors, this may be possible for a period as long as 3 or even 18 months.
Sobell and Sobell have largely applied this technique to respondents on treatment programs, which is probably why they propose such long Timeline (TL) periods. But, although a TL of several months might be appropriate for a treatment group (whose members are likely to be eager to recall their consumption), it may well be optimistic to expect a non-patient group to provide reliable estimates.
One disadvantage of the QF method (and any related technique) lies in the small number of items (usually 6 items) it covers, with the consequence that internal consistency is low. That said, the method can be, and usually is employed at an aggregated level to measure consumption in a total population. Then, the problem of low internal consistency is less severe, since this is compensated for by having a large number of individuals.
TLFB, by contrast, has the advantage that internal consistency can be expected to be high. But, since respondents might find it hard to remember consumption several weeks or months back in time, there may be a major problem of recall. An ideal TL at an individual level would be one that showed both satisfactory reliability and high validity with regard to actual recall.
Several previous studies have shown that TLFB has advantages in terms of reliability and coverage rate (see Sobell et al., 1986; Sobell et al., 1988; Breslin et al, 1996; and the overview in Sobell & Sobell, 1992). But the method also has the manifest disadvantage that it is time consuming. According to Sobell and Sobell (1992), a single test can take between 10 and 30 minutes according to length of TL. Given the TL periods that Sobell and Sobell envisage, however, even testing times in this range may be practically impossible. Accordingly, there is much to gain if TL length can be reduced. The differences between the Timeline proposed by Sobell and Sobell (1992) and the one used in this study is that the one used in this study is shorter and based on a normal, representative sample rather than persons undergoing treatment.
The purpose of the study was to investigate the reliability of a 28-day Timeline Follow Back (TLFB) technique and what possibilities subjects had to recall consumption back in time.
Data for a 28-day TL were gathered in a health interview conducted
as part of a wider-ranging birth-to-maturity study covering 212
individuals (122 males and 90 females). Ninety-one percent of
the individuals encompassed by this wider study, and who were
still alive at the age of 36 (ranging between 34-37), took part
in the data collection. Analysis in this study was based on valid
data for 105 males. The original sample has been shown to be representative
of all Swedish persons on a wide range of demographic variables
(Karlberg et al., 1968). It is important to notice that the study
is not based on a problem drinking sample but on a sample
containing a normal variety of drinking habits (see Wennberg,
Andersson & Bohman, 1998). In total, five individuals had
received treatment due to drinking problems.
Respondents were provided with a daily calendar, and asked about the last day they consumed alcohol. Starting with that day, they were requested to estimate their consumption day-by-day back in time. Reported consumption (of beer, wine, spirits, etc.) was converted by the researchers into units of gram pure ethanol per day for the last 28 days back in time. The interviews conducted, were evenly distributed over the year. This gave us reason to conclude that variations in the Timeline reports were caused by the subjects ability to recall consumption, rather than by the variation in consumption over the year.
The analyse of data included calculations of internal reliability in terms of Cronbach alpha and split-half reliability using the drinking days as the unit of analysis. Further, a non-parametric regression and an "ordinary" linear regression were employed to investigate whether the reported consumption dropped back in time. The non-parametric regression presented is based on a moving average, and the advantage with this method is that it is not dependent on the shape on underlying data. All data analysis was performed with the computer program SPSS version 6.1.
Internal consistency and split-half reliability for the 28-day
period were calculated. Cronbach alpha was 0.84 and split-half
reliability was 0.88.
A non-parametric regression (Loess; 50% of observations, 5 iterations) was employed to illustrate the subjects' ability to effectively recall consumption back in time. The result of this is shown in Figure 1. As shown by Figure 1, the reported consumption from the day of the interview (day 28 in the figure) is relatively stable for 14 days back in time. However, there is a severe drop in consumption around 14 days back in time. A comparison of the week before the interview and the fourth week back in time from the interview showed that the consumption was 18 percent lower in the fourth week back in time compared to the week before the interview.
Further, an ordinary linear regression (R=0.46; p<0.05) also showed a drop in consumption back in time from the interview. An estimation based on the regression function indicated that, for every additional day back in time, the reported consumption dropped by 0.19 g (i.e. around 1.5 % per day).
Figure 1. Reported consumption over 28 days (day 28 closest
to the interview) for males (n=105) at age 36 (dots representing
The purpose of this study was to test a 28-day Timeline with regard
to consumption recall over time and reliability.
A possible disadvantage of TLs of the length earlier described by Sobell and Sobell (1992) is that respondents might find it hard to recall their consumption back in time. The findings of the present study suggest that consumption can be reported for 14 days without any severe drop but that there is quite a heavy under-reporting bias when the TL exceeds 14 days. The 28-day TL showed satisfactory results with regard to split-half reliability and internal consistency. As the interviews were evenly distributed throughout the year, the possibility that the systematic changes in reported consumption were due to specific holidays, such as Christmas, can be ruled out. In fact, we had no reason to believe that the systematic drop in reported consumption had any other cause than the subjects' lack of ability to effectively recall consumption. Certainly, this does not imply that the level of reported consumption one day back in time had a 100 percent coverage rate of the true consumption.
Population surveys are generally intended to obtain a fair test at a reasonable cost. A 14-day Timeline administered by a professional interviewer to a single individual is likely to take 5-10 minutes. In terms of time and cost, this is probably more realistic than the type of TLs proposed by Sobell and Sobell (1992). Such a TL also seems more realistic with regard to what people can actually recall.
Babor, T. F. & Del Boca, F. K. (1992). Just the facts. Enhancing
measurement of alcohol consumption using self-report methods.
In R. Litten & J. Allen, Measuring alcohol consumption
(pp. 3-19). Humana Press.
Babor, T. H., Kranzler, H. R. & Lauerman, R. J. (1989). Early
detection of harmful alcohol consumption: Comparison of clinical,
laboratory and self-report screening procedures. Addictive
Behaviors, 14, 139-157.
Breslin, C., Sobell, L. C., Sobell, M. B. & Buchan, G. et
al (1996). Aftercare telephone contacts with problem drinkers
can serve a clinical and research function. Addiction, 91,
Karlberg, P., Klackenberg, G., Klackenberg-Larsson, I., Lichtenstein,
H., Stensson, J. & Svennberg, I. (1968). The development of
children in a Swedish urban community. A prospective longitudinal
study. Acta Paediatrica Scandinavica Supplement, 187.
Knupfer, G. (1966). Some methodological problems in the epidemiology
of alcohol beverage usage: Definition of amount of intake. American
Journal of Public Health, 56, 973-986.
Kühlhorn, E. (in prep.). Problem och nya grepp i kartläggningar
av alkoholkonsumtionen och alkoholskadorna [Problems and new approaches
in surveys of alcohol consumption and related injuries]. In Sverige
alkoholen och Europa [Sweden, alcohol and Europe], forthcoming.
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, 54,
Sobell, L. C. & Sobell, M. B. (1992). Timeline Follow Back.
A technique for assessing self-reported alcohol consumption. In
R. Litten & J. Allen, Measuring alcohol consumption
(pp. 41-72). Humana Press.
Sobell, L. C., Sobell, M. B., Leo, G. I. & Cancilla, A. (1988).
Reliability of a Timeline method: assessing normal drinkers' reports
of recent drinking and a comparative evaluation across several
populations. British Journal of Addiction, 83, 393-402.
Sobell, M. B., Sobell, L. C., Klajner, F., Pavan, D. & Basian,
E. (1986). The reliability of a Timeline method for assessing
normal drinker college students' recent drinking history: Utility
for alcohol research. Addictive Behaviors, 11, 149-161.
Wennberg, P., Andersson, T. & Bohman, M. (1998). The association
between different aspects of alcohol habits in adolescence, early
adulthood and early middle age: A prospective longitudinal study
in a representative cohort of males and females. Submitted.