Female Income, the Ego
Effect and the Divorce Decision:
Evidence from Micro Data
by
Phone:
(870)9723416
Fax:(870)9108187
Email:
randyk@astate.edu
and
Dale Bremmer
Department of
Humanities and Social Sciences
RoseHulman
Institute of Technology
Phone: (812)8778456
Fax:(812)8778909
Email:
Abstract
During
the 1960’s and 1970’s divorce rates in the
Female Income, the Ego
Effect and the Divorce Decision:
Evidence from Micro Data
During the 1960’s and 1970’s divorce rates in the
This research specifically focuses on the economic causes of divorce. However, there can be little doubt that there are many other factors involved in the decision to dissolve a marriage. For example, demographic factors like changes in the population’s age structure contribute to changes in expected divorce rates. In developed countries where life expectancies have significantly increased, the opportunity for lengthier marriages has also increased. As a natural consequence of a lengthier marriage, the number of divorces would be expected to rise. Development of improved contraceptives and easier access to them has reduced the number of children, thus decreasing the transaction cost to obtaining a divorce. The costliness of children is well known and when a marriage dissolves, the children become an even larger economic burden. Offsetting this burden to a certain degree is the more extensive safety net provided by developed economies. In addition, regime changes like the emergence of nofault divorce laws have reduced the transaction cost of litigation, and by doing so have increased the likelihood of divorce. Finally, one cannot ignore the personal aspect of the decision to end a marriage. Rapid and extensive societal changes can lead to marriage ending frictions that in other circumstances might well be ignored.
Although the factors mentioned above are clearly of importance, this research concentrates on three economic arguments that have persisted over the years. All three relate to the female’s ability to generate income in the labor market. The first argues that as the female increases her ability to generate income, she becomes financially more independent thereby making divorce more likely. Inextricably tied to this argument is the relative cost of child bearing. As a female’s connection to the labor force strengthens, it increases the transaction cost of child rearing. As a logical consequence, having fewer children reduces the transaction cost of divorce. So, by strengthening ties to the labor force, the female invariably weakens ties to the family. The second argument contends that, as female earnings become a larger share of family income, marital friction results and the likelihood of divorce increases. Finally, it has also been argued that the family unit places a high value on the ability of the married female to earn income and, therefore, strives harder to avoid divorce as the female’s ability to earn income rises.
The remainder of the paper proceeds in five additional parts. The next section of the paper provides a short literature review. It examines both some of the theoretical and some of the previous empirical work dealing with the economic issues of divorce. Part III provides a brief explanation of the statistical model used to accomplish the estimations. Part IV describes the data used in the estimations. Part V discusses the empirical results and, finally, some conclusions and caveats are offered in Part VI.
Theoretical Models
In their seminal work on marital instability, Becker, Landes, and Michael [1977] argue that the decision to divorce compares the value of being single to the joint value of being married. Following the modeling work of Weiss [1996], let _{} be the household production function that measures the joint value of staying married. This joint valuation is a function of the characteristics of both spouses, _{} for the male and _{} for the female. The characteristics _{} and _{} include the earnings of each respective spouse. The production function also includes K_{t}, which measures marriagespecific capital such as children. The final argument in the production function, θ_{t}, is an unobservable measure of the quality of the relationship. All the arguments of the production function have a subscript “t” to denote a particular time period. The value of these arguments will vary over the duration of the marriage as will the likelihood of getting a divorce.
Let the postdivorce value of being single equal _{} for the male and _{} for the female. Among other things, these variables include each spouse’s estimate of the option value of remarriage, a possibility that occurs only after the current marriage is dissolved. The transactions cost to obtaining a divorce, C_{t,} not only includes legal expenses, but it also captures the opportunity cost of the time lost in divorce negotiations and the emotional costs incurred during dissolution of the marriage.
Divorce occurs when the sum of the values of becoming single, minus the transaction cost of the divorce, is greater than the joint value of remaining married. Using the terms defined above, divorce is optimal when
_{}
Of primary interest to this research are the theoretical inferences about income. For example, an increase in female labor force participation increases female income, which, in turn, increases _{} and makes divorce more likely. Higher female salaries increase the opportunity cost of children, reduce K_{t}, and increase the likelihood of marriage dissolution. Also, higher female salaries threaten male egos, leading to marital stress that reduces θ_{t}, the quality of the relationship, and increases the probability of divorce.
Empirical Literature
Various empirical studies of divorce in the United States have used a wide assortment of techniques, variables and data to produce results that often appear contradictory and, sometimes even, counterintuitive. In explaining the decision to divorce, crosssectional, microlevel econometric studies have used independent variables such as marriage tenure; working status of the female; the husband’s earnings, age, and educational level; the wife’s earnings, age, and educational level; the number and age of children; and the occurrence of a previous divorce. Timeseries analyses of a country’s divorce rate have used macroeconomic variables such as the unemployment rate, the inflation rate and the rate of growth in real GDP to explain changes in the divorce rate.
Micro, crosssectional studies
Lombardo [1999], and Greene and Quester [1982] argue that wives facing a higher risk of divorce will hedge against that risk with higher levels of labor force participation, and that they will also respond by working longer hours. The basic argument made in these papers is that investment in nonmarket activities, such as child rearing, becomes relatively less attractive (it yields a lower expected return), and investment in human capital becomes relatively more attractive (it yields a higher expected return) as the probability of divorce increases. Studies by Johnson and Skinner [1986], and Shapiro and Shaw [1983] provide additional evidence for the above argument by finding that women increase their labor force participation prior to dissolution of a marriage. The above papers make a clear causality argument that an increase in the likelihood of divorce increases a female’s willingness to enter the labor force.
Spitze and South in two separate studies [1985, 1986] argue a different line of causality. Their conclusion is that an increase in female labor force participation leads to an increase in familial conflict and, consequently, an increase in divorce. Substantiating evidence for this view is provided by Mincer [1985]. In a survey of twelve industrialized nations he found that rising divorce rates clearly lag rising female labor force participation rates.
Previous studies of the impact of income on divorce have provided mixed results. Becker, Landes, and Michael [1977] find that a rise in expected female earnings increases the probability of divorce, while a rise in expected male earnings reduces the probability of divorce. D’amico [1983] recognizes two distinctly different possible effects of income on divorce. One hypothesis is that as the female’s wage relative to the male’s rises, conflict based on competition for status within the marriage will occur and will increase the likelihood of divorce. The second, opposing hypothesis is the notion that the pursuit of higher socioeconomic status is a familial one and that a wife earning a relatively higher wage than that of the husband may contribute to the overall status goal and solidify the marriage. D’amico’s results tend to confirm the latter hypothesis. Finally, Hoffman and Duncan [1995] find no support for the hypothesis that higher real female wages lead to increased divorce rates and a study by Sayer and Bianchi [2000] tends to confirm this finding.
However, Spitze and South raise another question that is closely related to the income issue. In a 1985 study, they produce evidence that the number of hours a wife works has a greater impact on the probability of divorce than do various measures of the wife’s income.
Macro, timeseries studies
In an empirical study of the growth of divorce rates in Great Britain, Smith (1997) finds no evidence that marriage dissolution increases because of the introduction of nofault divorce laws. He argues that, rather than being the vehicle of change, these types of legal modifications merely codify, react to, and regulate ongoing social and economic transformations. Smith finds that procedural and legal changes do have a powerful, albeit temporary impact. One procedural change that did increase the number of divorces was revised court settlement rules that reduced transactions costs and improved the financial position of females post divorce. Smith attributes the increased number of divorces in Great Britain to rising female labor force participation, higher female income and the subsequent reduction in the female’s economic dependency on marriage. In addition, technological change giving females greater control over fertility has had a significant impact. With improved fertility control resulting in fewer children, a significant transaction cost of divorce has been effectively reduced.
South [1985] finds little evidence that the divorce rate rises during periods of recession and falls during periods of expansion. He does find a positive, albeit small, effect of unemployment on the divorce rate. His model indicates that changes in the age structure and the labor force participation rate of women have significantly stronger impacts on the divorce rate than other macroeconomic variables.
Using a simple vector autoregressive (VAR) approach with macro, timeseries data, Bremmer and Kesselring [1999] show that the female labor force participation rate does not Granger cause divorce rates. However, they do provide statistical evidence that divorce rates Granger cause female participation in the labor force. They also show that past participation in the labor market influences women salaries.
In another timeseries study using macro data, Bremmer and
Kesselring (2003) use cointegration techniques to investigate the relationship
between divorce, female labor force participation, and median female
income. Though these variables had unit
roots, their first differences were stationary, and these variables were shown
to be cointegrated. Impulse functions
from this model reveal that an increase in divorce leads to a rise in female
labor force participation; but positive innovations to female labor force
participation imply a decline in the divorce rate. Impulse function analysis also shows that a
positive innovation to median female income leads to increased divorce and
increased labor force participation on the part of females.
The most pervasive thread tying the above literature together is the relationship between female earning capacity and its relationship to divorce. This argument takes several important forms. First is a line of reasoning that developed during the 1970’s and 1980’s when females experienced a substantial increase in their ability to successfully participate in the labor market. Unquestionably, female labor force participation increased during this time and, along with this increase came improved salaries and benefits. Consequently, academics argued that as females developed better access to income outside the marriage the likelihood of divorce increased. While this idea was written about in many ways it proved difficult to quantify. It also led to counter arguments that when a married female’s value in the labor market increased it might well strengthen a marriage and reduce the probability of divorce. The difficulty with quantifying these arguments is the very nature of the observable outcomes. It is possible to observe the income of married females. It is also possible to observe the income of divorced females. Unfortunately, it is not possible to simultaneously observe both outcomes for an individual female. So, while the data may indicate that divorced females participate in the labor force at substantially higher rates than married females, it provides no valid evidence that ready access to the labor market is a causal factor in divorce. In other words, this is a classic sample selection problem.
Over the years, academics of various disciplines began to develop another income related argument that they hoped would explain the increasing occurrence of divorce. As a married female’s contribution to household income grows, it gives her more say in the conduct of the marriage. So, conflict within the marriage is given an opportunity to flourish. Disagreements about the way that money is spent and, for that matter, disagreements over the conduct of everyday household responsibilities become much more likely. However, the sample selection problem arises again. It is possible to observe the amount that married females contribute to the household but it is not possible to directly observe the amount that a divorced female would have contributed to the household had she remained married. Consequently, econometric techniques accounting for these difficulties are required.
A typical statistical approach to solving the above problems begins by specifying a selection equation. For example, the selection equation for the divorce problem would be:
_{}
where_{} represents a vector of variables that predict the likelihood of divorce. In this case (as in most cases) the selection variable (_{}) is not observed. Instead, only its sign is observed:
_{}
In order to test the above hypotheses regarding divorce, three different income equations need to be estimated. So, there are three equations that will make similar (though not identical) use of the information from the selection equation. The three equations are:
_{}
_{}
_{}
where PY stands for personal earnings and FY stands for family income.
The estimation technique used for all three is very similar. So, model development proceeds by using the first equation as an example. The usual equation that one is interested in estimating is:
_{}
where _{} is the vector of independent variables used to predict _{}. Unfortunately, _{} is only observed when _{}=1. Also, for this derivation the standard assumptions are made about _{} and _{}. In other words, they have a bivariate normal distribution with zero means and a correlation of r.
What one truly wishes to estimate is:
_{}
which is equivalent to :
_{}.
Incorporation of the selection equation results in the above equation yields:
_{}
_{}
_{}
_{}
where: _{}. The equation in estimable form becomes:
_{}.[1]
Consequently, application of OLS to the model results in two problems. First, omitting _{}(commonly referred to as the inverse Mills ratio) creates a bias similar to that attributed to an omitted variable and, second, the disturbance term,_{}, is heteroscedastic. As a result, estimation proceeds in a twostep manner. First, following the methodology recommended by Heckman [1979], the selection equation is estimated and used to create a _{}for each observation. Then, _{}is estimated by regressing _{} on _{} and _{}. Finally, the correction for heteroscedasticity recommended by Greene [1981] is applied to the estimates. By following similar logic and making minor adjustments, the estimates for all three of the equations can be obtained.
In order to test the hypotheses specified earlier, an additional step must be taken. It is necessary to use the estimated coefficients obtained from the least squares regressions to produce expected values (observations) for every individual (both married and divorced) in the data set [Lee, 1978]. Once these observations have been obtained, the necessary relationships can be calculated.
The created variable used to test the first hypothesis is:
_{}.
Smaller values for _{} indicate that the female is better equipped to enter the labor force and, consequently, better prepared to live independently. If a female were the only participant in the marriage, it would make sense to argue that as _{} declines the likelihood of divorce would definitely increase. However, the male is also a participant in the marriage and while his interaction in the process has not been written about as frequently as the female’s, it should (might?) be just as important. If the male values the possible economic contribution of the female to the family’s economic welfare, he would have an increasing preference to remain married as _{} declines and should, therefore, seek to reduce the likelihood of divorce.
The other variable of interest is created in the following manner:
_{}.
As _{} declines, the percentage of total family income accounted for by female earnings increases. Some authors have argued that this increasing share of income on the part of the female can cause friction within the marriage and, consequently, lead to a greater likelihood of divorce. Others have argued that the female’s contribution could be highly valued by the other family members and should, therefore, reduce the likelihood of divorce. So, the question of whether _{} or _{} has a positive or negative impact on divorce becomes an empirical issue.
The following equation was formulated to provide a statistical test of these issues:
_{}
where, as before, _{}stands for divorce. This equation, estimated by a probit procedure, has been referred to in the literature as a structural equation [see Lee] because it includes the specific variables hypothesized to predict the binary outcome. In this particular case, the series of dummy variables representing the fifty different states is included in the equation because each state has different laws regarding marriage and divorce. Plus, there are still cultural and social factors at work in the different states that could lead to varying divorce outcomes.
Most of the variables were taken from the March Supplement (Annual Demographic File) of the Current Population Survey. The complete surveys for the years 1990, 1995 and 2000 were obtained from the web site maintained by the National Bureau of Economic Research. State per capita income was obtained from the Bureau of Economic Analysis web site under the heading of State and Local Personal Income. Finally, the Consumer Price Index for 1990, 1995 and 2000 came from the Bureau of Labor Statistics. The names and definitions of all of the variables are provided in Table 1.
The statistical procedures required observations for married females and for divorced females. Application of these restrictions to the data set resulted in 112,740 usable observations. Of this total, 16,760 represented divorced females and 95,980 represented married females.
Table 2 provides the results from the probit procedure that was applied to the divorce selection equation. All variables used in the various estimation procedures were included in this equation. This is the usual procedure for estimations of this type because any variable that affects income should also have an effect on the occurrence of divorce. A criticism that has been directed at models of this type is that unless variables in the selection equation can, reasonably, be excluded from the other equations, results can be unreliable (See Vella, 1998). Fortunately, this formulation of the model provides an entire set of variables that can reasonably be excluded from the income equations—the series of state dummy variables. The state dummies are included in the selection equation to account for differences in state laws and socioeconomic conditions that, theoretically, should affect the decision to obtain a divorce. They are excluded from the income equations in favor of state percapita income. This variable is included in the income equations to account for variation in labor market conditions on a geographical basis.
Due to the rather large number of variables included in the estimation procedure, the coefficients for the state dummy variables (S_), the occupational dummy variables (OC_) and the industry dummy variables (IN_) are omitted.[2] Of the 34 variables listed in Table 2, 28 are significant at the 1 percent or 5 percent level. Most of the variables behaved in predictable ways. For the most part, if the female in question had migrated within the U.S. during the previous year (MIG_), the probability of being divorced was significantly higher. On the other hand, if the migration was from abroad (MIG_ABM, MIG_ABN), the probability of divorce significantly decreased. While the results for the education dummies (ED_) provided some mixed evidence, generally, they indicated that having a lower level of education increases the likelihood of divorce. As family size increases the likelihood of divorce significantly declines and being of Asian descent also reduces the likelihood of divorce. Of course the primary purpose of this equation is not for statistical inference, but for use in estimating the earnings equations.
Table 3 provides the results for the selection corrected estimation of divorced female personal earnings. The coefficients and tscores for the thirteen occupation dummies and the fourteen industry dummies that were included in the estimation are not displayed.[3] The education variables reveal no significance until the equivalent of a bachelor’s degree is attained. Then the coefficients increase in size and significance as education rises. The estimated coefficients for the migration variables are negative (with one exception) and both of the coefficients that are significantly different from zero are negative. Not surprisingly, the coefficient on being a federal worker is significantly positive while the coefficients on being either a state or a local government worker are significantly negative.[4] Being a member of the labor force significantly increases earnings whether the participant is employed or unemployed. Of the ethnic variables, only the coefficient on being Hispanic is significantly different from 0 (at the 10% level of significance) and it is negative. It is a little surprising that coefficient for age is negative and significant. However, the rest of the variables behave in expected ways. As the number of children in the household increase (Under18), earnings significantly decline. There is also a significant negative relationship between selfemployment and earnings. On the other hand, living in a metropolitan statistical area, purchasing a home, increasing levels of state per capita income, and increasing family size all significantly increase earnings. Finally, for this equation, the estimated coefficient for the selection variable (lambda) is negative and significant. The adjusted Rsquared for this equation is 0.47 which is a very respectable showing considering that most of the variables used in the equation are binary.
Table 4 presents the estimation results for the selection corrected equation predicting personal earnings for married females. Once again the coefficients and tscores for the thirteen occupation dummies and the fourteen industry dummies that were included in the estimation are not reported.[5] The results for this equation are very similar to those for the divorced earnings equation. Education doesn’t have a significantly positive affect on earnings until a bachelor’s degree is obtained at which point it has a large impact. Graduate education has an even greater positive effect. Of the six migration variables, only three have significant coefficients and they are all negative. Once again, working for the federal government has a significant and positive impact on earnings, as does participation in the labor force whether employed or not. The estimated coefficients for the ethnic dummies vary considerably from the divorced equation. The coefficients for being African American and Asian are both positive and significantly different from zero at the one percent level. However, the coefficients on being Hispanic or Indian are not significantly different from zero. Once again, age has a significant and negative effect on earnings. The number of children in the household and selfemployment both reduce income while living in an MSA, buying a home, having a larger family size, and living in a higher income state all significantly increase earnings. The estimated coefficient for lamda (the selection variable) is negative and significant. Finally, the adjusted Rsquared is 0.48.
Table 5 presents the selection corrected results for the equation predicting family income. The coefficients and tscores for the thirteen occupation dummies and the fourteen industry dummies that were included in the estimation are not reported.[6] The education variables behave in the expected fashion (very similar to the personal earnings equations) with increasing levels of education producing higher levels of income once the high school graduate stage is reached. The migration variables also behave very similarly to the way they behave in the personal earnings equation with a recent migration (within the previous year) resulting in reduced income. The labor force participation variables behave quite differently. If the female is employed it has a positive and significant impact on family income, but if the female is unemployed there is a negative and significant impact on family income. The ethnic variables also behave very differently in this estimation. The estimated coefficients for Black, Indian and Hispanic are all negative and significant. The coefficient for Asian is insignificant. Increasing numbers of children and age significantly decrease family income. Living in an MSA, being selfemployed, buying a home, increasing family size, and living in a state with a higher per capita income all significantly increase family income. The estimated coefficient for the selection variable (lambda) is negative and significant. Finally, the adjusted Rsquared is 0.38.
Of course, the final probit provides the most intriguing results and they are provided in Table 6. Both of the created difference variables (_{} and _{}) have negative and significant estimated coefficients. The result for _{} indicates that as females become more successful at producing income in the divorced state, the more likely they are to become divorced. The result for _{} indicates that as female earnings becomes a larger portion of total family income, the likelihood of divorce increases. Interestingly, the idea that the family attaches a positive value to the female’s earnings and therefore, attempts to continue the state of marriage fails to overcome the previous two arguments. There may be some validity to this contention, but the positive effect (if present) is obviously not large enough to offset either of the other two effects.
In addition to the two difference variables, the entire series of state dummy variables was included in this estimation (they are excluded from the individual income equations) just as it was in the original selection probit (although the results do not appear in Table 6).[7] The reason for including these variables is primarily that the various states have different legal systems and, as mentioned earlier, other studies have found that the legal situation does have an important impact on the willingness and ability to dissolve a marriage. Curiously enough, 39 of the 50 estimated coefficients were significantly different from zero at the 10% level of significance.[8] So, even though the created income variables (_{} and _{}) are significant, the geographic dummies retain their importance. Finally, even though the structural probit makes use of many fewer variables than the selection probit, its pseudo Rsquared is almost double that of the selection estimation—0.39 compared to 0.22.[9]
The estimations reported in this paper tend to confirm arguments that have long been made about the causes of divorce. As females experience greater levels of success in the labor market, they also tend to experience higher levels of divorce. This occurs for two important reasons. First, greater financial independence clearly makes the decision to seek a divorce much simpler. In this respect, developing countries that have concentrated on guaranteeing equal economic status for females have reduced the burden of living with unhappy marriages strictly for economic reasons. However, it also appears that a female’s economic success may, indeed, cause friction within the family. The results of the estimations in this paper clearly indicate that as the female’s earnings become a larger portion of total family income, the likelihood of divorce increases even while controlling for general success in the labor market (_{}). Of course, over time the causes of this effect (fragile male egos?) could well change.
Finally, there is little doubt that many other contributory factors affect the decision to seek a divorce. This is the reason that the series of state dummy variables was maintained in the final probit equation. As all statisticians know, resorting to the use of binary variables reveals a certain degree of ignorance. In this particular case, that degree of ignorance turned out to be relatively important, as thirtynine of the estimated coefficients for the state dummy variables in the final probit were significantly different from zero. In other words, opportunities still exist in the search for a more perfect statistical model of marriage dissolution.
References
Becker, Gary S., Elisabeth M. Landes, and Robert T. Michael, “An Economic Analysis of Martial Instability, “ Journal of Political Economy, 1977, vol. 85, no. 6, 11411187.
Bremmer, Dale and Randy Kesslering, “The Relationship between Female Labor Force Participation and Divorce: A Test Using Aggregate Data,” unpublished mimeo, 1999.
Bremmer, Dale and Randy Kesslering, “Divorce and Female Labor Force Participation: Evidence from TimesSeries Data and Cointegration,” unpublished mimeo, 1999.
Current Population Survey, March 1990, 1995, 2000 [machinereadable data file] / conducted by the Bureau of the Census for the Bureau of Labor Statistics. Washington: Bureau of the Census [producer and distributor], 1990, 1995, 2000.
D’amico, Ronald, “Status Maintenance or Status Competition? Wife’s Relative Wages as a Determinant of Labor Supply and Martial Instability,” Social Forces, June 1983, vol. 61, no. 4, 11861205.
Duncan, Gregory M., “Sample Selectivity As a Proxy Variable Problem: On the Use and Misuse of Gaussian Selectivity Corrections,” In New Approaches to Labor Unions: Research in Labor Economics, Supplement 2, Edited by Joseph D. Reid, Jr., Greenwich: JAI Press, Inc., 1983.
Greene, William H., Econometric Analysis, Fourth Edition, Prentice Hall, 2000.
Greene, William H., “Sample Selection Bias As a Specification Error,” Econometrica (May 1981), 795798.
Greene, William H. and Aline Q. Quester, “Divorce Risk and Wives’ Labor Supply Behavior,” Social Science Quarterly, March 1982, vol. 63, no. 1, 1627.
Greenstein, Theodore N., “Marital Disruption and the Employment of Married Women,” Journal of Marriage and the Family, Aug. 1990, vol. 52, no. 3, 657677.
Hannan, Michael T., Nancy Brandon Tuma and Lyle P. Groenveld, Income and Independence Effects on Marital Dissolution: Results from the Seattle and Denver IncomeMaintenance Experiments,” American Journal of Sociology, 1978, vol. 84, no. 3, 611633.
Heckman, James J., “Sample Selection Bias As a Specification Error,” Econometrica, vol. 47 (January 1979), 153161.
Hoffman, Saul and Greg Duncan, “The Effect of Incomes, Wages, and AFDC Benefits on Martial Disruption,” The Journal of Human Resources, vol. 30, no. 1, 1941.
Johnson, N., and S. Kotz, Distribution in Statistics: Continuous Multivariate Distributions, New York: John Wiley & Sons, 1972.
Johnson, William R. and Jonathan Skinner, “Labor Supply and Martial Separation,” American Economic Review, June 1986, vol. 76, no. 1, 455469.
Kreider, Rose
M. and Jason M. Fields, "Number, Timing, and Duration of
Marriages and Divorces: 1996", U.S. Census Bureau Current
Population Reports, February 2002.
Lee, LungFei, “Unionism and Wage Rates: A Simultaneous Equations Model with Qualitative and Limited Dependent Variables,” International Economic Review, vol. 19 (June 1978), 415433.
Lombardo, Karen V., “Women’s Rising Market Opportunities and Increased Labor Force Participation,” Economic Inquiry, April 1999, vol. 37, no. 2, 195212.
Mincer, Jacob, “Intercountry Comparisons of Labor Force Trends and Related Developments: An Overview,” Journal of Labor Economics, 1985, vol. 3, no. 1, pt. 2, S132.
Sayer, Liana C. and Suzanne M. Bianchi, “Women’s Economic Independence and the Probability of Divorce,” Journal of Family Issues, vol. 21, no. 7, October 2000, 906943.
Shapiro, David and Lois Shaw, “Growth in Supply Force Attachment of Married Women: Accounting for Changes in the 1970's,” Southern Economic Journal, vol. 6, no. 3, September 1985, 307329.
Smith, Ian, “Explaining the Growth of Divorce in Great Britain,” Scottish Journal of Political Economy, vol. 44, no. 5, November 1997, 519544.
South, Scott, “Economic Conditions and the Divorce Rate: A TimeSeries Analysis of Postwar United States,” Journal of Marriage and the Family, February 1985, vol. 47, no. 1, 3141.
South, Scott and Glenna Spitze, “Determinants of Divorce over the Martial Life Course,” American Sociological Review, August 1986, vol. 51, 583590.
Spitze, Glenna, and Scott South, “Women’s Employment, Time Expenditure, and Divorce,” Journal of Family Issues, September 1985, vol. 6, no. 3, 307329.
Stanley, T.D. and Stephen B. Jarrell, “Gender Wage Discrimination Bias? A MetaRegression Analysis,” The Journal of Human Resources, Fall 1998, vol. 33, no. 4, 947973.
Vella, Francis, “Estimating Models with Sample Selection Bias: A Survey,” The Journal of Human Resources, Winter 1998, vol. 33, no. 1, 127169.
Weiss, Y. “The Formation and Dissolution of Families: Why Marry? Who Marries Whom? And What Happens Upon Divorce?” Handbook of Population and Family Economics. M.A. Rosenweig and O. Stark, eds. Elsevier: North Holland, 1996.
Table 1
List of Variables
Variable 
Definition 
S_[two
letter state code] 
Binary:
1 for selected state, 0 otherwise (Washington,
D.C. is included) 
ED_56 
Binary:
1 if highest grade attempted was fifth or sixth, 0 otherwise 
ED_78 
Binary:
1 if highest grade attempted was seventh or eighth, 0 otherwise 
ED_9 
Binary:
1 if highest grade attempted was ninth, 0 otherwise 
ED_10 
Binary:
1 if highest grade attempted was tenth, 0 otherwise 
ED_11 
Binary:
1 if highest grade attempted was eleventh, 0 otherwise 
ED_12 
Binary:
1 if highest grade attempted was twelfth, 0 otherwise 
ED_Hsgd 
Binary:
1 if high school graduate, 0 otherwise 
ED_Univ 
Binary:
1 if some college was attempted, 0 otherwise 
ED_BA 
Binary:
1 if college graduate, 0 otherwise 
ED_Grad 
Binary:
1 if some graduate school was attempted, 0 otherwise 
OC_Exec 
Binary:
1 if executive, 0 otherwise 
OC_Prof 
Binary:
1 if Professional, 0 otherwise 
OC_Tech 
Binary:
1 if Technician, 0 otherwise 
OC_Sales 
Binary:
1 if Sales, 0 otherwise 
OC_Clerical 
Binary:
1 if Clerical and administrative support, 0 otherwise 
OC_Household 
Binary:
1 if Household service worker, 0 otherwise 
OC_Guard 
Binary:
1 if Protective service worker, 0 otherwise 
OC_Oservice 
Binary:
1 if Other service worker, 0 otherwise 
OC_Craft 
Binary:
1 if Craft and repair worker, 0 otherwise 
OC_Machine 
Binary:
1 if Machine operator, 0 otherwise 
OC_Transport 
Binary:
1 if Transportation and material moving, 0 otherwise 
OC_Equipment 
Binary:
1 if Handler, equipment cleaners, etc., 0 otherwise 
OC_Farm 
Binary:
1 if Farming, forestry and fishing, 0 otherwise 
IN_Ag 
Binary:
1 if Agriculture, 0 otherwise 
IN_Mine 
Binary:
1 if Mining, 0 otherwise 
IN_Construction 
Binary:
1 if Construction manufacturing, 0 otherwise 
IN_DGS 
Binary:
1 if Durable goods manufacturing, 0 otherwise 
IN_NDGS 
Binary:
1 if Non Durable goods manufacturing, 0 otherwise 
IN_Trans 
Binary:
1 if Transportation, communication and public utilities, 0 otherwise 
IN_Whtr 
Binary:
1 if Wholesale trade, 0 otherwise 
IN_Retr 
Binary:
1 if Retail trade, 0 otherwise 
IN_Fin 
Binary:
1 if Finance, Insurance or Real Estate, 0 otherwise 
IN_Bserv 
Binary:
1 if Business and repair services, 0 otherwise 
IN_Pserv 
Binary:
1 if Personal services, 0 otherwise 
IN_Ent 
Binary:
1 if Entertainment and recreation services, 0 otherwise 
IN_Profs 
Binary:
1 if Professional and related services, 0 otherwise 
IN_Padm 
Binary:
1 if Public administration, 0 otherwise 
MIG_MM 
Binary:
1 if Migrated from MSA to MSA in previous year, 0 otherwise 
MIG_MNON 
Binary:
1 if Migrated from MSA to non MSA in previous year, 0 otherwise 
MIG_NONM 
Binary:
1 if Migrated from non MSA to MSA in previous year, 0 otherwise 
MIG_NN 
Binary:
1 if Migrated from non MSA to non MSA in previous year, 0 otherwise 
MIG_ABM 
Binary:
1 if Migrated from abroad to MSA in previous year, 0 otherwise 
MIG_ABN 
Binary:
1 if Migrated from abroad to non MSA in previous year, 0 otherwise 
WC_Fed 
Binary:
1 if Worked for federal government, 0 otherwise 
WC_Local 
Binary:
1 if Worked for local government, 0 otherwise 
WC_State 
Binary:
1 if Worked for state government, 0 otherwise 
LF_Work 
Binary:
1 if Working, 0 otherwise 
LF_Unemp 
Binary:
1 if Unemployed but in the labor force, 0 otherwise 
Black 
Binary:
1 if African American, 0 otherwise 
Indian 
Binary:
1 if American Indian, 0 otherwise 
Asian 
Binary:
1 if Asian, 0 otherwise 
Hispanic 
Binary:
1 if Hispanic, 0 otherwise 
Age 
Age
of the individual measured in years 
Under18 
Number
of children under 18 years of age living with the family 
MSA 
Binary:
1 if living in an MSA, 0 otherwise 
Selfemploy 
Binary:
1 if Self employed, 0 otherwise 
Homebuy 
Binary:
1 if Purchasing the home, 0 otherwise 
Statepcy 
Per
capita income of the state of residence 
Famsize 
Number
of people in the family 
1995 
Binary:
1 if Observation is from 1995, 0 otherwise 
2000 
Binary:
1 if Observation is from 2000, 0 otherwise 
Divorce 
Binary:
1 if Divorced, 0 otherwise 
Fincome 
Family
income measured in constant dollars 
Pincome 
Personal
income measured in constant 19821984 dollars 
Table 2
Estimation Results for the
Probit Selection Equation
(dependent variable =
Divorce)
Variable 
Coefficient 
tscore 
Constant 
0.1509 
0.732 
ED_56 
0.1522 
2.324^{**} 
ED_78 
0.0642 
1.117 
ED_9 
0.1505 
2.494^{**} 
ED_10 
0.1864 
3.254^{*} 
ED_11 
0.1403 
2.432^{**} 
ED_12 
0.1420 
2.020^{**} 
ED_Hsgd 
0.1072 
2.104^{**} 
ED_Univ 
0.1847 
3.577^{*} 
ED_BA 
0.0445 
0.830 
ED_Grad 
0.0675 
1.206 
MIG_MM 
0.0578 
3.547^{*} 
MIG_MNON 
0.0904 
2.004^{**} 
MIG_NONM 
0.1364 
2.542^{**} 
MIG_NN 
0.2473 
8.365^{*} 
MIG_ABM 
0.4063 
5.572^{*} 
MIG_ABN 
0.5810 
2.192^{**} 
WC_Fed 
0.0293 
0.720 
WC_Local 
0.0138 
0.576 
WC_State 
0.0801 
2.531^{**} 
LF_Work 
0.2836 
6.397^{*} 
LF_Unemp 
0.3325 
6.267^{*} 
Black 
0.4260 
21.769^{*} 
Indian 
0.3022 
5.695^{*} 
Asian 
0.4105 
10.159^{*} 
Hispanic 
0.0427 
2.158^{**} 
Age 
0.0134 
27.633^{*} 
Under18 
0.7474 
92.804^{*} 
MSA 
0.1631 
10.171^{*} 
Selfemploy 
0.2555 
11.475^{*} 
Homebuy 
0.5600 
43.586^{*} 
Statepcy 
0.0000 
1.185 
Famsize 
0.9525 
125.417^{*} 
1995 
0.1055 
7.305^{*} 
2000 
0.1649 
4.895^{*} 



n 
112,740 

Chi
squared (goodness
of fit) 
28,632.46^{*} 

Pseudo
R sqared (Mckelvey
and Zavoina) 
0.2243 

^{ *}Significant at the 1% level.
^{**}Significant at the 5% level.
Table 3
Estimation Results for the
Divorced Personal Income Equation
(dependent variable =
Personal Earnings)
Variable 
Coefficient 
tscore 
Constant 
5938.062 
3.961*^{} 
ED_56 
414.453 
0.363 
ED_78 
376.520 
0.349 
ED_9 
229.585 
0.838 
ED_10 
568.008 
0.467 
ED_11 
481.276 
0.434 
ED_12 
173.434 
0.144 
ED_Hsgd 
398.550 
0.376 
ED_Univ 
846.582 
0.785 
ED_BA 
3840.904 
3.595^{*} 
ED_Grad 
7318.364 
6.680^{*} 
MIG_MM 
511.214 
2.705^{*} 
MIG_MNON 
880.202 
1.609 
MIG_NONM 
1537.966 
2.486^{**} 
MIG_NN 
533.094 
1.470 
MIG_ABM 
651.284 
0.631 
MIG_ABN 
219.384 
0.051 
WC_Fed 
1994.064 
4.281^{*} 
WC_Local 
1099.755 
3.716^{*} 
WC_State 
960.939 
2.600^{*} 
LF_Work 
5402.567 
8.594^{*} 
LF_Unemp 
1444.790 
2.105^{**} 
Black 
395.019 
1.534 
Indian 
194.288 
0.331 
Asian 
611.861 
1.140 
Hispanic 
443.722 
1.865^{***} 
Age 
26.624 
2.883^{*} 
Under18 
1071.602 
3.753^{*} 
MSA 
1358.628 
6.372^{*} 
Selfemploy 
836.517 
2.624^{*} 
Homebuy 
2068.368 
8.241^{*} 
Statepcy 
0.360 
11.251^{*} 
Famsize 
780.981 
2.134^{**} 
1995 
224.029 
1.290 
2000 
27.884 
0.149 
Lamda
(IMR) 
1423.519 
2.689^{*} 
Rho 
0.164 

n 
16,760 

F 
241.95^{*} 

Adjusted
R^{2} 
0.471 

^{ *}Significant at the 1% level.
^{ **}Significant at the 5% level.
^{***}Significant at the 5% level.
Table 4
Estimation Results for the
Married Personal Income Equation
(dependent variable =
Personal Earnings)
Variable 
Coefficient 
tscore 
Constant 
6415.416 
17.545^{*} 
ED_56 
284.100 
0.991 
ED_78 
1.920 
0.007 
ED_9 
238.122 
0.827 
ED_10 
374.250 
1.327 
ED_11 
507.512 
1.832^{**} 
ED_12 
336.782 
0.979 
ED_Hsgd 
345.321 
1.423 
ED_Univ 
82.323 
0.334 
ED_BA 
2028.300 
8.007^{*} 
ED_Grad 
6270.188 
23.618^{*} 
MIG_MM 
14.550 
0.175 
MIG_MNON 
98.368 
0.440 
MIG_NONM 
997.990 
3.654^{*} 
MIG_NN 
297.523 
1.966^{**} 
MIG_ABM 
1112.558 
3.962^{*} 
MIG_ABN 
77.786 
0.101 
WC_Fed 
2428.852 
11.379^{*} 
WC_Local 
749.226 
6.593^{*} 
WC_State 
84.579 
0.525 
LF_Work 
5197.963 
27.379^{*} 
LF_Unemp 
1513.559 
6.089^{*} 
Black 
748.609 
6.650^{*} 
Indian 
134.426 
0.482 
Asian 
793.503 
5.587^{*} 
Hispanic 
13.274 
0.149 
Age 
11.689 
4.758^{*} 
Under18 
529.507 
10.081^{*} 
MSA 
915.309 
13.491^{*} 
Selfemploy 
931.276 
9.990^{*} 
Homebuy 
1277.754 
16.798^{*} 
Statepcy 
0.300 
24.607^{*} 
Famsize 
92.319 
1.744^{**} 
1995 
461.506 
7.190^{*} 
2000 
679.873 
9.799^{*} 
Lamda
(IMR) 
621.565 
3.027^{*} 
Rho 
0.795 

n 
95980 

F 
1450.42^{*} 

Adjusted
R^{2} 
0.484 

^{ *}Significant at the 1% level.
^{**}Significant at the 5% level.
Table 5
Estimation Results for the
Married Family Income Equation
(depedent variable = Family
Income)
Variable 
Coefficient 
tscore 
Constant 
21996.904 
10.538^{*} 
ED_56 
305.502 
0.218^{} 
ED_78 
429.783 
0.301 
ED_9 
157.097 
0.107 
ED_10 
98.911 
0.058 
ED_11 
487.994 
0.333 
ED_12 
2023.786 
1.305 
ED_Hsgd 
4495.666 
3.138^{*} 
ED_Univ 
7822.660 
5.430^{*} 
ED_BA 
15525.113 
10.604^{*} 
ED_Grad 
21973.639 
14.962^{*} 
MIG_MM 
1156.044 
5.255^{*} 
MIG_MNON 
384.954 
0.659 
MIG_NONM 
3809.466 
5.389^{*} 
MIG_NN 
1897.787 
4.740^{*} 
MIG_ABM 
4557.808 
6.199^{*} 
MIG_ABN 
335.830 
0.168 
WC_Fed 
512.926 
0.935 
WC_Local 
2357.779 
8.045^{*} 
WC_State 
1963.318 
4.740^{*} 
LF_Work 
3099.166 
6.306^{*} 
LF_Unemp 
1434.980 
2.240^{**} 
Black 
5554.957 
19.133^{*} 
Indian 
3469.821 
4.842^{*} 
Asian 
298.584 
0.802 
Hispanic 
4286.220 
16.786^{*} 
Age 
320.324 
40.466^{*} 
Under18 
7449.363 
49.628^{*} 
MSA 
4869.148 
26.902^{*} 
Selfemploy 
2363.281 
9.798^{*} 
Homebuy 
10163.315 
49.171^{*} 
Statepcy 
1.118 
35.063^{*} 
Famsize 
8795.796 
49.702^{*} 
1995 
1372.387 
8.270^{*} 
2000 
325.845 
1.832^{**} 
Lamda
(IMR) 
8807.076 
14.156^{*} 
Rho 
0.434 

n 
95,980 

F 
962.07^{*} 

Adjusted
R^{2} 
0.383 

^{ *}Significant at the 1% level.
^{**}Significant at the 5% level.
Table 6
Estimation Results for the
Structural Probit
(dependent variable =
Divorce)
Variable 
Coefficient 
tscore 
Constant 
0.10681 
0.913 
_{} 
0.00014 
16.978^{*} 
_{} 
0.00016 
121.822^{*} 



n 
112,740 

Chi squared (goodness of fit) 
54,877.90^{*} 

Pseudo R sqared (Mckelvey and Zavoina) 
0.385 

^{ *}Significant at the 1% level.
[1]For a very good explanation of this technique see Greene [2000].
[2] Full results will be supplied on request.
[3] Full results will be supplied on request.
[4] Stanley and Jarrell (1998, p. 963) in their metaregression on gender wage discrimination point to the importance of job classification variables that specify governmental employment in the estimation of female earnings.
[5] Full results will be supplied on request
[6] Full results will be supplied on request
[7] The coefficient estimates and tscores for the state dummy variables are omitted from Table 6 because of concerns for space. Full results will be supplied on request.
[8] Washington, D.C. was included in the estimations as a state and Kansas was excluded. Thus, 50 coefficients were estimated.
[9] In addition, several other goodness of fit measures all indicate that the “structural” probit provides superior statistical results to the estimates produced by the reduced form equation.