UNIVERSITY OF

ILLINOIS LIBRARY

AT URBANA CHAMPAIGN

300KSTACKS

Digitized by tine Internet Archive

in 2011 with funding from

University of Illinois Urbana-Champaign

http://www.archive.org/details/howtoaskquestion312blai

Faculty Working Papers

HOW TO ASK QUESTIONS ABOUT DRINKING AMD SEX: RESPONSE EFFECTS IW MEASURING CONSUi'IER BEHAVIOR

Ed Blair and Seymour Sudtoan

#312

College of Commerce and Business Administration

University of Illinois at Urbana-Champaign

FACULTY WORKING PAPERS College of Conunerce and Business Administration University of Illinois at Urbana-Champaign itoy 14, 1976

HOW TO ASK QUESTIONS ABOUT DRINKIi^IG AND SEX: RESPONSE EFFECTS IN MEASURING CONSUi'IER BEHAVIOR

Ed Blair and Seymour Sudnan

#312

Market researchers often encounter response effects in measuring consumer behavior. This paper reports a nationwide study of response effects. Almost 1200 respondents answered threatening behavioral ques- tions presented in various formats* Study results indicate that threat- ening questions requiring yes or no answers can be asked in any format, but that threatening questions requiring quantitative answers should be asked in open-ended, long questions with respondent- familiar wording.

How to Ask Questions About Drinking and Sex: Response Effects in Measuring Consumer Behavior

Ed Blair and Se)Tnour Sudman

Introduction

Market researchers tend to assume that errors of measurement are var- iables with a mean of zero and independent of the "true" score. It is assigned that there is some sort of error attached to a particular measurement of each individual, but that over a sample of homogeneous individuals, the mean of the measurement errors will tend toward zero and the group mean will thus approximate the true score for the group. Since market researchers rarely are interested in the single individual, the comfortable assumption of mea- surement error as a variable with a mean of zero can lead to ignoring prob- lems of measurement error and response effects. Response effects are defined as the difference between "true" score and obtained response, standardized by dividing th's difference by the standard deviation of response. Unfor- tunately, such an assumption often is not sensible, particularly in regard to questions which make the respondent uneasy.

Strong evidence supports the intuitive belief that response effects in surveys increase as questions become more threatening. Sudman and Bradbum (18) summarize the literature prior to 1974 on this topic, including studies by Cannell and Fowler (1), Clark and Wallin (2), Clark and Tifft (3), David (4), Ellis (5), Kahn (7), Kinsey, Pomeroy and Martin (8), Knudsen, Pope and Irish (9), Lev inger (10), Mudd, Stein and Mitchell (13), Poti, Chakraborti and Malaker (14), Sarason (IS, 16, 17), Thorndike, Hagen and Kemper (19),

-2-

U.S. National Center for Health Statistics (21), Wallin and Clark (22),

Yaukey, Roberts and Griffiths (23), and Young (24). Locander (11) and Locander,

Sudman and Bradburn (12) report more recent experiemnts.

Also, empirical evidence shows that the impact of question threat is mediated by several variables, particularly question structure and question length. Sudman and Bradburn (18) make the following conclusions about these variables and question threat. Question structure and question length do not affect response effects for non- threatening questions. For threatening ques- tions, closed-ended questions elicit negative response effects (underreporting), perhaps because closed endings increase question threat by forcing the respon- dent to choose one from a number of alternatives. Closed-ended questions also seem more sensitive to social desirability factors, resulting in depressed reporting for closed-ended questions about socially sensitive behavior or attitudes. Open-ended questions thus seem most appropriate for tlireatening topics. Response effects for threatening items decrease with increasing question length, suggesting that padded questions exceeding 30 words are most appropriate for threatening topics, and refuting the accepted maxim of making questions as short as possible. Sudman and Bradburn note that these conclu- sions necessarily generalize from several highly specific studies, and that large scale field research is needed to confirm these effects and to explore their interactions.

This paper reports a large scale investigation of question structure, question length and response effects. It also explores a new variable, wording familiarity. Sudman and Bradburn show that response effects for threatening items increase sharply with increasing average word length, a common surrogate for wording difficulty. Since standard questions which use simple, easy-to-understand words minimize response effects from this

-3-

sourcB;, a typical difficulty manipulation which increases wording difficulty is not sensible. Efforts to create a manipulation which would reduce wording difficulty led to the idea of asking respondents for their oivn words, par- ticularly where standard words were not coiamon words. This manipulation cannot be considered a simple difficulty manipulation. Familiar words may have relaxing effects which go beyond difficulty reduction in improving re- porting. Familiar words also may have threatening effects which damage reporting, particularly when a closed-form response card presents street language variations of a standard word. Previous literature offers no clues about which result to expect, and this study tentatively hypothesized that increasing familiarity should liave effects similar to decreasing difficulty. Issues of possible bias in using familiar v/ording are discussed in the metho- dology section of this paper.

H;>^q theses

This research investigated three hyi^otheses:

Hypothesis 1: Open-ended questions will elicit higher reporting for threatening behavioral items than will closed-ended questions. (Threatening behavioral items invariably elicit underreportings so that higher reporting can be interpreted as a reduction in negative response effect, rather than an increase in positive response effect. This point is substantiated in the validation of results section).

Hypothesis 2: Long questions (containing over 30 words) will elicit higher reporting for threatening behavioral items than will short questions.

Hypothesis 3: Familiar questions (defined in this study as questions using wording chosen by the respondent) will elicit Mgher reporting for threatening behavioral items than will questions employing standard, researcher- chosen wordings.

Methodology

A nationa'' sample of 1,200 adults (over 18 years of age) was drawn from the National Opinion Research Center's national master sample, using proba- bility sampling with quotas. In each of the fifty areas which fell into the sample, the best interviewer currently available to NOPX was used, to ensure maximum data quality. Interviewers who agreed to participate completed a practice case, and emj>16yiy;nt for the study was contingent upon a high-quality practice interview. 1,172 personal interviews were obtained for final analysis.

The same base questionnaire was used in all 1,172 cases. After answering questions about various leisure activities and about general happiness and well-being, respondents received questions on (in order) gambling, social activities, drinking alcohol, getting drunk, using marijuana, using stimulants and depressants, sexual activity, and descriptive items including income. Gambling, drinking, getting drunk, smoking marijuana, using stimulEints and depressants J and sexual activity were thought to be threatening topics of serially increasing threat (responses to final questions about uneasiness regarding various questionnaire sectio, s showed this presumption of threat order to be correct) . Placing these items in invariant order of increasing threat was considered necessary to minimize breakoffs and item refusals. The fixed order also was justified by Sudman and Bradburn's (18) conclusions that order effects were minor in causing response effects, and that increasing interviewer-respondent rapport should depress negative response effects on later items (thus promising a conservative test of the hypothesized negative response effects) .

Hypothesis testing occurred through a 2 x 2 x 2 factorial manipulation of threatening items on the base questionnaire. Open-ended questions versus

-5-

closed-ended questions formed a question structure manipulation. Closed-ended questions and open-ended questions were identical except for the closed-ended questions* provision of response categories. Long questions versus short questions formed a question length manipulation. Long questions and short questions were identical except for the long questions' use of at least 15 prefatory words. Familiar v/ording versus standard wording formed the final manipulation, question woi-ding. Familiar wordings and standard wordings were identical except that for the familiar wordings tlie interviewer asked the respondent to suggest the wording to be used. The 2x2x2 design re- sulted in eight distinct questionnaires, each employing one combination of factor levels throughout all threatening items.

The question m.anipulations are best illustrated through an actual exam- pie. The questionnaire contained an item asking how many times in the past year respondents had become intoxicated. In the closed, short, standard questionnaire form (the form expected to obtain the poorest reporting) , this item read:

In the past year, how often did you become intoxicated while drinking uny kind of alcoholic beverage?

Respondents were handed a card listing these response categories:

Never

Once a year or less

Every few months

Once a month

Every few weeks

Once a week

Several times a week

Daily

In the open, long, familiar forr.i (the form expected to obtain the best reporting),

respondents first provided their own word for intoxication through this item:

•6-

Sometimes people drink a little too much beer, wine or whiskey so that they act different from usual. Wliat word do you think we should use to describe people when they get that way, so that you will know what we mean and feel c mfortable talking about it?

The intoxication item then read:

Occasionaliys people drink on an empty stoniach or drink a little too much and beccme (respondent's word). In the past year, Iiow often did you becone (respondent's v;ord) while drinking any kind of alcoholic beveragt!?

No response categories were offered for either item.

There was soiae possibility that the familiarity manipulation might change the meaning of questions. A question about being high, with its broad frame of reference, may have a different meaning than a question about intoxication. The study results indicate that such shifting meanings did not occur. Descrip- tions used in soliciting respondents' words were quite specific, often using standard words such as intercourse or masturbation. Also, interviewers were instructed to use the standard word if respondents offered awkward or inap- propriate words. Shifts in meaning would have been possible only for those few respondents w}io didn't understand the standard terra, and for whom the familiar form f^efined that term.

All respondents answered one of the eight questionnaires. Each geo- graphic sf^guent sajapled in the study contained eight cases, so that every form was used once in every segment. Starting form was randomized across segments to avoid sequence effects (such as interviewer practice effects) . Interviewers tj'pically did three segments each, or three replications of each form.

-7-

Results and Discussion

The study results indicate that threatening items must be separated into two categories — those items which ask about performing a behavior even once within soma time span, and which require yes or no responses, versus those items vrfiich ask about the frequency or intensity (how often or how much) of & behavior, and which require scane quantitative answer. Questions requiring yes or no responses showed no systematic, interpretable effects tram the question length and wording familiarity manipulations (question structure was not amenable to manipvilation for these items). Quostiona requiring quantitative answers proved consis- tently sensitive to the question structure and question length manipu- lations, though generally insensitive to the wording familiarity Bianipulation. Of course, non-threatening items elicited virtually flat reporting across questionnaire manipulations.

Table 1 shows the stability of yes or no responses across form sianipulations. Percentages of respondents answering yes to the gambling items show ranges of less than 6 percentage points, indicating little variation in reporting. Percentages of respondents answering yes to the drinking and se;tual activity items show similar ranges of less than 7 points. Almost none of thase differences are statistically significant, and thsir shifting directionality suggests that none of them are prac- tically significant. This stability of reporting indicates that question length and wording fairdliarity do not influence vriiether respondents vdll report having perforated sensitive behaviors at least once. The desision to report any behavior always precedes the decision of how much behavior to repojrt, even if the yeo or no question is not asked explicitly. Apparently, respondents are not influenced by question length and by

wording familiarity in making that first decision.

Tables 2 and 3 indicate that the second decision, how much to report, is far more se sitive to question fori» manipulations. Long, open-ended questions were hypothesised as the best format for asking threatening items. These tables support that belief. Long, open-ended questions, with famil- iar wording where relevant, obtain the highest reporting in all of the Table 2 breakdowris. in addition, differences in reporting between these cells and the lowest cells are large. Long, open, familiar questions elicit more that double the reporting of short, closed, standard questions for all three drinking items. These consistent and sizable differences le;ive no doubt that open, long questions enhance reporting for threatening items cf a "how much" or "how many" nature. Even when the differences are tested on an item by item basis, as in Table 3, the differences between open forms and closed forms and between long forms and short forms are either significant or nearly significant. Familiar wordings are not sta- tistically superior to standard wordings, but the consistently good per- formance of the long, open and familiar format indicates that famili?": wordings shoui 1 be used unless inconv- lient.

Note that both Tablo 2 and Table 3 present results for the total popi'''.ation. This includes respondents who ware not asked these questions: for exaffiple, a respondent v/ho answered no to having drunk beer in the past year was not asked abcut the amount of beer consumed. Form effects for only those respondents who answered these questions are even stronger.

Interaction among question structure, question length, and question wording in their impact on response effects has been an unexplored topic. The desire to evaluate interaction effects motivated this study's factorial design. Table 3 suggests that these factors do not interact. Only one

-9-

first-order interaction is significant and no second-order interactions reach significance. The isolated closure by length interaction for the masturbation frequency item seems to possess little pragmatic value.

This study's results add to the understanding of response effects for threatening, questions, though the three hypotheses receive only mixed support. Threatening items which ask about having performed a behavior even once, and which roquire yes or no responses, are insensitive to question format manipulations . All three hypotheses must be rejected for these items. Threatening items which ask about frequency or intensity of behavior, and which require quantitative answers, are very sensitive to question structure tind question length. Hypothesis 1 and Hypothesis 2 are supported for these items. Hypothesis 3 must be rejected, though consist- ent results suggest the wisdom of using familiar wordings. Also, inter- action effects for these factors appear trivial.

Market researchers should find practical value in these findings. Product usage studies requesting data on sensitive articles such as many hygiene products probably encounter sizeable response effects because of the nature of the product. Any unwillingness to participate or to report product usage should remain insensitive to questionnaire manipulations. However, frequency and amount of usage should be more honestly reported in response to open-ended questions with innocuous padding. These product usage studies, as 'jell as many other marketing and social marketing stud- ies which deal with socially sensitive products or behaviors, should bene- fit from continued research and expanding knowledge about response effects.

Validation of Results While there are strong reasons to believe that increases in reporting are improvements in reporting for threatening behaviors, one would like

-10-

direct proof. This study could not get validating evidence on an individ- ual level. Hoivever, validating information is available on an aggregate level for the alcohol consumption items.

Since beer, vjine and liquor a^e taxed, precise aggregate figures on U.S. production are available. These figures provide estimates for annual consumption, though these estimates will not precisely match reported consumption. U.S. production figures do not consider exports and imports. This factor probably is not too impos'tant for beer and liquor, but domestic wine production figures should considerably understate wine consximption. Also, production figures include institutional consumption (military bases, dormitories) which is not covered in a sample of households. Third, much beer and liquor consumption occurs out of the home, where it is subject to greater underreporting, so that reported consumption should fall short of production figures.

Table 4 presents the comparison of production figures with reported consumption. The Brewers Almanac 197S (20) lists per capita production of 4,352 ounces of beer, 332 ounces of wine and 256 ounces of liquor. Re- ported consumT>tion is converted to th" same basa by assuming that a can of beer contains 12 ounces, a glass of wine 4 ounces and a drink of liquor 1 ounce. The highest reported beer and liquor consumption reach only 65 and 62 percent of the production figures. Reported wine consumption slightly exceeds wine production^ but not by the amount expected from foreign wine consumption. Some unden^eporting seems to remain even at the highest re- porting level, indicating that increased reporting is improved reporting, not positively biased reporting.

Table 5 offers further comparisons with this study's results. Louis Harris (6) did a 1974 study for the National Institute on Alcohol Abuse and Alcoholism which contained drinking questions almost identical to the drink-

-11-

ing questions in this study. Harris' reported results cannot be converted into a total consumption estimate, so Table 5 presents comparisons of the frequencies of consumption. For all three items, the Harris results are about equal to the lowest results of this study. This outcome verifies the contention'that short, closed questions with standard wording get reporting levels equal to those of other surveys, and that these reporting levels can be much improved by using long, opgn questions with familiar wording .

A final validation is provided by Locander's (11) earlier study. Locander did have validation information for questions about having been convicted of traffic violations and about having gone through bankruptcy. Reporting of these events ranged from 27 to 75 percent across groups, show- ing that negative response effects are persistent for threatening questions.

REFERENCES -1. ■ C.F. Cannell and F.J. Fowler, "A Comparison of a Self-Enumerative Procedure and a Personal Interview: A Validity Study," Public Opinion Quarterly, 27 (1963), 550-264 .

2. A.L. Clark and P. Wallin, "The Accuracy of Husbands' and Wives' Reports

of the Frequency of Marital Coitus," Population Studies, 18 (1964), 165-173.

3. J. P. Clark and L.L. Tifft, "Polygraph and Interview Validation of Self- Reported Deviant Behavior," American Sociological Review, 31 (1966), 516-523.

4. M. David, "The Validity of Income Reported by a Sample of Families Who Received Welfare Assistance during 1959," Journal of the American Sitatistical Association, 57 (1962), 680-685.

5. A. Ellis, "Questionnaire versus Interview Methods in the Study of Human Love Relationships," American Sociological Review, 12 (1947), 541-553.

6. L. Harris and Associates, Inc., Public Awareness of the xNIAAA Advertising Campaign and Public Attitudes Toward Drinking and Alcohol Abuse, prepared for The Nptional Institute on Alcohol Abuse and Alcoholism, February, 1974.

7. R.L. Kahn, A Comparison of Two Methods of Collecting Data for Social Research: The Fixed Alternative Questionnaire and the Open-ended Inter- yiew. Unpublished Ph.D. dissertation. University of Michigan, 1952,

8. A.C. Kinsey, W.B. Pomeroy and C.E. Martin, Sexual Behavior in the Human Male, Philadelphia, Samiders, 1948.

9. D.D. Knudsen, H. Pope and D.P. Irish, "Response Differences to Questions on Sexual Standards: An Interview-Questionnaire Comparison," Public Opinion Quarterly, 31 (1967), 290-297.

10. G. Levinger, "Systematic Distortion in Spouses' Reports of Prefered and Actual Sexual Behavior," Sociometry, 29 (1966), 291-299.

11. W. Locander, An Investigation of Interview Method, Threat, and Response

■ Distortion, Unpublished Ph.D. dissertation. University of Illinois, 1974.

12. W. Locander, S. Sudman and N. Bradbum, "An Investigation of Interview Method, Threat and Response Distortion," Journal of the American Statistical Association, 71 (June, 1976), in press.

13. E.H. Mudd, M. Stein and 11. E. Mitchell, "Paired Reports of Sexual Behavior of Husbands and Wives in Conflicted Marriages," Coinprehensiye Psychi atry , 2 (1961), 149-156.

14. S.J. Poti, C.R. Malaker and B. Chakraborti , "Reliability of Data Relating to Contraceptive Practices," in C.V, Kiser (ed.). Research in Fariuly Planning, Princeton, N.J., Princeton University Pre.ss, 1962, 51-65.

15. I.G. Sarason, "Effects of Anxiety, Motivational Instructions, and Failure on Serial Learning," Journal of Experimental Psychology, 51 (1956), 253-260.

16. I.G. Sarason, "Effects of Anxiety and Two Kinds of Motivating Instructions on Verbal Learning," Journal of Abnormal and Social Psychology, 54 (1957), 166-171.

17. I.G. SaraFon, "Relationships of Measures of Anxiety and Experimental Instructions to Word Association Test Performance, Journal of Abnormal and Social Psychology, 59 (1959), 37-42.

18. S. Sudman and N. Eradburn, Response Effects in Surveys, Chicago, Aldine, 1974.

19. R.L. Thomdike, E. Hagen and R.A. Kemper, "Normative Data Obtained in the House -to -House Administration of a Psychosomatic Inventory," Journal of Consulting Psychology, 16 (1952), 257-260.

20. United States Brewers Association, The Brewing Industry in the United States: Brewers' Almanac 1975.

21. U.S. National Center for Health Statistics, "Effect of Some Experimental Interviewing Techniques on Reporting in the Health Interview Survey," Vital and Healtn Statistics, Series 2, No. 41, Washington, D.C., Govern- ment Printing Office, 1971.

22. P. Wallin and A. Clark, "Cultural Norms and Husbands' and Wives' Reports of their Marital Partners Preferred Frequency of Coitus Relative to Their Own," SocioKetry, 21 (1958), 247-254.

23. D.W. Yaukey, B.J. Roberts and W. Griffiths, "Husbands' vs. Wives' Responses to a Fertility Survey," Population Studies, 19 (1965), 29-43.

24. B.A. Young, The Effects of Sex, Assigned Thei^apist or Peer Role, Topic Intimacy, and Expectations of Partner Compatibility on Dyadic Communi- cation Patterns, Unpublished Ph.D. dissertation. University of Southern California, 1269.

TABLE 1 PERCENTAGE OF RESPONDENTS ANSWERING YES TO ITEMS BY FORM

A. Gambling**

Have you played cards for money in the past )4ar?

Have you bet on sports?

Have you bet on elections?

Have you been in a betting pool?

Have you played dice games for money?

Have you bought a state lottery ticket?

Long

Short

29.2	30.3
19.8	17.4
6.9	12.4
17.2	16.2
6.2	7.8
25.2	23.9

B . Drinking

Have you ever drunk beer?

Have you drunk beer in the past year?

Have ycu ever drunk wine?

Have you drunk wino in the pasi year?

Have you ever dtxink hard liquor?

Have you drunk hard liquor in the past year?

Long- Long- Short- Short- familiar standard familiar standarc'

76.7

59.8 81.0

64.8

82.3

65.8 80.9

62. S

79.8

62.1

82.5

67.0

85.7

69.0

81.0

65.0

"Question structurs and wording familiarity were not manipulated for these items.

Question structure was not manipulated for these items. Wording familiarity was manipulated only for the hard liquor questions.

-2-

TABLE 1 CCont.)

C. Sexual Activity^

Have you enga^^ed in petting or kissing in past month?

Have you engaged in petting or kissing in past 24 hours?

Have you engaged in intercourse in past month?

Have you engaged in intercourse in past 24 hours?

Have you masturbated in

past inonth? 10.9 11.1 10.9 7.0

Hav3 you masturbated in

past 24 hours? 1.7 1.4 1.4 .3

Long- Lonr-	Short- Shori-
farniliar standard	f and liar standard
75. 6	76.2
41.8	35.1
67.9 66.1	69.4 69.2
18.1 16.4	16,9 15.8

Question structure vas not manipulated for these items. V'ording familiarity vas not manipulated for the petting and kissing items.

TABLE 2 MEANS FOR DRINKING AhlU SEXUAL ACTIVITY FREQUENCY ITEMS BY FORMS

Drinking Items

Total cans of beer drunk per person in past year

Loug__ Short Structure means

Total glasses o£ wine drunk per person in past year

Open	Closed	Length means
235	173	203
225	110	167
230	142

Long

Short

Structure means

Total drinks of hard liquor per person in past year

Open	Closed	Length means
91	57	74
63	36	49
77	47

	Open	Closed	Length means
Long Familiar Standard	158 lis	129 115	129
Short Familiar Standard	126 141	90 66	106
Structure means	135	100

wording familiarity was not manipulated for these items .

TABLE 2 CCont.)

SEXUAL ACTIVITY ITEMS

Total nusiber of times respondent engaged in petting or kissing in past year

Open Closed Length means 167 155 161 Short 149 139 144

Structure means 158 147

Long

Total number of times respondent engaged in intercourse in past year

Open Closed Length menas

76

Long Familiar	90	71
Standard	76	68
Short Famiiiar	69	75
Standard	71	63
Structure means	77	($9

69

Total number oi times respondent mastu., bated in past year

Long

Familiar

Standard

Short

Familiar

Standard

Structure means

Open	CI	osed	Length means
13.8		3.8	8.1
9.4		5.2
6.4		6.9	4.8
1.1		4.7
7.7		5.2

Wording familiarity was not manipulated for these items.

TABLE 3 ANALYSES OF VARIANCE FOR DRINKING AND SEXUAL ACTIVITY FREQUENCY ITEMS

U rink ins items

Total cans of beer dnank per person in past year

Source of Variation

Closure

Length

Closure-length

Residual

F	•^ d.f. 1 .	Significance
14.78	.001
2.52	1	.108
1.35	1 1149	.243

Total glasses of wine drunk per person in past year"

Source of Variation

Closure

Length

c.os'ire-length

Residual

F	cLf^	Significance
8.38	1	.004
5.57	1	.02
.08	1 1147	.99

Total drinks of hard liquor per person .n past year

Source of Variation

Closure

Length

Wording

Closure-Length

Closure-Wording

Length-Wording

Closure-Length-Wording

Residual

F	d.f.	sis3	[ificance
4.59			.03
1.99			.155
.96			.99
1.49			.221
,03			.99
.51			.99
1.00	1146		.32

TABLE 3 (Cont.)

SEXUAL ACTIVITY ITEMS Total number of times respondent engaged in petting or kissing in past year Source of Variation F d.f. Significance

Closure 1.72 1 .187

Length 3.73 1 .05

Closure-Length .015 1 ,99

Residual

1141

Total number of times respondent engaged in intercourse in past year

Source of Variation

Closure

Length

Wording

Closure -Length

Ciosure- Wording

Lengch-Wording

Closure- Length' !Vording

Residual 1124

F	d.f.	Significance
2.04		.U9
1.62		.20
1.47		.223
1.19		.275
.02		.99
.09		.99
1.36		.242

i t i i'

.r.U •'?

0^..

«;?<:.

.'	NO . :'
I	£<-..;
/	Vfc...
r	«l. 1
t	\^.
i	CO.
	rtc . J
,-.!!,

Total number of times respondent masturbated in past year

Source of Variation

Closure

Length

Wording

Closure-Length

Closure-Wording

Length-Wording

Closure-Length-Wording

Residual

F	d.£.	Significance
1.S8		.207
2,72		.095
1,83		.173
5.39		.02
1.31		.251
.34		.99
.12		.99

1140

a„

Wording familiarity was not manipulated for these items,

, t.'Js.„

LVX.

^^ ; j

;v.

ani.b'i

r o"^; ' J 'J •;• »•. i i.i ii- 1 .T RfK 10!' > ('4

TABLE 4 COMPARISON OF REPORTED DRINKING TO U.S. PRODUCTION

Oiances of reported, beverage U.S. taxed production consumption per person by of beverage in ounces form per person

A.	Beer		Open	Closed
		Long	2,820	2,076
		Short	2,700	1,320
B.	Wine		Open	Closed
		Long	364	228
		Short	252	144

4,352

332

C. Hard Liquor

Open

Closed

Long Familiar	158	129
Standard	115	115
Short Familiar	126	90
Standard	141	66

256

TABLE 5 COMPARISON OF REPORTED DRINKING FREQUENCY TO HARRIS DRINKING FREQUENCY DATA

Total number of times average respondent drank beverage in past year.

Estimates by form Harris estimate

A. Beer
		Open	Closed
	Long	74.22	52.41
	Short	69.84	38.25
B. Wine
		Open	Closed
	Long	37.14	20.71
	Short	28.33	17.31

50.49

16.32

C. Hard Liquor

Open

Closed

Long Familiar	47.76	40.31
Standard	35.99	41.86
Short Familiar	43.53	25.00
Standard	41.73	22.23

26.98

f\v