U.S. College Students’ Sexual Risk Behaviors Before and During the Early COVID-19 Pandemic

Abstract

The present study describes changes in young adults’ sexual behaviors during the early COVID-19 pandemic. Latent class growth analyses conducted with four waves of data collected between July 2019 to May 2020 in N = 775 college students (M_age = 18.61, SD = 0.33; 50.3% female, 90.2% White) revealed the presence of high and low risk classes in separate models for oral, vaginal, and anal sexual risk taking. As anticipated, vaginal and oral risk taking declined in Spring 2020. Membership in high-risk trajectories was attributable to high COVID-19-related financial problems, early sexual debut, low self-control, and being in a romantic relationship. Other COVID-19 factors and demographic control variables were not linked to trajectory membership. Thus, while many young adults’ sexual risk-taking changed during the early pandemic, their perceptions of and experiences with COVID-19 were not predictive of sexual risk trajectory membership.

For contemporary young adults, the emergence of the COVID-19 pandemic upset the unique developmental period in which it is typical for individuals to explore partnered sexual behaviors and relationships (Lindberg et al., 2020; Vasilenko et al., 2012). While there is growing evidence about the implications of the pandemic for college students’ mental health (Firkey et al., 2022; Lanza et al., 2022), its potential ramifications for sexual behaviors and thus distal implications for sexual health are comparatively poorly understood, which in turn obscures potential intervention needs. As detailed below, cross-sectional inquiries involving retrospective reporting conducted during the early pandemic suggest that it limited young adults’ involvement in partnered sexual behaviors, which in turn truncated opportunities to engage in risky practices. However, retrospective reports are prone to overestimation, meaning that reported changes could be exaggerated due to response biases (Gillmore et al., 2010) or to errors in recall, the likelihood of which increases significantly with the passage of time for common sexual behaviors (Graham et al., 2003). Recall errors are also more likely for individuals who experienced greater pandemic-related life disruptions (Hipp et al., 2020); as young adults were disproportionately impacted in this fashion relative to other age groups (Salari et al., 2020), their retrospective reports of their sexual behaviors during this period may be particularly suspect. Thus, prospective longitudinal investigations are needed in order to determine conclusively if young adults’ sexual behaviors changed during the early months of the pandemic, and if so, if these changes are associated with relevant covariates. Armed with such a design, we addressed these gaps in the literature, drawing upon four longitudinal waves of data collected as part of a study about risk-taking behaviors over the first year of college. From a person-centered approach, we explored longitudinal trajectories of composite indices of oral, vaginal, and anal sexual risk-taking behaviors between July 2019 and May 2020 in a sample of first-year traditional-age college students at a large, land grant university in Appalachia. We also examined the predictive value of several COVID-relevant covariates in conjunction with established predictors of sexual risk.

Longitudinal Changes in College Students’ Sexual Behaviors in 2019–2020

By emerging adulthood, most individuals have initiated partnered sexual behaviors, and will have varying degrees of lifetime experience with oral, vaginal, and/or anal intercourse (Copen et al., 2012). For many young people, this period is characterized by heightened sexual risk-taking relative to adolescence, including involvement with multiple sexual partners and in unprotected intercourse (Chandra et al., 2012; Halpern & Kaestle, 2014; Lefkowitz et al., 2019; Moilanen, 2015a; Vasilenko et al., 2018). Youth follow distinct trajectories of involvement in vaginal sexual risk-taking, corresponding to stable and volatile pathways at low, moderate, or high levels of risk, which in turn are linked to differential life course outcomes (e.g., Moilanen et al., 2010). Though such investigations illustrating sexual risk trajectories have primarily involved multi-year longitudinal designs with annual or biennial assessments and have emphasized coital risk-taking (i.e., vaginal intercourse with multiple sexual partners and/or unprotected penile-vaginal sex), they provided a basis for expectations about first year college students’ developmental trajectories of oral, vaginal, and anal risk in 2019–2020, encompassing the early phase of the pandemic.

How might COVID-19 change sexual risk-taking trajectories for this cohort? The primary anticipated deviation from these trends during the early pandemic concerns the presence of declines in risk-taking at the study’s latter waves in Spring 2020. This prediction is based on the findings of early, largely retrospective, cross-sectional, and descriptive studies of sexual behaviors conducted with convenience samples of adolescents and college students, which provide modest evidence of declines in partnered sexual involvement and risk-taking during the first year of the pandemic (Lanza et al., 2022). One April 2021 survey of undergraduate students at a university in Ohio revealed that respondents reported having fewer sexual partners during the first year of the pandemic relative to the year prior to the pandemic (Maziarz & Askew, 2022). In the one known prospective study involving a two-wave longitudinal design and a college student sample, Herbenick et al. (2022) determined that between January-February 2020 and April-May 2020 there were decreases in several partnered sexual behaviors, including oral, vaginal, and anal sex. During the same period (i.e., late May to early July 2020), a separate Mechanical Turk study of sexually experienced heterosexual college students revealed comparable findings: while 55% of the sample suggested that COVID-19 reduced their opportunities to have sex, approximately 52% of respondents indicated that there was no change in their number of sexual partners and 65% reported that there was no change in their use of condoms (Firkey et al., 2022). Thus, it appears that for some but not all young adults, the early pandemic period was characterized by reduced involvement in partnered sexual practices, which consequently truncated opportunities to engage in risky practices.

We expected to identify multiple risk trajectories for each risk behavior. We tentatively anticipated the presence of a stable low risk trajectory (i.e., comprised of individuals with limited partnered sexual experience throughout the first year of college) and an elevated declining risk pathway. We predicted that the classes would be relatively comparable across the three forms of sexual risk, and consistency in classes across behaviors (e.g., that young adults classified as low risk in one model would be classified similarly in the others; Moilanen, 2015b).

Predictors of Risk Trajectories

We also anticipated that there would be systematic differences between the profiles in terms of pandemic-related factors as well as in established (i.e., theoretically-relevant) covariates of sexual risk-taking. Although not specific to longitudinal changes in or latent classes of sexual risk-taking, the overarching theoretical grounding was Vasilenko et al.’s (2014) variable-centered developmental theoretical model for adolescent-era sexual behaviors, which stipulates that sexual behaviors are the direct product of individual, relational, and situational factors (see Figure 1). Individual factors include personality and attitudinal dimensions, and for the present study, we considered two traditional (i.e., early sexual debut and self-control) and five novel pandemic-relevant individual factors (i.e., personal experiences with COVID-19, concerns for the self and for close others, distress, and financial problems). Conceptually, individual covariates culminate in sexual risk-taking by shaping young adults’ self-selection into or out of relationships and/or situational contexts, which in turn facilitate involvement in or avoidance of partnered sexual risk-taking (Vasilenko et al., 2014; see also Cooper & Zhaoyang, 2014). We elaborate in the literature review below.

Figure 1. The Conceptual Model (Adapted From Vasilenko et al., 2014).

Note. Solid lines denote the discrete paths that were tested in the present investigation. Dashed paths are included in Vasilenko et al.’s (2014) original conceptual model but were not included in these analyses. Other model dimensions that are not relevant for the present investigation are omitted.

To our knowledge, no prior studies to date have considered both COVID-relevant and established covariates in longitudinal investigations of sexual behaviors in young adults during the pandemic era. Notwithstanding, it seems unlikely that only novel COVID-related dimensions would be the primary forces explaining young adults’ patterns of sexual behavioral involvement during the early months of the pandemic (i.e., that pandemic-related factors would be so vital that the effects of established covariates would diminish markedly). Thus, we considered both novel and established covariates in order to test this assumption. Such information has potential distal value for applied efforts, including for the identification of young adults who may particularly benefit from targeted interventions as well as for informing the content of preventative educational curricula.

COVID-19 Individual Factors

We explored five COVID-19 individual factors as potential covariates of sexual risk trajectories (personal experience with COVID-19, concern for self or others, distress, and financial problems). These selections were informed by the Health Belief Model (Hochbaum, 1958), which stipulates that various dimensions of individuals’ health-related beliefs and risk perceptions shape their decisions about health behaviors. Such an exploratory approach seemed appropriate given the paucity of data about how such factors are linked to variations in sexual risk-taking during the pandemic, and is bolstered by patterns in the general literature on COVID-19 prevention practices, indicating that some but not all COVID-related risk perceptions are predictive of COVID-protective actions (Bianchi et al., 2022; Wise et al., 2020). Our working hypothesis was that young adults who had personal experience with COVID-19 or viewed it as highly threatening would be particularly likely to enact protective practices such as social distancing in Spring 2020 (i.e., they would self-select out of contexts that would facilitate involvement in partnered sexual behaviors, which should preclude involvement in risky behaviors). Such a possibility was suggested in one longitudinal study of Belgian adults, which revealed that high perceived severity of COVID-19 was predictive of low numbers of social contacts (Wambua et al., 2022). Further, one national survey of US adults indicated that those who viewed social distancing as highly important reported less frequent casual sex between March and October 2020 (Gleason et al., 2021). We also posited that young adults who perceived the COVID-19 virus as less threatening may be more likely to forego COVID-19 protective actions, which would consequently remove a barrier to social contact with partners, thus enabling potential involvement in risky practices. In support of this, one early descriptive study of students at two campuses in Wisconsin revealed that students reporting low concerns about COVID-19 were more likely to attend parties and less likely to engage in social distancing relative to peers with high concerns (Rosenblum et al., 2022). In another investigation of college students in May 2020, young adults classified on health behavioral profiles characterized by sexual activity were less likely to report consistent social distancing relative to their peers in other profiles (Lanza et al., 2022). Finally, one Spring 2020 investigation of unpartnered adults in 23 nations revealed that low levels of perceived COVID-19 threat were linked to more frequent involvement in coitus and oral sex, as well as higher numbers of sexual partners (Rodrigues et al., 2023). Thus, per this logic concerning the self-selection processes included in Vasilenko et al. (2014), we predicted that personal experiences with COVID-19 and high perceived threat would correspond to trajectories marked by declining risk in Spring 2020.

Youth may acutely understand their personal risk for COVID-19 if they or close social ties (i.e., friends or family members) tested positive during Spring 2020. Such personal experience or close contact with COVID-19 has been consistently linked to heightened risk awareness in multiple studies (Cipolletta et al., 2022), and hypothetically, it should be followed by risk-averse decision-making in order to avoid viral transmission (Reinhardt et al., 2022). Young adults may also avoid partnered sexual behaviors if they feel highly concerned about themselves and their family and friends contracting COVID-19 (Bowling et al., 2022). Adolescents and young adults who are personally vulnerable to COVID-19 report higher perceived COVID-19 risk as well as greater adherence to protective behaviors, as do those who report strong desires to protect their family and friends from the virus (Yang et al., 2020). We examined the degree to which young adults reported experiencing distress symptoms (e.g., sleeplessness) and financial difficulties (i.e., pandemic-related problems with finances and accessing necessities such as food and supplies). There is inconsistent evidence that variations in distress predict COVID-19 protective behaviors such as social distancing, with some studies revealing that high levels of distress correspond to self-reported involvement in protective practices (Adewoyin et al., 2022; Bianchi et al., 2022; Chung et al., 2022), while others finding no association between such feelings and behaviors (Fullerton et al., 2022). Typically, high levels of perceived strain or feelings of distress directly reduce sexual interest and desire, which in turn should limit involvement in sexual behaviors, including risky practices (Ballester-Arnal et al., 2021). Notwithstanding, a recent investigation indicated that in the early months of the pandemic, high financial concerns were associated with higher self-reported sexual desire (Balzarini et al., 2022), suggesting the possibility that financial problems could be linked to maintenance of or escalation in risk-taking.

Traditional Covariates of Sexual Risk-Taking

As acknowledged above, it is unlikely that individual experiences and perceptions of COVID-19 are the only forces linked to young adults’ patterns of sexual behavioral involvement during the early months of the pandemic. On the assumption that young adults’ decisions about sexual risk are undoubtedly still shaped by the traditional types of covariates enshrined in developmental models, we considered two conventional individual-level predictors (early age at debut and self-control), and two relational-situational variables (young adults’ relationship patterns in Spring 2020 and whether they experienced a residential change in March 2020).

Consistent with precedents, we defined early sexual debut as reporting first experiencing oral, vaginal, or anal sex by age 15. Young adolescents who begin their sexual careers earlier than their peers tend to be highly disinhibited or lacking in self-regulation, which in turn may trigger other developmental processes that predispose them to maintain and/or escalate involvement in high-risk practices long after debut (Magnusson et al., 2019; Peragine et al., 2022). Early initiation and ongoing risk-taking create opportunities for these young adults to become very familiar with and accustomed to experiencing the rewards of partnered sexual encounters (Alley & Diamond, 2021; Wasserman et al., 2017). These benefits may be sufficiently motivating to make them willing to accept the potential risks and consequences of contracting COVID-19 during the process, which Osberg and Doxbeck (2021) offered as a speculative explanation for college student partying during the pandemic. Thus, in keeping with prior studies (e.g., Moilanen et al., 2010), we anticipated that young adults who reported early debut would be disproportionately represented in higher-risk classes, while peers who did not experience early debut would follow comparatively lower-risk pathways characterized by declines in risk in Spring 2020.

Self-control is a stable personality dimension that helps individuals act in ways that lead to avoiding risks and attaining goals (Moilanen, 2007). There is ample evidence that poor self-regulation is an antecedent of individual differences in sexual risk-taking in adolescence and emerging adulthood (Moilanen, 2015b; Moilanen et al., 2010; Raffaelli & Crockett, 2003). Consistent with prior evidence, we anticipated that young adults in high-risk classes would report lower levels of self-control relative to those in low-risk groups. Further, low self-control has been linked to low adherence to COVID-related health recommendations from the CDC (Rodriguez et al., 2023), including social distancing (Bieleke et al., 2023). This provides some support for the notion that self-control may dictate self-selection into or out of contexts of potential sexual risk-taking in Spring 2020.

Regarding relational and situational covariates, we explored how risk trajectories covaried with young adults’ romantic relationships and residential changes during the two Spring 2020 study waves. On-campus residential living grants many youth new autonomy to pursue partnered sexual experiences without concern about parental interference (Bailey et al., 2011). For many first year college students, this provides new opportunities to engage in partnered sexual activities, which typically occur with romantic partners (Fielder et al., 2013). Students who are in romantic relationships engage in coitus most frequently, are at elevated risk of involvement in unprotected sex (Bailey et al., 2011; Manlove et al., 2011), but have fewer annual partners than do uncommitted peers (Moilanen & Manuel, 2018). For many youth who moved home when campus closed in mid-March 2020, this sudden residential change may have eliminated this vital context for meeting potential partners or for engaging with established partners who do not reside locally, translating into notably reduced opportunities to engage in partnered sexual encounters and risk-taking (Lindberg et al., 2020). While approximately one third of surveyed California youth ages 13–17 reported not seeing their romantic partners in person during the first three months of the pandemic (Yarger et al., 2021), studies of university students indicated that early pandemic-related campus closures and social distancing requirements curtailed access to meeting new partners and engaging in hookups more than it limited interactions with established partners (Leistner et al., 2023; Maziarz & Askew, 2022). As such, young adults who were in romantic partnerships during the early pandemic were more likely to report experiencing sexual activity more frequently than those who were not partnered (Firkey et al., 2022; Herbenick et al., 2022). Herbenick et al. (2022) revealed stark reductions in involvement in oral, vaginal, and anal sex for college students who were unpartnered and for those who were partnered but geographically separated from their partners during the early pandemic, while changes were modest for partnered students who lived nearby their romantic partners. Thus, we predicted that first-year students who were not in romantic relationships at either Spring 2020 study wave and moved home with families would report lower risk during the latter part of the study, relative to those who were in relationships at those waves and did not move home.

Demographic Factors

We considered the direct effects of three demographic control variables, including gender, race/ethnicity, and sexual identity. Prior inquires have directly linked involvement in high-risk sexual practices during emerging adulthood to male gender, White race, and non-heterosexual identities (Ashenhurst et al., 2017; Kann et al., 2016; Moilanen, 2015b; Vasilenko et al., 2015). Thus, we included these controls in analyses.

The Present Study

In the present inquiry, we adopted a person-centered approach in order to examine potential individual differences in sexual risk behaviors throughout this period, seeking to link longitudinal classes to covariates which may have value for informing applied efforts with college students during and beyond the era of COVID-19. To our knowledge, variable-centered approaches have dominated pandemic-focused sexual behavioral research conducted to date, including in the studies discussed above, consistent with long-standing trends in the field in general. Such strategies involve statistical techniques that are optimal for building variable-focused conceptual models but are poorly-equipped to identify subgroups within the data. In turn, as this precludes drawing inferences about covariates of such classes, it ultimately complicates the targeted translation of findings into applied contexts. Person-centered approaches are advantageous for studying sexual behaviors, including changes in risk-taking, owing to the multidimensional nature of sexual behavioral development and the demonstrated complexity of associations between covariates and outcomes. Furthermore, while there are gaps in the literature concerning patterns of change in sexual risk-taking over time in general, these gaps are particularly acute concerning changes for young adults during the early months of the COVID-19 pandemic. Pre-pandemic person-centered investigations of developmental trajectories of sexual risk-taking indicate the presence of multiple risk pathways across durations of multiple years, commonly identifying distinct groups characterized by noteworthy variations in level of risk and temporal stability (Huang et al., 2012; Moilanen et al., 2010). These inquiries have also linked longitudinal trajectories of sexual risk to various covariates of group membership, facilitating the identification of potential intervention targets for specific types of individuals. It remains uncertain whether they are present over comparatively shorter durations in time (i.e., the present study’s nine-month duration versus eight or more years), particularly when the focal time period includes the early stage of the COVID-19 pandemic. Similarly, it is also unclear whether such patterns established for vaginal risk generalize to oral and anal sexual risk behaviors. Thus, the present study was well-positioned to provide novel information about how COVID-19 may be linked to patterns of involvement in three forms of sexual risk-taking during the portion of emerging adulthood in which sexual exploration is typical. The current study’s potential was further augmented by its large sample, its inclusion of known predictors as well as novel COVID-19 relevant potential covariates, and its four-wave prospective longitudinal design that involved two waves in Spring 2020 (i.e., the third wave reflected sexual risk-taking over the month prior to the official declaration of COVID-19 as a national emergency in mid-March, and the fourth assessed prior month risk-taking in early May).

Our first goal was to identify longitudinal trajectories of composite indices of oral, vaginal, and anal sexual risk-taking behaviors, which were assessed at four waves between July 2019 and May 2020. As described above, we anticipated the presence of multiple risk trajectories for each behavior, including a stable low risk trajectory and an elevated risk declining pathway. We also predicted that the oral, vaginal, and anal risk classes present would be relatively comparable across the oral, vaginal, and anal risk models, and that individuals would be consistent in terms of group membership across these three models. Our second goal was to examine the predictive value of several COVID-relevant covariates in conjunction with additional known predictors of sexual risk in emerging adulthood. As detailed above, we posited that there would be systematic differences between the profiles on these covariates, such that membership in low-risk or declining risk classes would be linked to individual and relational/situational factors that may facilitate purposeful avoidance of contexts of sexual possibility.

Method

Participants and Procedures

This study involved analysis of the four-wave longitudinal College Student Transition Study. The sample consisted of matriculating first year college students enrolled at a large university in the southeastern United States. The West Virginia University Office of Enrollment Management provided a list of the email addresses of all incoming first year students (N = 4,329). This study’s broad focus was illicit alcohol and substance use, which necessitated the exclusion of students if they were younger than 18 years at the baseline assessment (in order to eliminate the need for parental consent) or if they would be older than 21 years as of the planned final assessment (n = 457). Transfer students were also excluded (n = 17), resulting in a potential pool of N = 3,855 eligible young adults, for which available funding would permit a sample of n = 800. Researchers then used stratified random sampling strategies to ensure that the final sample varied in terms of pre-matriculation academic risk. Of those designated at high risk, n = 331 consented of approximately 1,300 invitations (25.5% response rate) and for those at low risk, n = 443 consented from approximately 1,100 invitations (40.3% response rate). Ultimately 775 participants ages 18–20 years (M = 18.61, SD = 0.33; 50.0% female, 49.4% male, 0.1% transgender female, 0.5% gender fluid or non-binary) completed consent procedures and the first study assessment. The majority of the participants were White (89.9%); the remainder of the sample identified as multi-racial (4.4%), Black or African American (2.2%), Asian or Pacific Islander (1.9%), Native American/Alaskan Native (0.5%), Unknown (0.5%), Middle Eastern (0.4%), and Indian (0.1%). Of the full sample, the majority identified as non-Hispanic (96.6%), heterosexual (88.0%; 5.7% identified as bisexual, 3.8% as homosexual/gay/lesbian, 1.4% as pansexual, and 1.0% as asexual) and 15.1% were first-generation college students. Most participants were from rural settings (56.8%) and nearly all (92.9%) lived in campus housing at matriculation.

The study protocols were designed in accordance with ethical standards and were approved by the West Virginia University Institutional Review Board. Following recruitment, all respondents affirmed their informed consent prior to participation via online surveys in Qualtrics. Data collection occurred in 2019–2020, with waves occurring approximately two to four weeks prior to matriculation (i.e., Wave 1; July-August 2019), in the middle of the fall semester (Wave 2; October 24 – November 8, 2019), near the middle of the spring semester (Wave 3; March 9 – April 6, 2020), and at the end of the spring semester (Wave 4; May 3–15, 2020). The pandemic disrupted the latter two study waves¹. Respondents received financial incentives for participation in each wave, ranging from $20 (for completing the baseline assessment only) to $50 (for completing all four waves; n = 385; 49.7% of the Wave 1 sample). Attrition analyses indicated that after Wave 1, 13.3% of the respondents missed one subsequent wave, 15.1% missed two subsequent waves, and 22% missed three subsequent waves. In keeping with conventional approaches for exploring systematic attrition (McCartney et al., 2006), we created a binary variable in order to compare those who were present for all four waves to those who were missing data at one or more timepoints between Waves 2 and 4 via t-tests and χ² analyses for all Wave 1 study variables. These revealed that attrition after Wave 1 was linked to male gender and identifying as asexual (versus heterosexual), and that those with missing follow-up data reported modestly higher levels of vaginal risk-taking at Wave 1 relative to participants who were present at all four waves. A total of n = 17 respondents withdrew from the university during the course of the study. Bias analyses revealed no differences on any study variables between this subgroup and their peers who were consistently enrolled, and thus their datapoints were retained for analyses.

Measures

Oral, Vaginal and Anal Sexual Risk Composites (Waves 1 – 4)

At all study waves, participants responded to parallel sets of questions concerning their experiences with oral, vaginal, and anal sex during the past 30 days. Respondents first indicated whether they had any involvement in each behavior (0 = no, 1 = yes). Subsequently, for each behavior, they identified the number of partners with whom they had engaged in this behavior in the past month (i.e., open-ended, resulting in a range of 0 to 20 partners across the four waves; a subset of participants who indicated no past month involvement skipped this item, and their missing values were set to zero as part of variable construction procedures). Finally, respondents reported how often they used condoms/barriers during each behavior in the last 30 days, using a response scale from 0 (I never use condoms or barriers) to 4 (I always used condoms or barriers), with an additional response option available for participants to denote if they had not engaged in the behavior during the last 30 days. Responses to these latter two items were dichotomized prior to forming risk composites for each behavior at each wave, consistent with other investigations (e.g., Moilanen, 2015b; Raffaelli & Crockett, 2003). The number of partners variable was recoded to indicate 0 = 0–1 partners in the last month and 1 = two or more partners in the last month, and the condom use variable was recoded such that 0 = always used condoms/barriers at all encounters or reported no encounters in the past 30 days and 1 = reported any encounters during the past month and used condoms/barriers inconsistently or not at all in the past 30 days. Possible aggregate scores included 0 (no involvement in the last month, or had one partner and always used condoms/barriers), 1 (reported having either two or more partners or any unprotected encounters in the past month), and 2 (reported having both two or more partners and any unprotected encounters in the past month). Preliminary analyses indicated that it was necessary to revise the anal risk composite to a binary variable, reflecting 0 (no involvement in the last month, or had one partner and always used condoms), and 1 (reported having two or more partners and/or any unprotected encounters in the past month). Detailed descriptive statistics of each composite are provided in Supplemental Table S1.

COVID-19 Individual Variables (Wave 4)

Personal Experience.

Participants responded to four items indicating whether they had ever been tested for COVID-19 (0 = no, 1 = yes), and three items assessing whether they themselves, any family members, or any close friends had tested positive for COVID-19 (all 0 = no, 1 = yes/unsure). These responses were aggregated by averaging. As the resulting composite demonstrated unacceptable kurtosis (i.e., 13.0% endorsed one item, and 3.6% endorsed two or more items), the index was recoded to a binary variable (0 = no personal experience, 1 = any personal experience).

COVID-19 Concern for Self and Family/Friends.

Two separate items assessed COVID-19 concerns for the self (i.e., “How concerned have you been about the following issues since the COVID-19 pandemic started? …Concern about you contracting COVID-19”) and for family and friends (i.e., “…Concern about family/friends contracting COVID-19”). The same response scale was used for both items, ranging from 1 (not at all concerned) to 5 (extremely concerned).

COVID-19 Distress.

Participants responded to three items assessing frequency of experiencing COVID-related distress since the start of the pandemic, using a five-point response scale ranging from 1 (never) to 5 (always). These items referred to “having trouble sleeping at night,” “feeling restless,” and “not being able to focus on school/work.” Responses to these items were averaged (α = .81).

Financial Problems due to COVID-19.

Students also responded to a 14-item questionnaire that was designed by the final author specifically for this investigation, concerning the financial, academic, social, and exercise-related problems that they experienced in the Spring 2020 semester due to the pandemic. The response scale ranged from 1 (not at all a problem) to 4 (serious problem). Exploratory factor analyses revealed that these items loaded onto four factors: a six-item financial difficulties factor (e.g., “getting groceries/supplies/food”) that explained 30.09% of the total variance, a four-item academic difficulties factor (e.g., “completing coursework virtually”) that explained 10.10% of the total variance, a two-item exercise difficulties factor (e.g., “accessing a gym to exercise/work out”) that explained 8.29% of the total variance, and a two-item social difficulties factor (e.g., “socializing with friends”) that explained 7.56% of the total variance. We retained the first financial difficulties factor, and calculated a score by averaging (Cronbach’s α = .72).

Individual Variables

Early Debut Status.

At Wave 1, respondents reported their age at debut of involvement in oral, vaginal, and anal sex. We recoded these responses into a binary variable reflecting initiating any of these sexual behaviors by age 15 years (0 = not early, i.e., at age 16 or later, 1 = early debut, i.e., prior to or at age 15).

Self-Control.

At Wave 1, participants completed the 13-item Brief Self-Control Scale (Tangney et al., 2004; sample item: “I often act without thinking through all of the alternatives”), using a five-point response scale ranging from 1 (not at all like me) to 5 (very much like me). Items were reverse-coded as needed prior to averaging, so that high scores corresponded to better self-control (α = .85).

Relational-Situational Variables

Relationship Status.

At each wave, respondents reported their current romantic relationship status (0 = not in a relationship, 1 = in a relationship). Given the focus of the analyses, we used these reports to create a categorical variable indicating changes in relationship status between Waves 3 and 4, such that 0 = no relationship at either wave (57.9%), 1 = in a relationship at Wave 3 only (5.5%), 2 = in a relationship at Wave 4 only (2.2%), and 3 = in a relationship at both Waves 3 and 4 (34.4%). In analyses, dummy codes were used to reflect these patterns, with the first group set to the comparison group.

Moved Home at Campus Transition.

Participants responded to an item assessing residential changes due to the mid-March campus closure. This was recoded into a dummy variable indicating 0 (remained on/near campus) and 1 (moved back in with parents or family).

Demographic Control Variables

At the baseline assessment, participants reported on demographic characteristics, including gender (0 = male, 1 = female), and race (0 = White, 1 = not White). Participants also responded to a categorical item about their sexual identity, indicating whether they identified as heterosexual/straight, homosexual/gay/lesbian, bisexual, pansexual, or asexual. This was represented with a dummy code, with heterosexual identification as the referent group.

Analysis Plan

Preliminary analyses included descriptive statistics (see Table 1) and bivariate correlations (see Supplemental Table S2). The first study goal (i.e., to identify longitudinal trajectories of composite indices of oral, vaginal, and anal sexual risk-taking behaviors) was addressed with latent class growth analysis (LCGA) in Mplus (Muthén & Muthén, 2017). As described above, we anticipated the presence of multiple risk trajectories for each behavior, including a stable low risk trajectory and an elevated risk declining pathway. As a special type of growth mixture modeling, LCGA models examine separate growth trajectories for different unobserved subpopulations while assuming all individuals’ growth parameters to be the same within one subpopulation (Jung & Wickrama, 2008). With this assumption, it is easier to clearly identify latent classes with different growth trajectories, and it also reduces computational burden (Jung & Wickrama, 2008). Separate LCGA models were estimated for each risk composite. Specifically, we estimated LCGAs for a three-category outcome for oral and vaginal sexual risk composites, and LCGAs for a binary outcome for the anal sexual risk composite. We started with a linear one-class LCGA model without covariates for each risk composite and then increased the number K of latent classes. We stopped adding classes when there were convergence issues or when adding more classes no longer improved model fit. We also estimated a series of unconditional quadratic LCGA models for each risk composite in the same way. We then compared model fit for all estimated models. Absolute model fit was evaluated in terms of χ², and relative fit through AIC, BIC, ABIC, entropy, and VLMR-LRT and BLRT ps. Models with smaller AIC, BIC, and ABIC values are preferable to those with larger values. Non-significant VLMR-LRT and BLRT ps indicate support of the retention of a model with K-1 number of classes. When different fit indices indicated competing optimal solutions, we based our final decisions mainly on BIC and BLRT as previous simulations demonstrated that the BIC performed the best among the information criteria-based indices and that BLRT was the best tool for deciding the number of classes among all indices and tests (Nylund et al., 2007). We also took into consideration parsimony, theories, and interpretability in addition to the statistical tools when we determined the optimal models. We predicted that the oral, vaginal, and anal risk classes present would be relatively comparable across the oral, vaginal, and anal risk models, and that individuals would be consistent in terms of group membership across these three models. This hypothesis was tested through χ² tests.

Table 1.

Descriptive Statistics

Variable	N	M (SD) / %	Range
Sexual Risk Composites
Vaginal Risk Composite Wave 1	770	.25 (.50)	0 – 2
Vaginal Risk Composite Wave 2	539	.32 (.57)	0 – 2
Vaginal Risk Composite Wave 3	473	.36 (.61)	0 – 2
Vaginal Risk Composite Wave 4	468	.20 (.47)	0 – 2
Oral Risk Composite Wave 1	769	.35 (.54)	0 – 2
Oral Risk Composite Wave 2	540	.49 (.65)	0 – 2
Oral Risk Composite Wave 3	474	.46 (.64)	0 – 2
Oral Risk Composite Wave 4	468	.24 (.51)	0 – 2
Anal Risk Composite Wave 1	768	.02 (.14)	0 – 1
Anal Risk Composite Wave 2	540	.03 (.17)	0 – 1
Anal Risk Composite Wave 3	472	.04 (.18)	0 – 1
Anal Risk Composite Wave 4	468	.02 (.14)	0 – 1
COVID-19 Individual Factors
Personal Experience	471	16.6%	0 – 1
COVID-19 Concern for Self	470	2.16 (1.11)	1 – 5
COVID-19 Concern for Family and Friends	469	2.98 (1.27)	1 – 5
COVID-19 Distress	462	3.07 (1.09)	1 – 5
Financial Problems due to COVID-19	471	1.65 (.56)	1 – 4
Traditional Individual Factors
Early Sexual Debut	737	20.2%	0 – 1
Self-Control	774	3.35 (.72)	1 – 5
Relational-Situational Factors
In Relationship at Wave 3 Only	416	5.5%	0 – 1
In Relationship at Wave 4 Only	416	2.2%	0 – 1
In Relationship at Waves 3 and 4	416	34.4%	0 – 1
Moved Home at Campus Transition	470	96.2%	0 – 1
Control Variables
Male Gender	769	49.7%	0 – 1
Non-White Race	769	9.8%	0 – 1
Identifies as Heterosexual	768	88.0%	0 – 1

The second study goal (i.e., to examine the predictive value of several COVID-relevant covariates in conjunction with additional known predictors of sexual risk in emerging adulthood) was addressed in two phases. First, we explored class mean and distributional differences in each individual covariate via ANOVA and χ² analyses. Subsequently, in a set of analyses separate from the LCGA step described above, we sought to predict class membership via logistic regressions in Mplus with all covariates included simultaneously. As with the LCGAs, we estimated separate logistic regression models for each form of risk. Continuous predictors were mean-centered for analyses. In general, as detailed above, we anticipated that there would be systematic differences between the profiles on these covariates, such that membership in low-risk or declining risk classes would be linked to individual and relational/situational factors that indicate purposeful avoidance of contexts of sexual possibility.

Study data were not missing completely at random (MCAR), Little’s MCAR χ²(250) = 343.81, p < .001. Follow-up analyses indicated that missingness at Waves 2–4 was linked to the same variables identified above regarding attrition. The effect sizes were generally small (i.e., η² < .015 and Cramér’s V < .11), with the exception of gender, which had an effect size between small and moderate (i.e., Cramér’s V range = .18 - .23). In the LCGAs, missing data were addressed through full information maximum likelihood (FIML) estimation. When data are MCAR or are missing at random (MAR), FIML produces the least biased results compared to other missing data methods (Enders & Bandalos, 2001). In the logistic regressions predicting class membership, Monte Carlo integration was employed in order to estimate missing data. No absolute fit statistics are provided for logistic models under these constraints (Muthen, 2012).

Results

Latent Class Growth Analysis Models

In pursuit of the first study goal, we estimated three linear LCGA models and three quadratic LCGA models for each risk composite. All quadratic models showed a better model fit than linear models with the same number of classes. Model fit statistics supported the retention of quadratic two-class models for all three risk behaviors (see Table 2).

Table 2.

LCGA Fit Statistics

Number of Classes	(df) X2	AIC	BIC	ABIC	LMRT p	BLRT p	E K
Vaginal Risk
One - Linear	(74) 844.36***	2965.90	2979.85	2970.32
Two - Linear	(74) 121.79***	2641.55	2669.45	2650.39	<.001	<.001	.67
Three - Linear	(71) 115.20***	2639.54	2681.39	2652.81	.06	.10	.75
One - Quadratic	(73) 785.40***	2948.75	2967.35	2954.65
Two - Quadratic	(72) 72.75	2611.99	2649.19	2623.79	<.001	<.001	.67
Three - Quadratic	(68) 61.61	2608.68	2664.48	2626.38	.06	.04	.76
Oral Risk
One - Linear	(76) 738.65***	3535.35	3549.30	3539.77
Two - Linear	(74) 194.26***	3251.06	3278.96	3259.91	<.001	<.001	.57
Three - Linear	(71) 176.72***	3247.82	3289.67	3261.09	.50	.10	.71
One - Quadratic	(75) 659.85***	3490.04	3508.64	3495.94
Two - Quadratic	(72) 116.48 ***	3177.49	3214.69	3189.29	<.001	<.001	.58
Three - Quadratic	(68) 97.24***	3174.94	3230.75	3192.64	.17	.03	.55
Anal Risk
One - Linear	(12) 57.85	535.36	544.66	538.31
Two - Linear	(10) 14.33	513.97	537.23	521.35	<.001	<.001	.94
Three - Linear	(7) 12.18	517.03	554.24	528.83	.24	.14	.95
One - Quadratic	(11) 55.45	533.57	547.52	537.99
Two - Quadratic	(8) 10.73	511.96	544.51	522.28	.01	<.001	.92
Three - Quadratic	(4) 2.52	513.55	564.71	529.78	.02	.12	.86

Note. Selected solutions are denoted in bold text.

^*p < .05

^**p < .01

^***p < .001.

For vaginal risk, the two-class quadratic model has the smallest BIC value among all models. The BLRT suggested that the three-class quadratic model provided a modestly better fit to the data than the two-class quadratic model (p = .04). However, the smallest class of the three-class model represented only approximately 2% of the sample with a 46% classification probability for this class, indicating a low certainty of classification. Further, its BIC and ABIC were larger than those for the two-class model, which we ultimately retained. The two groups consisted of one high-risk class (28%) and one low-risk class (72%). These results showed that the high risk class had a higher propensity to engage in risky vaginal behaviors than the low-risk class at baseline (i.e., the intercept of the low risk class was automatically fixed to zero by Mplus). The high risk class showed a decelerating increase and then a decrease in their propensity to engage in risky vaginal sexual behavior, while the low risk class showed decreasing propensity across waves. As indicated in Table S1, the majority of individuals reporting one risk reported involvement in unprotected vaginal sex. A plot of class propensities is provided in Figure 2.

Figure 2. Vaginal Risk Class Propensities.

Note. The low risk class fixed effects were intercept = 0, linear = 0.93, p = .07, quadratic = −0.33, p = .04. The high risk class fixed effects were intercept = 3.61, linear = 1.10, quadratic = −0.40, ps < .001.

For oral risk, the BLRT and AIC suggested that the three-class quadratic model provided a modestly better fit to the data than the two-class quadratic model. However, its BIC and ABIC were larger than those for the two-class model. In the three-class model, the classification probability of the smallest class (representing 14% of the sample) for this class membership was relatively low (50%), whereas classification probabilities in the two-class solution were both relatively high (92% and 82%). Therefore, we selected the two-class model, which was also the more parsimonious solution, as the optimal model. The two groups consisted of one high risk declining class (41%) and one low risk class (59%). These results showed that the high risk class had a higher propensity to engage in risky oral behaviors than the low risk class at baseline. Both classes showed an increase followed by a decrease in their propensity to engage in risky oral sexual behavior (see Figure 3).

Figure 3. Oral Risk Class Propensities.

Note. The low risk class fixed effects were intercept = −2.69, linear = 0.60, p = .07, quadratic = −0.31, ps < .001. The high risk class fixed effects were intercept = 0, linear = 1.56, quadratic = −0.58, ps < .001.

For anal risk, the BLRT, AIC, BIC, and ABIC suggested that the two-class quadratic model provided a modestly better fit to the data than the one-class quadratic model. The two groups consisted of one high risk class (2%) and one low risk declining class (98%). These results showed that the high risk class had a higher propensity to engage in risky anal sexual behaviors than the low risk class at baseline. The low risk class showed decreasing propensity to engage in risky anal sex, while this propensity for the high risk class was stable over time (see Figure 4).

Figure 4. Anal Risk Class Propensities.

Note. The low risk class fixed effects were intercept = −2.79, p = .04, linear = 1.06, p = .07, quadratic = −0.48, p = .02. The high risk class fixed effects were intercept = 0, linear = 1.39, p = .23, quadratic = −0.20, p = .66.

Consistency of Classes Across the Risk Behaviors

A summary of classifications across all three LCGA models is provided in Table 3. As anticipated, there was considerable consistency across the models, with 81.5% of respondents assigned to the same class in all three models. Notably, relatively few cases were classified into the high anal risk group. All of those following that trajectory were also members of one or more other high risk class, with more than half also being members of the high risk classes in both the vaginal and oral risk models. However, there were no distributional differences between class assignments within the high anal risk class, χ²(1) = 0.49, p > .05, Cramer’s V = .22. Amongst the majority of respondents who were in the low anal risk class, there were significant distributional differences in assignments to oral and vaginal risk classes, χ²(1) = 279.05, p < .001, Cramer’s V = .61. A substantial proportion reported both low oral and vaginal risk, while comparatively fewer were in the high risk class for both models, followed by a smaller proportion that reported high oral risk and low vaginal risk, and an even smaller proportion demonstrating low oral risk and high vaginal risk. In sum, individuals tended to demonstrate consistent risk involvement across behaviors.

Table 3.

Summary of Conjoint Risk Classifications (n = 773)

Anal Risk Class	Oral Risk Class	Low Vaginal Risk (n = 555)	High Vaginal Risk (n = 218)
Low Anal Risk (n = 763)	Low Oral Risk (n = 477)	57.6% (n = 445)	4.1% (n = 32)
	High Oral Risk (n = 286)	13.8% (n = 107)	23.2% (n = 179)
High Anal Risk (n = 10)	Low Oral Risk (n = 1)	0.0% (n = 0)	0.1% (n = 1)
	High Oral Risk (n = 9)	0.4% (n = 3)	0.8% (n = 6)

Covariates of Risk Class Membership

The logistic regressions predicting class membership are summarized in Table 4, and the results of class comparisons for each individual covariate are available in Supplemental Table S3. The general patterns across the two sets of analyses were generally consistent, and thus below we focus on the results of the comparatively rigorous logistic regressions.

Table 4.

Summary of Logistic Regressions Predicting Sexual Risk Classes

	Vaginal Model		Oral Model		Anal Model
Covariate	OR	p	OR	p	OR	p
Personal Experience	1.32	.482	.97	.913	.66	.566
Concern for Self	1.07	.693	1.01	.970	.62	.412
Concern for Family/Friends	.79	.072	.99	.907	.68	.321
COVID-19 Distress	.87	.288	.84	.114	.99	.973
Financial Problems	2.45	.010	1.72	.061	1.99	.521
Early Sexual Debut	5.69	.001	3.64	.001	1.90	.523
Self-Control	.67	.000	.76	.009	1.35	.563
In Relationship at Wave 3 Only	2.21	.275	2.72	.119	.05	.000
In Relationship at Wave 4 Only	3.47	.325	2.81	.385	.002	.000
In Relationship at Waves 3 and 4	5.41	.005	4.33	.002	1.11	.913
Moved Home	.64	.331	.34	.001	.15	.000
Male Gender	.77	.171	.84	.317	1.21	.892
Non-White Race	.78	.441	.74	.298	1.01	.992
Identifies as Heterosexual	2.02	.133	1.37	.333	.12	.000
AIC		10292.97		10429.88		9614.69
ABIC		10490.94		10627.85		9812.66

In the vaginal model, elevated odds of membership in the high risk class were attributable to high financial problems as a result of COVID-19, early sexual debut, low self-control, and being in a relationship at both Waves 3 and 4 (relative to being in a relationship at neither wave). The results of the oral risk logistic regression model were largely consistent with those for the vaginal risk model. High odds of membership in the high risk class was linked to early sexual debut, low self-control, being in a relationship at both Waves 3 and 4 (relative to being in a relationship at neither wave), and not moving home when campus closed.

Finally, we report the results of the logistic regression for the anal risk model, while urging considerable caution in its interpretation given the unbalanced group sizes resulting from the rare reporting of these risk behaviors and the inconsistency in associations in this regression model relative to those in the zero-order correlations (see Table S2). Low odds of membership in the high risk class were attributable to being in a relationship at either Wave 3 or Wave 4, with moving home when campus closed, and with identifying as heterosexual.

Sensitivity Analyses

In order to ascertain whether any group differences may be attributable to involvement in partnered sexual activity at Wave 4, we estimated three additional logistic regressions using the same procedures as described above. For each behavior, we modeled prior involvement with a dummy code reflecting any endorsement at the three prior study waves versus no endorsement. We also modeled only Wave 4 relationship status rather than relational patterns over the last two waves. These variations notwithstanding, some of the associations were common to the models predicting class membership (see Supplemental Table S4). For vaginal and oral sexual behavior, any prior involvement in that behavior at any previous wave was linked to higher odds of past month involvement at Wave 4, as was being in a romantic relationship at Wave 4. In the same models, moving home and non-White race was linked to reduced odds of last month sexual involvement at Wave 4. For anal sex, as before, we report the results of this model, while urging considerable caution in its interpretation given the unbalanced group sizes resulting from the rare reporting of involvement in anal sex. Notwithstanding, personal experience with COVID-19, high concern for family or friends, and heterosexual identity was linked to low odds of any last month anal sex at Wave 4.

Discussion

In Spring 2020, the COVID-19 pandemic had significant potential to disrupt young adults’ developmentally-typical sexual involvement and exploration. The immediate and distal consequences of this interruption are of concern to scholars, particularly given the ambiguity concerning the duration and timing of COVID-related barriers to sexual interactions (Lindberg et al., 2020). As part of disentangling this storyline, we sought to document changes in the sexual behaviors of young adults who were in their first year of college when COVID-19 led to the suspension of on-campus operations. We examined these longitudinal changes in the context of separate LCGA models for vaginal, oral, and anal sexual risk-taking composites. These analyses illustrated shifts in young adults’ sexual practices during Spring 2020, represented in terms of reductions in sexual risk-taking in May 2020 relative to prior study waves. We explored several pandemic-related individual factors as covariates of class membership, and examined associations with additional traditional covariates of sexual risk-taking. While caution is generally warranted for all models involving anal sex, owing to low base rates in these behaviors throughout the study period and regression model associations that did not mirror those in the bivariate correlations, in general these analyses revealed that the COVID-related variables had minimal associations with risk classes. In particular, likelihood of membership in the high vaginal risk class was linked only to reported financial problems due to COVID-19 (i.e., an effect that was not present in the oral and anal risk models), while moving home was associated with low likelihood of membership in the high risk oral and anal classes. Instead, across all models, membership in high-risk classes was comparatively strongly linked to traditional covariates, such as early sexual debut (i.e., vaginal and oral risk models) and sexual identity (i.e., anal risk model). Thus, the findings of this prospective longitudinal study confirm the conclusions of prior cross-sectional inquiries suggesting declines in some young adults’ sexual risk-taking during the early pandemic. It also extends those findings by illustrating that individual changes were largely a function of pre-pandemic individual factors and Spring 2020 relationship experiences, while COVID-related individual factors had limited value in explaining class differences. It is worth noting, however, that the unbalanced sample with relatively few participants who were non-White, non-heterosexual, and did not move home in Spring 2020 might have resulted in some bias in the findings. Therefore, replication is needed, particularly in samples with comparatively equitable group sizes for these predictor variables.

Sexual Risk Changes

The first study goal was to describe trajectories of young adults’ vaginal, oral, and anal sexual risk-taking across the first year of college. In general, we anticipated that multiple trajectories would be present in each behavior-specific model, and that these groups would vary in terms of level of risk across the study period. We also predicted that any classes demonstrating elevated risk would demonstrate declines in risk-taking during Spring 2020. These hypotheses were generally supported. For each behavior-specific model, fit indices supported the retention of two classes: a high risk class that evidenced a peak in risk-taking prior to the emergence of the pandemic in the United States in early spring, followed by a subsequent decline in risk as of May, and a low risk class that had relatively flat levels of involvement throughout the study period. Thus, in conjunction with supplemental analyses (see Table S1), these findings are consistent with propositions that COVID-19 would result in less partnered physical sex and thus reduced involvement in sexual risk-taking amongst many but not all young adults (Lindberg et al., 2020). Logically, youth who were not previously engaged in risky sexual practices simply continued to avoid taking sexual risks during the early pandemic (i.e., through abstinence or involvement in low-risk practices). In contrast, those who were engaged in risk-taking prior to the onset of the pandemic had the possibility of reductions in risk-taking during Spring 2020, yet analyses revealed the presence of a subset of young adults who continued to engage in partnered sexual risk-taking into May 2020 (i.e., the declines evidenced in the high risk classes in Spring 2020 did not culminate in zero risk for all respondents). Associations with covariates provide insights into possible pathways, which we discuss below.

Notably, these two-class solutions are relatively simpler than those in prior longitudinal studies of vaginal risk composites, which have typically revealed three or more longitudinal classes corresponding to comparatively nuanced patterns of risk involvement (Huang et al., 2012; Moilanen et al., 2010). This is likely attributable to the present study’s comparatively intense, shorter-term longitudinal design, as has been suggested in methodological investigations (Eggleston et al., 2004). To illustrate, this study involved four assessments of risk separated by one to three month intervals, spread across approximately nine months during the first year of college, which is markedly different from typical prior studies with four biennial assessments, distributed across eight or more years waves across middle adolescence and emerging adulthood. The pandemic likely further limited variability, though without a second cohort for comparison, we are unable to determine the degree to which this was an issue.

Additionally, as discussed above, prior studies have heavily emphasized vaginal or coital risk, while oral and anal risk have been relatively understudied in typically-developing young adult samples. This investigation also revealed that there was considerable consistency across the models. As expected, the classes in the oral and vaginal models were relatively similar in terms of longitudinal patterns, size, and membership (i.e., the majority of participants were classified into the same category in both models). The classes in the anal sex model were disproportionate, with exceptionally few young adults classified as high risk, and all of those youth were classified as high risk in at least one of the other models. This pattern is consistent with the larger literature, which demonstrates overlapping involvement across sexual risk behaviors in emerging adulthood (Lefkowitz et al., 2016; Moilanen, 2015b) and relatively low base rates of anal sexual risk-taking during the college years (Lefkowitz et al., 2019).

Predictors of Classes

The second study goal was to explore associations between classes and covariates. Based on Vasilenko et al.’s (2014) conceptual model, we considered several variables corresponding to individual perceptions of and experiences with COVID-19 that were all assessed in May 2020, in addition to pre-pandemic individual factors that are established covariates of sexual risk-taking in adolescence and emerging adulthood. Analyzing the data using LCGA precludes linking specific covariates variables to individual or class growth terms, as is possible in the latent growth curve analytic framework. Consequently, while we can link predictors to the likelihood of class membership, we cannot draw firm conclusions about the specific associations between covariates and growth parameters, such as the quadratic desistance demonstrated in all classes except for the high anal risk class. This limitation is particularly noteworthy in reference to the COVID-relevant covariates, which could not possibly be measured prospectively at the first study wave. It seems highly unlikely that COVID-relevant covariates that only become relevant around Waves 3 and 4 (i.e., March and May 2020) could have had any influence on young adults’ sexual behaviors at Waves 1 and 2 (i.e., in July and October 2019), and thus we interpreted their minimal effects primarily in reference to the latter two study waves.

We considered five individual factors that indicate young adults’ experiences with and perceptions of COVID-19, and posited that young adults who viewed COVID-19 as highly threatening would be likely to enact protective practices such as social distancing, which would preclude involvement in partnered sexual behaviors (including risky behaviors) in May 2020 (Rodrigues et al., 2023). Amongst these factors, only perceived financial problems resulting from COVID-19 covaried with class membership, and this effect emerged only in the vaginal model. Unexpectedly, higher perceived problems corresponded to higher likelihood of assignment to the high risk vaginal trajectory class, rather than to the low risk class as anticipated. Youth reporting high levels of financial difficulties may have been in financially precarious positions that necessitated continued in-person employment during the early pandemic, likely in service roles that may have precluded following social distancing recommendations (Cohen et al., 2020; Frechette & Reilly, 2021). Relative to the high perceived risk of regular COVID-19 exposure through strangers in the workplace, the additive risk from an in-person sexual encounter with a romantic partner may have seemed minimal (Schlager & Whillans, 2022); other investigations have revealed that young adults are less concerned about contracting COVID-19 through sexual encounters than they are about contracting COVID-19 in general (Trent et al., 2023). Alternately, this measure may indirectly assess individuals’ degree of tolerance of pandemic-related disruptions to everyday life that were commonly experienced in Spring 2020, which may indirectly tap into COVID-19 skepticism rather than perceived pandemic threat or stress as we had initially presumed. Highly skeptical young adults who viewed the early pandemic response as overblown may have found any divergence from pre-pandemic ways of life as highly burdensome. Such thinking has been linked to pandemic-related economic difficulties and low adherence to protective practices (van Prooijen et al., 2023): logically, if COVID-19 “is just a cold,” there is no reason to change one’s sexual or other social practices in order to stem viral transmission (Latkin et al., 2022). Relative to their peers in other geographic settings, college students in this study may have been more likely to be skeptical of the need to avoid partnered sex as a COVID-19 protective strategy, as West Virginia is known for its predominant highly conservative cultural milieu (Fullerton et al., 2022; Levin & Bradshaw, 2022).

Our hypotheses about the correspondence of such covariates with sexual risk classes was not supported: with the exception of financial problems, the other COVID-19 individual factors considered herein did not predict risk class membership (i.e., there were no associations with personal experiences with COVID-19, concerns for the self or close others, or COVID-related distress). While unexpected, these null findings are consistent with other studies considering similar individual factors and COVID-19 prevention behaviors (Bogg & Milad, 2020; Fullerton et al., 2022) and sexual behaviors during the first several months of the pandemic (Gleason et al., 2021). Collectively, this suggests that the present results are unlikely to be attributable to measurement limitations or an unusual sample, though replication remains needed. Thus, this study adds to the growing literature demonstrating that individuals’ choices about behavioral risks during pandemic conditions are disconnected from their personal experiences with COVID-19 and yet are connected to other individual-level constructs (e.g., personality, political orientations).

As such, while predictions about COVID-19 factors were not supported, those for traditional individual factors were generally confirmed and were consistent with theory. In keeping with research conducted prior to the pandemic’s onset, in the vaginal and oral models the strongest predictors of high risk class membership were a personal history of early sexual debut, low dispositional self-control, and having a romantic partner at both Spring 2020 study waves (Bailey et al., 2011; Lefkowitz et al., 2019). These findings align with patterns in other studies on vaginal risk trajectories and in the broad literature on sexual risk-taking in adolescence and emerging adulthood, linking stable low risk involvement to later age at debut and high self-regulation (Magnusson et al., 2019; Moilanen et al., 2010). Bivariate correlations suggested support for hypothesized effects of traditional individual factors via relational contexts, as predicted in Vasilenko et al.’s (2014) theoretical model: young adults who were in romantic relationships at both Spring study waves reported higher levels of self-control than their peers who were not partnered at these waves. It is known that high self-control covaries with high social competence, which in turn likely directly grants access to sexual partners in the context of romantic partnerships (Moilanen & Manuel, 2018). Relationship status was a particularly robust predictor, and findings were consistent with prior evidence that romantic relationships are a context for sexual behavior and risk taking during the transition to college (Bailey et al., 2011; Manlove et al., 2011; Moilanen & Manuel, 2018), and for ongoing sexual involvement during the early pandemic (Firkey et al., 2022; Herbenick et al., 2022). Young adults who were not partnered prior to mid-March have reported that pandemic control measures were barriers to meeting new partners, which limited their opportunities to engage in partnered sexual practices (Leistner et al., 2023; Maziarz & Askew, 2022). This explanation aligns with theory and seems applicable to the present results (Cooper & Zhaoyang, 2014; Vasilenko et al., 2014).

Moving home in Spring 2020 had modest effects in predicting class membership in the oral and anal risk models, such that young adults who did not move home when campus closed were more likely to be members of the high risk classes. Caution is warranted regarding this finding, as relatively few participants did not move home when campus closed in mid-March. Under the unique circumstances present in Spring 2020, such a non-normal distribution is logical given that the sample was comprised of 18 to 20 year old first year college students who were generally required to live in on-campus housing. Further, we estimated the logistic regression models without this variable and found that the pattern of results was unchanged when it was omitted (not presented; details are available from the corresponding author upon request). For young adults who moved home, parents may have imposed limits on social interactions, including physical visits with nearby romantic or sexual partners, in order to enforce social distancing norms and/or in accordance with lockdown requirements (Oosterhoff et al., 2020). While it is possible that this effect is driven by variations in external parental control across living arrangements, we propose that it could also function via self-selection processes. In particular, young adults who were engaging in high risk practices prior to the pandemic may have decided against moving home, either because anticipated parental limit-setting could disrupt their ongoing participation in partnered sexual practices, or because such a move would result in a sudden transition into a long-distance partnership (Lindberg et al., 2020; Moussa Rogers & McKinney, 2019). While we cannot test these explanations in the present dataset, the bivariate correlations provide some indirect supporting evidence in favor of self-selection, in that young adults who were in relationships at both Spring waves were less likely to move home than were their unpartnered peers. Ultimately, both possibilities warrant exploration in future research, ideally in samples with more balanced representation of students who did and did not move home.

Finally, associations with control variables were limited. There was a sole effect for sexual identity in predicting class membership in the anal risk model, which was consistent with the general literature (Chandra et al., 2011). There remains a need to replicate these findings in samples with comparatively better representation of sexual minority youth. In the interim, we urge caution regarding the findings of the anal risk model, owing to low base rates of anal risk-taking, the small size of the high risk class and to its considerable overlap with the vaginal and oral class models.

Strengths, Limitations and Future Directions

The present investigation provided novel insights into longitudinal classes of first year college students’ sexual risk-taking before and during the early COVID-19 pandemic. These findings correspond to those of cross-sectional studies with retrospective designs (Firkey et al., 2022; Herbenick et al., 2022), revealing that many young adults had low propensities for risk involvement throughout the nine-month investigation, while a separate subgroup demonstrated high propensities for sexual risk-taking that continued during the early pandemic. Use of a four-wave prospective design and examination of three forms of sexual risk-taking are considerable advantages relative to prior cross-sectional studies, in light of evidence that retrospective designs can lead to biased reporting of sexual encounters (Gillmore et al., 2010). This investigation was further strengthened by our consideration of known covariates of sexual activity patterns in emerging adulthood in addition to COVID-relevant variables. This permitted us to establish that likelihood of class membership was largely attributable to known covariates of sexual risk-taking, rather than personal experiences with or perceptions of COVID-19.

As with any other investigation, the present study possessed several noteworthy limitations. An obvious and uncontrollable limitation was the timing of the global pandemic relative to the study’s design. Although we were able to capitalize upon a prospective study that was already underway when the pandemic initially emerged, we were limited in that the study ended early in its first phase, which truncated our ability to explore ongoing changes in sexual behaviors during the time of COVID-19. These results suggest the need for this ongoing research, in pursuit of identifying any longer-term deviations in sexual behavioral patterns unique to this cohort. Such work will be crucial for addressing questions about potential rebound effects after initial lockdowns ended (Lindberg et al., 2020). The distal health implications of this “pause” in sexual activity may depend upon the degree to which young adults subsequently sought to “make up for lost time” (Brotto et al., 2022).

Further, while the study was augmented by the inclusion of multiple COVID-related individual factors, these measures may have been insufficiently sensitive, in that all the COVID-related items referred to aggregate experiences, perceptions, and feelings experienced since the start of the pandemic or throughout the Spring 2020 semester (i.e., their timeframe was out of alignment with the thirty-day span of all sexual risk items). These measures were also not conducive to capturing temporal changes in these dimensions, though at the time of data collection, no evidence was yet available regarding rapid changes in adults’ risk perceptions and feelings during this period (Li et al., 2021; Wise et al., 2020). Further, respondents completed the Wave 4 survey during the first two weeks of May, around the time that the state’s stay at home orders ended (West Virginia Office of the Governor, 2020). If respondents interpreted this policy change as a signal of reduced threat, this may have led to underreporting of prior concerns (Bourassa, 2021). While our consideration of vaginal, oral, and anal risk was advantageous, an additional limitation may stem from our reliance upon composite measures of sexual risk. In keeping with numerous prior investigations (Barker et al., 2019), we aggregated across two risky practices (i.e., multiple partners and unprotected encounters) that have confirmed value as vital indicators of outcomes such as pregnancy and STIs during adolescence and emerging adulthood. While the use of composites facilitates the identification of subtle associations between covariates and combinations of risk behaviors, it precludes examinations of longitudinal changes in discrete sexual risk-taking behaviors. In order to mitigate these concerns, future studies should include additional dimensions of sexual behaviors and risk, such as contraceptive use and involvement in high-risk practices, both independently and as part of risk composites. Further, and while not a unique limitation to this study, grouping young adults with zero risk (i.e., a score of 0 on the composite, representing those with either no sexual activity during the prior month or only protected encounters with a single partner) or with a single risky practice (i.e., a score of 1 one on the composite, including respondents with either multiple partners or unprotected encounters) may potentially obscure meaningful variations between subgroups within each level over time. Relatedly, survey length concerns precluded our inclusion of follow-up questions that would have permitted us to determine if respondents had the same partner(s) across sexual behaviors and/or over time. Though their cumulative levels of risk would vary considerably, in this context we were unable to distinguish between participants who had a single partner across all four waves from those who had a new partner at each of the study waves, as both would have had the same risk composite score at each timepoint. For future studies, researchers are advised to include questions that would facilitate making such distinctions.

A final limitation is that all of this study’s participants were college students at a single large, land-grant university which has long had high national rankings for its party culture (Princeton Review, 2020). While these findings may generalize to students at similar residential campuses with cultures that endorse risk-taking (Kuperberg & Padgett, 2016), they may be less applicable to emerging adults at other types of university settings where risky practices are a comparatively less prominent part of student life (e.g., commuter campuses with large proportions of adult learners or religiously-oriented colleges) or to those who do not attend college (Bailey et al., 2011; Vasilenko et al., 2018). The relative homogeneity of the sample may have compounded the short-term longitudinal design and have further contributed to the identification of fewer classes (and less variation in moderate to high risk class patterns) relative to those identified in investigations with diverse national samples and longer-term longitudinal designs. Such homogeneity resulted in low frequencies for the minority categories for some dichotomous predictor variables (e.g., non-White, non-heterosexual, did not move home), which may have biased the results of the logistic regressions. This is especially true with respect to engagement in high-risk anal sex. As such, the findings from the anal risk models should not be interpreted as widely generalizable, but rather as preliminary evidence that should inform future studies targeting diverse populations that engage in risky anal sex behaviors.

Conclusions

In sum, the present study makes novel contributions to the literature on the development of sexual risk-taking, documenting longitudinal classes in vaginal, oral, and anal sexual risk-taking across the first year of college in a large sample of emerging adults during the COVID-19 pandemic. While these findings varied relative to other investigations, they provided confirmatory evidence of widespread (but not ubiquitous) declines in risk-taking during the early pandemic. They also revealed the presence of a subgroup of youth that persisted in sexual risk-taking through Spring 2020. Overall, in the context of sexual behaviors in emerging adulthood, the present study extends the literature documenting the degree to which individuals’ health behaviors are largely disconnected from their experiences with and perceptions of COVID-19 (Bogg & Milad, 2020; Fullerton et al., 2022; Gleason et al., 2021). Instead, patterns of risk-taking behaviors were comparatively strongly linked to established covariates of sexual risk. While caution is warranted in the absence of corroborating evidence, the findings of the present investigation indicate that major pandemic-specific modifications to developmental theoretical models of sexual risk-taking are generally not needed in order to explain sexual risk-taking during the early pandemic.

In terms of implications for applied efforts, although many young adults had low involvement in risky sexual practices in Spring 2020, practitioners should not assume that sexual risk is by default mitigated in the event of lockdowns because all young adults will be motivated to avoid COVID-19 transmission. Indeed, a subgroup of young adults who were already engaging in risk prior to pandemic onset continued to engage in risk-taking during the early pandemic. Further, young adults’ individual experiences with and perceptions of COVID-19 did not have protective effects during this period, and perceptions of pandemic-related financial hardship may have ultimately facilitated risky behaviors for a subgroup of youth. Thus, we concur with others’ recommendations concerning the utility of varied intervention strategies, which may need to be tailored to address several risk factors. The present findings primarily speak to the value of comprehensive evidence-based sexuality education in early adolescence, which ideally target the promotion of self-control as part of delaying debut and enhancing knowledge of medically-accurate risk reduction strategies. However, as partnered sexual behaviors are a typical part of emerging adulthood, it is also necessary to continue to reduce barriers to free or low-cost forms of contraception and condoms/barriers through university or community clinics in order to help youth avoid negative outcomes such as mistimed pregnancies and STIs. Young adults with romantic partners have sexual encounters most frequently and are less likely to use condoms than their unpartnered peers (Bailey et al., 2011; Manlove et al., 2011). This group may particularly benefit from targeted intervention strategies that encourage advance preparation in anticipation of disrupted access to sexual and reproductive health services (Lewis et al., 2021), including access to long-acting reversible contraception methods (Aly et al., 2020).