Marriage, an important bond that forms the most basic unit of a society, has long been a research interest to economists. Unlike the romanticized version in arts, economists have quite a ‘cold’ tone in explaining the motivation and behavior of married couples. Existing literature looks at marriage as a ‘deal’ where both parties seek to engage in to maximize their utilities by contributing their available resources (Becker, 1973). This is the process of complementing each other to form a beneficial union. Emotional satisfaction is no doubt a part of the marital benefit a couple seeks to fulfill when entering a marriage. Emotional attachment between partners sure sets a foundation for a lasting co-living experience but it is not the only factor that drives marriage. In ‘A Theory of Marriage’, Becker discussed the economics gains as part of the equation when it comes to marital output. However, there is limitation to his theory as traditionally, men were the breadwinners and women took care of children and house works. While more recent researches aimed to update or complement Becker’s theory on marriage given the structural change in labor market, the classic economics view on marriage remains relevant.
Change in labor market is mainly driven by increase in educated women and technology advancement that liberates women from household duties (Chiappori, 2009). As women are now taking more share in labor market, they have become the second breadwinner in the family. They can be even better off without marriage if her potential mates are proven not worthy of her contribution. Besides, as men also aim to marry higher educated women as much as women do to men (Chiappori, 2009) and both men and women are less likely to marry those who are significantly less educated than them, higher educated women should become more desirable. This is partially supported by data from census surveys which shows most couples have equal education. In couples where there is education disparity, the most common gap is 1 level of education. Ideally, women should have more ‘power’ in selecting partner than before, meaning they should feel less pressure in aiming for equal or higher educated partners given they have improved their stand in society. However, data does not support this. Figure 1 shows the number of couples where women are more educated than men has increased from 14% in 1998 to 21% in 2020, in line with the increase of college educated women.
Despite the rise, this kind of marriage did not bring higher family income than marriages where men are more educated than women (Figure 2). Couples where men are more educated than women maintain their distance from the others throughout 23 years.
The above argument is based on the assumption that there are enough men for women to choose from. Unfortunately, that is not the case. Generally, there are less men than women and there are more women than men in all education levels. The scarcity of ‘quality’ men impairs women’s freedom to choose her compatible partner. This no doubt decreases the economic gains compared to the scenario where that same woman married equal or more educated partner. That comparison is a bit unrealistic if the woman is doctorate educated. There is no higher level than doctorate to do a fair comparison. A similar marriage in terms of education combination could do a better job in showing how marrying less educated men associated to the change in economic gains. Ideally, couples with same education combination of both men and women should have similar family income conditional on industry and occupation. However, the arguable presence of gender inequality in salary could play a role in driving down the family income of couples with women being more educated than men. Despite that, it still shows how much less benefit women enjoy from marrying down compared to the old-schooled marrying up.
With that, the two questions that this post aims to answer are set to quantify the impact of men availability to marrying less educated men (‘OLS A’) and how much lower family income of couples where women are more educated than men compared to the other (‘OLS B’).
Data
Main data: Micro data about labor and housing from 1998 to 2020 (ASEC) is the main data used for this project. ASEC included demographics and income information for each person in surveyed households. This gives convenience to gather data about each person’s spouse. ASEC comes with three different record types: household, family and person. These datasets can be linked together using unique ID. The ‘person’ data includes more than 4 million data entries across 23 years. All states and around 300 counties are included in the survey each year. On average, there are more than 181 thousand data entries each year and 20% of this are married women.
Supplement data from census.gov: “Table DP02” — population by selected characteristics (age, race, education) are used to work out the men/women ratio for each county from 2010 to 2019. “Table S2403” — population by industry of 2019 are used to work out the dominant industry for each county. The dominant industry is defined as the industry which holds the largest percentage of population. This data covers 13 main industries in 836 counties. However, on average, only 266 counties in ASEC are included in “Table S2403” which further decreased the sample size for the analysis. Data of 2019 is used for the ten-year period 2010–2019 as the key industry is not easily shift in short time periods.
OLS models
OLS A: Men availability and education gap in marriage
OLS model was used to measure how men availability is associated to the probability of marrying a lower educated men, with the county’s major industry is set as instrument variable.
OLS B: Family income gap in unequal educated couples
Higher: this binary variable indicates wives with higher educational attainment than their husbands. This variable is constructed by comparing wife’s education and husband’s education. Value of 1 indicates education disparity in favor of wives. Educational attainment in the data is coded as numeric in range from 31 to 46 with each higher level is increment by 1 from previous level. For this analysis, educational attainment is recoded to four big groups: high school (0 to 39), college (40 to 43), master and professional degree (44 and 45), and doctorate (46). As the original census codes are very closed to each other, each increment in school grade is not meaningful in reflecting the distinction between education levels. Regrouping them into broader groups helps discriminating educational attainment and would prevent the model from being too closed to data and hence, the results that it produced would be more representative for the population. These two variables are control variables in OLS B as education levels are correlated with earning.
Age: Wives’ ages. The age range in data is from 0 to 85, with ‘80’ refer to 80–84 years of age and ‘85’ refers to 85+ years of age. This way of grouping is to ensure confidentiality of survey participants and does not impact the regression result as these groups of age are minority in the data.
Availability:
Major industry: Major industry is the major industries of counties where the sampled couples reside. Each county’s major industry is the industry which has the highest proportion of population in it. Major industry is set as instrument variable in OLS A as it can determine the gender and education of its labor force which subsequently impact the availability in each county. It is very unlikely that Major industry can directly drive women getting married to lower educated men. This makes it a suitable instrument variable for OLS A.
Year: years of the surveys. As the proportion of higher educated women increased over the years, year is added as control variable to separate its correlation with this proportion from the regression result.
Family income: family income for the surveyed year.
Education combo: the set of education of each couple in the survey. With 4 levels of education, which are high school, college, master and doctorate, we have 6 unique sets of education. These are high school and college, high school and master, high school and doctorate, college and master, college and doctorate, master and doctorate.
Industry and Occupation: the industries and occupation of both men & women in the survey. Certain industries and occupations have higher income than the others. Hence, industry and occupation are included as control variables in OLS B to eliminate the correlation between them and family income.
Results
OLS A: impact of availability to probability of higher educated wives
Stage 1 regression: F-statistics of regressing men availability on the instrument variable — major industry of county — is 334 which is much higher than 10. This score met the condition for relevance assumption for instrument variable.
Stage 2 regression: Availability has significant negative impact on the probability of women get married to lower educated men. Specifically, in counties where there are more men available, the chance of women topping their husbands’ education decreases more than 27%. Younger women are more likely to be higher educated than their partners compared to older women. Since the data is time series, this could be interpreted as indication for the upward trend in women get married to less educated men.
OLS B: family income gap in unequal educated couples
The regression result shows a decrease of 6K in family income (equivalent to 6.6% of the average family income of $95K) of couples where men are less educated than women; holding education combo, industry, occupation number of children and year of survey constant.
By breaking down this result by education combo, the pattern in family income difference in unequal educated couples became clearer. In Figure 11 below, the Low-High marriage is the marriage where women are less educated than men and the High-Low marriage refers to couples where women are more educated than men. In terms of family income, the master-doctorate combo is the ideal combination but apparently, is hard to acquire as the market can’t provide abundance of doctorate. This can be shown in the sample size chart where the coverage of this combination is very small. In contrary, the less ideal combos, high school-college and college-master, take the majority of the samples. Hence, even though the difference in family income between marriage types in these two combos are not very large, it is much more pervasive in the samples than other combos.
Aside from this, only when women are doctorate educated, her marriage with less educated men results in higher family income compared to couples with similar education combination but having men being more educated than women. Even though this project does not have the ambition to discuss gender inequality, if we assume that gender pay gap plays a role then it should have pervasive impact across education levels. As matter of fact, women earn less than men even when they have the same education and they also contribute to family income less than men as shown in Figure 12.
Then, a possible explanation for family income difference when doctorate education involved is doctorate educated women contribute more to family income than less educated women do, which is also supported by data as in Figure 13. Doctorate educated women contribute roughly 40% to family income, 4% higher than all other education levels.
Ending notes
The increasing trend of marrying less educated men is clearly observed from ASEC data, together with the increasing education level of women through the passage of time. The proportion of high school educated women has been going down from 50% in 1998 to just above 30% in 2020. In contrary, the proportion of women in other higher education levels has been gradually rising since 1999.
However, this correlation does not necessarily indicate that higher educated women prefer less educated partner nor supported by current literature on marriage theory. Regression results showed the dependence of marrying lower educated partner on the availability. This availability is partially driven by the characteristics of the labor market as certain industries typically attract more or less men. In addition, the knowledge required to work in these industries could also have influence on the degree of the participants.
Despite being on increasing trend, the family income of these marriages is actually 6.6% less than that of other marriages. This is not economically significant, however, it still showed interesting pattern of income discrepancies between men and women.
The above results align with existing literature on marriage market where individuals seek compatible partners to increase their well-being. Education remains important in signaling one’s competence as it is strongly associated with income. Once the market is not in deficit of higher educated men, women are more likely to turn to men who are equal or higher educated. This kind of marriage is likely to bring more economic benefit than marrying less educated men, especially when women still earn less than men on average.
References
Becker, G. S. (1973). A Theory of Marriage: Part I. Journal of Political Economy, 81(4), 813–846. https://doi.org/10.1086/260084
Chiappori, P.-A., Iyigun, M., & Weiss, Y. (2009). Investment in Schooling and the Marriage Market. The American Economic Review, 99(5), 1689–1713.
Fry, R., & Cohn, D. (2010, January 19). Women, Men and the New Economics of Marriage. Pew Research Center’s Social & Demographic Trends Project. https://www.pewsocialtrends.org/2010/01/19/women-men-and-the-new-economics-of-marriage/
Torr, B. M. (2011). The Changing Relationship between Education and Marriage in the United States, 1940–2000. Journal of Family History, 36(4), 483–503.
