Local Labor Markets and Human Capital Investments

I. Introduction

Many college majors represent an investment in sector- or occupation-specific skills. Without the relevant major, entry into these sectors or occupations is difficult to impossible. Given wage differentials across sectors and occupations, these decisions may become important for an individual’s career and lifetime earnings.¹ These decisions also have important aggregate implications, as they help determine supply of skills in the labor market.

This work analyzes whether individuals choose sector-specific human capital investments, specifically college major, based on local labor demand, rather than national demand. The relevance of this question is underscored by two facts. First, there are dramatic differences in labor demand across local markets, and substantial geographic concentration of industries. In the United States, examples include the computer sector in Silicon Valley, finance in New York, and oil and gas in Wyoming.² Second, geographic mobility is limited and declining even among highly educated individuals. From 2001 to 2010, annual interstate migration of college-educated individuals was 2.1 percent, roughly half the rate in the 1980s (Molloy, Smith, and Wozniak 2011).³ The first fact suggests investments based on local demand may differ significantly from those based on national demand. The second suggests local demand may affect investments, given that college-educated individuals are increasingly less likely to move across markets.

This is the first study, of which I am aware, studying the impact of local, sector-specific labor demand on local, sector-specific human capital production (college major) across the entire United States. Individuals may make investments based on local rather than national demand because of migration frictions or because of information frictions that cause a lack of awareness of nationwide job prospects. Investments based on local demand may be individually optimal if they are explained by strong location preferences. However, if individuals make investments based on local demand due to information frictions, this suggests an important role for policy improvements. Regardless of the mechanism, the consequences for the aggregate economy are potentially large if individuals invest based on local demand, and this causes mismatch between the aggregate supply of sector-specific skills and demand for these skills.⁴

It is possible to directly observe the correlation between sector-specific human capital investments, local, and national labor demand. However, these correlations alone would not be convincing evidence for this local elasticity, as endogeneity concerns make causality difficult to establish. Local demand may respond to, not determine, local human capital investments.

Using four sector-specific exogenous shocks with differential local effects, I test for changes in the share of relevant majors following these shocks. I then test whether universities in areas more exposed to these shocks experience greater changes in the share of sector-relevant majors. I focus on computer science and computer engineering (CS/CE) majors after the dot-com crash in 2000, geology majors after the boom in oil and gas enabled by hydraulic fracturing (fracking), finance majors after the 2008 financial crisis, and business majors after Delaware became an international center for financial services in the early 1980s, following a U.S. Supreme Court decision and subsequent state legislation.

While there are clear differences between these shocks, they all had large sector-specific employment effects in some local markets and smaller or zero effects in others. Investing based on local demand would yield significantly different major choices compared to investing based on national demand. I also exploit that for each shock the timing was exogenous to the number of majors, and there is a close mapping to demand for a particular major.

I show that, on average, universities experience a change in the share of sector-relevant degrees after these shocks, using university-level data on completions by major from the Integrated Postsecondary Education Data System (IPEDS). At their lowest post-bust levels in 2009, the share of CS/CE degrees awarded had fallen 2.2 percentage points (51 percent) on average at universities across the United States. At their highest post-fracking boom levels in 2014, the share of geology degrees awarded had increased 0.12 percentage points (52 percent) on average. At their lowest post-financial crisis levels in 2013, the share of finance degrees awarded had fallen 0.36 percentage points (15 percent) on average. The creation of a financial services center in Delaware did not affect average share of business degrees awarded at universities in Delaware and nearby states. This is consistent with this shock having an effect on financial employment in Delaware, but not in the broader region.

Second, I show college majors respond differentially in areas more exposed to these labor demand shocks. I compare universities more geographically exposed to these shocks, to less-exposed universities whose students experience the same national shock. I estimate the effect of exposure by year, as well as use a more parametric dynamic specification, enabling identification of preexisting trends.

These differential local effects are important in explaining the shocks’ effects on majors nationally. Of the aggregate decline in CS/CE after the dot-com crash, over 20 percent can be explained by differential impacts on majors in top-quartile exposed areas, as opposed to the shock’s national impact on majors affecting all universities equally. Of the overall increase in geology majors after the fracking boom, 14 percent can be explained by differential effects in top-quartile-exposed counties. Of the national decline in finance degrees after the crisis, more than 45 percent can be explained by differential impacts in top-quartile-exposed metropolitan statistical areas (MSAs).

This work contributes to an established and growing literature on how individuals choose human capital investments, and particularly college majors (see Altonji, Blom, and Meghir 2012 for a review).⁵ A number of these papers have studied how majors respond to national economic conditions.⁶ Several studies analyze how major choice is affected by local demand.⁷

A related literature shows local shocks affect the extensive margins of high school completion and college enrollment (Betts and McFarland 1995; Cascio and Narayan 2015; Charles, Hurst, and Notowidigdo 2018). These reflect responses to the opportunity costs of an additional year of schooling. My work’s focus on majors reflects whether individuals tailor those large investments to labor demand, conditional on college attendance.

I contribute to these literatures by focusing on salient national sectoral shocks, which map closely to majors, and how these salient national shocks differentially affect major choice across local labor markets. Except for the financial crisis, I focus on the impact of sectoral shocks that have received little or no coverage in the literature.⁸ More generally, there are few papers studying changes in major choice after shocks that are highly sector-specific.⁹ The analysis of the finance shock in Delaware is the first analysis of how human capital investments respond to place-based and local economic development policies, which are prevalent around the world.¹⁰

This study also contributes to current policy discussions on CS majors. Presidential administrations have enacted policies to increase access to CS education (Smith 2016; Executive Office of the President 2017). Further, a recent report requested by the National Science Foundation addressed the current all-time high enrollment in CS and how universities should respond in the short and long run (National Academies 2018).¹¹ Understanding students’ decisions to major in CS, and how this is affected by geography and sectoral demand, is crucial for maintaining a strong CS-skilled workforce.

Most generally, I show individuals make investments that enhance their ability to benefit from local shocks. This complements recent work suggesting individuals are affected by local shocks because of low levels of migration (Bartik 2018; Yagan 2019).

This paper proceeds as follows. Section II describes the sectoral shocks and shows they have differential employment effects across local labor markets. Section III presents the data. Section IV describes the empirical strategy identifying the shocks’ impact on major composition and differential impacts across markets. Section V contains the results, and Section VI concludes.

II. Sectoral Shocks with Local Labor Market Impacts

I show these four shocks affected sector employment share differentially across markets, using the Quarterly Census of Employment and Wages. If students chose majors based on local labor demand, those in negatively shocked areas would substitute away from sector-relevant degrees. However, if students chose majors based on national demand, these substitutions would be smaller in magnitude.

The dot-com crash began in March 2000 with a steep decline in internet stock prices.¹² Figure 1A shows this differentially affected computer employment relative to other sectors, and the effect was much stronger in Silicon Valley. Between 2001 and 2002, the percent of workers employed in “Computer Systems Design and Related Services” fell 0.76 percentage points in Santa Clara County, California, the home of Silicon Valley. In the United States overall, this percentage fell only 0.1 percentage points.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Sectoral Shocks and Differential Effects on Local Labor Markets

For the second shock, I use the dramatic increase in oil and gas production from the introduction of hydraulic fracturing (fracking) and horizontal drilling in the mid-2000s (Figure 1B). Between 2005 and 2011, the percent of workers employed in natural resources and mining increased 15.6 percentage points in McKenzie County, North Dakota. This county experienced the greatest cumulative increase in the value of new fossil fuel production from 2004 to 2014. Nationally, the percent employed in natural resources and mining increased only 0.14 percentage points.

The third shock, the 2008 financial crisis and the subsequent Great Recession, had a differential effect on employment in “Finance, Insurance, and Real Estate” (FIRE) relative to other sectors in New York County (Manhattan). Between 2007 and 2010, the percent employed in FIRE in New York County fell 0.85 percentage points (Figure 1C). Nationally, the percent employed in FIRE had been falling before the 2008 financial crisis. The effect of the crisis and subsequent Great Recession on national FIRE employment relative to other sectors was much weaker than the effect in New York County.

For the fourth shock, I use the creation of an international headquarters for the finance industry in the state of Delaware, resulting from jurisdictional competition and firm relocation. This is likely the least well known of these shocks, but nonetheless represented a dramatic change in the finance industry. Weinstein (2018) analyzes labor market adjustment to this shock.

Prior to 1978, state usury laws determined the interest rate that credit card companies could charge the state’s residents. The US Supreme Court’s ruling in Marquette National Bank of Minneapolis v. First Omaha Service Corp. (1978) allowed a bank to export the highest interest rate allowed by the state in which it is headquartered. In 1981, Delaware eliminated its usury laws, with the passage of the Financial Center Development Act (FCDA). The FCDA also reduced other industry regulation and introduced a regressive tax structure for banks.¹³ As a result, many companies moved their finance or credit operations to Delaware, starting with J.P. Morgan in 1981. Between 1981 and 1991, the percent of Delaware workers employed in FIRE sectors increased 4.5 percentage points (Figure 1D), based on Bureau of Labor Statistics (BLS) Current Employment Statistics (CES). Nationally, the increase was only 0.3 percentage points.

Each of these sectoral shocks differentially affected sectoral employment share across local labor markets. I will study whether sector-relevant majors differentially respond in areas more exposed to the shocks, which would suggest investments based on local, not just national, demand.

III. Data

I obtain university-level data on bachelor’s degrees awarded by academic discipline. For the dot-com crash, the fracking boom, and the financial crisis, I use IPEDS data.¹⁴ For the dot-com crash, I classify computer and information sciences and support services (CS) and computer engineering (CE) majors as sector-relevant degrees. For the financial crisis, I classify finance majors as sector-relevant degrees.¹⁵ For the fracking boom, I classify geology majors as the sector-relevant degrees. I focus on geology for several reasons. First, geology is crucial for understanding where to drill. Given the fracking boom involved innovations in horizontal drilling and hydraulic fracturing, these skills were arguably especially in demand (Vita 2015). Second, geology is offered at universities around the country regardless of their fracking exposure. Excluding petroleum engineering degrees will very likely lead to underestimating the local elasticity. However, these degrees are offered by very few universities and mostly in fracking-exposed areas.¹⁶ Oil and gas companies certainly demand other more widely offered engineering degrees, such as chemical, mechanical, and civil engineering. However, these degrees are also demanded by other sectors that may have their own cycles during this period.

Studying Delaware’s finance labor demand shock requires earlier university-level data. I obtain Bachelor’s degrees awarded by university and academic discipline for 1966–2013 from the IPEDS Completions Survey. These data are accessed from the Integrated Science and Engineering Resources Data System of the National Science Foundation. I focus on business and management majors in this part of the analysis as degrees by four-digit CIP codes are not available in these earlier years.¹⁷

Using the American Community Survey (ACS), pooling the 2009–2017 surveys, Table 1 confirms these are the relevant degrees for the industry. I show the share of employed 23–25-year-olds working in the affected industry by field of degree.¹⁸ To focus on degrees that are similarly awarded, I limit to majors constituting at least 0.2 percent of all degrees awarded, which is the proportion for the oil-and-gas-relevant majors. There may be some very relevant majors that are not widely offered across the United States, making it difficult to study differential responses by shock exposure. For example, 46 percent of petroleum engineering majors work in the oil and gas sector, but they constitute only 0.06 percent of degrees, reflecting that few universities offer these degrees.

Recent graduates with majors classified as relevant are much more likely employed in the computer, oil and gas, and finance industries than graduates in majors with the highest proportion employed in the sector, outside of those classified as relevant. The sector-relevant majors are also much more likely to work in the sector than the average major, excluding classified sector-relevant majors, the top three majors not classified as relevant, and the bottom five not classified as relevant. If few students would switch between the sector-relevant major and these least-relevant majors, changes in the latter around the sectoral shock may imply the shock affected student composition at the university. I will use the Table 1 categories to implement a placebo analysis testing for this possibility.

To determine the exposure of the university’s local labor market to the dot-com crash and financial crisis, I obtain the share employed in computers and finance using the IPUMS USA 2000 Census 5 percent sample (Ruggles et al. 2015). I classify as computer-related industries the BLS-defined high-technology industries that are relevant for the computer industry.¹⁹ I include the FIRE industries, excluding insurance and real estate, as finance-related industries.²⁰ Using the person weights, I obtain the weighted sum of individuals by industry and metropolitan area.²¹ I merge the data on share employed in computers and finance to the university-level data using the 2013 MSA.

Many universities are not in MSAs, and some MSAs are not represented in the census. For these categories, the principal results assume percent employed in computers or finance is zero. Approximately 32 percent of universities and 25 percent of all degrees (in 2000) cannot be matched to MSA employment for one of the reasons above. For robustness, I exclude these universities from the sample.

For fracking exposure, I obtain annual data on the value of oil and gas production from wells drilled for the first time that year (in 2014 dollars), within 200 miles of each county, from Feyrer, Mansur, and Sacerdote (2017).²² Papers studying the impact of fracking on the local economy often use an instrument for fracking exposure, since the decision of where to frack within the shale may be correlated with local economic characteristics, and trends in the outcome. However, it is much less likely that the decision of where to frack is correlated with trends in oil-and-gas-related degrees awarded by local universities.²³ This measure of exposure captures geographic areas experiencing new production due to fracking by directly using drilling data; this may not be as cleanly identified using industry employment.²⁴

To determine the university’s local labor market exposure to Delaware’s finance shock, I calculate distance between the university and Wilmington, DE (the city where the shock was concentrated) using the university’s latitude and longitude.²⁵ Because this was a Delaware-specific shock, I limit the sample of universities to those in Delaware, New Jersey, Pennsylvania, Maryland, Washington DC, Virginia, and West Virginia.

I confirm the sector-relevant majors are awarded widely across the United States, in areas with lower and higher sectoral employment concentration (Online Appendix Figure A1). This validates the exercise of studying differential response of these majors by shock exposure. While their pre-shock share is lowest in the least exposed areas, it is still nontrivial. Comparing areas with medium exposure to those with the most exposure, the share of sector-relevant degrees is often quite similar before the shock. It is clear that the most-exposed areas do not award all of the sector-relevant degrees. For three out of the four shocks, the top 1 percent of exposed areas, unweighted by total degrees awarded, award no more than 3 percent of sector-relevant degrees in the year before the graduates were freshman at the shock’s onset. For the 2008 financial crisis, the top 1 percent of exposed areas produce approximately 12 percent of sector-relevant degrees.²⁶

IV. Identifying Sectoral Shocks’ Effects on Majors

I start by estimating: (1)

The regression estimates the share of sector-relevant majors over time, relative to the omitted year, t*. Because I include university fixed effects, this regression gives the average within university change in sector-relevant major share after the sectoral shock.

The variable Share(Majors_ct) denotes the share of relevant majors at university c in year t. The coefficients δ_t identify the average within-university change in share relevant majors in year t relative to t*, in which the graduates were freshmen at the shock’s onset (2003 for the dot-com shock, 2009 for the fracking boom, 2011 for the finance shock, and 1985 for the Delaware shock).²⁷ As a rough measure of the fracking boom onset, I use the year in which fracking success had been publicized in at least 25 percent of shale plays (2006), using publicity data from Bartik, Currie, Greenstone, and Knittel (2019).

The variable TotDegrees_ct denotes the total Bachelor’s degrees awarded by university c in year t. I weight the observations by TotDegrees_ct, ensuring changes at larger universities get more weight than those at smaller universities. I cluster standard errors at the university level.²⁸

The coefficients δ_t are the coefficients of interest for understanding the nationwide impact of the shock on relevant degrees. The main identification assumption is that a shock’s timing is not caused by changes in major or correlated with other factors differentially affecting sector-relevant majors. Estimating the effects by year provides important evidence on the strength of the identification assumption.

Estimating effects by year is important for two additional reasons. First, these were not one-time shocks. Their magnitude changed over time, and perceptions about a shock’s persistence may also have changed. Second, these specifications allow me to identify how quickly degree completions respond to the initial shock. I do not separately identify dynamic effects from an original shock relative to contemporaneous effects as the shock evolves. However, I will analyze changes in major composition in the years after the shock’s onset and relate those changes to demand.

I also estimate similar regressions constraining the phase-in and prior trends to be linear: (2)

To best capture immediate effects, I include only post-policy years within five years of the shock. I include the ten years preceding the shock and censor the trend variable (t – t*) at –5 (as in Lafortune, Rothstein, and Schanzenbach 2018). The ten years preceding the Great Recession include another recession and recovery. To best capture the boom immediately preceding the shock, for this shock I include only the five years preceding t*. The coefficients β_trend reflect whether universities experienced greater changes in sector-specific majors preceding these shocks.

Based on the coefficients in Equation 2, I obtain the effect of these sectoral shocks relative to the year preceding the shocks. I present results showing the effect for the first graduates exposed as freshmen (t*) and five years after the first graduates exposed as freshmen (t* + 5). The impact of the shock in year t* relative to t* – 1 is: (β_jump + β_trend). The impact in year t* + 5 relative to t* –1 is: (β_jump + 5β_phasein + 6β_trend).

Next, I identify whether universities in more exposed areas experience larger changes in major composition: (3)

The variable Exposure_m denotes the extent to which university c is exposed to the shock, given its location in area m. For the dot-com crash and 2008 financial crisis, this is the share of metropolitan area m’s employment in 2000 in the computer sector and the share in the finance sector, respectively. For the fracking boom, I use whether the county’s cumulative value of new oil and gas production from 2004 to 2014 is within the top quartile. For Delaware’s shock, Exposure is one for universities within 15 miles of Wilmington, DE.²⁹

The coefficients β_r identify the differential effect on majors in each year in areas more exposed to the industry. These effects are also estimated in years before t* as k_min < 0. The main identification assumption is that the timing and local exposure to the shocks are not caused by local changes in major or correlated with other factors differentially affecting majors in exposed areas.

I also estimate the counterpart to Equation 2: (4)

Specifications 3 and 4 identify differential changes in major composition at more exposed universities. These may be driven by students at these universities changing majors or by a changing composition of students at these universities. Either suggests these shocks have effects on human capital investment decisions—either where or what to study. Changes in the national proportion of sector-relevant degrees surrounding these shocks suggest significant numbers of students changed majors in response to the shocks, and differential local effects are not explained by students changing universities. The placebo analysis and robustness section present further evidence suggesting the results are not driven by changing student composition.

V. The Effect of Sectoral Shocks on Major Composition

A. Average Effect across All Universities

National sectoral shocks cause large within-university changes in major composition (Table 2). Figure 2 shows the nonparametric (Equation 1) and parametric (Equation 2) results closely match for most of the shocks. However, the speed of the response differs across shocks. This is not surprising given differences in the shocks’ evolution and potential differences across major in the cost of switching. This implies difference-indifference estimates from the parametric specification, which assume t* as the first treatment year, will not capture the true effect. For consistency, I report these results as well as differences-in-differences based on the nonparametric specification and identifying the first year in which sector-relevant major share appears to respond.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Sectoral Shocks and within University Changes in Major Composition, Average across All Universities

View this table:

• View inline
• View popup

Table 2

Effect of Sectoral Shocks on College Majors, Average across All Universities

The share of degrees awarded in CS/CE began a sharp decline starting in 2004, the year after the first graduates exposed as freshmen. The share continued to decline through 2009. For graduates in 2009, the share CS/CE majors was on average 2.2 percentage points lower (p ≤ 0.01) than the share at the same university in 2003, the year before the decline began. In 2003, on average 4.3 percent of a university’s degrees awarded were in CS/CE (weighted by total degrees). Thus, a decline of 2.2 percentage points reflects a 2.2/4.3 = 51 percent decline in the proportion of CS/CE degrees awarded as a result of the dot-com crash. This almost exactly reverses the increase in share CS/CE degrees during the dot-com boom, when the share CS/CE degrees increased on average 1.9 percentage points from 1995 through 2003.

The share of degrees awarded in geology increased first in 2008, two years after publicity of success in 25 percent of shale plays, and the year before the first graduates exposed as freshmen. The share continued to increase through 2014. For graduates in 2014, the share of geology majors was on average 0.12 percentage points higher (p ≤ 0.01) than the share at the same university in 2007, the year before the increase began. In 2007, on average 0.23 percent of auniversity’s degrees are awarded in geology. Thus, universities experience an average 0.12/0.23 = 52 percent increase in the proportion of geology degrees awarded as a result of the fracking boom.

The share of degrees awarded in finance fell for the first time in 2010, two years after the onset of the crisis, and the year before the first graduates exposed as freshmen. The share continued to fall through 2013. For graduates in 2013, the share finance majors was on average 0.36 percentage points lower than the share at the same university in 2009, the year before the decrease began (p ≤ 0.01). In 2009, on average 2.43 percent of a university’s degrees awarded are in finance. Thus, universities experience an average 0.36/2.43 = 15 percent decline in the proportion of finance degrees awarded.

Financial relocation to Delaware did not on average affect the share of business majors at all universities in Delaware, Maryland, New Jersey, Pennsylvania, Virginia, and West Virginia. This is consistent with this shock being highly localized, without broad effects on regional employment.

The response to the dot-com crash appears to operate with a greater lag relative to the other shocks.³⁰ Initial course investments presumably make switching majors costly, and this may be most costly in STEM fields. Lagged effects imply potentially very adverse effects for students entering during a boom, but graduating during a bust. In the case of a positive shock, it may mean students miss entering an industry at an advantageous time.

For each shock that affects major choice, sector-relevant majors in the pre-shock period were not trending in the same direction as in the post-shock period. For the dot-com crash and financial crisis, the pre-shock trend was the reverse of the post-shock trend.³¹ This is consistent with the periods preceding the dot-com crash and financial crisis being sectoral boom periods (Figure 1). These pre-trends mitigate concerns that the identification assumption is violated.

Effects on major composition during these pre-shock boom periods also suggest a relationship between demand and human capital investments, although subject to endogeneity concerns. Job growth may have responded to university specialization, rather than the reverse. I focus on the crashes since these shocks are more clearly exogenous.

B. Differential Effects at More Geographically Exposed Universities

Effects on sector-specific majors are larger at universities in more exposed areas. Figure 3 shows the coefficients from estimating Regressions 3 and 4. As above, I report differences-in-differences based on the nonparametric specification and based on the parametric specification in Table 3.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Effect of Sectoral Shocks on Universities, by University’s Geographic Exposure to the Shock

View this table:

• View inline
• View popup

Table 3

Effect of Sectoral Shocks on College Majors, by University’s Exposure to the Shock

For 2009 graduates, the dot-com crash reduced the share of CS/CE majors by an additional 1.7 percentage points at universities whose MSA computer employment share was higher by ten percentage points, relative to graduates from the same university in 2003 (Row 8).³² This effect is statistically significant at the 1 percent level. In 2003, on average 5.4 percent of degrees awarded are in CS/CE, among universities whose MSA computer employment share is at least 0.1 (weighted by total degrees). For these universities, this additional decline of 1.7 percentage points represents a 31 percent decline in their share of CS/CE degrees awarded.

For 2014 graduates, the fracking boom increased the share of geology majors an additional 0.1 percentage points at universities in top-quartile-exposed counties, relative to graduates from the same university in 2007 (p ≥ 0.05). In 2007; on average 0.28 percent of degrees are awarded in geology at universities in top-quartile-exposed counties. For these universities, this additional increase of 0.1 percentage points represents a 36 percent increase in their share geology degrees awarded.

For 2013 graduates, the 2008 financial crisis reduced the share of finance majors by an additional 0.25 percentage points at universities whose MSA finance employment share was higher by five percentage points, relative to 2009 graduates from the same university (p = 0.12).³³ In 2009, on average 3.5 percent of degrees awarded are in finance, among universities where MSA finance employment share is at least 0.05. For these universities, this additional decline of 0.25 percentage points represents a 7 percent decline in their share finance degrees awarded.³⁴

Five years after the first-exposed graduates, Delaware’s finance shock increased the share of business majors by an additional 5.9 percentage points at Wilmington-area universities, relative to graduates the year before the first-exposed graduates. In 1984, on average 21.6 percent of degrees awarded are in business, among universities within 15 miles of Wilmington, DE. For these universities, this additional increase of 5.9 percentage points represents a 27 percent increase in their share business degrees awarded.

Figure 3 shows pre-shock trends in the effect of exposure are not in the same direction as post-shock trends. Further, for three of the shocks, the pre-shock trend in the effect of exposure on sector-relevant majors was the reverse of the post-shock trend. This is consistent with the periods preceding the negative dot-com crash and financial crisis being boom periods for the industry, and the period preceding Delaware’s positive shock being a bust period for FIRE Employment in Delaware (Figure 1). Differential effects in more exposed areas during these pre-shock boom or bust periods also imply a relationship between local demand and human capital investments, although subject to the endogeneity concerns discussed above.

However, larger increases at exposed universities in the preceding booms may suggest new majors produced during the boom were more marginal at these universities. This may explain the greater decline in the bust rather than locally driven investments. However, except for the dot-com crash and the 2008 financial crisis, for the other shocks there was no national pre-trend in the opposite direction that created or eliminated marginal majors. These shocks also produce differential local responses, reducing concerns that results reflect more marginal majors at exposed universities. Further, if exposed universities produced the most marginal CS or finance majors during the boom, this may quite plausibly be explained by investments based on local demand.

E. Change in Student Composition vs. Change in Major Choice: Placebo Analysis

A university’s geographic exposure to shocks may also affect students’ application and enrollment decisions. Universities’ major composition may have changed because of changes in student composition, rather than students changing their major. However, total degrees awarded do not vary by university’s exposure to the shock (Online Appendix Figure A5 and Table A5).

Nonetheless, despite constant enrollment, the students selecting into exposed universities may change. To test this, I implement a placebo exercise, using majors for which we expect few students to be on the margin, and thus minimal substitution, between these and the sector-relevant majors. As a result, any change in these sector-distant majors timed with the sector-specific shock likely does not reflect substitution between majors, but may reflect changes in student composition at the university.³⁸

I identify majors for which we expect minimal substitution using those with the lowest likelihood of employment in the affected sector, based on the ACS and shown in Table 1. These majors arguably require interests and skills quite different from those that are required by the sector and the sector-relevant major. The students who would have chosen the sector-relevant degree preceding a negative shock (for example, CS/CE) are unlikely to be choosing the sector-distant degree after the negative shock (for example, education).

For each university, I obtain the percent of degrees awarded in each year in the five most sector-distant fields.³⁹ I then estimate Regressions 3 and 4 using this as the dependent variable. Sector-distant degrees do not appear to change in the opposite direction as the sector-relevant degrees and with the same timing (Figure 4, Online Appendix Table A9).

• Download figure
• Open in new tab
• Download powerpoint

Figure 4

Effect of Sectoral Shocks on Sector-Distant Degrees, by University’s Geographic Exposure to the Shock

While exposed areas experience a larger increase in share sector-distant degrees after the dot-com crash, this is part of a longer trend that began before the shock. Although the individual coefficients from the nonparametric specification are statistically significant only in the post-period, the parametric results suggest the trend in the post-period is not statistically distinguishable from the pre-trend. The trend also continues after 2009, when CS/CE degrees start differentially increasing in exposed areas.⁴⁰ This suggests the increase in sector-distant degrees does not reflect changing student composition due to a decline in CS/CE degrees, but is instead part of a pre-existing trend in exposed areas.

This pre-existing increasing effect of exposure on sector-distant degrees is explained by education degrees, the largest component of computer-distant degrees (Online Appendix Figure A6).⁴¹ Nationally, education degrees fell sharply over this period. In 1990–1991, education degrees represented 10 percent of all bachelor’s degrees awarded in the United States. By 2013–2014, they were roughly 5 percent (National Center for Education Statistics 2018).

Education degrees are also much more concentrated in low-computer employment MSAs. In 1990, the average education degree share in MSAs least exposed to the computer industry was roughly 15 percent, dividing universities into equally sized bins of share computer employment and weighting by total degrees awarded at the university. In the highest-computer exposure MSAs, this was roughly 6 percent. While the share of education degrees fell everywhere, these declines were largest in levels and percentages in MSAs with higher education degree share. Because these are also MSAs with low computer employment share, we see a differential trend in education degree share by computer exposure. Importantly, this starts before, and continues after, the differential negative effect of exposure on the share of CS/CE degrees.

There is some evidence of a differential decline in sector-distant majors following Delaware’s finance shock. As mentioned, I use a different data source for this shock, and I have only 21 broad major classifications for this analysis. Part of the reason we might see a response is because the broad groupings include some less-distant fields, so this might reflect substitution rather than compositional changes.⁴² However, Online Appendix Figure A7 shows an increase in share of out-of-state students around the time of Delaware’s legislation, though this was also part of a pre-existing trend.

With the possible exception of the Delaware shock, these results suggest changes in share sector-relevant majors do not simply reflect changes in the types of students selecting into exposed universities. Instead, the evidence is consistent with students changing major choice differentially in exposed markets.

VI. Conclusion

I study whether college majors are influenced by local rather than national labor demand. I test for changes in sector-relevant majors after sector-specific local labor demand shocks and whether these changes are greater at more geographically exposed universities. I analyze four sectoral shocks with local effects: the 2000 dot-com crash, the fracking boom, the 2008 financial crisis, and the shock making Delaware a global financial headquarters.

First, these sectoral shocks affect within-university sector-relevant major share, using university-level data on degree completions by academic discipline for 1966–2016. Second, universities in areas more exposed to these shocks experience greater changes in sector-relevant majors. Of the national change in sector-relevant degrees after these shocks, differential effects at the most-exposed universities explain 23 percent (dotcom), 14 percent (fracking), and 46 percent (financial crisis). These are impacts over and above those experienced by all universities regardless of exposure.

Investing in human capital based on local labor demand may yield mismatch between aggregate supply of skills and aggregate demand. This may help explain why young college-educated individuals have much higher unemployment rates than older individuals (National Center for Education Statistics 2015), a puzzle from earlier literature (Blanchflower and Freeman 2000). This local dependence may also affect aggregate productivity if individuals are not matched to the job in which they are most productive.

Policy implications depend on whether the local elasticity is explained by information frictions or location preferences. If students invest based on local demand due to location preferences, encouraging human capital investments based on national demand may increase mismatch for students with strong preferences. Identifying the mechanism explaining the local elasticity is an important area for research, as some recent initiatives have provided information on national demand to college-going students, while others provide information on local demand.⁴³ Most generally, the results show individuals make human capital investment decisions that enhance their ability to benefit from local economic shocks.