Re-Examining the Link Between Premarital Sex and Divorce

Background

Although scholars have long been interested in the causes of divorce, the predictive role of premarital sexual partnerships has received relatively little research attention. What we do know is primarily based on three studies. Kahn and London (1991) compared women who reported any premarital sexual experience to those with none, and found higher divorce risk in the former group. Expanding on this finding, Teachman (2003) showed that there is elevated divorce risk only for those with premarital partners other than their eventual spouse, that is, sex with a future spouse alone does not predict divorce (but see Paik, 2011). Furthermore, he found that adverse effects of premarital cohabitation with someone other than a future spouse disappeared when controlling for premarital sex. Finally, Paik (2011) showed that the adverse effect of early sexual debut on marital stability can be explained by the fact that those who have sex earlier are more likely to have premarital sexual partners other than future spouses. Other studies also have demonstrated in passing that premarital sex predicts divorce (e.g., Heaton, 2002; Manning & Cohen, 2012). In each of these studies, the link between premarital sex and divorce is significant when controlling for a variety of individual characteristics.

These three studies have suggested various possible explanations for their findings, which largely mirror the literature on cohabitation and divorce (e.g., Rosenfeld & Roesler, 2019; Sassler & Lichter, 2020). Premarital sex may be an indicator of permissive attitudes toward sex and marriage, low religiosity, or a predilection for sexual variety, all of which are linked to higher divorce risk (Kahn & London, 1991; Paik, 2011; Vaaler, Ellison, & Powers, 2009). In this case, the premarital sex-divorce link is best explained by pre-existing differences between individuals. Alternatively, the experience of premarital sex itself, especially with multiple partners, may contribute to the development of more permissive attitudes toward sex or a greater awareness of sexual alternatives, either of which may serve to undermine marital stability (Paik, 2011; Teachman, 2003). In this case, the premarital sex-divorce link is better understood in terms of causality.

Data and Methods

The data for this study come from the first, third, and fourth waves of the National Longitudinal Study of Adolescent to Adult Health (Add Health) (https://addhealth.cpc.unc.edu/) (Harris et al., 2019). The first wave of data collection occurred in 1994–1995 and consisted of 20,745 adolescents from grades 7–12, as well as 17,670 parents (mother if available, another adult in the household if not). Data for Wave III (N = 15,197) were collected in 2001–2002 when respondents were 18–26 years old. Wave IV took place in 2008–2009 and consisted of 15,701 respondents ages 24–32, with a response rate of 80.3%.

This study uses two separate samples from Add Health. The first sample is limited to those who had ever been married as of Wave IV (N = 7797), had a valid sample weight (N = 7393), and had sufficient data to calculate marital duration (N = 7286). We remove 67 cases where respondents, despite having been married, report no lifetime sexual partners, as this likely indicates an invalid response. An additional 166 cases were excluded due to reporting having no mother or father or any other household caregiver who serves this role at Wave I (e.g., grandparents, aunts/uncles), as some items regarding parental relationships are included in the analysis. A further 23 cases were removed due to reporting already being married as of Wave I. This leaves an analytic sample of 7030 respondents who had been in either opposite-sex (N = 6989) or same-sex (N = 41) marriages. The second sample is a subset of the 7030 which includes only respondents whose first marriages did not start until after Wave III (N = 3506).

Each sample contains distinct advantages and disadvantages. The first sample is larger, more representative, and offers a longer duration of observation of marriage, and is thus better suited to examining the effects of early-life factors and personal characteristics. The second sample is smaller and is selective of later marriages, but contains less ambiguity regarding the time ordering of sexual partners. This allows for greater approximation of the number of premarital partners needed in order to examine the shape of the association between sexual history and divorce risk.

Measures

Dependent Variable

The dependent variable in this study is the duration of first marriage in months. In Wave IV, respondents were asked to provide information regarding what month and year their marriages started and ended. The survey items did not distinguish between separation and divorce but instead relied on self-report as to when respondents viewed their marriages as ending. ¹ If respondents reported their marriages were ongoing, the end date was set as the date of the Wave IV interview and observations were right-censored.

Independent Variables

The key independent variables in this study are measures of premarital sexual partnerships. In Wave III, respondents were asked to provide their total lifetime number of vaginal sexual partners, as well as the total number of vaginal, oral, or anal partners they had had since 1995. In Wave IV, respondents were asked about their total lifetime number of sexual partners, including prior to the age of 18. These items, in conjunction with marital timelines, were used to develop a dichotomous indicator of whether respondents had any premarital sexual partners other than their spouse. For 434 cases (6%) in the full sample, there was insufficient information available to make this determination; values for these cases were imputed. Measures and coding criteria used to develop this indicator are shown in Table 1.

OPEN IN VIEWER

Table 1.

Coding Criteria for Premarital Sex Indicator.

Measures used

Wave IV

Total male + female lifetime vaginal, oral, or anal sexual partners

Total male + female vaginal, oral, or anal sexual partners prior to age 18

Divorce status

Report of any concurrent sexual partners during first marriage a

First marriage began prior to Wave III

Age at first marriage

Wave III

Total lifetime vaginal sexual partners

Total vaginal, oral, or anal sexual partners since the summer of 1995

Start and end dates for sexual relationships since the summer of 1995 a

Sufficient conditions to code respondents as having had premarital, nonspousal sexual partners (“1”)

In first marriage at Wave IV, report more than one lifetime partner, no nonspousal partners concurrent with marriage

Unmarried as of Wave III, more than one partner ever as of Wave III

Married prior to Wave III, report any sexual relationships having ended prior to when marriage started

Married after age 18, more than one partner prior to age 18

Sufficient conditions to code respondents as having had no premarital, nonspousal sexual partners (“0”)

No more than one partner reported on any Wave IV or Wave III measure

No more than one partner reported prior to Wave III, married by Wave III

No more than one partner reported prior to age 18, married by age 18

Note: Above criteria allowed for logical certainty of categorization for 94% of respondents. Missing data for the remaining 6% of cases were addressed using multiple imputation (m = 10).

^aMeasure only available for a subset of respondents.

For the subsample of those who did not marry until after Wave III, we first take the highest number of sexual partners reported across three items: lifetime vaginal partners as of Wave III, total sexual partners of any kind between 1995 and Wave III data collection, or partners of any kind prior to age 18 as reported at Wave IV. The partners counted in these three measures may overlap, so we use the highest number to capture as many partners as possible while ensuring no partner is counted more than once. We subtract one from this number (unless respondents report no partners) to account for the possibility that a single partner may be the respondent’s eventual spouse. Next we divide the sample intro three groups: 1) Those with no premarital partners (excluding future spouses) as of Wave III, 2) those with one to eight partners, and 3) those with nine or more. We selected these categories based on a series of analyses using dichotomous indicators of sexual partners, which showed divorce risk to increase substantially with one partner and again with nine, while remaining stable for partner counts between one and eight (see appendix). This approach also provides an improved model fit as measured by the deviance scores, compared to alternative approaches such as using quartiles or a continuous sexual partner variable.

It should be noted that because this second stage of analysis only includes sexual partners measured as of Wave III, any premarital partners in between Waves III and IV are not counted. This may result in some measurement error, leading to estimates biased downwards. Results from this analysis should thus be regarded as conservative.

Adolescent and Family Characteristics

To account for early-life experiences and individual characteristics, we include several measures of respondents’ Wave I beliefs and values, religious involvement, and other traits. We measure religious identity using a modified version of Lehman and Sherkat’s (2018) recent typology (the key modification is that Latter-Day Saints are coded to the “Sectarian” category due to considerations of small cell sizes). Wave I religious attendance is included as a continuous variable. Unfortunately, Add Health respondents from Wave I who reported no religious affiliation were not asked about religious behaviors; these cases were coded as having no attendance.

We include two additional dichotomous variables to capture beliefs and values about premarital sex. The first indicates whether or not the respondent had ever taken a virginity pledge (Bearman & Brückner, 2001; Uecker, 2008). The second indicates whether adolescents report that in an ideal relationship, they would expect to have sex, but not get married, in the next year. Specifically, respondents were given a set of cards identifying different relationship activities (e.g., going out together in a group, holding hands, meeting partner’s parents, getting married), and asked to return to the interviewer any cards indicating activities they would not want to happen in the next year in their ideal relationship. If they kept the card for having sex, but returned the card for getting married, they were coded as a “1” on this indicator. ²

The analysis also includes scales of Wave I depressive symptoms, delinquent behaviors, and decision-making style, as these all may be associated with both sexual behavior and marital outcomes. The depression scale uses 10 items on the frequency of experiencing various depressive symptoms, drawn from the CES-D depression inventory (e.g., “You felt you were too tired to do things”) (α = .72). The delinquency scale includes 15 items on frequency of various delinquent activities in the past year (e.g., damaging property, lying to parents about activities) (α = .83). Finally, adolescents were asked to assess their own personalities. We use five items pertaining to decision-making (e.g., “When you have a problem to solve, one of the first things you do is get as many facts about the problem as possible,”) to create a scale measuring a cautious approach to making decisions (α = .64) (Uecker, Regnerus, & Vaaler, 2007).

Two items are included to capture potential parental influences on children’s sexual and later marital outcomes. As positive parent–child relationships may serve as a protective factor against high-risk sexual behaviors (McElwain & Bub, 2018; Price & Hyde, 2009), we measure parent–child relationship quality using a scale of five items indicating the adolescent’s perception of closeness to each parent (e.g., “Most of the time, your mother is warm and loving toward you”) (α = .85). This scale used responses about mothers where available, fathers otherwise. We further include a cumulative measure of the number of activities respondents report doing with their parent in the past 4 weeks (e.g., gone shopping, played a sport).

Sociodemographic Controls

Based on past studies, a number of controls are included to account for potential spuriousness in the relationship between premarital sex and marital stability. From Wave I, these include sex, race, and highest level of parent education in four categories: 1) Less than high school, 2) high school diploma/GED, 3) some post-high school education, and 4) college degree or higher. We further include measures of the number of family structure transitions experienced as of Waves I and II, using a variable constructed through procedures outlined in Gaydosh & Harris (2018; see also Martinson & Wu, 1992). ³ From Wave IV, we add respondent education and indicators of the respondent’s relationship history. These include age at first marriage, any premarital birth, any premarital cohabitation, age at first sex in three categories (before age 16, at ages 16 or 17, and age 18 or later), and a binary indicator of the experience of nonconsensual sex as a minor. ⁴

Results

Descriptive statistics are shown in Table 2. Of respondents who had ever been married as of Wave IV, approximately 21% had a marriage which ended in divorce. Eighty-four percent of respondents reported having premarital, nonspousal sexual partners. For the late-marrying subsample, the proportion whose first marriage ended in dissolution is about 10%. Later marriage is a protective factor against dissolution (e.g., Teachman, 2002), and these marriages have shorter durations of observation, and hence a shorter period in which to observe divorce. In this group, about a quarter reported one or zero sexual partners as of Wave III, over half (57%) reported between one and eight partners, and 19% reported 9 or more.

OPEN IN VIEWER

Table 2.

Descriptive Statistics for Ever-Married Respondents as of Add Health Wave IV.

Variables	W1 (Full)	W1 (Late-Marrying)	Range	% imputed
	M (SD) or proportion
Marital dissolution by Wave IV a	.21	.10
Premarital sexual partners (excluding first spouse) a	.84			6.2
0		.24
1–8		.57
9+		.19
Child and family characteristics
No religion	.10	.10
Sectarian	.32	.28
Catholic	.26	.28
Moderate/liberal protestant	.27	.29
Other religion	.05	.05
Religious attendance frequency	2.81 (1.18)	2.85 (1.18)	1–4	0.1
Taken virginity pledge	.15	.15		0.3
Sex in perfect relationship	.27	.27		0.8
Depressive symptoms	0.52 (0.37)	0.49 (0.35)	0–3
Delinquent behaviors	0.27 (0.33)	0.26 (0.32)	0–3	0.4
Cautious decision-making style	3.67 (0.57)	3.67 (0.57)	1–5	0.1
Parent–child relationship quality	4.39 (0.66)	4.44 (0.63)	1–5
Parent–child activities in past 4 weeks	3.71 (1.90)	3.78 (1.91)	1–9	0.1
Sociodemographic controls
Female	.57	.55
White	.62	.62
Black	.13	.13
Hispanic	.16	.14
Other race	.09	.11
Parent has < HS education	.06	.07		3.4
Parent has HS diploma/GED	.16	.22		3.4
Parent has some college	.47	.29		3.4
Parent has a college degree	.32	.41		3.4
Family structure transitions	0.76 (1.10)	0.69 (1.08)	0–5	0.5
Respondent has < HS education	.06	.04
Respondent has HS diploma	.16	.12
Respondent has some post-HS	.47	.42
Respondent has BA or higher	.32	.43
Age at first marriage a	23.80 (3.17)	25.70 (2.31)	14.8–32.3
Premarital birth a	.21	.22		0.3
Premarital cohabitation a	.70	.75
First sex before age 16	.38	.34		0.9
First sex at ages 16–17	.32	.32		0.9
First sex at age 18 or older	.30	.34		0.9
Coercive sex as minor a	.12	.10		0.8
No. of observations	7030	3506
Marriage-months	389,197	131,769

Results for the first set of discrete-time models are shown in Table 3. The most important takeaway is that premarital sex is a highly significant predictor of divorce at the p < .001 level in every model. This effect remains robust even with the inclusion of the full set of early-life factors relating to beliefs or values, religious practice, family characteristics, individual attributes, and parent–child relationships. The effect size is both large and stable: across models, those with premarital sexual partners have more than twice the odds of divorce as do those without (ORs = 2.50—2.52). We thus find no evidence that the link between premarital sex and divorce is due to selectivity based on early-life religiosity or beliefs and values. In fact, although several variables significantly predict divorce in bivariate analyses (not shown), most do not predict divorce in full models, aside from those confirmed in past research: African Americans are at higher risk of divorce compared to whites, people with a college degree have lower divorce risk, experience of family transitions predicts higher risk of divorce, and age at marriage is strongly and negatively linked to divorce.

OPEN IN VIEWER

Table 3.

Odds Ratios from Discrete-Time Event History Models Predicting Dissolution of First Marriage.

Variable	M1	M2	M3
Any premarital sexual partners	2.61 (0.12)***	2.54 (0.14)***	2.52 (0.13)***
Child and family characteristics
Sectarian			0.89 (0.16)
Moderate/liberal protestant			0.94 (0.15)
Catholic			0.98 (0.17)
Other religion			0.86 (0.20)
Religious attendance frequency			0.98 (0.04)
Taken virginity pledge			1.01 (0.10)
Sex in perfect relationship			0.98 (0.09)
Depressive symptoms			0.98 (0.10)
Delinquent behaviors			1.10 (0.10)
Cautious decision-making style			0.96 (0.06)
Parent–child relationship quality			1.02 (0.06)
Parent–child activities in past 4 weeks			0.99 (0.02)
Sociodemographic controls
Female		0.90 (0.07)	0.92 (0.07)
Black		1.21 (0.11) †	1.27 (0.11)*
Hispanic		0.89 (0.11)	0.86 (0.12)
Other race		1.23 (0.14)	1.22 (0.14)
Parent has HS diploma/GED		1.09 (0.11)	1.08 (0.12)
Parent has some college		1.09 (0.11)	1.09 (0.11)
Parent has a college degree		0.90 (0.14)	0.91 (0.14)
Respondent has HS diploma		0.81 (0.14)	0.82 (0.14)
Respondent has some post-HS		0.95 (0.11)	0.97 (0.12)
Respondent has a BA		0.62 (0.16)**	0.64 (0.17)*
Family structure transitions		1.07 (0.03)*	1.06 (0.03) †
Age at first marriage		0.87 (0.02)***	0.87 (0.02)***
Premarital birth		0.99 (0.09)	0.99 (0.09)
Premarital cohabitation		1.04 (0.09)	1.02 (0.09)
First sex at ages 16–17		0.90 (0.07)	0.91 (0.07)
First sex at age 18 or older		0.95 (0.09)	0.98 (0.09)
Coercive sex as minor		0.95 (0.10)	0.96 (0.10)
Months	1.02 (0.00)***	1.02 (0.00)***	1.02 (0.00)***
Months squared	1.00 (0.00)***	1.00 (0.00)***	1.00 (0.00)***
Deviance	19,964	19,637	19,630

Analyses of the subsample of those who did not marry until after Wave III are shown in Table 4. To reiterate, the purpose of using this sample is to be able to capture the variation in premarital sexual partners more effectively. The key results here are mostly consistent across models: those with the highest number of premarital sexual partners as of Wave III (nine or more) have about triple the odds of divorce compared to those with none (ORs = 2.65—3.20). Notably, this effect becomes stronger as controls are added to the model, indicating such hypothesized selection factors as sociodemographic or religious characteristics actually suppress, rather than help explain, the effect of premarital sex for those with the highest number of partners.

OPEN IN VIEWER

Table 4.

Odds Ratios from Discrete-Time Event History Models Predicting Dissolution of First Marriage among Later-Marrying Respondents.

Variable	M1	M2	M3
1–8 premarital partners	1.50 (0.20)*	1.59 (0.22)*	1.64 (0.21)*
9+ partners	2.65 (0.20)***	3.00 (0.24)***	3.20 (0.24)***
Child and family characteristics
Sectarian			0.38 (0.31)**
Moderate/liberal protestant			0.45 (0.29)**
Catholic			0.46 (0.27)**
Other religion			0.70 (0.34)
Religious attendance frequency			1.06 (0.10)
Taken virginity pledge			1.22 (0.21)
Sex in perfect relationship			0.78 (0.18)
Depressive symptoms			0.76 (0.22)
Delinquent behaviors			1.21 (0.21)
Cautious decision-making style			0.95 (0.12)
Parent–child relationship quality			1.04 (0.13)
Parent–child activities in past 4 weeks			1.04 (0.04)
Sociodemographic controls
Female		1.23 (0.16)	1.23 (0.17)
Black		1.03 (0.22)	1.16 (0.22)
Hispanic		0.77 (0.28)	0.79 (0.29)
Other race		1.37 (0.24)	1.18 (0.23)
Parent has HS diploma/GED		0.87 (0.28)	0.90 (0.27)
Parent has some college		0.99 (0.25)	1.00 (0.25)
Parent has a college degree		0.77 (0.29)	0.75 (0.29)
Respondent has HS diploma		0.65 (0.30)	0.65 (0.29)
Respondent has some post-HS		0.72 (0.24)	0.72 (0.23)
Respondent has a BA		0.43 (0.30)**	0.43 (0.30)**
Family structure transitions		1.05 (0.05)	1.04 (0.05)
Age at first marriage		0.90 (0.04)**	0.91 (0.04)*
Premarital birth		0.85 (0.18)	0.91 (0.19)
Premarital cohabitation		1.44 (0.24)	1.42 (0.26)
First sex at ages 16–17		0.89 (0.18)	0.90 (0.18)
First sex at age 18 or older		1.35 (0.20)	1.37 (0.21)
Coercive sex as minor		0.85 (0.22)	0.85 (0.22)
Months	1.04 (0.01)***	1.04 (0.01)***	1.04 (0.01)***
Months squared	1.00 (0.00)*	1.00 (0.00)*	1.00 (0.00)*
Deviance	4999	4928	4889

Those with one to eight partners are also at greater risk of divorce, though this coefficient is weaker than for those with nine or more partners. Specifically, in the full model the odds of divorce for those with one to eight partners are 64% higher than those with no premarital partners (ORs = 1.50—1.64). This effect is also not attenuated as controls are added to the model, reinforcing the finding that explanations based on early-life experiences and personal characteristics are not supported. Additional analyses show those in this middle category to have a significantly lower divorce risk than those with nine or more partners, indicating three distinct groups. Taken together, these results suggest that the relationship between number of premarital partners and marital dissolution is nonlinear. They point rather to three tiers of divorce risk, with the lowest risk for those with no premarital, nonspousal partners, a modest increase for those with some, and a sharp increase for those with many. These results are more consistent with the notion that the effect of premarital sex on divorce becomes stronger, not weaker, as sexual partners accumulate.

Finally, to test for differences in the effect of premarital sex on divorce between men and women, we repeat all analyses from Tables 3 and 4 and include an interaction between gender and premarital sex variables (not shown). The results are straightforward: we do not any evidence of sex differences in the link between premarital sex and divorce risk in any model for either sample. Although there is theoretical reason to expect such a difference, our results suggest otherwise.

Discussion

Past research has established a clear link between premarital sex, particularly with partners other than an eventual spouse, and the risk of marital dissolution (Kahn & London, 1991; Paik, 2011; Teachman, 2003). Thus far, however, we know little about the nature of this link. The purpose of this study was to shed light on three areas which have not been fully addressed in past literature: the potential role of early-life factors or personal characteristics in explaining this relationship, the possibility of nonlinear effects of premarital sexual partnerships on divorce risk, and whether the relationship varies between men and women. A better understanding of these factors contributes to our knowledge of the links between early-life and later-life relationship experiences and outcomes, an important focus of theory and research on the family life course (Sassler, 2010).

Previous research pointed to a variety of individual and social variables to explain the relationship between premarital sex and divorce, including nontraditional views on sex and marriage, weaker religious attachments, and lower-quality family relationships (Kahn & London, 1991; Paik, 2011). We find no support for these explanations, and show that the effect of premarital sex remains highly significant after accounting for a wide range of individual and social differences between respondents. Needless to say, our analysis does not rule out the possibility that unmeasured variables might be driving selection. Future research should endeavor to test for the potential causal effects of premarital sex on divorce, and to elaborate on possible selection processes which have not been considered in past scholarship.

As expected, we find evidence of a nonlinear relationship between the number of sexual partners and the risk of divorce. Those in the highest category of partners (9+) consistently show the highest divorce risk by a substantial margin, followed by those with one to eight partners, with the lowest risk for those with none. In other words, we find distinct tiers of divorce risk between those with no, some, or many premarital, nonspousal sexual partners. As over half the sample falls into the middle category, these results may be interpreted as indicating two distinctive groups. At one extreme, having no premarital partners serves as a protective factor against divorce risk. At the other, those in the top quintile experience especially high levels of marital dissolution. These findings highlight the importance of incorporating the variation in sexual history into research in this area, rather than relying on dichotomous indicators of premarital sex. A possible implication here is that the robust effect of premarital sex found in past studies is being driven largely by a minority of respondents with especially high levels of both sexual partners and divorce rates. This reflects the fact that although partner counts of eight or less have become increasingly normative, having more partners may indicate distinctive characteristics which are not conducive to marital stability.

We find no evidence of gender differences in the link between premarital sex and divorce. This is a surprising finding—the domains of sexuality and marriage are highly gendered (e.g., Monin & Clark, 2011; Okami & Shackelford, 2001; Oliver & Hyde, 1993), and there are many plausible theoretical pathways by which the premarital sex-divorce relationship might be expected to differ between men and women. Future research might consider if the mechanisms linking premarital sex and divorce function the same way for men and women.

This study has contributed to our understanding of the link between premarital sex and marital dissolution by taking advantage of the mixed-sex sample and longitudinal design of the Add Health study in order to address some of the limitations of past literature. Yet Add Health has its own limitations with respect to the aims of this study. As all respondents at Wave IV were 32 years old or younger, the sample is selective of those who marry younger, and the time to observe divorce is somewhat limited. The results are therefore most reflective of early marriages and divorces. Also, at the start of data collection, respondents ranged from early to late adolescence, and many had already had sex. On a related note, in some cases it is unclear whether sexual partnerships occurred before, during or after marriage. To ensure proper temporal ordering, tests of the effects of the number of sexual partners were only performed on a late-marrying subsample, and the measure used did not capture all partners. Nevertheless, it represents a more detailed measure than almost all prior studies have used. Needless to say, a more exact measure for the full sample would be ideal. Similarly, we relied on proxies for beliefs and values about marriage and divorce as well as marital quality, so more precise measures would be preferable. Finally, though Add Health data are well-suited to ruling out some of the hypothesized selection processes explaining the premarital sex-divorce link, they are less appropriate for examining possible causal mechanisms. This would require more detailed timelines of sexual histories and more frequent waves of data collection than Add Health provides.

This study makes a contribution to our understanding of the link between premarital sex and divorce, which has been heretofore limited in key respects. It reinforces the finding that the effect of premarital sex is robust, and shows that it applies equally to men as well as women. Moreover, our findings help to rule out common explanations offered by past research. The results here also serve to highlight the importance of conceptualizing sexual history more broadly, taking into account its extent and timing, when considering how it might affect marriages. As relationship experiences over the life course have become more varied and complex in recent decades (Sassler & Lichter, 2020), it is increasingly vital that theory and analysis adjust to take account of these developments.