Marriage & Hypogamy:The effects of Hyper/Hypogamy on Female Marital Happiness and Divorce

Anthony RafettoQuantitative Methods in the Social Sciences

Columbia UniversityMasters of Arts Thesis

May 1, 2013

Abstract

Since the 1980s a reverse gender gap has emerged in post-secondary completion, with women obtaining Bachelor’s degrees at increasingly higher rates than men. Many suggest this creates a smaller market of marriageable men. Using IPUMS census data for female respondents, I find that as relative education levels increase for women, educational hypogamy has not increased drastically but divorce rates have increased and marriage has become less common. I also find that educational hypogamy has a negative association with marital happiness, and also may have a negative association on the likelihood of getting divorced, while educational hypergamy seems to have a positive association. The effects of age on the model were not very pronounced. Work hypogamy, meanwhile, is associated with an increased likelihood of divorce. It is likely that work hypogamy among women, as it is much rarer than educational hypogamy, is still stigmatized in some way.

Introduction

Since the early 1980s the gender gap in college graduation rates has shifted, with

women consistently graduating at higher rates than men (Goldin & Katz, 2007). Scholars

are unsure about what is causing this trend (Bowen, Chingos, & McPherson, 2011), and it

is unclear what effects the reversal of the gender gap in educational attainment may have

had on marital stability, with many studies suggesting different effects. Some research

suggests that, in marriages where the wife has more education than their husband, divorce

is 27% to 38% more likely (Bumpass, Castro, Martin, & Sweet, 1991; Goldstein &

Harknett, 2006; Phillips & Sweeney, 2006; Teachman, 2002). In addition to decreased

marital stability, the gender gap has certainly decreased the pool of “eligible”

(homogamous) men available to highly educated women.

In this paper, I review the literature on educational and income hypogamy, and

examine the effects of marital inequity on marital happiness, likelihood of cheating, and

divorce rates. To do this, I use a regression on General Social Survey (Smith, Marsden,

Hout, & Kim, 2011x) data, and a lagged logistic regression on Integrated Public Use

Microdata Series (Ruggles, Alexander, Genadek, Goeken, Schroeder, & Sobek, 2010)

data.

For this paper, homogamy is defined as marrying somebody who is a relative

equal, whether by having the same education level or the same income or work status.

Hypergamy, meanwhile, is defined as marrying up, or marrying somebody who has more

education or a higher income. Finally, hypogamy is defined as marrying down, or

marrying somebody who has less education or a lower income. In other words, if a

husband has less education but makes more money than his wife, she will be considered

to be married with educational hypogamy but work (or income) hypergamy.

Literature Review

Historically, men have had more education than women and have been willing to

marry those with less education than them. But whether this is due to necessity or

preference is hard to say because typically, people prefer to marry those similar to

themselves (Kalmijn, 1998), and there are possible downsides from both genders

marrying down/up. On the men’s side, husbands who make less money than their wives

are more likely to be unfaithful (Munsch, 2010), and domestic violence is also more

likely in these relationships (Atkinson, Greenstein, & Lang 2005; Melzer, 2002).

On the women’s side, some internet dating studies have shown that women are

less interested in “marrying down” in education than men (Hitsch, Hortaçsu, & Ariely,

2010). Ultimately, marriages where females have greater education are still relatively

rare (Schwartz & Han, 2013), and some experiments suggest that both sexes prefer

homogamous partners, and prefer to avoid relationships in which a woman has higher

status (Fisman, Iyengar, Kamenica, & Simonson, 2006; Hitsch, Hortaçsu, & Ariely,

2010). Many scholars suggest that non-homogamous marriages are less successful purely

because of tension caused by those differences. For instance, interracial relationships are

more likely to face outward disapproval that can strain the union (Bratter & Eschbach,

2006, Fu, Tora, & Kendall, 2001; Root, 2001).

Attitudes toward female hypogamy may be changing though. When male college

students were asked whether they would be bothered if their partners earned a higher

salary, almost 60% said “it wouldn’t bother me at all” in 1990, up from just 41% in 1980

(Willinger, 1993). In addition, while lower-earning husbands may be more likely to

abuse their wives, Atkinson, Greenstein, & Lang (2005) found this was only the case if

the husband held traditional values. Gender differences in work and education have

decreased quite rapidly, but men’s family roles appear to be much slower to change

(Goode, 1982; Hochschild, 1989).

Because schools have students with educational homogamy, it is intuitive that

couples that meet in school and then get married would be educationally homogamous.

With more students in school in general, educational homogamy would, in theory,

increase. It is not so straightforward though. In the U.S., most studies show a rise in

educational homogamy, but the results can vary (Hou & Myles 2008, Rosenfeld 2008,

Schwartz & Mare 2005). For instance, some studies suggest that trends in educational

homogamy do not appear to have any consistency (Kalmijn, 1998).

As time from graduation increases, a man will become less likely to marry with

educational homogamy, while a woman becomes less likely to marry at all (Schwartz &

Han, 2013). In addition, couples that marry older are less likely to get divorced, and

higher educated couples are less likely to as well (Schoen, 1975; Bramlett & Mosher,

2001). The interaction between the two variables is not entirely clear, as a higher

educated couple will likely be older. Ultimately, couples in which wives are more highly

educated than their husbands were historically more likely to divorce, but this association

has declined significantly, and in recent studies, divorce is no more common among these

couples (Schwart & Han, 2013). This may be due to a change in stigma as female

hypogamy becomes more prevalent.

Economically speaking, higher-earning men are more likely to marry higher-

earning women than they were historically (Sweeney & Cancian, 2004). They also value

a woman’s financial prospects, education, and intelligence more (Buss, Shackelford,

Kirkpatrick, & Larsen, 2001). While this could simply be because it is more likely for

women to have those characteristics, it could also be due to the tougher economic climate

and relative need for those attributes in a modern partner (Oppenheimer, 1988; Sweeney,

2002). When women’s socioeconomic status increases relative to men, women are more

likely to select a mate with money being less of a factor (Oppenheimer & Lew 1995,

Sweeney 2002). While trend studies show that the importance of men’s earnings for

marriage has not declined (Buss et al 2001, Sweeney 2002), Fernández, Guner, and

Knowles (2005) found that when women are in good shape economically, educational

homogamy is not as important. “Love” becomes more important than money.

Studies suggest that marital commitment and satisfaction are more useful for

predicting divorce than is the relative economic independence of partners. This suggests

that while it can prevent a woman from exiting an unsatisfactory marriage, it doesn’t

have a significant detrimental impact on a satisfactory marriage (Sayer & Bianchi, 2000).

Similarly, a woman’s participation in the workforce does not have a detrimental effect on

happy marriages, but can increase the risk of divorce in unhappy ones (Schoen, Astone,

Kim, Rothert, & Standish, 2002). However, time series data suggests that full-time

employment for a woman is associated with greater marital stability, and changes in their

employment do not affect marital quality, though unhappily married women are more

likely to join the workforce (Schoen, Rogers, & Amato 2006).

Hypothesis

Many of the effects of marrying up/down on divorce rates are still unknown, so I

explore satisfaction of such marriages and further explore the likelihood of divorcing. I

hypothesize that there are two primary experiences acting against each other in marital

satisfaction and prevalence: age and satisfaction. As age of first marriage increases,

divorce rates go down. But, on the other side, I predict that a woman is less satisfied in

marriage if she marries down. Ultimately, I predict that a woman is actually less likely to

divorce if she marries hypogamously, but will report being less happy in their marriage.

Divorce rates tend to be much lower among more educated and older people. Practically,

this is partly just a function of time: if you marry later you naturally have less chance to

get divorced.

Also, if you divorce at an older age, your chance of finding a better partner is

likely diminished, and your incentive to leave the person you are with becomes lower. I

predict that younger people are much more likely to ultimately get divorced. Finally, I

hypothesize that women will be less likely to marry hypogamously given the choice,

which could influence total marriage rates over the years.

Data

Data are from two sources. I use 1960, 1970, and 1980, 1990, 2000, and 2010

data from IPUMS which is composed of microdata that marks each record as an

individual person, to allow for analysis of specific persons in the context of their

household rather than households being the unit of observation. IPUMS data uses

household size, race, group quarters, and geography to create the representative strata.

Also, the data have person weights for each year to represent how many people in

America the particular respondent represents.

All IPUMS data comes from the Census Bureau every ten years, except 2010

data, which comes from the American Community Survey. Respondents range in age

from 15 to 100 years old, and only heterosexual marriage is considered. The sample

sizes range, but in each given year there are between 300,000 and just over 2 million

female respondents who reported being married, and between 14,000 and nearly 450,000

that reported being divorced (the highest were both in 2000, which had a total female

sample of nearly 4 million, while the lowest were all in 1960, which had a total female

sample of just over 450,000).

I also use all recorded data from the General Social Survey, which has performed

face-to-face interviews from 1972-2012, in order to answer the questions about attitudes

toward marriage. Sample size ranges from 1372 to 1613 each year, with oversamples of

black respondents in 1982 and 1987. Not every question is asked of every respondent,

and some questions are left out if irrelevant. For instance, the question on marital

happiness (hapmar) is only asked of currently married respondents.

Both sources serve a different purpose in completing a robust analysis. For

instance, GSS data is useful for assessing opinions, actions, and feelings about marriage

while IPUMS gives me a much larger sample to analyze likelihood of divorce. The GSS

data, for instance, has only between 60 and 500 divorced female respondents per year,

not nearly enough to perform an accurate lagged analysis from year to year.

My analysis is only on married/divorced women. For this, I drop men completely

from the analysis and use spouse variables included in a respondent’s information. If

there is no spousal information, the respondent is removed from the sample. For my

analysis of factors contributing to divorce, I take a record of all respondents who are

married in one panel year but not in the next, and perform a lagged logistic regression to

determine the effects of certain variables on the likelihood of their becoming divorced.

For this analysis, my primary variables include educational attainment, marital

status, age, age at first marriage, race, and salary. I do not have an age cutoff for the data

because I analyze purely based on marriage then divorce, and would like to see any

impact that age might have on those rates. Educational attainment will be divided into

categories of degree rather than number of years of schooling. Educational attainment

and income/work disparity are used as measures for homogamy. For example, spousal

educational attainment is recoded according to the same process as respondents’

educational attainment and subtracted from respondents’ educational attainment

indicating a difference in degree.

Because others have analyzed educational attainment differences and their effects

on marital dissolution, I also include a brief exploration of the nature of relationships that

are not homogamous and the participants in these relationships, as well as the hesitation

of becoming a participant in such a relationship (as measured by the relative quantity).

For instance, since there are significantly more women with bachelor’s degrees than men,

this may have an impact on likelihood of getting married.

Finally, I will test the satisfaction of non-homogamous relationships through

things like marital happiness and preponderance of cheating. To do so, I analyze the

effects of educational and work hyper/hypogamy on two variables: marital happiness, and

whether a respondent has cheated on her husband. For this analysis, I use a simple yes/no

variable for non-homogamy. For instance, if a woman has more education than her

husband, then they have educational hypogamy. I also include several variables such as

family income, frequency of church attendance, presence of children, and age at marriage

as control variables as many studies (Glenn & Weaver, 1978; Glenn & McLanahan,

1982) suggest these might be important considerations in marital happiness.

These two datasets should be sufficient to allow me to test my hypotheses on the

effects of hyper/hypogamy on marital dissolution, and to observe the trends of such

behavior over an extended period of time in America.

IPUMS Figures

The above table shows the percentage of all bachelor’s degrees held by men

versus women in the IPUMS data. The second line in each row records the frequency.

As can be seen from the above chart, it took until sometime between 1990 and 2000

before women held a majority of bachelor’s degrees. This is because, while women

graduated from college at higher rates starting in the early 1980s, there were a lot of older

men and women left in the population from when men graduated with higher rates.

When just looking at ages 26-35, the change occurs earlier, as can be seen from the chart

and matching graph below:

1960 1970 1980 1990 2000 20100

102030405060708090

% of Bachelor's Degrees by Gender

Women (26-35)Men (26-35)

Year

Perent

The above chart shows women aged 26-35 clearly overtaking men in bachelor’s

degree attainment sometime around 1990. When divided by race, data looks pretty

similar. Below is a table presenting the percentage of bachelor’s degrees held by men

and women (between 26-35 years old) for each race:

White (26-35) 1960 1970 1980 1990 2000 2010

Men 78.43 73.71 61.47 50.57 47.05 43.66

Women 21.57 26.29 38.53 49.43 52.95 56.34

Black (26-35) 1960 1970 1980 1990 2000 2010

Men 52.29 53.92 47.06 41.36 39.43 35.44

Women 47.71 46.08 52.94 58.64 60.57 64.56

Hispanic (26-35) 1960 1970 1980 1990 2000 2010

Men 79.61 73.66 61.08 49.49 45.16 40.49

Women 20.39 26.34 38.92 50.51 54.84 59.51

Asian (26-35 1960 1970 1980 1990 2000 2010

Men 66.91 64.62 58.36 50.14 49.51 44.50

Women 33.09 35.38 41.64 49.86 50.49 55.50

As is expected, the percentages for whites are most consistent with the overall

population, while black women have since the 1980s held a higher percentage of degrees

than black men. Asians seem to stay pretty level, though women still superseded men at

around the same time period. The trend for each race is consistent, with women’s share

of bachelor’s degrees rising and men’s falling. The following figures chart the data for

each respective race:

1960 1970 1980 1990 2000 20100

102030405060708090

% of Bachelor's Degrees among Whites

MenWomen

Year

Percent

1960 1970 1980 1990 2000 20100

10

20

30

40

50

60

70

% of Bachelor's Degrees among Blacks

MenWomen

Year

Percent

1960 1970 1980 1990 2000 20100

10

20

30

40

50

60

70

80

% of Bachelor's Degrees among Asians

MenWomen

Year

Percent

1960 1970 1980 1990 2000 20100

102030405060708090

% of Bachelor's Degrees among Hispan-ics

MenWomen

Year

Percent

Next, I code each marriage in my sample data as being either homogamous,

hypergamous, or hypogamous. Hypergamy is defined as a woman married to a man with

more education while hypogamy is when a woman is married to a man with less

education. Homogamy is when their two education levels are equivalent. If a couple has

equivalent education, their marriage is homogamous and the observation is coded 0. If a

woman has more education than their husband, they are said to marry hypergamously,

and it would be recorded by a 1. If they have less education than their husband, this

would be considered educational hypogamy, and would be recorded by a -1. An average

of 0 would indicate an average of complete homogamy, while an average of 1 would

indicate complete hypogamy, or a woman having less education than their husband.

Average Homogamy Level 1960 1970 1980 1990 2000 2010

Females 0.0059 -0.0811 -0.1311 -0.0929 -0.0519 0.0128

As can be seen in the table above, hypogamy is only slightly more common as the

years progress, and in fact reverses in 2010. This trend doesn’t entirely reflect the

shrinking percent of men with higher educations and smaller market for a

hyper/homogamously inclined woman, but it may shed light on the decreased frequency

of marriage as we see later.

1960 1970 1980 1990 2000 20100%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

Marriage Rates for Women

Ed HypogamyEd HomogamyEd Hypergamy

Year

Per

cen

t

From the above chart we can see that while educational hypergamy dropped

slightly from 1960 to 1980, it rose again into 2010. Homogamy, meanwhile, rose

marginally but has not changed much over time. Hypogamy rose when hypergamy

dropped, but went back down following the 1980s.

The above tables representing the rates of various marriages by race show Asian

and Hispanic women marrying down with much more frequency. Asian women reported

marrying with educational hypogamy at nearly 50% by 2010 (and just above 50% in

1980), and Hispanic women reported marrying with educational hypogamy at about 41%.

Whites, meanwhile, married with educational hypogamy only about 32% of the time in

2010.

It is interesting to see that while educational homogamy and hypergamy are only

moderately common among women, work hypergamy is extremely common, though

decreasing since 1970. This is fairly consistent with common intuition as women entered

the workforce in increasing numbers, but still make less than men on average. This

might also suggest that educational homogamy is less important of a factor: even if a

woman has more education than their husband, they still aren’t likely to make more

money than them. The chart below illustrates the increasing likelihood of a woman

marrying with work hypogamy, which more than quadrupled from 1960 to 2010 (7 to

29%).

1960 1970 1980 1990 2000 20100%

10%20%30%40%50%60%70%80%90%

100%

Marriage Rates for Women

Work HypogamyWork HomogamyWork Hypergamy

Year

Per

cen

t

Looking at divorce rates from each panel year, 6.7% of female respondents who

were married in 2000 were not married in 2010, while only 4.7% of female respondents

married in 1990 were divorced in 2000, and only 3.3% of female respondents married in

1980 reported being divorced in 1990. There is an upward trend in divorce rates as

women began obtaining Bachelor’s degrees at higher rates than men.

Divorce Rates Year-to-Year: 1980-90 1990-2000 2000-2010

3.3% 4.7% 6.7%

I also wanted to see whether other variables, like age at first marriage, might have

an impact on degrees of homogamy. From the chart below, it does appear that older

women are more likely to marry with educational hypergamy, but the effect is only

marginal. 28% of 20 year olds marry with hypergamy, for instance, while up to 40% of

older females marry with hypergamy (40% is the rate among 48 year olds). Homogamy,

meanwhile, drops as a respondent ages. This is quite intuitive, as a young lover is likely

to meet their partner during school. Below age 18, percent of educational homogamy

drops into the 20s from 36% as an 18 year old.

12 15 18 21 24 27 30 33 36 39 42 45 48 51 540%

20%

40%

60%

80%

100%

Educational Homogamy by Age at First Marriage

Ed HypogamyEd HomogamyEd Hypergamy

Age at First Marriage

Per

cen

t

So if hypogamy hasn’t increased steadily over time with the increase in females

with Bachelor’s degrees, does that mean that women are getting married less frequently?

The chart below suggests that to be the case, with almost 34% of respondents in 2010 not

married, and only 18% in 1960.

1960 1970 1980 1990 2000 20100%

10%20%30%40%50%60%70%80%90%

100%

Females by Marital Status

Never MarriedWidowedDivorcedMarried

Year

Per

cen

t

The effect is even more pronounced among 26-35 year olds, with only 8% not married in

1960 but 34% not married in 2010. Whether fear of marrying hypogamously causes it or

not, marriage rates have clearly dropped quite quickly.

1960 1970 1980 1990 2000 20100%

10%20%30%40%50%60%70%80%90%

100%

Females by Marital Status, Ages 26-35

Never MarriedWidowedDivorcedMarried

Year

Per

cen

t

GSS Figures

For the analysis of female respondent’s attitudes towards and behavior in

marriage, nearly all of the variables I use are recoded yes/no variables. Below is a table

illustrating the rates of marrying with education and work hypogamy:

There are at least a chunk of women who marry hypogamously, but it is not nearly as

drastic as might be expected. Here’s a table showing average marital happiness for

women as well as the standard deviation (.56). Happiness is recorded on a 1 to 3 scale,

with 1 being the most happy, and 3 being the least. The second graph shows the

happiness among marriage without regard for gender. So on a 1 to 3 scale, happiness is

right about 50% on average.

Below are two tables showing that there is a solid group of women who married

hypogamously spread out among the happiness scale. The following two tables show

that there are a greater proportion of very happy women who married without hypogamy,

so a regression will be useful to determine any kind of effect this has.

GSS Methods

Hapmar is the first variable I test in regression. The reason for this is because, I

hypothesize, while someone may not get divorced or cheat after they marry

hypo/hypergamously, they may regret it or be slightly less happy about it. For this

model, I added edhypergamy and wrkhypergamy as variables, which return 1 or 0 (1 if a

woman marries up, 0 otherwise. This is not a redundant addition of hypogamy because it

handles the equal situation in the opposite manner – if both are represented by a 0 then

the respondent would be considered to be married with homogamy.

Next, I use family income, church attendance, presence of children, and age at

marriage as control variables as many studies (Glenn & Weaver, 1978) suggest these

might be important considerations in marital happiness. Other studies suggest there is no

effect, and my regression did not prove any correlation among any of those variables

except the presence of children and family income. Below is my initial model, with all

variables initially showing as insignificant.

After removing the variables for age of marriage and belonging to church, there

still was not a significant association from educational hypergamy. The effects from

educational hypogamy and work hypogamy, however, are almost identical, with

respondents reporting themselves as .06 points less happy from both (on average),

holding other variables constant.

Logically, work hypergamy had a similar opposite association, increasing

reported happiness by .04 (on average), holding other variables constant. Age and

income had marginal associations. Having children slightly decreased reported happiness

(about .03 points on average, holding other variables constant), which is consistent with

expectations, though other studies don’t find very powerful effects (Glenn & Weaver,

1978). Below is the regression with all significant variables.

Next, I look at the model from a different angle: is a woman more likely to report

having slept with someone besides their husband if they married down? Here is the final

regression. Number of children, religious attendance, family income, and even

educational hypogamy proved to not be significant for the model and were removed:

The model suggests that both work hypogamy and work hypergamy have almost

the same association with cheating. This might seem counterintuitive at first, but

considering that work hyper/hypogamy technically means that one spouse is working and

the other isn’t, we can suppose that having one spouse at home increases (as opposed to

both working or both staying at home) is associated with an increase of a respondent’s

cheating by about .03 for hypogamy, and about .04 for hypergamy.

IPUMS Methods

To do the analysis on likelihood of divorce using IPUMS data, I start by

identifying women who are married in each year of the data, and isolate those who were

divorced in the next panel year. Because the spread between panels is a 10-year period,

this should be plenty of time for a divorce to occur. Then, I identify a criterion for

hypergamy, homogamy, and hypogamy. For this analysis, I use two different kinds.

First, I use work as a variable, as measured by the difference in personal income between

a respondent and her spouse. If the difference is greater than $1000, then there is

considered to be hypergamy or hypogamy. This limit is also useful for getting rid of

marginal responses.

I also use education, as measured by the differences in personal educational

attainment between a respondent and her spouse. I exclude any response that does not

report either a respondent or their spouse’s earning or education. Then, using control

variables including age, race, educational attainment, income, number of children,

number of times married, and age at first marriage, I run a lagged logistic analysis to

determine whether there is any significant impact on the likelihood of becoming

divorced. A lag model is useful for this scenario as it helps to predict a value of a

dependent variable in the present panel wave of the data using the previous panel wave’s

values. I use logistic regression because it is particularly useful for analysis where the

dependent variable is binary. The logit varies between 0 or 1, reflecting the odds of a

given occurrence – in my case a 0 or 1 that reflects whether or not a respondent is

divorced/separated.

To start, I evaluated my panel data to determine the number of respondents who

got divorced between one wave and the next. For instance, 6.7% of respondents who

were married when these data were collected in 2000 ended up divorced or separated in

2010. For an initial regression, I tested from 2000 to 2010 with no control variables

except for the educational hypergamy/hypogamy variables. The results are as follows:

Educational hypogamy and hypergamy appear to have significant explanatory

power for divorce between the two periods, as both hold p-values far less than .05.

Essentially, according to the regression, a change from no hypogamy to hypogamy is

associated with a change in the logit by -.11 points, while a change from no hypergamy to

hypergamy has a similarly opposite association, increasing the logit by .11 points as well.

This is consistent with my hypothesis that a woman who marries below themselves,

educationally speaking, is not necessarily more likely to get divorced.

It is important to remember that both sets of variables for education and work are

binary – they are not relative. This means that a woman either marries with hypogamy or

they don’t marry with hypogamy. These variables do not measure how far apart the

husband and wife actually are on the scale. This has benefits and downsides: creating a

scale implies attaching a relative value to the data – for instance, coding the difference

between a respondent with a bachelor’s degree and her husband with a master’s degree

would give a net hypergamy of 1, but what if the husband had a PhD? If I were to code

that as a 2, that would imply that having a PhD versus a Bachelor’s degree is twice as

significant as having a Master’s degree versus a Bachelor’s degree. Instead, I use dummy

variables stating a 1 or 0 if the condition is true.

Below is the regression for work homogamy. A woman is defined as being

married with work hypogamy if they make more than $1000 more than their husband,

based on their wages from salary, tips and the like. A woman is defined as being married

with work hypergamy if their husband makes more than $1000 more than them. I chose

to separate the difference by $1000 to filter out any couples that failed to report their

income as well as to limit any noisy differences between marginally different incomes.

While a year or two of school could be the difference between failing to complete a

bachelor’s degree or obtaining a master’s degree, a difference of a few dollars would not

be as significant to a two-person, two-income household.

An opposite association can clearly be seen in work hyper/hypogamy. In other

words, according to the regression, when a woman is hypogamously married in the panel

year 2000, the logit of their chances of divorce by the year 2010 goes up .19, with a

statistically significant p-value. Work hypergamy, while having a similarly opposite

association of -.08 on the logit, is not statistically significant with a p-value just shy

of .05. This is consistent with the theory that a woman who marries “down” is more

likely to get divorced, though not necessarily with my hypothesis.

The next step is to add the control variables to get a sense of how profound the

association actually is, and to add in the additional years. Finally, it is important to

remember that this data does not record information for previous spouses, which means I

can only determine whether, based on the information provided for her spouse in one

wave, a woman is more likely to be divorced in the next. While this could carry

problems for the objectivity of the information, it is an unavoidable weakness of survey

data. Below is my initial logistic regression, for respondents married in 2000 but

divorced in 2010.

As can be seen above, work hypogamy, educational hypogamy, educational hypergamy,

and age are all significant. The income of the respondent has almost no effect, and work

hypergamy is marginally insignificant, with a p-value of .058. Below I perform the same

regression but remove the respondent’s income, which raises the associations of work

hypogamy.

Work hypergamy remains marginally insignificant. Work hypogamy, meanwhile,

is associated with a 0.18 increase on the log of the odds of getting divorced, which

implies there is almost a 20% (e^0.18) increase in a hypothetical respondent’s chances of

getting divorced, all other variables being held constant. Educational hypogamy is

associated with a 0.08 decrease in the log of the odds of getting divorced (which

translates to about an 8% decrease in the chances of getting divorced as well, taking e^-

0.08). Educational hypergamy, meanwhile, has a near opposite association with the log

of the odds of getting divorced, increasing the chances by about 8% (e^0.08). Age has a

less than 1% negative association (e^-0.007). Interestingly, this suggests that a woman

who marries “down”, educationally speaking, would be less likely to get divorced, which

is consistent with my hypothesis that more education doesn’t necessarily increase divorce

rates. However, a woman who marries a man who makes less money than them is more

likely to get divorced.

Next, I perform the same regression on the respondents who were married in 1990

but divorced in 2000. The results are below.

While the associations appear similar, they have some slight variations in degree.

Income becomes significant with a p-value less than .001, but still has only a marginal

association (4.55e-06). The coefficient of the log of the odds on work hypogamy also

drops from 0.18 to 0.06, which would translate to only about a 6% increase in the

chances of getting divorced (e^0.06) as opposed to almost 20% (e^0.18) from the

previous model. Finally, the same regression is run on married respondents in 1980 that

were divorced in 1990, with the results below.

In this sample, work hypogamy and educational hypogamy lose significance. Work and

educational hypergamy, on the other hand, seem to have very similar effects to other

panels. Below is the data with the two insignificant variables removed:

This model suggests that a change from no work hypergamy to work hypergamy

has a -0.12 association with the log of the odds of a hypothetical married respondent in

1980 being divorced in 1990, all other variables being held constant. This would

translate into a 12% (e^-0.124) decrease in the chances of getting divorced. A change

from no educational hypergamy to educational hypergamy, meanwhile, would have a

0.05 positive association on the log of the odds of a hypothetical married respondent in

1980 being divorced in 1990, with all other variables held constant. This would imply an

increase in the chances of getting divorced by approximately 5% (e^0.53). The

associations of age and incwage would be less than 1% on average, all else being equal.

The panels, then, are not entirely consistent. The final panel, from 2000 to 2010,

has the highest sample of married respondents being divorced in the second panel (6.7%).

Limitations

One frustration with both sets of data is that it does not allow me to control for the

effects of degrees that are obtained after marriage, since between the waves I am not able

to know which happened first, or the effects the change could cause. Also, because for

GSS data one of the variables I’m testing is whether the wife works fulltime and her

spouse doesn’t, this gives me only a very crude and slightly affected way to test the

effects of hypo/hypergamy. Unfortunately, the GSS does not have a question asking how

much a spouse makes, only ones asking how much the respondent makes and how much

their household makes. IPUMs data, meanwhile, have information on both a

respondent’s income and their spouse’s income, but not on measures of happiness or

satisfaction. I considered subtracting the respondent’s income from the household

income, but thought it would be too full of noise to give me an accurate sampling.

Instead the question became: is a wife less happy when her husband is not working, but

she is? This is actually quite a rare phenomenon by GSS data. There are 1,142 cases of a

woman marrying with work hypogamy over the course of the GSS data, while there are

22,314 cases of women marring with work hypergamy over the same time.

Because the work hyper/hypogamy variables only record if a woman worked

fulltime and her husband did not, it doesn’t allow for degrees of influence and could

merely represent either wealthier families or families with some sort of dysfunction

(maybe the husband doesn’t work because he can’t, or because he has to take care of a

family member, or something else). However, even the black/white definition proved

useful in this model, which was good to see. The education hypo/hypergamy definition

was much better in this regard, as if a spouse had more years of education, then it would

be considered hypergamy.

Of course, this model could be confounded. We can’t say with certainty that a

woman who marries hypogamously isn’t also less likely to be happy in general. We

don’t know if the women originally married hypogamously or if they became

hypogamous. We also don’t know whether the husband’s lack of fulltime work is driving

the unhappiness of marriage in non-hypogamous ways. There is a strong possibility that

the husband might not be able to find work, which might be causing unhappiness in

general, and the husband’s status could just be an extension of that.

Conclusion

While the IPUMS data suggests that work hypogamy may have a negative

association with the likelihood of getting a divorce, the effect is not entirely constant

across the panels. On the other side, educational hypogamy seemed to have a negative

effect on the chances of getting a divorce, while educational hypergamy seemed to have a

positive effect. This was consistent with my hypothesis, but the effects of age that I

expected to see were not very pronounced. It is likely that work hypogamy among

women, as it is much rarer than educational homogamy, is still stigmatized in some way.

Ultimately, divorce is probably too drastic an action since a respondent was

probably aware of any marital differences when they initially married. It might play into

the changing rates of marriage these days, which at least superficially could help to

explain the large decrease in marriage rates over the years.

Works Cited

Atkinson, M. P., Greenstein, T. N., & Lang, M. M. (2005). For women, breadwinning

can be dangerous: Gendered resource theory and wife abuse.Journal of Marriage

and Family, 67(5), 1137-1148.

Bowen, W. G., Chingos, M. M., & McPherson, M. S. (2009). Crossing the finish line:

Completing college at America's public universities. Princeton University Press.

Bramlett, M. D., & Mosher, W. D. (2001). First marriage dissolution, divorce, and

remarriage. In National Center for Health Statistics.

Bratter, J. L., & Eschbach, K. (2006). ‘What about the couple?’Interracial marriage and

psychological distress. Social Science Research, 35(4), 1025-1047.

Bumpass, L. L., Sweet, J. A., & Cherlin, A. (1991). The role of cohabitation in

declining rates of marriage. Journal of Marriage and the Family, 913-927.

Buss, D., M., Shackelford, T. K., Kirkpatrick, L. A., & Larsen, R. J. (2001). A half

century of mate preferences: The cultural evolution of values. Journal of

Marriage and Family, 63(2), 491-503.

Fernandez, R., Guner, N., & Knowles, J. (2005). Love and money: A theoretical and

empirical analysis of household sorting and inequality. The Quarterly Journal of

Economics, 120(1), 273-344.

Fisman, R., Iyengar, S. S., Kamenica, E., & Simonson, I. (2006). Gender differences in

mate selection: Evidence from a speed dating experiment. The Quarterly Journal

of Economics, 121(2), 673-697.

Fu, X., Tora, J., & Kendall, H. (2001). Marital happiness and inter-racial marriage: A

study in a multi-ethnic community in Hawaii. Journal of Comparative Family

Studies, 32(1), 47-60.

Glenn, N. D., & McLanahan, S. (1982). Children and marital happiness: A further

specification of the relationship. Journal of Marriage and the Family, 63-72.

Glenn, N. D., & Weaver, C. N. (1978). A multivariate, multisurvey study of marital

happiness. Journal of Marriage and the Family, 269-282.

Goldin, C., & Katz, L. F. (2007).  The race between education and technology: The

evolution of US educational wage differentials, 1890 to 2005 (No. w12984).

National Bureau of Economic Research.

Goldstein, J. R., & Harknett, K. (2006). Parenting across racial and class lines:

Assortative mating patterns of new parents who are married, cohabiting, dating or

no longer romantically involved. Social Forces, 85(1), 121-143.

Goode, W. J. (1970). World revolution and family patterns (Vol. 963). New York:

Free press.

Gould, E. D., & Paserman, M. D. (2003). Waiting for Mr. Right: rising inequality and

declining marriage rates. Journal of Urban Economics, 53(2), 257-281.

Hitsch, G. J., Hortaçsu, A., & Ariely, D. (2010). Matching and sorting in online dating.

The American Economic Review, 130-163.

Hoschild, A., & Machung, A. (1989). The second shift: Working parents and the

revolution at home. New York, USA: Viking.

Hou, F., & Myles, J. (2008). The changing role of education in the marriage market:

Assortative marriage in Canada and the United States since the 1970s.Canadian

Journal of Sociology, 33(2).

Kalmijn, M. (1998). Intermarriage and homogamy: Causes, patterns, trends.Annual

review of sociology, 395-421.

Loughran, D. S. (2002). The effect of male wage inequality on female age at first

marriage. Review of Economics and Statistics, 84(2), 237-250.

Melzer, S. A. (2002). Gender, work, and intimate violence: Men's occupational

violence spillover and compensatory violence. Journal of Marriage and

Family,64(4), 820-832.

Oppenheimer, V. K. (1988). A theory of marriage timing. American Journal of

Sociology, 563-591. Oppenheimer V. K. (1995). American Marriage Formation

in the Eighties: How Important was Women's Economic Independence? In

Gender and Family Change in Industrialized Countries, ed. KO Mason, A-M

Jensen, 105-38. New York: Oxford University Press

Phillips, J. A., & Sweeney, M. M. (2006). Can differential exposure to risk factors

explain recent racial and ethnic variation in marital disruption?. Social Science

Research, 35(2), 409-434.

Root, M. P. (2001). Love's revolution: Interracial marriage. Temple University Press.

Rosenfeld, M. J. (2008). Racial, educational and religious endogamy in the United

States: A comparative historical perspective. Social Forces, 87(1), 1-31.

Ruggles, S., Alexander, J. T., Genadek T., Goeken R.,Schroeder, M. B., & Sobek, M. 

Integrated Public Use Microdata Series: Version 5.0 [Machine-readable

database]. Minneapolis: University of Minnesota, 2010.

Sayer, L. C., & Bianchi, S. M. (2000). Women's Economic Independence and the

Probability of Divorce A Review and Reexamination. Journal of Family

Issues, 21(7), 906-943.

Schoen, R. (1975). California divorce rates by age at first marriage and duration of first

marriage. Journal of Marriage and the Family, 548-555.

Schoen, R., Astone, N. M., Kim, Y. J., Rothert, K., & Standish, N. J. (2002). Women's

employment, marital happiness, and divorce. Social Forces, 81(2), 643-662.

Schoen, R., Rogers, S. J., & Amato, P. R. (2006). Wives’ Employment and Spouses’

Marital Happiness Assessing the Direction of Influence Using Longitudinal

Couple Data. Journal of Family Issues, 27(4), 506-528.

Schwartz, C. R., & Mare, R. D. (2005). Trends in educational assortative marriage

from 1940 to 2003. Demography, 42(4), 621-646.

Smith, T. W., Marsden, P., Hout, M., & Kim, J.  General social surveys, 1972-

2010[machine-readable data file] /Principal Investigator, Tom W. Smith; Co-

Principal Investigator, Peter V. Marsden; Co-Principal Investigator, Michael

Hout; Sponsored by National Science Foundation. --NORC ed.-- Chicago:

National Opinion Research Center [producer]; Storrs, CT: The Roper Center for

Public Opinion Research, University of Connecticut [distributor], 2011.

Sweeney, M. M. (2002). Two decades of family change: The shifting economic

foundations of marriage. American Sociological Review, 132-147.

Sweeney, M. M., & Cancian, M. (2004). The changing importance of white women's

economic prospects for assortative mating. Journal of Marriage and

Family, 66(4), 1015-1028.

Teachman, J. D. (2002). Stability across cohorts in divorce risk factors.

Demography, 39(2), 331-351.

Willinger, B. (1993). College Men's Attitudes Toward Family and Work in the 1980s.

Men, work, and family, 4, 108.