The validity of self-reports in alcoholism research

Vol. 7, pp. 123-132, 1982 Printed in the USA. All rights reserved.

AddictiveBehaviors,

0306-4603/82/020123-10$03.00/O

Copyright e 1982 Pergamon Press Ltd

THE VALIDITY OF SELF-REPORTS IN ALCOHOLISM RESEARCH J. MICHAEL The Rand Corporation,

POLICH*

Santa Monica, California

It is often assumed that many alcoholics underreport their drinking and behavioral problems. Nonetheless, previous studies using official records and collateral reports suggest that self-reports of concrete drinking problems are not biased, and that overreports equal or exceed underreports. New data are presented, based on collateral reports and blood alcohol measures for 632 alcoholics interviewed four years after treatment. Results indicate that the subjects accurately reported abstention and major alcohol-related events, such as jail terms and hospitalization. Compared with estimates from blood alcohol measures, 35% of recent drinkers underreported their consumption during the 24 hours before the interview, and 24% underreported their consumption during the previous month. However, an overall outcome classification based on a combination of consumption and other measures was not substantially affected by errors in consumption reports. These findings indicate that most types of selfreports are valid, and that broadly based outcome measures are not likely to be significantly biased by underreporting errors. Abstract-

Alcoholism research relies heavily on patient self-reports, because many aspects of alcoholism can be observed only by the person affected. Nonetheless, traditional models of alcoholism characterize “denial” as an important feature of the disorder, leading to a presumption that patients distort self-reported information (Gerard & Saenger, 1966). In a comprehensive review of evaluation methodology, Hill and Blane (1967) noted that this makes “evaluation based only on self-report somewhat suspect.” Despite these widely held beliefs, numerous empirical studies indicate that most selfreports are valid. The most compelling evidence comes from studies comparing selfreports with an external criterion representing the same behavior. This paper reviews these findings and presents new validity data from a large-scale, independent evaluation of alcoholics’ self-reports four years after treatment. The results substantiate the validity of most self-reports, but also show that the degree of validity depends on the type of alcoholic behavior assessed. PREVIOUS

RESEARCH

Previous studies that have examined self-reported may be grouped three broad classes.

drinking against external criteria

1. Official record studies One method of evaluating self-reports is compare them with law enforcement or treatment records, such as records of arrests or hospitalizations (Sobell, Sobell, & Samuels, 1974; Sobell & Sobell, 1975; Sobell & Sobell, 1978; Cooper, Sobell, Maisto, & Sobell, 1980; Cooper, Sobell, Sobell, & Maisto, 1981). Most such comparisons show at least two-thirds of subjects in exact agreement with the records. Where bias is present, there are usually more cases of overreporting (where the subject admits an event that is not reflected in the record) than cases of underreporting (where the subject This research was supported in part by the National Institute on Alcohol Abuse and Alcoholism (Contract No. ADM-281-76-0006) and in part by The Rand Corporation. *Reprint requests may be addressed to J. Michael Poiich, The Rand Corporation, 1700 Main Street, Santa Monica, California 90406. 123

124

J. MICHAEL POLICH

denies an event in the record). As a result, the mean rate of alcohol problems according to self-reports is frequently greater than the mean rate according to records.’ 2. Collateral reports Numerous studies have used a report of a collateral observer, such as a family member, friend, or counselor, as an external criterion for evaluating self-reports. Guze, Tuason, Stewart, and Picken (1963) found that subjects themselves were more likely than collaterals to report a subject’s drinking problems; but other studies have found better subject-collateral agreement (Bailey, Haberman, & Sheinberg, 1966; McCrady, Paolino, & Longabaugh, 1978; Miller, Crawford, & Taylor, 1979; Maisto, Sobell, & Sobell, 1979).’ Like official record studies, collateral studies usually show little or no net bias in self-reports, but they often reveal numerous disagreements running in both directions, suggesting that random errors may be present. If so, the most important problem with self-reports might be attenuation of correlations, since random errors reduce observed correlations but do not cause bias in overall means and percentages. 3. Blood alcohol tests A few recent studies have corn’ ‘:ed alcoholics’ reports of amount of drinking with a measure of blood alcohol concentration (BAC). Armor, Polich, & Stambul (1978) studied alcoholics tested at treatment admission: they found that a mean BAC computed from self-reports was about 25% lower than the actual mean BAC. Sobell, Sobell, & VanderSpek (1979, 1978) made a similar comparison at treatment followup, but included only subjects who had a positive BAC reading; they found that among 25 subjects who had alcohol in their blood, 52% (n = 13) underreported their amount of consumption. However, this 52% underreporting rate probably exaggerates the proportion of underreporters in a clinical followup population, because the method automatically excludes any subjects who drank moderate amounts that would have been metabolized by the time of the BAC test. Jalazo, Steer, & Fine (1978) also found a tendency toward underreporting among positive-BAC subjects who had been arrested for driving while intoxicated. General population surveys provide other evidence that some people underreport alcohol consumption. Survey-based estimates of national alcohol consumption understate official beverage sales records by about 50% (Armor et al., 1978; Room, 1971). Some of this discrepancy could be due to a lack of complete sampling coverage, because household surveys eliminate some high-consuming populations, such as transients and dormitory residents. Room (1971) concluded that more accurate survey questions could increase the coverage rate to at least 67%, and Polich and Orvis (1979) found a coverage rate of 83% in a sample of U.S. Air Force personnel. However, these results still suggest that self-reports could understate true consumption by 20% or more. METHODS

Data on the validity of self-reports were collected as part of a four-year followup of ’ For example, results reported in Sobell, Sobell. and Samuels (1974) show a total of 548 self-reported arrests among 70 subjects (mean = 7.8 per subject), but only 451 arrests were found in the official files (mean = 6.4). *Other studies have reported partial data from collateral reports. Gerard and Saenger (1966) stated that at a treatment followup. 3 of 125 subJects claiming abstinence and 2 of 34 subjects claiming controlled drinking were contradicted by collaterals. However, no data on overreports were presented. Knupfer (1967) and Orford (1973) compared spouse reports with subjects’ reports of drinking problems, but in both cases the subject was asked about specific events while the spouse was asked for a more general judgment.

Validity of self-reports

125

alcoholics at public treatment facilities. Subjects were randomly selected from persons who made contact in 1973 with one of eight treatment centers sponsored by the National Institute on Alcohol Abuse and Alcoholism (Polich, Armor, & Braiker, 1980, 1981). The original random sample contained a total of 922 male alcoholics, including 758 who were admitted to treatment and 164 who did not enter treatment but who were judged alcoholic by the treatment staff. In 1977, followup data were collected on this sample. Of the 922 cases, 113 were found to be deceased at the four-year followup, and 809 were presumed living; among these 809 cases, 668 were interviewed. These 668 subjects constitute the sample for whom self-reports can be compared with external criteria. Of the 141 subjects not interviewed, the largest group (88 cases) could not be located. Nonrespondents were very similar to respondents in baseline characteristics, and projections of possible bias showed that the sample results were unlikely to be affected to any appreciable extent by nonresponse bias (Polich et al., 198 1, Appendix A). The data were collected by Johns Hopkins University interviewers, who were independent of the treatment facilities. Subjects were assured that no individual data would be reported to the treatment facility or to anyone else except the research staff. Compensation of $10 was offered for the 90-minute interview. (Documentation of procedures and instruments is shown in Polich et al., 1981.) The interview ascertained the occurrence of major alcohol-related events, such as jail terms and hospitalization, in the past 6 months. If the subject had been drinking in the past 6 months, he was also asked the frequency with which he had alcohol dependence symptoms (tremors, morning drinking, etc.), and the amount of beer, wine, and distilled liquor he consumed on “typical” drinking days in the 30 days before his last drink. In addition, detailed questions were asked about the amount of alcohol consumed and the times when drinking began and ended during the day of the interview (“today”) and the day before the interview (“yesterday”). After the interview, all subjects were asked to take a breath test that would shov the amount of alcohol in the blood.3 Additional compensation of $5 was offered for the test. A predesignated random subsample of 164 interviewees was also asked to name a collateral (“a person who knows you well”) who could give another report on the subject’s drinking. Of the 668 respondents, 95% (n = 632) provided a usable breath sample. Of the 164 respondents designated for collateral interviews, 16 named collaterals who lived more than 100 miles from the interviewing site, which was too far away to make a collateral interview economically feasible; collateral interviews were attempted for all of the 148 remaining subjects. Of these 148 attempted collateral interviews, 86% (n = 128) were completed. The rest were not completed because the staff could not reach them at home; no collateral refused to be interviewed. The collateral interview contained questions directly parallel with the questions asked of subjects. Most collaterals were persons who had a close relationship with the alcoholic: 32% were wives; 3Breath tests to measure blood alcohol concentration (BAC) were taken using a portable device, the model SM-7 manufactured by Luckey Laboratories, San Bernardino, California. The device consists of a balloon and a volumetric bag attached to a glass tube. When used, 2100 cc of alveolar breath passes through the tube, which collects any alcohol that is present in the breath. The tube contents were analyzed by Valley Toxicology of Davis, California, a forensic laboratory licensed by the State of California. Two samples from each tube were independently analyzed by gas chromatograph. The laboratory also conducted a simultaneous analysis of a standard with each tube to ensure proper calibration. Results of breath tests have been shown to be reliable indicators of blood alcohol Levels(Glendening, Rush, & Duffett, 1971). In addition, the present study included a preliminary experiment in which nine untrained subjects used the SM-7 device under conditions simulating an interview. The results yielded a correlation of .996 between the breath-test estimate and a blood-test measurement of BAC (Polich et al., 1981, Appendix B).

126

J. MICHAEL POLICH

24% were other relatives living in the same area; 37% were friends or co-workers; 4% were alcoholism counselors; and 3% were landlords. COLLATERAL

INTERVIEW

RESULTS

Both subjects and collaterals were asked a broad range of questions to assess the subject’s drinking behavior over the 6 months before the interview. Table 1 shows results of comparisons between subject and collateral reports for that period. Among the 128 subjects for whom collateral interviews were campleted, 92 subjects (72%) reported that they drank some alcohol during the preceding 6 months. Only 82 collaterals (64Vo) reported that the subject drank in the same period. Hence, the picture given by collaterals tends to understate the subject’s involvement with alcohol, rather than revealing subject distortions. Indeed, in only one case did the subject deny drinking while the collateral reported it. This “subject underreport” constitutes 1% of the sample, as shown in the summary percentages in the right-hand portion of Table 1. By contrast, in 11 cases the subject admitted drinking while the collateral either denied it or was unsure. If the collateral information were accepted as an absolute criterion, these cases, constituting 9% of the sample, would be “subject overreports.” Thus, there is a high degree of agreement between alcoholics and their collaterals about whether or not the alcoholic drank, and the few cases of disagreement indicate more error in the collateral reports than underreporting by the subject. A similarly high level of agreement was found for overt alcohol-related problems among drinkers. As Table 1 shows, nearly 90% of the collaterais agreed with subjects about such visible events as being jailed or hospitalized-because of drinking. For these concrete problems, the rate of subject underreporting was 6% or less, and there was more overreporting than underreporting. In addition, few collaterals were uncertain about the occurrence of these events. However, the picture is quite different for alcohol dependence symptoms, such as tremors or morning drinking, because of uncertainty on the part of the collaterals. When asked whether or not the subject had tremors, 32% of the collaterals were unsure (29 cases); when asked about morning drinking, 39% were unsure (36 cases). Nonetheless, the discrepancy rates indicate more overreporting (29 to 45%) than underreporting (8 to 15070).~It is notable that most of the collaterals who were uncertain were reporting on a subject who had already admitted the symptom in question. The source of this uncertainty may be that the collaterals did not observe enough of the alcoholic’s behavior to make a report. A similar degree of uncertainty (26% unsure) was found when collaterals were asked to describe the subject’s typical level of alcohol consumption. It appears that when questions ask for highly visible events or global judgments, collaterals provide useful validating data; however, for more detailed measures collaterals do not make effective validators. RESULTS

OF

A bstention versus drinking A simple test of a respondent’s

BLOOD

ALCOHOL

TESTS

candidness is to compare

self-reports

of recent

‘The summary percentages in the right-hand portion of Table 1 were computed by treating a collateral “unsure” as a negative reply, since the collateral failed to report the symptom. This method illustrates the important fact that more subjects than collaterals reported symptomatic drinking by the alcoholic. However, other alternative methods could be used, such as omitting all pairs where the collateral was unsure. That method would omit many cases where the subject admitted symptoms, and hence would understate the alcoholics’ readiness to give information. By restricting attention to those pairs where the collateral was certain, it would show more underreports than overreports for tremors, but more overreports than underreports for morning drinking.

(on awakening)

I.

9

81 IO 20 14

colt. Yes

I5

5 5 6 9

Coil. No

Subject Yes

26

6 0 I I8

Coil. Unsure

I

25

35 66 55 26

I 5 4 I4

Coil. No

Subject No colt. Yes

Number of Cases

IO

0 4 6 II

Coil. Unsure

48

91 89 88 55

Percent in Agreementh

45

9 6 8 29

Overreport

8

I 6 4 I5

Underreport

Percent Discrepant’

Percent of Sample”

Subject and collateral reports of alcoholics’ drinking behavior at four-year followup.

___

__

-_---

aAgreements and discrepancieswere calculated by treating a collateral “yes” as an aftirmativc. and a collateral “no” or “unsure” as a negative. hAgreemenI indicales either (a) both collateral and subject were affirmative. or (b) both subject and collateral were negative. ‘Overreport indicates the subject was aflirmative but the collateral was negative. Underreport indicates the subject was negative but the collateral was affirmative. dFor the question about any drinking versusabstention in the past 6 months, results are shown for all subj&ts with a collateral interview (n = 128). For the other questions, which ask about drinking-related problems, results are shown only for those Fobjects who reported drinking in the past 6 months (n = 92).

Any morning drinking

Any drinkingd In jail, related to drinking In hospital, related to drinking Any tremors or “shakes”

Drinking Behavior (past 6 months)

Table

128

J. MICHAEL

POLICH

abstention with measurements indicating the presence or absence of ethanol in the blood. To do this, we asked each subject the date of his last drink. Among those who reported their last drink as occurring 6 months ago or earlier (n = 164), only 1% (2 cases) had a positive BAC; among those reporting the last drink 30 days to 6 months ago (n = lOO), 6% (6 cases) had a positive BAC; and among those reporting the last drink 2 to 29 days ago (n = 148), 5% (8 cases) had a positive BAC. In contrast, among those reporting a drink in the past 24 hours (n = 220), 59% (129 cases) had a positive BAC. These results indicate that self-reports of abstention are highly valid.5

Amount of alcohol consumption For a test of the accuracy of self-reported amounts of consumption, we compared self-reports with BAC levels among all subjects who reported any drinking in the 24 hours before the interview (Table 2). The actual BAC measure was derived directly from laboratory analysis of the breath samples. The self-report estimate of BAC was derived by computing the BAC that the subject should have had if his reports of the timing and quantity of his recent consumption were correct.6 The method considers the I Among the 91 cases who drank in the past 24 hours but had zero BACs, all but 13 reported drinking at such times that they should have eliminated all ethanol by the time of the breath test. The method used for estimating their expected BAC level is discussed below. ‘The self-report estimate of the BAC was derived by converting the self-reported quantity of ethanol consumed today into a BAC due to “today’s drinking,” and then reducing that BAC by the amount of expected elimination between the onset of today’s drinking and the time of the breath test. The estimate of BAC was computed by the formula B = Q/(.14W) - TE, where B is the estimated BAC in percent, Q is the selfreported quantity of ethanol consumed today in fluid ounces, W is body weight in pounds, T is the elapsed time in hours between the onset of today’s drinking and the time of the breath test (reduced by one hour to allow for absorption), and E is the average eiimination rate for alcoholics (estimated to be .02 percent per hour in studies cited by Wallgren and Barry, 1970). An additional adjustment was made to add an increment to the BAC for yesterday’s consumption if the subject’s report of yesterday’s pattern of drinking implied that he had a nonzero BAC level at the onset of today’s drinking. This procedure is parallel in principle to methods used in an earlier analysis of these same data (Polich et al., 1981), but it incorporates some refinements.

Self-report versus breath test estimates of blood alcohol concentration

Table 2. Self-Report Estimate of

BAC From Breath Test

BAC”

0

0

.Ol - .02 .03-.04 .OS-.06 .07 - .08 .09-.I0 .11-.20 Over .20 Total

.Ol-.02

=

.03-.04

78 IO 0 0 0 2 0

9 3 4 0 0 0 1

I

0

10 2 2 3 0 0 0 2

91

17

19

Mean, Self-Report = .044 Mean, Breath Test = .057 Correlation

(number of cases)

.358

.05-.06

10

.07-.08

20

.09-.I0

.ll-.20 6 1 I 1 0 0 0 0

16 3 3 4 3 I 5 4

9

39

Over .20

Total 139 27 11 IO 5 4 10 14

15

220

Summary Statisticsb Agreement (within .04) = 58% Underreports (.04 or more) = 35% Overreports (.04 or more) = 7%

aBAC level, in percent (g/ 100 ml). implied by the subjecr’s self-report of the amount consumed today, yesterday. and timing of drtnking periods. Assumes an alcohol elimination rate of .02% per hour. bBased on uncategorized data (continuously distributed. not grouped as in the table).

amount

consumed

Validity of self-reports

129

subject’s self-reported quantity of. consumption, body weight, and hours available for elimination, but it assumes a constant rate of elimination for all subjects. Although individual variations in elimination rate could create errors in the estimated BAC, previous analysis (Polich et al., 1981) found that the rate of underreporting was not changed by more than 5 percentage points when the assumed elimination rate was either increased or decreased within reasonable ranges. Underreports are represented by cases in the upper right portion of Table 2, where the self-report BAC is lower than the actual BAC. Overreports are represented by cases in the lower left portion of the table. The location of cases reveals a general tendency toward underreporting. This is confirmed by the means, which indicate an average BAC of .044 by self-report and .057 by breath test.’ * A large group of cases fall near the diagonal, showing instances of fairly close agreement between the self-report and the breath test. To analyze these cases further, the difference between the self-report and the breath test measure was computed, and only those cases showing a discrepancy of .04 or more were treated as errors. (For a 165 pound male, a BAC of .O4 corresponds to about two average-size drinks, i.e., approximately .92 fluid ounces or 21 grams of ethanol.) Using this criterion, the data indicate that 35% of the recent drinkers underreported their consumption, 7% overreported, and 58% were consistent with the BAC. Therefore, although most of the alcoholics appeared to accurately report their consumption in the past 24 hours, a substantial group underreported by a significant amount. It is important to note that these results apply only to alcoholics’ self-reports of their consumption during the past 24 hours. Subjects tended to report lighter drinking in the past 24 hours than on typical drinking days over the past month (a mean of 2.1 ounces of ethanol on the interview day, 5.0 ounces on the day before the interview, and 6.2 ounces on typical drinking days). Thus, the alcoholic’s report of his drinking in the past 24 hours frequently did not square with his account of his “typical” drinking in a l-month period. In contrast, when self-reports of typical drinking in each of the 6 months preceding the interview were examined, results showed that 6-month and l-month assessments were remarkably consistent (Polich et al., 1981, Chapter 3). Therefore, it seems that a self-report of consumption during the last month of drinking is probably adequate to capture Gmonth variations in reported consumption patterns, and is preferable to an assessment covering only the past day or two. For that reason, BAC results were also compared with the interview data for the past month. Each subject’s “actual quantity consumed today” was estimated from the BAC and elimination data.’ If the alcoholic’s report of his typical daily consumption in the past month was less than the actual quantity he consumed today, it was judged that he was probably underreporting. By this criterion, 24 percent of the subjects in Table 2

‘There are a few cases, located in the bottom row of Table 2, for which the estimated BAC is much greater than the actual BAC. All of these cases reported very large quantities of consumption on the day of the interview or the previous day (generally, 10 ounces of ethanol or more), leading to very high estimated BACs (in some instances, over .50). This suggests that these subjects had a true elimination rate substantially higher than the mean rate of .02, which was assumed for all subjects. However, it is probable that there exist other subjects whose true elimination rate is lower than the assumed rate of .02. This appears especially likely because the mean elimination rate for normal subjects is .015, as cited by Wallgren and Barry (1970). Because such errors in the elimination rate presumably run in both directions, it was decided not to eliminate these cases from the analysis. ‘The a&al quantity consumed today was estimated by a method parallel to that in footnote 6, i.e., by the formula Q = .14W(B + TE), where Q is the estimated actual quantity, W is weight, B is the actual BAC, T is time elapsed, and E is the elimination rate as defined in footnote 6.

J. MICHAEL

130

POLICH

underreported their consumption in the past month.9 Since most studies would use a period of at least one month to assess drinking, this 24% rate of underreporting may be the best estimate for use in evaluation studies at present. This underreporting rate had only a small effect on overall assessments of subjects’ drinking behavior. This is important, because pessimistic notions on this point are abundant, and some observers believe than any outcome assessment based on self-reports is questionable (Hill & Blane, 1%7). To investigate that issue in this sample, we constructed an overall classification of each subject’s drinking status based on (1) the presence or absence of any alcohol dependence symptoms (including tremors, morning drinking, blackouts, missing meals, loss of control over drinking, and continuous drinking over 12 hours); (2) the presence or absence of any adverse consequences of drinking (including alcohol-linked health problems, hospitalization, law enforcement incidents, and work and interpersonal problems); and (3) the typical amount of alcohol consumption, classified as under 2 ounces per day versus 2 ounces or more (Polich et al., 1980). Subjects who reported any symptoms or consequences were classified as problem drinkers (54% of the sample), while subjects without such problems were classified as either low consumers (8%) or higher consumers (loal,). The remaining 28% were abstainers. It is clear that self-report errors in consumption can make only a small difference in this classification. Indeed, even the most extreme assumptions about underreporting lead to an estimate that one-fourth of all low consumers- that is, 2% of the sample- could be misclassified because of underreporting (Polich et al., 198 1, Appendix B).‘O Thus, one should not overstate the impact of response errors. However, these results do suggest that it is advisable to use a multiple assessment strategy to determine outcomes, including several different measures of a subject’s drinking behavior. DISCUSSION

AND

CONCLUSIONS

Many observers familiar with alcoholism assume that alcoholics underreport or distort their drinking behavior. However, relatively few studies have tested this assumption by making an explicit comparison between the alcoholic’s self-report and an external criterion. The great majority of such studies have employed official records or collateral reports. Their results indicate that self-reports of concrete drinking problems are not biased. In fact, the number of overreports frequently equals or exceeds the number of underreports. The present study extends and modifies these findings in several ways. First, this study represents outcomes that were measured long after treatment (4 years) and were 9Underreporters for the past month were defined as those cases for which the difference between the actual quantity consumed today and the self-reported typical quantity consumed on drinking days in the past month was I ounce of ethanol or more. Fifty-three of the 220 cases met this criterion. Of these 53, 50 also underreported their consumption during the past 24 hours. Moreover, the discrepancies were usually large; in 45 of the 53 cases the past-month report was more than two ounces beneath the actual quantity consumed today (a discrepancy equivalent to four drinks). loThe proportion of underreporters was nearly constant (about 25%) for all levels of self-reported consumption, and was the same for those who reported problems as for those who reported no problems. Therefore, the sensitivity analysis assumed that 25% of all nonproblem drinkers who reported low consumption were in fact drinking much more. Only eight percent of the sample reported low consumption without problems. Thus only two percent of the sample (.25 x .OE) would be misclassified as low-consuming nonproblem drinkers as a result of self-report errors. The threshold of 2 ounces (59 ml) of ethanol was adopted to be a conservative limit for moderate drinking, corresponding to 4 drinks. However, if a higher limit had been adopted (such as 5 ounces), the sensitivity analysis would reveal a similar result: in that case, 13% of the sample would have been classified as lowconsuming nonproblem drinkers, and 3% of the sample (.25 x .l3) would be misclassified because of selfreport errors.

Validity of self-reports

131

assessed by an independent organization that was not connected with the treatment providers. Most other validity studies are more limited because they were (1) based on small samples, (2) carried out during treatment, (3) conducted in a clinical setting, or (4) carried out by personnel connected with a treatment facility. The consistency between this four-year study and previous validity studies (e.g., Sobell and Sobell, 1978; Maisto et al., 1979) suggests that self-reports of concrete drinking problems are generally valid, regardless of the research environment. However, the results reveal significant differences in validity among different types of self-reports. Collaterals and alcoholics agreed quite closely when questioned about abstinence, or about major alcohol-related events such as jail terms and hospitalization. About one-third of collaterals were unsure as to whether or not the alcoholic had suffered symptoms of alcohol dependence. Thus, evidence for the validity of symp toms was inconclusive. However, the alcoholic was more likely than the collateral to report symptoms and other types of alcohol problems. Results of blood alcohol measurements confirmed that self-reports of abstinence are highly valid. In contrast, some drinkers tended to understate their amount of recent consumption. Compared with a blood alcohol measure, 35% underreported their consumption in the past 24 hours, 7% overreported, and 58% were consistent. The preponderance of underreporting in the consumption data is consistent with studies done among alcoholics (Armor et al., 1978; Maisto et al., 1979; Sobell et al., 1979) and among normal populations (Polich & Orvis, 1979; Room, 1971). However, many subjects who underreported their drinking during the past 24 hours admitted heavier drinking on “typica1” days in the past month. Therefore, the proportion who underreported their drinking in the past month was only 24%. Moreover, under reporting of consumption made little difference when multiple measures were used to assess patient outcome; sensitivity analyses suggested that, at most, 2% of the sample were misclassified because of underreporting errors. These results suggest that alcoholics’ reports of most types of behavior may be used with confidence, but that multiple self-report measures are desirable to maximize the accuracy of outcome classifications. REFERENCES Armor, D.J., Polich, J.M., & Stambul, H.B. Alcoholism and Treatmenf. New York: Wiley, 1978. Bailey, M.B., Haberman, P.W., & Sheinberg, J. Identifying alcoholics in population surveys: a report on reliability. Quarterly Journal of Studies on Alcohol, 1966, 27, 300-315. Cooper, A.M., Sobell, M.B., Maisto, S.A., & Sobell, L.C. Criterion intervals for pretreatment drinking measures in treatment outcome evaluation. Journal of Studies on Alcohol, 1980, 41, 1186-l 195. Cooper, A.M., Sobell, M.B., Sobell, L.C., & Maisto, S.A. Validity of alcoholics’ self-reports: Duration data. international Journal of the Addictions, 1981, 16, 401-406. Gerard, D.L., & Saenger, G. Out-patient treatment of alcoholism. Toronto: University of Toronto Press, 1966. Glendening, B.L., Rush, A.C., & Duffett, N.D. The use of small gas chromatographs for breath testing of drunken drivers by the Highway Patrol utilizing the Public Health Laboratory (Kansas), Health Laborarory Science, 1971, 8, 131-141. Guze, S.B., Tuason, V.B., Stewart, M.A., & Picken, B. The drinking history: a comparison of reports by subjects and their relatives. Quarterly Journal of Studies on Alcohol, 1963, 24, 249-260. Hill, M.A., & Blane, H.T. Evaluation of psychotherapy with alcoholics. Quarterly Journal of Studies on Alcohol, 1967, 28, 76-104. Jalazo, J., Steer, R.A., & Fine, E.W. Use of breathalyzer scores in the evaluation of persons arrested for driving while intoxicated. Journal of Studies on Alcohol, 1978, 39, 1304-1307. Knupfer, G. The validity of survey data on drinking problems: A comparison between respondents’ selfreports and outside sources of information. Social Research Group, University of California, Berkeley, unpublished manuscript, 1967. McCrady, B.S., Paolino, T.J., & Longabaugh, R. Correspondence between reports of problem drinkers and spouses on drinking behavior and impairment. Journal of Studies on Alcohol, 1978,39, 1252-1257.

132

J. MICHAEL

POLICH

Maisto, S. A., Sobell. L.C., & Sobell, M.B. Comparison of alcoholics’ self-reports of drinking behavior with reports of collateral informants. Journal of Consulting and Clinical Psychology, 1979, 47, 106-l 12. Miller, W.R.. Crawford, V.L., & Taylor, C.A. Significant others as corroborative sources for problem drinkers. Addictive Behaviors, 1979, 4, 67-70. Morse, R.M., & Swenson, W.M. Spouse response to a self-administered alcoholism screening test. Journal of Studies on Alcohol, 1975, 36, 400-405. Orford, J. A comparison of alcoholics whose drinking is totally uncontrolled an& whose drinking is mainly controlled. Behaviour Research and Therapy, 1973, 11, 565-576. Pokomy, A.D., Miier, B.A., & Cleveland, S.E. Response to treatment of alcoholism: A follow-up study. Quarterly Journal of Studies on Alcohol, 1968, 29, 364-381. Polich, J.M. & Orvis, B.R. Alcohol Problems: Patterns and Prevalence in the U.S. Air Force. Santa Monica: The Rand Corporation, 1979. Polich, J.M.. Armor, D.J., & Braiker, H.B. Patterns of alcoholism over four years, Journal of Studies on Alcohol, 1980, 41, 397-416. Polich, J.M., Armor, D.J.. & Braiker, H.B. The Course of Alcoholism: Four Years After Treatment. New York: Wiley, 1981. Room, R. Survey vs. sales data for the U.S. Drinking and Drug Pmctices Surveyor, 1971, 3, 15-16. Sobell. L.C., & Sobell, M.B. Outpatient alcoholics give valid self-reports. Journal of Nervous and Mental Disease, 1975, 161, 32-42. Sobell, L.C., & Sobell, M.B. Validity of self-reports in three populations of alcoholics, Jourmi of Consulting and CIintcal Psychology, 1978, 46, 901-907. Sobell. M.B.. Sobell. L.C., & Samuels, F.H. Validity of self-reports of alcohol-related arrests by alcoholics. Quarterly Journal of Studies on Alcohol, 1974, 35, 276-280. Sobell. M.B., Sobell, L.C. & VanderSpek. R. Relationship between clinical judgment, self-report, and breath analysis measures of intoxication in alcoholics. Unpublished manuscript, 1978. Sobell, M.B., Sobell. L.C., & VanderSpek, R. Relationship among clinical judgment, self-report, and breath analysis measures of intoxication in alcoholics. Journal of Consulting and Clinical Psychology, 1979, 47, 204-206. Wallgren, H., & Barry, H. Actions of Alcohol, Vol. 1. Amsterdam: Elsevier, 1970.