Abstract

Some conservative groups argue that allowing same-sex couples to marry reduces the value of marriage to opposite-sex couples. This article examines how changes in U.S. legal recognition laws occurring between 1995 and 2010 designed to include same-sex couples have altered marriage rates in the United States. Using a difference-in-differences strategy that compares how marriage rates change after legal recognition in U.S. states that alter legal recognition versus states that do not, I find no evidence that allowing same-sex couples to marry reduces the opposite-sex marriage rate. Although the opposite-sex marriage rate is unaffected by same-sex couples marrying, it decreases when domestic partnerships are available to opposite-sex couples.

Introduction

Does the value of an institution depend on who else participates in that institution? Some conservative groups argue that this is the case with marriage and that allowing same-sex couples to marry reduces the value of marriage to opposite-sex couples.1 Marriage is of interest because it serves as a social and legal contract that facilitates family decision-making while providing legal and cultural safeguards.2 In economic models of marriage, people choose to marry when the benefits of being married outweigh the costs. As a result, if marriage becomes less valuable, marriage rates will decline. In this article, I analyze the effects of changing legal recognition laws on marriage rates in the United States.

The potential effect on opposite-sex marriage of allowing same-sex couples to marry is theoretically ambiguous. Allowing same-sex couples to marry could lower opposite-sex marriage rates if part of the value of marriage comes from its exclusivity. Alternatively, allowing same-sex couples to marry could increase opposite-sex marriage rates by increasing interest in the institution of marriage or by reducing the pressure on employers to provide marriage-like benefits to cohabiting couples (Rauch 2004; Trandafir 2013).

Few studies have examined how allowing same-sex couples to marry affects marriage rates. Langbein and Yost (2009) used data on the stock of marriages in the United States and found that the number of married people in a state does not change after same-sex couples are allowed to marry. However, the stock of marriages may be slow to change even if marriage rates change immediately. Another issue is that Langbein and Yost used data through 2004, which is the first year same-sex couples could marry in a U.S. state. Thus, the effect on marriage is identified by very few state-year observations.

Trandafir (2013) studied the effects of a Netherlands law that allows same-sex couples to marry and a separate law that allows all couples to enter into registered partnerships, which provide similar benefits to marriage. Trandafir found suggestive evidence that marriage rates rise after all couples can enter into registered partnerships but fall after same-sex couples can marry. Although women are less likely to be married after same-sex couples can marry, Trandafir concluded the experience of the Netherlands suggests no major effects of changing legal recognition laws on overall marriage rates given that controlling for heterogeneity greatly reduces the coefficient. Because both of these laws changed for the Netherlands as a whole with only a few years in between, Trandafir had difficulties disentangling the effects of the two laws. Furthermore, people in the United States, who are culturally very different than residents of the Netherlands, may react differently.

The advantage of studying legal recognition changes in the United States is that it provides a variety of policy experiments happening at different points in time from 1995 to 2010. In some U.S. states, same-sex couples are allowed to marry; in other states, they are allowed to enter into newly created forms of legal unions instead of marriage. These new forms of unions, generally called civil unions or domestic partnerships, are intended to mimic marriage but under a different name. In some states, such as Oregon and Vermont, only same-sex couples can enter into the new forms of unions; in other places, such as Maine and the District of Columbia, all couples can enter into the nonmarriage legal recognition.

Opening new forms of legal recognition to opposite-sex couples could result in lower marriage rates if some couples prefer an alternate form of legal recognition to marriage. An issue with this is that domestic partnerships do not offer the federal benefits of marriage, so people entering into domestic partnerships instead of marriage have fewer legal benefits. Thus, increasing the benefits of domestic partnerships may strengthen these couples’ legal and financial standing, but it might also lead more couples to choose to enter into domestic partnerships instead of marriage.

To analyze the effects of the changes in legal recognition, I use two data sources. First, I construct a state-level panel data set that contains marriage rates, legal changes, and other state characteristics. The advantages of using this data set are that the marriage rates come directly from the states and that they account for every marriage occurring in the state in a given year. As a result, I can consider how these laws affect both overall marriage rates and opposite-sex marriage rates.

However, there are a few disadvantages of using this type of data. The first is that people sometimes marry in states other than where they reside, which would confound any estimation strategy using marriage rates in a state. To the extent that marrying outside the state of residence is random, this would bias the estimates toward zero. Second, testing for heterogeneous responses is difficult without knowing individuals’ characteristics. Finally, the legal changes could affect the stock of marriages without affecting the flow if couples exit marriage after the legal changes. This would be missed if only data on marriage rates were used. To address these issues, I use individual-level data from the Current Population Survey (CPS) to examine how the stock of married couples changes in response to legal recognition laws.

With both data sets, I estimate difference-in-differences models as well as models with flexible time effects, which allow the effects of legal recognition changes to vary over time. Allowing the effects of these laws to vary over time is important for several reasons. First, because marriage decisions are typically made years in advance, the effects of these laws might not be seen immediately. Second, I can test for effects before changes in legal recognition, which allows an examination of whether differing time trends before the legal changes are a concern and whether people respond after the laws are passed but before they are enacted. Finally, the number of same-sex couples marrying is likely to be at its highest in the first few years of eligibility because of pent-up demand. Time-flexible specifications can allow for a comparison of immediate effects with longer-run effects.

I find that allowing same-sex couples to marry increases the overall marriage rate, but this increase appears to be driven entirely by same-sex couples marrying. Regardless of the identification strategy used, I find no evidence that allowing same-sex couples to marry has altered marriage rates for opposite-sex couples. Opposite-sex couples do, however, take advantage of the new forms of legal recognition when available. Marriage rates fall by about 5 % to 10 % when nonmarriage legal recognition is available to opposite-sex couples. These results are robust to a number of specifications, and I find no evidence that national marriage rates were affected after 2004, the year that Massachusetts became the first state to allow same-sex couples to marry.

Changes in Legal Recognition

Background

Table 1 shows state changes in legal recognition for couples classified into three categories: (1) laws giving same-sex couples the right to marry; (2) laws allowing same-sex couples but not opposite-sex couples to enter into domestic partnerships or civil unions3; and (3) laws allowing both same-sex couples and opposite-sex couples the right to enter into either domestic partnerships or civil unions.4 For ease of discourse, I will refer to all new forms of unions as domestic partnerships, given that there are no systematic differences between civil unions and domestic partnerships.

Although the rights granted to couples through these different types of unions vary by state, these new forms of legal recognition are designed to provide the same state-level benefits as marriage. Common rights include the ability to inherit a partner’s estate in the absence of a will, immunity from testifying against a partner in court, hospital visitation rights, family leave for a sick or dying partner, and the right for partners to share a nursing home room. States generally require fully insured employers to provide health insurance to employees’ partners for couples in these new forms of unions if the employers provide health insurance to married opposite-sex spouses.5

The 1996 Defense of Marriage Act (DOMA) denied all federal benefits of marriage to same-sex couples until the Supreme Court ruled that it was unconstitutional. Because this study uses data from 1995 to 2010, the married same-sex couples studied do not have access to federal benefits of marriage. Similarly, because domestic partnerships and civil unions are recognized only at the state levels, same-sex and opposite-sex couples taking advantage of them do not have federal benefits. The federal benefits of marriage include Social Security benefits for surviving spouses; the ability to file income taxes jointly, which may reduce the overall tax rate the couple faces; no estate taxes upon inheriting a deceased spouse’s assets; and the ability to petition for a spouse to immigrate to the United States.6 Because the Supreme Court overturned DOMA, same-sex couples are now able to experience all the federal benefits of marriage.

Although the fact that these new unions are not recognized by the federal government may hurt same-sex couples who cannot always enter into marriage, it is a possible advantage for many opposite-sex couples. Widows and widowers are eligible to receive the Social Security benefits their spouses would have received if they don’t remarry by the age of 60. Thus, civil unions and domestic partnerships can provide opposite-sex couples with state-level protection while not jeopardizing their Social Security survivor benefits. This may induce older widows and widowers to choose domestic partnerships instead of marriage. Marriage rates are, however, driven by first marriages by young people. According to Survey of Income and Program Participation (SIPP) data, 91.3 % of men and 93.3 % of women who married in 2008 were younger than 55 (Kreider and Ellis 2011). Although the data on marriage rates will not allow for examining heterogeneity based on age, one would expect the effects coming from older people to be relatively small because marriage rates are so low for older people.

Conceptual Framework

Theoretical Impact on Marriage Rates

The effect of allowing same-sex couples to marry is unclear ex ante. Allowing same-sex couples to marry could lower the value of marriage for opposite-sex couples if it severs the link between marriage and childbearing or if it reduces any value of marriage that comes from its exclusivity (Kurtz 2004). Reducing the value of marriage would induce couples on the margin to choose to remain unmarried instead of entering into marriage, which would result in lower opposite-sex marriage rates. On the other hand, some opposite-sex couples may value marriage more when marriage is available to all couples if they believe that excluding couples makes marriage a discriminatory institution. Additionally, allowing same-sex couples to marry could increase the value of marriage to opposite-sex couples by increasing interest in the institution of marriage or by reducing pressure on employers to provide marriage-like benefits to cohabiting couples (Rauch 2004; Trandafir 2013). Increasing the value of marriage would induce marginal couples to marry who would not have married otherwise, which would increase opposite-sex marriage rates.

The effect of allowing same-sex couples but not opposite-sex couples to enter into domestic partnerships is also theoretically ambiguous. If opposite-sex couples in these states view domestic partnerships as equivalent to marriage, then any value of marriage coming from the exclusivity of domestic partnerships would be diminished. On the other hand, some opposite-sex couples could see the value of marriage rise in ways similar to when same-sex couples are allowed to marry.

Opening new forms of legal recognition to opposite-sex couples could result in lower marriage rates if some couples prefer an alternate form of legal recognition. This could occur for people who are not religious and believe that marriage has religious meaning. People may also believe that a domestic partnership has lower dissolution costs, which could be the case if a failed domestic partnership is less emotionally costly than a failed marriage. This would make entering into a domestic partnership less risky than marriage. Finally, states passing domestic partnership legislation may make firms more likely to offer partnership benefits for couples who are not in any form of legal recognition because domestic partnership legislation would require firms to alter their benefit programs to include unmarried couples. Thus, even if couples do not officially register as domestic partners, they may have more benefits available to them and may therefore be less likely to marry.

I would expect any changes in marriage rates to lead to eventual changes in marriage stocks. However, marriage stocks would be slow to change because marriage stocks are already high. Changes to marriage stocks may not be detectable for many years. Marriage stocks could also change if people are more or less likely to exit marriage after any legal changes for any of the reasons outlined earlier.

The effects of changing legal recognition may vary based on demographic characteristics because certain groups of people tend to be more supportive of providing legal recognition to same-sex couples than others. For example, young people tend to be more supportive of allowing same-sex couples to marry and to enter into other forms of legal recognition (Jones 2013). Likewise, support for allowing same-sex couples to marry rises with education, and Democrats tend to be more supportive than Republicans (Pew Research Center 2013). One might expect the effects of allowing same-sex couples to marry to have a positive impact on groups that tend to be more supportive of marriage and a negative impact on groups that are less supportive of marriage. Similarly, one might expect people who are not religious to be more likely to choose to enter into domestic partnerships instead of marriage. Testing this is not possible with the marriage rate data, but when I consider marriage stocks, I will be able to test for differences based on education and age.

Welfare Implications

If allowing same-sex couples to marry or to enter into other forms of legal recognition reduces the value of marriage for opposite-sex couples, opposite-sex couples would be worse off as a result of these laws. Same-sex couples would be better off, so the overall welfare impacts would depend on the size of the relative impacts on each group.

Allowing opposite-sex couples to enter into the new forms of unions weakly improves their welfare because they can still enter into marriage but now have a new option as well. These nonmarriage forms of unions provide fewer rights than marriage, meaning that people are choosing to enter into legally inferior unions. This might suggest that nonmarriage legal unions should be strengthened; however, this may result in even more people shifting from marriage into nonmarriage recognition.

Data Sources and Identification Strategy

Data

To examine the impact of legal recognition changes on marriage, one can look at either stock measures or flow measures. The stock measure is the total number of marriages, and the flow measure is the number of people entering into marriage. I construct a state-level panel to examine the flows of marriage and use the March CPS to examine the stocks of marriages.

The data containing the marriage rate per 1,000 individuals for each state in a given year come from the Centers for Disease Control and Prevention (CDC) for 1995 to 2010. The main results use data from 1995 to provide a few years of data before the earliest law is passed. Most laws, however, were not changed until the 2000s. Later I will verify that the results are robust to including only the years 2000 to 2010.

As is common in the literature, I use the log of state-level marriage rates, which allows for interpreting the coefficients as percentage changes in marriage rates.7 All states have reported marriage rates for all years except for Oklahoma and Louisiana, which did not report marriage rates for a few of the years studied.

Marriage rates from the CDC are formed using all marriages in a given state and year. For states that allow same-sex couples to marry, I obtain the number of same-sex marriages occurring in a year from the state health departments, which keep data on same-sex marriages but do not report these data to the CDC. I then subtract this number from the total number of marriages occurring to calculate the opposite-sex marriage rate.8

I supplement the data on marriage rates with various state-level controls calculated using the 1995 to 2010 March CPS. For each state during each year of the data, I calculate the percentage of people aged 25 and older who have a high school diploma, the percentage who have completed some college, and the percentage who have completed college. I also calculate the percentage of people in the labor force who are unemployed. I control for the percentage of people in three broad age groups: 21–40, 41–60, and older than 60. Additionally, I calculate the percentages of people who are white and black, and the percentage of people who are female.

One possible concern is that states may change their definitions of legal recognition as a result of shifting attitudes toward the gay and lesbian community. If these attitude changes are correlated with changes in the value of marriage, the estimation strategy would falsely attribute the effects of changing attitudes to providing legal recognition for same-sex couples. To account for this, I control for the percentage of the state population that voted for the Democratic candidate in the previous presidential election, given that Democrats have tended to be more supportive of providing legal recognition to same-sex couples than Republicans. Although this measure is coarse, it serves as a rough proxy for changing attitudes.

The descriptive statistics are shown in Table 2. States that change their definitions of legal recognition look demographically similar to states that do not. Of the control variables, only the percentage of people voting for the Democratic candidate in the preceding presidential election appears to be statistically different. States that change their definitions of legal recognition have a higher percentage of people voting for Democratic presidential candidates. I control for demographic characteristics in certain specifications to make sure that changes in demographic characteristics are not driving any of the results.

In addition to using the March CPS to account for demographic changes in the construction of the state-level panel data set, I also use the March CPS from 1995 to 2011 to examine the stock of marriages. With this data set, I control for race, gender, and a cubic in age. I cannot identify the same-sex couples who enter into marriage in the CPS because the CPS codes all same-sex couples as being unmarried partners, so I focus only on the stock of opposite-sex marriages.9

Identification Strategy

Figure 1 shows national marriage rates for the study period. Because the downward trend in marriage rates during this period started in the early 1980s, one cannot simply compare what happens in a state after legal recognition. I account for the national time trend by including a control group that would be subject to the same time trend.

I estimate simple difference-in-differences models as well as models that allow the effects of legal recognition changes to vary over time. An issue with using time-flexible models is that many of these laws were passed only recently, and many states have not expanded their definitions of legal recognition, which results in large standard errors. By examining the more aggregated difference-in-differences estimator, I can better identify the average effects over time even though I no longer have estimates at each point in time.

I estimate two main equations. The first provides the difference-in-differences estimator:
formula
(1)
where y is the log of the marriage rate per 1,000 people, s indexes the state, t indexes the year, is a vector of time effects, v is a vector of state effects, X is a vector with the average demographic characteristics for each state in a given year, is an indicator variable equal to 1 in a state after a law of type j was passed, and is the state-level error term. Again, there are three potential types of laws: (1) those allowing same-sex couples to marry, (2) those allowing only same-sex couples to enter into new forms of recognition, and (3) those allowing all couples to enter into new forms of recognition. The coefficients provide the effect of legal recognition changes averaged over time.
I also want to be able to distinguish immediate effects of the laws from later effects. To do this, I estimate a model of the following form:
formula
(2)
where is an indicator variable equal to 1 in the kth period after a law of type j was passed, and the other variables are defined as before. The laws were passed in the year k = 0. can be interpreted as the effect of a type j law change k years after its passage.10 Estimating this model requires more from the data than the difference-in-differences model. For high values of k, only one state identifies the effects in some cases; thus, these estimates should be interpreted with caution, especially for high k. Also, standard errors will be too high to distinguish most of the coefficients from zero. I graph the coefficients to provide an idea of how the effects may vary over time.

Results

Marriage Flows

The results from estimating Eq. (1) with the state-level data are shown in Table 3. In the first two specifications, the dependent variable is the log of the overall marriage rate. In the next two specifications, the dependent variable is the log of the opposite-sex marriage rate. The second and fourth specifications control for demographic characteristics of the states, but the first and third do not. The top panel reports the unweighted estimates, and the bottom panel reports the estimates weighted by population.

The unweighted estimate on marriage in column 1 of Table 3 suggests that allowing same-sex couples to marry has increased the overall marriage rate by about 13.7 %. Controlling for demographic characteristics in column 2 causes the unweighted coefficient to decrease by less than 1 percentage point to 12.8 %. With the mean of marriages per 1,000 people being 8.98, these estimates suggest about 1.2 additional marriages per 1,000 people per year. Weighting the estimates by population size causes them to fall to 0.103 and 0.076 in specifications 1 and 2, respectively, but they remain significantly different from zero. The weighted estimates suggest an increase of 0.7 to 0.9 marriages per 1,000 people per year.

One must be careful in interpreting these results. States do not have residency requirements for marriage, and some same-sex couples in states where same-sex couples cannot marry do travel to states where they can marry.11 One would not expect marriage rates to increase by this much nationally if same-sex couples were allowed to marry across all states. Similarly, one would not expect the increases to be this high as more states allow same-sex couples to marry.

The coefficient on marriage for same-sex couples in column 3 (Table 3), where the dependent variable is the log of the opposite-sex marriage rate, is statistically indistinguishable from zero. Controlling for demographics in column 4 changes the coefficient very little, as does weighting the estimates. The point estimates range from –0.007 to 0.021. Although the statistical power is limited because of the small treated sample sizes, these results do not suggest any evidence that allowing same-sex couples to marry has had an effect on opposite-sex marriage rates.

Because the marriage rates are defined per 1,000 people, the estimates on opposite-sex marriage would all be biased downward if allowing same-sex couples to enter into legal recognition resulted in same-sex couples moving into a state. Dillender (2013) considered migration of same-sex couples and found no effects of the laws on the numbers of same-sex couples in a state.12

The results presented in Table 3 suggest that allowing opposite-sex couples to enter into domestic partnerships decreases marriage rates between 9 % and 11 %. The coefficients on domestic partnerships for all couples are significant at the 10 % level in three of the four specifications. This represents a decrease of roughly 1 marriage per 1,000 people per year. The weighted point estimates range from –0.095 to –0.130 and are significant in all four specifications. These results suggest that some opposite-sex couples enter into new forms of unions when they are available instead of entering into marriage. This is important for two reasons. First, domestic partnerships are legally inferior to marriage because domestic partnerships do not include any federal benefits. Second, these results suggest that opposite-sex couples may enter into marriage in the absence of alternate recognition when they would really prefer a nonmarriage form of legal recognition.

The coefficients on domestic partnerships for same-sex couples only are slightly positive and significant in one of the specifications. Weighting the estimates causes the estimates of the effect on domestic partnerships for all couples to rise, suggesting that allowing only same-sex couples to enter into domestic partnerships may have a positive effect; these results, however, are not robust to controlling for demographic characteristics or to the robustness checks presented later.

Figure 2 shows the unweighted coefficients from estimating Eq. (2), which allows for time-varying effects of the law changes. Note that the coefficients are not cumulative and that the size of all the effects are relative to more than six years before a law is passed.13 Likely because of the large standard errors, all coefficients are statistically indistinguishable from zero, with the exception of when same-sex marriages are included. The coefficients on overall marriage rates after same-sex couples can marry are statistically different than zero but not from each other.

Allowing for time-varying effects reveals that the number of same-sex couples marrying is at its highest in the first year that same-sex couples can marry and then decreases in the following years. After two years, the increase in marriage rates remains at about 9 %. The coefficients on domestic partnerships for all couples oscillate before the laws are passed, and they begin to fall after the law is passed. When the dependent variable is the opposite-sex marriage rate, there is no evidence that the coefficients on allowing same-sex couples to marry are different from zero or that they vary over time. The same is true for the effect of domestic partnerships for same-sex couples only. The coefficients appear to spike for years 11 and 12, but these are each identified from one observation each. Although drawing strong conclusions from these coefficients is difficult, there appears to be no evidence that allowing same-sex couples to marry or to enter into domestic partnerships has had a negative effect on opposite-sex marriage.

Two caveats are important to note regarding the results presented in this section. First, the standard errors are too large to rule out positive or negative effects of allowing same-sex couples to marry on marriage rates. However, I use a variety of tests to look for any evidence to support the claim that same-sex couples marrying reduces the number of opposite-sex couples marrying, and I consistently find no evidence of it. Second, the results shown here are the immediate effects of changing legal recognition. To the extent that marriage rates or the value of marriage may change gradually over time as a result of these laws, I will not be able to detect this.

Marriage Stocks

I next use data from the March CPS to examine the stock of marriages. There are three advantages of using the CPS data. First, I can test for heterogeneous responses because people’s marital status is directly matched to their demographic characteristics. Second, I can address another potential concern of the earlier analysis that stems from the fact that many people do not get married in their state of residence—perhaps because certain states are marriage destinations or because people want to marry in the state where their family lives. If seeing same-sex couples marrying really does lessen the value of marriage, one would technically expect a decrease in the number of people living in the state who get married and not necessarily a change in the number of marriages that take place in the state. Third, marriage stocks may change even if rates do not, if people are more or less likely to exit marriage after legal recognition. Data on marriage stocks allow for an examination of this possibility. A limitation of these data, however, is that I can look only at opposite-sex marriages because of the coding procedure of the CPS.

Columns 1 and 2 of Table 4 show the difference-in-differences results. The sample in these two columns contains everyone age 16 and older. The first column does not control for demographics; the second column does. In both specifications, all coefficients are indistinguishable from zero. Approximately 56 % of the sample is married; thus, an estimate of 0.005 on allowing same-sex couples to marry would indicate a 0.009 % increase in the likelihood of being married.

As stated earlier, one might expect older people to be negatively affected by same-sex couples marrying because they tend to be less supportive of providing legal recognition to same-sex couples. One might also expect older opposite-sex couples to be more likely to enter into domestic partnerships because marrying would cause widows and widowers to lose their Social Security survivor benefits. Columns 3 through 8 contain results for the sample, restricted to include various age groups. The restrictions are people younger than age 30, people at least 30 but younger than 60, and people at least 60 years old. The results provide no evidence of heterogeneous effects based on age. In all cases, the coefficients are insignificantly different from zero.

People of different educational levels may be affected differently as well. In columns 9 and 10 of Table 4, the sample is restricted to those who have attended at least some college. In columns 11 and 12, the sample is restricted to those who have not attended college. Only the coefficient on domestic partnerships for all couples is significant in any of the specifications. The coefficient of –0.015 suggests a –0.027 % decline in the likelihood of being married for people who have attended some college. This could indicate that more-educated people are more likely to enter into domestic partnerships, perhaps because education is negatively correlated with religion (Glaeser and Sacerdote 2008). The significance of the coefficient, however, is not robust to controlling for demographic characteristics.

Figure 3 plots the estimates from the time-flexible models. The point estimates appear to rise slightly after same-sex couples were allowed to marry, but there seems to be no evidence that the stock of marriages fall after some time, which the death-of-marriage argument would imply.

The insignificant coefficients on domestic partnerships for all couples may seem at odds with the estimates from the previous section that suggest that allowing opposite-sex couples to enter into alternate forms of recognition lowers the opposite-sex marriage rate. Two factors would minimize the estimated effects of domestic partnerships for all couples on the stocks of opposite-sex married couples from the CPS. First, the stock of married people is already high, so even if changes in flow measures take place immediately, the stock measures would be slow to change. Second, it is not clear how people who enter into domestic partnerships would report their relationship status in the CPS because the only two relationship statuses are unmarried partner and spouse. Domestic partners who reported that they are spouses would lead to a finding of no effect of extending new forms of legal recognition to opposite-sex couples.

Robustness

I now verify the robustness of the main results to various specifications as well as test for national effects of allowing same-sex couples to marry. I focus on the opposite-sex marriage rates results because rates would change before stocks and because I did not find evidence of changes in marriage stocks. However, the results regarding marriage stocks are similar to the previous estimates as well.

Are There National Effects of Allowing Same-Sex Couples to Marry?

The identification strategy described earlier makes the key assumption that legal changes will impact behavior only in states with laws allowing same-sex marriage. This may be more reasonable in relation to domestic partnerships than same-sex marriage. With domestic partnerships, opposite-sex couples may choose not to enter into marriage and instead take up this new type of legal union only when it is available to them, suggesting that examining state variation should be sufficient for attaining accurate estimates of the effect of domestic partnerships. With same-sex marriage, however, this may not be the case. Perhaps same-sex marriage anywhere affects the value of marriage and thus marriage rates everywhere. I cannot identify these types of effects by exploring state variation.

To consider the idea that same-sex marriage in any state may have national ramifications, I look at state trends in marriage rates over the last 15 years. If national marriage rates suddenly drop after same-sex couples begin marrying in various states, allowing same-sex couples could have national ramifications, implying that the identification strategy used earlier is flawed. The solid line in the top graph of Fig. 4 shows the year coefficients in Eq. (1) estimated without controlling for the passage of the laws but with the controls described earlier. The dashed line shows how these coefficients differ from the previous year. The solid line mirrors the shape of the national rates shown earlier. The dashed line hovers at slightly below zero for most of the period. The bottom graph in Fig. 4 shows the equivalent for opposite-sex marriage rates. In both figures, the trend seems not to have changed when Massachusetts began allowing same-sex couples to marry in 2004. Marriage rates continued to fall after Massachusetts began allowing same-sex couples to marry but at a similar rate as before. From 2008 to 2010, opposite-sex marriage rates have actually risen nationally, after demographic changes are accounted for. Although examining trends can provide no definitive evidence that allowing same-sex couples to marry has no national ramifications, these results suggest that allowing same-sex couples to marry has not drastically altered national-level marriage rates.14

Dropping Observations Before 2000

The main results use data from 1995 onward. I start with 1995 because that milestone marks a few years before the earliest law is passed. Most laws, however, were not changed until the 2000s, which makes the pretreatment period very long for several of the states studied. In columns 1 and 2 of Tables 5 and 6, I drop all years before 2000. Table 5 shows the unweighted estimates, and Table 6 shows the weighted estimates.

Regardless of whether demographic controls are included, the coefficients are statistically indistinguishable from the previous coefficients. There remains no evidence that allowing same-sex couples to marry results in the death of marriage, but there is still evidence that some marriage rates fall when opposite-sex couples can enter into domestic partnerships.

Unobserved Changes Over Time

A second key assumption is that in the absence of legal recognition changes, marriage rates would be trending similarly in states that alter legal recognition and states that do not. The identification strategy controls for state heterogeneity that is fixed over time, but a potential concern is that states offering legal recognition may be changing in unobserved ways differently from states that do not offer legal recognition and that these unobserved changes confound the estimation strategy. In this section, I verify the robustness of the results to two additional ways to account for unobserved heterogeneity that changes over time. The first involves greater care in choosing the control group. The second allows states that alter legal recognition to have different time trends than other states.

Choice of Control Group

Legal recognition can be extended to same-sex couples only in states without state-constitutional bans on legal recognition. Thus, states without bans on legal recognition might prove to be a better control group than all states without legal recognition. Columns 3 and 4 of Tables 5 and 6 replicate the results, using states that have neither legal recognition for same-sex couples nor state-constitutional bans on same-sex marriage as the control group.15

One would be concerned that unobserved state trends were confounding the estimation strategy if the results changed after choosing a more narrowly defined control group. All the coefficients are similar to the original estimates. This exercise indicates the choice of using all of the nontreatment states as the control group is not driving the results.16

State-Specific Time Trends

I next supplement Eq. (1) with linear state-specific time trends for those states that extend legal recognition. This means that identification comes from how marriage rates change apart from the state-specific trends as well as from national trends after legal recognition is extended. The new estimating equation is
formula
(3)
where M is the linear time trend for state s, and the other variables are defined as before.17

The estimates for the main coefficients and for the state-specific linear time trends are shown in columns 5 and 6 of Tables 5 and 6. None of the states that allow same-sex couples to enter into marriage have a time trend that is statistically different from the national time trend. As with the original estimates, there remains no evidence that allowing same-sex couples to marry results in opposite-sex couples marrying less.

Two states that have passed domestic partnerships for all couples have time trends that appear to differ from the national time trend in both the specifications with and those without demographic controls: Nevada has a negative linear time trend, whereas Maine has a positive time trend. The coefficients on domestic partnerships for all couples do appear to fall slightly when these state-specific time trends are included in the estimating equation. However, they are statistically indistinguishable from the previous estimates and are still statistically different from zero.

Of the states that allow only same-sex couples to enter into new forms of recognition, California and Oregon have trends that are statistically different from the national time trend: they are both positive compared with the national time trend, meaning that marriage rates in these states were rising relative to the rest of the nation before they passed laws allowing domestic partnerships. Accounting for state-specific time-trends, however, still yields coefficients on domestic partnerships for all couples that are statistically indistinguishable from zero.

Conclusion

There has been much debate about what allowing same-sex couples to marry will do to the institution of marriage. This article considers several possible avenues for how the legal changes that occurred during the first decade of the twenty-first century could have affected marriage. I find that allowing same-sex couples to marry increases overall marriage rates and that the effect on marriage rates is highest for the first few years after same-sex couples are allowed to marry. This increase is accounted for entirely by same-sex couples marrying. I find no effect on opposite-sex marriage rates of allowing same-sex couples to marry, which suggests that allowing same-sex couples the right to marry does not affect the value of marriage for opposite-sex couples. This is inconsistent with the end-of-marriage argument.

The evidence does suggest, however, that allowing opposite-sex couples to enter into new forms of legal recognition decreases marriage rates. Thus, in the absence of domestic partnerships, many opposite-sex couples may enter into marriage even though they would actually rather enter into nonmarriage legal recognition. Strengthening these domestic partnerships may make opposite-sex couples better off, on average; however, strengthening the partnerships would also likely induce more people to enter into the partnerships instead of marriage.

Acknowledgments

This work was completed as part of my dissertation at the University of Texas at Austin. I thank the Editors, multiple anonymous referees, Jason Abrevaya, Sandra Black, Daniel Hamermesh, Carolyn Heinrich, Gerald Oettinger, and Stephen Trejo for helpful comments.

Notes

1

For example, in June 2011, then–presidential candidate Rick Santorum stated that allowing same-sex couples to marry would “cheapen marriage and make it into something less valuable” (The Des Moines Register2011). In 2004, James Dobson stated “[Gay people] want to destroy the institution of marriage. [Same-sex marriage] will destroy marriage” (Snyder 2004). The end-of-marriage argument was largely the rationale behind Proposition 8, the California state constitutional amendment that restricted marriage to a union between a man and a woman.

2

Much of the work on marriage and economics stems from Becker (1973, 1974).

3

A few states in the third category allow opposite-sex couples to enter into civil unions and domestic partnerships if at least one member of the couple is at least 62 years old. Results are robust to the inclusion of the laws separately. I combine the laws because the coefficients on the two types of laws are similar if I estimate the effects separately, likely because marriage rates are driven by young people.

4

Colorado allows people to designate beneficiaries. Because these types of unions do not imply a romantic relationship—that is, any two unmarried people can enter into designated beneficiary agreements, including friends and siblings—and do not offer most of the benefits of marriage, I do not code Colorado as providing alternate recognition. All results are robust to the omission of Colorado or the estimation of a separate coefficient for the effect of designated beneficiary agreements. As of May 2013, Colorado offers more comprehensive civil unions exclusively to same-sex couples.

5

For a detailed example of a domestic partnership law, see the American Civil Liberty Union’s guide to civil unions in Illinois, which is available online (http://civilunions.aclu-il.org/).

6

For a complete listing of federal benefits of marriage, see Shah (2011).

7

For examples, see Bitler et al. (2004) and Brien et al. (2004). The results are not sensitive to this specification choice; results are similar if I use the marriage rates.

8

Because the District of Columbia does not keep statistics on the number of same-sex marriages, I omit it from the analysis of opposite-sex marriages after same-sex couples can marry there.

9

I omit same-sex couples from the sample as well as any couples whose gender or marital status changed. Before 2010, the CPS changed the sex of the spouse if two people of the same sex reported being married. Beginning in 2010, the CPS changed the marital status. The results are very similar if I do not try to account for same-sex couples.

10

Wolfers (2006) used a similar econometric model to study the effects of divorce laws.

11

These marriages would typically not be legally recognized in non–same-sex marriage states because of DOMA.

12

The results are very similar if the dependent variable is adjusted so that it no longer accounts for population and is instead the log of marriage rates in a state and year.

13

These estimates do not control for demographic characteristics. The graph looks similar when demographic controls are included.

14

As discussed earlier, one might expect different responses for people with different political attitudes. I tested for different reactions to the Massachusetts ruling for more-liberal and more-conservative states as measured by the percentage of the state population that voted for George W. Bush in 2004, which is the year when Massachusetts began allowing same-sex couples to marry and when one of the main issues in the presidential election was a federal Constitutional ban on allowing same-sex couples to marry. Bush supported the ban, while his opponent, John Kerry, did not. I found no evidence of differences.

15

The new set of control states is Delaware, Illinois, Indiana, Minnesota, New Mexico, New York, North Carolina, Pennsylvania, Rhode Island, West Virginia, and Wyoming.

16

An alternate method of choosing the control group is to use the synthetic control method from Abadie et al. (2010), which selects control groups that do not violate the parallel trends assumption. In results not shown (but available upon request), I implemented the synthetic control method for each state that has changed its legal recognition laws. The results from the synthetic control method were consistent with the results presented throughout this article. I found no evidence that allowing same-sex couples to marry reduces the opposite-sex marriage rates, but I did find evidence that marriage rates fall when opposite-sex couples can enter into the new forms of recognition.

17

Similar strategies have been used by Besley and Burgess (2004), Bitler et al. (2004), and others.

References

Abadie, A., Diamond, A., & Hainmueller, J. (
2010
).
Synthetic control methods for comparative case studies: Estimating the effect of California’s Tobacco Control Program
.
Journal of the American Statistical Association
,
105
,
493
505
. 10.1198/jasa.2009.ap08746
Becker, G. S. (
1973
).
A theory of marriage: Part I
.
Journal of Political Economy
,
81
,
813
846
. 10.1086/260084
Becker, G. S. (
1974
).
A theory of marriage: Part II
.
Journal of Political Economy
,
82
,
S11
S26
. 10.1086/260287
Besley, T., & Burgess, R. (
2004
).
Can labor regulation hinder economic performance? Evidence from India
.
Quarterly Journal of Economics
,
119
,
91
134
. 10.1162/003355304772839533
Bitler, M. P., Gelbach, J. B., Hoynes, H. W., & Zavodny, M. (
2004
).
The impact of welfare reform on marriage and divorce
.
Demography
,
41
,
213
236
. 10.1353/dem.2004.0011
Brien, M. J., Dickert-Conlin, S., & Weaver, D. A. (
2004
).
Widows waiting to wed? (Re)Marriage and economic incentives in Social Security widow benefits
.
The Journal of Human Resources
,
39
,
585
623
. 10.2307/3558989
Dillender, M. (
2013
).
Essays on health insurance and the family
(Unpublished doctoral dissertation).
Department of Economics, University of Texas
,
Austin
.
Glaeser, E. L., & Sacerdote, B. I. (
2008
).
Education and religion
.
Journal of Human Capital
,
2
,
188
215
. 10.1086/590413
Jones, J. M. (
2013
,
May
13
).
Same-sex marriage support solidifies above 50% in U.S
.
Gallup Politics.
Retrieved from http://www.gallup.com/poll/162398/sex-marriage-support-solidifies-above.aspx
Kreider, R. M., & Ellis, R. (
2011
).
Number, timing, and duration of marriages and divorces: 2009
(Current Population Reports P70-125).
Washington, DC
:
U.S. Census Bureau
.
Langbein, L., & Yost, M. (
2009
).
Same-sex marriage and negative externalities
.
Social Science Quarterly
,
90
,
292
308
. 10.1111/j.1540-6237.2009.00618.x
Pew Research Center
. (
2013
).
Growing support for gay marriage: Changed minds and changing demographics
(Report).
Washington, DC
:
Pew Research Center
. Retrieved from http://www.people-press.org/2013/03/20/growing-support-for-gay-marriage-changed-minds-and-changing-demographics/
Rauch, J. (
2004
).
Gay marriage: Why it is good for gays, good for straights, and good for America
.
New York, NY
:
Times Books
.
Shah, D. K. (
2011
).
Defense of Marriage Act: Update to prior report
.
Washington, DC
:
U.S. Government Accounting Office
. Retrieved from http://www.gao.gov/new.items/d04353r.pdf
Snyder, C. P. (
2004
,
October
23
).
Marriage, family advocate in state to support Coburn
.
The Oklahoman
. Retrieved from http://newsok.com/article/2871070
Trandafir, M. (
2013
).
The effect of same-sex marriage laws on different-sex marriage: Evidence from the Netherlands
.
Demography
. Advance online publication. doi:10.1007/s13524-013-0248-7
Wolfers, J. (
2006
).
Did unilateral divorce laws raise divorce rates? A reconciliation and new results
.
American Economic Review
,
96
,
1802
1820
. 10.1257/aer.96.5.1802