Trying to practice safer sex: development of the sexual risks scale.

In recent years, many researchers have examined psychosocial models of health behavior as these relate to HIV risks and safer sex (Fishbein & Middlestadt, 1989; Fisher, 1984; Jemmott & Jemmott, 1991; Kirscht & Joseph, 1989). Researchers typically have drawn upon common models, such as the Theory of Reasoned Action (Fishbein & Ajzen, 1975) and the Health Belief Model (Becker, 1974), and used common constructs, such as perceived norms, efficacy beliefs, and attitudes about safer sex. Yet few researchers have used established psychometric instruments to assess necessary constructs or rigorous scientific techniques for item-pool derivation. Although practical limitations of time and sample size undoubtedly influenced the latter decision, the former likely resulted from a shortage in the availability of such measures. The current study was intended to address this deficiency by providing a theoretically-based measure of constructs related to safer sexual activity. Drawing from the literature on health behavior, including models such as the Theory of Reasoned Action, the Theory of Trying (Bagozzi & Warshaw, 1990), the Information-Motivation-Behavioral Skills Model (Fisher, Fisher, William , & Malloy, 1994), and the Health Belief Model, we derived items for a scale to assess constructs related to safer sexual behavior. The final scale contains six subscales, each of which may be used individually or in conjunction with one another, depending upon the needs of particular researchers.

Theory of Reasoned Action

The Theory of Reasoned Action (TRA) was originally developed by Fishbein and Ajzen (1975) as a general model of behavioral prediction. Subsequently, this model has been adopted by researchers to explore determinants of health behavior (Fisher, 1984; Fishbein & Middlestadt, 1989). According to the model, behavior is a function of intention to behave in a particular manner. Intention, in turn, is a function of an individual's attitude about the act and the perceived norm regarding the behavior.

The attitudinal component of TRA consists of the individuals perception about consequences of the act, as well as the evaluation of those consequences. For instance, if an individual believes that unprotected sex may result in contracting HIV (perceived consequence of unprotected sex) and that contracting HIV is a dreadful possibility (evaluation of the consequence), then that individual is likely to have an aversive attitude toward improtected sex.

The normative component of TRA consists of the perceived behavioral norm and motivation to comply with the norm. For example, if the individual believes that only promiscuous people use condoms (perceived norm) and wishes to abstain from condom use to avoid seeming promiscuous (motivation to comply), he or she will likely have a positive normative conception of unprotected sex. Theorists have suggested that because intention is composed of both personal (attitudinal) and other (normative) elements, it is the best predictor of the couple's decision to practice safer sex (Davidson & Morrison, 1983).

Theory of Trying

The Theory of Trying (TT;, Bagozzi & Warshaw, 1990) is a modification of TRA. Bagozzi and Warshaw stated that behavior is not always preceded by careful deliberation or conscious decision; some behaviors, for instance, derive from habit. This may be especially true when alcohol or other awareness-impairing drugs interfere with reasoning (Kelly, St. Lawrence, & Brasfield, 1991; Leigh & Stall, 1993). Other behaviors may be influenced by unforeseen impediments to the behavioral goal (e.g., the individual may reach into her purse to retrieve a condom only to find that it has been ruined by hairpins). Thus, Bagozzi and Warshaw (1990) suggested that effort or trying to achieve the goal should replace behavior as the criterion measure in models of health behavior. Intention to try replaces TRA behavioral intention. Researchers have also noted that a person's past experience with trying may influence his or her recognition of behavioral barriers and thereby influence his or her ideas about the effectiveness or feasibility of trying to perform the behavior (Bagozzi, Yi, & Baumgartner, 1990; Bentler & Speckart, 1979; Warshaw & Davis, 1985). For example, a person may prefer to practice safer sex but may recognize that a stubborn partner will insist upon unprotected sex. Thus, this model incorporates the idea that behaviors such as condom use may not always fall neatly under the umbrella of personal choice and impediments may prevent one's intentions from being realized. Bagozzi and Warshaw suggested that expectations about success and failure interact with attitudes about the act. The modified model thereby includes expectations about whether a behavior will occur as a predictor of intention to try.

Information-Motivation-Behavioral Skills Model

The Information-Motivation-Behavioral Skills model of AIDS-preventive behavior (IMB; Fisher et al., 1994) is based upon the idea that information and motivation combine to influence behavioral skills. All three components are said to contribute to AIDS-preventive behavior. The information component consists of knowledge about HIV transmission and the effects of HIV on the body. This component is quite similar to TRA's perceived consequences of the act. The motivation component of IMB includes attitudes toward AIDS-preventive acts, subjective norms regarding AIDS-preventive acts, and behavioral intentions for AIDS-preventive acts. All of which are similar to TRA constructs (attitudes, norms, and intentions). The behavioral skills component includes efficacy beliefs and negotiation skills, both of which might be subsumed under TT's expectations (e.g., "It would be difficult for me to discuss condom use with a partner").

Health Belief Model

The Health Belief Model (HBM; Becker, 1974) is based upon the premise that perceived severity of consequences, perceived susceptibility to consequences, and perceived benefits of preventive behavior all contribute to health behavior. Severity of consequences and benefits of preventive behavior might be subsumed by TRAs perceived consequences and evaluation of consequences. For instance, if an individual believes that AIDS is a dreadful disease (HBM: perceived severity; TRA: evaluation of consequence of unprotected sex) and that using a condom may prevent trans, mission of HIV (HBM: perceived benefits; TRA: perceived consequences of using condom), then that individual may be more likely to use a condom than might someone with dissimilar beliefs. In terms of safer sexual behavior, HBM includes perceived susceptibility to HIV infection as an additional factor in risk aversion. To illustrate perceived susceptibility, let us consider beliefs that predominated the early stages of the United States AIDS crisis. AIDS was first portrayed as a disease that struck gay men, Haitians, and hemophiliacs (Shilts, 1987). Some members of the general public believed (and still believe) that HIV is a "gay" virus. These persons might feel that heterosexual sex does not pose a risk for HIV and may thereby eschew condom use as an unnecessary precaution.

Substance Use

Several researchers have indicated that substance use may influence adherence to safer sexual behavior Kelly et al., 1991. Leigh & Stall, 1993). Alcohol and drugs may impair awareness and lower inhibitions. Under such circumstances. individuals may forget or dismiss the need to practice safer sex or may have diminished ability to counter pressures to engage in risky sex.

The Current Research

The current study was intended to encompass aspects of each of the aforementioned behavioral theories. We used constructs from each theory for generation and classification of scale items to enhance the generalizability of final measures for future research purposes. We conducted a pilot study and two further studies directed toward these goals.

Pilot Study: Item-Pool Derivation

We conducted a preliminary study to derive items for an initial questionnaire. In this study, open-ended questions were used to elicit student beliefs and opinions about safer sexual activity

Method

Recruitment of Participants

For this and each study described in the current article, students volunteered to participate in research pertaining to attitudes, beliefs, and behaviors concerning sexuality and dating. Because sexuality was mentioned in our recruitment protocol, we were concerned that sexually liberal persons might be overrepresented in the sample. To explore this possibility, we included the Sexual Opinions Survey (Byrne & Fisher, 1983) within one of the studies (Study 1, N = 296). We found participant responses along this measure of erotophobia/erotophilia to be normally distributed, representing a wide range of sexual beliefs and values. Thus, we have reason to believe that the sample was fairly unbiased in terms of sexual openness.

As a precautionary measure to maintain participant anonymity within each study, students were permitted to take home surveys, which were then returned in sealed envelopes and dropped into a sealed repository. Students were informed that they would not be penalized should they choose not to complete or return the survey. For our initial pilot study, participants were 28 undergraduate students who completed the study as part of a research requirement for an introductory psychology course.

Materials and Procedure

Students were first given an excerpt from an educational HIV pamphlet that listed safer, moderate risk, and high-risk sexual behaviors. The students were invited to write paragraphs expressing their feelings about why they did (or would like to) practice safer sex and why they did not (or would not like to) practice safer sex. Students were also asked to write about any personal changes toward safer sexual behavior and reasons for these changes. Each student was asked to imagine that his or her sexual partner did not like safer sex. Students then wrote paragraphs explaining why they would or would not "give in" to the partner and practice unsafe behavior. Each student was then asked to imagine that his or her friends would appreciate the student's practicing safer sex. Each student wrote a paragraph about why he or she would or wouldn't comply with friends' wishes.

Results and Discussion

Fifteen participants were male, 12 were female, and 1 participant did not specify gender. Ages in this sample ranged from 18 to 40, with a median age of 20. Two participants identified as homosexual, 24 as heterosexual, and 1 as bisexual, and 1 participant did not specify sexual orientation. Most participants (68%) were White, 25% were Black, 1 participant was Hispanic, and 1 did not specify ethinicity. Most participants (68%) had a yearly personal income of less than $10.000.

Item Derivation and Refinement

Student paragraphs were reduced to individual statements pertaining to safer sexual activity. Because the current study was intended as a pilot study for the purposes of establishing a basic pool of items to be explored in subsequent analyses, classification of items was performed for limited descriptive purposes. The first author performed many classifications without consultation of a second rater, so interrater rehabilities are not available for these introductory classifications. The 184 statements were classified into five major categories:

(1) Attitudinal statements (45%c7c of all statements)

(2) Normative statements (12%)

(3) Intention to try to practice safer sex (18%)

(4) Expectation to practice safer sex (18%)

(5) Perceived susceptibility statements (7%)

Approximately 55% of all statements were supportive of safer sexual practices (e.g., "I try to wear a condom when I am having sex"), and 45% were nonsupportive (e.g., "In the heat of the moment, it is easy to ignore safer sex"). The bulk of student statements regarding safer sex could be characterized as attitudinal. These items included sentiments such as "I think `safer' sex would get boring fast." Normative statements in the item pool included ideas such as "My friends talk a lot about safer sex." Although few student statements suggested strong normative networks, the majority of normative items generated by students were supportive of safer sex. That is, when students believed that friends encouraged safer sex, students were motivated to comply Fewer students expressed indifference to normative influences (eg., "It's none of my friends' business whether or not I practice safer sex").

Intention to try to practice safer sex represented a substantial percentage of student statements. These included statements such as "Abstinence (not having sex is a way to reduce HIV/ AIDS risk that I just wouldn't try and "I intend to follow safer sex guidelines within the next year." Expectation items included efficacy beliefs, such as "If a sexual partner didn't want to use condoms, we would have sex without using condoms" as well as statements indicative of sexual communication/negotiation skills, such as "If my partner wanted me to participate in `risky' sex and I suggested a lower-risk alternative, we would have the `safer' sex instead."

Most student statements concerning perceived susceptibility to HIV/AIDS indicated a sense of invulnerability; that is, few students believed themselves to be at risk for HIV. Items included statements of perceived personal risk, such as "My sexual experiences do not put me at risk for HIV/AIDS."

To represent more thoroughly relevant theoretical constructs, we generated a number of additional statements and adapted statements from the attitude-behavior literature (Bienvenu, 1980; Brown, 1984; Hall, 1992; Hebert, Bernard, De Man, & Farrar 1989; Herzberger, Chan, & Katz, 1984; Sheeran, Abraham, Abrams, Spears, & Marks, 1990). Recent research has indicated that attitudes toward condom use fall along a number of dimensions, including the perceived reliability/effectiveness of condoms, embarrassment about condom purchase and use, stigma related to condom use, the expense/availability of condoms, offensiveness of condoms, convenience of condoms, and the effects of condoms on sexual pleasure and romance (Brown, 1984; Hebert et al., 1989; Helweg-Larsen & Collins, 1994; Sheeran et al., 1990). We ensured that all these dimensions were represented in our item pool. Although much attitudinal research on condom use is focused on condoms as a method of birth control, our items were focused on more general., attitudes toward condoms and attitudes toward condoms as a method of HIV risk reduction. We also included attitudinal items to address alternate forms of safer sex (e.g., "petting") and safer sex in general. For greater coverage of perceived susceptibility, we included comparative risk items, such as "I am less at risk for HIV/AIDS infection than most of my friends," and items concerning trading off/discounting risks, such as "I would rather enjoy unprotected sex than worry about getting AIDS." Because no students generated statements concerning substance use, we added items about alcohol and drug use. Past research has indicated that substance use may be an important variable to measure, yet our guiding theoretical models (e.g., Theory of Trying) do not explicitly include substance use; thus, only a few items were used to represent this supplementary construct. Substance use items were characterized by statements such as "When I socialize, I usually drink alcohol or use drugs" and "I might not practice safer sex if I'd been drinking, smoking pot, or using drugs."

Using author-generated statements, statements from the literature, and the aforementioned student-generated statements, a maximal number of unique items were derived. We did not include complex items that might warrant agreement/disagreement to more than one idea [e.g., "I would only participate in unprotected intercourse with a spouse (first idea), whom I would give myself over to (second idea)"]. All items in the pool were refined to create clarity and uniformity of presentation. Items were written in a manner that dearly stated any specific behaviors or normative referents to which the item applied. Items were then modified in accord with criteria utilized by Brown (1984), Edwards (1957), and Wang (1932): (a) half the items were worded positively, and half were worded negatively, (b) items were brief and technical jargon was avoided; (c) attitude items were affectively oriented and debatable; (d) universals (e.g., all, never) and double negatives were avoided, (e) items that were likely to be endorsed by almost everyone or almost no one were not used. We also tried to assure that items did not contain phrases that indicated participants were/are sexually active (e.g., "When I have sex, I take precautions to reduce my risk of HIV/AIDS" would be changed to "If I were going to have sex, I would take precautions to reduce my risk of HIV/AIDS".

Following item refinement, the item pool consisted of 108 items. These included attitudinal items (31%) normative items (19%), intention to try items expectation items (20%), susceptibility items (10%), and substance use items (4%). The proposed response format for all items consisted of five-point scales. To evaluate the practical and psychometric utility of items generated in our pilot study, another study was conducted.

Study 1: Criterion Testing and Item Elimination

We administered items from the pilot study, as well as a number of criterion measures, to a second sample of students to ascertain reliability and validity information on pilot study items/subscales. Analyses provided information about the most useful items for a final scale.

Method

Participants. Participants were 296 undergraduate students who responded to the measure to complete a research requirement for an introductory psychology course.

Materials and procedure. Items derived from our pilot study, along with social desirability items, were randomly ordered on a preliminary questionnaire. Social desirability was assessed using the Marlowe-Crowne Social Desirability scale (MCSD; Crowne & Marlowe, 1964), and scores on this measure were included for consideration in psychometric analyses. Specifically, we wished to eliminate items for which participant responses were primarily a function of social desirability response bias.

We also generated a behavioral risk survey for use as a criterion measure in Study 1. For precision of measurement, the behavioral risk survey was quite explicit Included were items assessing relationship status, relationship length, whether the individual and her or his partner had ever been tested for HIV, number of sexual partners, and frequency nature of sexual behavior e.g., anal, oral, vaginal, first-receptive versus "insertive" used barriers and spermicides: steady partner versus casual partner). Participants would, for example, first indicate how many times within the past six months they had engaged in anal sex with a steady partner. Each participant would then indicate on a five-point scale how often he or she had been "the inserter. Almost never/Almost always) and how often a condom was used (Almost never/Almost always) when he was the inserter and when he or she was not the inserter. The risk survey contained approximately 90 items, with these items being organized with respect to type of sexual activity and casual versus steady partners. Administration of the risk survey was counterbalanced, with some participants completing the behavioral risk survey before completing attitude/belief items, and some participants completing the risk survey after other items were completed.

Results and Discussion

Participants were 92 men, 202 women, and 2 persons who did not specify gender. Participant ages ranged from 18 to 45, with a median age of 19. Most (82%) of the students were White, 14% were Black, 2% Asian, 1% Hispanic, and 1% unspecified. The majority of students (82%) earned less than $10,000 per year. Four participants identified as homosexual, 281 as heterosexual, 8 as bisexual, and 1 as asexual, and 2 participants did not specify sexual orientation. Most participants (80%) had engaged in anal, oral, or vaginal sex within the past six months, with the median number of sexual encounters being 25. Only 7% of the sample identified as never having had sex, and the median number of sexual partners (in lifetime) was 3. Men's and women's self-reported sexual behavior was comparable along these dimensions.

Computation of behavioral risk scores. To compute behavioral risk scores, individual sexual behaviors were first ranked according to approximate risk value for HIV transmission. Rank ordering was based upon empirical data from the most recent research that could be found to detail rankings of the diverse range of behaviors explored within our study (Shernoff, 1988). For instance, anal sex was ranked as having greater risk value than vaginal sex, which in turn was ranked as having greater risk value than vaginal sex with a condom. Vaginal sex with a condom was ranked as having greater risk than oral-penile sex without a condom, and so on. Twenty-one sexual behaviors were ranked with consideration of type of sexual contact (e.g., anal, vaginal, oral-anal, oral-vaginal), receptive versus insertive participation, and use of latex/spermicidal barriers. Ordinal data were assigned abstract numerical risk indices based upon "anchoring" of data points as a function of relative HIV-transmission data from previous research. For instance, research indicates that anal sex is 2 to 3 times more risky than vaginal sex, and vaginal sex with a condom is said to be 10 times less risky than vaginal sex without a condom (Hearst & Hulley, 1988; Padian et al., 1987). Numerical values corresponding to sexual behaviors reflected such empirical findings. Initial rank ordering and assignment of numerical risk values were made independent of behavioral context (e.g., whether the partner was "a steady partner, was a casual partner, had been tested for HIV. Because some data on the behavioral risk survey were collected using five-point scales (e.g., "When my partner and I had anal sex in the past six months, I was the inserter Almost never/Almost always"), algebraic formulae were used to convert participant responses into approximate risk scores per act, which corresponded with anchored numerical risk values.

Using information concerning the context of the behavior (e.g., monogamous versus casual partner, number of sexual partners, whether partner had been tested for HIV), we expanded risk formulae differentially to weigh contextual factors that might influence HIV risk. For instance, if a person is in a steady relationship with a steady partner who has tested negative for HIV, the risk is minimal. If the person is engaging in sex with many partners, the chances of encountering a seropositive person or an efficient transmitter of FIN are increased (Hearst & Hulley, 1988; May. Anderson & Blower, 1988; Reiss & Leik, 1989).

For each participant, an overall risk score was computed from summary data of all acts, for which each act's risk per act score was multiplied by the frequency of act (e.g., "My partner and I had anal sex about (frequency) times in the past six months"]. Thus, persons who engaged in riskier sexual practices (given context, type of act, and frequency) had higher risk scores than persons who engaged in less risky practices. Risk scores, therefore, acted to approximate a criterion measure addressing the HIV risk of various behaviors. (Behavioral risk measures and detailed computational procedures are available from the authors.)

Item elimination. A number of statistical analyses were performed to assess the reliability and validity of individual items. Items that did not possess adequate psychometric qualities were eliminated from the item pool.

Items with restricted response ranges (e.g., no participant endorsed a value of five on a five-point scale) may result in ceiling and floor effects, thereby attenuating relationships among variables. Fourteen items (six attitudinal, three intention to try, four expectation, one susceptibility) were omitted from the item pool to minimize the potential for such range effects.

To minimize the potential for social-desirability response bias in future data sets, items with significant correlations (p [is less than] .05) to the MCSD were eliminated from the item pool. Eight items (three attitudinal, one normative, two intention to try, one expectation, one susceptibility) were eliminated for this reason.

Items that correlate with a scale's total score are, presumably, the best indicators of the construct that the Scale is intended to measure. Cronbach's alpha reliabilities were calculated for each of the major categories used in our pilot study. Those items that reduced the internal consistency of proposed subscales were candidates for omission from the item pool. Twenty-Seven items 112 attitudinal, 6 normative. 2 intention to try, 4 expectation, 1 susceptibility. 2 substance use) were eliminated because of low item-scale correlations (that is, inclusion of these items reduced the overall reliability of the subscale). These deletions were made with consideration of exploratory factor analyses, in that items that did not correlate with the proposed subscale, but that clustered together and possessed substantial face validity and predictive validity were considered for retention as separate/new subscales in the item pool. Multidimensional representation of attitudes toward condoms (as suggested by Helweg-Larsen & Collins, 1994) offered no psychometric advantage over the current unidimensional, attitudinal measure. Items representing issues such as reliability, embarrassment, and identity stigma associated with condom use failed to predict (correlate at p [is less than] .05 with) past risk behavior or intention to try to practice safer sex. The current attitudinal subscale is composed primarily of items concerning the influence of condoms (and safer sex) upon sexual pleasure, with a few items representing more general favorable/unfavorable attitudes toward safer sexual activity (e.g., "Generally, I am in favor of using a condom").

Principle components (PC) factor analyses revealed several clusters of items that were examined for usefulness as "new" subscales. These included items representing constructs such as social network involvement, perceived behavioral control, verbal communication skills, and discounting/trading off risks. Because none of the new subscales provided substantial contributions to the prediction of critical constructs (new subscales did not correlate with intention or risk behavior at p [is less than] .05), these items were ultimately eliminated from the pool. Nineteen items (5 normative, 8 expectation, 2 intention to try, 4 susceptibility) were eliminated for this reason.

Although detailed procedures were used for computation of behavioral risk scores for criterion analyses, subsequent analyses indicated that this level of detail provided no significant increment over simpler measures (e.g., frequency of condom use) in prediction of theoretical constructs such as intention to try and expectation. It is possible that our complicated formulae did not add accuracy to actual risk prediction. Another likely possibility is that students do not engage in such detailed deliberation prior to participating in risky sexual behavior. Casual sex, spermicide use, and insertive versus receptive intercourse may not be factors salient in the safer sexual practices of some persons. For these individuals, decisions to practice safer sex may be influenced only by dichotomies of condom use versus unprotected sex. Despite the fact that our behavioral risk measure was an unvalidated criterion measure, it was significantly (p [is less than].01) associated with simpler single item measures, such as frequency of condom use with steady/nonsteady partners (range Almost never/Almost always) and number of sexual partners. These single item measures are analogous to those commonly used in research on safer sexual practices and other health behaviors (Bagozzi & Warshaw, 1990; Fisher et al., 1979). Thus, the original behavioral risk measure, although more detailed than was necessary, nevertheless seems to be a reasonable criterion for the aforementioned analyses. For brevity of measures and generalizability to other studies, simpler measures were used in subsequent analyses.

Reliabilities and factor structure of resulting subscales. After item elimination, the item pool consisted of 13 attitudinal items (a = .90), 7 normative items ([Alpha] = .84), 7 intention to try items ([Alpha] = .76), 7 expectation items ([Alpha] = .82), 4 susceptibility items ([Alpha] = .80), and 2 substance use items ([Alpha] =.78.

Tabachnick and Fidell 1989, suggested that Principle Components, (PC) analyses using Varimax , for orthogonal dimensions) rotation be used for initial exploration of factor structure. We found results to be most interpretable when two separate analyses were performed. That is. when such a large number of items were entered into a single analysis, the overlap between correlated constructs (e.g., attitudes and norms are said to predict intention) resulted in less differentiation between subscales. The first analysis included items concerning attitudes, norms, susceptibility, and substance use. When entered into a PC analysis with Varimax rotation, extraction produced four components. All subscale items had the highest factor loadings on the appropriate components: attitudes (Factor 1), norms (Factor 2), susceptibility (Factor 3), and substance use (Factor 4). None of the 26 items had loadings of over .35 on more than a single subscale. Factor loadings and variance estimates are presented in Table 1.

[TABULAR DATA 1 NOT REPRODUCIBLE IN ASCII]

The second analysis included the conceptually similar expectation and intention to try items. As can be seen in Table 1, PC extraction using Varimax rotation yielded two components. Factor 1 was comprised of intention to try items, and Factor 2 was comprised of expectation items. Two expectation items had very high loadings on both components, but none of the 12 remaining items had loadings of over .35 on more than a single component. Because these analyses were exploratory our factor solutions seemed acceptable, and these 40 items were retained for the final questionnaire.

Study 2: Psychometric Properties of the Final Survey

This study was intended to verify the appropriateness of items retained in Study 1 for a final scale measuring theoretical constructs related to safer sexual behavior. Factor analyses were used to explore the underlying dimensions of subscales and internal reliability and predictive validity information for subscales and the total scale were provided. Using structural equation modeling, we examined the construct validity of subscales, and complex relationships between individual subscales were addressed in greater detail.

Method

Participants. Participants were 200 (61 men, 139 women) undergraduate students who completed the survey to fulfill a research requirement for an introductory psychology course or who completed the survey for course credit in upper level psychology courses.

Measures and procedures. Students completed a demographic questionnaire, as well as a survey containing items retained from Study 1. Instead of using items from the previous behavioral risk survey, we administered several simpler behavioral risk items. Two items assessed condom use with steady partners in the past year and with nonsteady partners in the past year. (a) "When I had sex (vaginal or anal intercourse) with a steady partner in the past year, we used a condom --" and (b) "When I had sex in the past year with someone who was not a steady partner, we used a condom --." A third item assessed condom use in the past two weeks: "When I had sex in the past two weeks, we used a condom --." These items had a five-point response format (never/always; also, no steady partner, no nonsteady partners, and have not had sex options were available for the three respective items). A final behavioral risk item was used to assess the number of sexual partners within the past year (response options: 0, 1, 2-4, 5-10, over 10).

Results and Discussion

The majority of participants (86%) were between 18 and 25 years of age, and most (72%) had a yearly personal income of less than $10,000. Most participants (76%) were White, 20% were Black, 2% were Hispanic, 2% were Asian, and 2% unspecified. Although many participants (56% of men and 36% of women) had not had sex within the past two weeks, most of the sample (82% of men and 81% of women) had engaged in sexual activity within the past year.

Factor analyses. Principle components analyses were conducted to examine whether the factor structure found in Study 1 was maintained across samples. Tabachnick and Fidell (1989) suggested that PC analyses with Varimax rotation be used for exploratory analyses, followed by examination by numerous methods (e.g., Principle Aids, Maximum Likelihood, Unweighted Least Squares) and rotations (e.g., Obhmin, Equamax, Image) for confirmatory analyses, until the most interpretable results are achieved. In our analyses, alternative methods to PC produced very similar results, with all but one or two items loading on the same components. Because PC analyses produced the most meaningful factor structure, we retained this method for the current analyses. PC analyses sometimes capitalize upon unique and error variance, and thus other methods of factor analyses that are used to analyze shared variance might result in slightly lower proportions of explained variance within the item pool.

Items concerning attitudes, norms, susceptibility, and substance use were entered into a single analysis with Varimax rotation. Extraction produced four distinct components, with all subscale items having the highest loadings on the appropriate components: attitudes (Factor 1), norms (Factor 2), susceptibility (Factor 3), and substance use (Factor 4). Factor loadings for these items are presented in Table 2.

Table 2

Scale Items and Item Loading

                                      Analysis  1

                          Factor   Factor    Factor    Factor

Attitude 11                .80       .08      .03      -.07
Attitude 1                 .72       .05      .10      -.20
Attitude 3                 .71       .14     -.03      -.06
Attitude 5                 .68      -.05     -.01      -.16
Attitude 2                 .67      -.01     -.09       .00
Attitude 13                .66       .10      .02       .14
Attitude 6                 .65       .12     -.05       .06
Attitude 8                 .65       .12     -.05      -.14
Attitude 12                .64      -.11      .07      -.23
Attitude 9                 .63       .11     -.11       .15
Attitude 7                 .59       .13     -.05       .05
Attitude 4                 .49       .19     -.08      -.29
Attitude 10                .35       .33      .00       .09
Norm 7                     .09       .80      .18      -.13
Norm 3                     .14       .79      .01      -.08
Norm 5                     .11       .78      .08      -.11
Norm 6                     .05       .73      .07      -.02
Norm 1                     .14       .59     -.14       .00
Norm 2                    -.03       .58     -.04       .08
Norm 4                     .06       .58     -.04       .07
Suscept 3                  .04       .01      .85       .08
Suscept 4                 -.11       .07      .85      -.02
Suscept 2                 -.10       .01      .79      -.01
Suscept 1                  .01      -.04      .77       .17
Substance 2               -.09      -.02      .05       .86
Substance 1               -.09      -.02      .15       .81

% Variance                24%       13%      10%        6%

                       Analysis 2

               Factor    Factor    Factor
                 1          2         3

Expect 4        .82      -.04       -.04
Expect 3        .81       .01        .08
Expect 5        .74       .26        .19
Expect 1        .69       .08       -.02
Expect 2        .63      -.19        .23
Intent 5        .08       .85        .10
Intent 4       -.09       .79       -.19
Intent 7        .09       .63       -.22
Intent 6       -.12       .09       -.84
Intent 3        .02       .13       -.81
Intent 1        .36      -.02       -.47
Intent 2        .34       .15       -.44

% Variance     38%       14%        9%

None of the 26 items had loadings of over .35 on more than a single subscale. The attitudinal subscale accounted for approximately 24% of total item variance, normative items accounted for approximately 13% of the variance, susceptibility accounted for approximately 10%, and substance use accounted for approximately 6% of the total variance.

When expectation and intention to try items were analyzed using PC extraction and Varimax rotation, four components were extracted. We found that oblique rotation (for correlated dimensions) enhanced interpretability by allowing correlations among the four components. When expectation items and intention to try items were analyzed using oblique Direct Oblimin rotation, extraction indicated that two items on the expectation subscale represented a separate abstinence construct. This abstinence dimension seemed largely unrelated to other dimensions represented in our analyses e.g., component correlations with the abstinence component were .07, .13, and .06. Subsequent regression analyses indicated that these items did not provide a significant increment over other expectation and intention items in prediction of condom use with steady partners. F change (14, 181) = 1.74. p = .18, ns. or with nonsteady partners, F change (14, 181)=2.23, p = .11. ns. accounting for less than 2% of the variance in condom use with either steady or nonsteady partners. Total abstinence was a consideration in the predicted condom use measures, despite the labels, used to refer to these measures. That is, item responses ranged from I (almost never used condoms to 4 (almost always used condoms), and abstinence (not having had sex) was assigned a value of 5. indicating that sexual behavior tended toward the "safer" direction. Thus, abstinence items did not add to prediction of our criterion measures, and inclusion of the two items reduced the internal reliability of the expectation subscale (from .80 to .73). The two items were therefore eliminated from the final scale.

The remaining expectation and intention to try items were subjected to another PC extraction with oblique Direct Oblimin rotation. This analysis revealed three correlated components. Factor loadings are presented in Table 2. Factor 1, which accounted for approximately 38% of the total item variance, was represented by expectation items. Factors 2 and 3 represented intention to try items. Intention to try items fell along two dimensions; one dimension represented items that explicitly mentioned "condom use," whereas the other dimension represented items concerning "safer sex." These items accounted for approximately 14% and 9% of the total item variance. Because the two groups of intention to try items were correlated (r = .52, p [is less than] .01), and the internal consistency of the total subscale was quite high (see Internal Reliability of Subscales), we decided to retain these dimensions within a single subscale. To address possible bias because of the overrepresentation of women in our sample, we conducted additional factor analyses by gender. PC analyses (Varimax rotation) of items concerning attitudes, norms, susceptibility, and substance use were performed separately for men and women. These analyses indicate that the aforementioned factor structure is basically maintained for women's responses (see Table 3). For women, a single attitude item loaded onto the norm component along with all norm items. Men's responses were essentially similar with regard to the first three components (see Table 4) but differed with regard to the substance use component in contrast to our original analyses, substance use items loaded upon the susceptibility Component for men, and the fourth component was comprised of two norm items. A single attitude item ("The proper use of a condom could enhance sexual pleasure") loaded onto the norm component.

Factor 3

Factor Loadings for Women

                                    Analysis 1

               Factor 1   Factor 2   Factor 3   Factor 4   Item Mean

Attitude 11       .76        .14        .02       -.08       3.21
Attitude 3        .75        .15       -.04       -.07       3.11
Attitude 2        .69       -.03       -.07       -.02       3.64
Attitude 8        .69        .11       -.08       -.11       3.64
Attitude 1        .68        .08        .08       -.30       3.21
Attitude 5        .68       -.04       -.02       -.11       3.04
Attitude 12       .66       -.14       -.03       -.21       2.95
Attitude 6        .63        .17        .01        .10       4.04
Attitude 13       .61        .10        .07        .20       3.86
Attitude 4        .61        .17       -.05       -.09       2.83
Attitude 9        .59        .19       -.09        .04       3.35
Attitude 7        .49        .15       -.02        .05       3.78
Norm 7            .11        .80        .21        .01       3.44
Norm 3            .11        .80        .02       -.10       3.11
Norm 5            .17        .77        .09        .04       3.35
Norm 6            .15        .72        .00        .13       2.96
Norm 4            .06        .63       -.01       -.12       3.49
Norm 1            .13        .62       -.11       -.15       3.42
Norm 2           -.02        .57       -.12        .07       3.15
Attitude 10       .30        .41        .11        .26       4.29
Suscept 4        -.10        .07        .98       -.09       1.83
Suscept 3         .05       -.01        .87        .07       2.19
Suscept 2        -.13        .03        .85       -.04       1.62
Suscept 1         .01       -.07        .76        .15       2.40
Substance 1      -.10       -.03        .06        .83       2.47
Substance 2      -.17       -.02        .02        .82       2.58

% Variance      24%         13%        11%        6%

                                    Analysis 2

                    Factor 1    Factor 2    Factor 3    Item Mean

Expect 3              .79        -.02         .11         4.26
Expect 4              .76         .02        -.03         4.30
Expect 2              .74        -.10        -.15         4.42
Expect 1              .67         .06        -.07         3.76
Expect 5              .61         .36         .20         3.91
Intent 1              .49        -.14        -.41         4.65
Intent 2              .40         .15        -.39         3.94
Intent 5              .04         .86         .07         4.07
Intent 4             -.20         .79        -.30         4.12
Intent 7              .28         .60        -.10         4.29
Intent 3             -.07         .07        -.88         4.22
Intent 6              .13         .17        -.77         4.29

% Variance           39%         14%        10%
Table 4

Factor Loadings for Men

                                       Analysis 1

               Factor 1   Factor 2   Factor 3   Factor 4   Item Mean

Attitude 11       .87       -.08        .00        .03       3.23
Attitude 1        .81        .05        .11       -.04       3.07
Attitude 7        .74        .01       -.07        .18       3.44
Attitude 5        .73        .06       -.05       -.26       3.10
Attitude 13       .71        .00       -.04        .20       3.59
Attitude 6        .68       -.15       -.17        .20       3.90
Attitude 12       .68        .15        .11       -.07       3.21
Attitude 9        .62       -.12       -.05       -.26       3.00
Attitude 2        .61        .09       -.05       -.17       2.98
Attitude 3        .61        .00       -.07        .22       2.92
Attitude 8        .58        .30       -.10       -.25       3.56
Attitude 10       .41        .10       -.25        .05       4.13
Norm 6           -.09        .85        .14       -.05       3.23
Norm 5            .03        .84       -.11        .06       3.39
Norm 7            .06        .79       -.12        .07       3.36
Norm 3            .19        .79        .02        .14       2.93
Norm 2           -.04        .58        .16        .18       3.00
Attitude 4        .28        .38       -.35       -.34       2.59
Suscept 3        -.05        .17        .78       -.02       1.98
Suscept 1        -.01        .03        .75        .11       2.36
Suscept 4        -.12        .20        .73       -.11       2.03
Substance 1      -.07       -.19        .65        .01       2.54
Suscept 2         .00        .06        .64        .03       1.87
Substance 2       .01        .23        .47        .11       2.59
Norm 4           -.04        .14       -.01        .86       3.02
Norm 1           .11         .25       -.09        .78       3.20

% Variance      23%         14%        11%         7%

                                   Analysis 2

                   Factor 1        Factor 2        Item Mean

Intent 7            .87             .16             4.54
Intent 6            .78             .13             4.10
Intent 3            .72             .08             4.18
Intent 4            .63            -.28             4.36
Intent 1            .60            -.23             4.66
Intent 5            .60            -.08             4.07
Intent 2            .52            -.32             3.90
Expect 5            .00            -.89             3.77
Expect 4            .01            -.87             3.77
Expect 3            .07            -.82             4.02
Expect 1.           .08            -.70             3.59
Expect 2           -.07            -.51             4.15

% Variance          40%            16%

PC analyses of intention to try and expectation items were again conducted using oblique Direct Oblimin rotation for correlated components. For women, two intention items loaded onto the expectation component, but loadings for these items were also high on the appropriate intention component. Loadings for other items were essentially analogous to our original analyses. Men did not seem to discriminate between intention to try to use condoms and intention to try to practice safer sex, in that all of these intention items loaded upon a single component. This finding offers some support for our decision to aggregate the two components (condom use versus safer sex of intention to try into a single subscale.

Internal reliability of subscales and of the final scale. Alpha reliabilities were performed on each of the six subscales. Reliabilities for subscales were as follows: attitudinal [Alpha] = .88, normative [Alpha] = .83. expectation [Alpha] = .80, intention to try [Alpha] = .80. susceptibility [Alpha] = .84, and substance use [Alpha] = .76. The final 38-item scale had an [Alpha] =.86.

Correlations among intended constructs. Table 5 presents Pearson r correlation coefficients between three measures of behavioral risk (condom use with steady partners, condom use with nonsteady partners, and number of sexual partners in past year) and each of the six proposed subscales, as well as with the final scale score. Al though most of these correlations were significant at the p [is less than] .05 level, only about half reached significance when Bonferroni's adjustment for tests Of multiple hypotheses (Dunn, 1961) was used. Based on this criterion, condom use with steady and nonsteady partners remained significantly associated with the final scale score. Also, perceived susceptibility was inversely associated with condom use and positively associated with number of sexual partners. This is discordant with predictions of HBM that imply high perceived susceptibility would enhance safer sexual behavior. Although no definitive causal statements can be made, logical inference might indicate that our findings support the idea that perceived susceptibility for our sample followed from reflections upon past risk behavior.

Table 5

Correlations Between Behavioral Risk and Subscales

                                      Condom Use

                              W/Steady        W/Nonsteady

to Try                         0.33(***)       0.06
   Men                         0.18            0.22
   Women                       0.39(***)      -0.01
Expectations                   0.17(*)         0.27(***)
   Men                         0.22            0.29(*)
   Women                       0.20(**)        0.24(**)
Attitudes                      0.23(***)       0.09
   Men                         0.16            0.30(*)
   Women                       0.31(***)      -0.02
Norms                          0.11            0.03
   Men                         0.05            0.14
   Women                       0.14           -0.02
Perceived Susceptibility      -0.27(***)      -0.41(***)
   Men                        -0.16           -0.48(***)
   Women                      -0.33(***)      -0.38(***)
Substance Use                 -0.10           -0.26(***)
   Men                        -0.04           -0.48(***)
   Women                      -0.13           -0.16
Total Scale                    0.33(***)       0.25(***)
   Men                         0.24            0.41(***)
   Women                       0.40(***)       0.17(*)

                           Number of
                            Partners

to Try                        -0.07
   Men                        -0.14
   Women                      -0.15
Expectations                  -0.16(*)
   Men                        -0.06
   Women                      -0.20(**)
Attitudes                     -0.08
   Men                        -0.13
   Women                      -0.03
Norms                         -0.01
   Men                        -0.18
   Women                      -0.07
Perceived Susceptibility       0.49(***)
   Men                        -0.38(**)
   Women                       0.55(***)
Substance Use                  0.18(*)
   Men                        -0.34(**)
   Women                       0.11
Total Scale                   -0.21(**)
   Men                        -0.22
   Women                       0.19(*)

(*) p <.05 (**) p < .01 (***) p < .002 (requisite for Bonferroni's adjustment for tests of multiple correlations)

When these correlational analyses were performed separately by gender, several marked differences were observed in construct relationships for men versus women. For condom use with steady partners, correlations were maintained for women, but no subscale significantly predicted condom use with steady partners for men. The scale's ability to predict the sexual behavior of men seemed to be concentrated along dimensions of perceived susceptibility and substance use, with these two dimensions predicting men's number of sexual partners and condom use with nonsteady partners

Table 6 presents correlations between critical theoretical constructs (e.g., correlation between attitudes and intention to try, between norms and intention to try). Again, most of these correlations were significant at the p [is less than] .05 level. Correlations that reached significance on the basis of Bonferroni's adjust meat include correlations of expectations, attitudes, and norms with the higher order construct of intention to try, as well as intercorrelations between attitudes and expectations, norms and expectations, perceived susceptibility and expectations, and attitudes and norms. These correlations may seem high given the distinctness of principle components in previous analyses. The separability of subscales in PC analyses was in part maintained by performing two separate analyses: one for orthogonal attitude, norm, susceptibility and substance use components and one including oblique intention and expectation components.

Table 6
Correlations Between Subscales

                    Intention       Expectation   Attitudes

Expect.              0.49(***)
Attitudes            0.51(***)      0.37(***)
Norms                0.32(***)      0.20(***)       0.25(***)
Suscept.            -0.16(*)       -0.26(***)      -0.08
Substance           -0.09          -0.19(**)       -0.18(*)

                     Norms   Susceptibility

Expect.
Attitudes
Norms
Suscept.             0.03
Substance           -0.06       0.16(*)

(*) p < .05

(**) p < .01

(***) p < .003 (requisite for Bonferroni's adjustment for tests of multiple correlations

Regression analyses. Factor scores from the preceding factor analyses were entered into several regression equations to assess the predictive validity of subscales with concern to HIV risk behavior. When the behavioral risk item concerning condom use with steady partners was regressed upon factor scores for attitudes, norms, susceptibility, substance use, expectation, intention--condoms, and intention--safer sex, these scores accounted for approximately 23% of the variance (using adjusted or shrunken [R.sup.2] values) in condom use with steady partners, F(7, 188) = 9.40, p [is less than] .001. A refined analysis (using only statistically significant predictors from the first analysis) was conducted. The best predictors of condom use with steady partners were intention--condoms ([Beta] = .42) and susceptibility ([Beta] = -.27), ps [is less than] .001. These two factor scores accounted for approximately 23% of the variance in condom use with steady partners, F(2, 193) = 30.83, p [is less than] .001.

When condom use with nonsteady partners was regressed upon factor scores, scores accounted for 21% of the total variance in nonsteady Condom use, F(7, 188) = 8.62, p [is less than] .001. Refined analyses included expectations ([Beta] = .19), susceptibility ([Beta] = -.35), and substance use ([Beta] = -.18), accounting for about 22% of the variance in condom use with nonsteady partners, F93, 192) 19.39, ps [is less than] .01.

Analyses using subscale totals, rather than factor scores, yielded similar results in terms of particular dimensions that best predicted condom use. Subscales accounted for approximately 14%, of the variance in condom use with steady partners, F(6, 189) = 6.45, and 21% of the variance in condom use with nonsteady partners, F(6, 189) = 9.85, ps [is less than] .001.

The scale total was entered into separate regression equations for prediction of condom use with steady and nonsteady partners. The scale total accounted for less variance than had factor scores or subscale totals for both condom use with steady (adjusted [R.sup.2] = .10) and nonsteady partners (adjusted [R.sup.2] =m), but prediction continued to be significant [steady: F(1, 194) = 23.74; nonsteady: F(1, 194) = 12.97, ps [is less than] .001].

In another analysis, subscale totals for theoretical predictors (attitudes, norms, expectations, and perceived susceptibility) were regressed upon the subscale total for the construct intention to try. Attitudes, norms, expectations, and susceptibility accounted for 39% of the total variance in intention to try, F(4, 195) = 32.19, p [is less than] .001.

Because correlational analyses indicated that perceived susceptibility to HIV may follow from reflections upon past risk behavior, perceived susceptibility was regressed upon behavioral risk variables. Taken together, condom use with steady partners ([Beta] = -.17), condom use with nonsteady partners ([Beta] = -.24), and number of sexual partners [Beta] = .31, accounted for 29% of the variance in perceived susceptibility, F(3, 189 = 26.99. ps [is less than] .01.

Structural equation modeling. LISREL-7 path analyses were performed upon the data to assess the relationships among various theoretical constructs. Our first model was used to address the proposed relationship between constructs as delineated by the Theory of Trying. According to this model, expectations, norms, and attitudes should all contribute to intention to try, which should in turn contribute to actual trying (conceptualized in this model as condom use with steady partners). The Chi-square statistic, [chi square] (3) = 1.37, p = .71, ns. and the Goodness of Fit Index, GFI = .997, indicated an excellent fit between the proposed model and the data. Note that in structural equation modeling, the researcher attempts to establish that there is no significant difference between the proposed model and the observed data. Thus, a nonsignificant Chi-square value, indicating no difference is indicative of a good fit between proposed and observed models. The structural equation model is presented in Figure 1. Path coefficients depicted in the model are analogous to standardized regression coefficients.

[Figure 1 ILLUSTRATION OMITTED]

Regression analyses were indicative of associations between condom use with nonsteady partners and expectations, susceptibility, and substance use. Our second model was used to address the relationships among these constructs. We also included number of sexual partners, which had been correlated with perceived susceptibility, in the model. For this analysis, we included only those cases in which the participant indicated having had nonsteady partners in the past year. This produced 68 cases for analysis. The best fitting model demonstrated that substance use may influence condom use with nonsteady partners, which may in turn influence perceptions of personal risk and expectations about the feasibility of safer sexual behavior. The number of past sexual partners also contributed to perceptions of personal risk. This model, presented in Figure 2, also yielded an excellent fit with our data, [chi square](6) = 1.98, p = .92, ns, GFI = .989.

General Discussion

Our final scale possessed internal reliability and both predictive and construct validity. The scale is unique in its focus upon safer sex for HIV prevention, rather than for contraception. The 38-item scale is comprised of subscales measuring attitudes about safer sex, normative beliefs, intention to try to practice safer sex, expectations about the feasibility of safer sexual activity, perceived susceptibility to HIV/AIDS, and substance use. The subscales are sufficiently concise and reliable for use independently or in conjunction with one another, an advantage that may enhance the utility of the scale for particular needs of HIV researchers. In fact, the reliability of subscales surpassed the reliability of scales commonly generated for purposes of evaluating such constructs Fisher et al., 1994; Helweg-Larsen & Collins. 1994). Previous measures of constructs related to condom use have typically been focused upon a single construct (e.g., attitudes) or have included items that presume sexual or condom use experience (Grimley, Riley, Bellis, & Prochaska, 1993; Sacco, Rickman, Thompson, Levine, & Reed, 1991, 1993; Wulfert & Wan, 1995). The wording used within the current scale is broadly applicable with regard to gender, sexual orientation, and sexual experience; thus, the scale may be used with diverse populations and in primary prevention assessments for persons whose sexual behavior patterns have not yet been established.

The subscales and the final scale generally correlated well with criterion measures of behavioral risk, and scale dimensions accounted for about one fifth of the variance in condom use with steady and nonsteady partners. Lower order constructs, including attitudes about safer sex, perceived norms, expectations concerning safer sexual efforts, and perceived personal susceptibility to HIV infection, account for about 40% of the variance in intention to try to practice safer sex.

Although intention to try was a significant predictor of condom use with steady partners, expectations were a better predictor of condom use with nonsteady partners. This is consistent with Bagozzi and Warshaw's (1990) suggestion that expectations come into play when barriers to successful action are involved. In sex with steady partners, relationship dynamics of intimacy, trust, and cooperation may allow individuals to implement intentions successfully. With nonsteady partners, however, skills of negotiation and communication may be more important Fear of rejection may prevent some persons from even suggesting condom use. Other individuals involved in casual sexual encounters may feel less comfortable implementing safer sexual behavior or countering partner resistance to safer sex. Encounters with nonsteady partners may also be unforeseen, and condoms may thereby be unavailable. Also, as indicated by our correlational analyses, alcohol and other awareness-impairing drugs may impede intentions to implement safer sexual activity with casual partners.

We attempted to explore further some relationships found in our initial analyses via structural equation modeling. Our first model supported the construct validity of the final scale. In accord with relationships set forth by the Theory of Trying (Bagozzi & Warshaw, 1990), attitudes about safer sex, normative beliefs, and expectations about the feasibility of safer sexual activity all contributed to the prediction of intention to try to practice safer sex. Intention to try, in turn, was used to predict risk behavior Thus, our results meshed nicely with the Theory of Trying. To the extent that TT concepts encompass the Information-Motivation Behavioral Skills Model, the scale offered support for this perspective as well. Because our criterion measures of HIV-risk behavior were retrospective, future researchers may wish to assess the replicability of our findings using prospective measures. Berk and associates (Berk, Abramson, & Okami, 1995) have indicated that retrospective reports of sexual behavior may be of limited reliability, in that participants in a recent investigation tended to overestimate slightly the frequency of past sexual behavior after a time lapse of only two weeks. These researchers used diaries in conjunction with retrospective reports, an approach that may be useful in future research.

In our second analysis, the limitations of our retrospective behavioral risk measure were more salient. In that model, expectations about the feasibility of safer sexual activity were related primarily to past experiences of condom use with nonsteady partners. Persons who had not successfully implemented condom use with partners in the past year did not expect to do so successfully in the future. Substance use appeared to be a factor in past HIV risk behavior, in that persons who typically used drugs or alcohol were less likely to use Condoms in sexual encounters with nonsteady partners. Correlational analyses had indicated that the perceived susceptibility aspect of the Health Belief Model seemed important in the prediction of safer sexual behavior. but that relationships were opposite those predicted by HBM. In contrast to ideas engendered by HBM, higher perceived susceptibility was related to lesser use of condoms with nonsteady partners and a greater number of sexual partners. This suggests that perceived susceptibility to HIV as measured by our survey was a successor rather than predecessor of sexual decision making. That is, persons who hadn't successfully implemented safer sexual behavior in the past thereby considered themselves to be at greater risk for HIV infection. Past research has indicated that perceived susceptibility has been less consistent than other HBM constructs in predicting HIV risk behavior (Wulfert & Wan, 1995). The instability of past findings may stem from conflicting correlations produced by reflections upon past behavior (positive association between risk behavior and susceptibility) and projections of future behavior (possibly an inverse association between susceptibility and risk behavior).

Our analyses indicated that specific subscales (e.g., substance use, expectations, perceived susceptibility) of our total scale may take on heightened importance in separate analyses in which condom use with nonsteady partners is used as a criterion measure. In previous exploratory analyses, many resulting relationships had been masked because condom use with steady partners and condom use with nonsteady partners were aggregated into a single score. Our results implied that different factors are important for prediction of safer sexual activity with steady versus nonsteady partners. Our findings of different predictor variables for steady versus nonsteady partners are consistent with a recent study by Galavotti et al. (1995). They found a proposed model to predict condom use with women's "main partner, but the model was less successful for prediction of condom use with other partners. The separability offered by Our internally reliable subscales was advantageous in this respect. Our separate gender analyses indicated that relationships among constructs may differ for men and women. Although attitudes. perceived susceptibility, intentions, and expectations predicted condom use with steady partners for women, these relationships did not hold for men. This is in contrast to past research that has indicated that attitude-behavior relationships are similar for men and women, but that women are more inhibited about buying and carrying condoms (Sacco et al., 1991, 1993). Gender norms are said to influence the latter finding, in that the decision to use condoms has traditionally been viewed as a male perogative, and men seem to exert more control over this decision (Sacco et al., 1993). Subscales tended to predict the sexual behavior of men more for number of partners and condom use with nonsteady partners, with substance use of male, but not female, participants being especially influential. Possibly men defer condom-use decisions to partners within steady relationships but may feel less compelled to comply with casual partners and when intoxicated. Past researchers have not generally performed separate analyses ,with regard to steady versus nonsteady partners and have targeted the attitudes of men and negotiation skills of women for intervention. Our separable subscales for measurement of such attitudes and skills (negotiation/communication skills are included on the expectation subscale) may offer benefits to researchers who wish to perform gender-specific assessments and interventions.

Factor analyses by gender indicated that our observed factor structure also differed somewhat when men's and women's responses were analyzed separately. Women's differentiation of condom use versus safer sex on the intentions components may relate to the aforementioned gender norms regarding sexual decision making. Because the current sample consists of about twice as many women as men, future researchers might explore the applicability of our findings among larger samples in which men and women are equally represented. Data concerning the applicability of findings to non-college populations and more ethnically diverse populations would also be valuable. For instance, do sexual risks for commercial sex workers, teens, gay men, or ethnic minorities? Our sampling of college students was largely a matter of convenience. Although HIV risk of college students is rising, with case reporting among adolescents and those who contract HIV through heterosexual contact being among the fastest rising groups, only about 1 in 500 college students is HIV positive (Centers for Disease Control, February 1996, personal communication). This prevalence rate is low relative to higher-risk samples of poverty-level, innercity ethnic minorities. Nevertheless, peer pressure, lack of maturity, and alcohol or drug use may put college students at risk for HIV (Centers for Disease Control, February 19%, personal communication). The perceived invulnerability to HIV expressed by college students (Wulfert & Wan, 1995) also renders college populations worthy of study with regard to IRV risk attitudes, behaviors, and beliefs. Wulfert and Wan (1995) noted that facilitating realistic risk appraisals among college students may be one avenue of intervention.

The current measure is not intended to replace more thorough measures of a single construct. For instance, Sacco et al. (1991) and Helweg-Larsen and Collins (1994) developed relatively comprehensive scales that are focused solely upon attitudes toward condom use. Our scale, however, offers a brief global measure of attitudes, as well as concise measures of constructs that are similar across various theoretical models. Wulfert and Wan (1995) recently noted that the Health Belief Model, the Theory of Reasoned Action, and social-cognitive models of health behavior share common emphases upon cognitive-affective reactions (attitudes in the current measure) and social context norms. These researchers found that those two constructs, along with the social-cognitive construct of efficacy beliefs (included in our expectations subscale successfully predicted intention to practice safer sex. Prochaska et al. (1994) suggested that models may be mutually enhanced when studied in conjunction with one another and that the ability of a chosen model to integrate core constructs of other models offers a more comprehensive study of HIV prevention. In an approach similar to that of Prochaska and associates, Grimley et al. (1993, combined a "stages-of-change" model with a decision-making model for investigation of contraceptive behavior. Their "trans-theoretical model" incorporated stages such as precontemplation (not yet considering behavior change), contemplation, and action (when actual changes are enacted) with "pro" and "con" decisions regarding contraceptive use at each stage. Findings demonstrated distinct patterns of perceived costs relative to perceived benefits within particular stages of change. The researchers also found that college students were more advanced in the stages with respect to contraceptive use for pregnancy prevention versus sexually transmitted disease prevention. Such differences indicate that use of measures such as the current measure, which is focused upon HIV prevention, may yield results that diverge from previous contraceptive research, especially when studied in the context of stages of change, when various attitudes, beliefs, and behaviors may differentially facilitate behavior change across stages (Galavotti et al., 1995; Prochaska et al., 1994). Thus, our Sexual Risks Scale may provide a useful vehicle for researchers who wish to study HIV-preventive attitudes, norms, expectations, intentions, perceived susceptibility, and risk behaviors within the framework of new or preexisting theoretical models. We hope that the scale will be a useful tool for health psychologists, community researchers, and other persons wishing to evaluate these dimensions in the contexts of basic and applied research.

References

Bagozzi, R. P. & Warshaw, P. R. (1990). Trying to consume Journal of Consumer Research, 17, 127-140.

Bagozzi, R. P. Yi, Y., & Baumgartner, J. 1990. The level of effort required for behavior as a moderator of the attitude-behavior relation. European Journal of Social Psychology, 20, 45-59.

Becker, M. 1974. The Health Belief Model and personal health behavior. Thorofare, NJ: Slack.

Bentler, P. M., & Speckart, G. (1979). Models of attitude-behavior relations. Psychological Review, 86, 452-464.

Berk. R., Abramson. P. R.. & Okami. P. (1995). Sexual activities as told in surveys. In P. Abramson & S. Pinkerton (Eds.), Sexual nature sexual culture (pp. 371-386). Chicago: University of Chicago.

Bienvenu, M. J., Sr. (1980). Counselor's and teacher's manual for the Sexual Communication Inventory. Saluda. NC: Family Life Publications.

Brown, I. S. (1984). Development of a scale to measure attitudes toward the condom as a method of birth control. The Journal of Sex Research. 20, 255-263.

Byrne, D., & Fisher, W. A. (1983). Adolescents, sex, and contraception. Hillsdale, NJ: Erlbaum.

Crowne, D. P., & Marlowe, D. (1964). The approval of motive: Studies in evaluative dependence. New York: Wiley.

Davidson, A. R., & Morrison, D. M. (1983). Predicting contraceptive behavior from attitudes: A comparison of within- versus between-subjects procedures. Journal of Personality and Social Psychology, 45, 997-1009.

Dunn, O. J. (1961). Multiple comparisons among means. Journal of the American Statistical Association, 56, 52-64.

Edwards, A. L. (1957). Techniques of attitude scale construction. New York: Appleton-Century-Crofts.

Fishbein, M., & Ajzen, I. (1975). Belief attitude, contention, and behavior. Reading, MA: Addison-Wesley.

Fishbein, M., & Middlestadt, S. E. (1989). Using the Theory of Reasoned Action as a framework for understanding in changing AIDS-related behaviors. In V. M. Mays, G. W. Albee, & S. F. Schneider (Eds.), Primary prevention of AIDS: Psychological approaches (pp. 93-110). Newbury Park, CA: Sage.

Fisher, J. D., Fisher, W. A., Williams, S. S., & Malloy, T. E. (1994). Empirical tests of an Information-Motivation-BehavioraI Skills Model of AIDS-Preventative Behavior with gay men and heterosexual university students. Health Psychology, 13, 238-250.

Fisher, W. A. (1984). Predicting contraceptive behavior among university men: The role of emotions and behavioral intentions. Journal of Applied Social Psychology, 14, 104-123.

Fisher, W. A., Byrne, D., Edmunds, M., Miller, C. T., Kelley, K., & White, L. A. (1979). Psychological and situation-specific correlates of contraceptive behavior among university women. The Journal of Sex Research, 15, 38-55.

Galavotti, C., Cabral, R. J., Lansky, A., Grimley, D. M., Riley, G. E., & Prochaska, J. O. (1995). Validation of measures of condom and other contraceptive use among women at high risk for HIV infection and unintended pregnancy. Health Psychology, 14, 570-578.

Grimley, D. M., Riley, G. E., Bellis, J. M., & Prochaska, J. O. (1993). Assessing the stages of change and decision-making for contraceptive use for the prevention of pregnancy, sexually transmitted diseases, and Acquired Immunodeficiency Syndrome. Health Education Quarterly, 20, 455-470.

Hall, J. (1992). Social norms and safer sex: A theory-driven AIDS risk reduction intervention. Unpublished manuscript.

Hearst, N., & Hulley, S. (1988). Preventing the heterosexual spread of AIDS: Are we giving our patients the best advice? Journal of the American Medical Association, 259, 2428-2432.

Hebert, Y., Bernard, J., De Man, A. F., & Farrar, D. (1989). Factors related to the use of condoms among French-Canadian university students. Journal of Social Psychology, 129, 707-709.

Helweg-Larsen, M., & Collins, B. E. (1994). The UCLA Multidimensional Condom Attitudes Scale: Documenting the complex determinants of condom use in college students. Health Psychology, 13, 224-237.

Herzberger. S., Chan. E., & Katz. J. 1984 The development of an assertiveness self-report inventory. Journal of Personality Assessment, 48, 317-323.

Jemmott, L. S., & Jemmott, J. B. (199). Applying the Theory of Reasoned Action to AIDS risk behavior: Condom use among Black women. Nursing Research. 40, 228-234.

Kelly, J. A., St. Lawrence, J. S., & Brasfield, T. L., (1991). Predictors of vulnerability to AIDS risk behavior relapse. Journal of Consulting and Clinical Psychology, 59, 163-166.

Kirscht, J. P., & Joseph, J. G. (1989). The Health Belief Model: Some implications for behavior change, with reference to homosexual males. In V. M. Mays, G. W. Albee, & S. F. Schneider (Eds.), Primary prevention of AIDS: Psychological approaches (pp. 111-127). Newbury Park, CA: Sage.

Leigh, B., & Stall, R. (1993). Substance use and risky sexual behavior for exposure to HIV. Issues in methodology interpretation, and prevention. American Psychologist, 48, 1035-1045.

May, R. M., Anderson, R. M., & Blower, S. M. (1988). The epidemiology and transmission dynamics of HIV-AIDS. Daedalus, 119, 162-201.

Padian, N., Marquis, L., Francis, D., Anderson, R., Rutherford, G. W., O'Malley, P. M., & Winkelstein, W. (1987). Male-to-female transmission of human immunodeficiency virus. Journal of the American Medical Association, 258, 788-790.

Prochaska, J. O., Velicer, W. F., Rossi, J. S., Goldstein, M. G., Marcus, B. H., Rakowski, W., Fiore, C., Harlow, L. L., Redding, C. A., Rosenbloom, D., & Rossi, S. R. (1994). Stages of change and decisional balance for 12 problem behaviors. Health Psychology, 13, 39-46.

Reiss, I.. & Leik, R. (1989). Evaluating strategies to avoid AIDS: Number of partners vs. use of condoms. The Journal of Sex Research. 26, 411-433.

Sacco, W. P. Levine. B., Reed. D. L., & Thompson. K. (1991). Attitudes about condom use as an AIDS-relevant behavior: Their factor structure and relation to condom use. Psychological Assessment. 3, 265-272.

Sacco. W. P., Rickman. R. L.. Thompson. K., Levine. B., & Reed, D. L. (1993). Gender differences in AIDS-relevant condom attitudes and condom use. AIDS Education and Prevention, 5, 311-326.

Sheeran. P. Abraham, S. C. S., Abrams, D., Spears, R., & Marks. D. (1990). The post-AIDS structure of students' attitudes to condoms: Age, sex and experience of use. Psychological Reports, 66, 212.

Shernoff, M. (1988). Integrating safer-sex counseling into social work practice. Social Casework, 69, 334-339.

Shilts, R. (1987). And the band played on. New York: Penguin.

Tabachnick, B. G., & Fidell, L. S. (1989). Using multivariate statistics. 2nd ed. New York: Harper-Collins.

Wang. C. K. A. (1932). Suggested criteria for writing attitude statements. Journal of Social Psychology, 3, 367-373.

Warshaw, P. R., & Davis, F. D. (1985). Disentangling behavioral intention and behavioral expectations. Journal of Experimental Social Psychology, 21, 213-228.

Wulfert, E., & Wan, C. K. (1995). Safer sex intentions and condom use viewed from a health belief, reasoned action, and social cognitive perspective. The Journal of Sex Research, 32, 299-311.

Manuscript accepted April 15, 1996

Portions of this data were presented at the 1995 conference of the Southeastern Psychological Association in Savannah GA. We thank Linda Bledsoe and Maureen Wilson for assistance in conducting Study 2 and Jamie McPherson of the Jefferson County Department of Health for encouraging development of the project. Please address all correspondence to Dann D. DeHart, Ph.D., Department of Psychology University of South Carolina, Columbia, SC 29208. Phone: [phone]. Fax: [phone]. E-mail: DEHARTD@GARNET. CLA.SC.EDU.