outcomes and future labor market outcomes, they do not believe there is a strong connection between

these domains. Supporting this observation, the individual-specific treatment effect on semester GPA is

only weakly correlated with the individual-specific treatment effects on finding a job before graduation

(corr=0.0497, p=0.065) and expected earnings at 35 (corr=0.0467, p=0.077). The one notable exception to

the lack of heterogeneity in panels (e) and (f) of Figure 1 is the 2020 cohort, which on average revised their

subjective probability of finding a job before graduation three times as much as other cohorts. However,

students expecting to graduate in 2021 and later still believe that COVID-19 will impact their job market

outcomes; even students who plan to graduate in 2023 expect to be 6 percentage points less likely to find a

job before graduating due to COVID-19.

5 Mechanisms behind the COVID-19 Effects

This section presents mediation analysis on the drivers of the underlying heterogeneity in the treatment

effects. The COVID-19 pandemic serves as both an economic and a health shock. However, these shocks may

have been quite heterogeneous across the various groups, and that could partly explain the heterogeneous

treatment effects we documented in the previous section.

5.1 Economic and Health Mediating Factors

We proxy for the financial and health shocks due to COVID-19 by relying on a small but relevant set

of exogenous/predetermined variables. Financial shocks are characterized based on whether a student lost

a job due to COVID-19, whether a student’s family members lost income due to COVID-19, the change in

a student’s monthly earnings due to COVID-19, and the likelihood a student will fail to fully meet debt

payments in the next 90 days. To measure health shocks, we consider a student’s belief about the likelihood

that they will be hospitalized if they contract COVID-19, a student’s belief about the likelihood that they

will have contracted COVID-19 by summer, and a student’s subjective health assessment. Finally, in order

to summarize the combined effect of each set of proxies, we also construct principal component scores as

one-dimensional measures of the financial and health shock to students.14

Table 3 reports summary statistics of the different economic and health proxies by demographic group.

Given the results in Figure 1, the remainder of the analysis will focus on three socioeconomic divisions:

parental income, gender, and Honors college status. Our data indicate that lower-income students faced

larger health and economic shocks as compared to their more affluent peers. In particular, they are almost

10 percentage points more likely to expect to default on their debt payments than their higher-income
14 Eigenvalues indicate the presence of only one principal component for each of the shocks.

11
counterparts. Additionally, lower-income students are 16 percentage points more likely to have had a close

family member experience an income reduction due to COVID-19. Regarding the health proxies, lower-

income students rate their health as worse than higher-income students and perceive a higher probability of

being hospitalized if they catch the virus. Finally, the differences in economic and health shocks between

lower and higher-income students, as summarized by the principle components of the selected proxy variables,

are statistically significant.

Columns (5)-(7) of Table 3 show that both economic and health shocks are larger for non-Honors students.

In fact, the average differences in the principal component scores for both the economic and health factors

is larger for these two groups than for the income groups. Finally, the last three columns of the table show

that women experienced larger COVID-19 shocks due to economic and health factors. These differences

are partly driven by the fact that, in our sample, females are more likely to report that they belong to a

lower-income household than males (50% vs. 42%).

In short, Table 3 makes clear that the impacts of COVID-19 on the economic well-being and health of

students have been quite heterogeneous, with lower-income and lower-ability students being more adversely

affected.

5.2 The Role of Economic and Health Shocks on Explaining the COVID-19
Effects

To investigate the role of economic and health shocks in explaining the heterogeneous treatment effects

(in section 4.2), we estimate the following specification:

∆i = α0 + α1 Demogi + α2 F inShocki + α3 HealthShocki + i , (2)

where ∆i is the COVID-19 treatment effect for outcome O on student i. Demogi is a vector including

indicators for gender, lower-income, Honors status, and dummies for cohort year and major. F inShocki and

HealthShocki are vectors containing the shock proxies or their principal component. Finally, i denotes an

idiosyncratic shock.

The parameters of interest are α2 and α3 . A causal interpretation of these parameters hinges on the

identification assumption that F inShocki and HealthShocki are independent of i . In our context, we

believe that this assumption is reasonable given that F inShocki and HealthShocki contain variables that

capture arguably exogenous changes due explicitly to the COVID-19 outbreak or were determined before the

pandemic began. For example, it is reasonable to assume that a family member losing a job or a student’s

baseline health status is independent of the idiosyncratic term. On the other hand, if a student losing a

12
job or missing a future debt payment is correlated with idiosyncratic factors that are also likely to affect

the change in the student’s outcomes (such as change in graduation timing), the identification assumption

would be invalid. In this case (where the omitted factors are positively correlated with the shock proxies),

the overall ability of the proxies to explain the treatment effect may then be interpreted as an overestimate

of the true role of each type of shock in shaping treatment effects. Thus, in the presence of unobservable

variables, we should be cautious in interpreting the coefficients on the proxy variables.

Table 4 shows estimates of equation (2) for four different outcomes (Appendix Table A2 shows the esti-

mates for additional outcomes). For each outcome, five specifications are reported ranging from controlling

for only demographic variables in the first specification to controlling for both economic and health factors

in the fourth specification. Finally, the last column includes only the principal component of each shock

to provide insight about overall effects, given that certain shock proxies show high levels of correlation (see

appendix Table A4 for the correlations within each set of proxies).

Several important messages emerge from Table 4. First, both shocks are (economically and statistically)

significant predictors of the COVID-19 effects on students’ outcomes. In particular, F-tests show that the

financial and health shock proxies are jointly significant across almost all specifications.15 This is also

reflected in the statistical significance of the principal components. Moreover, the fact that the effect of key

proxy variables remains robust when we simultaneously control for both shocks demonstrates the robustness

of our results. For example, we find that a 50 percentage point increase in the probability of being late on

debt payments increases the probability of delaying graduation and switching majors due to COVID-19 by

6.9 and 6.4 percentage points, respectively. These effects are large given that they represent more than half

of the overall COVID-19 treatment effect for these variables. Similarly, we find that an analogous increase

in the probability of hospitalization if contracting COVID-19 leads to a 6 and 5 percentage points increase

in the probability of delaying graduation and switching majors due to COVID-19.

Second, in terms of labor market expectations, we find that the change in the expected probability of

finding a job before graduation strongly depends on having a family member that lost income (which is also

correlated with the student himself losing a job). In particular, the size of this effect represents 32% of the

overall COVID-19 treatment effect. Therefore, this finding suggests that students’ labor market expectations

are driven in large part by personal/family experiences.

Third, although the proxies play an important role in mediating the pandemic’s impact on students, there

is still a substantial amount of variation in COVID-19 treatment effects left unexplained. Across the four

outcomes in Table 4, the full set of proxies explain less than a quarter of the variation in outcomes across

individuals. Appendix Figure A2 visualizes this variation by plotting the distribution of several continuous
15 The only exception is the financial shock when explaining changes in the probability of taking classes online.

13
outcomes with and without controls. While the interquartile range noticeably shrinks after conditioning on

the proxy variables, these plots highlight the large amount of variation in treatment effects remaining after

conditioning on the proxies.

Finally, our results show that the financial and health shocks play an important role in explaining the

heterogeneous effects of the COVID-19 outbreak. In particular, columns (4) and (9) demonstrate that

economic and health factors together can explain approximately 40% and 70% of the income gap in COVID-

19’s effect on delayed graduation and changing major respectively. The gap between Honors and non-Honors

students is likewise reduced by 27% and 39% for the same outcomes. Taken together, these results imply

that differences in the magnitude of COVID-19’s economic and health impact can explain a significant

proportion of the demographic gaps in COVID-19’s effect on the decision to delay graduation, the decision

to change major, and preferences for online learning. These results are important and suggest that by

focusing on mitigating the economic and health burden imposed by COVID-19, policy makers may be able

to significantly prevent COVID-19 from exacerbating existing achievement gaps in higher education.

6 Conclusions

This paper provides the first systematic analysis of the effects of COVID-19 on higher education. To

study these effects, we surveyed 1,500 students at Arizona State University, and present quantitative evidence

showing the negative effects of the pandemic on students’ outcomes and expectations. For example, we find

that 13% of students have delayed graduation due to COVID-19. Expanding upon these results, we show

that the effects of the pandemic are highly heterogeneous, with lower-income students 55% more likely to

delay graduation compared to their higher-income counterparts. We further show that the negative economic

and health impacts of COVID-19 have been significantly more pronounced for less advantaged groups, and

that these are partially responsible for the underlying heterogeneity in the impacts that we document. Our

results suggest that by focusing on addressing the economic and health burden imposed by COVID-19, as

measured by a relatively narrow set of mitigating factors, policy makers may be able to prevent COVID-19

from widening existing achievement gaps in higher education.

14
References

Angrist, J. D. and J.-S. Pischke (2009). Mostly harmless econometrics: an empiricist’s companion. Princeton:

Princeton University Press. OCLC: ocn231586808.

Arcidiacono, P., V. J. Hotz, A. Maurel, and T. Romano (2020, March). Ex Ante Returns and Occupational

Choice.

Chetty, R., J. N. Friedman, N. Hendren, and M. Stepner (2020, May). Real-Time Economics: A New

Platform to Track the Impacts of COVID-19 on People, Businesses, and Communities Using Private

Sector Data.

Heckman, J. J. and E. Vytlacil (2005, May). Structural Equations, Treatment Effects, and Econometric

Policy Evaluation. Econometrica 73 (3), 669–738.

Imbens, G. and D. B. Rubin (2015). Causal inference: for statistics, social and biomedical sciences : an

introduction. OCLC: 985493948.

Jacob, B., B. McCall, and K. Stange (2018, April). College as Country Club: Do Colleges Cater to Students’

Preferences for Consumption? Journal of Labor Economics 36 (2), 309–348.

Kuchler, T. and B. Zafar (2019, October). Personal Experiences and Expectations about Aggregate Out-

comes. The Journal of Finance 74 (5), 2491–2542.

Kuhfeld, M., J. Soland, B. Tarasawa, A. Johnson, E. Ruzek, and J. Liu (2020, May). Projecting the potential

impacts of COVID-19 school closures on academic achievement. Publisher: EdWorkingPapers.com.

Malmendier, U. and S. Nagel (2016, February). Learning from Inflation Experiences *. The Quarterly

Journal of Economics 131 (1), 53–87.

Oreopoulos, P., T. von Wachter, and A. Heisz (2012, January). The Short- and Long-Term Career Effects

of Graduating in a Recession. American Economic Journal: Applied Economics 4 (1), 1–29.

Rothstein, J. (2020, May). The Lost Generation? Labor Market Outcomes for Post Great Recession Entrants.

Shapiro, M. and P. Giustinelli (2019, July). SeaTE: Subjective ex ante Treatment Effect of Health on

Retirement.

Wiswall, M. and B. Zafar (2018, February). Preference for the Workplace, Investment in Human Capital,

and Gender. The Quarterly Journal of Economics 133 (1), 457–507.

15