No Longer Qualified? Changes in the Supply and Demand for Skills within Occupations

Abstract

Although labor market mismatch often refers to imbalances between supply and demand across occupations, mismatch within occupations can arise when skill requirements change rapidly, with important consequences for workers and the labor market. Using 200 million US online job postings, the authors show educational upskilling varied considerably by occupation during the Great Recession, persisted well beyond the initial recovery, and was correlated with rising demand for software skills. Developing an adjusted mismatch index, they demonstrate how the educational composition of vacancies becomes misaligned with that of unemployed workers within occupations, decreasing aggregate matching efficiency. Among occupations with persistent educational upskilling, the authors document lower job-finding rates for noncollege workers, suggesting rapidly changing educational requirements create a moving target for unemployed workers. Although they do not examine non-educational upskilling, the findings help reconcile prior studies showing little evidence of labor market mismatch with employer reports of skilled worker shortages after the Great Recession.

Keywords

labor demand job skills vacancies unemployment firm behavior

Although the term “mismatch” often refers to imbalances in the supply of and demand for labor across occupations, mismatch within occupations can also arise if the skill requirements for a job are changing over time. During the Great Recession, US employers rapidly increased requirements within occupations for a bachelor’s degree when hiring for open positions, a trend that became known as “educational upskilling” (Modestino, Shoag, and Ballance 2020). Although roughly one-third of educational upskilling during the last recession was shown to have been cyclical or temporary, as much as two-thirds of that increase appeared to have persisted during the initial recovery (Modestino, Shoag, and Ballance 2016), possibly driven by structural forces such as skill-biased technological change (Hershbein and Kahn 2018).

What are the broader implications of educational upskilling for workers and the labor market? If educational requirements within jobs increase gradually over time, labor supply can presumably adjust with minimal lags. However, employers may increase educational requirements more rapidly during periods of labor market disruption, such as when responding to recessions or adopting new technologies. This can create larger imbalances between labor supply and demand that take longer to resolve, such as the ongoing “race between education and technology,” with adverse impacts for less-educated workers (Autor, Goldin, and Katz 2020).

We study this question during the Great Recession, when the share of vacancies requiring at least a bachelor’s degree jumped by more than 10 percentage points (more than 70%) between 2007 and 2010. Figure 1 shows this increase was only partially reversed over the next three years before remaining relatively stable through 2019. This persistence in rising educational requirements suggests some unemployed workers lacking these newly demanded credentials would no longer qualify for jobs they once held, possibly extending their jobless spells—due to retraining or switching to another occupation—or causing them to exit the labor force entirely. If the supply of qualified workers lagged demand for an extended period, educational upskilling may have impaired matching efficiency both within affected occupations and in the aggregate, potentially explaining the slower labor market recovery after the Great Recession (Cavounidis, Dicandia, Lang, and Malhorta 2021).

Figure 1.

Requested Educational Qualifications versus Labor Market Slack

Using the near-universe of roughly 200 million US online job postings collected between 2007 and 2019 by Lightcast (formerly Burning Glass Technologies), we document a novel set of stylized facts about educational upskilling dynamics over the business cycle. We focus on educational upskilling, the increase in demand for workers with a bachelor’s degree, because obtaining a four-year college degree takes significant time and financial resources and completion is readily verifiable, making it a meaningful hurdle for employment. First, we explore how the increased demand for a bachelor’s degree varied considerably by occupation during the Great Recession and persisted beyond the initial recovery. Second, we examine how this persistence in bachelor’s degree requirements was correlated with rising demand for software skills within occupations, providing a direct link between the adoption of new technologies and structural educational upskilling. We acknowledge that other types of upskilling, such as on-the-job experience or industry certifications, may also be present but do not diminish the impact of educational mismatch occurring within occupations on the labor market outcomes of non-bachelor’s workers in occupations with large and persistent increases in the demand for a bachelor’s degree.

Finally, we are the first to document the impact of persistent educational upskilling on aggregate matching efficiency and its implications for workers. We develop an adjusted mismatch index to detect labor market imbalances caused by shifts in educational requirements within occupations over time. Using this adjusted mismatch index, we demonstrate how persistent educational upskilling shifts the composition of vacancies toward workers with a bachelor’s degree, creating misalignment with the educational composition of unemployed workers within occupations and decreasing matching efficiency in the aggregate. We further document lower job-finding rates for noncollege workers among occupations with persistent educational upskilling. Together, these contributions to the literature help reconcile prior studies finding little evidence of labor market mismatch (Davis, Faberman, and Haltiwanger 2012; Abraham 2015) with industry reports linking the dearth of skilled workers to slower hiring after the Great Recession (Weaver and Osterman 2017). Our findings suggest that search-and-matching models should account for rapidly changing educational requirements that present a moving target for unemployed workers to qualify for re-employment (Pissarides 2000).

Related Literature

Recent studies have found that changes in employer skill requirements during the Great Recession reflected both cyclical and structural forces. On the cyclical side, Modestino et al. (2020) demonstrated the share of job postings requiring four-year college degrees increased by 10 percentage points between 2007 and 2010. They estimated one-third of this educational upskilling was an opportunistic response to the greater availability of educated workers during the recession. Separately, they showed employer demand for college degrees and certain skills fell by roughly one-third of their initial increase as the labor market recovered between 2010 and 2014 (Modestino et al. 2016).

On the structural side, a complementary set of papers confirmed the remaining two-thirds of educational upskilling that occurred during the Great Recession persisted through 2014, possibly reflecting a structural change in job requirements (Hershbein and Kahn 2018; Zago 2018; Blair and Deming 2020). Persistent educational upskilling within occupations may reflect longer-term trends such as skill-biased technological change (Katz and Murphy 1992; Autor, Katz, and Krueger 1998; Autor, Levy, and Murnane 2003) or labor market polarization (Autor, Katz, and Kearney 2008; Acemoglu and Autor 2011; Autor and Dorn 2013). For example, Hershbein and Kahn (2018) found that rising IT capital investments, particularly in routine-cognitive occupations, were correlated with educational upskilling. These structural trends may have been accelerated by cyclical forces stemming from the Great Recession, as downturns tend to hasten long-term changes in the labor market (Charles, Hurst, and Notowidigdo 2012; Tüzemen and Willis 2013; Beaudry, Green, and Sand 2016; Jaimovich and Siu 2020).

Some scholars argue that educational requirements listed on job openings are not binding but instead reflect existing workers being overqualified within certain occupations (Cappelli 2015). Yet several studies find employers are willing to pay a premium for rising skill requirements within occupations induced by technology adoption. Bessen, Denk, and Meng (2022) showed that jobs requiring higher computer usage experience larger relative wage increases, contributing to growing wage inequality within occupations. Kogan, Papanikolaou, Schmidt, and Seegmiller (2021) demonstrated that technological change not only displaces low-skilled labor through automation but also depresses earnings growth among older high-skilled workers whose skills become obsolete. Braxton and Taska (2023) found technological change results in large earnings losses among displaced workers who switch to lower-paying jobs when skill demands in their prior occupations increase.

Despite these impacts on earnings, whether educational upskilling could be large or persistent enough to affect matching efficiency—either within affected occupations or in the aggregate—remains unclear. The U.S. Department of Labor’s O*NET database shows that computer, analytical, and quantitative skills have increased within job categories since 1979, but the increases were modest (Liu and Grusky 2013). Yet other studies show that states exhibiting greater mismatch in educational qualifications during the Great Recession also experienced greater job polarization, suggesting that shifting skill requirements can restrain job growth in the aggregate for an extended period, consistent with an outward shift in the Beveridge curve (Restrepo 2015; Zago 2018).

While standard indices constructed across occupations indicate labor market mismatch contributed to joblessness during the Great Recession, they fail to detect mismatch as a significant factor during the sluggish employment recovery. Şahin, Song, Topa, and Violante (2014) showed that labor market mismatch across occupations increased by 22% between 2007 and 2010, accounting for 29% of the rise in unemployment during the Great Recession. However, this standard mismatch index returned to pre-recession levels by 2012 (Burke 2015), despite aggregate matching efficiency remaining below pre-recession levels beyond 2015 (Hobijn and Perkowski 2016; Hall and Schulhofer-Wohl 2018). The anemic wage growth observed in the aggregate during the recovery period was also inconsistent with the labor mismatch hypothesis (Rothstein 2012; Abraham 2015). Instead, economists argued that weak aggregate demand, rather than skills mismatch or other structural factors, better explained the continued outward shift of the Beveridge curve after the Great Recession (Barlevy 2011; Lazear and Spletzer 2012; Rothwell 2012; Carnevale, Jayasundera, and Cheah 2012; Diamond 2013; Diamond and Şahin 2015; Marinescu 2017; Weaver and Osterman 2017). Yet these prior studies could not account for shifting educational requirements within occupations, possibly explaining why the economic literature contradicted employer reports claiming the high vacancy rate during the recovery reflected a lack of skilled workers (Bessen 2014).

Data Sources

We extend the literature to reveal new facts about educational upskilling within occupations over the full business cycle of the Great Recession, from 2007 through 2019, and demonstrate their broader implications for both workers and aggregate matching efficiency. We focus on occupational mismatch since workers can potentially qualify for similar jobs in other industries but are less able to qualify for different jobs in other occupations when aggregate demand falls.

To measure labor demand, we use data from more than 200 million US online job postings collected by Lightcast for 2007 and 2010–2019.¹ Using a proprietary algorithm to de-duplicate ads, Lightcast scrapes more than 40,000 sites from job boards, newspapers, government agencies, and employers, capturing more than 7 million unique job openings daily. Lightcast parses the text of each job posting to categorize occupation, industry, and educational requirements (e.g., bachelor’s degree) as well as specific types of common (e.g., communication), specialized (e.g., accounting), and software (e.g., Python) skills.

We use two versions of the Lightcast data. The first is the “main” vacancy data set used by researchers that provides unique job postings on a monthly basis. We pool this data by year to study changes over time in employer demand for education and specific skillsets by occupation. These time trends closely track movements in both aggregate vacancies from national surveys (e.g., Job Openings and Labor Turnover Survey [JOLTS]) and breakdowns by occupation and education distributions from state surveys (e.g., Minnesota).

Although the main Lightcast data set closely tracks vacancy trends from national and state surveys over time, the number of vacancies at a point in time is consistently lower. This is because Lightcast cannot capture job openings that are posted behind online paywalls or advertised physically (e.g., sign in the window). Moreover, whereas surveys explicitly ask employers about the number of openings, one online posting can represent multiple openings. Fortunately, Lightcast constructed a “normalized” (reweighted) data set that exactly matches the monthly number of industry vacancies as measured by JOLTS and then disaggregates this monthly count by using the occupational distribution within each industry from the main Lightcast data set.² We use this normalized version of the Lightcast data set to measure the number of vacancies by education level within occupations when constructing our mismatch indices.

To measure labor supply, we use microdata on unemployed workers collected by the Current Population Survey (CPS) from 2007 through 2019. The cross-sectional component provides the number of unemployed workers by occupation and education level to construct our mismatch indices. We also use the longitudinal dimension to track job-finding rates by worker education level within occupations experiencing temporary versus persistent educational upskilling.

Finally, we use other labor market data to measure changes over time in employment and wages. We use the American Community Survey (ACS) to disaggregate movements in the supply of labor by educational attainment within versus between occupations at various levels of the Standard Occupational Classification (SOC) system. We use the Occupational Employment Statistics (OES) to measure changes in wage levels and inequality over time within occupations experiencing persistent versus temporary or no educational upskilling.

Methods

Measuring Educational Upskilling within Occupations

Using the main vacancy data set, we first examine whether rising educational requirements during the Great Recession (2007–2010) were temporary or persisted throughout the subsequent recovery (2010–2019). Unlike prior studies, we measure persistence within occupations to reveal heterogeneity in labor imbalances masked by aggregate measures. For example, some occupations may have experienced mostly opportunistic (e.g., temporary) educational upskilling driven by the greater availability of college-educated workers during the recession. This could increase mismatch in the short term by temporarily lengthening unemployment spells for workers without a bachelor’s degree (Kuhn and Skuterud 2004), and resolve relatively quickly as the labor market initially recovered (2010–2013). Other occupations may have experienced more structural (e.g., persistent) educational upskilling driven by technology adoption, leading to larger labor market imbalances that resolved more slowly as workers either obtained the required credentials or switched occupations (2010–2019).

To operationalize this approach, we define an occupation as having experienced “significant” educational upskilling during the Great Recession if the change in the share of postings requiring a bachelor’s degree or higher was greater than the employment-weighted average increase that was observed economy-wide. Table 1 shows this share increased on average by 10.77 percentage points within occupations during the recession (2007–2010). Occupations that had below-average increases in this share during the Great Recession are designated as having experienced no significant educational upskilling.³

Table 1.

Employer Skill Requirements, Employment, and Wages for Occupations at the Three-Digit SOC Level

	Year				Recession	Short-term recovery	Longer-term recovery
	2007	2010	2013	2019	Δ2007–2010	Δ2010–2013	Δ2010–2019
Employer skill requirements
Number of job postings
Mean	139,603.70	127,078.90	218,317.00	356,284.30	−12,524.80	91,238.10	229,205.40
Standard deviation	(245,052.20)	(233,608.00)	(339,077.10)	(550,944.80)	(50,486.51)	(118,965.30)	(326,968.20)
Mean percentage of job postings requesting:
Bachelor’s degree or higher	10.01	20.77	18.92	19.69	10.77	−1.85	−1.08
Common skills	65.80	67.55	71.93	83.33	1.76	4.38	15.78
Specialized skills	57.12	78.07	80.35	88.45	20.95	2.29	10.38
Software skills	5.97	12.99	12.71	16.18	7.02	−0.27	3.19
Employment
Annual number of employed workers
Mean	1,054,938.00	986,104.00	999,048.40	1,054,253.00	−68,834.00	12,944.40	68,149.00
Standard deviation	(1,164,488.00)	(1,103,781.00)	(1,129,397.00)	(1,261,859.00)	(154,437.30)	(63,197.16)	(312,436.40)
Mean percentage with a bachelor’s degree or higher	31.80	28.81	34.29	36.98	−3.00	5.49	8.17
Wages
Median real wage
Mean	22.52	22.93	22.69	23.35	0.41	−0.24	0.41
Standard deviation	(12.16)	(11.91)	(11.99)	(11.86)	(3.71)	(1.50)	(1.95)
Ratio: 75th/25th percentile wages
Mean	1.63	1.66	1.72	1.71	0.03	0.05	0.04
Standard deviation	(0.31)	(0.22)	(0.25)	(0.22)	(0.28)	(0.08)	(0.09)
No. of three-digit occupations	94	94	94	94	94	94	94

Source: Data on employer skill requirements are from Lightcast, weighted by annual employment from the American Community Survey (ACS). Data on wages are from the Occupational Employment Statistics (OES).

Among the occupations that experienced above-average increases in the share of postings requiring a bachelor’s degree during the recession, we further differentiate between whether this increase in educational requirements was “persistent” or “temporary.” According to Table 1, on average 10% of the initial recessionary increase in the demand for a bachelor’s degree within occupations was reversed during the longer-term recovery period (2010–2019). We designate occupations as “temporary” educational upskillers if they experienced greater than a 10% reversion of their initial recessionary increase during either the short-term (2010–2013) or longer-term (2010–2019) recovery. We designate occupations as “persistent” educational upskillers if they experienced less than a 10% reversion of their initial recessionary increase during both the short- and longer-term recovery periods.⁴

Using these definitions, we classify each occupation in terms of educational upskilling behavior at both the two-digit and three-digit SOC levels. We then test whether persistent changes in educational requirements primarily reflect a compositional shift in job vacancies across, as opposed to an increase in demand for bachelor’s degrees within, the underlying detailed occupations.⁵ We also explore whether persistence in educational upskilling was more prevalent among occupations of a certain size, those with a higher initial share of educated workers, or those with greater productivity.

Finally, we use a difference-in-difference approach to test whether occupations that engaged in persistent educational upskilling also had persistent increases in the share of postings requiring other skills (e.g., common, specialized, or software) relative to occupations that exhibited temporary or no significant educational upskilling. To further explore how persistent educational upskilling might be driven by technology adoption, we examine whether occupations with persistent increases in requiring a bachelor’s degree were simply seeking more of the same software skills or requiring new software skills that might reflect structural changes in the job.

Detecting Educational Labor Market Mismatch within Occupations

We adapt the standard labor mismatch index developed by Şahin et al. (2014) to quantify potential hiring lost because of a misallocation of unemployed workers relative to the distribution of vacancies by education within occupations. The standard index is based on a Cobb–Douglas matching function, with hires increasing in the number of both unemployed workers ( $u_{it}$ ) and vacancies ( $v_{it})$ in market i at time t. Markets are typically defined by industry, occupation, or geography, but here we delineate them by occupation alone:⁶

M_{t} = 1 - \sum_{i = 1}^{I} \frac{φ_{i}}{\bar{φ_{t}}} {(\frac{v_{it}}{v_{t}})}^{δ} {(\frac{u_{it}}{u_{t}})}^{1 - δ}

(1)

The terms $u_{t}$ and $v_{t}$ refer to the total number of unemployed workers and vacancies in the economy, respectively. The parameters $δ$ and (1 − $δ$ ) capture the vacancy and unemployment elasticity of hires, respectively. The term $φ_{i}$ represents matching efficiency specific to occupation i, and $\bar{φ_{t}}$ represents a Current Employment Statistics (CES) aggregator of the market-specific matching efficiencies weighted by their respective vacancy shares.

By construction, the value of the index ranges from 0 (when all potential hires occur) to 1 (when no potential hires occur), depending on how closely the occupational composition of unemployed workers ( $u_{it}$ ) matches that of vacancies ( $v_{it}$ ), while accounting for different matching efficiencies across occupations. Zero mismatch should be considered an idealized benchmark for a social planner capable of costlessly reallocating unemployed workers across various occupational labor markets accordingly. However, since reallocation is typically not costless, this standard mismatch index can be interpreted as an upper bound (Şahin et al. 2014).

We extend the standard mismatch index in two important ways: one empirical and the other conceptual. Empirically, we use the richness of the Lightcast data to incorporate observed changes in educational requirements for job vacancies over time. Because of data limitations, Şahin et al. (2014) imputed labor demand by education for occupations using the pre-recession distribution of educational attainment for incumbent workers from the Bureau of Labor Statistics (BLS). Their approach holds this educational distribution fixed over time and assumes that the “educational requirements of newly created vacancies for each occupation is equal to the educational content in the existing jobs for that same occupation” (Şahin et al. 2014: 3549–50). By construction, their measure cannot detect labor market mismatch arising from changes in the educational demands of employers during the recession. In contrast, Lightcast’s normalized vacancy data set allows us to measure the observed number of vacancies demanded by education level within occupations each year to capture changes in mismatch arising from educational upskilling over time.

Conceptually, we develop an adjusted mismatch index to capture persistent educational upskilling within rather than across occupations. Şahin et al. (2014) measured occupational mismatch for distinct educational “sectors” by estimating their standard mismatch index using Equation (1) by two-digit occupation separately within an education group (e.g., college-educated). However, this approach detects only reductions in hiring caused by mismatch between the composition of job vacancies rather than that of unemployed workers across occupations (e.g., registered nurse versus sales representative) for a given education level (e.g., bachelor’s degrees).⁷ By construction, it cannot detect mismatch between the educational demands of employers and the educational attainment of unemployed workers within occupations (e.g., the share of registered nursing vacancies requiring bachelor’s degrees versus the share of job-seeking registered nurses with bachelor’s degrees), which can arise when education requirements shift rapidly in response to structural changes such as technology adoption.

This misalignment of vacancies and workers due to structural, rather than temporary, changes in educational requirements could have long-term implications for the efficient level of educational investments. To test this hypothesis, we adjust the standard mismatch index to capture persistent educational upskilling over time within occupations. Specifically, we treat vacancies for a given three-digit persistent educational upskilling occupation as pertaining to distinct labor markets according to whether the vacancy requires a bachelor’s degree. Similarly, we treat unemployed workers in that same three-digit occupation as now searching in different labor markets according to their degree status. For all other occupations—both temporary- and non-upskilling—we follow the standard approach and define the market solely on the basis of the three-digit occupation. More precisely, our adjusted index can be expressed as follows:

M_{tA} = 1 - \sum_{i = 1}^{I} \sum_{j = 0}^{1} \frac{φ_{i}}{\bar{φ_{t}}} {(\frac{v_{ijt}}{v_{t}})}^{δ} {(\frac{u_{ijt}}{u_{t}})}^{1 - δ} - \sum_{k = 1}^{K} \frac{σ_{k}}{\bar{σ_{t}}} {(\frac{v_{kt}}{v_{t}})}^{δ} {(\frac{u_{kt}}{u_{t}})}^{1 - δ}

(2)

In the above equation, $M_{tA}$ denotes the value of the adjusted mismatch index in month t. Persistent educational upskilling occupations are indexed by i, and all other occupations (those exhibiting temporary or no educational upskilling) are indexed by k. Education level is indexed by j, which takes a value of 1 if the vacancy requires a bachelor’s degree or higher (or if the worker has a bachelor’s degree or higher) and equals 0 otherwise. On the labor demand side, $v_{ijt}$ represents the number of vacancies in occupation i (exhibiting persistent upskilling) with given education requirement j (bachelor’s degree or not) in month t, $v_{kt}$ is the number of vacancies in occupation k (exhibiting either temporary or no upskilling) in month t, regardless of education requirement, and $v_{t}$ is the total number of vacancies in the economy in month t. On the labor supply side, $u_{ijt}$ represents the number of unemployed workers in occupation i with education level j (bachelor’s degree or not) as of month t, $u_{kt}$ refers to the number of unemployed workers in occupation k in month t, and $u_{t}$ denotes the total number of unemployed workers in the economy in month t.

The remaining terms are defined similarly to Equation (1). The term $φ_{i}$ still represents matching efficiency specific to occupation i, that is not specific to the education requirement of the vacancy, and $σ_{k}$ represents matching efficiency for occupation k.⁸ The term $\bar{φ_{t}}$ represents a CES aggregator of the market-specific matching efficiencies among the occupations indexed by i, weighted by their respective vacancy shares, and $\bar{σ_{t}}$ represents the analogous term for the occupations indexed by k.

Our approach assumes persistent educational upskilling reflects structural changes that would justify some increase in the share of job seekers with a bachelor’s degree (under costless retraining), whereas temporary upskilling might not justify such investment.⁹ We acknowledge that in practice, it is often not socially optimal to increase the share of job seekers with bachelor’s degrees to fully meet increased demand—even within persistent educational upskilling occupations—because producing more college graduates is costly in terms of both time and money. And even if it were socially optimal, private decisions might need to be subsidized if some of the benefits of obtaining additional education are external to the worker. Nonetheless, subsidizing existing workers within an occupation (e.g., healthcare) to obtain additional education (e.g., bachelor’s degree) could be more efficient than retraining existing college graduates in other occupations (e.g., sales), as assumed under the standard mismatch index. For future labor market entrants, college course offerings and choice of majors do respond to changes in job postings, especially for lower-cost course offerings (Conzelmann et al. 2024).

Quantifying the Impact of Educational Upskilling on Workers

We examine two potential impacts of educational upskilling on noncollege versus college-educated workers in occupations that experienced persistent educational upskilling versus those that did not. First, we calculate job-finding rates from unemployment by assigning each unemployed individual to their most recent occupation using a three-digit SOC crosswalk following Birinci, See, and Wee (2023). Each individual is then placed into one of six categories based on a combination of their occupation’s upskilling category—persistent, temporary, or non-upskilling—and their educational attainment—having earned a bachelor’s degree or not. For each category, the job-finding rate for month t is calculated as the share of unemployed people as of month t − 3 who were employed in month t, conditional on being observed in both months.¹⁰

Second, we compare changes in wage levels and inequality over time within occupations experiencing persistent versus temporary or no educational upskilling. The rapid increase in demand for educational requirements relative to the supply of educated workers within occupations might necessitate employers raising wages to attract workers with a bachelor’s degree to those positions, possibly increasing wage inequality within occupations between workers with and without a college degree. We test this hypothesis by comparing changes over time in median wages and the ratio of wages at the 75th and 25th percentiles for occupations with persistent educational upskilling relative to those without.

Results

Heterogeneity in Educational Upskilling within Occupations

We first examine the persistence in educational upskilling during the Great Recession, whether it varied across the labor market, and the degree to which it reflected a compositional shift in job vacancies across, versus increased demand for bachelor’s degrees within, the underlying detailed occupations. Figure 2 plots the share of postings requiring a bachelor’s degree by occupation at the two-digit SOC level over time, revealing stark differences in how educational upskilling unfolded over the business cycle. Relative to the economy-wide average, occupations with persistent educational upskilling (e.g., management and others, represented by the solid lines) experienced steeper increases in educational requirements during the recession (2007–2010); these increases endured throughout the recovery (2010–2019), with little sign of reversion. Temporary-upskilling occupations (e.g., community and social services and others, represented by dotted lines) showed large increases in the share of postings requiring a bachelor’s degree during the recession, yet those gains reversed by more than 10% during the recovery. The remaining occupations (e.g., production and others, represented by the dashed lines) experienced little or no upskilling during this period. Thus, the magnitude and degree of persistence in rising educational requirements was not widespread, as prior research has suggested, but instead varied considerably across the labor market. This heterogeneity could have adverse consequences for less-educated workers within the affected occupations as well as aggregate matching efficiency.

Figure 2.

Requested Educational Qualifications by Two-Digit SOC, 2007–2019

How much of the educational upskilling associated with a given broad occupation group is due to changes in education requirements within versus between the underlying sub-occupations? If most of the changes in bachelor’s degree requirements were occurring between the underlying sub-occupations, then it might be possible to detect labor market imbalances due to educational upskilling by calculating the standard mismatch index across those sub-occupations. To test this, we decompose the change in the share of postings requiring a bachelor’s degree for a given two-digit SOC occupation into separate components due to changes within versus between the underlying three-digit occupations for both the recession (2007–2010) and initial recovery (2010–2013) periods.

Figure 3, panel A, shows that during the recession, the increase in the share of postings requiring a bachelor’s degree for a given two-digit broad occupation was largely attributable to educational upskilling within the underlying three-digit sub-occupations, not the changing composition of job postings across those sub-occupations. By contrast, panel B shows that half or more of the reversion during the recovery period for the two-digit occupations that experienced temporary upskilling was attributable to changing composition across the underlying sub-occupations rather than reversion within those occupations.¹¹

Figure 3.

Decomposition of Change in Share of Postings Requesting a Bachelor’s Degree within versus between Three-Digit SOC

Disaggregating even further, Figure 4 confirms that the three-digit sub-occupations within a given two-digit broad occupation group also did not behave uniformly in terms of educational upskilling. For example, the broad legal occupation group is composed of lawyers, judges, and related workers (which experienced persistent upskilling) as well as legal support workers (which experienced temporary upskilling). In fact, 61 out of the 94 sub-occupations experienced no significant upskilling between 2007 and 2010, with the change in the share of postings requiring a bachelor’s degree falling below the economy-wide average threshold of 10.77 percentage points. Although a handful of these occupations did experience large percentage increases in the share of postings requiring a bachelor’s degree during the recession, it was from a very low base (most were below 10%) and all but one experienced some reversion in those demands during the recovery period, with one in seven ending up at or below their 2007 level.

Figure 4.

Change in Requested Educational Qualifications by Three-Digit SOC over Business Cycle

Among the 33 sub-occupations that did experience significant educational upskilling during the recession, only 15 were “persistent upskillers,” using our definition—meaning the reversion in the share of postings requiring a bachelor’s degree during the recovery period was less than the economy-wide average of 10%. The remaining 18 occupations that experienced significant educational upskilling during the recession were classified as “temporary upskillers”—exhibiting a degree of reversion during either the initial or longer-term recovery period that was greater than 10%. Notably, Figure 4 reveals no systematic relationship between the size of the initial increase in the share of postings requiring a bachelor’s degree during the recession and whether an occupation was classified as either a “temporary” or “persistent” upskiller during the recovery.

Figure 5 reveals that movements in labor supply were relatively small compared to those of labor demand during both the recession and recovery periods. This outcome is likely because only a fraction of the sudden double-digit surge in demand for college-educated workers could be filled from the pool of unemployed workers with a bachelor’s degree and because obtaining a bachelor’s degree would take several years for workers who no longer qualified for those positions. Decomposing the change in the share of employed workers with a bachelor’s degree among the broad two-digit occupations reveals that any increase came from changes in worker education levels rising within the underlying three-digit occupations rather than from a compositional shift in hiring between three-digit occupations. This pattern is consistent with greater occupational specialization arising from new education investments—either among incumbent workers or entering cohorts—rather than occupation switching among existing college graduates. Overall, however, the educational attainment of employed workers did not keep pace with the rapid shift in demand within occupations, suggesting that persistent educational upskilling could affect aggregate matching efficiency if unemployed workers were no longer qualified for their prior jobs.

Figure 5.

Decomposition of Change in Share of Employed Workers with a Bachelor’s Degree within versus between Three-Digit SOC

Characteristics of Occupations with Persistent Educational Upskilling

Examining the characteristics of occupations reveals that the recession likely accelerated the demand for workers with a bachelor’s degree within certain occupations. Table 2 shows that before the recession, occupations with persistent educational upskilling were growing—having a higher number of job postings and a greater share of postings as a percentage of employment compared with occupations that showed temporary or no significant upskilling. Persistent-upskilling occupations also had a higher pre-recession share of postings requiring a bachelor’s degree and other skills, such as specialized and software skills, as well as a higher share of employed workers with a bachelor’s degree and higher wage levels.

Table 2.

Characteristics of Occupations by Type of Educational Upskilling

	Pre-recession: 2007			Recession: Δ2007–2010			Short-term recovery: Δ2010–2013
	Persistent	Temporary	None	Persistent	Temporary	None	Persistent	Temporary	None
Employer skill requirements
Number of job postings
Mean	377,580.20	87,493.37	96,611.18	−34,412.30	−16,140.84	−5,907.76	163,836.10	54,006.57	84,878.48
Standard deviation	(107,435.70)	(201,309.80)	(129,978.40)	(59,536.49)	(35,374.03)	(51,228.46)	(147,004.00)	(54,006.58)	(105,673.80)
Mean percentage of job postings requesting:
Bachelor’s degree or higher	24.29	18.04	3.89	22.39	20.77	4.69	1.08	−4.28	−1.81
Common skills	60.66	70.28	65.66	14.35	5.84	−2.69	3.72	0.19	5.86
Specialized skills	67.61	56.66	54.64	20.39	25.30	19.37	1.17	1.10	3.64
Software skills	12.36	7.17	3.99	11.10	9.97	5.07	1.14	−3.58	−0.54
Employment
Annual number of employed workers
Mean	1,516,926.00	792,040.20	1,018,310.00	−64,923.00	−80,948.10	−54,574.30	42,615.00	−756.60	10,270.90
Standard deviation	(1,221,693.00)	(1,329,768.00)	(1,083,564.00)	(108,166.30)	(109,695.30)	(172,730.40)	(87,135.35)	(48,147.21)	(59,635.34)
Mean percentage with a bachelor’s degree or higher	71.87	53.91	15.16	2.16	1.10	0.87	1.08	1.12	1.26
Wages
Median real wage
Mean	33.12	27.78	18.29	1.67	−0.89	0.49	0.68	−0.33	−0.45
Standard deviation	(18.63)	(13.51)	(6.36)	(3.18)	(7.47)	(1.43)	(2.74)	(1.39)	(0.95)
Ratio: 75th/25th percentile wages
Mean	1.62	1.81	1.58	0.14	−0.01	0.01	0.06	0.06	0.05
Standard deviation	(0.46)	(0.21)	(0.27)	(0.53)	(0.05)	(0.22)	(0.10)	(0.10)	(0.07)
No. of three-digit occupations	15	18	61	15	18	61	15	18	61

Notes: Occupations with a percentage point change in bachelor’s degree share between 2007 and 2010 that is greater than the aggregate are defined as having significant upskilling. Those that also experience less than a 10% decline during the initial recovery (2010–2013) and in the longer term (2010–2019) are defined as persistent upskillers.

The pattern of changes over time in Table 2 highlights other distinguishing features of occupations for which the increase in education requirements during the Great Recession was “sticky.” For example, although all types of occupations raised requirements for various skillsets during the recession period (2007–2010), persistent-upskilling occupations were the only ones to continue to raise requirements for software skills during the initial recovery (2010–2013). Moreover, the share of employed workers with a bachelor’s degree, along with the median wage and wage inequality, increased more rapidly among persistent versus temporary upskillers during the recession period. These trends confirm that employers who raised educational requirements within the persistent-upskilling occupations were able to fill those jobs with qualified workers to some degree, although they had to pay a premium to do so.

Table 3 calculates the correlation between educational upskilling and these pre-recession characteristics for three-digit occupations. Although the annual share of postings requiring a bachelor’s degree is highly correlated with both the 2007 share of employed workers with a bachelor’s degree and real median wages, the change in the share of postings requiring a bachelor’s degree—either annually or using three-year stacked differences—is much less so. Also, the size of the occupation in terms of total employment is not highly correlated with educational upskilling, confirming that the increased demand for education is not driven by a handful of large occupations. Examining skill clusters, the increase in the share of postings requiring a bachelor’s degree during the recession was most highly correlated with an increase in the share requiring software skills (corr = 0.584), followed by specialized skills (corr = 0.484) but not common skills (corr = 0.171)—suggesting both technology and specialization played a role.

Table 3.

Correlation between Educational Upskilling and Occupational Characteristics

	BA share annual level	BA share annual change	BA share three-year change
	2010–2019	2010–2019	2007–2010
Percentage of postings requiring a BA
Mean	20.321	−0.120	10.766
Standard deviation	(20.296)	(2.535)	(9.702)
Correlation with pre-recession level of occupational characteristics
BA share of employed, 2007	0.878	0.092	0.233
Real median wage, 2007	0.627	0.033	0.122
Total employment, 2007	0.018	0.045	0.038
Correlation with contemporaneous measures of other skill requirements
Share of postings requiring software skills	0.583	0.346	0.584
Share of postings requiring specialized skills	0.489	0.283	0.484
Share of postings requiring common skills	0.402	0.304	0.171
No. of three-digit occupations	94	94	94

Notes: BA, bachelor’s degree.

Relationship between Educational Upskilling and Technology Skills

To what degree does persistent educational upskilling reflect structural changes in the underlying skills required for the job? Table 4 reports the results of our difference-in-difference analysis of changes in skill requirements for occupations with persistent versus temporary educational upskilling over time, relative to occupations with no significant changes in the share of postings requiring a bachelor’s degree. Each column is a separate regression in which the dependent variable is the share of postings requiring a particular skill. The independent variables of interest are an indicator for whether the occupation experienced persistent or temporary educational upskilling. The coefficients are measured relative to the omitted category of occupations with no significant educational upskilling to control for other changes in the labor market (e.g., immigration) that might affect the demand for particular skills.

Table 4.

Change in Skill Requirements within Occupations by Type of Educational Upskilling over Time

	Percentage point change relative to occupations that exhibited no educational upskilling
	Software skills			Specialized skills			Common skills
	Recession period	Short-term recovery	Long-term recovery	Recession period	Short-term recovery	Long-term recovery	Recession period	Short-term recovery	Long-term recovery
	Δ2007–10	Δ2010–13	Δ2010–19	Δ2007–10	Δ2010–13	Δ2010–19	Δ2007–10	Δ2010–13	Δ2010–19
Diff persistent vs. no upskilling	0.047(0.011) ***	0.010(0.005) **	0.029(0.012) **	0.013(0.039)	−0.025(0.013) *	−0.065(0.014) ***	0.108(0.049) **	−0.034(0.021) *	−0.045(0.021) **
Diff temporary vs. no upskilling	0.045(0.015) **	−0.026(0.007) **	−0.003(0.016)	0.041(0.055)	−0.047(0.012) ***	−0.056(0.020) ***	0.060(0.023) **	−0.058(0.029) **	−0.058(0.030) **
Diff (persistent–temporary)	0.003	0.036	0.031	−0.028	0.023	−0.008	0.048	0.024	0.014
(F-test p-value)	(0.883)	(0.041) **	(0.069) *	(0.228)	(0.270)	(0.708)	(0.539)	(0.477)	(0.686)
Baseline Diff for no upskilling occupations	0.051	−0.005	0.022	0.194	0.036	0.126	−0.027	0.059	0.185
No. of three-digit occupations	94	94	94	94	94	94	94	94	94

Source: Authors’ calculations based on vacancy data provided by Lightcast, weighted by 2007 employment level from the American Community Survey (ACS).

Notes: Each column is a separate regression in which the dependent variable is the share of postings requesting a particular skill and the omitted category is a dummy variable for occupations with no upskilling. Baseline difference for occupations with no upskilling listed for reference.

Statistical significance is indicated at the *** 1%, ** 5%, and * 10% levels, respectively.

We find that software skills are a distinguishing feature of persistent educational upskilling. During the recession, both persistent- and temporary-upskilling occupations increased the share of postings requesting software and common skills (e.g., communication), relative to occupations with no significant increase in the share of postings requiring a bachelor’s degree. During the recovery period, the demand for common skills showed significant reversion among both persistent- and temporary-upskilling occupations relative to those experiencing no significant educational upskilling. In contrast, the demand for software skills continued to increase among occupations that had experienced persistent educational upskilling but showed significant reversion among temporary-upskilling occupations. Overall, the sharp increase during the recession and the subsequent persistence in the demand for software skills followed a pattern that was strikingly similar to the demand for bachelor’s degrees, suggesting that employers were using the bachelor’s requirement not simply as a screening tool but possibly as an indicator that workers had acquired or could learn emerging software skills associated with the job.

Figure 6 confirms that occupations exhibiting persistent educational upskilling were also those that showed persistent upskilling in terms of software skills, even during the longer-term recovery period (2010–2019). This was true even for occupations beyond the usual technology-driven sectors such as engineering, mathematical, and computer science occupations. For example, occupations such as financial specialists; health diagnosing and treating practitioners; and advertising, marketing, promotions, public relations, and sales managers experienced large increases in the share of postings requesting software skills.

Figure 6.

Change in Share of Postings Requesting Software Skills by Three-Digit SOC over Business Cycle

Moreover, occupations that experienced persistent educational upskilling also requested a greater variety of software skills and at a higher frequency compared to temporary educational upskillers. Figure 7 plots the initial level in 2010 versus the change (2010–2019) in the share of postings requesting the top 10 individual software skills within three-digit occupations during the business cycle for persistent versus temporary upskillers. Panel A shows that occupations experiencing persistent educational upskilling sharply increased their demand for software skills specific to engineering, statistics, accounting, finance, business intelligence, and human resources software as well as database, customer relationship management, and application programming interface (API) tools. In contrast, panel B shows that occupations experiencing temporary education upskilling showed little or no increase in software requirements during the longer-term recovery and often asked for many of the same skills across occupations such as graphic and visual design software, geospatial information and technology, and scripting languages.¹²

Figure 7.

Initial Level versus Change in Share of Postings Requesting Top 10 Software Skills within Three-Digit SOC during Business Cycle

To what extent might the sudden increase in the demand for software skills within occupations that experienced persistent educational upskilling present a barrier to workers who were displaced from their jobs during the recession? Table 5, panel A, reports the number of unique software skills requested for each of the occupations that experienced persistent educational upskilling. Employers requested upward of 200 unique software skills on average in 2010 and continued to increase the number of software skills requested during the recovery period. In particular, occupations with initially lower levels of software skills in sectors such as healthcare and education experienced the largest increases in the demand for software skills during the recovery. Clearly, it would be impossible for an unemployed worker to acquire all of these diverse software skills and thus be qualified for every job opening within their prior occupation. Thus, employers might use a bachelor’s degree as a proxy for a worker’s ability to learn new software skills, which could explain why educational requirements persisted throughout the recovery as firms increased their adoption of new software technology.

Table 5.

Change in Software Skills for Persistent Educational Upskilling Occupations, 2010–2019

SOC	Persistent-upskilling occupation	A. Number of unique software skills requested			B. Software skill with largest percentage point (pp) change in share of postings for each persistent upskilling occupation
		2010	2019	Change (%)		2010	2019	Change (pp)
152	Mathematical Science Occupations	219	242	11	Scripting Languages	3.99	21.98	17.99
151	Computer Occupations	251	261	4	Cloud Solutions	1.25	12.28	11.03
112	Advertising, Marketing, Promotions, Public Relations & Sales Managers	214	238	11	Customer Relationship Management (CRM)	2.05	9.59	7.54
172	Engineers	215	246	14	Engineering Software	15.70	23.15	7.45
132	Financial Specialists	202	229	13	Accounting and Finance Software	5.69	9.98	4.29
251	Postsecondary Teachers	155	200	29	Education Software and Technology	1.03	4.94	3.90
131	Business Operations Specialists	237	256	8	Human Resources Software	3.98	7.45	3.47
254	Librarians, Curators, and Archivists	120	148	23	Library and Archiving	1.63	4.35	2.72
113	Operations Specialties Managers	229	248	8	Accounting and Finance Software	3.11	5.30	2.19
299	Other Healthcare Practitioners & Technical Occupations	71	116	63	Health Information Management and Medical Records	0.64	2.46	1.82
111	Top Executives	200	236	18	Customer Relationship Management (CRM)	0.60	2.21	1.61
291	Healthcare Diagnosing or Treating Practitioners	159	197	24	Health Information Management and Medical Records	0.31	1.88	1.57
252	Preschool, Elementary, Middle, Secondary & Special Education Teachers	116	177	53	Education Software and Technology	0.58	2.02	1.44
253	Other Teachers & Instructors	101	153	51	Education Software and Technology	0.40	1.63	1.23
231	Lawyers, Judges & Related Workers	133	177	33	Tax Software	0.14	1.06	0.92

Source: Data on employer skill requirements are from Lightcast.

Notes: SOC, Standard Occupational Classification.

Moreover, the demand for certain unique software skills increased sharply within occupations, even those not usually considered technical or technology driven. Panel B of Table 5 lists the top software skills that had the largest percentage point change in the share of postings within each persistent-upskilling occupation. For example, the demand for customer relationship management (CRM) software skills increased by nearly 8 percentage points for advertising, marketing, promotions, public relations, and sales managers. This increasing specialization within occupations for certain types of software skills might replace routine tasks and perhaps be complementary with cognitive tasks that require a bachelor’s degree (Braxton and Taska 2023). Overall, our findings are consistent with the hypothesis that technological advances are driving the persistent educational upskilling observed within occupations, particularly those that use specialized software (e.g., engineering software) or those for which new software rapidly diffuses (e.g., managers), possibly changing the nature of job tasks.

Implications of Upskilling for Labor Market Mismatch

Did persistent educational upskilling within occupations affect matching efficiency, either in the aggregate or within certain sectors of the labor market? Unemployed workers in occupations with persistent educational upskilling may no longer qualify for the positions they once held if they lack the necessary skills or credentials to meet these new hiring requirements, possibly increasing occupational mismatch. To test this hypothesis, we estimate our adjusted mismatch index from Equation (2). We first partition each of the 15 persistent-upskilling occupations into two distinct sub-occupations by education: one that is open to workers with a bachelor’s degree and the other that is open to workers without a bachelor’s degree. For example, financial operations vacancies that require a bachelor’s degree and unemployed financial operations workers who hold a bachelor’s degree are assigned to a separate financial operations bachelor’s sub-occupation. Financial operations vacancies without that requirement and financial operations workers without that degree are assigned to a different financial operations non-bachelor’s sub-occupation. Occupations experiencing temporary or no significant upskilling are not partitioned since changes in the demand for a bachelor’s degree were more transitory or much smaller, respectively.

We then calculate our adjusted mismatch index accounting for persistent educational upskilling within occupations using Equation (2) and compare it to the standard mismatch index from Equation (1). Figure 8 shows that the level of the adjusted mismatch index (panel A) is higher than that of the standard index (panel B), but this is somewhat mechanical since the adjusted index is calculated at a slightly more disaggregated level than the standard index.¹³ More relevant to our research question is the relative comparison of the changes in these two indices over the business cycle. Both signaled an increase in labor market mismatch during the recession (2007–2010), although the rise was steeper for the adjusted versus the standard mismatch index. The pattern during the recovery period was even more striking. Between 2010 and 2013, the standard mismatch index fell from 0.121 to 0.065—nearly a 50% drop. By contrast, the adjusted mismatch index declined more modestly from 0.203 to 0.179—decreasing by roughly 10%. After 2013, the standard index was relatively flat, while the adjusted index rose slightly before leveling off. Overall, the adjusted mismatch index exhibited a less cyclical pattern than did the standard index, aligning with industry reports that some unemployed workers were no longer qualified for their jobs.

Figure 8.

Adjusted Mismatch Index Accounting for Educational Upskilling within Occupations

How does educational upskilling affect occupational mismatch within educational sectors as vacancies are reallocated from the non-bachelor’s sector to the bachelor’s degree sector over time? For comparison, Figure 9 compares the standard mismatch index across occupations separately for each educational sector, showing the misallocation of vacancies across the same set of occupations for unemployed workers with a bachelor’s degree (panel A) versus unemployed workers without a bachelor’s degree (panel B).¹⁴ Note that the level of mismatch is consistently higher in the bachelor’s versus the non-bachelor’s sector. Thus, while having more education makes workers more adaptive, it also makes them more specialized and potentially less substitutable across occupations, with this second effect being dominant. For example, a worker with a bachelor’s degree in engineering is not likely to be able to switch costlessly to a job as a healthcare practitioner. In contrast, a worker with a high school degree may have a more general set of skills (e.g., customer service) that can be applied to a wider range of occupations (e.g., waitstaff versus sales).

Figure 9.

Standard Mismatch Index across Occupations by Educational Sector

Figure 9 also shows that changes over time in the standard mismatch index also vary by educational sector and are consistent with the educational upskilling trends documented earlier. As employers raised education requirements during the recession, vacancies were reallocated from the non-bachelor’s degree sector to the bachelor’s degree sector, largely among the persistent- and temporary-upskilling occupations. Yet, we showed earlier that the rate at which the share of postings for a bachelor’s degree increased was more rapid than the rate at which the supply of workers with a bachelor’s degree increased. This difference would be expected to sharply increase mismatch in the bachelor’s sector between 2007 and 2010, as shown in panel A, relative to little or no increase in mismatch in the non-bachelor’s sector, as shown in panel B.

As the labor market tightened during the initial recovery, employers reduced education requirements in the temporary-upskilling occupations so that some jobs in the degree sector were reallocated back to the non-degree sector. This shift would be expected to result in an initial decrease in mismatch index for the bachelor’s degree sector between 2010 and 2013, as shown in panel A. However, many of the persistent-upskilling occupations continued to increase the share of postings requiring a bachelor’s degree during the longer term, consistent with the subsequent increase in mismatch in the degree sector later in the recovery (after 2013). By contrast, panel B reveals less cyclical movement in the mismatch index for the non-degree sector since most non-degree occupations exhibited little or no significant upskilling.

Impact of Educational Upskilling on Workers without a Bachelor’s Degree

What are the implications of educational upskilling for workers? If workers were able to transition across educational levels and occupations more easily than the mismatch index indicated, we might overestimate the degree to which educational upskilling constrained hiring during the recession and recovery period. To test this, we examine two potential impacts of educational upskilling on workers. First, we compare whether workers without a bachelor’s degree had greater difficulty in finding a job relative to workers with a bachelor’s degree in occupations that experienced persistent educational upskilling versus those that did not. Figure 10 confirms that job-finding rates fell sharply during the Great Recession for all workers and did not start to recover until early 2011, well after the recession was officially over. Yet within each of our three upskilling categories, the job-finding rates for workers without a bachelor’s degree declined more steeply during the recession compared to those for workers with a bachelor’s degree, consistent with the relative decline in demand for workers by education level. This gap is most pronounced for persistent-upskilling occupations, consistent with employers increasing their demand for workers with a bachelor’s degree more sharply during the recession, and keeping those demands higher for longer, compared to occupations with temporary or no significant educational upskilling.

Figure 10.

Impact of Educational Upskilling on Job-Finding Rates by Educational Attainment

After 2011, job-finding rates improved for all workers as the economy recovered. However, the job-finding rates for workers without a bachelor’s degree increased faster than for their more-educated peers among occupations with temporary or no significant upskilling and exceeded that of workers with a bachelor’s degree by 2016. This pattern is consistent with the narrative that workers with a bachelor’s degree once again became a luxury rather than a necessity for these occupations as the labor market tightened between 2016 and 2019, although we cannot rule out that the tight labor market had stronger cyclical impacts on occupations with temporary or no significant upskilling (Autor, Dube, and McGrew 2023). By contrast, job-finding rates among workers without a bachelor’s degree were consistently lower than those of workers with a bachelor’s degree throughout the recovery period for persistent-upskilling occupations.¹⁵

The second impact we examine is on the wage rates of workers at the top versus the bottom of the distribution within occupations. Other researchers have noted that the sluggish aggregate wage growth during most of the recovery period seems inconsistent with the mismatch hypothesis (Rothstein 2012; Abraham 2015). The rapid increase in the demand for educational requirements relative to the supply of educated workers among occupations with persistent educational upskilling would suggest employers raise wages to attract workers with a bachelor’s degree to those positions, possibly increasing wage inequality between workers within occupations. Based on the same difference-in-difference approach as before, the results in Table 6 demonstrate that median wages increased among occupations with persistent educational upskilling relative to occupations with temporary or no upskilling during both the recession and recovery periods. The latter distinction is important because if occupational mismatch is present, then employment growth should be positively correlated with wage growth (Abraham 2015). In addition, changes in the ratio of wages at the 75th relative to the 25th percentile indicate that rising wages among persistent-upskilling occupations occurred at the top rather than the bottom of the wage distribution, suggesting that workers with a bachelor’s degree were likely the recipients of higher wages. This finding is consistent with recent evidence that workers who are not displaced from their occupation by technological change experience larger earnings gains (Braxton and Taska 2023).

Table 6.

Change in Real Wages within Occupations by Type of Educational Upskilling over Time

	Change relative to occupations that exhibited no educational upskilling
	Real median wages			Wage inequality
	Recession	Short-term recovery	Long-term recovery	Recession	Short-term recovery	Long-term recovery
	Δ2007–10	Δ2010–13	Δ2010–19	Δ2007–10	Δ2010–13	Δ2010–19
Diff persistent vs. no upskilling	0.084**(0.033)	0.147*(0.085)	0.124(0.087)	−0.010(0.150)	0.008(0.038)	0.094*(0.053)
Diff temporary vs. no upskilling	−0.022(0.028)	0.060(0.073)	0.016(0.075)	−0.059(0.128)	−0.01771(0.033)	0.016(0.046)
Diff (persistent–temporary)	0.106**	0.087	0.108	0.049	0.026	0.078*
(F-test p-value)	(0.032)	(0.244)	(0.156)	(0.705)	(0.438)	(0.094)
Baseline Diff for no upskilling occupations	0.033	−0.026	0.028	0.015	0.049	0.037
No. of three-digit occupations	94	94	94	94	94	94

Source: Authors’ calculations based on vacancy data provided by Lightcast, weighted by 2007 employment level from the American Community Survey (ACS). Data on wages are from the Occupational Employment Statistics (OES).

Notes: Each column is a separate regression in which the dependent variable is the change in wages and the omitted category is a dummy variable for occupations with no upskilling. Changes in real median wages calculated as the difference in log wages. Changes in wage inequality measured as the ratio of wages at the 75th percentile to the 25th percentile. Regressions also control for share of workers with a bachelor’s degree within each occupation as calculated from the 2005–07 ACS. Baseline difference for occupations with no upskilling listed for reference.

Statistical significance is indicated at the *** 1%, ** 5%, and * 10% levels, respectively.

Conclusions and Policy Implications

Using a novel database of roughly 200 million US online job postings, we find that movements in the demand for college-educated workers varied much more across occupations over the business cycle than was previously known. Many occupations (e.g., construction) experienced little or no educational upskilling, while others (e.g., community and social services) experienced only temporary educational upskilling that was mostly confined to the recession period. Only a subset of occupations (e.g., business and financial) exhibited a pattern of persistent educational upskilling that extended well after the Great Recession. Moreover, this persistence in requiring a bachelor’s degree was driven by educational upskilling within occupations rather than the changing composition of vacancies across occupations over time toward those with a higher share of postings requiring a bachelor’s degree.

Examining specific skillsets further reveals that the demand for software skills was a distinguishing feature of occupations with persistent educational upskilling. Relative to occupations that showed little or no educational upskilling, those experiencing temporary or persistent educational upskilling increased the share of job postings requiring software skills between 2007 and 2010, consistent with prior research indicating that recessions accelerate skill-biased technological change (Hershbein and Kahn 2018; Jaimovich and Siu 2020). Between 2010 and 2013, however, more than half of the increase in software skills was reversed among occupations experiencing temporary educational upskilling, whereas those experiencing persistent educational upskilling continued to increase their demand for software skills, demonstrating the complementarity between education and technology.

Other indicators suggest persistent educational upskilling had different consequences for workers with bachelor’s degrees compared to those without. Although the education levels of employed workers did increase, suggesting employers succeeded in hiring more-qualified workers, supply did not keep pace with demand for occupations that experienced persistent educational upskilling. As a result, the gap in job-finding rates between workers with and without a bachelor’s degree grew especially wide during the recession—and persisted for longer during the recovery—for occupations experiencing persistent educational upskilling, relative to those with either temporary or no upskilling. Moreover, wages increased among occupations with persistent educational upskilling, primarily at the top of the wage distribution, consistent with the need to attract workers with a bachelor’s degree.

Finally, we are also the first to document that educational upskilling contributed to reducing aggregate matching efficiency during the sluggish labor market recovery after the Great Recession. We develop an adjusted mismatch index to account for persistent educational upskilling within occupations and find that this produces a pattern of labor market mismatch that is less cyclical and more aligned with employer observations. Whereas the standard mismatch index calculated across occupations shows a marked increase during the Great Recession and a relatively quick recovery in the years immediately after, our adjusted mismatch index stays elevated for an extended period during the labor market recovery. This finding is consistent with prior evidence showing greater movements in the standard mismatch index across occupations in the bachelor’s degree sector as jobs were reallocated across educational sectors during the recession (Şahin et al. 2014).

We acknowledge, however, that our mismatch estimates are based on the number of unemployed job seekers in various occupations, ignoring the job-seeking behavior of both employed workers and individuals not in the labor force. We also acknowledge that we may underestimate mismatch due to educational upskilling by assuming no long-term impact arising from either occupations in which the share of postings requiring a bachelor’s degree reversed more than average (e.g., temporary upskilling) or occupations in which the initial increase in bachelor’s degree requirements was large in percentage terms but from a small initial base (e.g., no significant upskilling). Finally, this article focused only on educational upskilling, specifically that related to requiring a bachelor’s degree. There may have been other disruptions for workers in occupations classified as having temporary or no significant upskilling along the education dimension, for example if upskilling occurred along non-education dimensions for those workers. Nonetheless, based on our analysis, any such disruptions did not pose a persistent barrier in job-finding rates or wage growth for less-educated workers in such occupations.

Taken together, our findings contribute to the literature by identifying educational upskilling related to technological change as a factor in reducing aggregate matching efficiency, in ways not previously recognized by economists. Specifically, our findings suggest that lower matching efficiency in the US labor market after the Great Recession may reflect a shift in demand toward more specialized jobs that require particular software skills, thus leading to imbalances between the demand for and supply of educational credentials. This finding is supported by recent research showing that unemployed workers displaced by technology direct their job search toward new occupations in which their skills are still employable but wages are lower (Braxton and Taska 2023). As a result, search-and-matching models of the labor market should account for periods of persistent educational upskilling, when workers are more likely to be chasing a moving target for re-employment (Kambourov and Manovskii 2009; Alvarez and Shimer 2011; Carrillo-Tudela and Visschers 2023).

Last, our findings also contribute to debates about workforce development and related educational policies by documenting the adverse impacts of persistent educational upskilling on workers without a bachelor’s degree. For example, recognizing that meaningful shifts in educational requirements occur only in certain occupations, rather than economy-wide, can guide workforce development practitioners to better target sector-based training (Holzer 2015). Similarly, understanding that educational requirements can shift rapidly should incentivize educational institutions and training providers to partner more closely with employers in monitoring job qualifications, adjusting curriculum development, and advising students, particularly during recessions. Moreover, distinguishing between persistent and temporary shifts in educational demands within occupations could help policymakers identify which human capital investments are worthwhile in the long run (e.g., vocational versus baccalaureate degrees) and encourage employers to shift from credential toward skills-based hiring in the short run, especially in occupations with well-defined skill requirements (e.g., healthcare). Finally, knowing that persistent educational upskilling is likely to affect certain groups of workers more than others can help career counselors tailor their coaching for job seekers based on the suitability of their qualifications for various jobs and retraining opportunities.

Supplemental Material

sj-pdf-1-ilr-10.1177_00197939251413182 – Supplemental material for No Longer Qualified? Changes in the Supply and Demand for Skills within Occupations

Supplemental material, sj-pdf-1-ilr-10.1177_00197939251413182 for No Longer Qualified? Changes in the Supply and Demand for Skills within Occupations by Alicia Sasser Modestino, Mary A. Burke, Shahriar Sadighi, Rachel Sederberg, Tomer Stern and Bledi Taska in ILR Review

Footnotes

ORCID iDs

Alicia Sasser Modestino

Mary A. Burke

This work was funded by a seedling grant from the NULab for Texts, Maps, and Networks at Northeastern University. The views expressed herein are those of the authors and do not indicate concurrence by the Federal Reserve Bank of Boston, the principals of the Board of Governors, or the Federal Reserve System.

An Online Appendix is available at .

For general questions as well as for information regarding the data and/or computer programs used to generate the results presented in the article, please contact the corresponding author, Alicia Modestino, at a.modestino@northeastern.edu.

1

Lightcast data are unavailable for 2008 and 2009 because of operational changes when the company was founded. See the for more details.

2

This normalized Lightcast data set is only available for 2007 and 2010–2017.

3

We follow the prior literature and use the percentage point change in educational requirements to avoid designating occupations with large percentage changes from a small initial base as posing a significant barrier for workers without a bachelor’s degree.

4

Results are qualitatively similar using a more restrictive definition of persistent upskilling that does not allow for any reversion in the share of postings requiring a bachelor’s degree.

5

We decompose the net increase in the share of vacancies requiring a bachelor’s degree for a given two-digit occupation into movements within versus between the underlying three-digit occupations for the 2007–2010 period.

6

assumed each unemployed worker searches for jobs in a particular occupation and each employer encounters only the workers searching in that occupation.

7

In the Online Appendix, we also replicate to calculate the standard mismatch index separately by educational sector using the Lightcast data.

8

These parameter values are found in and vary only across three-digit occupations—there is no common component across the persistent upskillers versus other occupations. The parameters δ and (1 −δ) capture the vacancy and unemployment elasticity of hires as before.

9

This produces more conservative estimates of mismatch within occupations since increases in education requirements for temporary-upskilling occupations were partially reversed and those for non-upskilling occupations were small or nonexistent.

10

Approximately 8% of unemployed workers in the CPS cannot be assigned an occupation because of missing data or gaps in the crosswalk. Results are qualitatively similar for one- and two-month job-finding rates.

11

The results are qualitatively similar when decomposing changes within three-digit occupations based on the underlying occupations at the six-digit level. See .

12

Similarly, the increase in the share of postings for “common” skills such as communication and “specialized” skills such as budget management reflected an increased prevalence for existing skills rather than requests for new skills or a greater variety of skills across occupations.

13

By construction, the mismatch index is increasing in the level of disaggregation.

14

In the Online Appendix, we replicate this prior analysis from using the Lightcast data.

15

Comparing education levels of new hires relative to continuing employees, shows that occupations experiencing persistent educational upskilling were more successful in hiring new workers with a bachelor’s degree during the initial recovery compared to occupations with temporary or no significant upskilling.

References

Abraham

Katharine G.

2015. Is skill mismatch impeding U.S. economic recovery? ILR Review 68(2):291–313.

Acemoglu

Daron

Autor

David

. 2011. Skills, tasks, and technologies: Implications for employment and earnings. In Card

David

Ashenfelter

Orley

(Eds.), Handbook of Labor Economics, Volume 4A, pp. 1043–71. Amsterdam: North Holland Publishing.

Alvarez

Fernando

Shimer

Robert

. 2011. Search and rest unemployment. Econometrica 79(1):75–122.

Autor

David H.

Dorn

David

. 2013. The growth of low-skill service jobs and the polarization of the US labor market. American Economic Review 103(5):1553–97.

Autor

David H.

Dube

Arindrajit

McGrew

Annie

. 2023. The unexpected compression: Competition at work in the low wage labor market. NBER Working Paper No. 31010. Cambridge, MA: National Bureau of Economic Research.

Autor

David H.

Goldin

Claudia

Katz

Lawrence

. 2020. Extending the race between education and technology. NBER Working Paper No. 26705. Cambridge, MA: National Bureau of Economic Research.

Autor

David H.

Katz

Lawrence F.

Kearney

Melissa S.

2008. Trends in U.S. wage inequality: Revising the revisionists. Review of Economics and Statistics 90(2):300–323.

Autor

David H.

Katz

Lawrence F.

Krueger

Alan B.

1998. Computing inequality: Have computers changed the labor market? Quarterly Journal of Economics 113(4):1169–213.

Autor

David H.

Levy

Frank

Murnane

Richard J.

2003. The skill content of recent technological change: An empirical exploration. Quarterly Journal of Economics 118(4):1279–334.

10.

Barlevy

Gadi

. 2011. Evaluating the role of labor market mismatch in rising unemployment. Economic Perspectives 35(3):82–96.

11.

Beaudry

Paul

Green

David A.

Sand

Benjamin M.

2016. The great reversal in the demand for skill and cognitive tasks. Journal of Labor Economics 34(S1):S199–S247.

12.

Bessen

James

. 2014. Employers aren’t just whining—the skills gap is real. Harvard Business Review, August25.

13.

Bessen

James

Denk

Erich

Meng

Chen

. 2022. The remainder effect: How automation complements labor quality. Boston University School of Law Research Paper No. 22-3.

14.

Birinci

Serdar

See

Kurt

Wee

Shu Lin

. 2023. Job applications and labor market flows. Federal Reserve Bank of St. Louis Working Paper No. 2020-023. https://doi.org/10.20955/wp.2020.023

15.

Blair

Peter Q.

Deming

David J.

2020. Structural increases in demand for skill after the Great Recession. AEA Papers and Proceedings 110:362–65. doi: 10.1257/pandp.20201064

16.

Braxton

J. Carter

Taska

Bledi

. 2023. Technological change and the consequences of job loss. American Economic Review 113(2):279–316.

17.

Burke

Mary A.

2015. The Rhode Island labor market in recovery: Where is the skills gap? Federal Reserve Bank of Boston: Current Policy Perspectives No. 15-7.

18.

Cappelli

Peter H.

2015. Skill gaps, skill shortages, and skill mismatches: Evidence and arguments for the United States. ILR Review 68(2):251–90.

19.

Carnevale

Anthony P.

Jayasundera

Tamara

Cheah

Ban

. 2012. The college advantage: Weathering the economic storm. Report. Georgetown University, McCourt School on Public Policy, Center on Education, and the Workforce.

20.

Carrillo-Tudela

Carlos

Visschers

Ludo

. 2023. Unemployment and endogenous reallocation over the business cycle. Econometrica 91(3):1119–53.

21.

Cavounidis

Costas

Dicandia

Vittoria

Lang

Kevin

Malhotra

Raghav

. 2021. The evolution of skill use within and between jobs. Working Paper No. 29302. Cambridge, MA: National Bureau of Economic Research.

22.

Charles

Kerwin Kofi

Hurst

Erik

Notowidigdo

Matthew J.

2012. The masking of the decline in manufacturing employment by the housing bubble. Journal of Economic Perspectives 30(2):179–200.

23.

Conzelmann

Johnathan G.

Hemelt

Steven W.

Hershbein

Brad J.

Martin

Shawn

Simon

Andrew

. 2024. Skills, majors, and jobs: Does higher education respond? Upjohn Institute Working Paper No. 24-400. Kalamazoo, MI: Upjohn Institute.

24.

Davis

Steven J.

Faberman

R. Jason

Haltiwanger

John C.

2012. Recruiting intensity during and after the Great Recession: National and industry evidence. American Economic Review 102(3):584–88.

25.

Diamond

Peter A.

2013. Cyclical unemployment, structural unemployment. IMF Economic Review 61(3):410–55.

26.

Diamond

Peter A.

Şahin

Ayşegül

. 2015. Shifts in the Beveridge Curve. Research in Economics 69(1):18–25.

27.

Hall

Robert E.

Schulhofer-Wohl

Sam

. 2018. Measuring job-finding rates and matching efficiency with heterogeneous jobseekers. American Economic Journal 10(1):1–32.

28.

Hershbein

Brad

Kahn

Lisa

. 2018. Do recessions accelerate routine-biased technological change? Evidence from vacancy postings. American Economic Review 108(7):1737–72.

29.

Hobijn

Bart

Perkowski

Patryk

. 2016. The industry-occupation mix of US job openings and hires. http://dx.doi.org/10.2139/ssrn.2858603

30.

Holzer

Harry J.

2015. Job market polarization and U.S. worker skills: A tale of two middles. Working Paper. Washington, DC: The Brookings Institution. https://www.brookings.edu/wp-content/uploads/2016/06/polarization_jobs_policy_holzer.pdf

31.

Jaimovich

Nir

Siu

Henry E.

2020. Job polarization and jobless recoveries. Review of Economics and Statistics 102(1):129–47.

32.

Kambourov

Gueorgui

Manovskii

Iourii

. 2009. Occupational specificity of human capital. International Economic Review 50(1):63–115.

33.

Katz

Lawrence F.

Murphy

Kevin M.

1992. Changes in relative wages, 1963–1987: Supply and demand factors. Quarterly Journal of Economics 107(1):35–78.

34.

Kogan

Leonid

Papanikolaou

Dimitris

Schmidt

Lawrence D.W.

Seegmiller

Bryan

. 2021. Technology, vintage-specific human capital, and labor displacement: Evidence from linking patents with occupations. NBER Working Paper No. 29552. Cambridge, MA: National Bureau of Economic Research.

35.

Kuhn

Peter J.

Skuterud

Mikal

. 2004. Internet job search and unemployment durations. American Economic Review 94(1):218–32.

36.

Lazear

Edward P.

Spletzer

James R.

2012. The United States labor market: Status quo or new normal? Economic Policy Symposium – Jackson Hole, Wyoming. Federal Reserve Bank of Kansas City, pp. 405–51.

37.

Liu

Yujia

Grusky

David B.

2013. The payoff to skill in the third industrial revolution. American Journal of Sociology 118(5): 1330–74.

38.

Marinescu

Ioana

. 2017. The general equilibrium impacts of unemployment insurance: Evidence from a large online job board. Journal of Public Economics 150: 14–29.

39.

Modestino

Alicia Sasser

Shoag

Daniel

Ballance

Joshua

. 2016. Downskilling: Changes in employer skill requirements over the business cycle. Labour Economics 41:333–47.

40.

———. 2020. Upskilling: Do employers demand greater skill when workers are plentiful? Review of Economics and Statistics 102(4):793–805.

41.

Restrepo

Pascual

. 2015. Skill mismatch and structural unemployment. Unpublished manuscript. http://pascual.scripts.mit.edu/research/01/skill_mismatch.pdf

42.

Rothstein

Jesse

. 2012. The labor market four years into the crisis: Assessing structural explanations. ILR Review 65(3):437–500.

43.

Rothwell

Jonathan

. 2012. Education, job openings, and unemployment in metropolitan America. Metropolitan Policy Program Report. Washington, DC: The Brookings Institution. https://www.brookings.edu/wp-content/uploads/2016/06/29-education-gap-rothwell.pdf

44.

Şahin

Ayşegül

Song

Joseph

Topa

Giorgio

Violante

Giovanni L.

2014. Mismatch unemployment. American Economic Review 104(11):3529–64.

45.

Tüzemen

Didem

Willis

Jonathan L.

2013. The vanishing middle: Job polarization and workers’ response to the decline in middle-skill jobs. Economic Review (Q1):5–32. https://www.kansascityfed.org/publicat/econrev/pdf/13q1Tuzemen-Willis.pdf

46.

Weaver

Andrew

Osterman

Paul

. 2017. Skill demands and mismatch in U.S. manufacturing industries. ILR Review 70(2):275–307.

47.

Zago

Riccardo

. 2018. Job polarization, skill mismatch and the Great Recession. http://dx.doi.org/10.2139/ssrn.3004759