Skip to main content

Advertisement

SpringerLink
Log in

Business Cycles, Race, and Investment in Graduate Education

  • Original Article
  • Published:
Journal of Economics, Race, and Policy Aims and scope Submit manuscript

Abstract

This paper examines how macroeconomic factors influence household decision making with regard to human capital investment. We provide evidence suggestive of a causal relationship between macroeconomic indicators and the decision to pursue graduate education. Overall, we find graduate school enrollment is counter-cyclical with the business cycle and the magnitude of the relationship between macroeconomic indicators and the specific type of graduate school programs varies. In particular, we find differential racial effects of the business cycle on graduate school enrollment. The magnitude of the effects of the business cycle on graduate school enrollment is greater for some under-represented minority groups.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Log in via an institution

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Institutional subscriptions

Fig. 1
Fig. 2
Fig. 3
Fig. 4
Fig. 5
Fig. 6

Similar content being viewed by others

Notes

  1. Source - U.S. Census Bureau: http://www.census.gov/compendia/statab/cats/labor-force-employmentearnings.html

  2. Bureau of Labor Statistics.

  3. The Asian subsample is very small in the NLSY data set. This precludes us from drawing robust conclusions from any between group differences with the Asian subsample.

  4. We utilize national-level unemployment rates to be consistent with our other macroeconomic indicators. However, in the robustness checks section, we perform a similar analysis using state level unemployment rates and find consistent results.

  5. http://www.nber.org/cycles/cyclesmain.html

  6. We focus on graduate school demand side issues in our analysis. The data indicate that supply side issues (i.e., available graduate school slots) did not fluctuate substantially over the time period studied. Data from the Council of Graduate Schools indicate that the overall US graduate school acceptance rate was 41% in 2000 and 41% in 2011 with a maximum acceptance rate of 46% during the time period studied. (“Graduate Enrollment and Degrees: 1986–2004”, 2005; “Graduate Enrollment and Degrees: 2001–2011”, 2012). Further, the overall US graduate school acceptance rate had a \((\rho =-0.5574)\) correlation with the lagged unemployment rate and a \((\rho =+ 0.1058)\) correlation with the lagged S&P 500 Index.

  7. Income is lagged one year, but for simplicity, we include it in \(X_{it}\). Further, log of income is given a value of zero for respondents with no income. There are three variables in our data set that have missing data (income, AFQT score, and live in a rural area). To identify any potential issues with missing data, we use Little’s test. Based upon the Little’s test results, we characterize these missing variables as missing at random (MAR) since the pattern of data missingness can be predicted from other control variables in the data set. Since the missingness is conditional on other variables for which we control, we do get a random subset.

  8. We acknowledge that dropping the income and debt variables only provides a weak test of the robustness of the key results. While it addresses endogeneity bias, one could argue that it introduces omitted variable bias. However, there is not an appropriate instrument in the data to enable us to better address this issue.

  9. Given the panel nature of the data, it is theoretically possible to utilize a fixed effects model. Coefficients estimated using the fixed effects logit model could be more likely to capture the causal relationship because unobservable time-invariant characteristics are held constant. However, in the fixed effects logit model, identification hinges on changes in the macroeconomic variable causing changes in graduate enrollment. Thus, we lose a substantial portion of our data (over 95%) in this model such that we cannot obtain significant results.

  10. Since we utilize a sample of individuals with an undergraduate degree but no graduate degree, we deem the lagged structure to be the most appropriate. However, as a robustness check, we run additional models that also include current unemployment levels/S&P 500 levels. We find that the results of these specifications are consistent with our primary model specifications and these results are included in Appendix B.

  11. In the unemployment model, the one- and two-year lags are statistically significant. In the S&P500 Index model, only the two-year lag is statistically significant. (Any lags that are outside of the gray area (− 0.5 to 0.5) are statistically significant at 5% level.) Thus, we include both the one- and two-year lags in our models.

  12. “Trends in Graduate Student Financing” 2015 indicates that the average net (excluding grant aid) price of attendance for graduate students was increasing between 1995–2012.

  13. We utilize AFQT scores as an indicator of mental ability because it is a more uniform and unbiased measure of mental ability than undergraduate GPA, which is not available for our sample.

  14. For the white and Latino subgroup comparison, W \(=\) 20.88. For the white and African American subgroup comparison, W \(=\) 51.28. For the African American and Latino subgroup comparison, W \(=\) 87.86.

  15. In specifications using only the two-year lag, we find results that are consistent with our main results with regard to coefficient sign, magnitude, and significance.

  16. Due to the NLSY oversampling of African Americans and Latinos, we have over 100 unique individuals in each of these groups in the sample. We exclude Asians from our NLSY analysis by racial group due to the limited number of unique individuals in the sample (71 respondents).

  17. The Ph.D. and professional school columns of panel C are left blank due to the small sample size.

  18. Using data from the Current Population Survey (CPS), Ihrke and Faber (2012) show that the percent of individuals that moved from a different state between 1995 and 2000 was 8.4% while the percent of individuals that moved from a different state between 2005 and 2010 was 5.6%.

  19. The average of the variances of the state unemployment rates is similar to the variance of the national unemployment rate. Between 2000 and 2011, the variance in the national unemployment rate level was 3.54, while the state unemployment rate levels had an average variance of 3.31.

  20. Income is lagged one year, but for simplicity, we include it in \(X_{it}\).

  21. Source: https://nces.ed.gov/statprog/handbook/pdf/cps.pdf

  22. The observations in our sample broken down by year are as follows: 2000—11,644; 2001—15,197; 2002—15,972; 2003—15,804; 2004—15,913; 2005—16,025; 2006—15,860; 2007—14,815; 2008—14,783; 2009—13,022; 2010—13,861; and 2011—14,015.

  23. From “NCES Handbook of Survey Methods: Current Population Survey (CPS) - October Supplement,” there are both sampling errors and nonsampling errors for the data. For data cleaning, we have dropped the observations that have at least one missing value or nonresponse value in any of the variables. In addition, we validate the data by checking, for example, if an observation’s “Hispanic” variable indicates (s)he is Hispanic, but his/her “race” variable is “Black” only, then we exclude the observations like this. We check region and state and drop the observations with inconsistencies (e.g., A person lives in New York but “region” is “Pacific Division”). We also check school enrollment (after we limit to college graduates) and drop the observations if the current school attendance variable indicates a person is currently enrolled in a college or high school.

  24. Only total graduate enrollment is used as the CPS does not have data on type of graduate school enrollment. The CPS only asked about the type of degree program in which a respondent was enrolled in 1994 and this question was dropped from future waves of the survey.

  25. Income is lagged one year, but for simplicity we include it in \(X_{it}\).

References

  • Allum JR, Bell NE, Sowell RS. Graduate enrollment and degrees: 2001 to 2011. (Tech. Rep.). Council of Graduate Schools and Graduate Record Examinations Board. 2012.

  • Aughinbaugh A, Gardecki RM. Attrition in the National Longitudinal Survey of Youth 1997. (Working Paper). 2007.

  • Bedard K, Herman DA. Who goes to graduate/professional school? The importance of economic fluctuations, undergraduate field, and ability. Econ Educ Rev 2008;27:197–210.

    Article  Google Scholar 

  • Betts JR, McFarland LL. Safe port in a storm: the impact of labor market conditions on community college enrollments. J Hum Resour 1995;30(4):741–65.

    Article  Google Scholar 

  • Blackburn K, Varvarigos D. Human capital accumulation and output growth in a stochastic environment. Econ Theory 2008;36:435–52.

    Article  Google Scholar 

  • Boffy-Ramirez E, Hansen B, Mansour H. The effect of business cycles on educational attainment (Working Paper). 2010.

  • Box GE, Jenkins GM, Reinsel GC. Time series analysis: forecasting and control, 4th edn. Hoboken: Wiley; 2008.

    Book  Google Scholar 

  • Chauvet M. Stock market fluctuations and the business cycle. J Econ Soc Meas. 235–57. 1998.

  • Dellas H, Koubi V. Business cycles and schooling. Eur J Polit Econ 2003;19(4):843–59.

    Article  Google Scholar 

  • Dellas H, Sakellaris P. On the cyclicality of schooling: theory and evidence. Oxf Econ Pap 2003;55:148–72.

    Article  Google Scholar 

  • Estrella A, Mishkin FS. Predicting U.S. recessions: financial variables as leading indicators. Rev Econ Stat 1998;80(1):45–61.

    Article  Google Scholar 

  • Goh W. Is enrollment into graduate school affected by the business cycle? (Manuscript, Stanford University). 2009.

  • Graduate Enrollment and Degrees: 1986–2004. (Tech. Rep.) Council of Graduate Schools and Graduate Record Examinations Board. 2005.

  • Graduate Enrollment and Degrees: 2001–2011. (Tech. Rep.) Council of Graduate Schools and Graduate Record Examinations Board. 2012.

  • Herman DA. The impact of short-term economic fluctuations on kindergarten enrollment. Econ Educ Rev 2010;29:423–31.

    Article  Google Scholar 

  • Heylen F, Pozzi L. Crises and human capital accumulation. Can J Econ 2007;40(4):1261–85.

    Article  Google Scholar 

  • Higa FB. Community college enrollment and the 2007–2009 financial crisis (Brown University, Working Paper). 2010.

  • Ihrke DK, Faber CS. Geographic mobility: 2005 to 2010. (Tech. Rep.). United States Census Bureau. 2012.

  • Oyer P. The making of an investment banker: stock market shocks, career choice, and lifetime income. J Financ 2008;63(6):2601–28.

    Article  Google Scholar 

  • Trends in Graduate Student Financing: selected years, 1995–96 to 2011–12. (Tech. Rep.). U.S. Department of Education. (NCES 2015-026). 2015.

  • Wei M. Human capital, business cycles and asset pricing. (Manuscript, Columbia University). 2004.

Download references

Acknowledgments

We would like to thank Michael Dowell and Marie Mora for useful comments and discussions and Howard Yu for research assistance.

All rights reserved. This research was conducted with restricted access to Bureau of Labor Statistics (BLS) data. The views expressed here do not necessarily reflect the views of the BLS. All errors are our own.

Author information

Authors and Affiliations

  1. The Charles H. Dyson School of Applied Economics and Management, Cornell University, 201K Warren Hall, Ithaca, NY, 14853, USA

    Vicki L. Bogan

  2. Edgeworth Economics, LLC, 1111 19th Street, NW 12th Floor, Washington, DC, 20036, USA

    Di Wu

Authors
  1. Vicki L. Bogan
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. Di Wu
    View author publications

    You can also search for this author in PubMed Google Scholar

Corresponding author

Correspondence to Vicki L. Bogan.

Ethics declarations

Conflict of Interest

The authors declare that they have no conflict of interest.

Appendices

Appendix A: Definition of Primary Variables Used in Analysis

Education-Related Variables :

  • Graduate School Enrollment — A dummy variable that is given a value of 1 if the respondent enrolls in a graduate program in a given year and is set to 0 otherwise. The graduate school programs include full-time master’s degree programs, doctoral degree programs, and professional degree programs.

  • Full Time Master’s Degree — A dummy variable that is given a value of 1 if in a given year the respondent enrolls in a full-time master’s degree program including Master of Arts (M.A.) or Master of Science (M.S.). The variable is set to 0 otherwise.

  • Full Time Ph.D. — A dummy variable that is given a value of 1 if the respondent enrolls in a full-time Doctor of Philosophy (Ph.D.) program in a given year. The variable is set to 0 otherwise.

  • Full Time Professional Degree — A dummy variable is given a value of 1 if in a given year the respondent enrolls in a full-time professional degree program including Master of Business Administration (M.B.A.), Master of Public Administration (M.P.A.), Doctor of Medicine (M.D.), Juris Doctorate (J.D.), or professional degree program in another field. The variable is set to 0 otherwise.

Business Cycle-Related Variables :

  • Unemployment Rate — The annual national unemployment rate in a given year.

  • State Unemployment Rates — The annual state unemployment rates in a given year.

  • Log of S&P 500 Index — The natural logarithm of the S&P 500 Index in a given year.

Respondent Characteristic Variables :

  • Male Dummy Variable — A dummy variable that is given a value of 1 if the respondent is male. The variable is set to 0 for female.

  • Age — The age of the respondent.

  • \(\text {Age}^{2}\) — The squared age of the respondent to control for non-linear effects of age.

  • Married Dummy Variable — A dummy variable that is given a value of 1 if the respondent is married in a given year. The variable is set to 0 otherwise.

  • Number of Children — The number of children of the respondent.

  • Log of Income — The natural logarithm of the household’s total income in the previous year. Total income includes salary, wages, investment income, business income, and other income. (This continuous income level variable is used in NLSY Specifications only. CPS specifications utilize income level dummy variables.)

  • Log of AFQT Score — The natural logarithm of the respondent’s highest Armed Forces Qualification Test (AFQT) score. AFQT raw scores range from 0 to 100,000.

  • Have College Loans Dummy Variable — A dummy variable that is given a value of 1 if the respondent still owes college loans in a given year. The variable is set to 0 otherwise.

  • Undergraduate Major Dummy Variables — Dummy variables for the following majors in a given year are used: business, art, social science, science, engineering, law, and pre-med.

  • African American Dummy Variable — A dummy variable that is given a value of 1 if the respondent is African American. The variable is set to 0 otherwise.

  • Hispanic/Latino Dummy Variable — A dummy variable that is given a value of 1 if the respondent is Hispanic or Latino. The variable is set to 0 otherwise.

  • Native American Dummy Variable — A dummy variable that is given a value of 1 if the respondent is Native American. The variable is set to 0 otherwise.

  • Asian Dummy Variable — A dummy variable that is given a value of 1 if the respondent is Asian. The variable is set to 0 otherwise.

  • Other Race Dummy Variable — A dummy variable that is given a value of 1 if the respondent does NOT classify him/herself as white, African American, Latino/Hispanic, Asian, or Native American. The variable is set to 0 otherwise.

  • North Central Region Dummy Variable — A dummy variable that is given a value of 1 if the respondent lives in the north central region of the United States in a given year. The variable is set to 0 otherwise.

  • South Region Dummy Variable — A dummy variable that is given a value of 1 if the respondent lives in the southern region of the United States in a given year. The variable is set to 0 otherwise.

  • West Region Dummy Variable — A dummy variable that is given a value of 1 if the respondent lives in the west region of the United States in a given year. The variable is set to 0 otherwise.

  • Rural Dummy Variable — A dummy variable that is given a value of 1 if the respondent lives in an rural area in a given year. The variable is set to 0 otherwise.

  • CPS Only: Cohort Dummy Variables — Three dummy variables for age cohorts within the CPS data set. One dummy variable that is given a value of 1 if the respondent is between age 20 and 34 and is set to 0 otherwise. One dummy variable that is given a value of 1 if the respondent is between age 35 and 49 and is set to 0 otherwise. One dummy variable that is given a value of 1 if the respondent is over 49 and is set to 0 otherwise.

Appendix B: Robustness Checks

Endogeneity Robustness Checks

Table 17 Key marginal effects of unemployment rate on graduate enrollments—specification without log (income) and has college loans controls
Full size table
Table 18 Key marginal effects of log (S&P 500 Index) on graduate enrollments—specification without log (income) and has college loans controls
Full size table
Table 19 Key marginal effects of unemployment rate on graduate enrollments by race—specification without log(income) and has college loans controls
Full size table
Table 20 Key marginal effects of S&P 500 Index on graduate enrollments by race—specification without log(income) and has college loans controls
Full size table

Contemporaneous Macro Variables Robustness Checks

Table 21 Key marginal effects of unemployment rate on graduate enrollments—specification includes contemporaneous unemployment rate
Full size table
Table 22 Key marginal effects of log (S&P 500 Index) on graduate enrollments—specification includes contemporaneous log of S&P 500 Index
Full size table
Table 23 Key marginal effects of unemployment rate on graduate enrollments by race—specification includes contemporaneous unemployment rate
Full size table
Table 24 Key marginal effects of S&P 500 Index on graduate enrollments by race—specification includes contemporaneous log of S&P 500 Index
Full size table

Rights and permissions

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Bogan, V.L., Wu, D. Business Cycles, Race, and Investment in Graduate Education. J Econ Race Policy 1, 142–175 (2018). https://doi.org/10.1007/s41996-018-0004-x

Download citation

  • Received:

  • Revised:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s41996-018-0004-x

Keywords

Search

Navigation