Internet sexuality research with rural men who have sex with men: can we recruit and retain them?

HIV seroprevalence in rural areas has increased steadily since 1982 (Lam & Liu, 1994). In rural areas, men who have sex with men (MSM) account for a majority of the HIV/AIDs cases (Centers for Disease Control and Prevention, 2001). Quantitative research studies with rural MSM are sparse, probably because traditional methods of recruiting are difficult with such highly stigmatized groups, and the travel expenses required to make face-to-face contact are high. The emergence of the internet, both as a virtual social community and as a research tool, presents an excellent opportunity for contacting rural populations. Rural people, particularly rural MSM, are rapidly increasing their use of the internet (Bell, Reddy, & Rainie, 2001), rural MSM in particular. In Wyoming (Williams, Bowen, & Horvath, 2005), MSM indicate that the internet is a primary resource for making contact with other MSM and that internet-delivered risk reduction interventions would be welcomed.

The internet's accessibility, affordability, anonymity (Cooper, 1998), and acceptability (King, 1999) make it especially attractive for research with hidden and stigmatized groups. Anyone with access to a computer and a modern is a potential participant, and respondents may access the research at a time and place convenient to them (Binik, 2001; Mustanski, 2001). Internet data collection is rapid and reasonably inexpensive, and large numbers of participants can be recruited from diverse locations (Cooper, Scherer, & Mathy, 2001). Samples appear to provide more variability than face-to-face interviews (Ross, Tikkanen, & Mansson, 2000; Tikkanen & Ross, 2000). Also, surveys can be tailored to the participants' responses and branching is invisible, thus lowering the participant burden.

Conversely, the internal validity of internet research may be threatened if participants feel less pressure to complete surveys (Frick, Bachtiger, & Reips, 2001), resulting in high "hit" to completion rates (Birnbaum & Mellers, 1989; Ross, Danebeck, Mansson, Tikkanen, & Cooper, 2003) or early abandonment (Crawford, Couper, & Lamias, 2001). Studies with urban MSM support this concern. For example, Ross, Rosser, Stanton, and Konstan (2004) recruited Latino MSM for an internet study and received 33,024 clicks on their banner ads. While 1,742 (5%) of the "click through" men qualified for the study and said they wanted to enroll, only 1,546 (89%) initiated the questionnaire. The survey completion rate for initiators was 65.4%. Similar hit to initiation (0.7%) and survey completion rates (51.2%) were reported by Ross et al. (2003) in their internet study of Swedish MSM. On the other hand, Elford, Bolding, Davis, Sherr, and Hart (2004) conducted an online survey of MSM in London, and although they did not report absolute completion rates, 97% of 1,250 men provided sufficient data to examine HIV status and sexual risk behaviors.

Three important outcomes should be noted from these studies: web banners attract urban MSM, including minorities; banners successfully recruit large numbers of initiators; and survey completion rates are often low. Ross et al. (2004) identified some correlates of completion among their urban participants. The men identified their preferred reimbursement method prior to initiating the questionnaire, and dropouts were more likely to have rejected the $20 compensation or offered it to a charity. Other correlates of dropping out included having not met men for sex on the internet, identifying as bi- or heterosexual, and using the internet less at home. Bull, Lloyd, Rietmeijer, and McFarlane (2004), in an internet-based prevention program for MSM, cited lack of contact information and inadequate incentives as primary issues leading to only 13% retention at the one-month follow-up. Meaningful compensation has also been identified as a successful retention strategy in non-internet studies (Leonard et al., 2003).

Internet research may be ideally suited to recruiting rural MSM, because participation can occur anonymously and the cost of traveling long distances to remote areas can be eliminated. On the other hand, if drop-out rates are high, the generalizability of the data, even among internet-using rural MSM, will be problematic. Although preliminary research suggests that banner ads and reimbursement are useful in internet research with urban MSM, little is known about their effects on recruiting or retaining rural MSM. This study aimed to examine the utility of internet banner ads for recruiting rural MSM for an online sexuality survey and to identify demographic, monetary, and behavioral correlates of survey initiation and completion.

METHOD

Sample

The sample included 1,045 men 18 years of age or older who self-identified as gay or bisexual and lived in rural areas. Rural areas were defined as towns of less than 75,000 people more than 1 hour's drive from an urban area (Bowen, Williams, & Horvath, 2004). The sample was "cleaned" to eliminate duplicate participation by a single individual using a number of variables. Specifically, surveys were compared for time proximity (surveys submitted within 2 to 15 minutes), similar e-mail addresses, passwords, age, and zip code. Duplicates who completed the survey were provided with reimbursement if appropriate and deleted from the database. This project was approved by the University of Wyoming Institutional Review Board.

Banner ads resulted in three groups of men recruited from a popular internet dating site from December 18-31, 2002, and from January 18-31, 2003. The "expected pay, paid" group responded to banners that ran between December 18 and 25. Banners indicated a "Rural Men's Sex Survey" and rolled over to "Click here to see of you qualify to earn $25." Interested men (i.e., men who clicked the banner) completed a brief set of screening questions, and those who qualified were told they would be reimbursed $25. The "expected pay, not paid" group was recruited from December 25 to January 31, and responded to the same banners as the "expected pay, paid" group. Men who clicked the banner were told immediately that, due to limited funds, they would not be reimbursed. This unplanned and brief deception was due to the large response in the first eight days that depleted the project's reimbursement funds. The men were assured that their information was important and they could continue to see if they qualified. Those who chose to continue proceeded to the screening questions. The "no pay expected, not paid" group was recruited in January and responded to new banners in which no money was advertised. These men were welcomed and immediately completed the screening questions.

Measures

Screening questions. The screening questions included eligibility questions (age, sexual identity, size of town of residence), along with nonintrusive questions about internet use. Previously completing questionnaires online was coded "yes" or "no." Motivation for completing online questionnaires included the following four yes/no questions: "find out what other men say," "hear good things about the study," "earn money," and "support the gay/bisexual community." Frequency of chat room use to meet men was recoded into "once per week or more" and "less than once per week." Reasons for chatting online included the following five yes/no questions: "get information," "to chat," "make friends," "find long term partner," and "find a sex partner." Finally, the time of day in which the questionnaire was completed was recorded and grouped into 4 time periods: 6 a.m. to noon, noon to 6 p.m., 6 p.m. to midnight, and midnight to 6 a.m.

Payment effects. Two variables were developed to examine payment expectation and reimbursement effects. The expectation variable compared the group of men who expected pay but were not paid to the rest of the sample. The reimbursement variable compared the men who expected pay and were paid to the rest of the sample.

Wyoming Rural AIDS Prevention Project Questionnaire (WRAPP). The WRAPP questionnaire resulted in 338 discrete responses. Demographic measurements included ethnicity, income, marital status, full-time employment, student status, and education. Age was recoded into five categories: 18-24, 26-29, 30-34, 35-39, and over 40 years of age. Respondents reported their sexual identity as gay, bisexual, or heterosexual. Only four men identified as heterosexual, so they were deleted from the analyses. Due to the small minority sample, ethnicity was coded as White (non-Hispanic) or all others. Income was recoded into more or less than $25,000 per year. Job status in the past 30 days included two separate variables: "regular full-time" or "not," and "student" or "not." Marital status included three categories: single/never married, living with a partner of the opposite sex, or other.

Three HIV-risk behaviors used in the analyses included number of sex partners in the last 90 days, HIV testing, and number of STDs in the last year. These were chosen because they were presented early in the questionnaire, so many men who chose to terminate early had completed these questions. Number of sex partners was recoded as an ordinal variable with six levels--none, 1, 2, 3, 4 to 5, more than 6--to reduce skew. Two HIV-testing variables were created to compare untested versus tested (positive or negative) and HIV-negative versus HIV-positive/unknown. The number of STDs was coded as 0, 1, or 2 or more.

Procedure

Men who clicked a banner ad were directed to a welcome page that briefly introduced the project. The welcome page differed slightly for the three groups. The "expected pay, paid" group was encouraged to complete the screening questions to "see if you qualify for $25." The "expected pay, not paid" group was told that due to limited funds, no reimbursement was available. The "no pay expected, not paid" group was welcomed with no mention reimbursement. Each page encouraged the men to click the "see if I qualify" button, which directed them to the screening questions.

Men who qualified were routed to the "Welcome to the WRAPP Project" page, which either reconfirmed eligibility or non-eligibility for reimbursement or did not mention reimbursement. The page included a "next" button to proceed to the consent form that included a complete description of the study. Consent was indicated by the participant developing his own password and pressing the "I agree" button. The participant was then automatically directed to the WRAPP Questionnaire, which took about 25 minutes to complete. Finally, participants in the "expected pay, paid" group chose a reimbursement method. Participant reimbursement options included an online gift certificate, direct payment through an online service, or "no reimbursement." Men who did not qualify were routed to a "thank you" page and then back to the WRAPP home page.

Analyses

The first analyses examined predictors of completing the screening questions only (initiators) or providing consent and continuing to the WRAPP questionnaire (participants). Completion was coded as a dichotomous variable because of the bimodal skew of number of questions completed. The second analyses included only participants who provided informed consent and proceeded to the WRAPP questionnaire. The dependent variable was the absolute number of responses (M = 278.50, SD = 72.31, Median = 306, mode = 311; minimum = 42, maximum = 338). The number of responses variable was reflected and log-transformed due to its negative skew (-2.04) and kurtosis (2.84), so that a lower score indicates more responses completed.

In both sets of analyses, differences in demographic variables across the two groups were examined using the Mann-Whitney U test. Spearman's rho correlations were conducted to identify variables to be entered into the logistic regression, due to the large number of demographic and behavioral variables. Given the exploratory nature of the study, I chose to err on the side of inflated Type 1 error and considered p < .05 to be a significant relationship. Stepwise logistic and stepwise linear regressions were used to examine the contribution of significant variables to initiation/participation and number of responses, respectively.

RESULTS

Recruiting Rural MSM on the Internet

The banner ads resulted in 1,045 rural MSM clicking the banners to find out more about the study in period of four weeks. Banners that advertised pay resulted in 674 men clicking through, while banners that did not advertise pay resulted in about 50% fewer men (n = 371). The sample was geographically diverse, with men representing 49 states, plus 27 men from Canada, 10 from Australia/New Zealand, and 5 from other countries. The statewide distribution appears to be generally representative of state populations. Rhode Island was the only state not represented, possibly because the distance requirements of 60 minutes from an urban center could not be met. Three states were represented by only two men (Connecticut, Delaware, and Nevada), and Hawaii had only four. The remaining states were represented by a minimum of six men (Maryland) and a maximum of 69 (California).

Initiators versus Participants

The sample included 314 initiators who completed the screening questions only and 731 participants who completed the screening questions plus the consent and some WRAPP questionnaire items. Table 1 illustrates the sample size for each demographic category, percentage of men in the category who became participants, and the results of the Mann-Whitney U test comparing the percentage of initiators to participants across demographic categories. No significant differences were found between the groups.

Correlates of initiation versus participation. Correlations were conducted between the dependent variable (initiator or participant) and the following independent variables: age, sexual identity, town size, completion of other online questionnaires, four motivation questions, five reasons for using chat rooms, time of day, and the two payment effect variables (expectation and reimbursement). Five variables were significant at p < .01: "support the gay community as motivation to complete an online questionnaire," "use chat rooms to 'chat' with other men," "use chat rooms to find friends," and the two payment effect variables. Two variables were significant at p < .05: sexual identity and frequency of using chat rooms. The intercorrelations between significant independent variables were all less than r = .47.

Contribution of independent variables to participation. Stepwise logistic regression was performed in two blocks. In Block 1, the two payment effect variables were entered, and in Block 2, the five significant demographic and behavioral variables were added. Valid data were available for 231 initiators and 653 participants.

In Block 1, only the expectation variable entered the equation, and it showed a statistically significant improvement over the constant-only block, [X.sup.2.sub.(df = 1)] = 75.08, p <.000, indicating that expectations distinguished between initiators and participants. Block 2 included two steps. "Chat to make friends" entered first with an improvement [X.sup.2.sub.(df = 1)] = 8.74, p <.003 and sexual identity entered second with an improvement [X.sup.2.sub.(df = 1)] = 4.92, p <.03, for an overall Model [X.sup.2.sub.(df = 3)] = 88.74, p <.000. The prediction success was modest and variable, with 17% of initiators and 96% of participants correctly classified for an overall prediction rate of 75%. The Nagelkerke pseudo [R.sup.2] suggests that the full model accounts for 14% of the variance. No outliers were found. Table 2 shows the regression coefficients, Wald statistics, odds ratios, and 95% confidence intervals for odds ratios for each predictor in Blocks 1 and 2.

Level of participation. The sample included 722 rural MSM who provided informed consent, initiated the WRAPP questionnaire, and had valid data. A participant was considered to be a completer if he responded to the social-cognitive theory questions regarding "the last man you had anal sex with," or, if he denied having anal sex, if he answered the comparable questions about oral sex. Table 3 illustrates the sample size for each demographic category, the percentage of men in the category who completed the questionnaire, and the results of the Mann-Whitney U test comparing the percentage of drop-outs (N = 170) to completers (N = 552). Income was the only demographic variable that was significantly different across groups, with men making more than $25,000 slightly more likely to complete the survey (79%) than the lower-income men (73%). The three reimbursement groups differed significantly in their completion rates: 84% of the paid group, 70% of the unpaid group, and 62% of the expected pay, not paid group completed the questionnaire.

Correlates of number of responses. Correlations between the transformed dependent variable (number of responses), the two payment effect variables, demographics, behavioral variables, and HIV-risk variables were examined. Three variables correlated at p < .01 with the dependent variable: expectation (r = -.17), payment (r = .12), and number of sexual partners in the last 90 days (r = -.15). Four variables correlated at p < .05: having a college degree (r = -.10), being motivated by money to complete online questionnaires (r = -.10), chatting to find friends (r = .09), and being tested for HIV (r = .10). The intercorrelations between the seven significant independent variables were all less than r = .38.

Contribution of independent variables to number of responses. The stepwise linear regression was performed in two steps. The first step included the two payment effect variables, and the second step added the significant demographic, behavioral, and HIV-risk behavior variables. Four outliers and three participants with missing values were excluded, making a final sample of 569 men. Measures of collinearity were within acceptable limits.

Table 4 displays the regression coefficients (B), the standardized regression coefficients, partial correlations, and [R.sup.2] and the adjusted [R.sup.2] for variables in the equation. R for regression was significantly different from zero at the end of each step. Four independent variables contributed significantly to prediction of number of questions (reflected, [log.sub.10]) completed. In Model 1, reimbursement entered the equation; [R.sup.2] = .03, F (1,567) = 18.06, p < .000. In Model 2, number of sex partners was added; [R.sup.2] (change) = .019, F (1,566) = 11.60, p = .001. In Model 3, unknown HIV status was added; [R.sup.2] (change) = .007, F (1, 565) = 4.36, p = .04, and in the final step, being motivated by money was added; [R.sup.2] (change) = .009, F (1,564) = 5.50, p = .02. After the four independent variables entered the equation, the statistics were R = .26, F (4, 564) = 10.08, p < .000. Altogether, the model accounted for only 7% (6% adjusted) of the variability in the number of responses. Post-hoc F tests of the correlations between number of responses and the non-significant independent variables indicated that they were not significant (p > .05).

DISCUSSION

As internet research expands, questions have arisen among researchers about the need to examine the internal and external validity and generalizability of internet studies (Cooper et al., 2001). A number of recent internet research studies have examined the external validity of internet samples by comparing them to non-internet samples. When taken together, the results of these studies suggest that internet convenience samples do not show systematic differences from non-internet samples (Rhodes, DiClemente, Cecil, Hergenrather, & Yee, 2002; Ross et al., 2000; Tikkanen & Ross, 2000). This study begins to address two additional methodological issues: the utility of internet banners ads for recruiting hidden, highly stigmatized populations, and the identification of factors associated with internet survey participation and completion.

This study includes one of the largest samples of rural MSM to date and supports using the internet for recruiting rural gay and bisexual men. Banner ads that indicate payment for a "rural men's sex study" resulted in more than 600 men accessing the Welcome page in only two weeks. Banner ads that did not indicate payment were about half as successful in the same amount of time. Additionally, the sample included rural MSM from across the United States and from Canada, Australia, and New Zealand. These findings demonstrate the utility of internet research for hidden populations in rural areas. Recruiting and "interviewing" required only the purchase of banners, the monitoring of reimbursement by one staff person once per day, and the downloading of the data.

Sampling issues that need further study include the limited representation of minority participants, MSM who identify as heterosexual, and men who identify as gay but have not had sex with other men. The banners in this study were generic ads for a sex survey for rural MSM. One approach to increasing participation by unrepresented groups might be to use specific messages on the banner ads. For example, Ross et al. (2004) successfully recruited Hispanic men using ethnicity-specific banners. Kalichman and Coley (1995) found that negatively-framed video messages presented by African American women to African American women resulted in 63% of these women getting an HIV test within two weeks. Ethnic- or sexual identity-specific message framing on banner ads in terms of gain or loss for engaging in health promotion might also be useful for increasing sample diversity.

The overall participation and completion rates in this study were very high. Almost 70% of the men who completed the screening questions decided to participate, and more than 80% of paid men completed the survey. Although I have no data to address the reason for the high participation rates directly, I suspect it is related to the novelty of research participation for rural MSM. As noted above, the sample size is large for a rural MSM study, and there are few large quantitative studies with rural MSM. Thus, the men may not have been saturated by research studies nor had negative experiences with research. It will be interesting in the future, as rural MSM have more opportunity to participate via internet studies, to see if their completion rates decrease.

The differential effects of payment and expectation of payment on survey initiation and completion are worth noting. "Truth in advertising" did not affect completion of screening questions, but it was a significant factor in the men continuing to provide consent and beginning the study questionnaire. This is interesting because the unpaid men who responded to banners that advertised payment learned that payment would not be offered prior to completing any questions. It appears that expecting one outcome and receiving another contributed to negative feelings about the questionnaire as the participant completed the screening questions. This expectation effect was no longer important for men who agreed to participate in the complete study questionnaire. As found in offline studies (Leonard et al., 2003), money was an important motivator of the men's responses. The effect of reimbursement is strengthened by the finding that men who said they were motivated by money to complete questionnaires online completed more questions than those who were not. The men in the WRAPP study were reimbursed $25. Further research might examine the effects of different reimbursement rates or different ways of advertising the reimbursement. For example, in this study, we could have pointed out that reimbursement of $25 for less than 30 minutes of work is the equivalent of $50/hour.

Behavioral and demographic factors also differ in their influence on initiation/participation and participation/completion. Although comparison of predictors of these two dependent variables is not completely equivalent, since more variables were available for predicting participation/completion, a few findings are relevant. Gay men were more likely to consent to the study than bisexual men, but once men of either identity consented to the study, sexual orientation did not affect the number of questions completed. It is possible that bisexual men identify less with HIV-risk factors than gay men, so the potential burden of survey completion may outweigh their interest in the study and its outcomes. Also, after completing the screening questionnaires, men who went online to chat with friends were more likely to decide to participate than those who went online for other reasons. This may have occurred because the men were online to relieve boredom, and the questionnaire provided an activity as well as something to talk about later. It is interesting that time of day did not affect participation, since anecdotal reports suggest that when men are online to seek sex partners, they might be less likely to participate in questionnaires. The lack of a time of day effect and the chatting effect may be uniquely rural findings. Most rural men must travel to meet sex partners, so it is possible that their online behaviors are less time-dependent (i.e., they can't go online, find a partner, and meet him all in the same evening). Thus, seeing and completing a survey in the evening as opposed to the afternoon may not interfere with specific goal-directed behaviors.

Although the number of survey responses completed was not related to demographic variables or reasons for chatting, it was related to having more sexual partners and having been HIV tested. One possible explanation is that men with more sexual partners are more concerned about their HIV risk and this leads them to get HIV tested. These men might use a survey as a means of examining their own risk behaviors. A similar explanation is that those men are just more interested in sex, and a sex survey can be a mechanism for "talking" about sexual behavior. In support of the first hypothesis, a number of men e-mailed the project stating that they had learned a lot about themselves and had developed a better understanding of their risk behaviors. Future studies might examine the effect of advertising that feedback will be provided at the end of the survey as an inducement to participate, or possibly using risk behavior questions as part of the screening and providing a risk estimate as an inducement to continue.

Finally, although I have noted a number of areas that affect recruiting and participation of rural internet sex survey participants, limits to the overall generalizability of the study should be noted. Rural minorities, heterosexual MSM, and men who identify as gay or bisexual but have not had sex with men were poorly represented. The HIV-risk behaviors of minority MSM, especially African American and non-gay identified MSM, are high (Rock et al., 2002), making it imperative to identify efficient methods for recruiting and retaining these men in HIV-risk research. Men who identify as gay but have not had sex are also important targets for prevention. If we can find ways to recruit them, prevention messages may result in sexual initiation that includes safer sex practices. Finally, although a number of potential facilitators of recruiting and retaining participants were identified, the overall variance accounted for by each regression was low. Future research should refine these factors and identify other predictors of study completion.

Note. This project was partially supported by grant #RO1-MH63667 from the National Institute of Mental Health. The opinions expressed herein are solely those of the author and do not reflect those of the Institute. I'd like to thank Keith Horvath and the journal reviewers for comments on earlier versions of the manuscript.

Manuscript accepted June 20, 2005

REFERENCES

Bell, P., Reddy, P., & Rainie, L., (2001). Rural areas and the internet. Washington, D.C., Pew Internet and American Life Project

Binik, Y. M. (2001). Sexuality and the internet: Lots of Hyp(otheses)--Only a Little Data. The Journal of Sex Research, 38(4), 281-282.

Birnbaum, M. H., & Mellers, B. A. (1989). Mediated models for the analysis of confounded variables and self-selection samples. Journal of Educational Statistics, 14, 146-158.

Bowen, A. M., Williams, M. W., & Horvath, K. (2004). Using the internet to recruit rural MSM for HIV risk assessment: Sampling Issues. AIDS & Behavior, 8, 311-319.

Bull, S., Lloyd, L., Reitmiejer, C., & McFarlane, M. (2003). Recruitment and retention of participants for randomized control trials online: The experience of the Smart Sex Quest project. Abstracts of the 2003 STD/HIV Prevention and Internet Conference, CDC. Washington, D.C.

Centers for Disease Control and Prevention. (2001). Percentage distribution of AIDS cases in male adults and adolescents, by risk exposure and population of area of residence diagnosed through 2001, United States. Retrieved February 19, 2005, from http://www.cdc.gov/hiv/graphics/ images/L206/1206_8.gif.

Cooper, A. (1998). Sexuality and the internet: Surfing into the new millennium. Cyberpsychology & Behavior, 1 187-193.

Cooper, A., Scherer, C., & Mathy, R. (2001). Overcoming methodological concerns in the investigation of online sexual activities. CyberPsychology & Behavior, 4, 437-447.

Crawford, S. D., Couper, M. P., & Lamias, M. J. (2001). Web surveys: Perceptions of burden. Social Science Computer Review, 19, 146-192.

Elford, J., Bolding, G., Davis, M., Sherr, L., & Hart, G. (2004). Web-based behavioral surveillance among men who have sex with men: A comparison of online and offline samples in London, UK. Journal of Acquired Immune Deficiency Syndrome, 35, 421-426.

Frick, A., Bachtiger, M-T, & Reips, U-D. (2001). Financial incentive, personal information, and drop out in online studies. In U-D. Reips & M. Bosnjak (Eds.), Dimensions of internet science. Berlin: Zagreb Pabst Science Publishers.

Kalichman, S. C., & Coley, B. (1995). Context framing to enhance HIV-antibody-testing messages targeted to African American women. Health Psychology, 14, 247-254.

King., S. A. (1999). Internet gambling and pornography: Illustrative examples of the psychological consequences of communication anarchy. CyberPsychology & Behavior, 2, 175-193.

Lam, N. S., & Liu, K., (1994). Spread of AIDS in rural America, 1983-1990. Journal of Acquired Immune Deficiency Syndrome, 7, 485-490.

Leonard, N. R., Lester, P., Rotheram-Borus, M. J., Mattes, K., Gwadz, M., & Ferns, B. (2003). Successful recruitment and retention of participants in longitudinal behavioral research. AIDS Education and Prevention, 15(3), 269-281.

Mustanski, R. B. (2001). Getting wired: Exploiting the internet for the collection of valid sexuality data. The Journal of Sex Research, 38, 292-301.

Rhodes, S. D., DiClemente, R. J., Cecil, H., Hergenrather, K. C., & Yee, L. J. (2002). Risk among men who have sex with men in the United States: A comparison of an internet sample and a conventional outreach sample. AIDS Education and Prevention, 14, 41-50.

Rock, W., Johnson, D., Lu, S., Jordan, W., Gildon, B., Currier, J., et al. (2002). HIV risk behaviors among African American men in Los Angeles County who self-identify as heterosexual. Journal of Acquired Immune Deficiency Syndrome, 31, 354-360.

Ross, M. W., Daneback, K., Mansson, S-A., Tikkanen, R., & Cooper, A. (2003). Characteristics of men and women who complete or exit from an on-line internet sexuality questionnaire: A study of instrument dropout biases. The Journal of Sex Research, 40(4), 396-402.

Ross, M., Tikkanen, R., & Mansson, S. (2000). Differences between internet samples and conventional samples of men who have sex with men: Implications for research and HIV interventions. Social Science and Medicine, 47, 749-758.

Ross, M. W., Rosser, B. R. S., Stanton, J., & Konstan, J. (2004). Characteristics of Latino men who have sex with men on the internet who complete and drop out of an internet-based sexual behavior survey. AIDS Education and Prevention, 16, 526-637.

Tikkanen, R., & Ross, M. W. (2000). Looking for sexual compatibility: Experiences among Swedish men in visiting gay chat rooms. CyberPsychology & Behavior, 3, 605-616.

Williams, M. L., Bowen, A. M., & Horvath, K. J. (2005). The social/sexual environment of gay men residing in a rural frontier state: Implications for the development of HIV prevention programs. The Journal of Rural Health, 21, 48-55.

Anne Bowen

University of Wyoming

Address correspondence to Dr. Anne Bowen, Psychology Department, University of Wyoming, Dept. 3415, 1000 E. University, Laramie, Wyoming, 82071-3415; e-mail: [e-mail].

Table 1. Comparison of Demographics Across Initiators
and Participants

                       N    Participants (a)   [Z.sup.2]      p

Age
  18-24               361         71.7         -.34 (b)      .74 (c)
  25-29               167         64.1
  30-34               127         66.1
  35-39               126         69.0
  [greater
  than or
  equal to] 40        264         73.5

Sexual Identity
  Gay                 850         71.6           -2.57       .01
  Bisexual            195         62.6

Reimbursement Group
  Paid                450         78.4           -1.93       .05
  Expected,           224         45.5
    but not
    Paid
  No Pay              371         74.4

(a) Row percent.

(b) Mann-Whitney U, with Z transformation comparing initiators
to participants across demographic categories.

(c) Significance based on the Monte Carlo procedure.

* p < .05

*** p < .001

Table 2. Stepwise Logistic Regression of Payment Expectations,
Demographics, Motivation, and Reasons for Chatting on
Participation in an Online Internet Questionnaire

Variable                                                   95%
in The                Regression                Odds    Confidence
Equation              Coefficient     Wald      Ratio    Interval

Model 1                  1.56       76.07 ***   4.76    3.35-6.76
Expectation (a)

                      [chi square] = 75.08,p <.000,
                      Nagelkerke [R.sup.2] = .12

Model 2

Expection                1.55       74.85 ***   4.73    3.33-6.73
Sexual Identity          0.43        5.04 *     4.51    1.06-2.25
 (Gay vs. Bisexual)
Chat for Friends        -0.51        7.29 **     .60    0.42-0.87
 (No vs. Yes)

(a) Expectation variable = "expected pay, not paid" vs. "other"

* p < .05

** p < .01

*** p < .000

Table 3. Level of Participation: Percentage
of Participants Who Completed the Questionnaire

                                  %
                        N    Complete (a)    Z (b)   P

Age
  18-24                259       72.2       -1.06   0.28
  25-29                107       82.2
  30-34                84        71.4
  35-39                87        77.0
  [greater
    than or
    equal
    to] 40             194       77.3

Sexual Identity
  Gay                  609       77.2       -1.88   0.07
  Bisexual             122       67.2

Ethnicity
  White
   (non-Hispanic)      632       75.6        -.00   1.00
  All Others           93        76.3

College Degree
  Yes                  302       75.8       -0.60   0.59
  No                   428       75.2
Income
< 25,000               385       73.0       -1.95   0.05
[greater than
  or equal
  to] 25,000           338       79.0
Reimbursement Group
  Paid                 353       83.6       -4.11   0.000
  Expected, but
    not Paid           102       61.8
  No Pay               276       70.3

(a) Row percent.

(b) Mann-Whitney U with Z transformation, comparing
    "drop-outs" to "completers" across demographic categories.

(c) Significance based on the Monte Carlo procedure.

* p < .05

*** p < .001

Table 4. Stepwise Linear Regression of Payment
Expectations, Demographics, Motivation, and
Reasons for Chatting, and HIV-Risk Variables on
the Number of Questions (reflected, [log.sub.10])
Completed in an Online Questionnaire

Variables                                        Partial
                              B       [beta]   Correlation

Reimbursement (yes) (a)    -.08 ***    -.15        -.15
Number sex partners
 (last 90 days)            -.02 ***    -.14        -.14
HIV tested (no)             .06 *       .10         .10
Motivated by money (yes)   -.06 *      -.10        -.10
Constant                   1.55

Model: [R.sup.2] =.07, Adj. [R.sup.2] =.06, R =.26***

(a) Reimbursement variable = "expected pay, paid" vs. "other."

* p < .05

*** p < .001