Demographic and Sexual Risk Predictors of Delay Discounting of Condom-Protected Sex

Abstract

Objective:

Sexual delay discounting describes the decreased likelihood of condom-protected sex if a condom is not immediately available, which can be quantitatively summarised using the Sexual Delay Discounting Task (SDDT). The present studies determined the extent to which condom use likelihood as assessed by the SDDT is associated with self-reported sexual risk behaviours and demographics in two online samples of adults.

Design:

Study 1 (n = 767) assessed demographics, sexual risk behaviour, and delay discounting, and examined relations between these variables using correlation and regression. Study 2 (n = 267) examined whether real-world instances of unprotected sex because a condom was not immediately available predicted greater sexual discounting.

Main Outcome Measures:

sexual delay discounting, condom use

Results:

Both studies observed significant positive relations between sexual delay discounting and self-reported sexual risk behaviours, and found that males tended to show greater sexual discounting. In Study 2, 46% of the sample self-reported having unprotected sex because a condom was not immediately available, and these individuals showed significantly greater sexual delay discounting.

Conclusion:

These results extend prior findings by demonstrating that delay is a critical variable underlying real-life sexual risk behaviour among non-clinical samples. The SDDT is an ecologically-valid measure of these processes.

Worldwide, there were approximately 1.8 million adults who became infected with HIV in 2017 (UNAIDS, 2018). Although the adoption of antiretroviral therapy has reduced overall mortality rates (e.g., Samji et al., 2013), HIV remains a chronic health condition associated with substantial costs (Djawe et al., 2015). Approximately 90% of all HIV infections in adults are transmitted via sexual contact (vaginal or anal intercourse; CDC, 2015). When used properly, male condoms reduce the risk of sexual HIV transmission by 70–80%, but many individuals engage in intercourse with risky partners without using condoms at all, or use them inconsistently (Smith et al., 2015; Weller & Davis-Beaty, 2002). Globally, condom use remains a major component of prevention approaches for sexually transmitted infections (STIs) (UNFPA, WHO, UNAIDS, 2015). Therefore, understanding factors that contribute to inconsistent condom use is an important goal for preventing the transmission of HIV and other STIs.

The reasons behind inconsistent use of condoms are complex and multifactorial, but one factor that may affect condom use decisions is delayed condom availability. For example, a person may generally prefer to use a condom in a casual sex scenario if there is one immediately available. However, if there is a delay to obtaining a condom, the same individual may choose to have unprotected sex immediately rather than wait to have sex with a condom. Delay as a factor affecting condom use is of particular interest, given that promoting condom use in such scenarios can be targeted by multiple points of intervention, including both providing readily available condoms and attempting to directly affect choice and improve decision-making when a condom is not immediately available. Delay discounting provides a framework for describing the effects of delayed consequences on choice, and refers to the observation that delaying a consequence decreases its value or impact on voluntary behaviour (Mazur, 1987). Delay discounting is typically studied by examining a series of choices between smaller-sooner and larger-later hypothetical monetary outcomes (Du et al., 2002; Green et al., 1994; Rachlin et al., 1991). Greater preference for smaller-sooner over larger-later outcomes reflects greater delay discounting and is associated with addiction and related diseases (Bickel et al., 2012). The Sexual Delay Discounting see (SDDT) was developed in order to apply this framework to sexual HIV risk, using hypothetical condom-use decisions rather than monetary decisions (Johnson & Bruner, 2012). In the SDDT, participants select hypothetical casual sex partners from an array of photographs and then report their likelihood of having immediate unprotected sex in a casual sex scenario versus waiting a specified delay (e.g., ranging from 1 hour to 3 months) to have sex with a condom. Larger decreases in the reported likelihood of condom-protected sex when condoms are delayed indicate greater discounting.

Research using the SDDT has found that reported condom use likelihoods tend to decrease as a function of increased delay to condom-protected sex (i.e., sexual delay discounting) (Bolin et al., 2016; Collado et al., 2017; Dariotis & Johnson, 2015; Herrmann et al., 2014; Herrmann et al., 2015; Johnson & Bruner, 2012; Johnson & Bruner, 2013; Johnson et al., 2015a; Johnson et al., 2017; Johnson et al., 2016; Koffarnus et al., 2016; Meredith et al., 2016; Quisenberry et al., 2015; Strickland et al., 2017; Wongsomboon & Robles, 2017). This research has also shown that sexual delay discounting is greater for partners that are rated as more sexually desirable, or perceived as less likely to have a STI (e.g., Johnson & Bruner, 2012), greater for men than women (Collado et al., 2017; Johnson & Bruner, 2013), and greater in cocaine- or opioid-using samples relative to non-drug-using samples (Herrmann et al., 2014; Johnson et al., 2015a; Koffarnus et al., 2016). The SDDT has established test-retest reliability (Johnson & Bruner, 2013), but also increases with acute alcohol and cocaine administration (Johnson et al., 2017; Johnson et al., 2016). Overall, these data suggest delay to condom-protected sex and individual differences in sensitivity to delay may be fundamental contributors to sexual HIV risk. Thus, the SDDT may be a valuable tool to identify risk factors that may exacerbate the effect of delay on condom use.

The present report summarises the results of two studies that were conducted to evaluate the extent to which condom use likelihood as assessed by the SDDT, which is hypothetical, corresponds to real-life sexual risk behaviours (e.g., unprotected sex) and determine situational or individual risk factors associated with increased discounting of condom-protected sex. We utilized online crowdsourcing technology to facilitate data collection. The use of crowdsourcing technology is a cost-effective approach to collecting experimental data from larger sample sizes than would typically be feasible in traditional in-person laboratory settings. Data collected using crowdsourcing technology has already provided useful insights into various decision making and health behaviours (Bickel et al., 2014; Jarmolowicz et al., 2012; Johnson et al., 2015b; vanDellen et al., 2017). A particular advantage of using crowdsourcing to collect data regarding sexual risk behaviours is that respondents, who are fully anonymous, may be more forthcoming regarding their past history and reported likelihoods of using a condom under various situations. Thus, we used respondents from Amazon Mechanical Turk crowdsourcing website across two studies in order to assess utility of the SDDT as an index of sexual risk behaviour and as a potentially useful tool for understanding decision-making mechanisms of sexual HIV and other STI risk.

Study 1

Previous studies have shown that greater sexual delay discounting is associated with greater self-reported sexual risk behaviour in individuals with cocaine use disorder (Johnson & Bruner, 2012), in high-risk adolescents and young adults from disadvantaged backgrounds (Dariotis & Johnson, 2015), in men who have sex with men (Herrmann et al., 2015), and in college students (Collado et al., 2017). The observed relation between sexual delay discounting and self-reported sexual risk behaviour adds confidence that the hypothetical condom use likelihoods reported by participants during the SDDT reflect clinically meaningful risk behaviours outside of the laboratory, but some open questions remain. For example, only one prior study which observed a relation between decisions on the SDDT and real-life sexual risk behaviours also assessed monetary delay discounting (Johnson & Bruner, 2012). When examining sexual delay discounting among cocaine users, the authors observed that sexual delay discounting was more strongly related to self-reported sexual risk behaviour than is monetary discounting, demonstrating domain specificity of money versus sexual outcomes (Johnson & Bruner, 2012). We hypothesize that the same previously observed relations will hold true in a non-clinical sample of adults, specifically, that real-life sexual risk behaviour will be associated with increased sexual delay discounting, and have a stronger relation with sexual discounting than monetary discounting, but this requires empirical examination. Further, examination of how basic demographic factors, including education, income, and cigarette smoking, may relate to SDDT is challenging among previously examined sub-populations (e.g., college students, substance users) as preexisting differences in prevalence and demographics may challenge generalizability. Based on prior research, it is hypothesized that males will show greater sexual delay discounting relative to females, and that smokers may show greater discounting relative to non-smokers, but other demographic factors have been little examined (Bickel et al., 2012; Collado et al., 2017; Johnson & Bruner, 2013). Thus, purpose of Study 1 was to examine real-life self-reported sexual risk behaviour, sexual delay discounting, and monetary discounting in a non-clinical sample of adults in order to evaluate the domain-specificity of the relationship between discounting and sexual risk behaviour and evaluate potential demographic predictors of sexual delay discounting.

Method

Participants

The Johns Hopkins University School of Medicine IRB approved all study procedures. Participants were recruited using the crowdsourcing website Amazon Mechanical Turk (MTurk; www.mturk.com). Informed consent was provided before a qualification survey. In order to qualify for the full survey, participants were required to be 18–28 years of age, reside in the United States, report consuming caffeine or alcohol in the last month, indicate feeling comfortable answering questions about drug use and sex, correctly answer two attention check questions during the screening questionnaire, and have a worker approval rating on MTurk of ≥95%. Qualified participants were compensated up to $3.00 for completing the main survey ($1.50 for completing the survey and $1.50 for correctly responding to attention check questions). Data from this sample are summarised in a prior report whose aim was to examine the relation between energy drink consumption and risk behaviour among young adults (Meredith et al., 2016). The present analyses examining the relation between discounting and real-life sexual risk behaviour have not been previously reported.

Survey

The survey was described as a behavioural health and decision-making study and contained questions about demographics, food and beverage consumption, substance use, risk behaviours, personality measures, and assessments of delay discounting.

Sexual Risk Behaviours

Participants were asked how many times in their lifetime they had sexual intercourse (a) without using a condom; for this question, married participants were asked how many times they had sex without a condom with someone other than their spouse, (b) with someone who was drunk or high, (c) while drunk or high, (d) with someone who they did not know very well, (e) that was later regretted (questions adapted from Miller, 2008). There were four response options where 1 = Not at all, 2 = Just once, 3 = 2–5 times, 4 = 6 or more times.

Sexual Delay Discounting Task (SDDT)

The SDDT assessed participants’ likelihood of having condom-protected sex in a casual sex scenario when a condom was immediately available versus when it was delayed (e.g., Johnson & Bruner, 2012). Participants started this task by viewing a diverse array of photographs (30 male; 30 female), and selecting photographs of any and all individuals with whom they would be willing to have sex, assuming they were not currently in a committed relationship, they liked the person’s personality, and that sex carried no risk of pregnancy. From this subset of selected photographs, participants made further distinctions by identifying the photograph of the individual they (1) most wanted to have sex with, (2) least wanted to have sex with, (3) thought most likely to have an STI, and (4) thought least likely to have an STI. For each of these four partner conditions (order counterbalanced), participants used a visual analog scale (VAS) to first rate the likelihood of having condom-protected vs. unprotected sex if a condom was immediately available (‘I will definitely have sex with this person without a condom’ = 0 on VAS to ‘I will definitely have sex with this person with a condom’ = 100 on VAS). Then, participants rated their likelihood of having unprotected sex immediately vs. waiting to have condom-protected sex (e.g., ‘I will definitely have sex with this person right away without a condom’ = 0 on VAS to ‘I will definitely wait 1 hour to have sex with this person with a condom’ = 100 on VAS) across several delays presented in ascending order (1 hour, 3 hours, 6 hours, 1 day, 1 week, 1 month, 3 months). A survey programming error resulted in the 1 month delay scenario being presented to only part of the sample, thus the one-month delay scenario was excluded from the analysis.

Monetary Delay Discounting

A 27-item Monetary Choice Questionnaire (MCQ) assessed discounting of hypothetical monetary outcomes (Kirby et al., 1999). On each question, participants chose between a smaller amount of money today and a larger amount of money after various delays (e.g., Would you prefer $25 today, or $60 in 14 days?). Discounting was summarised as the proportion of larger-later choices. Fewer larger-later choices indicate greater discounting (Myerson et al., 2014).

Data analysis

Data from the SDDT were evaluated for orderliness. Specifically, we determined the percentage of systematic and nonsystematic data according to prior criteria (Johnson & Bickel, 2008; Johnson et al., 2015a; Johnson et al., 2015b). Data were considered nonsystematic if across consecutive trials (in which delay to condom availability increased) likelihood of condom use increased by more than 20% of the previous trial, and/or if condom-use likelihood increased by more than 10% from the first to the last trial. We included all discounting data in the present analysis, even if characterised as non-systematic.

Data from the SDDT were summarised for each partner condition using (1) the reported likelihood of condom use with no delay, in order to assess baseline preference for condom use, and (2) Area Under the Curve (AUC) as an aggregate measure of discounting over all delays (Myerson et al., 2001), where lower AUC value indicates less likelihood of waiting for a condom (i.e., greater discounting of condom-protected sex). Before calculating AUC, likelihood of condom use at each delay was expressed as a proportion of the participants’ baseline likelihood of condom use with no delay. This standardization procedure isolates the effect of delay on condom use likelihood (e.g., Meredith et al., 2016). Because of this, individuals who reported a 0% likelihood of using an immediately available condom were excluded from AUC calculation in that partner condition. If the likelihood of immediate condom use is 0%, there is no ability to evaluate the potential decreasing effect of delay.

We used correlation analyses to examine the relationships between variables of (1) likelihood of condom use with no delay, (2) Sexual Delay Discounting Task (SDDT) AUC, and (3) proportion larger-later choices on the MCQ and the following predictor variables: (1) sex (male or female), (2) age, (3) income, (4) education, (5) tobacco smoking status (past month yes or no), and (6) each of the five sexual risk questions described above. Point-biserial correlations were conducted for dichotomous variables (sex, smoking status), Spearman’s rank order correlations were conducted for ordinal predictors (income, education, sexual risk questions), and Pearson’s r was assessed for age. We also conducted a linear regression with all of the above variables as predictors in the same model for each outcome to account for shared variance. Analyses for likelihood of condom use with no delay and SDDT AUC were conducted separately for each partner condition.

Results and Discussion

Participants

A total of 1014 individuals qualified for the study and began the main survey. Of these, n=140 did not complete the main survey or failed to pass attention check questions, and an additional n=107 were not presented with the SDDT because they did not select at least two hypothetical sexual partners from the array of photographs (see task description), resulting in a final sample of 767 individuals. Participant demographic characteristics for Study 1 are presented in online supplementary materials (Supplementary Table 1).

Discounting and Sexual Risk Behaviours

Median condom use likelihood for each of the four partner conditions are displayed in Figure 1. Figure 1 shows that median condom use likelihood (expressed as a proportion of the likelihood of condom use when there was no delay) tended to be lower for the more preferred partner conditions (i.e., ‘most want to have sex with’, ‘least likely to have an STI’) relative to the less preferred partner conditions (i.e., ‘least want to have sex with’, ‘most likely to have an STI’). Specifically, likelihoods of condom use with no delay (i.e., proportion VAS) were lower for the ‘most want to have sex with’ partner (M = .66, SD = .40) compared to the ‘least want to have sex with’ partner (M = .82, SD = .30), as well as the ‘least likely to have an STI’ partner (M = .66, SD =.39) relative to the ‘most likely to have an STI’ partner (M = .92, SD = .20).

Figure 1 Caption.

Figure 1 shows median condom use likelihood at each delay (0, 1 hour, 3 hours, 6 hours, 1 day, 1 week, and 3 months) across each partner condition. Condom use likelihood at each delay is expressed as a proportion of the likelihood of condom use when there was no delay.

Percent of sexual delay discounting data considered systematic according to the aforementioned criteria ranged from 91–93% across the four partner conditions. Overall mean proportion larger-later choices on the MCQ was .43 (SD = .19). Mean SDDT AUC values, where smaller values indicate greater discounting of condom-protected sex (i.e., less likelihood of waiting for a condom), were also lower for the ‘most want to have sex with’ partner (M = .44, SD = .41) relative to the ‘least want to have sex with’ partner (M = .69, SD = .38), and the ‘least likely to have an STI’ partner (M = .48, SD = .41) relative to the ‘most likely to have an STI’ partner (M = .77, SD = .34), which is consistent with previous literature.

Table 1 shows individual correlations between likelihood of condom use with no delay, SDDT AUC, and proportion larger-later choices on the MCQ with demographic variables and self-reported sexual risk behaviours. Participants’ biological sex was significantly correlated with likelihood of condom use with no delay for three of four partner conditions, as well as SDDT AUC for all partner conditions, showing that females tended to report a higher likelihood of condom use relative to males. Participant sex, however, was not significantly correlated with proportion larger-later choices on the MCQ. Age and income were not consistently related to likelihood of condom use with no delay, SDDT AUC, or proportion larger-later monetary choices. Education was significantly correlated with likelihood of condom use with no delay (three out of four partner conditions), SDDT AUC (all partner conditions), and proportion larger-later choices on the MCQ, where higher level of education was associated with greater likelihood of condom use and greater proportion larger-later monetary choices. Significant correlations between smoking status and condom use for three out of four partner conditions (both with and without a delay to condom use) and monetary choices indicate non-smokers tended to report higher likelihood of using immediately available condoms, and discounted delayed condom-protected sex and delayed monetary rewards significantly less than smokers.

Table 1.

Correlation coefficients for likelihood of condom use with no delay, sexual delay discounting, and monetary delay discounting in Study 1.

Variable	Likelihood of condom use with no delay				Sexual delay discounting (AUC)				MCQ
	Most want to have sex with	Least want to have sex with	Most likely to have an STI	Least likely to have an STI	Most want to have sex with	Least want to have sex with	Most likely to have an STI	Least likely to have an STI	Proportion Larger Later Choices
Sex1	−.188***	−.089*	−.022	−.196***	−.189***	−.279***	−.180***	−.163***	−.061
Age2	.063	.067	.059	.075*	.031	.069	.080*	.043	−.007
Income3	−.015	.008	.002	−.020	−.065	−.105**	−.049	−.036	.042
Education3	.160***	.025	.096**	.099**	.127**	.108**	.080*	.124**	.178***
Smoking status (past month)1	−.130***	−.057	−.082*	−.085*	−.181***	−.072	−.116**	−.103**	−.231***
Frequency of unprotected sex (with someone other than a spouse)3	−.158***	−.094**	−.081*	−.189***	−.126**	−.103**	−.090*	−.120**	−.019
Frequency of sex with someone who was drunk or high3	−.099**	−.059	−.029	−.090*	−.112**	−.059	−.074*	−.108**	−.049
Frequency of sex while drunk or high3	−.105**	−.053	−.058	−.103**	−.120**	−.066	−.076*	−.117**	−.047
Frequency of sex with someone not known very well3	−.231***	−.105**	−.091*	−.242***	−.160***	−.120**	−.100**	−.139***	−.065
Frequency of sex that was later regretted3	−.070	−.082*	−.026	−.117**	−.077	.003	−.058	−.053	−.068

Note:

¹Point-biserial correlation. Negative values for correlation with participant sex indicate male sex was associated with lower values of the outcome variables (i.e., lower likelihoods of condom use, lower AUC, and fewer larger later choices). Negative values for smoking status correlations indicate smoking was associated with lower values of the outcome variables.

²Pearson correlation

³Spearman’s rank-order correlation. AUC = Area-under-the-discounting curve. MCQ = Monetary Choice Questionnaire of delay discounting. For sexual risk frequency questions, there were four ordinal response options where 1 = Not at all, 2 = Just once, 3 = 2–5 times, 4 = 6 or more times.

^*p < .05

^**p < 0.01

^***p < .001

Table 2 displays standardised β values for linear regression models including each demographic variable and sexual risk question as predictors of likelihood of condom use with no delay, SDDT AUC, and proportion larger-later choices on the MCQ. The overall regression models were statistically significant for each outcome except the likelihood of condom use with no delay in the ‘most likely to have an STI’ partner condition. In each significant regression model, participant male sex was a significant predictor of lower likelihood of condom use with no delay, and lower SDDT AUC, indicating lower likelihoods of condom use among males both with and without delay. Participant male sex was also a significant predictor of decreased proportion larger-later monetary choices, but the magnitude of this effect (β = −.072, p = .045) was smaller relative to the condom use outcomes with significant regression equations (β range = −.099 to −.275, all ps < .01). Participant sex was the most robust demographic predictor of condom use outcomes, but older age, higher level of education, and non-smoking status were significant predictors of greater condom use likelihood across more than one condom use outcome. Higher education and non-smoking status also significantly predicted greater proportion larger-later monetary choices. The regression results for demographic variables were generally consistent with the individual correlations. However, whereas self-reported sexual risk frequency questions were consistently correlated with condom use outcomes in the individual correlations, when sexual risk frequency questions were included with other predictors in the regression models, there were fewer significant relationships with condom use outcomes as assessed by the SDDT. Self-reported frequency of unprotected sex and frequency of sex with someone not known very well significantly predicted lower likelihood of condom use with no delay in the more desirable partner conditions (i.e., ‘most want to have sex with’, ‘least likely to have an STI’) but not the other two partner conditions. When included with other predictors in the regression models, frequency of unprotected sex was the only self-reported sexual risk frequency question to predict greater SDDT AUC in any partner condition (i.e., ‘least want to have sex with’ β =−.105, p = .018).

Table 2.

Linear regression analyses (β) predicting likelihood of condom use with no delay, sexual delay discounting, and monetary delay discounting in Study 1.

	Likelihood of condom use with no delay				Sexual delay discounting (AUC)				MCQ
Variable	Most want to have sex with	Least want to have sex with	Most likely to have an STI	Least likely to have an STI	Most want to have sex with	Least want to have sex with	Most likely to have an STI	Least likely to have an STI	Proportion Larger Later Choices
Sex1	−.179***	−.099**	−.018	−.197***	−.199***	−.275***	−.190***	−.173***	−.072*
Age	.058	.096*	.036	.093*	.028	.065	.110**	.048	−.037
Income	−.003	.046	.014	−.015	−.066	−.076*	−.060	−.049	.024
Education	.134***	−.006	.069	.068	.112**	.085*	.024	.114**	.139***
Smoking Status (Past month)1	−.053	−.040	−.078	−.018	−.129**	−.028	−.090*	−.041	−.206***
Frequency of unprotected sex (with someone other than a spouse)	−.136**	−.083	−.057	−.157***	−.077	−.105*	−.082	−.081	.032
Frequency of sex with someone who was drunk or high	−.020	−.090	.091	.059	.002	.002	−.021	−.030	−.034
Frequency of sex while drunk or high	.076	.110	−.019	.017	.009	−.013	.003	−.014	.057
Frequency of sex with someone not known very well	−.213***	−.042	−.024	−.194***	−.079	−.066	−.022	−.074	.010
Frequency of sex that was later regretted	.044	−.027	−.002	−.015	−.032	.026	−.043	−.017	−.078
R2	.130	.032	.021	.123	.103	.118	.073	.078	.079
F	11.323***	2.534**	1.614	10.600***	7.219***	9.450***	5.889***	5.287***	6.508***

Note:

¹Negative values for coefficients for participant sex indicate male sex was associated with lower values of the outcome variables (i.e., lower likelihoods of condom use, lower AUC, and fewer larger later choices). Negative values for smoking status coefficients indicate smoking was associated with lower values of the outcome variables. AUC = Area-under-the-discounting curve. MCQ = Monetary Choice Questionnaire assessing monetary delay discounting. For sexual risk frequency questions, there were four ordinal response options where 1 = Not at all, 2 = Just once, 3 = 2–5 times, 4 = 6 or more times.

^*p < .05

^**p < .01

^***p < .001

One purpose of Study 1 was to examine whether condom use likelihood as assessed by the SDDT was more strongly related to sexual risk behaviours than monetary discounting. Proportion larger-later choices on the MCQ was not significantly correlated with any self-reported sexual risk behaviour, whereas likelihood of condom use with no delay and SDDT AUC were significantly correlated with self-reported sexual risk behaviour. The present data also support some shared predictors of monetary and sexual discounting. The significant correlation of monetary delay discounting with cigarette smoking is consistent with previous research (e.g., Bickel, Odum, & Madden, 1999), but the relationship between sexual delay discounting and cigarette smoking is a novel finding which underscores the potential for the same variables to correlate with both sexual and monetary discounting. Overall, these data suggest that monetary and sexual delay discounting have some common predictors, but that condom use likelihood as assessed by the SDDT may be more sensitive to sexual domain-specific risk behaviours.

Another purpose of the present analyses was to determine the extent to which likelihood of condom use as assessed during the SDDT was related to self-reported sexual risk behaviour outside of the experimental context. Individual correlations showed significant relationships between self-reported sexual risk behaviours and likelihood of condom use in hypothetical casual sex scenarios with and without delays, suggesting that responding on the SDDT is related to self-reported sexual risk behaviours. On the other hand, when demographic variables were included in the regression models, the relation between SDDT AUC and self-reported sexual risk behaviours was diminished, suggesting that variance accounted for by demographic factors in the model may overlap with the variance shared by sexual delay discounting and self-reported sexual risk behaviour. This may be due in part to the generality of the sexual risk behaviours assessed. For example, one of the items most consistently related to condom use likelihood on the SDDT was past frequency of unprotected sex, but we would only expect a subset of unprotected sex be due to lack of an immediately available condom (e.g., Fehr et al., 2015; Higgins & Smith, 2016; Sheeran et al., 1999). Thus, in Study 2, we directly assessed whether or not someone had ever had unprotected sex because there was not a condom readily available.

These data also provide further evidence that participant sex (male/female) is a consistent predictor of condom use likelihood in hypothetical casual sex scenarios with and without delays to condom use. This finding is therefore not limited to populations of college students or individuals with substance use disorders (Collado et al., 2017; Johnson & Bruner, 2013), but also applies in a non-clinical sample with varying levels of income and education. Though the present sample was relatively older than a previous study examining sexual delay discounting in college students (Collado et al., 2017; age in years M = 19.72, SD = 1.99), it should be noted that the present sample excluded adults 29 or older and was relatively homogeneous in terms of race and ethnicity. Future work that assesses sexual delay discounting across populations should be careful to control for participant sex. That said, although participant sex was consistently a predictor of condom use likelihood with and without delays across the different regression models, the R² values were relatively low (R² range between .02 and .13) suggesting other factors also contribute to condom use likelihood with and without delays aside from the demographic and self-reported sexual risk behaviours assessed in the present study.

Study 2

Results from Study 1 suggest that the SDDT is related to self-reported sexual risk behaviours outside of the laboratory. On the other hand, Study 1 also suggests the specific self-reported sexual risk questions assessed may have captured more general sexual risk regardless of condom use or availability. As such, one goal of Study 2 was to assess the relation between self-reported discounting-like sexual risk behaviour and condom use likelihood on the SDDT. Thus, we asked participants in Study 2 whether they had ever had unprotected sex because a condom was not immediately available. We hypothesized that condom use likelihood as captured by the SDDT was associated with this real-life sexual risk behaviour.

In addition, past research has shown that situational factors such as acute alcohol intoxication (Johnson et al., 2016) and acute cocaine administration (Johnson et al., 2017) increase arousal and interact with partner type to decrease condom use likelihood (Johnson et al., 2017; Johnson et al., 2016). Therefore, Study 2 administered a sexual arousal vignette to determine whether acute increases in sexual arousal were associated with changes in sexual discounting. We hypothesized that increased sexual arousal might result in greater sexual delay discounting (i.e., decreased likelihood of waiting for a condom).

Study 2 also examined the effect of administering additional, shorter delays to condom-protected sex. Although longer delays most often used in prior studies (i.e., 1 hour, 3 hours, 6 hours, 1 day, 1 week, 1 month, 3 months) are more similar to the range of delays in traditional monetary discounting studies, a shorter range of delays may have greater ecological validity for condom-use decisions (e.g., options to get a condom from elsewhere in the residence, or make a brief trip to a convenience store). Delays to condom-protected sex of 2 min, 5 min, 15 min, 30 min and 1 hour were examined in a recent study examining the effect of acute cocaine on sexual delay discounting (Johnson et al., 2017) and reported likelihood of condom use decreased in an orderly manner across short delays. This suggests that where clinically relevant, such as in the case of acute drug effects or acute arousal, short delays may capture useful variability in terms of likelihood of condom use. Then again, it is possible that exposure to shorter delays may affect responding at subsequent delays via an anchoring or framing effect. Thus, Study 2 compared sexual delay discounting across participants who received the more commonly used longer delays relative to those who were also exposed to shorter delays in order to determine whether exposure to shorter delays would systematically increase or decrease reported likelihood of condom use at longer delays.

Method

Participants

The Johns Hopkins University School of Medicine IRB approved all study procedures. As in Study 1, participants were recruited using Amazon MTurk. Participants provided informed consent before beginning the survey. The description of the study in the informed consent indicated that the survey included descriptions of hypothetical sexual scenarios, some of which were explicit. In consenting, participants affirmed that they understood the description of the study and were at least 18 years old. Participants were compensated $1.00 for completing the survey.

Survey

The study description explained the survey purpose was to learn more information about sexual decision making, and noted that participants may be asked questions about background, sexual preferences, and condom use in hypothetical sexual scenarios. The survey contained questions about demographic information, sexual risk, and the SDDT.

Sexual Risk

Participants were asked how many sexual partners they had in their lifetime and within the past six months, and whether they ever had sexual intercourse with a condom or without a condom. To assess discounting-like sexual risk behaviour, participants were asked (upper case letters also in original): ‘Have you ever had sexual intercourse WITHOUT a condom because you didn’t have one available? That is, you had sex WITHOUT a condom in a situation where you normally use a condom, but didn’t use one because you didn’t have one handy?’

Sexual Delay Discounting Task (SDDT)

As in Study 1, participants selected hypothetical sexual partners from an array of photographs. In order to select a condition in which maximal discounting was anticipated (Johnson et al., 2015; 2017) and to increase brevity, Study 2 required participants to complete the SDDT only for the partner they indicated they most wanted to have sex with. All participants used a VAS to rate their level of sexual arousal from 0–100, their likelihood of condom use if a condom with no delay, and their likelihood of condom use if a condom was delayed.

The SDDT varied between participants along two dimensions in Study 2: arousal manipulation and delay lengths examined. In order to test the effect of acute sexual arousal, before answering questions about arousal and condom use likelihood, approximately half of the sample read a descriptive sexual vignette in which the participant was introduced to a potential partner (i.e., ‘Dan’ or ‘Ellen’ depending on the sex of the selected photograph) by a mutual friend at a party, and eventually went back to his/her place where sexual activity progressed from kissing to touching genitals under clothes (George et al., 2009; Johnson et al., 2017). In scenarios involving female partners, either the participant or ‘Ellen’ (depending on sex of participant) was portrayed as taking oral contraceptives to specify she is ‘not going to get pregnant if you keep fooling around.’ Participants who did not see the vignette read the generic task introduction that asks the participant to imagine that they have just met this person, they are getting along great and are interested in having sex right now, and that there is no chance of pregnancy. All participants rated their likelihood of condom use at the originally published (‘standard’) delays of 1 hour, 3 hours, 6 hours, 1 day, 1 week, 1 month, and 3 months to use a condom. To test the effects of additional potentially meaningful delays, approximately half of the sample viewed delays of 2 min, 5 min, 15 min, and 30 min before viewing the standard delays. Thus, at survey start, participants were randomly assigned to one of four groups: (1) no vignette with standard delays (n = 68), (2) vignette with standard delays (n = 68), (3) no vignette with shorter + standard delays (n = 65), or (4) vignette with shorter + standard delays (n = 66).

Data Analysis

As in Study 1, data were evaluated for orderliness. Data were considered nonsystematic according to the criteria outlined in Study 1, but again were not excluded from analysis. As in Study 1, data from the SDDT were summarised using (1) the reported likelihood of condom use with no delay and (2) AUC where likelihood of condom use at each delay was expressed as a proportion of the participant’s likelihood of condom use with no delay, which requires the exclusion of individuals whose condom use is zero with no delay from AUC calculations so as to isolate the effect of delay on condom use likelihood.

To determine whether seeing the shorter delays significantly affected reported likelihood of condom use at the longer, standard delays, we calculated AUC based only on standard delays for all participants. We then compared AUC for standard delays across participants who saw both shorter and standard delays relative to those who saw only the standard delays using an independent samples t-test. We conducted linear regression analyses for outcome variables of sexual arousal rating, likelihood of condom use with no delay, and SDDT AUC calculated using the standard delays with predictors of: vignette (exposed yes or no), whether or not the participant endorsed having not used a condom because one was unavailable, sex, age, income, education, and tobacco use (past six months yes or no).

Results and Discussion

Participants

Three-hundred twenty-four were consented and began the survey. We excluded 23 respondents for failure to complete the survey, 15 for not selecting at least one sexual partner for the SDDT, one who indicated that we should not use their data, and one who indicated misunderstanding the task. Study 1 did not exclude females who selected female partners for the SDDT, but female-to-female sexual HIV transmission is very rare (CDC, 2014). In order to focus on populations with a greater likelihood of HIV transmission, Study 2 excluded 17 women who chose a female partner. This resulted in a final sample size of 267. Demographic information is presented in online supplementary materials (Supplementary Table 2).

Discounting and Sexual Risk Behaviours

Ninety-four percent of participants reported having at least one sexual partner in their lifetime. Median number of lifetime sexual partners was six; median partners in the past six months was one. Eighty-eight percent endorsed ever having sex with a condom; 85% endorsed ever having sex without a condom. Forty-six percent reported having had sex without a condom when they normally would have, but did not because a condom was not immediately available.

Ninety-three percent of the sexual delay discounting data were considered systematic according to the aforementioned criteria. Participants who viewed shorter delays before the standard delays did not have significantly different AUC calculated using standard delays than participants who saw only the standard delays (t(246) = 1.11, p = .27). These data suggest that using shorter delays in the SDDT will not significantly affect ratings at longer delays. Thus, researchers may assess likelihood of condom use at a wide range of shorter, potentially clinically relevant delays without concern for differentially affecting ratings at delays used in prior research studies. Further, we can confidently combine participants from both delay conditions using standard delay AUC for the purposes of examining the effects of the arousal vignette.

Table 3 displays standardised β values for linear regression models including predictors of (1) vignette (exposed yes or no), (2) whether or not the participant endorsed having not used a condom because one was unavailable, and (3) demographic variables for outcomes of sexual arousal, likelihood of condom use with no delay, and Sexual Delay Discounting (SDDT) AUC (AUC corrected for initial likelihood of condom use, standard delays only). The overall regression models were statistically significant for each outcome. The significant, positive β value (β = .262, p <.001) for vignette as a predictor of sexual arousal suggests that the arousal vignette resulted in higher ratings of sexual arousal before completing the SDDT. Although the vignette significantly increased sexual arousal, the vignette did not significantly decrease likelihood of condom use with no delay (β = .034, p =.575) or increase sexual delay discounting (β = .104, p =.094). Participant sex was a significant predictor of all outcomes such that male participants tended to rate higher levels of sexual arousal, lower likelihoods of condom use with no delay, and greater sexual delay discounting (i.e., lower AUC) relative to females.

Table 3.

Linear regression analyses (β) predicting sexual arousal, likelihood of condom use with no delay, and sexual delay discounting in Study 2.

Variable	Sexual arousal	Likelihood of condom use with no delay	Sexual delay discounting (AUC)
Vignette	.262***	.034	.104
Self-reported unprotected sex due to lack of condom	.069	−.083	−.236***
Sex1	.168**	−.160**	−.161**
Age	.097	.077	.072
Income	−.008	.108	.084
Education	.084	−.027	−.065
Tobacco Use (Past six months)1	−.019	−.039	−.051
R2	.120	.055	.109
F	5.034***	2.162*	4.204***

Note:

¹Negative values for coefficients for participant sex indicate male sex was associated with lower values of the outcome variables (i.e., lower likelihoods of condom use, lower AUC). Negative values for smoking status coefficients indicate smoking was associated with lower values of the outcome variables. AUC = Area-under-the-discounting curve.

^*p < .05

^**p < .01

^***p < .001

Participants who reported having had unprotected sex because a condom was not readily available (i.e., self-reported discounting-like sexual risk behaviour) did not show different ratings of sexual arousal, or lower likelihood of condom use with no delay, but did show significantly greater sexual delay discounting (i.e., lower AUC; β = −.236, p <.001). This self-reported discounting-like sexual risk behaviour was the strongest predictor of sexual delay discounting of any in the model, including participant sex (β = −.161, p =.009). Age, income, education, and tobacco use did not significantly predict any outcome within this sample.

We hypothesised that increasing sexual arousal through exposure to a sexual arousal vignette might decrease willingness to wait for a condom. Contrary to expectation, there was a non-significant trend for those who read the arousal vignette to show less sexual delay discounting (β = .104, p =.094). In other words, participants who read the vignette reported higher likelihoods of waiting to use a condom. One explanation for a trend in the opposite anticipated direction is that the vignette provided details that may suggest the hypothetical sexual partner is promiscuous. For example, the friend who introduces you warns, ‘(s)he’s probably not interested in a serious relationship but that (s)he’s a lot of fun.’ Thus, by providing many details about the casual sex scenario in the vignette, we may have inadvertently created a scenario in which the partner in the vignette is perceived as more likely to have an STI. As was demonstrated in Study 1, condom use likelihoods tend to be higher and sexual delay discounting tends to be less for partners perceived as most likely to have an STI. Therefore, future manipulations of sexual arousal should attempt to control for perceived STI risk across conditions. It may also be that hypothetical sexual outcomes are less sensitive to acute changes in arousal. However, recently published studies with alcohol and cocaine suggest that both sexual arousal and hypothetical sexual decision making as assessed by the SDDT were sensitive to acute pharmacological manipulation. Thus, the arousal manipulation in the present study was insufficient to produce changes in discounting, either due to relatively small magnitude of the change in arousal or through confounding arousal with STI risk. It may also be that the increase in sexual arousal observed in previous pharmacological manipulations with cocaine and alcohol was not the mechanism of increased sexual delay discounting under these circumstances.

We also hypothesised that those individuals who reported real-life behaviour consistent with a discounting of condom-protected sex framework, i.e., those who reported not using a condom because they did not have one immediately available, would show greater sexual delay discounting. The data confirm this, but also show that this risk behaviour did not affect likelihood of condom use with no delay. This supports the idea that likelihood of using immediately available condoms and the extent to which condom-use likelihood is affected by delay are relatively independent processes. In addition, nearly half of the sample endorsed engaging in this risk behaviour, suggesting a meaningful subset of a non-clinical sample are engaging in sexual risk because of a delay to condom-protected sex.

General Discussion

The overall aim of the present studies was to determine the relationship between demographic factors and self-reported sexual risk behaviours to condom use likelihood as assessed by the Sexual Delay Discounting Task (SDDT). Across two studies, the present data provided evidence that participant sex (male/female), but not other demographic factors, was a reliable predictor of condom use likelihood as assessed by the SDDT. Further, the present data support the SDDT as a useful tool to assess and describe the effects of delay on condom use likelihood, and that behaviour measured during the SDDT was related to self-reported sexual risk behaviours in a non-clinical sample. Therefore, further study of the environmental and individual differences that may contribute to sexual delay discounting is warranted and may benefit understanding of sexual risk behaviours that contribute to STI-related public health concerns.

The present results further validate the SDDT as an index of sexual risk behaviour in a non-clinical sample. Significant relationships between self-reported sexual risk behaviours (e.g., real-life instances of unprotected sex) and behaviour as assessed by the SDDT were observed in both Studies 1 and 2. The strongest relation between self-reported risk behaviour and the SDDT was with self-reported instances of discounting-like sexual risk behaviour. Specifically, those who self-reported having previously had sex without a condom because a condom was not immediately available (Study 2) also showed greater sexual delay discounting, whereas those individuals did not show significantly lower likelihood of condom use with no delay. Further, nearly half of the sample reported engaging in this risk behaviour. The implications of this finding are that a meaningful subset of a non-clinical sample engages in discounting of condom-protected sex, and that the SDDT is capturing individual differences in sexual risk behaviours that correspond to real-life risk behaviours that may negatively impact public sexual health.

In both studies, male sex was a predictor of lower likelihood of condom use with and without delay relative to females. Previous studies in individuals with cocaine use disorder and college students have also found greater discounting in males relative to females (Collado et al., 2017; Johnson & Bruner, 2013). The mechanism of this difference is unclear, particularly because the instructions attempted to reduce the concern of pregnancy for both sexes. Still, an instructional set may be insufficient to overcome perceived risk of pregnancy in females. Although speculative, it may be the case that differences in sexual discounting between males and females are driven by greater negative consequences of unprotected sexual activity for females (e.g., infertility, ectopic pregnancy) (Brady, 2003). Other demographic factors were less consistently related to sexual delay discounting. There was some evidence in Study 1 to suggest that education and smoking status may be related to sexual delay discounting, but the strength of this relation was less than what was observed for participant sex, and these were not significant predictors in Study 2. Study 2 had a smaller sample size and more inclusive tobacco use criterion (i.e., past six months vs. past month), which may explain this difference. The relation between other demographic predictors and sexual delay discounting may be explored further, but the present results suggest that future studies should control for participant sex for when looking for group differences or the effects of a specific manipulation on sexual delay discounting.

The outcomes of the present studies must be taken in the context of some shortcomings. Crowdsourcing technology utilised in the present study allowed for inexpensive data collection from the largest sample size in which sexual delay discounting has been examined thus far, and allowed for the sampling of individuals who were not a special population (e.g., substance users, college students), which contributes to the prior literature. Even so, the sample examined was somewhat homogeneous in terms of race and ethnicity, and MTurk workers may be a distinct population. The correspondence between data collected online and laboratory findings, however, is encouraging. Task features of the SDDT must also be considered with respect to generalizability. For example, participants completing the SDDT assign perceived STI risk (e.g., most likely to have an STI) to individuals whom they have already indicated as potential hypothetical partners in the SDDT. In real-world scenarios, judgments regarding perceived STI risk likely affect the initial selection of potential partners as well as the subsequent likelihood of condom use. Because the focus of the SDDT is on the role of delay to condom availability, the task does not provide insight as to the role of perceived risk on initial partner selection. Another consideration is that our statistical approach treated condom use as captured by the SDDT as an outcome variable rather than a predictor variable. We took this approach because in the present study, the real-life sexual risk behaviours and demographic statuses temporally preceded the participant’s completion of the SDDT. Ultimately, though, it is hypothesized that delay discounting is a fundamental decision-making process with individual differences that may precede or explain risk behaviours in a real-world setting. Future research should evaluate sexual delay discounting as an a priori predictor of sexual risk behaviour to evaluate this possibility. In spite of these limitations, the present series of studies support the use of the SDDT as a way to understand and ameliorate the spread of STIs through increased access to condoms or interventions designed to reduce discounting. Overall, these data suggest that delay discounting of condom-protected sex is an unrecognised but major contributor to potentially harmful sexual risk behaviour.