Do Expectations of Divorce Predict Union Formation in the Transition to Adulthood?

Abstract

Objective:

This study describes the association between explicit expectations to divorce and subsequent first union formation over the transition to adulthood (ages 18–28).

Background:

Expectations for marriage in young adulthood predict union formation. Even before marrying, young adults may express a perceived risk of eventual divorce, and expectations of divorce may also have implications for union formation over the transition to adulthood.

Method:

Data from the 2005–2015 years of the Panel Study of Income Dynamics Transition to Adulthood Supplement (n = 2052) were used to estimate the association between expectations to divorce and entry into first premarital cohabitation and first marriage using discrete-time logistic and multinomial logistic survival models.

Results:

As hypothesized, greater expectations for divorce predicted slower entrance into first marriage, even controlling for expectations for marriage and various sociodemographic characteristics, and predicted a greater likelihood of both remaining single and being first observed cohabiting instead of marrying in young adulthood for both men and women.

Conclusion:

Despite desiring to marry, young adults may delay marriage if they are concerned about their risk of future divorce.

Young adults often have high hopes for their futures, including imagined family lives (Anderson, 2016b; Plotnick, 2007) but also likely recognize early on that their lives will not be perfect. Those who have witnessed family instability or divorce think about their own potential for divorce (Boyer-Pennington, Pennington, & Spink, 2001; Waller & Peters, 2008), and even those who have not witnessed divorce or dissolution first-hand recognize the possibility that their plans may not work out quite as they imagined (Arocho & Purtell, 2018). How are youth’s pictures of the future, good and bad, associated with how they behave? Young adult expectations for entering unions predict behavior (Arocho & Kamp Dush, 2017; Willoughby, 2014; Willoughby & Dworkin, 2009). However, little attention has been paid to the less positive expectations youth may hold, especially their expectations to experience divorce. Considering the prevalence of divorce in the United States (Cherlin, 2010a) it is perhaps not surprising that even unmarried young adults may consider themselves to be at risk of divorce in the future (Arocho & Purtell, 2018), and explicit expectations of divorce might also be associated with behaviors in young adulthood.

Expectations, Marriage, and Cohabitation in the U.S.

Expectations for marriage, specifically stronger expectations to marry one day or to marry sooner rather than later, predict behaviors, such as risk-taking, in young adulthood (Willoughby & Dworkin, 2009). Importantly, they also predict early entrance into marriage (Arocho & Kamp Dush, 2017; Willoughby, 2014). Understanding how expectations are associated with behavior is relevant because marriage, especially in the U.S. context, has changed dramatically in recent history (Cherlin, 2004). Marriage has experienced practical and ideological changes (Cherlin, 2010a, 2010b), as evidence by decreasing marriage prevalence (Manning, Brown, & Payne, 2014), rising median age at first marriage (Manning et al., 2014), and relatively high divorce rates (Cherlin, 2010a). At the same time, cohabitation has become the modal first union for young adults in the U.S. (Manning et al., 2014) as the overall prevalence of cohabitation has grown (Cherlin, 2010a).

With the high cultural significant placed on marriage in the U.S. (Cherlin, 2010b), individual may not be able to help but form a constellation of marital attitudes (Willoughby, Hall, & Luczak, 2015) when exposed to marriage rhetoric so often. Despite the aforementioned changes in union formation and dissolution, marriage expectations in the U.S. remain high, especially among young people (Anderson, 2016b). Even so, the median age at first marriage continues to rise and the role of marriage continues to change; what once was a step on the path towards adulthood has become for many a capstone to be placed once a couple has everything else in line (Cherlin, 2004). Thus, to better understand the imperfect association between expectations for marriage and actual marriage, perhaps we should examine another, related expectation: expectations that one’s hypothetical marriage may end in divorce.

Cohabiting couples reference concerns about divorce as a barrier to marriage (Miller, Sassler, & Kusi-Appouh, 2011; Waller & Peters, 2008). Across multiple European contexts, those seeking to avoid divorce seem to favor less formal unions, with a rise in the divorce rate of older generations appearing to contribute to a rise in cohabitation amongst younger generations (Perelli‐Harris, Berrington, Sánchez Gassen, Galezewska, & Holland, 2017). Indeed, the role of cohabitation in the association between expectations for divorce and entry into marriage should not be ignored. Cohabitation can play many roles in the lives of young adults in their quests to form the family lives they desire. For those who recognize the risk of divorce and the associated costs, cohabitation may be seen as a safe place to test a relationship (Anderson, 2016a) before transitioning to marriage; for others, cohabitation may seem the preferred or safer relationship option altogether (Waller & Peters, 2008). In an older sample of young adults, professing greater acceptability of divorce both predicted and was predicted by cohabitation experience (Axinn & Thornton, 1992), further suggesting an important link between the two.

Expectancy-value theory

To understand the behaviors one may enact in pursuit of a goal, expectancy-value theory (Wigfield & Eccles, 2000) suggests we must understand both the individual’s value of the goal and their expectations that they can be successful in achieving it. Someone with a high-value goal but who believes themselves unable to achieve it will not be motivated to work towards the goal. Similarly, someone who knows they could achieve a goal that holds no value to them also will not work towards the goal. Thus, the combination of the two beliefs should be important to understanding outcome behaviors. In the present study I considered marriage as the goal in question. The value of this outcome was measured in this study as expectations to marry. Expectations of success—in this case direct, personal evaluations of one’s risk of future divorce—provided a counterpoint to expectations of marriage. Previous research on the predictors of expectations to divorce suggest that various background and personal factors theorized to predict expectations of success also predict expectations of divorce (Arocho & Purtell, 2018) and in this study, I continue this work by applying expectations to later behavior.

Research question 1: How are divorce expectations associated with marriage in young adulthood?

Interviews with cohabiting couples revealed that they often consider the risk of divorce when thinking about marrying their partners (Miller et al., 2011). Even without explicitly verbalizing it, those who are exposed to divorce may be wary of marriage because of the high costs associated with its failure (Waller & Peters, 2008). These concerns likely extend even into young adulthood, and I hypothesize that those with higher expectations of divorce in early adulthood will be the least likely to marry.

Research question 2: How are divorce expectations associated with cohabitation, relative to marriage, in young adulthood?

Overall, young adults are more likely to enter into a cohabiting relationship than a marriage over the transition to adulthood (Manning et al., 2014). However, many youth believe in a “divorce-proofing” quality of cohabitation (Anderson, 2016a), and some young adults cite cohabitation as desirable specifically because of anxiety about the possibility of divorce (Arnett, 2015; Perelli‐Harris et al., 2017). Thus, I expect that youth with higher expectations of divorce will be more likely to cohabit over the transition to adulthood, especially when compared to the likelihood of entering marriage as their first union.

Method

I used the Panel Study of Income Dynamics (PSID) Transition into Adulthood Supplement (TAS; Panel Study of Income Dynamics, public use dataset). As the TAS data were not designed specifically to capture detailed family formation histories in this age period, there are limitations to the measurement of relationships (discussed in more detail below) that should be noted. However, these data are uniquely qualified for the current study as respondents were asked to explicitly quantify their perceived personal risk of divorce, a rare question in such a large and diverse sample of unmarried young adults.

The overall PSID started as a nationally-representative sample of American families in 1968. As children in the families aged and started their own households their new families were added back into the PSID study, and their children were eventually added similarly. Starting in 2005 and repeated biennially through 2015, respondents assessed as children in the 1990s and early 2000s joined the TAS when they turned 18 or finished secondary education and were interviewed until age 28. The TAS has attained high response rates each year, ranging from 87 to 92% (Institute for Social Research, 2018). My analytic sample was restricted to TAS respondents (total N = 2,893) included in at least two waves and who had valid transition dates when a transition was indicated, n = 2,052. Seven respondents were missing either relationship type start date, and 368 respondents had cohabited, married, or both before their first interview. The remaining respondents were removed due to attrition; due to the analytic method, respondents had to contribute at least two interviews’ worth of data. All imputation and analyses models were weighted using the provided individual longitudinal weights for each respondent’s first TAS interview.

Independent Variables: Expectations of divorce and marriage

Youth reported their expectations of marriage and divorce separately at each wave of the TAS study. They were asked “What do you think are the chances that you will get [married/divorced]?” and were able to respond with no chance, some chance, about 50–50, pretty likely, and it will happen to each. Respondents who reported that they had “no chance” of marriage were not asked to report expectations to divorce. This was a small portion of respondents (180 observations of 132 individuals). For analyses, they were given “no chance” of divorce as they did not expect to marry in the first place.

Dependent Variable: Transitions

At each wave of the TAS study respondents reported their current marital and cohabitation status and indicated the date they married or began living with their partner, respectively. Additionally, marriages that occurred between waves were recorded in a retrospective marital history collected on all respondents between ages 15 and 44. Unfortunately, cohabitations between waves were not recorded similarly, so I was able to only record cohabitations that respondents were in during an interview. Given that the average time between interviews was 23.89 months (analysis not shown), and the mean length of cohabitation is about 22 months (Copen, Daniels, & Mosher, 2013), there were likely cohabitations that took place between waves and either broke up before the next interview or transitioning to marriage and were captured as first marriages. Thus, it is possible that respondents who were observed cohabiting represented more stable cohabitations than average, or that those observed in marriages that actually started as cohabitation represented relationships that transitioned faster than others. This is an unfortunate limitation of these data, given the prevalence of cohabitation in young adulthood (Manning et al., 2014). It should be noted that observed marriages were limited to different-gender relationships. The question on cohabitation did not specify the gender composition of the relationship, but without information on the partner’s gender I was unable to differentiate between same- and different-gender cohabitations.

Two individuals reported the month of a transition as summer or autumn, which I recoded to months (June and September, respectively). For respondents missing the month of cohabitation but not year, June was substituted (8 respondents). Respondents missing year of a transition were not included in the analyses (n = 1 for marriage, 7 for cohabitation). Respondents who did not experience a transition during the observation period were censored (ceased being observed) at last interview. In analyses models, respondents were considered “at risk” of transitioning starting at age 18, but because some respondents were not interviewed until later, I specified delayed entry as the first interview date (StataCorp, 2017).

Control Variables

I controlled for a variety of other variables, both time-varying and invariant, to better elucidate the unique association between expectations of divorce and these role transitions. For variables that included exact dates of transitions, I measured change at the monthly level. Other variables that changed over time but not with specific dates were updated biennially at the date of the interview. Some time-invariant variables were also included.

Monthly variables.

Pregnancy and childbearing play a role in union type and timing (Lichter, Sassler, & Turner, 2014). Respondents were asked “How old were you when you first took on a parenting role / had your first child?” I generated a variable representing a change to parent in the June the year the respondent was the age they first assumed a parental role.

Educational attainment plays an important role in relationship readiness, type, and timing (Carroll et al., 2009; Cherlin, 2010a; Manning et al., 2014). Completed education was measured through a combination of variables. Respondents reported the date they graduated from high school, received a GED, or left school if they did not receive either a diploma or GED. At each wave, respondents were able to report on two college enrollment spells, including a present spell if they were currently enrolled. For spells that had ended, respondents reported if they had graduated or left for another reason, and if they had graduated the degree they had received. I constructed a monthly history of completed degrees; relatively few respondents achieved graduate or professional degrees, thus I classified degrees as less than high school, high school or GED (reference), and Bachelor or higher. Similar to education, employment, and the associated ability to provide for a family, is often seen as a prerequisite for being “ready” for marriage (Carroll et al., 2009). Much like college enrollment spells, respondents were able to report on up to five jobs at each interview. I used the dates of employment and college enrollment to construct a monthly economic engagement calendar with three categories: enrolled in education (working or not), employed (not enrolled), and neither.

Parents’ marital experiences are associated with both the expectations and attitudes young adults hold and their union behavior (Amato & DeBoer, 2001; Arocho & Kamp Dush, 2017). To ascertain a time-varying measure of the respondents’ parents’ marital status, I turned to the PSID marital histories collected on all adults ages 15–44 (Institute for Social Research, 2018). I identified the respondents’ primary caregiver in the CDS 1997 wave and collected their month of entry into first marriage, exit from first marriage, and entry into second marriage. From this, I created a time-varying indicator denoting the caregiver being never married, in first marriage, out of first marriage, and in second or higher order marriage at each month. If a respondent was observed past the parent’s last interview, the values were set to missing after that year unless the parent had died, in which case the last value was retained for the remaining TAS interviews.

Biennial variables.

Variables measured only at interviews were updated at the month of the interview and maintained that value until the next interview. These variables included the importance a respondent placed on religion, their present savings, current student loan debt, and how much they worried about their future. Religiosity was measured with the question “How important is religion to you?” with responses from 1 “not at all important” to 4 “very important.” Respondents who did not identify as religious were given a distinct code. Across the years, more than half of respondents with a religious identity responded that it was “very important,” thus I collapsed responses into: not religious, religious but less than very important, religious and very important. Savings was measured as the amount respondents held in savings or checking accounts, with responses for $0, $1-$999, and $1,000+. Similarly, respondents reported if they held student loan debt, and if so, how much, and I coded these responses in the same categories as savings. The measure of worry was the average score on a scale originally developed for the Michigan Study of Adolescent and Adult Life Transitions (Institute for Social Research, 2018). The questions for this scale included “How often do you worry that you may not have enough money to pay for things?” “How often do you feel discouraged about the future?” and “How often do you worry that you will not have a good job in the future?” and was scored 1–7 with higher scores indicating worrying more often (α = .73–.79 for analytic sample each year). A measure of overall wellbeing was derived from averaging 14 items assessing psychological, social, and emotional wellbeing (α = .87–.89); these scales came from the Midlife in the United States study (Institute for Social Research, 2018). Religion predicts union formation type and timing (Eggebeen & Dew, 2009), and savings and student loans were included as a measure of financial wellbeing, another predictor of union formation (Smock, Manning, & Porter, 2005). The measure of worry provided both an insight into respondents’ readiness for marriage (Carroll et al., 2009) and pessimism, and the wellbeing measure was meant to pick up on previously-identified associations between wellbeing and expectations of divorce (Arocho & Purtell, 2018).

Invariant variables.

Finally, I controlled respondents’ first reported race and ethnicity and the respondents’ first report of their mother’s education. Race/ethnicity was coded as NonHispanic White, NonHispanic Black, Hispanic, and NonHispanic Other. Due to the sample families’ original selection in the 1960s, present American racial diversity was not reflected in these data. Gender, race, and ethnicity are associated with marital ideals and union formation (Cherlin, 2010a; Crissey, 2005), and mother’s education was included as a proxy for background socioeconomic status, also important to consider for union formation timing and type (Cherlin, 2004; South, 2001).

Missing Data

For missing data within waves and missing waves between interviews, I used multiple imputation using chained equations to estimate the values in regression models over 20 datasets (White & Royston, 2009) separately by gender. I combined estimates from each dataset into a comprehensive estimate using the MI suite of commands in Stata15. Some researchers recommend transforming survival data into wide form (one line per individual, with each variable recorded for each time point) for imputation and then reshaping into long (person-period, with one line per observation) format for analyses (Young & Johnson, 2015). However, due to the use of month-level data, reshaping these data produced thousands of variables. It should be noted that the suggestion for using wide-form imputation was made in the context of continuous-time survival analyses. In discrete-time analyses such as those conducted here, observations are considered independent, as the object of analysis is actually each month of each respondents’ life, rather than respondents as a whole (Singer & Willett, 2003). Due to the data limitations and the nature of observations in this analysis type, I elected to impute using the long-form data. Based on best-practice recommendations for imputing survival data (White & Royston, 2009), I included the event indicator and the cumulative hazard estimate in the imputation models; I also included the values of time (Singer & Willett, 2003). I did not impute waves past the respondents’ final interview; that is, I elected to practice traditional censoring whereby respondents who dropped out of the study early or reached the end of the observation period without a transition were censored and not followed beyond that time.

Plan of Analysis

I used discrete-time event history analysis to estimate entry into first cohabitation and first marriage. Respondents’ monthly histories were used to create a person-period dataset with one line of data for each month of observation. Respondents exited all analyses after they experienced a transition or at the month of last interview. I tested various parameterizations of time (Singer & Willett, 2003) for the models (not shown) and chose to use a linear term for all analyses. Where the choice was not immediately clear from model fit changes between specifications (Singer & Willett, 2003), I tested quadratic and cubic models and results were substantively unchanged (not shown). Finally, I specified the cluster (StataCorp, 2017) estimation option for all models to adjust standard errors to account for interdependence between respondents descended from the same original 1968 PSID families.

First, I conducted a competing-risk model using a three-category outcome of censored, first marriage without cohabiting (direct marriage), and first cohabitation. I tested this model first with censored as the reference outcome, and then tested a model with married as the reference. Then, to better get at overall marriage timing in young adulthood regardless of cohabitation, I tested a binary model of transition to first marriage versus no transition. This model included a slightly larger subsample (n = 2309), as respondents who had been excluded from the previous model because of cohabitation prior to entering the study were now included.

Results

Table 1 displays the nonimputed, unweighted descriptive statistics of the subsample eligible for the competing-risk analyses: those with at least two interviews and valid dates of reported marriages or cohabitations (between ages 18 and 28, after first interview, not missing year). This represented 2,052 individuals and 102,253 monthly observations. Respondents contributed an average of nearly 50 months of observation before forming a union or being censored at their final interview, ranging from 1 to 103 months.

Table 1:

Nonimputed, Unweighted Descriptive Statistics. Observations (person months) = 102253, individuals = 2052.

	Mean/%	SD	Min	Max	Missing
Male	0.50
Age at First Interview	19.53	1.03	17.75	25.83
Age at Last Interview	24.81	2.23	19.75	28.25
Months of Observation	49.83	26.41	1	103
Age Cohabitation	22.50	1.94	18.25	27.67
Age Marriage	22.96	2.03	18.58	27.17
Invariant Variables
Race					0
Non-Hispanic White	0.41
Non-Hispanic Black	0.38
Non-Hispanic Other/Multiracial	0.10
Hispanic	0.11
Mother Education					0.02
Less HS	0.24
HS or GED	0.29
Some Col	0.25
College+	0.20
Biennial Variables
Worry	3.49	1.57	1.00	7.00	0.07
Wellbeing	13.67	2.51	4.00	18.00	0.07
Religious Importance					0.07
Not Religious	0.15
Religious, Moderately or Less Important	0.37
Religious, Very Important	0.41
Savings					0.08
None	0.22
$1–1K	0.39
$1K+	0.30
Student Loans					0.08
None	0.61
$1–1K	0.15
$1K+	0.15
Likelihood of Divorce	1.78	0.77	1.00	5.00	0.07
Likelihood of Marriage	3.87	1.05	1.00	5.00	0.07
Monthly Variables
Economic Engagement					0.09
Enrolled	0.46
Employed, Not Enrolled	0.34
Neither Employed nor Enrolled	0.11
Completed Education					0.01
Less than High School	0.07
High School or GED	0.80
College+	0.12
Parent	0.16				0.0008
Parent Marital Status					0.08
Never Married	0.11
First Marriage	0.46
Ended First Marriage	0.19
Second+	0.16

Half of the sample was male; the sample was 19.53 years old on average at the first interview (ranging 17.75 to 25.83) and 24.81 at their last interview (range 19.75–28.25). Nearly half of the sample was Non-Hispanic White (41%), with 38% Non-Hispanic Black, 10% reporting “Other” or Multiracial background but not Hispanic, and 11% with Hispanic heritage. Regarding mother’s education, 24% reported less than high school, 29% reported high school or equivalent, 25% reported some college, and 20% reported more (2% missing). For the 190 marriages observed in the model, the average age was 22.96. For the 536 observed cohabitations, it was 22.50.

In the pooled observations (person-months), the mean rating of likelihood of divorce was 1.78 and marriage was 3.87. Respondents reported a mean score of 3.49 out of 1–7 for worry and 13.67 out of 4–18 for wellbeing. For 15% of observations respondents reported no religious identity; 37% of the time they reported a religious identity but said it was only moderately important or less, and 41% of the time their religion was very important. During 22% of observations respondents reported having no savings, and 30% of the time they reported more than $1,000 in saving (8% missing). Sixty-one percent of the time respondents reported having no student loans, and 15% of the time they owed more than $1,000 (8% missing). Nearly half of the observations (46%) were of respondents enrolled in college, whereas 34% of observations were of individuals working but not enrolled, and 11% of the time they were neither. Over the course of the study 7% of observations were when respondents had no high school degree, 80% where when respondents had a high school diploma or Associate’s degree, and the remaining 12% were after respondents had received a Bachelor’s degree or more (1% missing). Respondents were themselves parents 16% of the time (<1% missing). Respondents’ parents were married in their first marriages for 46% of observations, were never married 11% of the time, had ended their first marriage 19% of the time, and had been married a second time or more the other 16% of the time (with about 8% missing).

Survival Analyses

Results of the competing-risk model are shown in Table 2. First, with no union (censored) as the reference, marital expectations positively predicted forming either type of union. A one-step increase in marital expectations predicted 51% greater risk of marriage and 15% greater risk of cohabitation compared to remaining single in a given month. Divorce expectations were negatively associated with marriage, such that a one-step increase in divorce expectations was associated with 35% lower risk of marriage compared to remaining single. Switching the reference to marriage as one’s first union, marital expectations continued to predict marriage, with a one-step increase in marital expectations associated with 34% lower risk of remaining single in a given month, though expectations for marriage did not predict marriage over cohabitation. Expectations for divorce predicted greater risk of both staying single (54%) and being first observed cohabiting (58%) over being first observed in a marriage. In the binary model estimating the hazard of marriage compared to remaining unmarried (Table 3) regardless of cohabitation, expectations for both marriage and divorce functioned in the same direction as the multinomial models. Greater expectations for marriage were associated with 82% greater odds of marriage, and greater expectations of divorce with 30% lower odds of marriage.

Table 2:

Results of Survival Analyses Predicting Competing Risks of Marriage and Cohabitation. Observations = 102,253, individuals = 2,052.

	Marriage	Cohabitation	Censored	Cohabitation
	vs. Censored		vs. Marriage
Likelihood of Divorce	0.65 *	1.03	1.54 *	1.58 *
Likelihood of Marriage	1.51 *	1.15 *	0.66 *	0.76
Parental Marital Status (ref: First Marriage)
Never Married	1.40	0.96	0.71	0.68
Ended First Marriage	1.01	1.01	0.99	1.00
Second or Higher Marriage	1.15	1.06	0.87	0.92
Mother Education (ref: High School or GED)
Less High School	0.63	0.82	1.60	1.32
Some College	0.67	0.83	1.48	1.22
College+	0.84	0.80	1.19	0.95
Race (ref: NonHispanic White)
NonHispanic Black	0.21 ***	0.86	4.79 ***	4.10 ***
NonHispanic Other/Multiracial	0.57	0.86	1.76	1.51
Hispanic	1.07	1.13	0.93	1.06
Male	1.06	0.72 *	0.94	0.67
Had a Child	3.28 ***	2.44 ***	0.30 ***	0.74
Economic Engagement (ref: Enrolled)
Employed, Not Enrolled	2.97 ***	1.77 **	0.34 ***	0.60
Neither Employed nor Enrolled	3.67 ***	1.08	0.27 ***	0.29 **
Earned Degree (ref: High School or GED)
Less High School	0.35	1.47	2.84	4.18 *
College+	2.11 **	1.01	0.47 **	0.48 *
Religious Importance (ref: Somewhat Important)
Not Important	0.93	1.16	1.08	1.25
Very Important	1.73 *	0.68 *	0.58 *	0.39 ***
Worry about the Future	0.89	1.03	1.12	1.15
Wellbeing	0.99	1.00	1.01	1.01
Savings (reference: $1–1K)
No savings	0.55	0.86	1.82	1.56
Over $1K	0.70	0.76	1.43	1.09
Student Loans (reference: $1–1K)
No loans	0.93	1.08	1.07	1.15
Over $1K	0.48 *	1.63 *	2.08 *	3.39 **
Time	1.01	1.00	0.99	1.00

Note: Relative Risk Ratios reported.

^*p < 0.05,

^**p < 0.01,

^***p < 0.001

Table 3:

Results of Survival Analyses Predicting Entrance into Marriage. Observations = 133,270, individuals = 2,309.

	Odds	Ratio
Likelihood of Divorce	0.70	**
Likelihood of Marriage	1.82	***
Parental Marital Status (ref: First Marriage)
Never Married	0.73
Ended First Marriage	0.81
Second or Higher Marriage	1.15
Mother Education (ref: High School or GED)
Less High School	0.78
Some College	0.85
College+	0.76
Race (ref: NonHispanic White)
NonHispanic Black	0.31	***
NonHispanic Other/Multiracial	0.71
Hispanic	0.86
Male	0.86
Had a Child	2.51	***
Economic Engagement (ref: Enrolled)
Employed, Not Enrolled	2.76	***
Neither Employed nor Enrolled	2.48	**
Earned Degree (ref: High School or GED)
Less High School	0.51	*
College+	2.00	***
Religious Importance (ref: Somewhat Important)
Not Important	0.84
Very Important	1.33	*
Worry about the Future	0.94
Wellbeing	0.97
Savings (reference: $1–1K)
No savings	0.69
Over $1K	1.00
Student Loans (reference: $1–1K)
No loans	0.84
Over $1K	0.73
Time	1.02	***

Note:

^*p < 0.05,

^**p < 0.01,

^***p < 0.001

Neither marital status or maternal education predicted variations in union formation. Regarding race and ethnicity, only NonHispanic Black respondents were significantly different than NonHispanic White respondents (reference), in that in each model NonHispanic Black young adults were significantly less likely to marry than NonHispanic White respondents. Males were only different than females when looking at cohabitation compared to marriage; males had 28% lower risk of cohabitation over marriage compared to females.

When respondents had children they were more likely to partner in either cohabitation or marriage, and there were not significant differences comparing the union types. Compared to those enrolled, employed respondents were more likely to partner (though not more or less likely to marry over cohabiting), and those who were neither enrolled nor employed were more likely to marry than students. Those with less than high school education were more likely to cohabit over marrying compared to those with high school education, and less likely to marry overall. Those with a college education were more likely to marry, both overall and over cohabiting. Those who felt religion was very important to them were more likely to marry and less likely to cohabit in all models. Worry, wellbeing, and amount of savings were not significantly associated with union formation in any model. When respondents carried the greatest student loans, however, they were less likely to marry compared to both remaining single and cohabiting. Finally, the overall risks of union formation in the competing risks models were not significantly associated with time. However, when looking only at marriage, respondents’ overall odds of marriage increased by about 2% for each month of age.

Conclusion

Expectations for divorce may play a role in behavior even before unions are formed. As hypothesized, greater expectations for divorce predicted significantly slower entry into marriage, even controlling for expectations for marriage and various sociodemographic characteristics associated with union formation. Furthermore, expectations for divorce predicted a lower likelihood of marriage being one’s first observed union compared to both remaining single and cohabitation. Other research in various contexts and samples has suggested that cohabitation may be preferable to risking divorce (Perelli‐Harris et al., 2017); my results supported this. This is evidence that expectations for divorce may play a role in marriage over the transition to adulthood, extending previous suggestions regarding cohabiting adults and parents to a younger, more general population (Miller et al., 2011; Waller & Peters, 2008). Contrary to my hypothesis, divorce expectations did not play a role in first cohabitation formation on its own. It may be that, given that many young adults do not initiate cohabitations with intentions to marry (Cohen & Manning, 2010), expectations for possible divorce are too distal to matter for cohabitation itself.

Limitations

Cohabitation is more common than marriage in young adulthood (Manning et al., 2014) and cohabitations are often short-lived and may be entered into multiple times (Cohen & Manning, 2010), thus the limitation to the cohabitation measure was unfortunate. This issue also clouded the interpretation of the competing-risks model: young adults first observed in marriages may have cohabited before but not been captured as such (given that the majority of recent marriages started in cohabitation [Manning et al., 2014], this is likely the case). There are likely still differences in respondents who were first observed in cohabitations versus marriages (for instance, those observed in marriages may have transitioned more quickly through premarital cohabitation than others), but a more complete measure of cohabitation would contribute to further understanding of the role of expectations for divorce in union formation, as well as contribute to our understanding of associations between cohabitation experience and attitudes that may continue to predict behaviors like divorce (Axinn & Thornton, 1992).

It should also be noted that the unions of this sample were formed at relatively young ages, especially marriage (Manning et al., 2014). However, it is not unreasonable to assume that expectations for divorce would continue to function similarly later in development. Indeed, qualitative interviews with adult cohabitors into their thirties revealed how concerns about divorce delayed transitioning into marriage (Miller et al., 2011). In addition, a more detailed measure of relationship expectations, and associated cognitions such as aspirations, would benefit future studies. Expectations are not necessarily the same as desires or aspirations (though some work on marital attitudes have suggested they are very close, [Plotnick, 2007]), and future work would do well to differentiate between expectations, aspirations, and overall values to more thoroughly understand the cognitions underpinning behavior. The current measures of expectations were also skewed; other studies have suggested that most people are optimistic towards avoiding divorce (Arocho & Purtell, 2018; Campbell, Wright, & Flores, 2012). Expectations of divorce are also only one part of expectations regarding marriage (Willoughby et al., 2015). Thus, additional measures of attitudes towards marriage readiness, appropriate context, and the importance of marriage may provide a more comprehensive picture. Furthermore, expectations for success may entail more than simply avoiding divorce, so other measures of “success” may prove useful. Finally, these data are not population representative. Although the original PSID families were sampled to represent the American population in the 1960s, the present sample, especially the young adults, do not reflect current American diversity, such as in newer immigrant families.

Contribution

These results suggest that even those who desire to marry may not, at least in part, because of feelings towards divorce (Miller et al., 2011; Waller & Peters, 2008). Efforts to promote marriage (Wood, Moore, Clarkwest, & Killewald, 2014) may do well to address the reasons for pessimism towards as-yet hypothetical unions. For instance, expectations for divorce are associated with various factors like poor education and employment (Arocho & Purtell, 2018) parental divorce (Boyer-Pennington et al., 2001), and several aspects of wellbeing (Arocho & Purtell, 2018). Helping young adults to address issues like education, mental health, or relationship skills they may feel they are lacking may help individuals feel more positive towards their relationship futures, and thus feel safer in marrying if that is their desire.