Economics coursework and long-term behavior and experiences of college graduates in labor markets and personal finance.

Allgood, Sam ; Bosshardt, William ; Van Der Klaauw, Wilbert 等

I. INTRODUCTION

If economics is a unique way of thinking, as many economists like to claim, some level of coursework in economics may provide enough specific knowledge and skills to lead students to make different choices in their adult roles as consumers, workers, and voters/citizens. Of course the same could be claimed for training in business and other areas, even though (or perhaps because) it could be argued that most content and training in business courses is more applied and pragmatic than what is presented in economics courses.

Some empirical evidence--though not really a great deal--suggests that training in economics and business is associated with different adult behaviors and outcomes. For example, Golec (1996) finds that mutual funds perform better if mutual fund managers have an MBA, and Chevalier and Ellison (1999) also report a positive correlation between mutual fund performance and the educational background of mutual fund managers. Black, Sanders, and Taylor (2003) found that economics majors earn almost 20% more than graduates in other social science majors, and about 10% more than those who major in business administration. In fields with large numbers of graduates, they find that only engineering majors earn more than economics majors. Similarly, Hecker (1995) reported that males majoring in economics had median earnings 3% higher than all majors for the age cohort of 25-34, and 53% more than average for the cohort aged 35-44. Craft and Baker (2003) find that lawyers with an undergraduate major in economics earn 13% more than otherwise comparable lawyers with other undergraduate majors, and economics is the only major with a statistically significant effect on earnings, (1) Hamermesh and Donald (2008), however, find that higher earnings by economics majors are partly due to economics majors working more hours than others.

The relationship between financial literacy/ knowledge and adult behavior and outcomes in financial markets has been addressed in several recent studies--many sponsored by Federal Reserve Banks. Hilgert, Hogarth, and Beverly (2003) and Braunstein and Welch (2002) find that knowledge about credit, saving, and investment is significantly related to recommended financial practices in cash-flow management, credit management, saving, and investment. Unfortunately, there is also evidence of widespread financial illiteracy in U.S. households, and particularly among elderly women (see Lusardi and Mitchell 2007, 2008).

Employee education programs have been found to affect workers' savings, retirement, and investment behaviors (Bayer, Bernheim, and Scholz 2009; Bernheim and Garret 2003). State mandates for precollege education in personal finance appear to succeed in at least exposing students to more information and sources on financial education, and students from states with such mandates report higher levels of savings and net worth as adults (Bernheim, Garrett, and Maki 2001; Tennyson and Nguyen 2001).

Much less research exists linking college coursework or majoring in economics to personal finance outcomes, (2) but Christiansen, Joensen, and Rangvid (2008) find that those trained in economics (in college courses or employee education programs) are more likely to invest in stocks than those who receive training in other areas. Guiso and Jappelli (2005) report that, among college graduates, having a degree in economics is positively associated with knowledge about financial assets. Chert and Volpe (1998, 107) concluded that business majors had better knowledge about personal financial literacy than other students, but that in general college students were "not knowledgeable about personal finance.'' (3)

In this paper we use a large and unique data set to investigate the relationships between taking college coursework in economics or majoring in economics with long-term labor market and personal finance behaviors and outcomes. Our information on long-term behaviors and outcomes is drawn from a detailed questionnaire we mailed in January 2003 to over 25,000 graduates who attended four public universities in 1976, 1986, or 1996. Hamermesh and Donald (2008) point out that most efforts to investigate the link between students' choice of majors and earnings use data collected only a few years after graduation from college, but some of the subjects in our sample graduated from college more than 25 years before completing our survey. We compare the graduates in three groups of majors--economics, business, and all other majors--and also by the number of economics courses the graduates completed.

While most of the earlier research on labor market outcomes has focused on differences in earnings, we are able to investigate earnings and six other outcomes that have important consequences to individuals and families. Including information about layoffs and hours worked, for example, provides a richer description of a person's labor market experience. (4)

We explicitly account for gender differences in our analysis of the labor market outcomes and behaviors, because Katz, Goldin, and Kuziemko (2006) report that, although women are now more likely to graduate from college than men, women still comprise less than a third of economics majors, and in recent years that percentage appears to have fallen slightly (Siegfried 2007). One possible reason for this outcome is that the returns to studying economics may differ for women and men. (5)

In addition to labor market behaviors, we analyze variables covering such personal finance decisions as the number of credit cards held, payment patterns on the credit cards, homeownership and levels of home equity, savings levels and motivations, portfolio allocation, and life insurance coverage. We find large differences between majors for many of the labor market and personal finance variables.

In line with much of the existing literature, our attempts to account for selection bias in the choice of college majors, as well as the number of economics courses students take, are based primarily on procedures for dealing with selection-on-observables, where students with alternative sets of characteristics non-randomly self-select into different college majors and thereby into different course selections. The potential that our findings could also be influenced by selection based on unobservables cannot be dismissed, so we must caution against interpreting the partial correlations we estimate as purely causal. Nevertheless, the relationships we discover between economics majors and coursework and our variables on long-term outcomes and behaviors in labor markets and personal finance are interesting and important, and our findings contribute to this relatively new and underdeveloped area of research.

In Section II of the paper we provide a description of our data set. The basic statistical model and empirical results for labor market outcomes are presented in Section III. Results on personal finance decisions are reported in Section IV. In Section V we discuss statistical issues related to the potential endogeneity of our measures of economics training. Section VI provides a brief conclusion and summary of results. Appendices A and B provide additional information on our attempts to address issues of non-response bias and endogeneity.

II. THE DATA AND DESCRIPTIVE STATISTICS

Our survey on labor market and personal finance outcomes and behaviors was mailed in January 2003 to graduates from four public universities--Florida Atlantic (FAU), Nebraska-Lincoln (UNL), North Carolina (UNC), and Purdue--who attended the schools in 1976, 1986, or 1996. For each of the annual cohorts, the sample included up to 1,000 students from each of three different groups at each of the four schools. The three groups were based on students' final major, which we classified as economics, business, or general (i.e., anything other than economics and business majors). Business majors include all students with majors originating in business schools except economics. Whenever a group of majors at a school for a given year was larger than 1,000 students, a random sample of 1,000 students was drawn. That was usually the case for the general majors, and often the case for business majors. For all of the economics annual cohorts at all four universities, the number of majors was less than 1,000, so surveys were mailed to all of the economics majors enrolled in these years.

In addition to many specific questions about their labor market experiences, personal finance decisions, and voting behavior and other political activities since leaving school, the survey included questions on the graduates' impressions about their undergraduate coursework in economics and other fields, and general questions about their background. (6) In short, the survey was long, detailed, and in several places highly personal or even intrusive.

From the 25,292 surveys mailed, 1,313 were returned because of invalid addresses. We received 2,165 completed surveys, for an overall response rate of 9.0% (excluding surveys returned because of bad addresses), a rate that is typical or somewhat better than usual for this kind of mail survey. The response rate by school ranged from 5.8% to 11.4%, and also varied across the three groups of majors: 13.1% for the economics majors and about 8.5% for business and general majors. The response rate for the 1996 cohort was 10.0%, for the 1986 cohort 8.3%, and for the 1976 cohort 8.8%.

Transcript data were obtained from registrars' offices at the four universities, providing basic demographic information including gender and race, as well as information on students' overall GPA, semester GPAs, economics courses taken, and grades in economics courses. The transcript information was relatively easy to obtain for the vast majority of survey recipients at all four universities for the 1986 and 1996 cohorts. For the 1976 cohort, however, transcript information was only available in electronic formats at two institutions. At the other two schools we attempted to collect transcript information from copies of printed records for every business and general major who returned a survey, and for 100 additional transcripts in each of these two groups of majors, chosen at random from the non-respondents. We were successful in obtaining transcript information for all economics majors. For the entire mailing sample, transcript information was available for 23,127 former students, including all but six of the survey respondents.

Unfortunately, there are differences in the data that the four schools record on transcripts. For example, not all schools provide scores on college entrance exams, such as the SAT, and the schools that do not provide entrance exam scores on transcripts also provide very little data for students prior to matriculation (such as high school GPA). Consequently, our pooled data analysis could not use information that predated a student's enrollment at one of the universities.

Table 1 lists and describes all of the variables used in our estimations. (7) The first set of variables is based on a set of questions concerning labor market outcomes and behavior, including post-graduate education. Specifically, these variables include measures for employment status, levels of annual compensation, and the form of that compensation (salary vs. commission, etc.) at the time of the survey. Other variables in this set provide information on past layoff experience, self-employment experience, graduate education, and the individual's expected retirement age.

The second set of variables in Table 1 deals with various personal finance decisions, including home ownership and net equity, current and expected (at retirement) savings, personal motivations for saving, investment portfolio allocations, life insurance coverage, the number of credit cards held, and the monthly pay-off rate on those credit cards.

The third set of variables in Table 1 lists and describes our explanatory variables. Binary variables are used to indicate which school was attended, the different time cohorts (1976, 1986, and 1996), the three groups of majors (economics, business, and other), the U.S. geographic region of respondents' current residence, gender, race/ethnicity, and family status. Some responses were missing for the variables indicating race/ethnicity and family status. To avoid deleting these observations, missing values were replaced with a value of zero and then an indicator variable was used to identify each record with missing values. (8) For brevity, the estimated coefficients for the indicator variables are not listed in the regression tables, but they were similar to the estimated impacts at the means of the non-missing values. Estimations based on samples that only included non-missing values for the explanatory variables resulted in comparable estimates for all explanatory variables.

The EconCourses variable is a count of the number of economics courses on a student's transcript, excluding repeated courses but counting courses regardless of how well or poorly the student performed in the class. We also used the squared value for EconCourses as an explanatory variable. CumGPA variable is a graduate's cumulative GPA at the end of their undergraduate program.

Table 2 provides descriptive statistics for all individuals to whom we mailed questionnaires and for whom we obtained transcript data (labeled as Sample I, this was our target population), and for the subsample used in our estimations, comprised of individuals who returned the survey (Sample If). We report descriptive statistics with and without weights that were constructed to account for the non-random sampling method (based on college majors), described earlier, which we used to construct our mailing list. (9)

For Sample I, unweighted means show that 22% of the former students were enrolled in 1976, and 39% each in both 1986 and 1996. Business and general majors are each 45% of the sample, with the remaining 10% majoring in economics. The graduates (in all majors) were 46% female and 7% Black. On average, the students in Sample I took 2.9 economics courses and had a cumulative GPA of 2.97 (on a 4.0 scale).

Comparing these means to the weighted means shows the influence of the non-random nature of our sampling design, with the general majors group being underrepresented and business and economics majors overrepresented. Because business and economics majors take more economics courses than students in the general group of majors, the average number of economics courses taken in the actual target sample (2.9) is higher than in the weighted target sample (1.7). Economics and business majors also comprise smaller proportions of our sample when the weights are used. The weighted adjustments do not have large effects on the mean values for most other variables, however.

If survey non-response is random across sampled individuals, we would expect average characteristics for Samples I and II to be very similar. For most variables, however, the two samples differ in average values, reflecting heterogeneity in response rates across subpopulations. For example, survey respondents (Sample II) were more likely to be economics majors, white, and to have slightly higher GPAs. The differences between the two samples are statistically different at the 10%, or higher, confidence level for all variables except General, Female, and Northeast.

If non-response rates vary across the groups defined by different values of the characteristics reported in Table 2 but are random within each subgroup (i.e., conditional on the values of these characteristics), then as long as we adequately control for the heterogeneity in these characteristics (and thus for potential differences in behavior across groups) the validity of the results presented below should not be compromised due to non-random non-response. If response rates vary with unobserved characteristics that are related to the dependent variables, however, or with the values of these variables themselves, that could lead to biased estimates in which some of the differences in labor market and personal finance decisions and outcomes that we attribute to college major or course-taking choices are actually due to non-random survey response. We investigate the potential for sample selection bias due to non-random nonresponse in more detail in Section V, where we discuss estimates based on alternative statistical approaches used to account for such bias. Generally these estimates offer little evidence of significant non-response biases.

Table 3A provides unweighted sample means for survey respondents (Sample II) on all of the explanatory variables, reported separately for the three groups of undergraduate majors. (10) The three annual cohorts are all well represented in Sample II, although the 1996 cohort has the highest representation. UNL and Purdue alumni each represent slightly less than a fourth of the sample, FAU about 15%, and UNC graduates the remaining 39%. About 43% of the survey respondents were business majors, a similar number were general majors, and 13% were economics majors. The average number of economics courses taken by all students in Sample II was about 3.4, with differences across majors clearly reflecting degree requirements for economics and business majors. A typical business major took four economics courses and general majors took about one class, on average. Economics majors averaged almost nine courses. Across all three groups of majors, cumulative GPA at the end of the last term shown on the transcripts was 3.10 on a 4.0 scale.

Females comprise about 45% of Sample II, but are underrepresented in the economics and business majors (24% and 38%, respectively), and overrepresented in the general majors (58%). In contrast, minorities represent about 5% of graduates in all three groups of majors. In each category of majors, about 70% of respondents were married or living with someone when they completed our survey, and reported having one child, on average. Many graduates live in states where they were born, raised, and went to college, so given the locations of the four universities used in our study, over 85% of the survey respondents lived in the South and Midwest in 2003.

Table 3B reports summary statistics for the labor market and personal finance variables. (11) In Sample II (all survey respondents, regardless of major), almost 40% reported working more than 50 h a week; 34% were paid at least partly based on commission; 21% have experienced a layoff; over 70% reported earning an annual wage or salary income of more than $40,000, and over 20% over $100,000. More than a fourth reported at least some current or past self-employment activity. Nearly 40% had completed a graduate degree--mostly MBAs (11%) or other Masters degrees (19%), with smaller numbers of law degrees (6%) or a Ph.D. (3%). Our sample of respondents hopes to retire early, with an average expected retirement age of 60.

Over three-fourths of the sample owned a house, with 31% of the homeowners reporting a net home equity under $50,000, 47% with equity values between $50,000 and $250,000, and 22% over $250,000. Average and median current savings were about $600,000 and $65,000, respectively, reflecting a right skew in this distribution. For households with positive savings, investments were mostly divided across money market accounts, individual stocks, and mutual funds, with about 27% holding mutual funds and 16% owning individual stocks. Respondents expect to retire with average assets over $2.1 million (median $1 million). Although saving for precautionary reasons is given slightly more importance, the average importance across the various reasons to save (to buy a home, car, or vacation; for retirement, education expenses, or bequests; precautionary motives; or repaying of debt) is similar. The average (median) number of credit cards held by respondents was about 3.2 (2), and almost 60% reported paying off all of their credit card balances every month over the preceding year. (12) Nearly 90% of the respondents had life insurance, of which 60% was self-purchased (not provided by employers).

The data show important differences in labor market and personal finance decisions and outcomes between those who majored in economics and those who majored in business or other subjects. On average, economics majors work more and earn more than business majors, who in turn work and earn more than the graduates with other majors. A higher fraction of both economics and business majors receive at least part of their annual earnings in the form of commissions/bonuses. Business and economics majors are more likely to complete MBA degrees and economics majors are more likely to have attended law school relative to both other groups, but a much higher proportion of general majors obtain other Masters degrees and Ph.Ds.

Economics and business majors are more likely to invest in individual stocks, have one fewer credit card (on average), and are more likely to pay off their credit card balances each month. They are also more likely to have life insurance, but especially employer-provided life insurance--they are actually less likely to purchase policies as individuals, quite possibly because they are more likely to work for employers who provide policies as a fringe benefit. The average (median) savings of economics majors was $2.1 million ($100,000), for business majors about $600,000 ($80,000), and for the general majors $174,000 ($50,000). The same ranking holds for the average amount graduates expect to have saved by the time of retirement. There were only small differences in the reported reasons for saving across the different majors.

Some variation in these outcomes may reflect differences in individual characteristics across the groups of majors. For example, economics majors are more likely to be from one of the later annual cohorts, more likely to have attended UNC, and more likely to be male. To control for such differences in observed characteristics, we next report results using regression analysis, and then we investigate the relationship between these outcomes and the number of economics courses taken, exploring differences in economics training both across and within each group of majors. (13)

III. EMPIRICAL SPECIFICATIONS--LABOR MARKET OUTCOMES

Given the binary nature of all but one labor market outcome variable, we begin by estimating a set of standard probit equations relating each of those dependent variables to a set of individual characteristics, including a measure of economics training. (14) RetireAge is treated as a continuous variable and analyzed using ordinary least squares (OLS). In all estimations sample weights were used to account for the non-random sampling design. We first discuss estimates for specifications that include indicators for Economics and Business majors as explanatory variables (with General majors the excluded group). We then replace the variables for majors with variables indicating the number of economics courses completed by respondents (EconCourses) and the square of this variable (EconCourses (2)).

For each labor market outcome we present estimates based on two different sets of control variables. The first, corresponding to panel A of each table, includes two cohort and three university dummies, as well as binary variables for gender, race, and ethnicity. The second, shown in panel B, includes an extended set of controls, adding regional indicators of current residence and indicators of the individual's marital status and number of children. More importantly, the regression includes a measure of the individual's overall ability. One of the main and most frequently discussed sources of endogeneity bias in estimating the effect of education on labor market outcomes is heterogeneity across individuals in unobserved ability. Individuals of higher ability may self-select into different majors and may take more or fewer economics courses. At the same time, one would expect those with higher ability to have better labor market outcomes and to exhibit different personal finance behavior even in absence of any difference in economics training. If such ability differences are not controlled for, they will be incorrectly attributed to their economics background.

In an attempt to overcome ability biases in our estimations, we control for the individual's overall performance in college courses. (15) Using the student's final GPA score in college may be problematic due to differential grading practices across course fields, universities, and cohorts (Johnson 2003). To account for grade inflation and differential grading practices, we instead measure each student's performance relative to all other students in the same course or subject. We do so by regressing, separately for each university, all individual student course grades on course subject dummies, (16) course level (whether upper level course), year dummies, and individual fixed effects. The estimated individual fixed effects serve then as our ability measures.

A. The Economics Major

Table 4 presents estimates using binary indicators for Economics and Business majors as explanatory variables. For the probit estimations marginal effects at the variable means are reported, with z-scores in parentheses. For the OLS regressions coefficients and t-statistics are reported.

The economics major is positively and significantly associated with earning more than $40,000 or more than $100,000 a year, but also with working more than 50 h/week and with working partly for commission/bonuses and having engaged in some form and level of self-employment. It is negatively related to completing a graduate degree. The effects on layoff experience and planned retirement age are not significant.

The magnitudes of the earnings, self-employment, and graduate education differentials are relatively large and important.

Compared to the general majors, economics majors have a 17% higher chance of earning more than $40,000 a year, a 14% greater chance of earning more than $100,000 a year, and an 8% higher chance of engaging in self-employment. They are considerably more likely than general majors to work partly for commission/bonus (20%), but almost 8% less likely to complete an advanced degree.

We tested the null hypothesis that the marginal effect in these labor market outcomes for economics majors is equal to the marginal effect for business majors, and were unable to reject the null hypothesis in all but two cases: (1) the likelihood of having ever been laid off is about 7 percentage points higher for business majors, and (2) business majors are 16% less likely to pursue an advanced degree than general majors, which is twice the gap between the economics and general majors. The general (and not particularly surprising) conclusion is that economics majors are far more like business majors than general majors in terms of labor market outcomes.

Adding additional explanatory variables, as reported in panel B of Table 4, none of the point estimates for Economics and Business are substantially affected. As mentioned earlier, a student's ability and the other included variables may be directly related both to the labor market outcomes considered and to graduates' decisions to major in economics, business, or other subjects, and therefore to the number of economics courses they take. To the extent that is true, conditioning on these variables could reduce endogeneity biases in the estimates. And to the extent that the additional regressors help explain some of the variability in the outcome variables, their inclusion could increase the precision of our estimates. Although the coefficients associated with our ability measure and some of the other included covariates often were highly significant, we found little effect of adding these variables on the estimated associations between college major and labor market outcomes. The robustness of the estimates can be viewed as evidence that biases due to non-random selection into the economics and business majors (at least based on observables considered here) are small.

B. Economics Coursework

The estimates reported in Table 5 indicate that taking more economics courses is associated with a higher probability of earning more, working more than 50 h/week, being paid a commission, and having experienced a layoff. Economics coursework is negatively related to completing a graduate degree and to expected retirement age. As indicated by tests of the joint significance of its two associated coefficients (bottom two rows of Table 5), the number of economics courses taken is significantly related with all outcomes except self-employment. In the earnings, hours worked, commission, and graduate degree estimations, the results are predictably in line with the results from Table 4 for both economics and business majors, who take more economics than the group of other majors, on average. The more interesting cases concern the other dependent variables, for which signs and/or significance results for the economics and business majors appear different in Table 4. There are considerably more business majors than economics majors in Sample II, so the results for the laidoff variable in Table 5 reflect the finding that business majors were significantly more likely to have experienced a layoff, while economics majors (who take more economics courses than business majors) were less likely (though not significantly) to have experienced a layoff. Similarly, business majors were significantly more likely to report an earlier planned retirement, while the coefficient for economics majors was negative but insignificant. The self-employment outcome was significant for economics majors and insignificant for business majors, and is insignificant in Table 5.

Although the marginal effects of the squared term for economics coursework are only significant in three cases, there is general evidence of a non-linear association in which the effects of the number of economics courses taken and labor market outcomes become weaker as the number of economics courses increases. Specifically, while one additional economics course is initially associated with considerably higher probabilities of earning more than $40,000, earning more than $100,000, working more than 50 h, being paid in part based on commission, or having been self-employed, these marginal effects turn negative at various levels, ranging from 7 to 10 courses. Put differently, the last economics course or two in the economics major may not have large or even positive effects on these outcome variables.

The marginal effect estimates for the specification with additional control variables, reported in panel B of Table 5, again shows that none of the point estimates--in this case for economics courses or courses squared--is substantially affected. So as noted above in the discussion of Table 4, there is again little evidence of selection bias based on these observables.

C. Gender and Economics Majors and Coursework

Almost half of our sample is female, but only a fourth of the economics majors in our sample are women. (17) To see if the labor market behaviors associated with majoring in or taking courses in economics (relative to those associated with a general major or to taking no economics courses, respectively) are different for women and men, we interacted the indicators for majoring in economics and business, or the variables for economics courses, with an indicator variable to identify females. The indicator was also included directly to capture average differences in outcomes and behavior across genders. All of these estimates, reported in Table 6, use the full set of regressors used in the B panels in Tables 4 and 5.

Coefficients for the gender indicator are significant in four regressions (Salary >40K, Salary >100K, Hours, and Commission). In other words, even among those with a general degree and those with zero economics courses taken, women in our sample had different labor market outcomes: relative to male graduates they earned less, worked fewer hours, and were less likely to be paid a bonus/commission. However, none of the interaction terms are significant at the 5% level, and only one is significant at the 10% level. The one marginally significant case deals with layoffs: while male graduates who majored in economics are equally likely to have experienced a layoff as male graduates with a general major, female graduates who majored in economics are Jess likely to have experienced a layoff than their counterparts with a general major. This may reflect a higher layoff risk for male economics majors, or greater exposure to risk due to higher labor market participation rates relative to female economics majors. But on most outcomes, our results suggest that studying economics is associated with similar changes in labor market outcomes for men and women, even over a data set that reaches back over 25 years. (18)

D. Advanced Degrees

Black, Sanders, and Taylor (2003) find that students who complete a bachelors degree in economics are more likely to pursue a higher degree, and Nieswiadomy (1998, 2006) reports that economics majors comprise a large share of law school applicants and do especially well, compared to other majors, on the LSAT entrance exam. We were therefore surprised to find that the economics majors in our sample were less likely to complete advanced degrees. In Table 7, we report marginal effect estimates from weighted probit regressions for different kinds of post-graduate degrees. Economics and business majors in our sample are more likely than general majors to pursue MBAs (by 18% and 11% points, respectively), but both are less likely to pursue other Masters degrees (by 9% and 13% points) and the Ph.D. (by about 2% points each). (19) Both economics and business majors in our sample were equally likely to obtain a law degree as other majors.

Using the number of economics courses rather than Economics major as an explanatory variable, there is only one change in these results: taking economics classes increases the likelihood of obtaining a law degree, but at a decreasing rate. This finding primarily reflects a higher law school attendance rate among General majors who take more economics courses. The marginal effect is positive until about nine courses, after which it turns negative. The likelihood of obtaining an MBA increases through seven classes, but then declines. Relative to those who took no economics courses, those who took a few economics courses are more likely to pursue a Law degree or an MBA, but less likely to obtain other Masters degrees or the Ph.D. These findings may reflect the higher salaries earned by economics and business majors, which increase the opportunity cost of going to graduate school. They are also consistent with the falling share of U.S. students pursuing doctorates in economics and business. Unfortunately, it is difficult to say whether these findings reflect specific preferences or characteristics of graduates from the four universities in our sample, or apply to college graduates more generally.

IV. EMPIRICAL SPECIFICATIONS--PERSONAL FINANCIAL DECISIONS

Tables 8-11 present our estimates of the association between our measures of economic training (college major choice and coursework in economics) and personal finance decisions and outcomes, grouped in four categories: real estate, savings, financial investments, and credit cards and life insurance. In all of these estimations, we again use the full set of explanatory variables featured in Panel B of Tables 4 and 5. (20) However, because differences in personal finance decisions may simply reflect differences in the average incomes earned by graduates with different majors, we report two sets of estimates for each variable: one controls for self-reported income using a set of nine dummy variables corresponding to earnings intervals; the other estimation does not control for income.

A. Real Estate

Table 8 reports estimates for three real estate variables: owning a home (vs. renting), and for those who own rather than rent having net equity in the home that is either less than $50,000 or more than $250,000. Majoring in economics or business are both associated with a 4% point increase in the probability of homeownership, with the difference for the business major significant at the 5% level. After controlling for income, however, these differences disappear. Similarly, without controlling for income we find that taking more economics courses is initially (less than five classes) associated with an increased likelihood of homeownership; but for higher numbers of courses the marginal effects turn negative. After controlling for income the relationship between economics coursework and home ownership becomes weaker but retains statistical significance.

The story is a little different in terms of net equity. Not controlling for income, majoring in economics or business, or having taken at least some economics courses, are all associated with higher equity. Controlling for income, most of those results become insignificant, but business majors and those who took some economics are still less likely to have equity of less than $50,000. Particularly with regard to estimations that do not control for income, our results are consistent with those reviewed by Martin (2007), who reports a positive relationship between financial education programs and home ownership.

B. Saving

In Table 9, we consider self-reported levels of savings and several possible motivations for saving (namely, short-term vs. long-term objectives, precautionary motives, and repaying debt). Economics and business majors in Sample II saved approximately 57% more than the group of other majors. Not controlling for income, economics majors expect to save more than both business and other majors when they retire. These effects are smaller after controlling for income, but still significant. Bernheim, Garrett, and Maki (2001) found similar effects for those who completed a secondary curriculum on consumer decision making; our results are based on more general forms of college-level economic and business education.

Economics majors viewed specific short-term reasons for saving (such as buying a home or car, or to pay for a vacation) and precautionary motives for saving as more important than both business majors and the group of other majors. Both the economics and business majors considered long-term reasons to save (including retirement, future education expenses, and investments) more important than the group of general majors. Both the business and economics majors also place less importance than general majors on saving to repay debt. In a sample of British youth (ages 16-21), Furnham and Goletto-Tankel (2002) found that more education led to more positive attitudes about saving. Our results support the few earlier studies that suggest the specific field of study students pursue, as well as overall level of education, is important.

C. Financial Investments

Households make important financial decisions about how to allocate savings across various kinds of financial assets and investments. Martin (2007) suggests that two reasons many households do not own equities are: (1) a lack of information about the existence of equity markets, and (2) the presence of significant fixed costs (time, money, psychological) that create barriers to entering equity markets. Apparently a wide range of educational programs--school-based, retirement and employee education seminars, and so forth--may overcome both of these. Our results in Table 10 show that majoring in economics or business, or taking undergraduate economics courses, are correlated with how individuals allocate their investments, particularly in terms of owning stocks and using money market funds. These positive associations hold even after we control for differences in income, and they support the findings of Christiansen, Joensen, and Rangvid (2008), noted earlier, that people trained in economics are more likely to invest in stocks. At the same time, business majors and graduates who take some economics courses were significantly less likely to leave funds in checking and savings accounts at banks, presumably in pursuit of higher returns--but these results weaken somewhat after controlling for income.

D. Credit Cards and Life Insurance

Lee and Kwon (2002) found that holding of credit cards increases with education, and they hypothesize that those with more education are more likely to use credit cards for convenience in making purchases, rather than a means of financing consumption. We analyze the association between economics coursework and majors and the number of credit cards held using a Poisson model, with estimates of the incidence-rate ratios (IRR) presented in the first two columns of Table 11. We find the average number of credit cards held by economics majors was 62% of the number held by the general majors, with the comparable number for business majors being 67%. These values change very slightly and are still significant after controlling for income. The estimates of the IRR in the first column of Table 11 imply that, evaluated at the mean number of economics courses, an additional economics course reduces the incidence rate (the number of cards held) by a factor of 0.975, while evaluated at six courses the incidence rate falls by a factor of 0.945.

Estimates in the third and fourth columns of Table 11 show that economics coursework and majoring in business are positively associated with completely paying off credit card balances each month--the effect of the courses squared variable does not result in an overall negative effect until 12 courses, which is beyond the level taken even by the vast majority of economics majors.

Bernheim et al. (2006) offer evidence that people employed at Boston University, including professors, have too little life insurance, on average. We find that economics majors are about 5% more likely to have life insurance than the group of General (non-business, non-economics) majors--either employer provided or self-purchased. Business majors are about 2% more likely to have insurance than the group of General majors. There is no significant variation in those who self-purchase life insurance across majors or economics courses taken. Given the higher overall life insurance coverage rate for economics majors, this implies that they are more likely to have insurance provided by employers. However, the differences across majors and economics course work can be entirely attributed to the higher associated income levels.

V. PERFORMANCE IN ECONOMICS COURSES

In addition to our measures of economics training in college (economics or business majors, or the number of economics courses taken), we also investigated whether a student's academic performance in economics courses was associated with differential labor market or personal finance behavior. Put differently, in addition to exposure to economics, does it matter how well the content of economics courses is understood, as measured by course grades? We analyzed this by including in our models a normalized college GPA based solely on the grades obtained in economics courses taken. Specifically, we computed individual fixed effects from course-grade regressions identical to those used to compute the general ability measure discussed in Section III, but with the estimation sample including only grades in economics courses taken. As in Section III, our measure accounts for differential grading practices across schools, course subjects, and cohorts.

As shown in Table 12, these measures of how well students did in their economics courses had relatively little effect on our estimates for college major and economics courses taken. (21) Perhaps more importantly, with only a few exceptions the estimates indicate that how well students do in economics courses is not significantly related to outcomes or behaviors after controlling for college major or number of courses taken. (22) The cases in which there was a statistically significant relationship with higher economics grades were: (1) a higher rate of receiving some income paid as commission, (2) greater savings, (3) holding fewer credit cards, and (4) a higher pay-off rate on credit cards. But most of our results suggest that it is the general exposure to economics, rather than performance in the courses measured by the 22. Comparable results were obtained when relating outcomes to the number of economics courses taken. course grades, that matters for future outcomes and behaviors. (23)

VI. STATISTICAL ISSUES AND INTERPRETATION

We laced two major statistical issues in conducting and interpreting the results from these estimations: non-random survey non-response bias and the endogeneity of choice of major or economics courses taken. Although differences between respondents and non-respondents are to some extent captured by differences in observed characteristics, including these variables as linear controls in the regressions may not adequately capture their impact on these outcomes if the linear controls are too restrictive. And individuals in the different groups in our sample may also differ in terms of unobserved traits.

We controlled for possible non-response bias using propensity score methods and a control function approach. Our results were not altered using either method, so we do not present those estimates here, but they are available on request. We also experimented with an instrumental variable approach to account for possible selectivity in unobserved traits, but unfortunately were not successful in finding an instrument that was both credible and sufficiently strong (see Appendix A for further details).

Two of our key explanatory variables--majoring in economics and the number of economics courses taken--may be endogenous with respect to the labor market and personal finance behaviors and outcomes we are investigating. Earlier studies on the effects of studying economics on earnings or other labor market outcomes have ignored the issue of endogeneity (Black, Sanders, and Taylor 2003; Christiansen, Joensen, and Rangvid 2008; Hamermesh and Donald 2008). (24) This is also true for most previous studies relating financial literacy to financial behaviors. In Sections Ill and IV, to account for possible selection bias, we included a set of controls which included a measure of individual ability. While this reduces the potential for bias due to selection-on-observables (where students with alternative sets of characteristics non-randomly self-select into different college majors and into different course selections), it does not deal with potential selection-on-unobservables. We therefore attempted instrumental variable methods to address the potential endogeneity of these variables, but again we were not able to find an instrument strong enough to provide sufficiently precise estimates for causal inference, as discussed in Appendix B. Therefore, while the associations we report are important, we suggest caution in claiming a causal relationship.

VII. CONCLUSIONS

In arguing that outcomes in competitive markets were efficient, neoclassical economics assumed (often implicitly) that individuals innately knew or learned by experience how to make good decisions as consumers, workers, and voters. Many educators, civic leaders, and business and financial executives have long questioned that assumption, and over the past decade a growing body of research by economists has raised concerns about poor decision making by households in which adults have little or no training in economics or personal finance. Understanding how different kinds of education affect household behavior is therefore an important concern for economists and policy makers.

We find a wide range of evidence that relates studying economics at the undergraduate level to different outcomes and behaviors than those reported by graduates who majored in other fields. To summarize, economics coursework and majoring in economics are significantly related to higher levels of earnings, home equity, and savings. They are also associated with working more hours and negatively related to completing graduate degrees, except the MBA. Among college graduates with positive savings, those with more economics coursework invest more in individual stocks and money market accounts, and are more likely to have employer-provided life insurance. They have fewer credit cards, which are more often paid in full each month. These relationships appear not to be strongly related to how well a student did in the economics courses they completed.

With a few interesting exceptions, most of these findings also hold for graduates who majored in business. The exceptions are that business majors are more likely than economics majors to have been laid off by an employer, and less likely than economics majors to have been self-employed or to have employer-provided life insurance. Both business and economics majors are less likely than other majors to complete a graduate degree (except the MBA), but business majors are about twice as unlikely to do that as economics majors. Economics majors expect to save even more than business majors by retirement, on average, and they view both short-term savings goals and precautionary motives for saving as more important than business majors. On the other hand, among those who own their homes, business majors are somewhat less likely than economics majors to have net equity of less than $50,000.

Finally, our results suggest that exposure to economics through course-taking is more important for later outcomes than how well a student did in the economics courses they completed.

APPENDIX A

Non-response Bias

We analyze the sensitivity of our estimates using two different approaches to control for possible survey nonresponse bias. First we explored correcting for non-random non-response using propensity score methods (Rosenbaum and Rubin 1983). We used the parameter estimates from a probit model to estimate for each individual the probability of being included in the sample of survey respondents (the propensity score). The propensity score estimates were then used to form weights in weighted least squares regressions (Horvitz and Thompson 1952; Little and Rubin 1987; Wooldridge 2002). The weights are defined as P(1 =1)/P(1=1/X ), where P(1=1) represents the proportion of students in Sample I (mailing sample) who were included in our sample of survey respondents (Sample II), while P(I = l/X) is the estimated propensity score for an individual with observed characteristics X.

Under certain conditions, the use of these weights corrects the distributions of all variables so that these distributions become representative of the mailing population. Specifically, the approach relies on a conditional independence assumption in which, conditional on a set of variables, being in the sample of respondents can be treated as random (independent of the values of all other variables). The results we obtained using this approach were very similar and not significantly different from the unweighted linear probability model estimates.

Our second method for correcting for survey nonresponse bias is a control function approach. This involves including a selectivity bias correction term in the regression (Heckman and Robb 1985), approximated by a polynomial in the estimated propensity score (Newey, Powell, and Walker 1990; Vella 1998). Adding a linear, quadratic, or cubic polynomial in the propensity score did not quantitatively or qualitatively alter the estimates. Therefore, in this paper we only report estimates that ignore potential selectivity bias from survey non-response. Estimates obtained with both response bias correction methods are available on request.

Finally, we explored an instrumental variable approach similar to that proposed by Hamermesh and Donald (2008) for addressing survey non-response biases. They used an indicator of current membership in a university's alumni association as instrumental variable for survey response, based on the idea that those with closer ties to the university are more likely to participate in the survey. We did not have access to information about alumni association membership for our sample, but we explored using the distance from individuals" current residence (in 2003) to the college attended as instrumental variable for survey response. We constructed two dummy variables: living in-state and close to campus, and living in-state but far from campus--which left living out of state as the residual category. These instruments had a negative effect on survey response, which clearly raises questions about the validity of the instrument.

Item non-response rates varied between 1% and 12% for labor market outcome variables, with an average of 4%. For the personal finance variables item non-response rates varied from 2% to 32%, with an average of 7%. Therefore, claiming that the estimates reported in the paper are unbiased relies on an implicit missing-at-random (MAR) assumption, in which the missing status of an observation for the dependent variable is randomly assigned conditional on the observed values of all regressors (i.e., within subgroups defined by the conditioning variables). Note that this assumption allows the rate at which the dependent variable is missing to be different across the values of the included covariates, as it only imposes it to be constant within each subgroup. Although this assumption is generally imposed in empirical studies, it is rarely discussed explicitly and hard to test without auxiliary information.

APPENDIX B

Endogenous Regressors

Our attempts to control for endogeneity biases went well beyond the addition of controls to account for possible selection-on-observables. Our identification strategy for estimating causal effects requires a variable that is correlated with the number of economics courses taken and with the choice of economics or business as major, but that is not directly related to the labor market or personal finance outcomes we use. An important determinant in the decision to major in economics or to take additional economics courses is the student's relative performance in the first economics course taken. From a student's perspective this variable contains a significant random component beyond the stochastic element in the student's test/grade performance in the course, because it depends on the quality of the instructor and several other course characteristics, including course enrollment, class size, the use and quality of teaching assistants, and so forth. These factors are likely to vary over time and across universities, and across course sections at the same university in the same semester or quarter.

We exploit this random component while directly controlling for students' overall ability as measured by the individual's course-grade fixed effect (as discussed earlier), using three different measures of an individual student's relative performance in their first economics course. The first measure is the student's grade in the first non-transferred course in economics (usually Principles but otherwise lowest numbered course) relative to the student's overall cumulative GPA at the end of the semester in which the course was taken. The second measure is the same except that it is based on the average grade in all non-transferred economics courses taken during the first semester when an economics course was taken by the student. The third is similar to the first except that it measures the student's performance in the first economics course relative to the average course grade of all students who took the course. Our results are not sensitive to the specific instrument choice, but unfortunately while the instruments all have the expected positive effects on total economics courses taken and economics major choice, our instruments are not strong enough, causing our second stage estimates of EconCourses and Economics to have large standard errors.

We also considered as an instrument the overall popularity of the economics major among those who were already in college (non-freshmen) at the time the student entered college, measured by the fraction of economics majors among total degrees awarded during the first three years the student was in college. Even though there was considerable variation in the popularity of the economics major by college entry year and school, after controlling for cohort and university fixed effects the instrument was too weak to draw precise inferences.

First and second stage estimates of all instrumental variable estimations are available from the authors upon request.

REFERENCES

Allgood, S., W. Bosshardt, H., W. van der Klaauw, and M. Watts. "What Students Remember and Say about College Economics Years Later." American Economic Review: Papers and Proceedings, 94, 2004, 259-65.

Altonji, J. G. "The Effects of High School Curriculum on Education and Labor Market Outcomes." Journal of Human Resources, 30, 1995, 409-38.

Arcidiacono, P. "Ability Sorting and the Returns to College Major." Journal of Econometrics, 121, 2004, 343-75.

Barron, J. M., and M. E. Staten. "Usage of Credit Cards Received Through College-Marketing Programs." Journal of Student Financial Aid, 34, 2004, 7-28.

Bayer, P. J., B. D. Bernheim, and J. K. Schotz. "The Effects of Financial Education in the Workplace: Evidence from a Survey of Employers." Economic Inquiry, 47, 2009, 605-24.

Bernheim, B. D., and D. M. Garrett. "The Effects of Financial Education in the Workplace: Evidence from a Survey of Households." Journal of Public Economics, 87, 2003, 1487-519.

Bernheim, B. D., D. M. Garrett, and D. Maki. "Education and Saving: The Long-term Effects of High School Financial Curriculum Mandates." Journal of" Public Economics, 80, 2001, 435-65.

Bernheim, B. D., S. Bernstein, J. Gokhale, and L. J. Kotlikoff. "Saving and Life Insurance Holdings at Boston University--A Unique Case Study." National Institute Economic Review, 198, 2006, 75-96.

Black, D. A., S. Sanders, and L. Taylor. "The Economic Reward for Studying Economics." Economic Inquiry, 41, 2003, 365-77.

Blundell, R., L. Dearden, A. Goodman, and H. Reed. "The Returns to Higher Education in Britain; Evidence from a British Cohort." Economic Journal, 110, 2000, F82-99.

Braunstein, S., and C. Welch. "Financial Literacy: An Overview of Practice, Research, and Policy." Federal Reserve Bulletin, 87, 2002, 445-57.

Brown, R. H., N. Liang, and S. J. Weisbenner. "Individual Account Investment Options and Portfolio Choice: Behavioral Lessons from 401(K) Plans." Journal of Public Economics, 91, 2007, 1992-2013.

Chen, H., and R. P. Volpe. "An Analysis of Personal Financial Literacy among College Students." Financial Services Review, 7, 1998, 107-28.

Chevalier, J., and G. Ellison. "Are Some Mutual Fund Managers Better Than Others? Cross-Sectional Patterns in Behavior and Performance." The Journal of Finance, 54, 1999, 875-99.

Christiansen, C., J.S. Joensen, and J. Rangvid. "Are Economists More Likely to Hold Stocks?" Review o.1" Finance, 12, 2008, 465-96.

Craft, R. K., and J. G. Baker. "Do Economists Make Better Lawyers? Undergraduate Degree Field and Lawyer Earnings." Journal of Economic Education, 34, 2003, 263-81.

Furnham, A. F., and M.-P. Goletto-Tankel. "Understanding Savings, Pensions and Life Assurance in 16-21-year-olds." Human Relations, 55, 2002, 603-28.

Golec, J.H. "The Effects of Mutual Fund Managers' Characteristics on Their Portfolio Performance, Risk and Fees." Financial Services Review, 5, 1996, 133-48.

Grogger, J., and E. Eide. "Changes in College Skills and the Rise in the College Wage Premium." Journal of Human Resources, 30, 1995, 280-310.

Guiso, L., and T. Jappelli. "Awareness and Stock Market Participation." Review of Finance, 9, 2005, 537-67.

Hamermesh, D. S., and J. Lee. "Stressed Out on Four Continents: Time Crunch or Yuppie Kvetch?" Review of Economics and Statistics, 89, 2007, 374-83.

Hamermesh, D. S., and S. G. Donald. "The Effect of College Curriculum on Earnings: Accounting for Non-Ignorable Non-Response Bias." Journal of Econometrics, 144, 2008, 479-91.

Hecker, D. "Earnings of College Graduates, 1993." Monthly Labor Review, 118, 1995, 3-17.

Heckman, J. J., and R. Robb. "Alternative Models for Evaluating the Impact of Interventions: An Overview." Journal of Econometrics, 30, 1985, 239-67.

Hilgert, M. A., J. M. Hogarth, and S. G. Beverly. "Household Financial Management: The Connection between Knowledge and Behavior." Federal Reserve Bulletin, 88, 2003, 309-22.

Horvitz, D. G., and D. J. Thompson. "A Generalization of Sampling without Replacement from a Finite Universe." Journal of the American Statistical Association, 47, 1952, 663-85.

James, E., N. Alsalam, J. C. Conaty, and D.-L. To. "College Quality and Future Earnings: Where Should You Send Your Child to College?" American Economic Review, 79, 1989, 247-52.

Johnson, V. E. Grade Inflation: A Crisis in College Education. New York: Springer-Verlag, 2003.

Katz, L., C. Goldin, and L. Kuziemko. "The Homecoming of American College Women: The Reversal of the College Gender Gap." Journal of Economic Perspectives, 20, 2006, 133-56.

Kotlikoff, L. J., and B. D. Bernheim. "Household Financial Planning and Financial Literacy: The Need for New Tools," in Essays on Saving, Bequests, Altruism, and Life-cycle Planning, edited by L. J. Kotlikoff. Cambridge, MA: MIT Press, 2001, 427-77.

Lee, J., and K.-N. Kwon. "Consumers' Use of Credit Cards: Store Credit Card Usage as an Alternative Payment and Financing Medium." Journal of Consumer Affairs, 36, 2002, 239-62.

Little, R. J. A., and D. B. Rubin. Statistical Analysis with Missing Data. New York: John Wiley and Sons, 1987, 5-60.

Lusardi, A., and O. S. Mitchell. "Financial Literacy and Retirement Preparedness: Evidence and Implications for Financial Education Programs." Business Economics, 42, 2007, 35-44.

--. "Planning and Financial Literacy: How Do Women Fare?" American Economic Review, 98, 2008, 413-17.

Martin, M. "A Literature Review on the Effectiveness of Financial Education." WP 07-3, Federal Reserve Bank of Richmond, 2007.

Neal, D. A., and W. R. Johnson. "The Role of Premarket Factors in Black-White Wage Differences." Journal of Political Economy, 104, 1996, 869-95.

Newey, W. K., J. L. Powell, and J. R. Walker. "Semiparametric Estimation of Selection Models: Some Empirical Results.'" American Economic Review, 80, 1990, 324-28.

Nieswiadomy, M. "LSAT Scores of Economics Majors." Journal of Economic Education, 29, 1998, 377-79.

--. "LSAT Scores of Economics Majors: The 2003-2004 Class Update." Journal of Economic Education, 37, 2006, 244-47.

Rose. H., and J. R. Betts. Math Matters: The Links Between High School Curriculum. College Graduation. and Earnings. San Francisco, CA: Public Policy Institute of California, 2002.

Rosenbaum, P. R., and D. Rubin. "The Central Role of the Propensity Score in Observational Studies liar Causal Effects." Biometrika, 70, 1983, 41-55.

Salemi, M., J.J. Siegfried, K. Sosin, W. B. Walstad, and M. Watts. "Research in Economic Education: Five New Initiatives." American Economic Review, 91, 2001. 440-45.

Siegfried, J.J. "Trends in Undergraduate Economics Degrees, 1991 to 2006." Journal of Economic Education, 38, 2007, 360-64.

Tennyson, S., and C. Nguyen. "State Curriculum Mandates and Student Knowledge of Personal Finance." Journal of Consumer Affairs, 35, 2001, 241-62.

Thaler, R. H., and S. Benartzi. "Save More Tomorrow: Using Behavorial Economics to Increase Employee Savings." Journal of Political Economy, I12, 2004, S164-87.

Vella, F. "'Estimating Models with Sample Selection Bias: A Survey." The Journal of Human Resources, 33, 1998, 127-69.

Wooldridge, J. Econometric Analysis of Cross Section and Pattel Data. Cambridge MA: MIT Press, 2002.

(1.) Other economists have investigated the relationship between college coursework, choice of major, and career earnings, but with no special focus on the economics major or coursework. Grogger and Eide (I995l find that changes in the mix of college majors, over time, from low-skill subjects to high-skill subjects, help to explain the increase in the wage premium for college graduates. Arcidiacono (2004) finds large earnings differences across college majors, and James et al. (1989) conclude that the choice of major and coursework taken has a larger effect on future earnings than the choice of what college to attend. Conversely, Hamermesh and Donald (2008) found only small differences in earnings across different majors. There are also studies on the relationship between high school coursework and earnings. For example, Altonji (1995) finds relatively small effects of taking individual courses on subsequent earnings. Rose and Betts (2002), however, using similar methods but more recent data and a more detailed classification of the high school curriculum, found relatively large positive effects on earnings from taking mathematics courses.

(2.) As with labor market outcomes, many studies have been published on the effects of general education on individuals' behavior in the area of personal finance, with no particular focus on coursework in economics or personal finance. For example, Kotlikoff and Bernheim (2001) find that workers with less education have lower levels of financial literacy. Lee and Kwon (2002) report that consumers' use of store credit cards to finance purchases is negatively related to education levels. Barron and Staten (2002) show that college students exhibit different payment patterns for credit card accounts than non-students in the same age cohort. And Guiso and Jappelli (2005) found that individuals with more education were more likely to know about various kinds of financial assets.

(3.) Some observed behavior--such as consumers holding savings in accounts paying 2% interest while carrying credit card balances that are being charged 18% or more--have led economists to use models from behavioral economics to understand or predict that behavior (e.g., see Brown, Liang, and Weisbenner 2007; Thaler and Benartzi 2004).

(4.) As mentioned above, Hamermesh and Donald (2008) conclude that higher earnings by economics majors are partly due to working more hours than other majors. Hamermesh and Lee (2007) evaluate stress that is caused, in part, by working more hours. Conversely, there may be greater security and less stress in occupations that rarely face layoffs.

(5.) We could not perform a similar analysis for racial minorities due to the small number of minorities in our sample.

(6.) For more information on the survey see Allgood et al. (2004), which describes survey results on the graduates' perceptions of their undergraduate experiences with economics courses and instructors.

(7.) The questionnaire is available on request.

(8.) Race was missing for 31% of the sample, while family status was missing for 1.3% of individuals.

(9.) The sample weight for each cohort-school-major combination is defined as the ratio of the group's share in the student population in each college in 1976. 1986, or 1996 (proxied by the total number of degrees awarded within 4 years of each date), divided by their share in the total sample of potential respondents (the share in Sample I).

(10.) Differences across majors in weighted means were again very similar.

(11.) Because the number of survey responses for each of these variables differs (due to item non-response), regressions using these dependent variables have different numbers of observations. In Appendix A we discuss the potential effects of non-random item non-response on our estimates.

(12.) Our survey question asked how many months in the last year did you not pay off your credit card balances. Some respondents answered more than 12 (which could be interpretable for those who have more than one card), but we top-coded those responses as 12.

(13.) Even among graduates with general majors, 55% of our survey respondents took at least one economics course, and 22% took at least two economics courses.

(14.) Estimated marginal effects from linear probability models were very similar.

(15.) For a related attempt to control for ability bias using an individual's performance on the Armed Forces Qualification Test (AFQT) see Neal and Johnson (1996).

(16.) We differentiated courses both within and across general subject fields, resulting in 50 different course subjects.

(17.) This is lower than the share of current female graduates reported by Siegfried (2007) because it reflects enrollments from earlier decades, when there were fewer female graduates.

(18.) With almost 40% of students in our sample being UNC graduates, we also analyzed the extent to which the associations between economics training and outcomes differed between UNC and the three other universities. Across the various outcomes considered in this paper, we found very few statistically significant differences, indicating that our results do not solely apply to UNC alumni.

(19.) Specifications for these estimations again included the full set of covariates used in the B Panels in Tables 4 and 5. Results were again similar when we did not include the individual ability measure.

(20.) As with the labor market outcomes, the inclusion of additional controls, including ability measures, bad relatively small effects on the point estimates for college major and economics coursework in all regressions.

(21.) To the extent that this measure captures an aptitude for economics, rather than newly acquired economics knowledge, this indicates that our findings are robust to the inclusion of a second control for unobserved ability, in addition to the general ability measure discussed earlier. We therefore have little evidence to suggest that, after controlling for a student's overall ability, self-selection in taking economics courses is based on any preexisting ability or predisposition to do well in the courses, or is leading to significant biases in our estimates.

(23.) It may also be the case that course grades are relatively uninformative of the acquired knowledge or understanding.

(24.) Notable exceptions are Blundell et al. (2000) and Arcidiacono (2004). The first study uses matching methods to estimate major-specific earnings differentials based on a comparison group of individuals who did not attain a college degree. Similar to our approach, they account for selection-on-observables but not for selection-on-unobservables. Arcidiacono (2004) adopts a structural approach in which the choice of major is modeled jointly with its effect on labor market outcomes.

SAM ALLGOOD, WILLIAM BOSSHARDT, WILBERT VAN DER KLAAUW and MICHAEL WATTS*

* We thank the Board of the Calvin K. Kazanjian Economics Foundation for the grant that made this work possible, and the AEA Committee for Economic Education for bringing us together to write the proposal, as described by Salemi et al. (2001). April Fidler provided major assistance in project coordination, administration, and data entry. Georg Schaur worked extensively with data organization and preliminary tabulations. We have benefited from helpful comments from two anonymous referees and from participants at the 2006 SEA and 2009 SEE meetings and seminar participants at Waseda University. The views expressed are those of the authors and do not necessarily reflect those of the Federal Reserve Bank of New York.

Allgood: Associate Professor, Department of Economics, University of Nebraska-Lincoln, Lincoln, NE 68588. Phone 402-472-3367, Fax 402-472-9700, E-mail sallgood@UNLnotes.UNL.edu

Bosshardt: Associate Professor, Department of Economics, Florida Atlantic University, Boca Raton, FL 33431. Phone 561-297-2908, Fax 561-297-2542, E-mail wbosshar@FAU.edu

van der Klaauw: Assistant Vice President, Microeconomic and Regional Studies Function, Federal Reserve Bank of New York, 33 Liberty Street, New York, NY 10045. Phone 212-720-5916, Fax 212-720-1844, E-mail Wilbert. Vanderklaauw@ny.frb.org

Watts: Professor of Economics, Department of Economics, Purdue University, West Lafayette, IN 47907. Phone 765-494-8543, Fax 765-496-6068, E-mail mwatts@ purdue.edu

doi: 10.1111/j.1465-7295.2009.00270.x

TABLE 1
Description of Variables

Dependent variables--labor market

Hours = I if response to "How many hours a
 week do you usually work for pay?" was
 50 or greater, 0 otherwise

Commission = 1 if response to "What percentage of
 your annual income is paid as a
 commission or bonus based on how much
 you produce or sell?" was greater than
 zero. 0 if response to question was zero

Salary > 40K/Salary > 100K = 1 if response to "What was your
 individual (not family) wage or salary
 income in 2001, before paying taxes?
 Please check one of the following:"
 indicated an income of $40,001-$100,001
 or greater

Laidoff = 1 if response to "Since leaving
 college, have you ever been laid off or
 fired at a job'?" was yes, 0 if no

Self-Employ = 1 if response to "Are you currently
 running your own business, or have you
 ever run your own business after leaving
 college?" was yes, 0 if no

MoreThanBA = 1 if individual has more than a
 bachelors degree, 0 otherwise

Law/MBA/Masters/Ph.D. Indicators for highest year of school or
 college completed

RetireAge Response to "At what age do you expect
 to retire, or did you retire?"

Dependent variables--personal finance

OwnHome = 1 if answered "own" to "Do you own or
 rent your home?"

HomeEquity = 1 if less than $5000; = 2 if $5,000-
 $25,000; = 3 if $25,001-$50,000; = 4 if
 $50,001-$100,000; = 5 if $100,001-
 $250,000; = 6 if >$250,000 was response
 to "If you sold all of the real estate
 you own, including the land and house
 where you live if you own that property,
 and then paid off any money you owe on
 all of that property, how much money
 would you receive?"

Stocks/MutualFunds/Bonds/ Response to "What percent of your
MoneyMarket/CD/Bank financial savings and investments, other
 than employer-sponsored retirement
 accounts, are held in the following
 forms?"

#Cards Response to "How many bank credit cards
 (such as Visa, Master Card, American
 Express, Discover) do you have?"

Payoff = 1 if the response to "In how many
 months over the past year did you NOT
 pay off the full balance on all of those
 cards?" was zero, 0 otherwise

Savings Response to "What is the total value of
 your financial savings, investment, and
 retirement accounts?"

RetSavings Response to "If you have not already
 retired, how much money do you expect to
 have in your savings, investment, and
 retirement accounts when you retire?"

ReasonST/ReasonLT/ Respondents were asked to rate possible
Precaution/PayDebt reasons for saving as extremely
 important (5) to not important (1).
 Reasons listed were to pay for a home,
 car, or vacation; retirement; education
 expenses; bequests; invest in financial
 assets; unexpected events; and paying
 off debt. ReasonST is the average of
 responses for buying a home, car or
 vacation. ReasonLT is the average of
 responses for retirement, education,
 bequests. and invest in financial assets

Life Insurance = 1 if have employer provided or
 self-purchased life insurance

Self-PayLifeInsure = 1 if have self-purchased life
 insurance

 Independent variables

Cohort76/Cohort86/Cohort96 = 1 if in cohort, 0 otherwise. Cohort76
 is for 1976, Cohort86 is for 1986,
 Cohort96 is for 1996

FAU/Purdue/UNL = 1 for respondent's school, and 0
 otherwise

Business/General/Economics = 1 for respondent's major, and 0
 otherwise

EconCourses Number of economics courses found on
 student's transcript, removing repeated
 courses

EconCourses (2) Number of economics courses squared

Female = 1 if female, 0 otherwise. Reported by
 school in most cases. In some cases,
 educated guesses were made based on
 name. Androgynous names were left male

Black = 1 if race is black, 0 otherwise.
 Missing was coded as zero

Other = 1 if race is not black or Caucasian, 0
 otherwise. Missing was coded as zero

Livetogether = 1 if the answer to question "What is
 your current family status?" indicated
 they were living with partner (married
 or unmarried); 0 otherwise or missing

Children Response to "How many children to do you
 have?"

CumGPA Cumulative GPA at the end of the
 respondent's academic career

NorthEast/South/ = 1 if current state of residence is in
MidWest/West this Census region; 0 otherwise

TABLE 2
Sample Means (a)

 Sample I Sample II

Variable Unweighted Weighted Unweighted Weighted

Cohort76 0.222 0.222 0.306 0.324
Cohort86 0.393 0.393 0.321 0.344
Cohort96 0.385 0.385 0.373 0.332
UNL 0.185 0.185 0.231 0.226
FAU 0.265 0.265 0.146 0.139
Purdue 0.202 0.202 0.235 0.243
UNC 0.348 0.348 0.388 0.392
Economics 0.099 0.025 0.126 0.031
Business 0.450 0.181 0.428 0.148
General 0.451 0.794 0.446 0.820
EconCourses 2.886 1.652 3.368 1.852
 (2.780) (2.221) (3.050) (2.476)
Female 0.455 0.498 0.451 0.533
Black 0.069 0.083 0.045 0.051
Other 0.006 0.006 0.006 0.006
CumGPA 2.971 2.967 3.101 3.102
 (0.522) (0.551) (0.496) (0.520)
Northeast 0.052 0.048 0.057 0.054
South 0.574 0.573 0.513 0.508
Midwest 0.308 0.309 0.355 0.357
West 0.067 0.070 0.076 0.081
N 22,426 22,426 2,159 2,159

(a) All but two variables are dummy variables. For those two variables
the standard deviation is reported below the mean (in parentheses).
Means for Black and Other are for observations that included a
response for race.

TABLE 3A
Unweighted Sample II Means, by Major--Explanatory Variables (a)

Variable Full Sample Economics Business General

Cohort76 0.306 0.202 0.307 0.333
Cohort86 0.321 0.412 0.268 0.347
Cohort96 0.373 0.386 0.424 0.320
UNL 0.231 0.154 0.269 0.216
FAU 0.146 0.063 0.157 0.160
Purdue 0.235 0.099 0.253 0.255
UNC 0.388 0.684 0.320 0.369
Economics 0.126 1.000 0.000 0.000
Business 0.428 0.000 1.000 0.000
General 0.446 0.000 0.000 1.000
EconCourses 3.368 8.893 4.073 1.132
 (3.050) (2.011) (1.544) (1.823)
Female 0.451 0.243 0.382 0.575
Black 0.045 0.041 0.040 0.052
Other 0.006 0.014 0.003 0.006
Livetogether 0.704 0.699 0.706 0.705
Children 1.127 1.000 1.132 1.159
 (1.268) (1.155) (1.332) (1.235)
CumGPAEnd 3.101 3.066 3.100 3.113
 (0.496) (0.528) (0.446) (0.531)
Northeast 0.057 0.094 0.048 0.055
South 0.513 0.636 0.482 0.508
Midwest 0.355 0.206 0.396 0.357
West 0.076 0.064 0.075 0.080
N 2,159 272 924 963

(a) All but three variables are dummy variables. For those three
variables the standard deviation is reported below the mean (in
parentheses). Means for Black and Other are for observations that
included a response for race.

TABLE 3B
Unweighted Sample II Means, by Major--Outcome Variables (a)

 Full Sample Economics

Labor market

Hours 0.399 0.516
Commission 0.341 0.454
Laidoff 0.206 0.163
Salary > 40K 0.711 0.825
Salary > 100K 0.219 0.350
Self-Employ 0.273 0.315
MoreThanBA 0.423 0.468
Ph.D. 0.031 0.033
Law 0.058 0.097
Masters 0.185 0.134
MBA 0.105 0.193
RetireAge 60.347 60.380
 (8.832) (8.739)

Personal finance

OwnHome 0.774 0.759
HomeEquity < 50K 0.308 0.291
HomeEquity [50K,250K] 0.471 0.444
HomeEquity > 250K 0.221 0.265
Savings (000)--mean 606 2,135
 (8,619) (21,100)
Savings (000)--median 65 100
Investment allocations
Bank (%) 17.96 17.71
 (28.12) (29.24)
MoneyMarket (%) 13.29 16.84
 (22.59) (24.37)
CD (%) 2.78 2.77
 (10.40) (9.12)
Bonds (%) 3.64 4.34
 (10.30) (12.81)
Stocks (%) 16.01 21.37
 (24.62) (28.39)
MutualFunds (%) 26.74 26.86
 (31.09) (31.52)
Remainder (%) 19.57 10.11
 (32.75) (23.31)
RetSavings (000) mean 2,147 3,080
 (7,968) (6,098)
RetSavings (000) median 1,000 1,500
Reasons for saving
ReasonST 3.276 3.333
 (0.766) (0.713)
ReasonLT 3.446 3.602
 (0.791) (0.759)
Precaution 3.851 3.903
 (0.970) (0.980)
PayDebt 3.369 3.193
 (1.432) (1.453)
#Credit Cards--mean 3.211 2.586
 (22.187) (1.831)
#Cards--median 2 2
Payoff 0.585 0.590
Insurance coverage
Life insurance 0.897 0.920
Self-PayLifeInsure 0.601 0.588

 Business General

Labor market

Hours 0.459 0.303
Commission 0.434 0.212
Laidoff 0.234 0.190
Salary > 40K 0.777 0.615
Salary > 100K 0.251 0.152
Self-Employ 0.272 0.263
MoreThanBA 0.330 0.499
Ph.D. 0.007 0.053
Law 0.043 0.062
Masters 0.122 0.260
MBA 0.149 0.039
RetireAge 59.770 60.898
 (8.210) (9.9399)

Personal finance

OwnHome 0.781 0.771
HomeEquity < 50K 0.290 0.330
HomeEquity [50K,250K] 0.483 0.466
HomeEquity > 250K 0.227 0.204
Savings (000)--mean 586 174
 (6,126) (332)
Savings (000)--median 80 50
Investment allocations
Bank (%) 16.95 19.09
 (26.98) (28.94)
MoneyMarket (%) 14.15 11.30
 (23.37) (20.96)
CD (%) 2.64 2.94
 (10.63) (10.52)
Bonds (%) 3.61 3.45
 (9.98) (9.76)
Stocks (%) 17.88 12.40
 (24.96) (22.42)
MutualFunds (%) 27.06 26.38
 (30.62) (31.48)
Remainder (%) 17.71 24.43
 (31.29) (35.78)
RetSavings (000) mean 2,235 1,737
 (8,079) (8,368)
RetSavings (000) median 1,000 1,000
Reasons for saving
ReasonST 3.255 3.279
 (0.767) (0.779)
ReasonLT 3.492 3.357
 (0.762) (0.818)
Precaution 3.796 3.889
 (0.991) (0.944)
PayDebt 3.243 3.542
 (1.414) (1.425)
#Credit Cards--mean 2.790 3.800
 (3.712) (33.127)
#Cards--median 2 2
Payoff 0.628 0.542
Insurance coverage
Life insurance 0.902 0.885
Self-PayLifeInsure 0.595 0.610

(a) Most variables are dummy variables. For all other variables, the
standard deviation is reported below the mean (in parentheses).

TABLE 4
Estimated Marginal Effects--Economics/Business Major

Variable Salary > 40K Salary > 100K Hours

Panel A

Economics 0.171# 0.136# 0.125#
 (4.84)# (4.55)# (3.09)#

Business 0.166# 0.101# 0.123#
 (6.87)# (5.43)# (4.68)#

Panel B

Economics 0.166# 0.130# 0.131#
 (4.59)# (4.52)# (3.20)#

Business 0.164# 0.096# 0.127#
 (6.72)# (5.29)# (4.82)#

N 2,083 2,083 1,954

Variable Commission Laidoff Self-Employ

Panel A

Economics 0.204# -0.006 (a) 0.083#
 (5.31)# (0.19) (2.20)#

Business 0.225# 0.063# (a) 0.038
 (8.79)# (2.98)# (1.65)

Panel B

Economics 0.200# -0.005 (a) 0.081#
 (5.24)# (0.16) (2.14)#

Business 0.227# 0.067# (a) 0.036
 (8.82)# (3.18)# (1.53)

N 1,899 2,126 2,108

Variable MoreThanBA RetireAge

Panel A

Economics -0.075 (a) -0.755
 (1.84) (1.10)

Business -0.162# (a) -1.011#
 (6.18)# (2.11)#

Panel B

Economics -0.075 (a) -0.566
 (1.83) (0.82)

Business -0.176# (a) -0.998#
 (6.50)# (2.08)#

N 2,145 1,960

Note: RetireAge is estimated with weighted OLS; other regressions are
weighted probits. Marginal effects are evaluated at the mean and
reported for the probits. z-scores or t-statistics are in parentheses.
Estimates statistically significant at the 5% level are presented in
bold font. See Table 1 for variable definitions. Panel A estimated
models include the following control variables: Cohort76, Cohort86,
UNL, FAU, Purdue, Female, Black, Other. Panel B estimated models add
the additional control variables: Livetogether, Children, Ability
(grade) fixed effect and three region dummies.

(a) For this regression we are able to reject the null hypothesis that
the marginal effect of business equals the marginal effect of
economics at the 5% level. We cannot reject the null hypothesis for
all cases not indicated by the superscript "a."

Note: Estimates statistically significant at the 5% level are
presented in bold font is indicated with #.

TABLE 5
Estimated Marginal Effects--EconCourses

Variable Salary > 40K Salary > 100K Hours

Panel A

EconCourses 0.044# 0.021# 0.040#
 (3.15)# (2.65)# (3.00)#

[EconCourses.sup.2] -0.002 -0.001 -0.002
 (1.51) (1.02) (1.28)

Panel B

EconCourses 0.047# 0.020# 0.042#
 (3.36)# (2.60)# (3.11)#

[EconCourses.sup.2] -0.003 -0.001 -0.002
 (1.73) (0.96 (1.44)

N 2,083 2,083 1,954

Joint significance

Panel A 0.000 0.000 0.000

Panel B 0.000 0.000 0.000

Variable Commission Laidoff Self-Employ

Panel A

EconCourses 0.073# 0.027# 0.015
 (5.77)# (2.34)# (1.25)

[EconCourses.sup.2] -0.005# -0.003# -0.001
 (3.50)# (1.98)# (0.44)

Panel B

EconCourses 0.073# 0.031# 0.014
 (5.76)# (2.68)# (1.15)

[EconCourses.sup.2] -0.005# -0.003# 0.000
 (3.47)# (2.26)# (0.34)

N 1,899 2,126 2,108

Joint significance

Panel A 0.000 0.054 0.068

Panel B 0.000 0.021 0.068

Variable MoreThanBA RetireAge

Panel A

EconCourses -0.056# -0.707#
 (3.94)# (2.71)#

[EconCourses.sup.2] 0.005# 0.054
 (3.28)# (1.88)

Panel B

EconCourses -0.062# -0.711#
 (4.28)# (2.74)#

[EconCourses.sup.2] 0.005# 0.054
 (3.59)# (1.88)

N 2,145 1,960

Joint significance

Panel A 0.000 0.008

Panel B 0.000 0.007

Note: RetireAge is estimated with weighted OLS: other regressions are
weighted probits. Marginal effects are evaluated at the mean and
reported for the probits. z-scores or t-statistics are in parentheses.
Estimates statistically significant at the 5% level are presented in
bold font. See Table I for variable definitions. Panel A estimated
models include the following control variables: Cohort76. Cohort86.
UNL, FAU, Purdue, Female, Black, Other. Panel B estimated models add
the additional control variables: Livetogether, Children, Ability
(grade) fixed effect and three region dummies. Joint significance is
test of the joint significance of EconCourses and [EconCourses.sup.2].
Reported p-values are based on the [chi square] (2)-statistic for
probit regressions, and for OLS it is based on the F-statistic F(2, n
-k), where n is the number of observations and k is the number of
explanatory variables.

Note: Estimates statistically significant at the 5% level are
presented in bold font is indicated with #.

TABLE 6
Estimated Marginal Effects with Gender Interactions

Variable Salary > 40K Salary > 100K

Specification (i)

Economics 0.174# 0.121#
 (3.89)# (3.77)#

Business 0.147# 0.101#
 (4.22)# (4.57)#

Female -0.288# -0.139#
 (8.71)# (6.05)#

Female * economics -0.042 0.033
 (0.47) (0.62)

Female * business 0.040 -0.012
 (0.78) (0.37)

Specification (ii)

EconCourses 0.066# 0.023#
 (3.36)# (2.47)#

[EconCourses.sup.2] -0.004# -0.001
 (1.97)# (1.02)

Female -0.230# -0.118#
 (5.30)# (4.01)#

Female * EconCourses -0.030 -0.007
 (1.05) (0.38)

Fem * [EconCourses.sup.2] 0.001 0.000
 (0.37) (0.06)

N 2,083 2,083

Variable Hours Commission

Specification (i)

Economics 0.141# 0.221#
 (2.95)# (4.83)#

Business 0.143# 0.250#
 (4.22)# (7.48)#

Female -0.240# -0.075#
 (6.98)# (2.33)#

Female * economics -0.033 -0.055
 (0.39) (0.83)

Female * business -0.041 -0.049
 (0.81) (1.12)

Specification (ii)

EconCourses 0.050# 0.076#
 (2.90)# (4.56)#

[EconCourses.sup.2] -0.003 -0.005#
 (1.66) (2.91)#

Female -0.205# -0.040
 (4.59)# (0.92)

Female * EconCourses -0.023 -0.006
 (0.83) (0.24)

Fem * [EconCourses.sup.2] 0.002 0.000
 (0.78) (0.03)

N 1,954 1,899

 Self-
Variable Laidoff Employ

Specification (i)

Economics 0.024 0.091#
 (0.62) (2.00)#

Business 0.082# 0.032
 (2.88)# (1.03)

Female -0.024 -0.010
 (0.88) (0.32)

Female * economics -0.116 -0.035
 (1.94) (0.48)

Female * business -0.029 0.010
 (0.78) (0.22)

Specification (ii)

EconCourses 0.032# 0.020
 (2.12)# (1.27)

[EconCourses.sup.2] -0.003 -0.001
 (1.69) (0.42)

Female -0.016 0.020
 (0.45) (0.49)

Female * EconCourses 0.010 -0.007
 (0.44) (0.29)

Fem * [EconCourses.sup.2] -0.003 -0.001
 (1.23) (0.29)

N 2,126 2,108

Variable MoreThanBA RetireAge

Specification (i)

Economics -0.052 -0.525
 (1.07) (0.64)

Business -0.178# -0.909
 (4.92)# (1.43)

Female -0.046 -0.129
 (1.29) (0.19)

Female * economics -0.092 -0.124
 (1.08) (0.09)

Female * business 0.008 -0.224
 (0.15) (0.24)

Specification (ii)

EconCourses -0.065# -0.633
 (3.38)# (1.91)

[EconCourses.sup.2] 0.005# 0.035
 (2.70)# (1.02)

Female -0.075 -0.567
 (1.57) (0.72)

Female * EconCourses -0.009 -0.394
 (0.30) (0.75)

Fem * [EconCourses.sup.2] 0.004 0.089
 (1.16) (1.54)

N 2,145 1,960

Note: RetireAge is estimated with weighted OLS; other regressions are
weighted probits. Marginal effects are evaluated at the mean and
reported for the probits. z-scores or t-statistics are in parentheses.
Estimates statistically significant at the 5% level are presented in
bold font. See Table 1 for variable definitions. All estimated models
include controls for Cohort76, Cohort86, UNL, FAU, Purdue, Female,
Black, Other, Livetogether, Children, Ability (grade) fixed effect and
three region dummies.

Note: Estimates statistically significant at the 5% level are
presented in bold font is indicated with #.

TABLE 7
Estimated Marginal Effects-Likelihood of
Pursuing Advanced Degrees

Variable Ph.D. Law Masters MBA

Specification (i)

Economics -0.018# -0.001 -0.094# 0.177#
 (2.73)# (0.08) (2.96)# (7.50)#

Business -0.027# -0.014 -0.128# 0.109#
 (4.93) (1.49) (6.58)# (8.63)#
 2016 2127 2145# 2127#

Specification (ii)

EconCourses -0.014# 0.014# -0.059# 0.025#
 (3.79)# (3.09)# (4.85)# (5.82)#

[EconCourses.sup.2] 0.001# -0.001 0.005# -0.002#
 (2.72) (1.76) (3.65)# (4.40)#

N 2,145 2,145 2,145 2,145

Joint Significance 0.000 0.000 0.000 0.000

Note: Regressions are weighted probits. Marginal
effects, evaluated at the mean, are reported for the probits.
z-scores are in parentheses. Estimates statistically significant
at the 5% level are presented in bold font. See Table I for
variable definitions. All estimated models include controls
for Cohort76, Cohort86, UNL, FAU, Purdue, Female, Black,
Other, Livetogether, Children, Ability (grade) fixed effect
and three region dummies. A test of the null hypothesis that
the marginal effect of Economics equals the marginal effect
of Business was not rejected for all regressions in this table.
Joint significance is test of the joint significance of Econ
Courses and [EconCourses.sup.2]. Reported p-values are based on
the [chi square](2)-statistic.

Note: Estimates statistically significant at the 5% level
are presented in bold font is indicated with #.

TABLE 8
Real Estate

 OwnHome

Variable Without Income With Income

Specification (i)

Economics 0.040 0.006
 (1.42) (0.20)

Business 0.038# -0.001
 (1.99)# (0.04)

Specification (ii)

EconCourses 0.029# 0.018#
 (2.91)# (1.96)#

[EconCourses.sup.2] -0.002# -0.002#
 (2.47)# (1.99)#

N 2112 2055

Joint significance 0.010 0.134

 Home Equity < 50K

Variable Without Income With Income

Specification (i)

Economics -0.106# -0.076
 (2.77)# (1.81)

Business -0.096# -0.069#
 (3.50)# (2.41)#

Specification (ii)

EconCourses -0.034# -0.031#
 (2.31)# (2.02)#

[EconCourses.sup.2] 0.002 0.002
 (1.28) (1.40)

N 1580 1551

Joint significance 0.005 0.055

 HomeEquity > 250K

Variable Without Income With Income

Specification (i)

Economics 0.081# 0.050
 (1.99)# (1.11)

Business 0.056# 0.017
 (2.29)# (0.69)

Specification (ii)

EconCourses 0.022 0.015
 (1.84) (1.23)

[EconCourses.sup.2] -0.001 -0.001
 (1.19) (1.04)

N 1580 1551

Joint significance 0.055 0.446

Note: OwnHome and HomeEquity indicators are estimated as weighted
probit models. Marginal effects are evaluated at the mean and reported
for the probits. z-scores or t-statistics are in parentheses.
Estimates statistically significant at the 5% level are presented in
bold font. See Table I for variable definitions. All estimated models
include controls for Cohort76, Cohort86, UNL, FAU, Purdue, Female,
Black, Other, Livetogether, Children, Ability (grade) fixed effect and
three region dummies. The models with income controls include a set of
nine dummy indicators for whether pre-tax income in thousands exceeded
20, 30, 40, 50, 60, 80, 100, 120, and 140. A test of the null
hypothesis that the marginal effect of Economics equals the marginal
effect of Business was not rejected for all regressions in this table.
Joint significance is test of the joint significance of EconCourses
and [EconCourses.sup.2]. Reported p-values are based on the [chi
square]-(2)-statistic.

Note: Estimates statistically significant at the 5% level are
presented in bold font is indicated with #.

TABLE 9
Savings

 Log(Savings)

 Without With
Variable Income Income

Specification (i)

Economics 0.572# 0.333#
 (4.94)# (2.93)#

Business 0.534# 0.338#
 (6.82)# (4.52)#

Specification (ii)

EconCourses 0.075 0.034
 (1.77) (0.84)

[EconCourses.sup.2] -0.002 -0.001
 (0.40) (0.14)

N 1,769 1,749

Joint significance 0.000 0.144

 Log(RetSavings)

 Without With
Variable Income Income

Specification (i)

Economics 0.551# 0.314#
 (4.57)# (2.68)#

Business 0.380# 0.193#
 (4.62)# (2.46)#

Specification (ii)

EconCourses 0.105# 0.074
 (2.63)# (1.91)

[EconCourses.sup.2] -0.002 -0.002
 (0.62) (0.57)

N 1,486 1,449

Joint significance 0.000 0.000

 ReasonST

 Without With
Variable Income Income

Specification (i)

Economics 0.107 (a) 0.099 (a)
 (1.84) (1.63)

Business -0.021 -0.021
 (0.56) (0.54)

Specification (ii)

EconCourses -0.029 -0.027
 (1.41) (1.32)

[EconCourses.sup.2] 0.004 0.004
 (1.85) (1.73)

N 2,098 2,047

Joint significance 0.130 0.167

 ReasonLT

 Without With
Variable Income Income

Specification (i)

Economics 0.216# 0.177#
 (3.66)# (3.04)#

Business 0.143# 0.124#
 (3.69)# (3.15)#

Specification (ii)

EconCourses 0.031 0.024
 (1.44) (1.12)

[EconCourses.sup.2] -0.001 -0.001
 (0.49) (0.36)

N 2,048 2,000

Joint significance 0.020 0.087

 Precaution

 Without With
Variable Income Income

Specification (i)

Economics 0.100 (a) 0.093 (a)
 (1.31) (1.17)

Business -0.042 (a) -0.049 (a)
 (0.87) (0.96)

Specification (ii)

EconCourses -0.052# -0.051#
 (2.12)# (2.06)#

[EconCourses.sup.2] 0.007# 0.007#
 (2.56)# (2.54)#

N 2,118 2,062

Joint significance 0.032 0.032

 PayDebt

 Without With
Variable Income Income

Specification (i)

Economics -0.311# -0.315#
 (2.66)# (2.60)#

Business -0.276# -0.274#
 (3.86)# (3.67)#

Specification (ii)

EconCourses -0.083# -0.084#
 (2.21)# (2.19)#

[EconCourses.sup.2] 0.006 0.006
 (1.45) (1.47)

N 2,103 2,049

Joint significance 0.026 0.033

Note: Regressions are weighted OLS. t-statistics are in parentheses.
Estimates statistically significant at the 5% level are presented in
bold font. See Table l for variable definitions. All estimated models
include controls for Cohort76, Cohoit86, UNL, FAU, Purdue, Female,
Black, Other, Livetogether, Children, Ability (grade) fixed effect and
three region dummies. The models with income controls include a set of
nine dummy indicators for whether pre-tax income in thousands exceeded
20, 30, 40, 50, 60, 80, 100, 120, and 140. Joint significance is test
of the joint significance of EconCourses and [EconCourses.sup.2].
Reported p-values are based on the F-statistic F(2,n -k), where n is
the number of observations and k is the number of explanatory
variables.

(a) A test of the null hypothesis that the effect of Economics equals
the effect of Business could not be rejected at the 10% level or
better for this regression equation.

Note: Estimates statistically significant at the 5% level are
presented in bold font is indicated with #.

TABLE 10
Financial Investments

 Stocks MutualFunds

 Without With Without With
Variable Income Income Income Income

Specification (i)

Economics 11.540# 8.687# -0.578 -1.556
 (3.35)# (2.47)# (0.15) (0.39)

Business 10.626# 9.233# 2.294 1.483
 (4.54)# (3.85)# (0.89) (0.56)

Specification (ii)

EconCourses 4.321# 3.994# 0.988 0.617
 (4.09)# (3.74)# (0.86) (0.53)

[EconCourses.sup.2] -0.276# -0.278# -0.070 -0.044
 (2.75)# (2.75)# (0.64) (0.39)

N 1,799 1,770 1,799 1,770

Joint significance 0.000 0.000 0.619 0.836

 Bonds MoneyMarket

 Without With Without With
Variable Income Income Income Income

Specification (i)

Economics 2.921 2.362 9.418# 7.334#
 (0.88) (0.68) (2.85)# (2.15)#

Business 0.158 -0.178 8.787# 7.134#
 (0.07) (0.08) (3.94)# (3.09)#

Specification (ii)

EconCourses -0.072 -0.095 3.127# 2.643#
 (0.08) (0.10) (3.13)# (2.59)#

[EconCourses.sup.2] 0.056 0.05 -0.174 -0.148
 (0.63) (0.55) (1.85) (1.55)

N 1,799 1,770 1,799 1,770

Joint significance 0.295 0.450 0.000 0.002

 CD Bank

 Without With Without With
Variable Income Income Income Income

Specification (i)

Economics 5.178 6.275 -4.447 -1.992
 (1.01) (1.20) (1.33) (0.59)

Business -1.79 -1.888 -5.392# -3.928
 (0.53) (0.54) (2.47)# (1.76)

Specification (ii)

EconCourses -0.704 -1.03 -2.517# -1.995#
 (0.48) (0.69) (2.57)# (2.02)#

[EconCourses.sup.2] 0.128 0.166 0.157 0.13
 (0.92) (1.17) (1.66) (1.37)

N 1,799 1,770 1,799 1,770

Joint significance 0.378 0.265 0.004 0.046

Note: Tobit estimates. t-statistics are in parentheses. Estimates
statistically significant at the 5% level are presented in bold font.
See Table 1 for variable definitions. All estimated models include
controls for Cohort76, Cohort86, UNL, FAU, Purdue, Female, Black,
Other, Livetogether, Children, Ability (grade) fixed effect and three
region dummies. The models with income controls include a set of nine
dummy indicators for whether pre-tax income in thousands exceeded 20,
30, 40, 50, 60, 80, 100, 120, and 140. Joint significance is test of
the joint significance of EconCourses and [EconCourses.sup.2].
Reported p-values are based on the [chi square](2)-statistic.

Note: Estimates statistically significant at the 5% level are
presented in bold font is indicated with #.

TABLE 11
Credit Cards and Life Insurance

 #Cards Payoff

 Without With Without With
Variable Income Income Income Income

Specification (i)

Economics 0.624# 0.634# 0.046 0.030
 (10.79)# (10.22)# (1.10) (0.69)

Business 0.670# 0.654# 0.078# 0.062#
 (14.70)# (15.19)# (2.97)# (2.23)#

Specification (ii)

EconCourses 1.013 0.985 0.032# 0.032#
 (1.05) (1.14) (2.21)# (2.14)#

[EconCourses.sup.2] 0.995# 0.998 -0.003 -0.003
 (4.32)# (1.91) (1.79) (1.93)

N 2,064 2,015 2,009 1,962

Joint significance 0.000 0.000 0.069 0.100

 Has Life Insurance Self-Pay Life Insurance

 Without With Without With
Variable Income Income Income Income

Specification (i)

Economics 0.046# 0.026 -0.016 -0.022
 (2.54)# (1.41) (0.41) (0.61)

Business 0.023 0.004 0.017 0.009
 (1.70) (0.27) (0.66) (0.38)

Specification (ii)

EconCourses 0.012 0.005 -0.002 -0.001
 (1.59) (0.68) (0.12) (0.05)

[EconCourses.sup.2] -0.001 0.000 0.001 0.001
 (0.69) (0.14) (0.64) (0.56)

N 2,094 2,038 2,094 2,038

Joint significance 0.046 0.374 0.392 0.423

Note: Estimates for #Cards are Poisson regression estimates of the
IRR. The incidence-rate ratio measures the relative change in the
outcome, corresponding to a unit change in the regressor. z-scores or
t-statistics are in parentheses. Estimates statistically significant
at the 5% level are presented in bold font. Estimates for Payoff and
the two life insurance variables are marginal effect estimates
(evaluated at the mean) obtained using weighted probit estimation. See
Table 1 for variable definitions. All estimated models include
controls for Cohort76, Cohort86, UNL, FAU, Purdue, Female, Black,
Other, Livetogether, Children, Ability (grade) fixed effect and three
region dummies. The models with income controls include a set of nine
dummy indicators for whether pre-tax income in thousands exceeded 20,
30, 40, 50, 60, 80, 100, 120, and 140. A test of the null hypothesis
that the marginal effect of Economics equals the marginal effect of
Business was rejected for all regressions in this table. Joint
significance is test of the joint significance of EconCourses and
[EconCourses.sup.2]. Reported p-values are based on the [chi
square](2)-statistic.

Note: Estimates statistically significant at the 5% level are
presented in bold font is indicated with #.

TABLE 12
Estimates When Including Performance in Economics Courses

 Economics Business

Salary > 40K 0.171# (4.78)# 0.168# (6.90)#
Salary > 100K 0.121# (4.16)# 0.087# (4.61)#
Hours 0.133# (3.21)# 0.128# (4.79)#
Commission 0.187# (4.88)# 0.217# (8.32)#
Laidoff -0.005 (0.16) 0.068# (3.14)#
Self-Employ 0.083# (2.16)# 0.037 (1.55)
MoreThanBA -0.080 (1.93) -0.180# (6.45)#
RetireAce -0.625 (0.91) -1.049# (2.18)#
Ph.D. -0.018# (2.70)# -0.027# (4.87)#
Law -0.004 (0.34) -0.016 (1.72)
Masters -0.092# (2.84)# -0.126# (6.33)#
MBA 0.170# (7.13)# 0.104# (7.97)#
OwnHome 0.040 (1.43) 0.039# (2.00)#
Home Equity < 50K -0.103# (2.65)# -0.093# (3.34)#
HomeEquity > 250K 0.084# (2.02)# 0.059# (2.29)#
Log(Savings) 0.529# (4.54)# 0.498# (6.34)#
Log(RetSavings) 0.524# (4.28)# 0.357# (4.24)#
ReasonST 0.112 (1.92) -0.016 (0.43)
ReasonLT 0.204# (3.40)# 0.132# (3.34)#
Precaution 0.107 (1.37) -0.037 (0.75)
PayDebt -0.306# (2.60)# -0.273# (3.75)#
Stocks 11.454# (3.30)# 10.547# (4.44)#
MutualFunds -0.879 (0.22) 2.011 (0.77)
Bonds 3.003 (0.89) 0.235 (0.10)
MoneyMarket 9.962# (2.98)# 9.27# (4.07)#
CD 5.152 (1.00) -1.814 (0.52)
Bank -4.328 (1.28) -5.293# (2.39)#
#Cards 0.662# (9.37)# 0.710# (12.44)#
Payoff 0.031 (0.75) 0.066# (2.45)#
LifeInsurance 0.047# (2.62)# 0.025 (1.81)
Self-Pay Life -0.019 (0.48) 0.015 (0.57)
 Insurance

 Performance in
 Economics Courses

Salary > 40K -0.037 (1.59)
Salary > 100K 0.016 (1.24)
Hours -0.007 (0.33)
Commission 0.043# (2.16)#
Laidoff 0.000 (0.01)
Self-Employ -0.004 (0.20)
MoreThanBA 0.018 (0.75)
RetireAce 0.246 (0.62)
Ph.D. 0.001 (0.26)
Law 0.011 (1.23)
Masters -0.010 (0.60)
MBA 0.006 (0.89)
OwnHome -0.002 (0.12)
Home Equity < 50K -0.015 (0.63)
HomeEquity > 250K -0.006 (0.33)
Log(Savings) 0.159# (2.56)#
Log(RetSavings) 0.098 (1.64)
ReasonST -0.023 (0.68)
ReasonLT 0.050 (1.53)
Precaution -0.026 (0.64)
PayDebt -0.017 (0.28)
Stocks 0.341 (0.19)
MutualFunds 1.202 (0.61)
Bonds -0.255 (0.15)
MoneyMarket -1.922 (1.10)
CD 0.088 (0.03)
Bank -0.473 (0.28)
#Cards 0.717# (16.64)#
Payoff 0.059# (2.51)
LifeInsurance -0.012 (1.02)
Self-Pay Life 0.012 (0.55)
 Insurance

Notes: Regressions are weighted probit unless otherwise noted.
Marginal effects are evaluated at the mean and reported for the
probits. z-scores or t-statistics are in parentheses. Estimates
statistically significant at the 5% level are presented in bold font.
RetireAge, Log(Saving), Log(ExpectSave), STReason, LTReason,
Precaution, PayDebt are estimated with weighted OLS. Stocks,
MutualFunds, Bonds, MoneyMarket, CD, and Bank are estimated with
Tobit. Estimates for #Cards are Poisson regression estimates of the
IRR. Estimates statistically significant at the 5% level are presented
in bold font. See Table 1 for variable definitions. Estimated models
include the following control variables: Cohort76, Cohort86, UNL, FAU,
Purdue, Female, Black, Other, Livetogether, Children, Ability (grade)
fixed effect and three region dummies.

Note: Estimates statistically significant at the 5% level are
presented in bold font is indicated with #.