Sexual Risk Behaviors and Alcohol: Higher Base Rates Revealed using the Unmatched-Count Technique.

Alcohol consumption and sexual behavior often correlate. Researchers interested in HIV/AIDS and other STDs have tried to capitalize on this association in looking for factors that covary with behavior that puts people at risk for these negative outcomes. Many believe that alcohol consumption contributes to risky sexual behavior. Alcohol-related interventions might help increase safer sex behaviors such as condom use. However, determining the nature of the relationship between alcohol and risky sex has been difficult. Studies have yielded mixed results. Dingle and Oei (1997) reviewed 20 articles that dealt directly with the link between alcohol and risky sexual behavior. Seven studies (35%) supported and five studies (25%) partially supported the hypothesis that alcohol use may influence risky sexual behavior, while eight studies failed to support the hypothesis. Base rates for drinking combined with sexual risk taking are also difficult to determine. Problems in assessing sexual behavior may contribute to the inconsistency in findings on the relationship between risky sex and alcohol. Higher base rates than have previously been reported might mean that the association between drinking and risky sex is stronger and more consistent than research has determined. Accurate base rates for sexual risk behaviors after drinking might also help in developing targeted interventions among populations where the risk is greatest. This information could enhance the design of interventions aimed at increasing condom use.

Assessing Sexual Behavior

Most studies of sexual behaviors rely on self-reports. Researchers have questioned the validity of these self-reports ever since Kinsey first embarked on his sexual surveys. Sexual behavior is often highly private. Risky sexual behavior is often laden with negative evaluation. Thus, people might give socially acceptable responses when asked to reveal what they do, think, and feel when engaging in sexual activity (Catania, Gibson, Chitwood, & Coates, 1990). Brody (1995) claimed that the use of self-report data for sexual behaviors that confer risk for HIV is suspect, suggesting that participants in behavioral research are prone to intentional misrepresentations. Despite these concerns, self-reports remain the most prevalent method for assessing the topography of sexual behavior because of the ethical and practical problems involved in the use of more direct methods (Weinhardt, Forsyth, Carey, Jaworski, & Durant, 1998).

Two reviews of methodological issues in risky sexual behavior research have appeared in the past few years (Catania et. al., 1990; Weinhardt et al., 1998). Catania et al. urged researchers of HIV and other STDs to be rigorous in trying to find accurate assessment instruments. They conclude that current estimates of high-risk sexual behavior are underestimates. Weinhardt et al. reviewed the empirical literature since Catania et al., concluding that self-reports of sexual behavior still remain problematic. The authors made a number of recommendations to improve the validity of self-reported sexual behavior, including using appropriate measures for behaviors of interest, using easily understood language, using techniques that improve recall of behavior, and asking questions in a direct fashion. Nevertheless, generating accurate base rate estimates for risky sexual behavior and understanding the relationship between risky sexual behavior and drinking remains difficult. The sensitive nature of assessing for risky sexual behavior is the likely cause for these difficulties.

A Sensitive Behavior--Employee Theft

Wimbush and Dalton (1997) studied base rates for employee theft. Previous studies had found widely varying rates of employee theft, determining that it was difficult to assess (Camara & Schneider, 1994; Dalton & Daily, 1994; Murphy, 1993). Camara and Schneider noted base rate estimates of between 3% and 62%. Since employee theft is so negatively evaluated, people likely provided untrue and evasive answers (Chaudhuri & Mukerjee, 1988). Even guarantees of anonymity might not be enough to encourage truthful answering to produce accurate base rates. Thus, Wimbush and Dalton used the unmatched-count technique (UCT), providing participants a chance to answer sensitive items without ever having to directly admit to a given behavior.

In the UCT method, participants receive a series or set of five or six statements and respond by indicating the number of statements that are true for them. One of the statements is the item of interest. Half the sample receives the item of interest in their set of six statements, while the other half receives only the remaining five items. The base rate estimate for the item of interest is determined through random assignment of participants and comparisons between the two halves of the sample. This protocol is described more thoroughly in the Methods section of this paper. The key element of the UCT is that participants do not respond directly to the sensitive items. Instead, they report the number of tree items in a set that may include the sensitive item of interest.

Wimbush and Dalton (1997) found significantly higher percentages of participants endorsed engaging in employee theft on the UCT protocol than on a conventional anonymous self-report survey. The greater endorsement appeared both for admitting any theft and also for theft at different monetary values. The UCT protocol and the self-report surveys, however, showed no difference in percentages of endorsement for a control item that was not considered to be a sensitive behavior. The authors concluded that the higher percentages found using the UCT technique were more accurate and represented better estimates of the base rates of employee theft.

The Present Study

We extended the use of the UCT protocol to the sensitive areas of sexual behavior, particularly risky sexual behavior. We hypothesized that the UCT technique would yield significantly higher percentages of persons who endorse sensitive items like having sex without a condom and having sex without a condom after drinking over an anonymous self-report questionnaire. An alcohol item, drinking until intoxication, served as the control item. For the majority of college students, drinking is not considered a sensitive activity that carries with it a negative evaluation (Crundall, 1995; Senchak, Leonard, & Greene, 1998). Therefore, we expected no difference in base rate estimates between the two methods for this alcohol item.

If, as has been posited by Catania et al. (1990), current base rate estimates of high risk sexual behavior are underestimates, then the increased percentages of endorsements in the data from the UCT protocol will provide a more accurate representation of risky sexual behavior. These new and higher base rate estimates could help explain the mixed findings from studies looking at the relationship of alcohol to risky sex. They might also help public health officials assess risk within a given population so they can target interventions to those most in need.

METHOD

Participants

Three hundred forty-six college students participated in this study for extra credit in their undergraduate psychology classes. One hundred forty-five (41.9%) were male and 201 (58.1%) were female. The mean age for the participants was 21.93 years (SD = 6.48). The participants came from diverse ethnic backgrounds (52.9% White, 8.4% Black, 13.6% Hispanic, 13.9% Asian, and 11.2% Other or Mixed). The subjects were randomly divided into three groups. There were no apparent group differences in age, education, ethnicity, or gender (p [is greater than] .10).

Procedure

After being randomly divided, Group 1 (n = 102) received a true/false conventional self-report survey. Groups 2 (n = 122) and 3 (n = 122) were UCT protocol groups, with Group 2 receiving Form A and Group 3 receiving Form B.

Conventional self-report survey protocol. Group 1 received a series of behavioral statements. They were informed that the surveys were designed to encourage honest responding and that their responses were completely anonymous. There was no identifying number or code attached to this form. They responded true or false to each item. Several of the items assessed alcohol use, sexual behavior, and alcohol and sexual behavior paired together. These items appear in Appendix A.

UCT protocol. Base rate estimates by UCT require two randomly assigned groups (Dalton, Daily, & Wimbush, 1997). One of the groups received a set of five nonsensitive items (Dalton & Wimbush, 1994). The participants were told that this questionnaire was designed to encourage honest responding, and were asked not to respond directly to whether any particular item is true. Rather, they were asked to indicate how many of the five items were true for them. Thus, while indicating how many items were true, they never directly endorse any particular item. Someone who responded three, for instance, was indicating that three of the five items were true for him or her.

Another group received the same series of nonsensitive items. However, an extra item was added to this series, so that the series contained six items. This extra item was the sensitive item of interest. Participants in this group were given the same instructions as the first group. They were asked to indicate how many of these six items were true for them. As is the case with the first group, a response of four indicates that four of the six items were true for that participant. Again, there is no way to determine exactly which four of the six were true. Since the groups were randomly assigned, the difference in the mean responses of these two groups must be a function of some persons in the second group endorsing the sensitive item. The base rate estimate for the sensitive behavior is determined from this difference. Randomization and adequate sample size minimized the chance that confounds (i.e., group differences on demographic characteristics, etc.) rather than the proportion of persons in the second group endorsing the sensitive item account for the mean differences. Larger samples enhance estimate stability and accuracy. Dalton and Wimbush (1994) suggest that accuracy and stability are compromised if the groups do not contain at least 40 to 50 subjects respectively.

In this study, the two UCT groups contained 122 subjects each. Group 2 received Form A, while Group 3 received Form B (see Appendix B). Each Form contained five sets of items. The sensitive items and the control item were split between the two groups to further ensure that any hidden group differences were not accounting for differences in mean scores on any particular set.

Example: Sex without a condom

Form A, Set 1 contained five nonsensitive statements:

1. I have been to Spain.

2. I would consider myself a sports fan.

3. I have a brother.

4. I have more than one sister.

5. I read the book "The Pelican Brief".

If a participant gives a response of two to this set, then two of the five items are true for this person. However, they do not circle or check the two true items, but simply put a two next to the set. Even if the participant's names or defining demographic characteristics were on the measure, it would be impossible to ascertain exactly which two of the items were true for this particular individual. Set 1 on Form B contains the exact same five nonsensitive statements as Set 1 on Form A (these nonsensitive items are derived from the original Dalton and Wimbush, 1994, study). But Set 1 on Form B includes an additional item: I've had sex without a condom. The addition of this sensitive statement is the key to the UCT technique. The 122 participants who received Form A had a mean of 1.65 true items for the five items in Set 1, while the 122 different participants who received Form B averaged 2.35 true items out of the six items that made up Set 1 on Form B. The mean of the Form B group is expected to be larger because of the additional item (sex without a condom). As stated earlier, given the random assignment to forms, the difference in the average responses to Set 1 between the two groups is a function of those respondents who have endorsed the additional item, sex without a condom (Wimbush & Dalton, 1997).

The base rate estimate for the behavior of interest is found by subtracting the two means: p = mean (Form B, Set 1) - mean (Form A, Set 1), where p is the proportion of participants endorsing the sensitive behavior. In this example, p = 2.35 - 1.65 = 0.70. Thus, the base rate estimate in this population for having sex without a condom is 70%.

RESULTS

A criterion for the effectiveness of UCT is whether it results in higher base rates on sensitive behaviors than more conventional self-report surveys (Dalton & Wimbush, 1994, Wimbush & Dalton, 1997). This will be particularly true for the assessment of risky sexual behavior. The findings from the conventional survey and UCT protocols on the sex and/or alcohol behaviors are contained in Table 1.

Table 1. Comparison of Admissions Using Conventional Survey and Unmatched Count Protocols

                                           Binomial
                 Conventional              test
     Item         survey (%)    UCT (%)    (modern        Factor
   (N = 346)      (N = 102)    (N = 244)   method)

Sex without
  a condom            59           70      p=.048(*)       1.19
Drank until
  intoxication        77           70      p=.186 NS       .91
Sex + alcohol         48           49      p=.872 NS       1.03
Sex without
  a condom
  after
  drinking            36           65      p < .001(***)   1.78
Had sex               74           84      p = .03(*)      1.14

Note. A similar pattern of significant results were obtained using the more conventional Fisher's test for group comparison. The factor score is obtained by dividing the UCT proportion by the conventional survey proportion. The 1.78 factor score for sex without a condom while drinking indicates that UCT participants are 1.78 times more likely to admit to having sex without a condom while drinking than conventional survey participants.

(*) = p < .05.

(**) = p < .01

(***) = p < .001. NS = not significant at the 05 level.

Statistics

Because participants either engaged in the behavior of interest or did not, these data are well suited for analysis comparing two independent binomials. This analysis tests the hypothesis [p.sub.1] = [p.sub.2], where [p.sub.1] is the proportion of endorsements in the conventional survey and [p.sub.2] is the proportion of endorsements in the UCT. The most popular procedure for testing this hypothesis is Fisher's exact test. However, Fisher's test has less power compared to other procedures (Wilcox, 1996). Storer and Kim (1990) compared several methods that have better power properties than Fisher's. They recommend a modem, robust binomial test, which Wilcox (1997) named Twobinom. This procedure has better statistical properties than Fisher's, including both better power and better control over Type 1 error. Analysis using this modern binomial test returns a significance level for the hypothesis that the proportions are equal.

Findings

Significant differences were found between the two groups on the measures had sex (p [is less than] .05), had sex without a condom (p [is less than] .05), and had sex without a condom after drinking (p [is less than] .001). No differences were found on the items drink to intoxication (p = .872, NS) and had sex after drinking (p = .186, NS). For example, 36% of the participants receiving the conventional survey endorsed having sex without a condom after consuming alcohol, while the UCT protocol revealed a base rate estimate of 65% for this behavior. Thus, the UCT protocol revealed almost twice the percentage of persons engaging in risky sexual behavior after drinking as did the conventional self-report survey.

DISCUSSION

The UCT protocol revealed significantly higher base rates than a conventional self-report survey for having had sex, having sex without a condom, and having sex without a condom after drinking. The two techniques did not reveal a difference in base rates among participants for drinking to intoxication and having sex after drinking. Thus, the UCT protocol revealed higher estimates of risky sexual behaviors.

The higher base rates revealed using the UCT are consistent with the hypotheses for sensitive behaviors. There was no difference between methods for the base rate estimate of our control item, drinking until intoxication. In college populations where binge drinking remains somewhat normative, drinking until intoxication does not carry a strong negative evaluation. Thus, college students do not intentionally underreport this behavior to look better. A negative social evaluation connected to sexual activity, however, still seems to exist among some college students--more UCT participants endorsed having sex than survey participants. The lack of difference on the sex after drinking item suggests that sex serves as a threshold of sorts; those admitting to sexual activity will also admit to sex after drinking.

Safer sex or condom use, however, appears to be the area of highest sensitivity. The UCT protocol revealed significantly higher base rate estimates on both items related to condom use. The format of the UCT, not directly asking participants to endorse sensitive behaviors, makes it more likely for persons to admit to not using a condom and not using a condom when drinking. In the critical area of the association of risky sex and drinking in the same event, almost twice as many participants endorsed this behavior under the UCT protocol. This result suggests that college students feel pressure not to admit having risky sex after drinking. The idea of going out, getting loosened up with alcohol, and then having sex without a condom appears to be particularly laden with socially undesirable characteristics. Perhaps this is the result of safer-sex education coupled with people's sense that they are acting differently than they would act if they had not been drinking. It seems that people know that they should use a condom and so find it difficult to acknowledge not using one, even if granted anonymity on a self-report survey.

A base rate estimate for sex without a condom after drinking of 65% via the UCT protocol appears noteworthy. It almost doubles the base rate found by the conventional survey. This large discrepancy suggests that the UCT protocol may provide an opportunity for altering estimates of previous base rates for alcohol and risky sex occurring together in the same event. Previous research has suggested that alcohol use and sex without a condom do not necessarily occur together at the same time (Halpem-Flesher, Millstein, & Ellen, 1996). Rather, people who drink more tend to have more risky sex (Temple & Leigh, 1992). It has been suggested that both alcohol use and risky sexual behaviors are part of some underlying construct such as sensation seeking. However, the data derived from the UCT protocol could help better explain the relationship between alcohol and risky sex. If, as this study suggests, conventional assessment of risky sex after drinking only captures about half the actual behavior, previous studies have a powerful bias toward not finding effects. Thus, the fact that some studies have detected an association is powerful evidence for the positive co-occurrence of drinking and having sex without a condom in the same event.

Drinking, therefore, might be an important mechanism involved in risky sex. Attitudes toward condoms significantly decline when people are drinking (Gordon, Carey, & Carey, 1997). Alcohol expectancies for condom use have been found to mediate the relationship between alcohol and risky sex (LaBrie & Earleywine, 2000). Perhaps persons drinking alcohol are disinhibited from doing what they might intend to do--use a condom during sex. Or, persons might drink so that they can avoid responsibility for negotiating condom use, providing a condom, and actually using a condom correctly. Thus, alcohol-related interventions might increase safer sex. These interventions could include multiple steps: creating positive attitudes and expectancies toward condom use both generally and when drinking, building up efficacy for condom negotiation and use when drinking, educating persons that condom use is possible when drinking, and teaching that drinking does not lower one's responsibility for protecting themselves and others from HIV and other STDs.

The present study is limited in its relatively small sample size and college student population. While the sample size meets the criteria established by Dalton and Wimbush (1994), a larger sample would allow for more generalizability of results. The college student population does not necessarily generalize to the population at large. However, college students themselves continue to be at risk for HIV and other STDs. Also, in this study the only characteristics that were looked at for group equivalence were race, gender, age, and education. Other factors, such as religiosity, could impact one's attitudes and behaviors regarding risky sex and drinking. Future studies employing the UCT in assessing sexual behavior might seek to include factors such as religiosity, values, and sensation seeking when checking for group equivalence.

Finally, although the UCT is able to reveal accurate base rates, the nature of the method does not allow researchers to link other data to individuals who do or do not engage in the risk behaviors. Since it is not known exactly who endorsed the sensitive items it is impossible to figure out how much other factors like sensation seeking contribute to an individual's risky behavior. However, UCT can assess two things occurring together if asked in a single item, such as drinking and risky sex. A researcher could easily put a personality factor (such as a sensation-seeking item) and a behavior in the same sensitive item. Also, the higher base rates revealed by UCT suggest that it is a useful technique that could be employed in future sex research. The more accurate and substantially higher base rate estimates for risk behaviors revealed by UCT may be a help in adjusting other findings which employ more conventional methodologies.

The UCT protocol reveals higher estimates of base rates for not using a condom and not using a condom while drinking than conventional assessment tools. This result is similar to the results found in using UCT to assess for base rates of employee theft. It appears that the UCT, therefore, is a better assessment tool for determining base rates of behaviors that are sensitive, carry negative evaluation, and, thus, are underreported in survey data. This protocol is a potentially helpful tool for sex researchers. In HIV- and STD-related research, there may be a significant underreporting of risky behavior that masks real associations between risky sex and behaviors such as drinking. Accurate base rate estimates are essential to understanding if these associations exist, and for deciding how best to allocate resources and target interventions toward populations and places where risk is greatest. The UCT method reveals that risky sex after drinking is one such place. Thus, interventions aimed at drinking may be important for slowing the rate of HIV and other STD infections, saving persons from great suffering and society numerous resources.

REFERENCES

Brody, S. (1995). Patients misrepresenting their risk factors for AIDS. International Journal of STD and AIDS, 6, 392-398.

Camara, W. J., & Schneider, D. L. (1994). Integrity tests: Facts and unresolved issues. American Psychologist, 49, 112-119.

Catania, J. A., Gibson, D. R., Chitwood, D. D., & Coates, T. J. (1990). Methodological problems in AIDS behavioral research: Influences on measurement error and participation bias in studies of sexual behavior. Psychological Bulletin, 108, 339-362.

Chaudhuri, A., & Mukerjee, R. (1988). Randomized responses: Theory. and techniques. New York: Marcel Dekker.

Crundall, 1. A. (1995). Perceptions of alcohol by student drinkers at university. Drug and Alcohol Review, 14, 363-368.

Dalton, D. R., Daily, C. M., 8,: Wimbush, J. C. (1997). Business ethics research with full disclosure and no deception: The unmatched block technique. Journal of Business Ethics, 16, 1049-1057.

Dalton, D. R., 8,: Wimbush, J. C. (1994). Using the unmatched count technique (UCT) to estimate base-rates for sensitive behavior. Personnel Psychology, 47, 817-828.

Dingle, G. A., & Oei, T. P. S. (1997). Is alcohol a cofactor of HIV and AIDS? Evidence from immunological and behavioral studies. Psychological Bulletin, 122, 56-71.

Gordon, C. M., Carey, M. P., 8,: Carey, K. B. (1997). Effects of a drinking event on behavioral skills and condom attitudes in men: Implications for HIV risk from a controlled experiment. Health Psychology, 16, 490-495.

Halpern-Flesher, B. L., Millstein, S. G., & Ellen, J. M. (1996). Relationship of alcohol use and risky sexual behavior: A review and analysis of findings. Journal of Adolescent Health, 19, 331-336.

LaBrie, J. W., & Earleywine, M. E. (2000). Alcohol expectancies for condom use mediate the relationship between alcohol and safe-sex. Manuscript submitted for publication.

Murphy, K. R. (1993). Honesty in the workplace. Pacific Grove, CA: Brooks/Cole.

Sencahk, M, Leonard, K. E., & Greene, B. W. (1998). Alcohol use among college students as a function of their typical social drinking context. Psychology of Addictive Behaviors, 12, 62-70.

Storer, B. E., & Kim, C. (1990). Exact properties of some exact test statistics for comparing binomial proportions. Journal of the American Statistical Association, 85, 146-155.

Temple, M. T., & Leigh, B. C. (1992). Alcohol consumption and unsafe sexual behavior in discrete events. The Journal of Sex Research, 29, 207-219.

Weinhardt, L. S., Forsyth, A. D., Carey, M. P., Jaworski, B. A., & Durant, L. E. (1998). Reliability and validity of self-report measures for HIV-related sexual behavior: Progress since 1990 and recommendations for research and practice. Archive of Sexual Behavior, 27, 155-180.

Wilcox, R. R. (1996). Statistics for the social sciences. San Diego, CA: Academic Press.

Wilcox, R. R. (1997). Introduction to robust estimation and hypothesis testing. San Diego, CA: Academic Press.

Wimbush, J. C., & Dalton, D. R. (1997). Base rate for employee theft: Convergence of multiple methods. Journal of Applied Psychology, 82, 756-763.

Manuscript accepted March 21, 2000

APPENDIX A

Conventional Self-Report Survey Items

1. I have been to Ireland.

2. I think drinking alcohol decreases condom use.

3. My shoe size is over 7.

4. I've had sex after drinking.

5. I went to a private high school.

6. I've had sex without a condom.

7. I often watch television late at night.

8. I've had sex without a condom after consuming alcohol.

9. I like the editorial section of the newspaper.

10. I've had sexual intercourse.

11. I use the internet almost every week.

12. I've consumed alcohol until I was intoxicated.

APPENDIX B

Unmatched Count Technique Behavior Sets

Instructions: Next to each set below place the number that matches the number of items in the set that are true for you.

FORM A

   Set 1
   I have been to Spain.
   I would consider myself a sports fan.
   I have a brother.
   I have more than one sister.
   I read the book "The Pelican Brief."

   Set 2
   I type reasonably well.
   I own at least one cordless phone.
   I own one or more handguns.
   I have one or more cats.
   I've consumed alcohol until I was intoxicated.
   I know what a "balk" is in baseball.

   Set 3
   I read the local paper almost daily.
   I read USA Today several times a week.
   I usually choose sugar-free soft drinks.
   I don't smoke cigarettes.
   I have more than two children.

   Set 4
   I take vitamins almost every day.
   I read the book "The Prince."
   There's a shotgun in my house.
   I've had sexual intercourse.
   I've lived outside the U.S.
   I've lived in at least 3 different states.

   Set 5
   I have been to Ireland.
   My shoe size is over 7.
   I went to a private high school.
   I often watch television late at night.
   I've had sex without a condom after consuming alcohol.
   I like the editorial section of the newspaper.

FORM B

   I have been to Spain.
   I would consider myself a sports fan.
   I have a brother.
   I have more than one sister.
   I've had sex without a condom.
   I read the book "The Pelican Brief."

   I type reasonably well.
   I own at least one cordless phone.
   I own one or more handguns.
   I have one or more cats.
   I know what a "balk" is in baseball.

   I read the local paper almost daily.
   I read USA Today several times a week.
   I usually choose sugar-free soft drinks.
   I don't smoke cigarettes.
   I've had sex after drinking.
   I have more than two children.

   I take vitamins almost every day.
   I read the book "The Prince."
   There's a shotgun in my house.
   I've lived in at least 3 different states.
   I've lived outside the U.S.

   I have been to Ireland.
   My shoe size is over 7.
   I went to a private high school.
   I often watch television late at night.
   I like the editorial section of the newspaper.

Address correspondence to Joseph LaBrie, Department of Psychology, PO Box 45041, Los Angeles, CA 90045; e-mail: [e-mail].