Oi Methodology 2020 05

Summary

A May 2020 research paper, Real-Time Economics: A New Platform to Track the Impacts of COVID-19 on People, Businesses, and Communities Using Private Sector Data, by Raj Chetty, John N. Friedman, Nathaniel Hendren, Michael Stepner and the Opportunity Insights Team. The paper describes a publicly available Economic Tracker built from anonymized data supplied by credit card processors, payroll firms, job posting aggregators and financial services firms, reporting consumer spending, employment, business revenues and other indicators by county and industry. The authors report that state shutdowns and re-openings had little or no impact on economic activity in their analysis. The data and methods section explains how Homebase data are used to build hourly employment, hours worked and wage series and compares them with the QCEW and CPS. The paper closes with a list of references.

Summary drafted by a model from the document's text below and checked by script against that text before publication. It is a navigation aid, not a reading of what the document proves. Where AI is used

Full text

Real-Time Economics: A New Platform to Track the Impacts of
COVID-19 on People, Businesses, and Communities Using Private
Sector Data*

Raj Chetty, John N. Friedman, Nathaniel Hendren, Michael Stepner,
and the Opportunity Insights Team

May 2020

Abstract
We build a new, publicly available economic tracker that measures economic activity at a high-
frequency, granular level. Using anonymized data from several large businesses – credit card
processors, payroll firms, job posting aggregators, and financial services firms – we construct
statistics on consumer spending, employment rates, incomes, business revenues, job postings,
and other key indicators. We report these statistics in real time using an automated pipeline
that ingests data from these businesses and reports statistics publicly on the data visualization
platform, typically with a lag of three days or less since the relevant transactions occur. We
present fine disaggregations of the data, reporting each statistic by county and by industry and,
where feasible, by initial (pre-crisis) income level and business size. We illustrate how the tracker
can be used by measuring the economic impacts of the COVID-19 crisis on people, businesses,
and communities and estimate the causal effects of recent local policy decisions. Going forward,
we hope this tracker will serve as a public good that facilitates more precise targeting of policies
and rapid diagnosis of the root causes of economic crises.

*
We thank Gabriel Chodorow-Reich, John Grigsby, Erik Hurst, Lawrence Katz, James Stock, and Ludwig Straub
for helpful comments. We also thank the corporate partners who provided the underlying data used in the Economic
Tracker, who as of this version include: Affinity, Burning Glass, Earnin, Facteus, Homebase, Second Measure, Womply,
and Zearn. We are very grateful to Ryan Rippel of the Gates Foundation for his support in launching this project and
to Gregory Bruich for early conversations that helped spark this work. The work was funded by the Chan-Zuckerberg
Initiative, Bill & Melinda Gates Foundation, Overdeck Family Foundation, and Andrew and Melora Balson. The
project was approved under Harvard University IRB 20-0586.

The Opportunity Insights Economic Tracker Team consists of Matthew Bell, Gregory Bruich, Tina Chelidze, Lu-
cas Chu, Westley Cineus, Sebi Devlin-Foltz, Michael Droste, Shannon Felton Spence, Federico Gonzalez, Rayshauna
Gray, Abby Hiller, Matthew Jacob, Tyler Jacobson, Margaret Kallus, Laura Kincaide, Cailtin Kupsc, Sarah LaBauve,
Maddie Marino, Kai Matheson, Kate Musen, Danny Onorato, Sarah Oppenheimer, Trina Ott, Lynn Overmann, Max
Pienkny, Jeremiah Prince, Daniel Reuter, Peter Ruhm, Emanuel Schertz, Kamelia Stavreva, James Stratton, Eliza-
beth Thach, Nicolaj Thor, Amanda Wahlers, Kristen Watkins, Alanna Williams, David Williams, Chase Williamson,
and Ruby Zhang.
I Introduction

Economic policy decisions have traditionally been made based on employment and business ac-
tivity data collected from national surveys of households and businesses. Although such statistics
have great value in understanding the economy, they have two limitations that have become appar-
ent in the face of the fast-moving COVID-19 pandemic. First, such data are often available only
with a significant time lag. For instance, the Employment Situation Summary (i.e., jobs report)
released by the Bureau of Labor Statistics on May 8 presents information on employment rates
as of the week ending April 12; the next update will not come for another month. Second, due
to limitations in sample sizes, such statistics typically cannot be used to assess granular variation
across geographies or subgroups; most statistics are typically reported only at the state level and
breakdowns by demographic subgroups or sectors are often unavailable.
In this paper, we address these challenges by building a new, freely accessible platform that
tracks economic activity in real time at a granular level. Using anonymized and aggregated data
from several large businesses – credit card processors, payroll firms, job posting aggregators, fi-
nancial services firms – we construct statistics on consumer spending, employment rates, incomes,
business revenues, job postings, and other key indicators. We report these statistics in “real time”
using an automated pipeline that ingests data from these businesses and reports statistics publicly
on the data visualization platform, typically with a lag of three days or less since the relevant
transactions occur. We present fine disaggregations of the data, reporting each statistic by county
and by industry and, where feasible, by initial (pre-crisis) income level and business size.
Many firms already analyze their own data internally to make business decisions and some
firms have begun sharing aggregated data with policymakers and researchers in the current crisis.
Our contribution is to (1) collect these disparate data sources into a single, publicly accessible
platform that eliminates the need to write contracts with specific companies to access relevant
data; (2) systematize these data sources by documenting the samples they cover and adjusting for
selection biases, seasonal fluctuations, and other statistical issues; and (3) provide the combined
series in an interactive data visualization tool that facilitates comparisons across outcomes, areas,
and subgroups. The key technical problem that our platform solves is that it allows companies
to share data without disclosing sensitive information about their business or clients by combining
data from multiple sources into a single series. Hence, the platform serves as a coordination device

1
for the use of private sector data to inform public policy, one that we hope will expand over time
and be a useful resource for economic policy in this crisis and beyond.
We illustrate the value of the tracker by analyzing the impacts of recent policy decisions in the
COVID-19 crisis, focusing in particular on state shutdowns and re-openings. Perhaps surprisingly,
we find these policies have little or no impact on economic activity. The decline in economic activity
– consumer spending, the number of small business open, employment – occurred in most cases
before states “shut down,” consistent with other recent work examining data on hours of work and
movement patterns (Bartik et al. 2020, Villas-Boas et al. 2020). Moreover, we show that recent
policies ending these shut-downs in certain states such as Georgia and South Carolina have not been
associated with significant increases in economic activity. These findings suggest that the primary
barrier to economic activity is the threat of COVID-19 itself as opposed to legislated economic
shutdowns. This simple analysis illustrates the utility of the tracker: this finding would not be
evident in traditional government survey data for several months, but is easily observed in private
sector data a few days after policy changes are made.
Our work builds on and contributes to a rapidly evolving literature on the economic impacts of
COVID-19 as well as a long literature in macroeconomics on the measurement of economic activity
at business cycle frequencies. Several recent papers have used private sector data analogous to
what we assemble here to analyze labor market trends (e.g., Bartik et al. 2020, Kahn, Lange, and
Wiczer 2020), spending patterns (e.g., Alexander and Karger 2020, Baker et al. 2020, Chen, Qian,
and Wen 2020), business revenues (e.g.,), and social distancing (e.g., Allcott et al. 2020, Chiou
and Tucker 2020, Goldfarb and Tucker 2020, Mongey, Pilossoph, and Weinberg 2020). These
papers have identified a number of important patterns that we observe in our data as well, such as
larger reductions in income and employment for lower-income workers and concentrated impacts in
certain industries such as food and accommodation (e.g., Cajner et al. 2020). Each of these papers
typically analyzes one or two of these data sources, obtained through a data use agreement with
the relevant firm. We combine many of these datasets into a unified, freely accessible platform that
is automatically updated to pull the most recent data from companies’ internal databases. This
approach eliminates the need to obtain specific permissions to use data from each company, thereby
providing a public good that we hope will support the work of researchers, policy makers, and the
general public.1

1. Additionally, building on work by Gupta et al. (2020), we build a systematic, quantitative list of key policy
changes made in the COVID-19 crisis at the federal, state, and local levels that can be used by researchers and
policymakers to identify causal effects of policy changes and uncover mechanisms underlying economic outcomes.

2
Going forward, we envision two roles for such a platform to support macroeconomic policy. First,
the data can permit precise targeting of policies to subgroups and areas that are most affected by
a crisis by directly revealing which groups have been impacted most. Second, the data can be
used to learn rapidly from heterogeneity across areas – which are often hit by differential shocks
and pursue different local policy responses. This approach can permit rapid diagnosis of the root
factors underlying an economic crisis, potentially facilitating more effective macroeconomic policy
responses.
The paper is organized as follows. The next section describes our data and the methods we use
to construct the indices. In Section 3, we illustrate how the tracker can be used by presenting some
simple illustrative event studies of key outcomes around legislated shutdowns and re-openings at the
state level. Section 4 concludes by discussing policy implications and potential future applications
of the tracker. Technical details on data, methods, and supplementary analyses are available in an
online appendix. This paper will be updated to include further analysis as we obtain and analyze
more data, both from the data providers described below and other companies whose data are
currently being incorporated into the platform.

II Data and Methods

In this section, we describe the data sources and methods we use to construct the aggregated
series in the Economic Tracker. We organize the section around the central outcomes we study:
employment and earnings, consumer spending, business activity, education, and public health out-
comes related to the COVID-19 crisis. For each of these outcomes, we describe the underlying data
sources in turn, characterizing their samples and variable definitions to the extent permissible by
the data providers.

II.A Employment and Earnings
II.A.1 Homebase

We form our series for Hourly Employment and Hours Worked at small businesses using data
from Homebase. Homebase is a company that provides virtual scheduling and time-tracking tools,
focused on small businesses in sectors such as retail, restaurant, and leisure/accommodation (Table
1 discussed below provides precise industry compositions). For our purposes, they have information
on hours worked as well as wages of employees.

3
We receive de-identified data on employees at Homebase clients at the establishment-worker-day
level. We then aggregate to the county-day, metro-day, state-day, national-day level. Individuals
who work at multiple establishments are treated as distinct employees. We exclude firms that began
using Homebase in 2020 and salaried employees. We suppress estimates for geographies with fewer
than 10 Homebase clients in January 2020. The series runs from January 5, 2020 to the present.

Hourly Employment Our Hourly Employment measure is constructed as a seven-day moving
average percent change in number of hourly employees relative to January 2020. We construct this
by taking the sum of the previous six days of number of hourly employees at an establishment and
the current day’s number of hourly employees at an establishment, and then dividing by 7. We then
index each location relative to their pre-COVID-19 employment by dividing each moving average
value by their mean value during the January indexing period. We then subtract 1 to center the
series around 0. We aggregate across establishments to produce estimates at county, state and
national geographies, weighting by number hours worked in January at the establishments. We
assign location based on the zip code of establishment and crosswalk to counties and metros.

Comparison to QCEW In order to understand the context for this series, we compare cov-
erage rates to the distribution of employees and employers in the Quarterly Census of Employment
and Wages (QCEW). Table 1 compares the Homebase average monthly employees by industry
distribution to the QCEW for the period January 2018-September 2019. Figure 1 compares the
Homebase and QCEW distribution of firm sizes in the food and beverage industry for the period
January-March 2019. We can see that Homebase clients are not representative of the national
employment distribution, primarily consisting of small food and beverage and retail establishments
and their employees.

4
TABLE 1: Distribution of Average Number of Monthly Employees by Industry, QCEW
and Homebase

FIGURE 1: Comparing Average Employment Share by Establishment Size (Food & Drink)

When examining trends in employment in its key industries, Homebase patterns generally track
with the QCEW. Figure 2 and Figure 3 compare the national aggregate series of month-to-month
employment change in the food and beverage industry and retail industry, respectively. Homebase
employment change is calculated as establishment employment change at the weekly level then
averaging to the month, weighted by firm size in the prior month; QCEW employment growth is
computed nationally.

5
FIGURE 2: Month-to-Month Employment Growth (Food & Drink)

FIGURE 3: Month-to-Month Employment Changes (Retail)

Additionally, we compare Homebase wage distributions to the Current Population Survey
(CPS), which reports surveyed hourly earnings that exclude tips and overtime pay. The Home-
base data excludes tips but could include overtime pay. All wage data has been inflated to October

6
2019 levels. Figure 4 compares that national distributions of hours worked by wage level in the
food and beverage industry, for the period August-October 2019 (the most recent three months
of the CPS). The distributions appear quite similar; the wages offered by food and beverage firms
in the Homebase data reflect the national distribution in the industry. Figures 5 and 6 show that
national aggregate time trend in hours worked by wage bin in the food and beverage industry for
CPS and Homebase respectively. Both sets show relatively stability in hours worked by wage bin
and Homebase appears reflective of national wage trends.

FIGURE 4: Share of Total Hours Worked by Wage (Food & Drink)

7
FIGURE 5: CPS Share of Hours Worked by Wage Over Time (Food & Drink)

FIGURE 6: Homebase Share of Hours Worked by Wage Over Time (Food & Drink)

Hours Worked The key measure is the seven-day moving average percent change in total hours
worked relative to January 2020. We construct daily values as a 7-day moving average by taking the

8
sum of the previous six days of total hours worked at an establishment and the current day’s total
hours worked at an establishment, and then dividing by 7. We then index each location relative to
their pre-Covid-19 employment by dividing each moving average value by their mean value during
the January indexing period. We then subtract 1 to center the series around 0. We aggregate
across establishments to produce estimates at county, state and national geographies, weighting by
number hours worked in January at the establishments. We then produce metro-level estimates for
select large metros.

Average Hourly Wages The key measure is change in individuals’ hourly wages rates by week
relative to the second week of January 2020. We produce week-on-week change by comparing
hourly wages to the prior week for individuals who are observed in both time periods. We restrict
the sample to individuals making at least $1 above the state minimum wage (since it would not be
possible to see downward wage movement for individuals already at a lower bound). We aggregate
across individuals to produce estimates at county, state, metro, and national geographies.

Individual Earnings The key measure is the seven-day moving average percent change in total
earnings relative to January 2020. We construct daily values as a 7-day moving average by taking
the sum of the previous six days of total employee earnings at an establishment and the current
day’s total employee earnings at an establishment, and then dividing by 7. We then index each
location relative to their pre-Covid-19 employment by dividing each moving average value by their
mean value during the January indexing period. We then subtract 1 to center the series around 0.
We aggregate across establishments to produce estimates at county, state and national geographies,
weighting by number hours worked in January at the establishments. We then produce metro-level
estimates for select large metros.

II.A.2 State Unemployment Benefit Claims

We download and display claims data from the Office of Unemployment Insurance at the De-
partment of Labor at the state-week level and national-week level. Future versions will provide
information at the county-week level. Location is defined by the county and state of the filer’s
reported residence.

Unemployment Claims We provide both new employment claims and total employment claims.
Total claims are the count of new claims plus the count of people receiving unemployment insurance

9
benefits in the same period of eligibility as when they last received the benefits.

Unemployment Claims Rate We also construct an unemployment claims rate per 100 people
by taking the total number of claims filed, multiplying by 100, and dividing by the 2019 Census
population estimates2

II.B Consumer Spending
II.B.1 Affinity

We receive anonymized aggregate consumer spending data from Affinity Solutions Inc. Affinity
Solutions is a company that collects consumer purchasing information from card-based transactions
and uses this data to provide marketing insights. Information collected via Affinity is on the
purchaser side of the market and therefore provides a measure of consumer spending.

All Consumer Spending We receive and present data on credit and debit transactions at the at
the county-merchant category code-day level. We then crosswalk the MCCs to industry codes and
then aggregate up to the metro-industry-day, state-industry-day, and national-industry-day levels.
We mask all observations for county-industry combinations that have an average of less than
$XX in spending per day between January 4 and January 31. The series runs from January 7, 2020
to the present.
We construct daily values as a 7-day moving average as the sum of the previous six days spending
and the current day’s spending divided by 7. We then seasonally adjust the series by dividing each
calendar date’s 2020 value by its corresponding value from 2019. For the 29th of February, we use
an average between the 28th and the 1st of 2019 to adjust. We then index the seasonally adjusted
series relative to pre-Covid-19 spending by dividing the seasonally adjusted series by the mean
seasonally adjusted 7-day moving average from January 4-31.
Industry is received as merchant category codes (MCC), which crosswalk to categories to Ap-
parel & General Merchandise; Arts, Entertainment & Recreation; Grocery; Health Care; Restaurant
& Hotels; and Transportation. These categories do not sum to total spending.
Location is assigned to a county by Affinity using the address of the cardholder.

Comparison to CEX (And Womply Small Business Spending Below) In order to
understand the representativeness of this series, we compare the Affinity aggregate spending distri-

2. See the variable “popestimate2019” available here.

10
butions to the Consumer Expenditures Survey (CEX) from the Bureau of Labor Statistics. Figure
7 compares aggregate spending during the 2019, across NAICS codes. We crosswalk the Universal
Classification Codes (UCCs) in the CEX to NAICS using an internally generated crosswalk. Affin-
ity aggregate spending is skewed towards retail spending and has an overrepresentation of finance
and insurance; food and construction spending are underrepresented.
To understand how the spending series reflects national spending trends, we compare the na-
tional aggregate spending series to the Monthly Retail Trade Survey (MRTS), a national survey of
retailers and food service providers. Figure 8 compares the MRTS and Affinity month-to-month
spending series for the period January- December 2019, indexed to January 2019; Figure 9 is the
same but for the retail industry. Overall, Affinity data closely mirrors national patterns in food
and drink and retail spending over time.

FIGURE 7: Aggregate Spending by Industry (2019), Womply, Affinity, and CEX

11
FIGURE 8: Month-to-Month Spending Change, Affinity and MRTS (Food & Drink)

FIGURE 9: Month-to-Month Spending Change, Affinity and MRTS (Retail)

12
II.C Business Activity
II.C.1 Burning Glass

We receive job posting data at the geography-week level from Burning Glass which sources job
posting data from approximately 40,000 online job boards in the United States.

Job Postings We construct a measure of job postings by aggregating to the county-week, metro-
week, state-week, and national-week levels, with industry and qualification cuts at each level.
Burning Glass removes duplicate postings in the data and assigns postings to geographic areas
using an internal algorithm. Postings with or without associated employers are included. The
dates used for the sample are January 1, 2020 to present.
Pooled estimates at all geographic levels are derived directly from the Burning Glass data, as
are industry and qualification cuts at the state level and for the largest 200 counties. For other
counties we impute industry postings as follows: we first use the state-industry data to get the
share of postings in each industry in a given week and state. As the county total posts include
posts that are missing an industry classification, we create a county-level variable that is number
of posts that are not missing industry. We then multiply this number by the state-industry shares
to get county-industry imputations.
We use this data to construct the number of job postings relative to January 2020. We index
to pre-COVID-19 job posting levels by dividing a count of the week’s unique job postings by
the average weekly postings from January 4-31, 2020 for each geography-industry and geography-
qualification cell, with the exception of the imputed county-industry values (for counties outside
the largest 200). For values in those counties, we compare the imputed value to the average weekly
imputed postings from January 4-31, 2020.
Qualifications required are defined by ONET Job Zones, which are mutually exclusive categories
describing occupations as needing little or no preparation, some preparation, medium preparation,
considerable preparation or extensive preparation. A Burning Glass algorithm defines the ONET
Job Zone. In the tracker ONET Job Zones are referred to “required education” for brevity. See this
link for further details about this classification.
Industry is defined using select NAICS supersectors, which are aggregated from 2-digit NAICS
classification codes assigned by a Burning Glass algorithm. We include Construction (20); Ed-
ucation & Heath Services (65); Leisure & Hospitality (70); Manufacturing (30); Transportation
& Trade (40). Note that these categories are not exhaustive and therefore do not sum to total

13
postings.

Comparison to JOLTS Burning Glass constructs a fairly comprehensive scan of online job
postings, but may miss certain job postings that do not appear online or are not captured by their
web scanning. In order to understand the representativeness of this series, we compare the occu-
pation distributions to the Bureau of Labor Statistics’ Job Openings and Labor Market Turnover
(JOLTS) survey. Figure 10 compares the national aggregate industry distributions of the Burning
Glass data and JOLTS categorized 2-digit NAICS codes for January 2020. The distributions are
quite similar in general.

FIGURE 10: Industry Distributions for JOLTS and Burning Glass

II.C.2 Womply

We receive data on purchases at small businesses from Womply. Womply is a company that collects
commercial transactions data. Information collected via Womply is on the seller side of the market
and therefore provides a measure of small business revenue and activity.
We receive aggregate business sales data at the county-industry-day and state-industry-day level.
Womply aggregates this data up from deidentified credit card transaction level data at businesses
served by its payment processing partners. Prior to aggregation, they restrict to businesses meeting
the following criteria:

14
A businesses with 30 or more transactions in a quarter and more than on transaction in 2 to
3 months.

A business must have annual revenue that is less than the SBA thresholds by industry AND
within 2x the IQR (interquartile range) of the revenue Womply observes.

We aggregate data to the county-industry-week, metro-industry-week, state-industry-week and
national-industry-week level. We drop counties with less than 25,000 residents, as defined by the
2019 Census population estimates. At the state and national levels we include breakdowns for the
level of poverty of the zip codes of Womply businesses (i.e. for a given state, you can view data
separately for businesses located in high/middle/ low income zip codes as defined by the bottom
25%/middle 50%/top 25% of 2010 poverty levels in the national distribution). The series runs from
January 7, 2020 to the present.

Small Business Revenue The Small Business Revenue statistic measures the net businesses
revenue relative to January 2020. Net business revenue is the sum of all credits (generally purchases)
and debits (generally returns). We construct daily values as a 7-day moving average as the sum
of the previous six days of net revenue and the current day’s net revenue divided by 7. We then
seasonally adjust the series by dividing each calendar date’s 2020 value by its corresponding value
from 2019. For the 29th of February, we use an average between the 28th and the 1st of 2019
to adjust. We then index the seasonally adjusted series relative to pre-Covid-19 net revenue by
dividing the seasonally adjusted series by the mean seasonally adjusted 7-day moving average from
January 4-31.
Industry is received as Womply transaction categories, which are based on MCCs and include
some amount of additional Womply data on online business categories. We crosswalk these cate-
gories to two-digit NAICS codes, using an internally generated Womply category-NAICS crosswalk,
and then aggregate to selected NAICS Supersectors: Construction (20); Education & Heath Ser-
vices (65); Leisure & Hospitality (70); Manufacturing (30); Transportation & Trade (40). Note
that these categories do not sum to total businesses.
Location is assigned by Womply as the county and state of the business at which the transaction
occurred.

Comparison to CEX In order to understand the context for this series, we compare the
Womply aggregate spending distributions to the Consumer Expenditures Survey (CEX) from the

15
Bureau of Labor Statistics. Figure 7 compares aggregate spending during the 2019, across NAICS
codes. We crosswalk the Universal Classification Codes (UCCs) in the CEX to NAICS using an
internally generated crosswalk. Womply total revenue, which is a proxy for spending, is somewhat
skewed towards non-retail and food industries.

Small Businesses Open We construct a measure of changes in the number of businesses open
relative to January 2020. Businesses are tagged as not open if have had no transactions for 3
consecutive days. We construct daily values as a 7-day moving average as the sum of the previous
six days open businesses and the current day’s open businesses divided by 7. We then seasonally
adjust the series by dividing each calendar date’s 2020 value by its corresponding value from 2019.
For the 29th of February, we use an average between the 28th and the 1st of 2019 to adjust. We
then index the seasonally adjusted series relative to pre-Covid-19 businesses openings by dividing
the seasonally adjusted series by the mean seasonally adjusted 7-day moving average from January
4-31.

II.D Education
II.D.1 Zearn

We construct information on education outcomes from Zearn Inc. Zearn is an online education
company that partners with schools to provide online lessons. Zearn is directly integrated into
school curriculums and therefore provides a measure of the extent to which students of different
backgrounds are engaged in classroom activities throughout the COVID-19 shutdown. For more
information on Zearn see: https://about.zearn.org/
We receive data at the school-week level and aggregate to the county-week, city-week, state-
week, and national-week level. At the state and national levels we include breakdowns for the level
of poverty of the zip codes of Zearn schools (i.e. for a given state, you can view data separately
for students who attend schools located in high/middle/ low income zip codes as defined by the
bottom 25%/middle 50%/top 25% of 2010 poverty levels in the national distribution).
We drop schools who did not use Zearn for at least one week from January 6th-Feburary 7th and
schools that never have more than 5 students using Zearn during our analysis period. We winsorize
any values reflecting an increase of greater than 300% at the school level and cap all points in the
final series to a maximum 150% increase in student participation and a 300% increase in badges. To
reduce the effects of school break weeks and large fluctuations, we set any week for a given school

16
that reflects a 50% decrease (increase) greater than both points on either side of it to the average
of those three points The data are masked such that any county with fewer than 2 districts, fewer
than 3 schools, or fewer than 50 students on average using Zearn during the pre-period is excluded.
We fill in these masked county statistics with the commuting zone average whenever possible. The
series runs from January 6, 2020 to the present.

Online Math Participation The key outcome is change in student participation relative to
January 2020. Student participation is the number of students using Zearn in a given week.
We index to pre-COVID student participation by dividing weekly participation at the school
level by average weekly participation during the base period January 6th-February 7th and then
subtract 1 to center the data around 0% change. We aggregate school level-estimates to county,
state and national level, weighting by the average number of students using the platform at each
school during the base period.
Location is defined by the zip code of the school. When a zip code corresponds to multiple
counties, we assign the school to the county with the highest business ratio, as defined by HUD-
USPS ZIP Code Crosswalk Files. We generate city values for a selection of large cities using a
custom city-county crosswalk, available in data downloads. We assigned cities to counties and
ensured that a significant portion of the county population was in the city of interest. Some large
cities share a county, in this case the smaller city was subsumed into the larger city.
To better understand the context for this series, Figure 11 states’ true poor shares to the poor
shares calculated when weighting by the number of students in the zip codes of schools using Zearn.
Overall the relationship is strong, suggesting the locations of Zearn schools are reflective of state
economic characteristics.

17
FIGURE 11:Actual Poor Share Versus Zearn Zip-Code-Weighted Poor Share, by State

Student Progress in Math We also construct a measure of student progress in math using the
number of badges students earn from reaching milestones in a given week. We index to pre-COVID
student progress by dividing weekly total badges at the school level by average weekly badges during
the period January 6th-February 7th and then subtract 1 to center the data around 0.

II.E Public Health
II.E.1 COVID Cases

We construct this series using New York Times publicly available data available daily at the county,
state and national level. Their series attempts to count only lab-confirmed cases, but states have
differing reporting standards. See the NYTimes data description for a complete discussion of
methodology and definitions.
We generate metro values for a selection of large metros using the custom metro-county cross-
walk, with the exception of Kansas City which uses cases Kansas City proper rather than mapping
directly from counties. The NYTimes counts all New York City counties as one entity. For these
counties we instead use case data from New York State Department of Health data. The New York
specific data is available starting March 22nd.
They key outcome is number of newly confirmed cases per day per 1,000 people. We use the
Census Bureau’s 2019 population estimates to define population. We suppress data where new

18
counts are negative due to adjustments in official statistics. Data is for January 21 to present

II.E.2 COVID Death Rate

We construct this series using New York Times publicly available data available daily at the county,
state and national level. Their series attempts to count only lab-confirmed COVID deaths, but
states have differing reporting standards. This series is constructed identically to “COVID cases”,
but the key outcome is number of newly confirmed deaths per day per 1,000 people. See the
description for that series and NYTimes data description for a complete discussion of methodology
and definitions. Data is for January 21 to present.

II.E.3 COVID Tests

We construct this series using The COVID Tracking Project publicly available data on number of
total number of tests, available daily at the state and national level. Details on reporting by state
are available on their site.
The key outcome is new tests performed per day per 1,000 people. We use the Census Bureau’s
2019 population estimates to define population. We suppress data where new counts are negative
due to adjustments in official statistics.

II.E.4 Time Outside Home

We construct this series using data from Google’s “Google COVID-19 Community Mobility Re-
ports.” We download and present data at the county-day, state-day and national-day level. We
also generate metro-day values for a selection of large metros using the metro-county crosswalk.
The key outcome is time spent outside of residential locations relative to January 2020. At each
geographic level, we provide breakdowns of time spent at parks, retail and recreation, grocery and
transit locations. Details on these place types and additional information about data collection is
available from Google. As described, indexing is done relative to median value, for the corresponding
day of the week, during the 5-week period Jan 3–Feb 6, 2020, and place type is assigned by Google.
In order to construct the “Time Outside Home” variable, we use the Google provided data that
describes changes in amount of time spent as residential locations, and invert it to describe time
spent outside. As there are no base hours in the dataset, we use the American Time Use survey to
get the mean time spent inside the home (excluding time asleep) and outside the home in January
2018 for each day of the week. We multiply time spent inside the home in January with Google’s
residential percent change to get an estimate of time spent inside the home for each date. The

19
remaining waking hours in the day are our estimate for time spent outside the home. We then use
the average time spent outside the home in January to get the percent change outcome.
Note that Google Mobility trends do may not precisely reflect time spent at locations but rather
“show how visits and length of stay at different places change compared to a baseline”. We call this
time spent at a location for brevity.

III Illustrative Application: Government Shut-Downs

We illustrate the value of the tracker by analyzing the impacts of state shutdowns and re-openings
in the COVID-19 crisis, using event studies around these events to analyze their impacts.
Figure 12 plots several outcome around the day in which a state-level shutdown was imple-
mented. Figure 12A plots the average series for states that shut down “early,” defined as those
that issued a stay-at-home order and non-essential business closure in the week of March 19-26.
Figure 12B plots the average series across states that issued stay at home orders after this period.
Consistent with previous analyses (Bartik et al. 2020, Villas-Boas et al. 2020), hours of work begins
to fall prior to the formal date of the state-level shut-down. We also see similar declines in other
series such as consumer spending, small business spending, and time spent at work well before the
shutdowns. Broadly, these patterns suggests that the decline in economic activity was not driven
directly by the formal shut-downs themselves, but rather a general response to the onset of the
national COVID-19 epidemic.

20
FIGURE 12: Change in Consumer Spending, Small Business and Hours Worked Measures
Around Stay-At-Home Order
A. States Issuing Stay-at-Home Order and Business Closure Order on Same Day in Week 19-26 March

Change From Jan (%)
-20

-40

-60

-30 -20 -10 0 10 20 30
Days Before Order
Consumer Spend. Small Bus. Open Hours Worked at Small Bus.

Emp. at Small Bus. Time at Work

B. States Issuing Stay-at-Home Order or Business Closure Order After 26 March

Change From Jan (%)
0

-20

-40

-60
-30 -20 -10 0 10 20 30
Days Before Order
Consumer Spend. Small Bus. Open Hours Worked at Small Bus.

Emp. at Small Bus. Time at Work

Several states began efforts to “re-open” their economies by ending shut-down orders in late
April and early May. Figure 13 evaluates the early impacts of these policy changes by plotting the
same set of outcomes shown in Figure 12 for 4 states that have implemented re-opening policies:
Georgia, Oklahoma, South Carolina, and Alaska.

21
FIGURE 13: Change in Consumer Spending, Small Business and Hours Worked Measures
Around Partial Re-Opening
A. Alaska B. Georgia

Non-Essential Stay-at-Home Partial
Non-Essential Stay-at-Home Partial
Bus. Closed Order Re-Opening
50 Bus. Closed
50
Order Re-Opening

Change Jan. to Apr. (%) Change Jan. to Apr. (%)
0 0

-50 -50

-100 -100
1 Feb 16 Feb 1 Mar 16 Mar 1 Apr 16 Apr 1 Feb 16 Feb 1 Mar 16 Mar 1 Apr 16 Apr
Date Date
Consumer Spend. Small Bus. Open Hours Worked at Small Bus.
Consumer Spend. Small Bus. Open Hours Worked at Small Bus.
Emp. at Small Bus. Time at Work
Emp. at Small Bus. Time at Work

C. Oklahoma D. South Carolina

Stay-at-Home Non-Essential Partial Non-Essential Stay-at-Home Partial
Order Bus. Closed Re-Opening Bus. Closed Order Re-Opening
50 50

Change Jan. to Apr. (%) Change Jan. to Apr. (%)
0 0

-50 -50

-100 -100
1 Feb 16 Feb 1 Mar 16 Mar 1 Apr 16 Apr 1 Feb 16 Feb 1 Mar 16 Mar 1 Apr 16 Apr
Date Date
Consumer Spend. Small Bus. Open Hours Worked at Small Bus. Consumer Spend. Small Bus. Open Hours Worked at Small Bus.

Emp. at Small Bus. Time at Work Emp. at Small Bus. Time at Work

Figure 13A shows that we do not see an increase in economic activity in Georgia after it lifted
its stay-at-home order on April 24. Consumer spending, employment and hours at small businesses,
the number of small businesses that are open, and time spent at work all remain relatively similar
to their levels prior to April 24. We find similar patterns in other states as well, as illustrated in
Panels B-D. This suggests the primary factor limiting economic activity are choices being made by
individuals and businesses in response to the threat of COVID-19 itself, as opposed to government
policies that impose restrictions on economic activity.
The simple analysis in Figures 12 and 13 illustrates the utility of the tracker. These findings
would not be evident in traditional government survey data for several months, but are easily
observed in private sector data a few days after policy changes are made.

IV Conclusion

Modern data held by private companies provide an unprecedented capacity to measure economic
activity at a granular level very rapidly. These data have become increasingly integral for corpora-

22
tions in improving business decisions. In this paper, we have constructed a freely available platform
that harnesses the same data with the aim of supporting policymakers, non-profits, and the public
seeking to make better decisions.
In these uncertain and unprecedented times, we hope this real-time economic tracker provides
valuable information for understanding the state of the economy and facilitating the national re-
covery. We look forward to expanding upon this tracker as additional companies contribute data
and reporting analyses that emerge from these data. More broadly, we hope the approach proposed
here will serve as a template to permit real-time responses to changes in economic conditions going
forward.

23
Supplementary Appendix

In this appendix, we describe additional details about information reported in the tracker: key
dates in the COVID-19 crisis and geographic definitions.
Key Dates for COVID-19 Crisis. The Economic Tracker also includes information about key
dates relevant for understanding the impacts of the COVID-19 crisis. At the national level, we
highlight three key dates:

First U.S. COVID-19 Case: 1/20/2020

National Emergency Declared: 3/13/2020

CARES Act Signed in to Law: 3/27/2020

At the state level we highlight dates when:

Schools closed statewide: Sourced from COVID-19 Impact: School Status Updates by MCH
Strategic Data, available here. Compiled from public federal, state and local school informa-
tion and media updates.

Nonessential businesses closed: Sourced from the Institute for Health Metrics and Evalua-
tion state-level data (available here), who define a non-essential business closure order as:
”Only locally defined ’essential services’ are in operation. Typically, this results in closure
of public spaces such as stadiums, cinemas, shopping malls, museums, and playgrounds. It
also includes restrictions on bars and restaurants (they may provide take-away and delivery
services only), closure of general retail stores, and services (like nail salons, hair salons, and
barber shops) where appropriate social distancing measures are not practical. There is an
enforceable consequence for non-compliance such as fines or prosecution.”

Stay-at-home order goes into effect: Sourced from the New York Times stay at home order
data, available here.

Stay-at-home order ends: Sourced from the New York Times reopening data, available here.
Defined as the date at which the state government lifted or eased the executive action telling
residents to stay home.

Partial business reopening: Sourced from the New York Times reopening data, available here.
Defined as the date at which the state government allowed the first set of businesses to reopen.

24
These dates are updated as of 5/4/2020.
Geographic Definitions. For many of the series we convert from counties to metros and zip codes
to counties. Unless mentioned as otherwise the crosswalks are as follows:

ZIP Codes to County We use the HUD-USPS ZIP Code Crosswalk Files to convert from
zip code to county. When a zip code corresponds to multiple counties, we assign the entity to the
county with the highest business ratio, as defined by HUD-USPS ZIP Crosswalk.

County to Metro Areas We generate metro values for a selection of large cities using a
custom metro-county crosswalk, available in Appendix Table 1. We assigned metros to counties
and ensured that a significant portion of the county population was in the metro of interest. Some
large metros share a county, in this case the smaller metro was subsumed into the larger metro.

25
Appendix Table I: Metro-County Crosswalk
County FIPS
Metro County State
Code

Albuquerque Bernalillo NM 35001
Atlanta Fulton GA 5049
Austin Travis TX 48453
Bakersfield Kern CA 6029
Baltimore Baltimore MD 24005
Boise Ada ID 16001
Boston Suffolk MA 25025
Charlotte Mecklenburg NC 37119
Chicago Cook IL 17031
Cleveland Cuyahoga OH 39035
Colorado Springs El Paso CO 8041
Columbus Franklin OH 39049
Dallas Dallas TX 48113
Denver Denver CO 8031
Detroit Wayne MI 26163
El Paso El Paso TX 48141
Fort Worth Tarrant TX 48439
Fresno Fresno CA 6019
Honolulu Honolulu HI 15003
Houston Harris TX 48201
Indianapolis Marion IN 18097
Jacksonville Duval FL 12031
Kansas City Jackson MO 29095
Las Vegas Clark NV 32003
Los Angeles Los Angeles CA 6037
Louisville Jefferson KY 21111
Memphis Shelby TN 47157
Miami Dade FL 12025
Milwaukee Milwaukee WI 55079
Minneapolis Hennepin MN 27053
Nashville Davidson TN 47037
New Orleans Orleans LA 22071
New York City New York NY 36061
New York City Kings NY 36047
New York City Queens NY 36081
New York City Bronx NY 36005
New York City Richmond NY 36085
Oakland Alameda CA 6001
Oklahoma City Oklahoma OK 40109
Omaha Douglas NE 31055
Philadelphia Philadelphia PA 42101
Phoenix Maricopa AZ 4013
Portland Multnomah OR 41051
Raleigh Wake NC 37183
Sacramento Sacramento CA 6067
Salt Lake City Salt Lake UT 49035
San Antonio Bexar TX 48029
San Diego San Diego CA 6073
San Francisco San Francisco CA 6075
San Jose Santa Clara CA 6085
Seattle King WA 53033
Tampa Hillsborough FL 12057
Tucson Pima AZ 4019
Tulsa Tulsa OK 40143
Virginia Beach Virginia Beach City VA 51810
Washington District Of Columbia DC 11001
Wichita Sedgwick KS 20173
References
Alexander, Diane, and Ezra Karger. 2020. “Do stay-at-home orders cause people to stay at home?
Effects of stay-at-home orders on consumer behavior.” Federal Reserve Bank of Chicago Work-
ing Paper No. 2020-12 (April). doi:10.21033/wp-2020-12. https://www.chicagofed.org/
publications/working-papers/2020/2020-12.

Allcott, Hunt, Levi Boxell, Jacob C Conway, Matthew Gentzkow, Michael Thaler, and David Y
Yang. 2020. “Polarization and Public Health: Partisan Differences in Social Distancing during
the Coronavirus Pandemic.” NBER Working Paper No. 26946 (April). doi:10.3386/w26946.
http://www.nber.org/papers/w26946.

Baker, Scott R, R. A Farrokhnia, Steffen Meyer, Michaela Pagel, and Constantine Yannelis. 2020.
“Income, Liquidity, and the Consumption Response to the 2020 Economic Stimulus Payments.”
NBER Working Paper No. 27097 (May). doi:10 . 3386 / w27097. http : / / www . nber . org /
papers/w27097.

Bartik, Alexander W., Marianne Bertrand, Feng Li, Jesse Rothstein, and Matt Unrath. 2020. “Labor
market impacts of COVID-19 on hourly workers in small- and medium-sized businesses: Four
facts from Homebase data” (April). https://www.chicagobooth.edu/research/rustandy/
blog/2020/labor-market-impacts-from-covid19.

Cajner, Tomaz, Leland D. Crane, Ryan A. Decker, John Grigsby, Adrian Hamins-Puertolas, Erik
Hurst, Christopher Kurz, and Ahu Yildirmaz. 2020. “The U.S. Labor Market during the Be-
ginning of the Pandemic Recession.” Working Paper (May).

Chen, Haiqiang, Wenlan Qian, and Qiang Wen. 2020. “The Impact of the COVID-19 Pandemic
on Consumption: Learning from High Frequency Transaction Data.” Working Paper (April).
doi:http://dx.doi.org/10.2139/ssrn.3568574. https://ssrn.com/abstract=3568574.

Chiou, Lesley, and Catherine Tucker. 2020. “Social Distancing, Internet Access and Inequality.”
NBER Working Paper No. 26982 (April). doi:10 . 3386 / w26982. http : / / www . nber . org /
papers/w26982.

Goldfarb, Avi, and Catherine Tucker. 2020. “Which Retail Outlets Generate the Most Physical
Interactions?” NBER Working Paper No. 27042 (April). doi:10.3386/w27042. http://www.
nber.org/papers/w27042.

Gupta, Sumedha, Thuy D Nguyen, Felipe Lozano Rojas, Shyam Raman, Byungkyu Lee, Ana Bento,
Kosali I Simon, and Coady Wing. 2020. “Tracking Public and Private Responses to the COVID-
19 Epidemic: Evidence from State and Local Government Actions.” NBER Working Paper No.
27027 (April). doi:10.3386/w27027. http://www.nber.org/papers/w27027.

28
Kahn, Lisa B, Fabian Lange, and David G Wiczer. 2020. “Labor Demand in the Time of COVID-19:
Evidence from Vacancy Postings and UI Claims.” NBER Working Paper No. 27061 (April).
doi:10.3386/w27061. http://www.nber.org/papers/w27061.

Mongey, Simon, Laura Pilossoph, and Alex Weinberg. 2020. “Which Workers Bear the Burden of
Social Distancing Policies?” NBER Working Paper No. 27085 (May). doi:10.3386/w27085.
http://www.nber.org/papers/w27085.

Villas-Boas, Sofia B, James Sears, Miguel Villas-Boas, and Vasco Villas-Boas. 2020. “Are We #Stay-
ingHome to Flatten the Curve?” UC Berkeley: Department of Agricultural and Resource Eco-
nomics CUDARE Working Papers (April). https://escholarship.org/uc/item/5h97n884.

File and source

File: oi-methodology-2020-05.pdf
Size: 1,317,962 bytes
SHA-256: ccb48e5ad85766b96e6eaa952b5d2246ddf07844f495e865d5c25f1daf5ad2fa
Our copy: oi-methodology-2020-05.pdf
Original: opportunityinsights.org

Oi Methodology 2020 05

Summary

Full text

File and source

Related records