The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
Federal Reserve Bank of Chicago
The COVID-19 Pandemic and
Asian American Employment
Bo E. Honoré and Luojia Hu
REVISED
July 2021
WP 2020-19
https://doi.org/10.21033/wp-2020-19
Working papers are not edited, and all opinions and errors are the
responsibility of the author(s). The views expressed do not necessarily
reflect the views of the Federal Reserve Bank of Chicago or the Federal
Reserve System.
*
The COVID-19 Pandemic and Asian American
Employment∗
Bo E. Honoré†
Luojia Hu‡
First draft September 2020; this version July 2021
Abstract
Recent studies have documented the disparate impact of the COVID-19 pandemic on
labor market outcomes for different racial groups. This paper adds to this literature
by documenting that the employment of Asian Americans - in particular those with
no college education - has been especially hard hit by the economic crisis associated
with the onset of the pandemic. This can only partly be explained by differences in
demographics, local market conditions, and job characteristics, and it also cannot be
entirely explained by possible different selection into education levels across ethnic
groups. The burden on Asian Americans is primarily borne by those who are not
US-born.
Key Words:Employment, Pandemic, Asian Americans, Racial Disparity.
JEL Code: J21, J70, J71.
∗
This research was supported by the Gregory C. Chow Econometric Research Program at Princeton University and by the National Science Foundation (Grant Number SES-1530741). We thank Laura Pilossoph
and seminar participants at the Federal Reserve Bank of Chicago and System Applied Micro Conference
and Day Ahead Conference for constructive and helpful comments. Daniel Aaronson and Brian Phelan
generously shared the program for mapping occupations to job tasks. Aastha Rajan and Sharada Dharmasankar provided excellent research assistance. The opinions expressed here are those of the authors and
not necessarily those of the Federal Reserve Bank of Chicago or the Federal Reserve System.
†
Mailing Address: Department of Economics, Julis Romo Rabinowitz Building, Princeton, NJ 08544.
Email: honore@Princeton.edu.
‡
Mailing Addresses: Economic Research Department, Federal Reserve Bank of Chicago, 230 S. La Salle
Street, Chicago, IL 60604. Email: lhu@frbchi.org.
1
1
Introduction
A number of papers have documented the disparate impact of the COVID-19 pandemic on
labor market outcomes across racial groups. This includes Bartik, Bertrand, Lin, Rothstein,
and Unrath (2020), Cortes and Forsythe (2020), and Dam, Gaur, Karahan, Pilossoph, and
Schirmer (2021). This literature has primarily focused on Blacks and Hispanics. At the
same time, there is evidence that the health of Asian Americans in New York City was
disproportionately impacted by the pandemic. See, for example, Marcello, Dolle, Tariq,
Kaur, Wong, Curcio, Thachil, Yi, and Islam (2020). Finally, there has been increasing
focus on Anti-Asian sentiment. The rhetoric surrounding the “China-virus” is a memorable
example of this, and this has lead to the recent passing of the US Senate anti-Asian American
hate crime bill (Wu (2021)). The contribution of this paper is to study the disparity across a
broader group of ethnicities which includes Asian Americans. We find that Asian Americans
were also disproportionately hard hit by the onset of the pandemic in terms of employment.
The simplest facts are displayed in Figure 1. The figure shows that the employment of
minority groups (Blacks, Hispanics, and Asian Americans) was more negatively impacted
by the pandemic than the employment of Whites. The decline in employment was sharpest
at the start of the pandemic. For example, the employment rate for all men dropped by
9 percentage points from the first to the second quarter of 2020 and it only recovered by
5.8 percentage points from the second to the fourth quarter. While all groups experienced
a fall in employment, the effect was most dramatic for low-educated Asian Americans. For
example, Asian American men with a high school degree or less experienced a 31 percentage
point drop in employment in the second quarter.
The empirical patterns regarding Asian American employment seem to have been largely
overlooked. The report by Mar and Ong (2020) is a notable exception. They report aggregate
statistics to compare the unemployment rate for Asian Americans and Whites before and
after the onset of the crisis. The contribution of this paper is to use micro data to investigate
the extent to which the different employment patterns across the groups (especially Asian
Americans) reported above can be explained by differences in demographics, local labor
market conditions, and job characteristics at the individual level. In order to allow for the
2
Figure 1: Fraction Working by Group
possibility that the labor market treatment of Asian Americans is related to their appearance,
we initially disaggregate the group of Asian Americans into those of (“Chinese-looking”) East
Asian descent, those of South-East Asian descent, and the remaining group of Asians, and
compare those groups to Whites, Blacks, and Hispanics.
We argue that the differences in the effect of the pandemic on employment cannot be explained by differences in other demographic characteristics or local labor market conditions.
We document these findings in Section 2 below, and we demonstrate that the results are
qualitatively robust to whether one uses a linear fixed effects estimator or a modified conditional logit maximum likelihood estimator that controls for fixed effects. In that section, we
also compare the effects of the pandemic on different ethnic groups to the effects of the Great
Recession. We find that in contrast to the pandemic, there was no notable difference in the
3
effects of the 2008 economic downturn on Asians-Americans and Whites. This is perhaps
not surprising since the COVID-19 pandemic affected very different sectors of the economy
than previous downturns.
In the current pandemic, some industries are considered “essential” and the employment
in those industries has been impacted very differently from those considered non-essential.
At the same time, it is easier to work from home in certain occupations than in others. Since
workers of different ethnicities tend to work in different types of jobs, it is important to
control for industry or occupation when estimating the differential impact on employment
across different ethnicities. One problem with including industry or occupation as explanatory variables in a regression for the probability of working, however, is that individuals may
answer questions about the industry or occupation in a way that depends on their current
employment status. This will make variables for industry or occupation endogenous. In
Section 3, we therefore focus on transitions in and out of employment controlling for an
individual’s industry and occupation in their previous interview month. We find that the
large differential change in the probability of employment across ethnicity is mainly driven by
differences in the probability of remaining employed and less by differences in the probability
of gaining employment. We also find that differences in occupation and industry can explain
up to half of the differential between Asians and Whites in the probability of remaining employed. For example, after controlling for other demographics, local labor market conditions,
and occupation, low-educated East and South-East Asian American men experienced a 12
percentage point higher probability of losing their job in the second quarter of 2020 than
comparable Whites. Without controlling for occupation, the corresponding number is 19
percentage points.
In Section 4, we explore various explanations for the remaining large differences between
Asian Americans and other groups. One potential explanation is that the well-documented
increase in Anti-Asian sentiments in the United States around the onset of the pandemic has
lead to increased labor market discrimination against Asian Americans. Another possible
difference between groups is household structure. For example, if Asian Americans are more
likely to live in multi-generational households, then concerns for the health of older family
members might make them more likely to withdraw from the labor market in response to the
4
pandemic. Finally, many Asian American are recent immigrants. This raises the question of
whether the larger effect on Asian Americans is driven by their country of birth. A number
of interesting findings emerge from the analysis in this section. First, we find no evidence
that the differential impact on Asians is larger in occupations that are characterized by
higher intensity of interpersonal tasks. We also do not find evidence that Asian Americans
are especially hard hit in states with larger anti-Asian sentiments. Second, the patterns do
not seem to be driven by individuals who stop working due to health concerns about elderly
household members. Finally, we find that the impact on Asian Americans is largely borne
by recent immigrants.
The effect of the pandemic on employment differs significantly across education groups
as well as ethnicity. At the same time, Asian Americans tend to be higher educated than
other groups. This raises the question of whether the especially large effect on low-educated
Asian Americans is driven by different selection into educational attainment across ethnic
groups. In Section 5, we find that this cannot fully explain the differences in the effect of
the pandemic on employment across ethnic groups. Section 6 concludes the paper.
2
The Probability of Working
In this section, we first present some simple facts about the differences in employment outcomes across gender, education and ethnic groups before and during the pandemic. We then
turn to simple regressions models that control for other observable factors.
2.1
Data and Summary Statistics
We use the Current Population Survey (CPS) Basic Monthly micro data from January of
2019 to March of 2021, and restrict the data to individuals aged 25-65. We include data
from 2019 as well as 2020 to control for seasonality in working that varies by ethnicity.
The simplest facts regarding employment were already displayed in Figure 1. The figure
displays the fraction of each group that reports working between January 2019 and March
5
2021.
1
Overall, the employment rate for men declined by 9 percentage points in the second
quarter of 2020 relative to the previous quarter. The minority groups, Blacks, Hispanics, and
Asians, suffered steeper declines ranging between 10 and 14 percentage points. By the fourth
quarter of 2020, the decline in employment for Asian Americans had recovered to close to
that of Whites (at approximately a 2.5 percentage point lower than the first quarter of 2020),
while the negative effects on Blacks and Hispanics persisted at more than 4 percentage points
below their employment rate in the first quarter of 2020. The sharp decline in employment
in the second quarter of 2020 is also evident from Figure 2, which displays the fraction of
individuals employed in a given month relative to the same month in 2019.
Figure 2: Fraction Working by Group Relative to Same Month in 2019
The most striking feature of Figures 1 and 2 is a dramatic decline in employment for
1
Additional details are presented in the Appendix Tables 1 and 2.
6
Asian Americans with a high school degree or less. In the first quarter of 2020, 77% of Asian
men in this group reported working. In the second quarter, the rate fell by 31 percentage
points to 46%. By contrast, the changes for comparable Whites, Blacks and Hispanics were
approximately 9, 10 and 12 percentage points, respectively. The patterns in the changes for
women with a high school degree or less are similar to those for men.
These patterns found in Figure 1 also appear if one changes the definition of working
to also include those who reported “has a job, but not at work last week”. For both men
and women, the largest drops in employment were for Asian Americans with a high school
degree or less. The results are displayed in Figure 3 (See also Appendix Tables 3 and 4).
For the rest of this paper, we focus on the first definition of working. The reason is that
the alternative definition seems to be subject to a great deal of mis-classification since the
beginning of the pandemic (See US Bureau of Labor Statistics (2020)).
Figure 3: Fraction Working by Group (Alt. Definition)
7
2.2
Controlling for Other Characteristics
In this subsection, we present the results from estimating linear probability models for the
probability of working. We use data from April 2019 to March 2021 (covering twelve months
before the onset of the pandemic and twelve months after). Our main goal is to document
how working depends on ethnicity before and during the pandemic. Since Asian Americans
are a very heterogeneous group, we initially disaggregate the group into three categories:
East Asians (“AsianE” in the tables), South-East Asians (“AsianSE”), and other Asians
(“AsianOther”).2 The variables of interest will be ethnicity dummies and - especially - their
interactions with the COVID periods, starting with the second quarter of 2020 (the pandemic
quarters). Below, and in the tables, we refer to the last three quarters of 2020 and the first
quarter of 2021 as “Cr2”, “Cr3”, “Cr4”, and “Cr5”.
Whether someone is working is likely to be influenced by the demographic characteristics of the individual. Therefore, for each combination of sex and educational group, we
estimate a linear probability model for working that controls for age, age squared, marital
status, number of children, number of children under the age of 5, and interactions between
indicators of (a) each month of the year, (b) state of residence, and (c) each of the pandemic
months starting in April of 2020. This controls for geographic variation in the impact of the
pandemic on the local economy.
To allow for the possibility that the seasonal effects differ by group, we include interactions between indicators of (a) each quarter of the year, (b) each ethnicity, and (c) each
pandemic quarter starting with the second quarter of 2020. The parameters of interest are
the coefficients on the interactions between the ethnicity variables and pandemic quarters.
The estimated coefficients for the key parameters of interest are reported in Table 1.3 The
2
We define AsianAll based on the IPUMS CPS variable “race”. Specifically, AsianAll equals 1 if race is
either recorded as “asian only”, “hawaiian/pacific islander only”, “white-asian”, or “black-asian”. AsianAll
is divided into three groups: AsianE, AsianSE and AsianOther. The first is the subset that are either born,
or has a parent born, in Japan, North Korea, Taiwan, Hong Kong, South Korea, Mongolia, China or Macau.
The second is the subset that are not in AsianE and are either born, or has a parent born, in Brunei,
Cambodia, Indonesia, Laos, Malaysia, Burma, Myanmar, Philippines, Singapore, Thailand, Timor Leste, or
Vietnam. The third group, AsianOther, is the remaining set.
3
We use the population weights from the CPS. The standard errors for all the regressions are clustered
at the household level. This accounts for within-household correlation in unobservables. It also accounts for
the fact that the CPS has a panel structure, with individuals being interviewed for four consecutive months,
8
results in Table 1 are most striking for individuals with a high school degree or less (the first
and fourth columns). Generally speaking, Asian Americans in this group were much harder
hit by the pandemic than any other group. This is especially true in the second quarter of
2020, and particularly for East and South East Asians. Controlling for demographics, East
Asian men and women have an estimated additional 21 and 16 percentage point drop in
the probability of working in the second quarter relative to their white counterparts4 . The
point estimates for the corresponding drop in the probability of working between the first
and third quarter were approximately 12 and 6 percentage points. This pattern also holds
for South East Asians, although to a lesser degree. The differentials between Whites and
Blacks or Hispanics are generally much smaller.
The results for the top end of the educational attainment distribution (columns 3 and 6 of
Table 1) are very different. For men with a college degree or more, the decline in employment
is similar across ethnic groups, with the exception that Blacks suffered a larger decline than
the other groups. For women in this educational group, East Asians stand out for having
experienced the largest declines, while the other groups are quite similar. The magnitudes of
the differences between ethnic groups are dwarfed, however, by the differences for the lower
end of the education distribution.
Since the largest differential effects of the pandemic by ethnicity were in the first two
quarters of the pandemic (“Cr2” and “Cr3”), we only report estimates associated with those
two quarters in the remainder of the paper.
It is clear from Table 1 that the magnitudes of the estimated effects for East Asians and
South East Asians are quite different. On the other hand, the overall patterns are similar.
Moreover, some of the estimates are noisy when we restrict estimation to the subsample
of individuals with a high school degree or less. Table 2 presents the results for the same
model as in Table 1, but with East Asians and South East Asians merged into one group,
“AsianEast”. As expected, the overall pattern is the same, with the point estimates for
“AsianEast” of the same magnitude as the point estimates for the dis-aggregated East and
not interviewed for eight, and then again interviewed for another four months.
4
For comparison, the probability of working in the quarter prior to the onset of the pandemic was 72%
and 54% for two groups.
9
Table 1: Controlling for Demographics and Time-varying Labor Market Conditions
Black*Cr2
Hispanic*Cr2
AsianE*Cr2
AsianSE*Cr2
AsianOther*Cr2
Black*Cr3
Hispanic*Cr3
AsianE*Cr3
AsianSE*Cr3
AsianOther*Cr3
Black*Cr4
Hispanic*Cr4
AsianE*Cr4
AsianSE*Cr4
AsianOther*Cr4
Black*Cr5
Hispanic*Cr5
AsianE*Cr5
AsianSE*Cr5
AsianOther*Cr5
Men
HS or less
Men
Some Coll.
Men
College+
Women
HS or less
Women
Some Coll.
Women
College+
-0.029
(0.018)
-0.043
(0.013)
-0.211
(0.060)
-0.180
(0.043)
-0.160
(0.041)
-0.051
(0.017)
-0.033
(0.013)
-0.119
(0.061)
-0.082
(0.042)
-0.050
(0.040)
-0.035
(0.016)
-0.015
(0.012)
-0.100
(0.061)
0.020
(0.036)
-0.033
(0.040)
0.013
(0.017)
0.010
(0.012)
-0.083
(0.055)
0.023
(0.034)
-0.049
(0.039)
-0.031
(0.021)
-0.042
(0.020)
-0.120
(0.066)
-0.140
(0.045)
-0.048
(0.049)
-0.015
(0.021)
-0.014
(0.018)
0.027
(0.066)
-0.019
(0.047)
-0.010
(0.047)
0.009
(0.019)
0.016
(0.018)
-0.074
(0.055)
0.009
(0.038)
-0.009
(0.046)
-0.017
(0.020)
-0.035
(0.018)
-0.070
(0.068)
0.001
(0.038)
0.016
(0.051)
-0.077
(0.019)
-0.017
(0.018)
0.006
(0.025)
-0.050
(0.034)
0.016
(0.018)
-0.049
(0.018)
-0.021
(0.017)
0.042
(0.023)
-0.028
(0.034)
0.026
(0.015)
-0.044
(0.017)
-0.029
(0.016)
0.009
(0.022)
-0.007
(0.027)
0.031
(0.017)
-0.036
(0.016)
-0.031
(0.015)
0.002
(0.023)
0.020
(0.025)
0.017
(0.015)
-0.024
(0.018)
-0.038
(0.016)
-0.155
(0.048)
-0.208
(0.040)
0.088
(0.043)
-0.068
(0.018)
-0.011
(0.015)
-0.061
(0.049)
-0.027
(0.044)
0.107
(0.046)
-0.043
(0.018)
-0.023
(0.015)
-0.070
(0.050)
0.014
(0.041)
0.018
(0.046)
-0.033
(0.018)
-0.012
(0.015)
-0.068
(0.054)
0.057
(0.040)
0.039
(0.045)
-0.061
(0.020)
-0.020
(0.020)
-0.156
(0.064)
-0.035
(0.049)
-0.070
(0.054)
-0.066
(0.019)
-0.054
(0.020)
-0.054
(0.065)
0.018
(0.046)
-0.019
(0.058)
-0.030
(0.018)
-0.036
(0.019)
-0.032
(0.063)
-0.072
(0.046)
0.082
(0.051)
-0.038
(0.018)
-0.033
(0.020)
0.020
(0.056)
0.012
(0.041)
-0.023
(0.048)
-0.008
(0.016)
-0.041
(0.019)
-0.022
(0.027)
-0.003
(0.030)
0.054
(0.024)
0.009
(0.016)
0.011
(0.017)
0.013
(0.025)
-0.017
(0.027)
-0.016
(0.024)
0.009
(0.015)
0.000
(0.017)
0.012
(0.026)
-0.089
(0.027)
0.023
(0.024)
0.002
(0.015)
-0.014
(0.017)
0.010
(0.024)
0.015
(0.027)
0.058
(0.023)
Observations
253,718
167,370
235,258
223,847
193,556
288,601
Robust standard errors are clustered at the household level. Coefficients for Age, Age2 ,
Marital Status, No. of Children, No. of Children under 5 and Main Effects not reported.
Fixed effects for each combination of state and month starting in April 2020 included.
Cr2, Cr3, Cr4, and Cr5 refer to the second quarter of 2020 through the first quarter of 2021.
10
South East Asians. Also as expected, the coefficients on “AsianEast” are more precisely
estimated than when this group is dis-aggregated. For most of the rest of the paper, we
therefore aggregate East and South East Asian Americans into one group.
Table 2: Main Results Combining East and South East Asians.
Men
HS or less
Men
Some Coll.
Men
College+
Women
HS or less
Women
Some Coll.
Women
College+
-0.192
(0.035)
-0.097
(0.036)
-0.134
(0.038)
0.000
(0.039)
-0.015
(0.021)
0.018
(0.020)
-0.188
(0.031)
-0.051
(0.034)
-0.075
(0.040)
-0.005
(0.039)
-0.015
(0.021)
-0.000
(0.019)
Observations
253,718
167,370
∗
Same specification as in Table 1.
235,258
223,847
193,556
288,601
AsianEast*Cr2
AsianEast*Cr3
2.3
Logit or Linear Probability Model?
The outcome “working” is binary. This leads to the question of whether the fixed effects
linear probability model will adequately capture its relationship with the explanatory variables. In this section, we therefore report the results from estimating a logit version of the
model in Table 2. This is less straightforward than it might seem. On one hand, estimating
a coefficient for each fixed effect could be challenging because of the incidental parameters
problem.5 On the other hand, the traditional conditional likelihood approach to eliminating the fixed effects is computationally infeasible in this case. We therefore use a modified
version of the conditional likelihood approach.
The point of departure for our approach is the familiar fixed effects logit model pioneered
by Rasch (1960, 1961). Specifically, let i index a group of observations with a particular
value of the state-month combination, and let j denote an observation within group i. We
assume that the dependent variable, y (working), for observation j in group i is independent
conditional on the explanatory variables and on a group-specific effect. We also assume that
5
See, for example, Arellano and Hahn (2007) for a discussion of how incident parameter can be a problem
when the number of observations per group is large, but small relative to the sample size.
11
the distribution of each y is
P yij =
i
1| {xij }Jj=1
, αi =
exp x0ij β + αi
1 + exp x0ij β + αi
= Λ x0ij β + αi ,
(1)
where Λ (·) the logistic cumulative distribution function and Ji denotes the number of observations in group i.
In this case, the distribution of (yi1 , . . . , yiJi ) conditional on (xi1 , . . . , xiJi ) and on
Ji
P
yij
j=1
is
P
i
i
i
{yij }Jj=1
{xij }Jj=1
= {cij }Jj=1
,
Ji
X
yij =
j=1
Ji
X
j=1
!
cij
exp
=P
P
dj ∈Bi exp
Ji
0
j=1 cij xij β
P
Ji
j=1
dj x0ij β
,
(2)
o
n
Pi
Pi
cij .
dt = Jj=1
where cit ∈ {0, 1} and Bi = (d1 , ..., dJi ) : dj ∈ {0, 1} , Jj=1
One could, in principle, estimate β by maximizing a conditional likelihood function based
on (2). This, however, would be extremely computationally intensive. For example, our
sample of women with a college degree or more contains a group (state-month combination)
with 13,627 observations of whom 9,835 reported working. This means that for this group,
the number of terms in the denominator of (2) is 13,627
, which is of the order of magnitude
9,835
of 103500 . This makes estimation based on the likelihood function in (2) infeasible. We
therefore note that for any two individuals in group i, j1 and j2 ,
P (yij1 = 1| xij1 xij2 , yij1 + yij2
exp (xij1 − xij2 )0 β
.
= 1) =
1 + exp (xij1 − xij2 )0 β
This can be used to form a likelihood function based on pairs of observations within each
group. In principle, one could use all pairs of (j1 , j2 ) in a group, but this can be computationally demanding. We therefore pair each observation for whom yij1 = 1 with a randomly
chosen observation (ij2 ) with yij2 = 0 in the same group. We also pair each observation for
whom yij1 = 0 with a randomly chosen observation with yij2 = 1 in the same group. The
12
estimator of β is then based on maximizing the pseudo log-likelihood function
exp yij1 (xij1 − xij2 )0 b
log
0
1
+
exp
(x
−
x
)
b
ij
ij
1
2
=1
Ji
n X
X
i=1 j1
over b.
Table 3 displays the estimated coefficient in a fixed effects logit version of the model in
Table 2.6 The conclusions to be drawn from the logit estimates are not substantially different
from those reported in Section 2. Consider, for example, a heterogeneous population of East
Asians in which the probability of working prior to the pandemic is uniformly distributed
between 50% and 90%7 . The estimated average effects of the pandemic on the probability
of working (relative to the effect for Whites with the same education and gender) for this
population are declines in the probability of working by 15.4, 13.5, 1.8, 18.7, 7.3, and 1.2
percentage points across the six combinations of education and gender. These are quite
similar to the results for the linear probability model in Table 2. For the rest of the paper,
we therefore use the linear probability model.
Table 3: Main Results Combining East and South East Asians (Logit).
AsianEast*Cr2
AsianEast*Cr3
Observations
2.4
Men
HS or less
Men
Some Coll.
Men
College+
Women
HS or less
Women
Some Coll.
Women
College+
-0.717
(0.154)
-0.468
(0.169)
-0.632
(0.235)
0.030
(0.179)
-0.091
(0.137)
0.139
(0.146)
-0.861
(0.139)
-0.246
(0.112)
-0.354
(0.142)
-0.053
(0.138)
-0.059
(0.109)
0.036
(0.115)
253,718
167,370
235,258
223,847
193,556
288,601
Comparison to the 2008 Recession
It is difficult to know the mechanisms behind the dramatic impact on Asians, especially East
Asians, with lower educational attainment. One possibility is that this is a typical feature of
6
The standard errors reported in Table 3 account for the group structure, but they are not clustered at
the household level. The estimation also does not employ sampling weights.
7
The marginal effects in a logit differ across observations with different explanatory variables. The overall
probability of working is roughly 70% and we therefore report an average marginal effect across individuals
with probabilities between 50% and 90%.
13
economic downturns. In order to investigate this possibility, we estimate the same model as
in Table 1 using data from 2006 through 2011, with the crisis variable defined as a dummy
variable for the “Crisis” years 2008, 2009, 2010 and 2011. The results are presented in Table
4.8 They suggest that there is generally no differential in the impact of the recession between
Asians and Whites. This is in sharp contrast to the COVID-19 crisis. The maximum Tstatistic of the 18 coefficients that measure the differential Asian-versus-White impact of the
crisis is around 2.5, and a joint test of significance across all 6 combinations of education and
gender yields a chi-square test statistic of around 14. Compared to a chi-square distribution
with 18 degrees of freedom, this is statistically very insignificant.
Table 4: 2008 Recession
Black
Hispanic
AsianE
AsianSE
AsianOther
Black*Crisis
Hispanic*Crisis
AsianE*Crisis
AsianSE*Crisis
AsianOther*Crisis
Observations
Men
HS or less
Men
Some Coll.
Men
College+
Women
HS or less
Women
Some Coll.
Women
College+
-0.110
(0.006)
0.037
(0.004)
0.013
(0.016)
-0.005
(0.014)
-0.000
(0.016)
-0.005
(0.007)
-0.012
(0.005)
0.038
(0.019)
0.001
(0.017)
-0.008
(0.020)
-0.076
(0.006)
-0.007
(0.006)
-0.040
(0.020)
-0.038
(0.015)
0.011
(0.014)
-0.019
(0.008)
0.001
(0.007)
-0.010
(0.025)
0.017
(0.018)
-0.006
(0.018)
-0.039
(0.006)
-0.009
(0.006)
-0.033
(0.008)
-0.040
(0.011)
-0.005
(0.007)
-0.027
(0.008)
-0.009
(0.007)
-0.013
(0.011)
0.003
(0.013)
-0.011
(0.009)
-0.039
(0.006)
-0.070
(0.006)
0.001
(0.016)
0.001
(0.015)
-0.108
(0.018)
-0.019
(0.007)
-0.006
(0.007)
0.003
(0.019)
0.007
(0.018)
-0.000
(0.022)
-0.003
(0.006)
0.001
(0.007)
-0.077
(0.021)
0.035
(0.016)
-0.021
(0.021)
-0.016
(0.007)
-0.014
(0.009)
0.001
(0.025)
0.017
(0.020)
-0.032
(0.026)
0.039
(0.006)
-0.017
(0.008)
-0.078
(0.011)
0.046
(0.011)
-0.119
(0.013)
-0.036
(0.007)
-0.006
(0.010)
-0.011
(0.014)
-0.018
(0.014)
0.004
(0.015)
1,042,906
641,817
781,625
1,013,730
773,306
862,519
8
Here, we allow for different effects for East Asians and South East Asians since the sample sizes are
much larger than what we consider for the COVID-19 pandemic.
14
3
Transition In and Out of Employment
The analysis in Section 2 focused on the probability of working. We now turn to the probability of working conditional on whether the individual worked in the previous month. There
are two distinct motivations for this. The first is that it is economically interesting to know
whether the large decline in the probability of working for Asian Americans is driven by
the probability that the employed stopped working or by the probability that those not employed started working. The second motivation is that it is interesting to investigate how
the empirical results in Section 2 change if one controls for industry and/or occupation.
The Current Population Survey does include information on the industry and occupation to
which an individual belongs. In principle, these variables should be well-defined whether or
not an individual is currently working as long as they have worked in the past. As such,
it should only be missing for individuals who have no attachment to the labor market, and
one thus might be able to justify ignoring those individuals from the empirical analysis.
However, if – contrary to protocol – the non-response to the questions about industry or
occupation is a direct result of not working, then ignoring those individuals will result in
selection bias. Indeed, we find strong evidence that non-response to the questions about
industry or occupation is caused by not working. For example, consider individuals who
worked in the previous month, and hence should in principle report industry and occupation
status in the current month regardless of their current employment status. In this group,
30% of those not currently working report occupation as missing. The corresponding number
for industry is 28%. One solution to this problem is to use lagged industry or occupation
as an explanatory variable, but whether this variable is missing might be the consequence
of lagged unemployment, which in turn could be highly correlated with current employment
status. Again, this could lead to endogeneity in whether or not lagged industry or occupation
is missing. In contrast, the transitions in and out of employment are much more likely to be
driven by recent events which are less likely to be related to whether or not lagged industry
or occupation is missing. This is our second motivation for now focusing on transitions.
The basic findings regarding month-to-month transitions out of employment are presented in Figures 4 and 5. Figure 4 displays the fraction of individuals working in the
15
previous month, who are also working in the current month for Whites, Blacks, Hispanics
and Asians (combined across all sub-groups). The figure shows a large drop in the probability of remaining employed in April of 2020 for all groups. The decline is larger for those
without a college degree and especially large for Asian-Americans.
The 2019 part of the figures illustrate that seasonality might play a role. In Figure
5, we therefore plot the probability of remaining employed in a given month relative to the
probability in the same month in 2019. The same pattern emerges: Asian-Americans without
a college degree experienced the biggest job loss.
Figure 4: Fraction Remaining at Work by Group
The corresponding figures for the monthly probability of being employed condition on
not being employed in previous month are based on much smaller sample sizes and therefore
too noisy to be informative, and we therefore turn to regressions. We first estimate the
16
Figure 5: Fraction Remaining at Work by Group Relative to Same Month in 2019
econometric model for working from Section 2 separately for the samples of individuals who
were or were not working in the previous month. The results are presented in Tables 5 and
6.
Table 5 confirms the results in Figures 4 and 5. After controlling for demographics,
Asian-Americans without a college degree suffered larger job loss in the second quarter of
2020 than other groups. The estimated effects on the probability of entering employment
reported in Table 6 tend to be statistically insignificant.
We now turn to the fact that some industries and occupations were particularly hard hit
by the pandemic. This could lead to omitted variable bias if Asian Americans tend to work
in those industries or occupations. For example, in the sample used below, East Asians are
overrepresented in the occupation “hosts and hostesses, restaurant, lounge, and coffee shop”
17
Table 5: Remaining Employed
Black*Cr2
Hispanic*Cr2
AsianEast*Cr2
AsianOther*Cr2
Black*Cr3
Hispanic*Cr3
AsianEast*Cr3
AsianOther*Cr3
Observations
Men
HS or less
Men
Some Coll.
Men
College+
Women
HS or less
Women
Some Coll.
Women
College+
-0.054
(0.016)
-0.026
(0.011)
-0.191
(0.040)
-0.070
(0.040)
-0.011
(0.013)
-0.004
(0.009)
-0.006
(0.021)
0.019
(0.027)
-0.030
(0.016)
-0.021
(0.014)
-0.063
(0.037)
-0.088
(0.041)
0.004
(0.014)
0.005
(0.011)
-0.028
(0.028)
-0.003
(0.031)
-0.054
(0.015)
-0.012
(0.011)
-0.018
(0.014)
-0.003
(0.010)
-0.005
(0.012)
-0.003
(0.011)
-0.006
(0.011)
0.003
(0.010)
-0.022
(0.019)
-0.009
(0.016)
-0.182
(0.041)
-0.021
(0.044)
-0.033
(0.016)
0.008
(0.013)
-0.054
(0.029)
0.078
(0.052)
-0.016
(0.016)
-0.042
(0.016)
-0.074
(0.039)
-0.119
(0.050)
-0.049
(0.015)
0.003
(0.014)
0.001
(0.024)
0.093
(0.030)
-0.013
(0.013)
-0.028
(0.015)
-0.021
(0.014)
0.020
(0.015)
-0.010
(0.011)
-0.019
(0.014)
-0.006
(0.012)
0.005
(0.017)
121,000
89,044
139,353
79,880
86,877
148,575
Table 6: Transition Into Employment
Black*Cr2
Hispanic*Cr2
AsianEast*Cr2
AsianOther*Cr2
Black*Cr3
Hispanic*Cr3
AsianEast*Cr3
AsianOther*Cr3
Observations
Men
HS or less
Men
Some Coll.
Men
College+
Women
HS or less
Women
Some Coll.
Women
College+
-0.006
(0.018)
0.005
(0.023)
-0.037
(0.043)
-0.248
(0.069)
0.024
(0.019)
-0.009
(0.024)
0.040
(0.050)
-0.116
(0.072)
-0.010
(0.034)
-0.058
(0.038)
-0.039
(0.057)
0.005
(0.069)
-0.040
(0.033)
-0.002
(0.042)
0.066
(0.070)
0.006
(0.068)
0.058
(0.049)
0.028
(0.053)
-0.000
(0.053)
0.061
(0.053)
-0.024
(0.045)
0.035
(0.052)
0.014
(0.050)
0.061
(0.065)
-0.013
(0.015)
-0.021
(0.012)
-0.007
(0.027)
0.005
(0.030)
-0.002
(0.016)
-0.003
(0.014)
0.056
(0.032)
0.053
(0.039)
-0.052
(0.020)
-0.027
(0.023)
-0.071
(0.039)
0.060
(0.048)
0.024
(0.022)
-0.017
(0.025)
0.042
(0.043)
-0.224
(0.061)
-0.041
(0.031)
0.011
(0.027)
0.014
(0.027)
0.011
(0.025)
0.033
(0.030)
0.026
(0.027)
0.003
(0.029)
0.016
(0.027)
52,476
26,976
24,118
72,926
46,894
51,916
by a factor of 3 relative to the rest of the population. In the group with a high school degree
or less, the difference becomes a factor of almost 5. In order to control for the potential
resulting omitted variable bias in the estimate for being Asian-American, we include a fixed
18
effects for interactions between occupation (including “missing”) and pandemic months. The
results are presented in Tables 7 and 8.9
Table 7: Remaining Employed Controlling for Time-varying Effects of Occupation
Black*Cr2
Hispanic*Cr2
AsianEast*Cr2
AsianOther*Cr2
Black*Cr3
Hispanic*Cr3
AsianEast*Cr3
AsianOther*Cr3
Observations
Men
HS or less
Men
Some Coll.
Men
College+
Women
HS or less
Women
Some Coll.
Women
College+
-0.037
(0.017)
-0.015
(0.011)
-0.120
(0.041)
-0.060
(0.040)
-0.010
(0.014)
-0.003
(0.009)
-0.011
(0.021)
0.014
(0.029)
-0.003
(0.015)
-0.020
(0.014)
-0.067
(0.037)
-0.097
(0.041)
-0.004
(0.015)
0.004
(0.011)
-0.021
(0.028)
-0.008
(0.033)
-0.043
(0.014)
-0.010
(0.011)
-0.012
(0.014)
-0.013
(0.010)
-0.006
(0.013)
-0.006
(0.011)
-0.010
(0.011)
0.000
(0.010)
-0.037
(0.019)
0.014
(0.017)
-0.081
(0.039)
-0.025
(0.045)
-0.036
(0.017)
0.011
(0.015)
-0.050
(0.029)
0.065
(0.051)
-0.010
(0.015)
-0.026
(0.016)
-0.046
(0.036)
-0.083
(0.045)
-0.049
(0.015)
0.003
(0.014)
-0.010
(0.024)
0.100
(0.032)
-0.013
(0.012)
-0.018
(0.014)
-0.020
(0.014)
0.018
(0.015)
-0.016
(0.012)
-0.015
(0.014)
-0.018
(0.012)
-0.003
(0.018)
121,000
89,044
139,353
79,880
86,877
148,575
A comparison of the results in Tables 5 and 7 reveals that an individual’s occupation can
explain up to approximately half of the difference between Whites and the two groups of
Asian Americans. For example, for East and South-East Asian American men with a high
school degree or less, the estimated effect on the probability of remaining employed (relative
to comparable Whites) is reduced from −0.19 to −0.12 when one controls for time-varying
occupation effects. The corresponding reduction for women is from −0.18 to −0.08.
The same comparison for the probability of entering employment (Tables 6 and 8) suggests that controlling for occupation makes only a small difference relative to the estimation
uncertainty.
9
We repeated the same exercise using industry rather than occupation. The results are very similar,
although it appears that controlling for occupation makes the bigger difference. We refrain from using all
combinations of industry, occupation, and month because this would lead to many “cells” with only one
observation, and the fixed effects approach would effectively drop these observations from the sample.
19
Table 8: Transition Into Employment Controlling for Time-varying Effects of Occupation
Black*Cr2
Hispanic*Cr2
AsianEast*Cr2
AsianOther*Cr2
Black*Cr3
Hispanic*Cr3
AsianEast*Cr3
AsianOther*Cr3
Observations
4
Men
HS or less
Men
Some Coll.
Men
College+
Women
HS or less
Women
Some Coll.
Women
College+
0.010
(0.017)
-0.022
(0.022)
-0.065
(0.044)
-0.226
(0.064)
0.013
(0.018)
-0.020
(0.023)
0.029
(0.045)
-0.142
(0.066)
0.012
(0.030)
-0.042
(0.034)
0.003
(0.053)
0.059
(0.079)
-0.033
(0.032)
0.009
(0.042)
0.068
(0.068)
0.042
(0.068)
0.028
(0.045)
0.023
(0.050)
0.033
(0.051)
0.045
(0.048)
-0.020
(0.041)
0.085
(0.049)
0.004
(0.048)
0.077
(0.064)
-0.006
(0.014)
-0.026
(0.011)
-0.045
(0.023)
-0.010
(0.029)
-0.000
(0.014)
-0.016
(0.012)
-0.011
(0.028)
0.040
(0.034)
-0.037
(0.018)
0.008
(0.020)
-0.068
(0.037)
0.021
(0.040)
-0.000
(0.020)
-0.016
(0.021)
0.017
(0.042)
-0.118
(0.057)
-0.027
(0.027)
0.005
(0.023)
-0.011
(0.024)
0.013
(0.022)
0.036
(0.026)
-0.002
(0.025)
-0.001
(0.026)
-0.016
(0.024)
52,476
26,976
24,118
72,926
46,894
51,916
Other Possible Explanations
4.1
Anti-Asian Sentiment
It is possible that the experience of Asian American workers during the pandemic reflects the
impact of increased discrimination due to their perceived association with China. Such an
explanation would be consistent with Kaushal, Kaestner, and Reimers (2007). That paper
finds that earnings of Arab and Muslim men declined dramatically after the September 11th
terrors’ attack. It would also be consistent with the finding in Sakong (2018) that economic
downturns are often associated with increased racial prejudice.
We take two approaches to investigating this. The first approach is to combine the CPS
data with a state specific measure of anti-Asians bias, and see whether Asian Americans in
states with larger bias were harder hit by the pandemic than those in states with less bias.
The second approach is to investigate whether Asian American working in jobs that involve
more intensive interpersonal interactions are harder hit than those in other occupations.
We construct a measure of racial bias using data from the Project Implicit Database.10
10
https://implicit.harvard.edu/implicit/
20
Specifically, we use data from respondents of the “Asian Implicit Association Test” from
2004-2020 and construct a variable “implicit bias” as the average IAT score by month and
state (see, e.g. Darling-Hammond, Michaels, Allen, Chae, Thomas, Nguyen, Mujahid, and
Johnson (2020)). We then augment the specifications in Tables 7 and 8 by adding interactions
between this bias variable and the pandemic-Asian interactions. In order to avoid potential
reverse causality, we use the average values of the implicit bias in a state in 2018. When
doing this, we find no evidence that Asian-Americans in states that had stronger anti-Asian
bias in 2018 saw an especially large decline in employment during the pandemic.
Our second approach for detecting Anti-Asian sentiment is to look at whether AsianAmericans are especially negatively impacted in occupations that involve intensive interpersonal interactions. To do this, we combine the CPS data with the mapping from occupation
to tasks developed by Aaronson and Phelan (2020).11 Specifically, we use whether the
variable “interpersonal task share” exceeds 0.5 (corresponding to the 75th percentile). This
includes, for example, the occupations “sales representatives, wholesale and manufacturing”,
“bartender”, “host and hostesses, restaurant, lounge and coffee shops”, and “real estate brokers and sales agents”. On the other hand, the lower tail of “interpersonal task share”
includes, for example, the occupations “industrial truck and tractor operators”, “welding,
soldering and brazing workers” and “electrical, electronic, electromechanical assemblers”.
The variable, interpersonal task share (“IP” in the table), is interacted with the indicator
variables for being for being Asian American, as well as their interactions with the pandemic
quarters.12 These are used as additional explanatory variables in the regression in Table 7.
The estimates are reported in Table 9. Only one of the estimated coefficients in Table 9 is
negative and statistically significant at the 5 percent level. With 24 estimated parameters,
this is not very different from what one might expect if all the parameter values are truly 0.
11
Aaronson and Phelan (2020) in turn builds on Acemoglu and Autor (2011), who developed the measures
from the O*NET database.
12
Since the other ethnicity groups did not experience as dramatic a drop in employment at the onset of
the pandemic, we did not attempt to decompose their decline into whether or not they were in an occupation
with high level of interpersonal interactions.
21
Table 9: Job Characteristics
Men
HS or less
Men
Some Coll.
Men
College+
Women
HS or less
Women
Some Coll.
Women
College+
-0.018
(0.085)
-0.050
(0.068)
-0.057
(0.078)
-0.097
(0.077)
-0.130
(0.067)
-0.008
(0.077)
-0.089
(0.063)
-0.056
(0.070)
-0.038
(0.021)
0.017
(0.017)
0.026
(0.017)
0.029
(0.014)
0.026
(0.075)
-0.013
(0.071)
-0.060
(0.061)
0.053
(0.064)
0.094
(0.057)
-0.123
(0.084)
0.027
(0.036)
-0.029
(0.031)
0.013
(0.021)
-0.047
(0.019)
-0.019
(0.017)
-0.014
(0.027)
AsianEast*IP*Cr2
AsianOther*IP*Cr2
AsianEast*IP*Cr3
AsianOther*IP*Cr3
4.2
Multi-generational Households
Another potential explanation for the differential downturn in employment across ethnicities
is that some groups are more likely to live in multi-generational households. If people are
concerned about the health of the older members and stop working during the pandemic as
a result of this, then this might induce different patterns in employment across ethnicities.
The results in Table 10 suggest that this is not the explanation for our findings. When we
restrict the sample to individuals in households where there is no member older than 65
years, the estimates are very close to those in Table 7.
4.3
Country of Birth
One distinct feature of Asian Americans is that they are more likely to be foreign born than
the other ethnicities. For example, 18% of the total sample is foreign born, while the fraction
for Asian Americans is approximately 2/3. This suggests that the labor market differences
between Asian Americans and Whites could be associated with language obstacles, other
cultural differences associated with being foreign born, or discrimination against foreigners.
In order to test this, we re-estimate the main model in Table 7 separately for the sample of
individuals who are born in the United States and for sample of those who are not.
The results for the US born are presented in Table 11. It is striking that that all of the
estimates related to Asian Americans interacted with the second quarter of 2020 are now
statistically insignificant.
Table 12 presents the results for non-U.S. born individuals. The estimation uncertainty
22
Table 10: Estimation Excluding Individuals With Household Members Older Than 65
Black*Cr2
Hispanic*Cr2
AsianEast*Cr2
AsianOther*Cr2
Black*Cr3
Hispanic*Cr3
AsianEast*Cr3
AsianOther*Cr3
Observations
Men
HS or less
Men
Some Coll.
Men
College+
Women
HS or less
Women
Some Coll.
Women
College+
-0.043
(0.017)
-0.012
(0.012)
-0.104
(0.044)
-0.043
(0.045)
-0.009
(0.014)
-0.002
(0.010)
-0.010
(0.023)
0.013
(0.030)
0.002
(0.015)
-0.015
(0.015)
-0.065
(0.037)
-0.087
(0.045)
-0.001
(0.016)
0.006
(0.012)
-0.026
(0.031)
0.030
(0.028)
-0.045
(0.015)
-0.013
(0.012)
-0.011
(0.014)
-0.012
(0.010)
-0.011
(0.013)
-0.009
(0.011)
-0.017
(0.012)
0.004
(0.010)
-0.045
(0.020)
0.007
(0.018)
-0.077
(0.042)
-0.031
(0.054)
-0.039
(0.018)
0.014
(0.015)
-0.061
(0.034)
0.040
(0.059)
-0.008
(0.016)
-0.019
(0.016)
-0.049
(0.039)
-0.110
(0.055)
-0.047
(0.016)
0.009
(0.015)
0.002
(0.028)
0.071
(0.031)
-0.013
(0.013)
-0.016
(0.015)
-0.019
(0.014)
0.021
(0.015)
-0.018
(0.012)
-0.010
(0.014)
-0.023
(0.013)
-0.001
(0.019)
110,295
82,332
131,911
70,353
78,161
136,530
Table 11: Estimation for U.S. Born
Black*Cr2
Hispanic*Cr2
AsianEast*Cr2
AsianOther*Cr2
Black*Cr3
Hispanic*Cr3
AsianEast*Cr3
AsianOther*Cr3
Observations
Men
HS or less
Men
Some Coll.
Men
College+
Women
HS or less
Women
Some Coll.
Women
College+
-0.025
(0.018)
0.012
(0.017)
-0.033
(0.086)
-0.042
(0.065)
-0.017
(0.015)
-0.013
(0.016)
0.037
(0.042)
-0.038
(0.062)
0.002
(0.016)
-0.009
(0.018)
0.017
(0.061)
-0.038
(0.050)
0.005
(0.016)
0.000
(0.014)
0.063
(0.042)
-0.076
(0.059)
-0.038
(0.017)
0.016
(0.014)
0.019
(0.023)
-0.018
(0.022)
0.000
(0.015)
-0.019
(0.014)
0.005
(0.019)
0.013
(0.015)
-0.044
(0.021)
0.013
(0.023)
0.124
(0.099)
-0.120
(0.103)
-0.030
(0.019)
-0.002
(0.021)
-0.105
(0.078)
0.091
(0.097)
-0.010
(0.016)
-0.030
(0.019)
0.027
(0.063)
-0.081
(0.062)
-0.054
(0.017)
0.002
(0.017)
-0.089
(0.063)
0.102
(0.043)
-0.005
(0.013)
-0.001
(0.017)
-0.024
(0.026)
0.015
(0.024)
-0.018
(0.012)
-0.023
(0.018)
-0.003
(0.021)
-0.002
(0.033)
93,461
79,539
115,339
62,046
77,928
127,059
is much greater in this sample, but – as one might expect – the point estimates of the impact
of the pandemic on Asian Americans are larger for this subgroup than for the full sample.
We conclude that the group that stands out is foreign born, low educated Asian Ameri23
cans.
Table 12: Estimation for Non-U.S. Born
Black Cr2
Hispanic Cr2
AsianEast Cr2
AsianOther Cr2
Black Cr3
Hispanic Cr3
AsianEast Cr3
AsianOther Cr3
Observations
5
Men
HS or less
Men
Some Coll.
Men
College+
Women
HS or less
Women
Some Coll.
Women
College+
-0.077
(0.050)
-0.036
(0.029)
-0.139
(0.058)
-0.050
(0.059)
0.024
(0.038)
0.034
(0.024)
0.028
(0.035)
0.059
(0.044)
-0.043
(0.076)
0.057
(0.060)
-0.036
(0.079)
-0.267
(0.102)
-0.038
(0.057)
-0.003
(0.033)
-0.057
(0.053)
0.024
(0.044)
-0.058
(0.026)
-0.055
(0.025)
-0.020
(0.022)
-0.014
(0.016)
-0.012
(0.028)
0.010
(0.023)
-0.020
(0.019)
-0.016
(0.019)
-0.023
(0.063)
0.067
(0.052)
-0.109
(0.074)
-0.004
(0.081)
-0.020
(0.049)
0.034
(0.042)
0.009
(0.053)
0.115
(0.070)
-0.004
(0.079)
0.043
(0.060)
-0.012
(0.072)
-0.093
(0.107)
0.021
(0.056)
0.061
(0.052)
0.074
(0.059)
0.152
(0.077)
-0.004
(0.040)
-0.040
(0.034)
0.008
(0.026)
0.041
(0.029)
0.003
(0.038)
-0.002
(0.029)
-0.020
(0.022)
-0.001
(0.027)
27,539
9,505
24,014
17,834
8,949
21,516
The Role of Education
The analysis so far has been done separately for different education groups. One could argue
that education is a choice made by an individual, and that this would make it endogenous.
Table 13 shows the distribution of education by ethnicity for both genders in our sample.
It is very clear that Asian Americans have higher education on average than other groups.
In other words, the selection into education level potentially differs across the ethnicities.
For example, the group of Asians Americans with a high school degree or less might be very
different in terms of unobservables from other groups with the same level of education.
To investigate whether the results for ethnicity are biased by selection into different education groups, we estimate a model for the probability of working with the same explanatory
variables as in Section 2, but now using the whole sample without conditioning on education groups. The first two columns of Table 14 show that Blacks, Hispanics and Asians
all experienced a bigger impact of the crisis on their employment than Whites. The effects
24
Table 13: Distribution of Education (Aged 25-65)
Whites
Blacks
Hispanic
Asians
Overall
Men
HS or less
0.334
0.458
0.601
0.224
0.390
Men
Some Col.
0.262
0.285
0.214
0.160
0.251
Men
College+
0.403
0.257
0.185
0.616
0.359
Women
HS or less
0.264
0.369
0.533
0.232
0.321
Women
Some Col.
0.277
0.315
0.237
0.158
0.268
Women
College+
0.459
0.316
0.230
0.610
0.410
are especially strong for East and South-East Asians who had the largest initial drops in
employment for both men and women. This is consistent with the findings in Section 2.
Table 14: Results For the Whole Sample (Not Conditional on Education)
Men
Black Cr2
Hispanic Cr2
AsianEast Cr2
AsianOther Cr2
Black Cr3
Hispanic Cr3
AsianEast Cr3
AsianOther Cr3
Observations
6
All
Women
Born In the US
Men
Women
Born Outside the US
Men
Women
-0.054
(0.011)
-0.051
(0.009)
-0.061
(0.017)
-0.015
(0.016)
-0.046
(0.011)
-0.029
(0.009)
-0.005
(0.016)
0.009
(0.015)
-0.036
(0.011)
-0.042
(0.010)
-0.060
(0.016)
0.054
(0.020)
-0.046
(0.010)
-0.019
(0.010)
-0.009
(0.016)
0.010
(0.020)
-0.042
(0.013)
-0.026
(0.013)
0.029
(0.031)
-0.037
(0.028)
-0.049
(0.012)
-0.044
(0.013)
-0.013
(0.031)
0.014
(0.029)
-0.033
(0.011)
-0.019
(0.014)
0.027
(0.033)
0.026
(0.034)
-0.044
(0.011)
-0.011
(0.014)
0.083
(0.030)
-0.025
(0.033)
-0.075
(0.028)
-0.047
(0.019)
-0.075
(0.025)
0.018
(0.024)
-0.045
(0.027)
-0.027
(0.017)
-0.010
(0.023)
-0.001
(0.022)
-0.039
(0.030)
-0.083
(0.021)
-0.104
(0.025)
0.044
(0.029)
-0.072
(0.030)
-0.067
(0.021)
-0.081
(0.024)
-0.018
(0.029)
656,346
706,004
546,034
584,391
110,312
121,613
Conclusion
This paper has documented that Asian Americans with no college education were especially
hard hit by the onset of the pandemic. The negative effect on Asian Americans with no
college education remains after controlling for differences in demographics, local labor market
conditions, and job characteristics. The extra burden is primarily borne by individuals who
were born outside the United States. The results add to the growing evidence that the
25
pandemic has had very different effects across different ethnicities. Here, we have studied
employment. Whether the results generalize to other economic outcomes is an interesting
topic for future research.
The paper illustrates the importance of treating Asian Americans as a distinct minority. Highly educated Asian Americans are similar to Whites in terms of labor market outcomes, but lower educated Asian Americans are more similar to other disadvantaged minority groups. This also highlights the vast heterogeneity within Asian Americans. Since
Asian Americans is the fastest growing ethnic group in the United States, we expect these
considerations to be even more important in future research.
References
Aaronson, D., and B. J. Phelan (2020): “The Evolution of Technological Substitution
in Low-Wage Labor Markets,” Federal Reserve Bank of Chicago Working Paper, No.
2020-16, July.
Acemoglu, D., and D. Autor (2011): “Skills, Tasks and Technologies: Implications for
Employment and Earnings,” vol. 4 of Handbook of Labor Economics, chap. 12, pp. 1043
– 1171. Elsevier.
Arellano, M., and J. Hahn (2007): “Understanding Bias in Nonlinear Panel Models:
Some Recent Developments,” in Advances in Economics and Econometrics, Ninth World
Congress. Cambridge University Press.
Bartik, A. W., M. Bertrand, F. Lin, J. Rothstein, and M. Unrath (2020): “Measuring the Labor Market at the Onset of the COVID-19 Crisis,” Brookings Papers on
Economic Activity, pp. 239–268.
Cortes, G., and E. Forsythe (2020): “The Heterogeneous Labor Market Impacts of the
Covid-19 Pandemic,” .
Dam, D., M. Gaur, F. Karahan, L. Pilossoph, and W. Schirmer (2021): “Black
26
and White Differences in the Labor Market Recovery from COVID-19,” Federal Reserve
Bank of New York Liberty Street Economics.
Darling-Hammond, S., E. K. Michaels, A. M. Allen, D. H. Chae, M. D. Thomas,
T. T. Nguyen, M. M. Mujahid, and R. C. Johnson (2020): “After “The China
Virus” Went Viral: Racially Charged Coronavirus Coverage and Trends in Bias Against
Asian Americans,” Health Education & Behavior, 47(6), 870–879, PMID: 32911985.
Kaushal, N., R. Kaestner, and C. Reimers (2007): “Labor Market Effects of September 11th on Arab and Muslim Residents of the United States,” The Journal of Human
Resources, 42(2), 275–308.
Mar, D., and P. Ong (2020): “COVID-19’s Employment Disruptions to Asian Americans,” Discussion paper, UCLA.
Marcello, R. K., J. Dolle, A. Tariq, S. Kaur, L. Wong, J. Curcio, R. Thachil,
S. S. Yi, and N. Islam (2020): “Disaggregating Asian Race Reveals COVID-19 Disparities among Asian Americans at New York City’s Public Hospital System,” medRxiv.
Rasch, G. (1960): Probabilistic Models for Some Intelligence and Attainment Tests. Denmarks Pædagogiske Institut, Copenhagen.
(1961): “On General Laws and the Meaning of Measurement in Psychology,”
in Proceedings of the Fourth Berkeley Symposium on Mathematical Statistics and
Probability, Volume 4: Contributions to Biology and Problems of Medicine, pp. 321–333,
Berkeley, Calif. University of California Press.
Sakong, J. (2018): “Racial Prejudice is Not Normal: A Collage of Empirical Evidence,”
https://www.jungsakong.com/research.
US Bureau of Labor Statistics (2020): “Frequently asked questions: The impact of the
coronavirus (COVID-19) pandemic on the Employment Situation for April 2020,” https:
//www.bls.gov/covid19/employment-situation-covid19-faq-april-2020.htm, Accessed: 2020-09-18.
27
Wu, N. (2021): “Senate passes anti-Asian American hate crime bill,” Discussion paper,
Politico.
28
Appendix Tables
Appendix Table 1: Fraction of Men Aged 25 to 65 Working by Group in 2020-21
Period
Jan-Mar, 2020
Apr-Jun
Jul-Sep
Oct-Dec
Jan-Mar, 2021
HS or less
0.708
0.621
0.681
0.684
0.657
Some Col.
0.789
0.699
0.740
0.742
0.748
College+
0.873
0.819
0.821
0.857
0.861
Overall
0.795
0.723
0.754
0.769
0.764
Blacks
Jan-Mar, 2020
Apr-Jun
Jul-Sep
Oct-Dec
Jan-Mar, 2021
0.578
0.482
0.543
0.558
0.545
0.735
0.636
0.666
0.686
0.676
0.858
0.731
0.757
0.774
0.804
0.697
0.589
0.631
0.651
0.651
Hispanics
Jan-Mar, 2020
Apr-Jun
Jul-Sep
Oct-Dec
Jan-Mar, 2021
0.800
0.679
0.736
0.764
0.752
0.828
0.692
0.766
0.794
0.746
0.893
0.782
0.796
0.824
0.850
0.823
0.702
0.755
0.781
0.769
Jan-Mar, 2020
Apr-Jun
Jul-Sep
Oct-Dec
Jan-Mar, 2021
0.768
0.459
0.631
0.684
0.681
0.763
0.559
0.712
0.754
0.722
0.871
0.799
0.838
0.865
0.864
0.832
0.692
0.774
0.805
0.802
Jan-Mar, 2020
0.715
0.786
Apr-Jun
0.608
0.682
0.671
0.732
Jul-Sep
Oct-Dec
0.686
0.742
Jan-Mar, 2021
0.666
0.736
∗
Statistics are calculated using population weights.
0.873
0.805
0.815
0.847
0.855
0.790
0.700
0.740
0.758
0.753
Whites
Asians
All
29
‘
Appendix Table 2: Fraction of Women Aged 25 to 65 Working by Group in 2020-21
Period
Jan-Mar, 2020
Apr-Jun
Jul-Sep
Oct-Dec
Jan-Mar, 2021
HS or less
0.558
0.445
0.502
0.527
0.519
Some Col.
0.673
0.571
0.611
0.649
0.652
College+
0.779
0.687
0.689
0.757
0.762
Overall
0.690
0.593
0.620
0.667
0.670
Blacks
Jan-Mar, 2020
Apr-Jun
Jul-Sep
Oct-Dec
Jan-Mar, 2021
0.558
0.422
0.450
0.495
0.486
0.679
0.540
0.588
0.635
0.628
0.784
0.715
0.736
0.771
0.770
0.669
0.554
0.585
0.627
0.625
Hispanics
Jan-Mar, 2020
Apr-Jun
Jul-Sep
Oct-Dec
Jan-Mar, 2021
0.526
0.387
0.465
0.488
0.472
0.698
0.563
0.587
0.630
0.629
0.765
0.626
0.668
0.713
0.724
0.619
0.488
0.547
0.574
0.565
Asians
Jan-Mar, 2020
Apr-Jun
Jul-Sep
Oct-Dec
Jan-Mar, 2021
0.556
0.324
0.459
0.505
0.521
0.647
0.448
0.570
0.607
0.604
0.671
0.631
0.635
0.657
0.687
0.641
0.537
0.587
0.613
0.635
Jan-Mar, 2020
0.549
0.676
Apr-Jun
0.419
0.559
0.481
0.601
Jul-Sep
Oct-Dec
0.510
0.642
Jan-Mar, 2021
0.501
0.642
∗
Statistics are calculated using population weights.
0.765
0.678
0.686
0.743
0.751
0.672
0.566
0.600
0.642
0.644
Whites
All
30
Appendix Table 3: Fraction of Men Aged 25 to 65 Working by Group in 2020-21 (Alt. Def.
of Work)
Period
Jan-Mar, 2020
Apr-Jun
Jul-Sep
Oct-Dec
Jan-Mar, 2021
HS or less
0.729
0.662
0.709
0.707
0.684
Some Col.
0.812
0.740
0.772
0.768
0.775
College+
0.892
0.847
0.857
0.875
0.877
Overall
0.816
0.759
0.787
0.791
0.786
Blacks
Jan-Mar, 2020
Apr-Jun
Jul-Sep
Oct-Dec
Jan-Mar, 2021
0.596
0.526
0.567
0.574
0.567
0.749
0.678
0.701
0.711
0.704
0.879
0.774
0.783
0.801
0.824
0.715
0.632
0.659
0.672
0.673
Hispanics
Jan-Mar, 2020
Apr-Jun
Jul-Sep
Oct-Dec
Jan-Mar, 2021
0.824
0.720
0.757
0.782
0.784
0.855
0.737
0.804
0.822
0.775
0.909
0.816
0.825
0.842
0.873
0.846
0.743
0.781
0.801
0.799
Asians
Jan-Mar, 2020
Apr-Jun
Jul-Sep
Oct-Dec
Jan-Mar, 2021
0.784
0.561
0.650
0.701
0.706
0.788
0.625
0.741
0.776
0.762
0.895
0.840
0.856
0.881
0.879
0.854
0.750
0.794
0.822
0.823
Jan-Mar, 2020
0.736
0.808
Apr-Jun
0.652
0.725
Jul-Sep
0.697
0.765
Oct-Dec
0.707
0.768
Jan-Mar, 2021
0.693
0.764
∗
Statistics are calculated using population weights.
0.892
0.837
0.847
0.866
0.872
0.811
0.739
0.770
0.780
0.776
Whites
All
31
Appendix Table 4: Fraction of Women Aged 25 to 65 Working by Group in 2020-21 (Alt.
Def. of Work)
Period
Jan-Mar, 2020
Apr-Jun
Jul-Sep
Oct-Dec
Jan-Mar, 2021
HS or less
0.576
0.487
0.535
0.550
0.539
Some Col.
0.695
0.619
0.649
0.677
0.678
College+
0.808
0.735
0.751
0.782
0.785
Overall
0.715
0.640
0.667
0.692
0.692
Blacks
Jan-Mar, 2020
Apr-Jun
Jul-Sep
Oct-Dec
Jan-Mar, 2021
0.582
0.463
0.484
0.523
0.506
0.707
0.599
0.640
0.661
0.663
0.806
0.762
0.787
0.798
0.795
0.693
0.602
0.630
0.655
0.651
Hispanics
Jan-Mar, 2020
Apr-Jun
Jul-Sep
Oct-Dec
Jan-Mar, 2021
0.543
0.424
0.490
0.507
0.493
0.720
0.616
0.623
0.666
0.658
0.792
0.687
0.727
0.739
0.746
0.639
0.534
0.584
0.598
0.589
Asians
Jan-Mar, 2020
Apr-Jun
Jul-Sep
Oct-Dec
Jan-Mar, 2021
0.575
0.405
0.482
0.524
0.538
0.676
0.526
0.599
0.627
0.627
0.704
0.667
0.669
0.675
0.704
0.670
0.589
0.618
0.632
0.652
Jan-Mar, 2020
0.568
0.700
Apr-Jun
0.461
0.611
Jul-Sep
0.511
0.641
Oct-Dec
0.532
0.670
Jan-Mar, 2021
0.521
0.670
∗
Statistics are calculated using population weights.
0.794
0.726
0.743
0.767
0.773
0.696
0.613
0.645
0.667
0.666
Whites
All
32
Appendix Table 5: Fraction of Men Remaining Employed by Group in 2020-21
Period
Jan-Mar, 2020
Apr-Jun
Jul-Sep
Oct-Dec
Jan-Mar, 2021
HS or less
0.951
0.888
0.945
0.946
0.941
Some Col.
0.960
0.910
0.948
0.948
0.954
College+
0.967
0.944
0.956
0.969
0.974
Overall
0.960
0.920
0.950
0.957
0.960
Blacks
Jan-Mar, 2020
Apr-Jun
Jul-Sep
Oct-Dec
Jan-Mar, 2021
0.917
0.815
0.913
0.934
0.926
0.954
0.871
0.930
0.939
0.934
0.952
0.892
0.948
0.954
0.967
0.939
0.857
0.929
0.942
0.942
Hispanics
Jan-Mar, 2020
Apr-Jun
Jul-Sep
Oct-Dec
Jan-Mar, 2021
0.935
0.870
0.933
0.933
0.928
0.941
0.877
0.956
0.947
0.946
0.967
0.931
0.954
0.961
0.954
0.943
0.885
0.943
0.941
0.937
Jan-Mar, 2020
Apr-Jun
Jul-Sep
Oct-Dec
Jan-Mar, 2021
0.951
0.700
0.940
0.940
0.946
0.930
0.801
0.923
0.942
0.954
0.962
0.934
0.965
0.974
0.973
0.955
0.883
0.955
0.963
0.966
Jan-Mar, 2020
0.942
0.955
Apr-Jun
0.868
0.896
0.938
0.946
Jul-Sep
Oct-Dec
0.941
0.946
Jan-Mar, 2021
0.936
0.950
∗
Statistics are calculated using population weights.
0.965
0.937
0.956
0.968
0.972
0.954
0.905
0.947
0.953
0.954
Whites
Asians
All
33
‘
Appendix Table 6: Fraction of Women Remaining Employed by Group in 2020-21
Period
Jan-Mar, 2020
Apr-Jun
Jul-Sep
Oct-Dec
Jan-Mar, 2021
HS or less
0.940
0.841
0.921
0.933
0.947
Some Col.
0.946
0.874
0.934
0.944
0.954
College+
0.949
0.906
0.930
0.961
0.964
Overall
0.946
0.885
0.929
0.951
0.958
Blacks
Jan-Mar, 2020
Apr-Jun
Jul-Sep
Oct-Dec
Jan-Mar, 2021
0.909
0.790
0.895
0.910
0.922
0.925
0.855
0.891
0.948
0.931
0.953
0.893
0.933
0.948
0.958
0.931
0.853
0.910
0.937
0.939
Hispanics
Jan-Mar, 2020
Apr-Jun
Jul-Sep
Oct-Dec
Jan-Mar, 2021
0.918
0.810
0.909
0.900
0.908
0.937
0.842
0.924
0.928
0.936
0.933
0.863
0.909
0.962
0.941
0.927
0.836
0.913
0.926
0.925
Asians
Jan-Mar, 2020
Apr-Jun
Jul-Sep
Oct-Dec
Jan-Mar, 2021
0.926
0.695
0.888
0.908
0.948
0.917
0.769
0.943
0.940
0.940
0.937
0.911
0.941
0.966
0.970
0.932
0.863
0.933
0.950
0.961
Jan-Mar, 2020
0.928
0.940
Apr-Jun
0.818
0.862
Jul-Sep
0.912
0.926
Oct-Dec
0.920
0.942
Jan-Mar, 2021
0.933
0.947
∗
Statistics are calculated using population weights.
0.947
0.901
0.930
0.960
0.962
0.940
0.872
0.925
0.945
0.951
Whites
3-8 All
34
Appendix Table 7: Fraction of Men Entering Employment by Group in 2020-21
Period
Jan-Mar, 2020
Apr-Jun
Jul-Sep
Oct-Dec
Jan-Mar, 2021
HS or less
0.111
0.151
0.126
0.116
0.104
Some Col.
0.127
0.189
0.162
0.141
0.146
College+
0.196
0.195
0.245
0.184
0.170
Overall
0.137
0.174
0.172
0.141
0.132
Blacks
Jan-Mar, 2020
Apr-Jun
Jul-Sep
Oct-Dec
Jan-Mar, 2021
0.075
0.110
0.114
0.086
0.089
0.142
0.188
0.149
0.120
0.149
0.233
0.216
0.182
0.182
0.238
0.111
0.145
0.134
0.111
0.128
Hispanics
Jan-Mar, 2020
Apr-Jun
Jul-Sep
Oct-Dec
Jan-Mar, 2021
0.213
0.253
0.194
0.185
0.239
0.209
0.196
0.253
0.199
0.172
0.319
0.270
0.275
0.220
0.213
0.224
0.242
0.221
0.193
0.221
Asians
Jan-Mar, 2020
Apr-Jun
Jul-Sep
Oct-Dec
Jan-Mar, 2021
0.150
0.132
0.151
0.146
0.134
0.179
0.206
0.197
0.160
0.190
0.278
0.211
0.249
0.167
0.197
0.220
0.183
0.206
0.157
0.172
All
Jan-Mar, 2020
Apr-Jun
Jul-Sep
Oct-Dec
Jan-Mar, 2021
0.123
0.163
0.138
0.124
0.128
0.141
0.191
0.174
0.145
0.152
0.219
0.207
0.242
0.185
0.186
0.149
0.181
0.176
0.144
0.147
Whites
∗
Statistics are calculated using population weights.
35
‘
Appendix Table 8: Fraction of Women Entering Employment by Group in 2020-21
Period
Jan-Mar, 2020
Apr-Jun
Jul-Sep
Oct-Dec
Jan-Mar, 2021
HS or less
0.057
0.089
0.096
0.069
0.063
Some Col.
0.085
0.129
0.115
0.110
0.086
College+
0.144
0.133
0.208
0.126
0.113
Overall
0.093
0.116
0.147
0.100
0.086
Blacks
Jan-Mar, 2020
Apr-Jun
Jul-Sep
Oct-Dec
Jan-Mar, 2021
0.086
0.096
0.102
0.076
0.063
0.119
0.097
0.145
0.104
0.122
0.161
0.162
0.221
0.196
0.127
0.112
0.110
0.141
0.107
0.095
Hispanics
Jan-Mar, 2020
Apr-Jun
Jul-Sep
Oct-Dec
Jan-Mar, 2021
0.072
0.097
0.111
0.090
0.084
0.094
0.135
0.138
0.120
0.125
0.176
0.144
0.210
0.101
0.142
0.090
0.112
0.137
0.098
0.100
Asians
Jan-Mar, 2020
Apr-Jun
Jul-Sep
Oct-Dec
Jan-Mar, 2021
0.073
0.095
0.144
0.096
0.085
0.123
0.127
0.124
0.111
0.103
0.111
0.113
0.126
0.084
0.091
0.102
0.110
0.131
0.092
0.091
Jan-Mar, 2020
0.067
0.093
Apr-Jun
0.093
0.124
0.104
0.124
Jul-Sep
Oct-Dec
0.077
0.111
Jan-Mar, 2021
0.070
0.099
∗
Statistics are calculated using population weights.
0.143
0.134
0.199
0.124
0.114
0.096
0.114
0.143
0.100
0.091
Whites
All
36
@FedFRASER