Made available through Montana State University’s ScholarWorks 
Seasonal agricultural activity and crime
Diane Charlton, Alexander James, Brock Smith
This is a pre-copyedited, author-produced PDF of an article accepted for publication in American 
Journal of Agricultural Economics following peer review. The version of record [Seasonal 
agricultural activity and crime. American Journal of Agricultural Economics 104, 2 p530-549 
(2021)] is available online at: https://doi.org/10.1111/ajae.12260.
Seasonal Agricultural Activity and Crime
Diane Charlton∗ Alexander James† Brock Smith‡§
August 16, 2021
Keywords: Crime; Immigration; Seasonal Agriculture; Farm Workers; Seasonal Employment
JEL Classification: Q10; K14; R23; F22; E24
Abstract
Sudden shocks to labor demand have sometimes been shown to increase local crime
rates. We build on this literature by estimating the causal effect of labor-intensive
seasonal agricultural activity on crime. We analyze a unique data set that describes
criminal activity and fruit, vegetable, and horticultural (FVH) employment by month
and U.S. county from 1990-2016. We find the FVH labor share is associated with re-
duced property and violent crime rates, and possibly the number of property crimes
committed within county-years. Examining heterogeneities based on ethnicity, labor-
intensive FVH activity decreases the rate of non-Hispanic arrests and victimization and
increases the number of Hispanic arrests and victims (consistent with rising local His-
panic populations). Taken together, results are broadly consistent with the idea that
agricultural harvest of labor-intensive crops enhances local labor market opportunities
that reduce incentives to commit crimes. Results are robust to a battery of alternative
specifications that address the inherent challenges associated with measuring seasonal
agricultural labor.
∗Department of Agricultural Economics & Economics, Montana State University. Bozeman, MT., 59717
Corresponding Author. Email: diane.charlton@montana.edu
†Department of Economics, University of Alaska Anchorage. Anchorage, AK. 99508.
‡Department of Agricultural Economics & Economics, Montana State University. Bozeman, MT., 59717.
§This research were funded in part by the USDA National Institute of Food and Agriculture (NIFA), Hatch
projects MONB00100 and 1026193.
Seasonal Agricultural Activity and Crime
Fruit, vegetable, and horticultural (FVH) production is characterized by large seasonal
changes in labor demand, much of which is met by migrant laborers. An estimated 38% of
seasonal FVH workers were classified as migratory from 1990-2016, and approximately 48%
were unauthorized immigrants over the same span.1 Many Americans believe that immigrants,
and especially “illegal” immigrants are more likely to commit violent crimes than the rest of
the US population.2 Such sentiments—along with anecdotal evidence—can lead to fear that
agricultural activity causes crime. For example, Huron, California, a quintessential Central
Valley town populated by farm workers, has been called “knife-fight city” in reference to the
ubiquity of knives used to harvest head lettuce during the spring combined with high poverty
and crime (Martin, Fix, and Taylor 2006). Further to the west, when agricultural guest worker
housing units were constructed in Spreckels, California, local residents raised concerns that
the presence of seasonal farm workers would increase crime and subsequently reduce home
values (Mohan 2017). Nevertheless, rigorous investigation is required to test whether these
fears and beliefs can be validated by a causal relationship between seasonal farm activities
and crime. In this paper, we identify the short-term effects of seasonal, labor-intensive FVH
activity on crime using monthly sector-specific employment data and crime reporting data at
the county level from 1990-2016.
1Based on authors’ analysis of the National Agricultural Workers Survey (NAWS). The NAWS defines a
worker as migratory if they reported jobs that were at least 75 miles apart or who reported moving more than
75 miles to obtain a farm job during a 12-month period.
2Data collected from a 2018 Grinnell College National Poll that asked 1,000 U.S. adults, “Compared to
the U.S. population overall, do you think the rate of violent crime committed by illegal immigrants in the
United States is higher, lower, or about the same?”. While 30% of respondents answered “higher” just 20%
answered “lower”. Detailed results are available at: https://www.pollingreport.com/immigration.htm. This
idea is reinforced by earlier survey data from 2000 in which 73.4% of respondents thought that it was “very
likely” or “somewhat likely” that crime rates would increase as a result of increased immigration into the
United States (Spenkuch 2014).
2
Seasonal Agricultural Activity and Crime
The relationship between crime and agricultural activity is theoretically ambiguous. On
the one hand, migrant farm workers are predominantly male, and tend to be poor and rel-
atively uneducated—all of which is associated with increased criminal activity (Campaniello
and Gavrilova 2018; Kelly 2000; Lochner and Moretti 2004).3 On the other hand, there is
empirical evidence that immigrants are equally (Reid et al. 2005; Bell, Fasani, and Machin
2013; Bianchi, Buonanno, and Pinotti 2012; Chalfin 2014) or less (Baker 2015; Butcher and
Piehl 2007; Stowell et al. 2009; Wadsworth 2010) likely to commit crimes than natural-born
citizens, and an estimated 78% of FVH workers in the United States were foreign born between
1990-2016.4,5 To the extent that agricultural activity creates economic spillovers that enhance
local labor market opportunities (which we do find evidence of), this could also reduce in-
centives to commit crimes (Blakeslee and Fishman 2018; Carr and Packham 2019; Freedman
and Owens 2016, 2018; Foley 2011; Gould, Weinberg, and Mustard 2002; Lin 2008; Watson,
Guettabi, and Reimer 2019).6 However, in certain instances significant and sudden increases
in economic activity have also been shown to increase crime, as with the U.S. shale energy
boom (Gourley and Madonia 2018; James and Smith 2017; Komarek 2018; Street 2018), a
result James and Smith (2017) hypothesize could be explained by migration and subsequent
371.1 percent of FVH workers in 2016 who reported working the previous year had an annual income below
$25,000, and 33.4 percent had income below $15,000. In the online supplementary appendix, we summarize
statistics on income distribution, labor migration, weeks not working, and other variables of potential interest
for FVH workers in the National Agricultural Workers Survey (NAWS).
4Based on authors’ analysis of the Department of Labor, Employment & Training Administration (2017)
for fiscal year 2016.
5One notable exception is (Spenkuch 2014) who finds that increasing the U.S. county migrant share of the
population is associated with more burglaries, larcenies, and grand theft-auto (but no effect on violent crimes).
He further finds that these effects only hold for immigrants from Mexico, who he posits have relatively poor
labor market opportunities and so might be prone to commit financially-motivated crimes.
6Whether agricultural activity creates meaningful short run economic spillovers remains an open question.
However, the literature finds little to no evidence of long-run economic spillovers from the agricultural sector
to other local non-farm sectors (Foster and Rosenzweig 2004; Hornbeck and Keskin 2015; Weber et al. 2015)
3
Seasonal Agricultural Activity and Crime
“social disorganization” in which the sudden inflow and outflow of people disrupts social cohe-
sion, making it easier to successfully commit crimes without being caught (Freudenburg 1986;
Sampson and Groves 1989).7 It is also possible that migration increases crime rates among
non-immigrant populations. For example, immigration has been shown to increase crime rates
among non-migrant residents who face increased employment competition (Borjas, Grogger,
and Hanson 2010). Furthermore, migrant farm workers often lack (perceived or actual) legal
protections, thus potentially contributing to their frequent victimization (Moyce and Schenker
2018; Wallis 2019).
Despite these theoretical ambiguities and the attention this topic has received in popu-
lar media outlets, to our knowledge we are the first to examine how seasonal labor-intensive
agricultural activity impacts local crime rates.8 We fill this gap by combining Uniform Crime
Reporting (UCR) data on crime counts and seasonal agricultural employment at the county-
by-month level over the period 1990-2016. We analyze how seasonal variation in FVH em-
ployment is associated with seasonal patterns of crime. Causal inference is facilitated by the
granularity of our data. While there may be unobserved factors that are correlated with agri-
cultural activity and crime (such as population density or income), such factors tend to be
specific to particular counties or years. But by exploiting county-by-month data, all identi-
fying variation comes from within the county-year level. Our key identifying assumption is
that within county-year variation in unobserved determinants of criminal activity is uncor-
7Whereas well acquainted neighbors with established friendships are more likely to keep a watchful eye
on their neighbor’s house, they may be less likely to recognize an ongoing burglary at a stranger’s house.
Relatedly, committing a crime in an environment of strong social organization is especially risky as witnesses
are more likely to recognize the perpetrator’s face and identify them to police.
8To our knowledge, the most closely related paper to ours is Blakeslee and Fishman (2018), which estimates
effects of weather-driven agricultural income shocks on crime in India. Unlike the present study, Blakeslee and
Fishman (2018) analyzes this relationship in a developing country, and does not study seasonal crime effects.
4
Seasonal Agricultural Activity and Crime
related with variation in our measure of labor-intensive agricultural activity after controlling
for month-by-year fixed effects. As long as this assumption holds, we are able to estimate the
short run (e.g., within year) causal effect of agricultural activity in FVH sectors on crime, but
this comes at the cost of foregoing analysis of longer run, more permanent effects that might
also be important.
Our analysis is made up of two parts. First, we estimate the marginal effect of an increase
in the seasonal agricultural employment share of the labor force on measures of property and
violent crime. Understanding marginal effects is important from a policy perspective, but
this baseline analysis potentially masks important non-linearities in the relationship between
seasonal employment and criminal activity. To address this, we supplement our baseline
specification with a semi-parametric one that describes how crime rates change each month
relative to the month when FVH-intensive counties have their peak seasonal farm workforce.
We find that a one percentage point increase in the seasonal FVH employment share is
associated with roughly five fewer property crimes per 100,000 members of the labor force9
(compared to a sample average of 388 property crimes per 100,000). Consonant results are
found for the violent crime rate, though with inconsistent statistical significance. Our semi-
parametric specification reinforces these findings. Relative to five months before peak seasonal
employment, property and violent crime rates are roughly 12% lower during the peak seasonal
employment month in treated counties (defined as those with significant seasonal FVH em-
ployment shares) relative to control counties.
We also investigate potential mechanisms driving these negative crime effects. As dis-
9Because population data are not available at the county-by-month level, we proxy crime rates with the
number of crimes divided by the labor force. See Section for further discussion.
5
Seasonal Agricultural Activity and Crime
cussed above, two possibilities are that migrant workers are less likely to commit crimes than
the non-migrant population, and that the non-migrant population may commit fewer crimes
during seasonal FVH activity due to economic spillovers. To explore these possibilities we
conduct an analysis of National Incident-Based Reporting System (NIBRS) data, which iden-
tify (in some cases) the ethnicity of both the victims and arrestees associated with a crime
(though the NIBRS program has much lower participation rates from police precincts than the
UCR program). A number of important insights emerge. First, during labor-intensive FVH
seasons both the victimization and arrest rate fall among non-Hispanics. Since the majority
of seasonal agricultural workers are Hispanic, this suggests that the crime rate falls for the
non-migrant population during labor-intensive seasons, consistent with the hypothesis that
seasonal agricultural activity broadly improves local economic conditions and reduces the in-
centive to commit crimes. Second, while we do not find effects on Hispanic arrest rates, we
find that the Hispanic victimization rate rises in response to seasonal labor-intensive FVH ac-
tivity. In this case, the corresponding offenders are largely Hispanic, though there is evidence
that some of these offenders are also non-Hispanic or of unknown ethnicity.
To further investigate mechanisms, we examine whether crime is affected by corn harvesting
season in corn-intensive counties rather than FVH-intensive counties. We choose corn because
it is widely grown but is not nearly as labor-intensive to harvest. We do not find evidence
of crime effects in this case, suggesting that the crime reductions we find for FVH are driven
specifically by labor market impacts rather than overall income effects (assuming that corn
harvests provide a contemporaneous increase in income in corn-producing counties).
This study contributes most directly to the literature examining the social and economic
6
Seasonal Agricultural Activity and Crime
effects of agricultural activity. To our knowledge we are the first to analyze how seasonal pat-
terns in labor-intensive agricultural activity are associated with seasonal patterns in crime. We
also contribute more generally to the literature examining the effects of labor demand shocks
and immigration on local crime rates. As mentioned above, this literature includes analysis of
the American shale energy boom, which is associated with increased crime (James and Smith
2017), and also studies of various international immigration shocks, which predominately find
a negative association with crime, though results vary.
Data
Employment Data
Our employment data come from the Quarterly Census of Employment and Wages (QCEW),
a census of all establishments that are covered by unemployment insurance compiled by the
Bureau of Labor Statistics (BLS). The QCEW provides month-by-county-by-industry employ-
ment counts for all counties and years from 1975-present. Industries are classified by NAICS
codes and employment counts are available at the six-digit level.10 One pitfall of these data is
that when there are a small number of employers in a given county-industry-year combination
(or some other reason that employers could be identifiable), wage and employment data are
suppressed. When employment data for any of our seasonal agricultural sectors are suppressed
we will under-measure the seasonal employment share. In the online supplementary appendix
we discuss this issue further and we provide a robustness check in which we drop observations
10We drop any county-year observations that report zero total employment for any month within the year,
though this is very rare.
7
Seasonal Agricultural Activity and Crime
with suppressed seasonal agricultural sectors.
To the extent that employers do not report unauthorized workers for unemployment in-
surance, we may also under-count seasonal farm workers in the QCEW. However, since the
Immigration Reform and Control Act (IRCA) was passed in 1986, employers are held legally
responsible for knowingly hiring unauthorized workers (Taylor and Charlton 2018). Thus,
farm employers have incentive to require workers to provide some form of legal documenta-
tion. Many unauthorized workers provide a false social security number to their employers
and would therefore be counted in the QCEW.11
Another concern with the QCEW is that farm employers in some states with few em-
ployees are not required to report workers for unemployment insurance, and consequently,
farm employers with few employees in these states do not record their workers in the QCEW.
Furthermore, employers in some states are required to report H-2A agricultural guest workers
for unemployment insurance while employers in other states are not.12 According to Rural
Migration News (2020), farm employers of all sizes in Washington and California must report
all employees for unemployment insurance, including H-2A workers, but farm employers in
Florida do not. Florida, North Carolina, Georgia, Washington, and California employed half
of all H-2A workers in 2016 (Martin 2017).13 This may cause us to under-count the seasonal
11It is estimated that undocumented immigrants contributed $13 billion to Social Security funds and $3
billion to Medicare in 2016 by submitting false social security numbers to employers for payroll (Roberts
2019).
12H-2A is a non-immigrant guest worker visa for seasonal farm workers. Program take-up was extremely
low between 1986-2010, but rose rapidly from 2011-2018. Nevertheless, in 2016 H-2A workers made up only 7
percent of the national farm workforce (Martin 2017).
13According to conversations with several of the leading researchers in the field of farm labor economics,
including administrators of the NAWS, there is no known database indicating which states have a threshold
number of employees below which agricultural employers do not report to Unemployment Insurance. According
to a phone call with the North Carolina Department of Commerce, farm employers in North Carolina do not
report H-2A workers in the QCEW or any farm employees if total employees is fewer than ten employees in 20
8
Seasonal Agricultural Activity and Crime
farm work force in key states, particularly if crew leaders for farm labor contractors (FLCs) are
considered individual employers. However, we drop Florida from our analysis due to irregular
crime data (see Section ), and we know that two of the other leading states in H-2A employ-
ment, California and Washington, report H-2A workers in the QCEW along with other farm
workers in the same sector. We address the concern that QCEW under-counts seasonal farm
workers by repeating our analysis using only counties in California and Washington where we
know that all employees must be reported in the QCEW.
Additionally, because H-2A agricultural guest workers could differ from other seasonal farm
workers given that they have legal temporary guest visas and are subject to the correspond-
ing regulations, we repeat our analysis using H-2A guest worker shares as the explanatory
variable.14 The results from these robustness checks are reported in the online supplementary
appendix.
An important caveat for this study is that, while we have county-by-month data on crime
counts and employment, we do not have monthly estimates of population, which creates a
challenge in estimating rates of crime (as opposed to counts). If the harvest-season employment
spike draws workers who work seasonally and remain in the same county even after their
employment ends, the increase in employment will exceed the true proportional increase in
population. For this reason we calculate monthly crime rates as the number of crimes per
total labor force, which is distinct from total employment in that it includes people who
weeks of a calendar year or payroll less than $20,000 per year. The Georgia Department of Labor was unable
to disclose any information about what Georgia employers do or do not report.
14Data on the number of H-2A workers per county-month come from the Office of Foreign Labor Certification
(OFLC) Disclosure data. Most observations include the worksite county and state for each H-2A application.
However, county names were sometimes misspelled, or employers reported the city in place of the county.
Marcelo Castillo (USDA, Economic Research Service (ERS)) generously shared with us the data that he
matched to work site county using data matching methods across employers in multiple years.
9
Seasonal Agricultural Activity and Crime
are not employed but are looking for work, and so is less sensitive to economic swings and
more representative of the working-age population, though the caveat remains that seasonal
employment spikes could draw in workers who are otherwise out of the labor force altogether.
We draw county-by-month labor force counts from the Bureau of Labor Statistics, which
constructs labor force estimates based on several sources, including the Current Population
Survey, American Community Survey, the Current Employment Statistics Survey, and state
unemployment insurance data. In addition, we run a robustness check that attempts to
account for seasonal farm workers who remain in the same county while they are not working.
Using data from the National Agricultural Workers Survey (NAWS) we find the percentage
of seasonal farm workers who report that they are settled in one location and the annual
average share of the year that these workers report that they did not work, and we adjust
our labor force denominator accordingly. This exercise is further discussed and presented in
the online supplementary appendix. Further, although measurement error in population is
of consequence in the interpretation of our estimated effects of seasonal farm labor shares on
crime rates, it is not of concern for our analysis of crime counts.
Crime Data
Crime data are drawn from Uniform Crime Reporting (UCR), which is a compilation of
incident counts by over 16,000 law enforcement agencies. We use the “Offenses Known and
Clearances by Arrest”, which contains counts of reported crimes at the month-by-agency level
for several types of offenses. It is important to note that the UCR data is restricted to serious
crimes. This is not to say, however, that an abundance of less serious criminal activity is
10
Seasonal Agricultural Activity and Crime
not also important. Our main outcomes of interest are rates of all crimes, violent crimes,
and property crimes. Violent crimes include homicide, rape, robbery, and aggravated assault.
Property crimes include burglary, larceny and motor vehicle theft. We aggregate agency-level
crime counts to the county level for our analysis.
One key issue with UCR data is that agencies are not required to report crimes. However
the data do indicate the number of months reported for a given agency and year. We drop any
agency-year combination with less than 12 months of reporting. Therefore we ensure that the
jurisdictional populations are equivalent for each month within a county-year. The month-
level design of this study makes the UCR reporting problem much less problematic than
designs that aggregate to the county-year level because agencies can be added or removed
from a county or experience large changes in reporting on a year-to-year basis. Since all of our
regressions include county-by-year fixed effects, all identifying variation is within the county-
year level where these issues do not apply. Further, some counties are not included at all in
the UCR, and this can vary by year. In our main sample, an average of 2,587 counties are
included per year. Missing counties are typically low in population.15
One remaining issue with UCR data is that in some cases even an agency that indicates
12 months of reporting loads a disproportionate number of crimes on a single month. Most
commonly in this case, agencies will have zero counts for all months except December, but
it sometimes happens for other months as well. To address this, we first drop all counties
in Florida and Alabama from our analysis since this issue is extremely common in those
15Missing counties are fairly evenly spread throughout the country geographically (aside from dropping all
counties in Florida and Alabama, as discussed below). Since rural counties are more likely to be missing, the
most rural states like Mississippi, Montana, and South Dakota have the highest share of missing counties.
11
Seasonal Agricultural Activity and Crime
states. For remaining counties, within each year we find the month with the highest number
of crimes. If the ratio of crimes in this month to the average of all other months within
the year is greater than 10, we drop that county-year combination from the analysis (we
perform this step separately for violent and property crimes). This step drops less than 1%
of observations. The threshold ratio of 10 is meant to remove especially extreme outliers that
could skew results. Our methods of dropping counties with misreported crimes data should
only introduce sample selection bias if selection is correlated with both seasonal farm labor
share and seasonal patterns of crime, which seems unlikely.
Table A1 provides basic summary statistics for the crime and employment data used in
our analyses. This table also provides statistics for seasonal employment count and seasonal
employment share for the peak seasonal employment month for the treatment group used in
our semi-parametric specification discussed in the next section.
Methodology
Production of fruit, vegetable, and horticultural (FVH) crops is characterized by high seasonal
variation in labor demand.16 Low-skilled Mexican immigrants, who make up a large share of
seasonal farm workers, respond strongly to geographic variation in labor demand Cadena and
Kovak (2016), so locations that experience an especially strong harvest are likely to draw a
large number of immigrant workers. We exploit this feature of the farm-worker labor market
and estimate the relationship between the seasonal employment share and criminal activity
16According to the 2012 Agricultural Census, there were 2.7 million workers hired on farms. There
were 3 million workers reported in the 2002 Agricultural Census, 2.6 million in 2007, and 2.4 million in
2017. Data come from the U.S. Department of Agriculture, National Agricultural Statistics Service (NASS)
https://quickstats.nass.usda.gov. Retrieved on April 27, 2020.
12
Seasonal Agricultural Activity and Crime
Table 1: Summary Statistics
Variable
Property Crimes per 100,000 Labor Force 388.3
(285.7)
[838332]
Violent Crimes per 100,000 Labor Force 171.6
(152.1)
[814020]
Labor Force 48969
(158387)
[838332]
Seasonal Employment 57.01
(670.9)
[838332]
Seasonal Employment Share (pp) 0.08
(0.68)
[838332]
Has Non-zero Seas. Employment Indicator 0.15
(0.36)
[838332]
Seasonal Employment in Peak Month (T Group Only) 2109
(4306)
[14964]
Seasonal Emp. Share in Peak Month (T Group Only) 0.07
(0.07)
[14964]
The table shows means of each variable for the baseline property and violent crime
rate regression samples. Standard errors are shown in parenthesis and sample size
used in the main regression specifications are in brackets.
13
Seasonal Agricultural Activity and Crime
using two different specifications. The first specification (which we call our baseline parametric
specification) measures the marginal effect of variation in the seasonal farm labor share on
county-month level crime rates. Our second specification (which we call our semi-parametric
specification) estimates how crime varies each month relative to the month when agricultural-
intensive counties have their peak seasonal farm workforce.
Baseline Parametric Specification
We estimate the marginal impact of seasonal agricultural labor on crime outcomes with the
following equation:
Yimy = α + β ∗ Seasonal Shareimy + µmy + γiy + imy, (1)
where Yimy is the outcome of interest for county i in month m of year y, Seasonal Shareimy
is the share of the labor force taken up by seasonal agricultural laborers (defined below),
measured in percentage points. Month-by-year fixed effects are given by µmy, and γiy is
county-by-year fixed effects. Month-by-year fixed effects control for any nation-wide month-
specific shocks in crime. County-by-year effects control for any factors constant over a calendar
year within a county. Therefore, all identifying variation comes from monthly shifts in seasonal
agricultural labor shares within a county-year, controlling for any monthly national shocks.
β is the coefficient of interest and represents the average change in crimes associated with a
one percentage point increase in the seasonal agricultural labor share. Standard errors for all
regressions are clustered at the county level.
14
Seasonal Agricultural Activity and Crime
To measure the seasonal agricultural labor share, we begin by identifying twelve Fruit
Vegetable and Horticultural (FVH) sectors in the QCEW data. These sectors, by NAICS title,
are apple orchards, grape vineyards, strawberry farming, berry (except strawberry) farming,
orange groves, citrus (except orange) groves, other vegetable and melon farming (excluding
potatoes), other non-citrus fruit farming, fruit and tree nut combination farming, food crops
grown under cover (Greenhouse), and nursery and floriculture production.17 These sectors
consist of crops with high shares of seasonal labor demand.
The QCEW employee counts for the sectors listed above do not include labor hired through
farm labor contractors (FLCs), who hire farm workers and contract them to work on individual
farms for short-term jobs. FLCs provide a service to reduce labor market frictions when many
workers are needed in various locations for short periods. We account for employees of FLCs
using multiple methods. In our main specification, we include the employees hired under the
NAICS title farm labor contractors and crew leaders in the counties where they are reported.
However, given that FLCs may transport workers to different counties to work on multiple
farms throughout the year, we perform a robustness check in which we estimate the number
of FLC workers contracted in each county based on the share of labor expenditures per county
attributed to contract labor in the Agricultural Censuses in 2002, 2007, and 2012.18
Of course, each of the twelve FVH sectors contain permanent laborers, in addition to
seasonal ones, and workers who may work continuously throughout the year on multiple
17Employment on potato farms is reported separately from other vegetables, and potato harvests are gen-
erally highly mechanized. Therefore, we excluded potatoes from our analysis. See, for example, Patterson
(2015) for a cost and return study for potato production in Idaho.
18We linearly interpolate shares of labor contracted through FLCs between 2002-2007 and 2007-2012 to
impute FLC shares in years between censuses.
15
Seasonal Agricultural Activity and Crime
farms. We estimate the number of seasonal laborers in a given month by performing the
following steps for each of the twelve FVH sectors and FLCs: first, for a given set of twelve
monthly observations within a county-year, we identify the month with the lowest employment
count, and assume this count is the number of “permanent” jobs for that county-year group
of observations. Then for a given county-month observation, the difference between total
employment in the specified sector that month and the permanent employment count is our
estimate of the number of seasonal workers in the specified FVH sector. We then sum together
seasonal employment from all twelve seasonal sectors and FLCs to yield a total seasonal
employment count. Total seasonal employment is then divided by the total labor force to
yield the seasonal share in Equation 1.
Semi-Parametric Specification
Observing that employment in seasonal agricultural sectors typically displays a distinct peak
period corresponding to harvest season, we alternatively perform a semi-parametric empirical
design that estimates how crime is affected over time relative to the peak. To do this we first
define a “treatment group” of counties that typically have high shares of seasonal agricultural
labor, and then for each of these counties identify a “peak” month where seasonal labor shares
are highest.
To define a treatment group, for each set of twelve monthly observations within a county-
year, we find the month with the highest share of seasonal agricultural employees, as defined
above. We then find the average of this yearly maximum seasonal share over all years in the
sample (1990-2016). We then include a county in the treatment group if this average maximum
16
Seasonal Agricultural Activity and Crime
share exceeds 4%, which is roughly the 95th percentile among counties that have non-zero
seasonal labor.19 This yields 47 treatment counties, which are shown in red in figure A1.
There is a high concentration of treatment counties in the Central, Salinas, and Imperial
Valleys of California and in the major apple-growing regions of Washington state.20 This
is not surprising since seasonal farm labor demands are particularly high in these regions.21
We drop counties that are below the 4% threshold but are above 1%, as these counties are
potentially impacted by seasonal labor, though this does not meaningfully change the results.
Figure 1: Treatment counties
For each treated county, we find the peak calendar month for seasonal farm labor, defined
as the month with the highest average seasonal labor share across all years in the sample.2223
19While the choice of threshold is necessarily arbitrary, results are qualitatively similar when using a thresh-
old of 2% or 6%, though somewhat weaker for the former and stronger for the latter, as expected. These
results are available upon request.
20See for example, Washington Grown. 2020. “Crops by County.” http://www.wagrown.com/crops-by-
county/ Last visited March 31, 2020.
21Seasonal farm labor demand is also high in Florida, but we dropped Florida from the analysis for two
primary reasons. The first is that Florida does not report all seasonal farm labor in the QCEW. The second
is that Florida does not have consistent records in the UCR crime data.
22The month with the highest share of agricultural labor is not necessarily always the same month within
a given county each year. We choose a single calendar month per county to simplify the analysis.
23Peak seasonal farm employment months generally run from early summer through the fall. For 44 of our
17
Seasonal Agricultural Activity and Crime
With the treatment group and peak month for each treated county defined, we estimate the
following equation:
∑6
Yimy = α + βs(λsTi) + µmy + γiy + imy, (2)
s=−4
where Ti is an indicator equal to one if county i is in the treatment group, and λs is an
indicator equal to one if the observation is s months after the peak seasonal labor month. All
other variables are defined similarly to Equation 1. βs then represents the average effect of
being s months after the peak month, where five months before the peak month is the omitted
category.
We estimate both the parametric and semi-parametric specifications to estimate the change
in violent and property crime rates, as well as the natural log of crime counts, associated
with monthly changes in the seasonal farm workforce. These are complimentary outcomes in
evaluating the overall impact on crime. Assuming agricultural activity reduces incentives to
commit crimes (by enhancing legal economic opportunities) and that migrants are less likely
to commit crimes than other groups, then our estimated effects on crime counts and crime
rates will be negative. If agricultural activity does not alter local incentives to commit crimes,
and migrants commit crimes at similar rates as non-migrants, then our estimated effects on
crime counts will be positive but effects on crime rates will be zero. If on the other hand,
seasonal workers commit crimes at lower but non-zero rates, effects on crime counts will be
positive but effects on crime rates negative. And finally, if seasonal workers commit crimes at
47 treated counties, the peak month is between May and October, with the most common being July.
18
Seasonal Agricultural Activity and Crime
higher rates or permanent residents commit more crimes in response to increased agricultural
activity, then effects on both crime counts and rates will be positive. Importantly, we should
not expect crime count effects to be negative unless non-migrant residents commit fewer crimes
in response to increased agricultural activity.
Results
We present the findings from our primary specifications in the sections that follow. First,
we present the results from the parametric specification and then the results from the semi-
parametric specification.
Baseline Parametric Specification Results
Panel A of Table A2 presents the results from estimating Equation 1 for the property crime
rate per 100,000 labor force participants, log of property crime rate, and log of property
crime count. Seasonal agricultural labor share is associated with a statistically significant
decrease in the property crime rate. The coefficient of -4.89 implies that increasing the seasonal
agricultural employment share of labor force by one additional percentage point is associated
with 4.89 fewer property crimes per 100,000 labor force participants. This reduction is roughly
1.5% of the sample median property crime rate of 335 per 100,000. We also find a statistically
significant reduction on log property crime rates, implying a one percentage point increase in
seasonal employment share is associated with a reduction in property crime rates of roughly
1%. Somewhat surprisingly given the influx of temporary laborers, we do not find evidence of
effects on property crime counts, and the point estimate is in fact negative. In sum, we find
19
Seasonal Agricultural Activity and Crime
that the increased seasonal labor force share is not associated with an increase in the number
of property crimes, and therefore the property crime rate declines as the size of the labor force
increases.
Panel B of Table A2 shows the estimated effects on violent crimes. For non-transformed
violent crime rates we find a negative and insignificant effect, though for log violent crime
rates the negative effect is significant at a 10% level.24 Unlike for property crimes, here we do
find a positive and significant increase in the count of violent crimes. The estimate of 0.004
implies that a one percentage point increase in seasonal agricultural labor share is associated
with a 0.5% increase in violent crime counts. The results for violent crime counts and rates are
not contradictory; they collectively imply that the number of violent crimes tends to increase
with the influx of seasonal farm labor, but the increase in the labor force is sufficiently large
that the measured crime rate falls.
Semi-parametric Results
Before presenting the results of the semi-parametric specification (Equation (3)), we first
demonstrate that our definition of seasonal labor described in Section indeed produces a
distinct spike in observed seasonal farm labor in our treatment counties. Figure A2 shows
the estimated coefficients from Equation (3) using seasonal agricultural labor share as the
dependent variable. The results imply that the seasonal agricultural employment share is
on average six percentage points higher relative to control counties during the peak month
24Note that the sample sizes for the log-transformed rate regressions are smaller due to observations with
zero crimes, but using this reduced sample for the non-transformed rate regressions does not meaningfully
change the result.
20
Seasonal Agricultural Activity and Crime
Table 2: Property & Violent Crime Results
A. Property Crime Results
(1) (2) (3)
Property Crime Rate Ln(Property Crime Rate) Ln(Property Crime Count)
Seasonal emp. share -4.894∗∗∗ -0.012∗∗∗ -0.004
(1.277) (0.003) (0.003)
N 838,332 814,987 814,987
B. Violent Crime Results
(1) (2) (3)
Violent Crime Rate Ln(Violent Crime Rate) Ln(Violent Crime Count)
Seasonal emp. share -0.207 -0.004∗ 0.004∗∗∗
(0.407) (0.002) (0.001)
N 814,020 765,667 765,667
Notes: Based on county-by-month data from from Uniform Crime Reports (UCR) and the Quarterly Census of Employment
and Wages (QCEW) for the years 1990-2016. The dependent variable is given in the column header. Each regression includes
month-year and county-year fixed effects. Standard errors clustered by county are given in parentheses. *, **, *** represent
significance at 10%, 5% and 1%, respectively.
(or zero “months since peak”) than this same difference five months before (the reference
category).
8
6
4
2
0
−5 −4 −3 −2 −1 0 1 2 3 4 5 6
Months since peak
Figure 2: Seasonal agricultural employment share,
semi-parametric results
Notes: Based on county-by-month data from from Uniform Crime Reports
(UCR) and the Quarterly Census of Employment and Wages (QCEW) for
the years 1990-2016. The graph plots coefficients and 95% confidence inter-
vals from estimating Equation (2) with seasonal employment share as the
dependent variable.
21
Coefficient
Seasonal Agricultural Activity and Crime
Figure A3 shows our semi-parametric results for property and violent crimes. They are
largely consistent with the parametric results shown in Table A2. For property crimes, the
effects on crime rate and log crime rate both experience a dip in the peak month. The negative
coefficient estimates are statistically significant at the 5% confidence level for the peak month
and one month before.
For log of property crime counts, effects are generally negative relative to five months
before peak and intermittently statistically significant, though the dip is much less pronounced.
While we interpret this as merely suggestive and inconclusive evidence for reductions in crime
counts, it is an interesting and unexpected result. A speculative interpretation of this finding
is that a seasonal rise in agricultural activity improves local economic conditions and reduces
the incentives to commit financially-motivated crimes. We explore this idea further in section
below.
Effects on violent crime rates are largely insignificant, though there is a positive and
statistically significant estimate for two months before peak. Similar to the results for property
crimes, the log of the violent crime rate shows a significant dip corresponding to the peak
month. The difference in results when using log of violent crime rate could indicate that
counties with more seasonal agricultural labor tend to have lower violent crime rates overall
and experience large percentage drops in violent crime rates during labor-intensive seasons.
For log violent crime counts, there are no statistically significant effects. This is somewhat in
contrast to the results using the parametric specification, which showed a small, statistically
significant increase in the log violent crime count associated with increased seasonal farm
employment. Overall, the semi-parametric results are consistent with the parametric results
22
Seasonal Agricultural Activity and Crime
Property Crime Rate Ln(Property crime rate) Ln(Property crime count)
20 .05 .05
0 0
0
−20 −.05
−.05
−40 −.1
−.1
−60 −.15
−80 −.2 −.15
−5 −4 −3 −2 −1 0 1 2 3 4 5 6 −5 −4 −3 −2 −1 0 1 2 3 4 5 6 −5 −4 −3 −2 −1 0 1 2 3 4 5 6
Months since peak Months since peak Months since peak
Violent crime rate Ln(Violent crime rate) Ln(Violent crime count)
15 .04 .06
.02 .04
10
0 .02
5
−.02 0
0
−.04 −.02
−5
−.06 −.04
−5 −4 −3 −2 −1 0 1 2 3 4 5 6 −5 −4 −3 −2 −1 0 1 2 3 4 5 6 −5 −4 −3 −2 −1 0 1 2 3 4 5 6
Months since peak Months since peak Months since peak
Figure 3: Property & violent crime, semi-parametric results
Notes: Based on county-by-month data from from Uniform Crime Reports (UCR) and the Quarterly Census of Employment
and Wages (QCEW) for the years 1990-2016. The graph plots coefficients and 95% confidence intervals from estimating
Equation (2). The dependent variables are indicated in the figure headers. Standard errors are clustered by county.
in Table A2, with the exception of finding no statistically significant effects on violent crime
counts.
Extension: Year-by-Year Estimates
Because the identifying variation in our baseline specification shown in Equation (1) is within-
year variation, we can identify marginal effects separately for each year, and test whether
effects change from 1990-2016. This may be consequential since migration of farm workers
declined significantly over this time, due in part to demographic changes in the farm workforce
(Fan et al. 2015). We do this for each of our six main outcomes in figure A4. Estimates are
generally consistent with the overall effects shown in Table A2, and fairly trendless throughout
23
Coefficient Coefficient
Coefficient Coefficient
Coefficient Coefficient
Seasonal Agricultural Activity and Crime
the sample period. One exception is that property crime effects are trending down in the first
six years of the sample (and non-transformed property crime rate effects gently trend up
thereafter, though remain negative throughout). Also, violent crime effects experience a large
positive spike in 2016, the last year of the sample.
Mechanisms
As discussed in the Introduction, there are at least two reasons to think that seasonal agricul-
tural activity reduces crime. First, if increased agricultural activity creates non-agricultural
sectoral spillovers and improves labor market opportunities, this possibly raises the opportu-
nity cost of engaging in illegal activity and reduces crime. Second, according to some esti-
mates, immigrants are less likely to commit crimes than non-immigrants (Butcher and Piehl
2007; Stowell et al. 2009; Wadsworth 2010; Baker 2015), and the large majority of migrant
farm workers are immigrants. But these effects may be partially offset by the effects of social
disorganization and the inward migration of people possibly prone to committing crimes (as
discussed earlier). In addition, non-migrant residents may commit crimes against immigrants
either because they are perceived to lack any legal power or means of retribution, or because
they view migrants as threats to their security or well-being and react preemptively. While
explicitly testing the viability of each of these theories is beyond the scope of this paper, here
we explore whether these proposed mechanisms are consistent with available data.
24
Seasonal Agricultural Activity and Crime
Property crime rate Violent crime rate
5 10
0
5
−5
0
−10
−15 −5
1990 1995 2000 2005 2010 2015 1990 1995 2000 2005 2010 2015
year year
Ln(Property crime rate) Ln(Violent crime rate)
.01 .02
0 .01
−.01 0
−.01
−.02
−.02
−.03
1990 1995 2000 2005 2010 2015 1990 1995 2000 2005 2010 2015
year year
Ln(Property crime count) Ln(Violent crime count)
.01 .03
.02
0
.01
−.01
0
−.02
−.01
1990 1995 2000 2005 2010 2015 1990 1995 2000 2005 2010 2015
year year
Figure 4: Crime effects by year
Notes: Based on county-by-month data from from Uniform Crime Reports (UCR) and the Quarterly Census of Employment
and Wages (QCEW). The graphs plot coefficients and 95% confidence intervals from estimating Equation (1) separately for each
year from 1990-2016. The dependent variables are indicated in the figure headers below each panel. Standard errors are
clustered by county.
25
Coefficient Coefficient Coefficient
Coefficient Coefficient Coefficient
Seasonal Agricultural Activity and Crime
Economic Spillovers
First, we investigate whether seasonal agricultural activity creates short-term economic spillovers
into non-farm sectors that could reduce the propensity of permanent residents to commit
crimes. There are two necessary conditions that must hold for this “economic” channel to
explain our core set of results. The first is that broad labor market conditions improve as the
seasonal farm-labor share increases. The second is that the non-migrant crime rate falls in
response to seasonal agricultural activity.25 To test the first condition, we re-estimate equa-
tion (2) with total non-agricultural employment as the dependent variable. The results are
consistent with the idea that seasonal agricultural activity generates local economic spillovers.
Specifically, figure A5 shows that non-agricultural employment peaks in tandem with peak sea-
sonal agricultural employment. The fact that non-agricultural employment follows the same
inverted “V” pattern as the seasonal employment share could be indicative of increased popu-
lation driving service sector employment, upstream or downstream linkages to the agricultural
sector, or both.
Ethnicity of Arrestees and Victims
The second necessary condition for the economic channel to hold is that seasonal agricultural
activity reduces the non-migrant crime rate. While we do not observe a non-migrant crime
rate, we can use crime data that includes information on ethnicity to shed light on this
mechanism. Since 95.3% of migrant FVH workers are Hispanic according to NAWS, the non-
25The extant literature offers an abundance of evidence that economic improvement reduces the incentive to
commit property crimes (Lin 2008; Baker 2015; Freedman and Owens 2018; Carr and Packham 2019; Watson,
Guettabi, and Reimer 2019), but it has little to say about the local economic effects of seasonal agricultural
activity.
26
Seasonal Agricultural Activity and Crime
.1
.05
0
−5 −4 −3 −2 −1 0 1 2 3 4 5 6
Months since peak
Figure 5: Non-agricultural employment
Notes: Based on county-by-month data from Uniform Crime Reports (UCR) and the
Quarterly Census of Employment and Wages (QCEW) for the years 1990-2016. The
graph plots coefficients and 95% confidence intervals from estimating Equation (2) with
the natural log of non-agricultural employment as the dependent variable. Standard
errors are clustered by county.
Hispanic crime rate can serve as a rough proxy for the crime rate of a subset of the population
that is non-migrant (only a subset because of course there are non-migrant Hispanics as well).
To do this, we use data from the National Incident-Based Reporting System (NIBRS), which
provides far more detail on individual crimes than UCR, including the ethnicity (if known)
of arrestees and victims. However, since precinct participation is voluntary and imposes a
heavier reporting burden, it has much lower coverage than UCR. As of 2016, the last year of
our data, precincts representing roughly 31% of the US population participated in NIBRS. We
use data going back to 2000, as coverage gets especially sparse going further back (about 15%
of the population is covered in 2000). These results should therefore be viewed with caution
but do provide interesting suggestive evidence regarding the mechanisms of our negative crime
effects.
27
Coefficient
Seasonal Agricultural Activity and Crime
We identify the count of crimes with at least one Hispanic arrestee and the count with
at least one non-Hispanic arrestee at the county-month level. We also identify the count of
crimes with at least one Hispanic victim, and the count with at least one non-Hispanic victim.
We then run our main specification with these counts or corresponding crime rates as the
dependent variable.26,27
For completeness, we measure crimes in four ways: 1) crime rate (crimes per labor force),
2) the crime count, 3) the natural log of the crime count, and 4) the inverse hyperbolic sine of
the crime count to account for observations with zero crimes.28 We find that, while increasing
the seasonal employment share has no effect on aggregate crimes (see the first row of Table
A3), it causes a significant reduction in both the number and rate of crimes committed by
non-Hispanics. We also find that the seasonal employment share is negatively associated with
the rate of non-Hispanic victimization. Given that such a small fraction of migrant farm
workers are non-Hispanic, this is consistent with the idea that the positive economic spillovers
documented above reduce the incentive to engage in illegal activity.
In contrast to the results for non-Hispanics, increasing the seasonal employment share
increases the number of Hispanic victims, but not the rate of Hispanic victimization. While
this may simply reflect an increase in the local Hispanic population, it is worth noting that
we do not document a corresponding increase in the aggregate number of Hispanic arrests.
26In this context, a crime rate is measured as the number of crimes committed by people of Hispanic (or
non-Hispanic) ethnicity, relative to the labor force.
27As with the UCR data, we drop all precincts in a given calendar year that do not have full reporting for
the year. Any county-year combination with no fully reporting precincts are not included in this analysis.
28Zero-crime observations are much more common in this analysis than when using UCR data for a few
reasons. First, we only have data for ethnicity of an arrestee, and roughly half of crimes in NIBRS do not
result in an arrest. Second, the ethnicity of the arrestee/victim is often not known or otherwise not given.
Third, many parts of the country have low Hispanic population.
28
Seasonal Agricultural Activity and Crime
This is important given that a 1 percentage point increase in the seasonal employment share
amounts to roughly 437 additional workers. Even if migrant farm workers committed crimes
at a relatively low rate, one might expect to find the number of crimes committed by this
group increases during picking season. These results are consistent with the idea that migrant
farm workers commit very few crimes such that, even when their population swells, no new
crimes can be detected.
We do however find that the number of Hispanic victims increases in response to an
increase in the seasonal employment share of the labor force. In the last three panels of Table
A3, we analyze crimes that have at least one Hispanic victim and at least one arrestee that
is Hispanic, non-Hispanic, or of unknown ethnicity. The effects on crimes against Hispanics
appear to be primarily driven by Hispanic offenders. However, depending on how crime counts
are measured, there is evidence that some of these crimes are committed by non-Hispanics and
people of unknown ethnicity as well. Taken together, these data are consistent with the idea
that agricultural activity both creates economic opportunities that reduce crime and attracts
a migrant population that is less prone to criminal activity than other groups, at least during
the labor-intensive season of agricultural production. The data is also consistent with the idea
that migrant farm workers are victimized by both Hispanic and non-Hispanic people.
Effects of Corn Harvests
To further disentangle the economic and migration mechanisms we analyze crime effects of
labor shares in corn farming. We choose corn because it is a crop that is grown extensively
in the United States, has a common harvest season typically in the Fall, but is not nearly as
29
Seasonal Agricultural Activity and Crime
Table 3: Effects by Ethnicity of Reported Offender and Victim
(1) (2) (3) (4)
Crime Rate Crime Count Ln(Crime Count) IHS(Crime Count)
All Crimes
Seasonal Emp. Share -1.962 0.677 0.002 0.002
(3.114) (1.156) (0.003) (0.003)
N 251018 251018 246646 251018
Non-Hispanic Arrests
Seasonal Emp. Share -1.523∗∗∗ -0.227∗∗∗ -0.005∗∗ -0.005∗∗
(0.358) (0.068) (0.002) (0.002)
N 251018 251018 208520 251018
Hispanic Arrests
Seasonal Emp. Share -0.061 0.023 0.006 0.007∗
(0.236) (0.059) (0.004) (0.004)
N 251018 251018 101998 251018
Non-Hispanic Victim
Seasonal Emp. Share -4.967∗∗∗ -0.921∗ 0.003 0.002
(1.441) (0.490) (0.003) (0.003)
N 251018 251018 215258 251018
Hispanic Victim
Seasonal Emp. Share 0.121 0.161∗∗ 0.009∗ 0.013∗∗
(0.278) (0.076) (0.005) (0.005)
N 251018 251018 109997 251018
Unknown Off|Hisp Victim
Seasonal Emp. Share 0.034 0.090 0.005 0.010∗∗
(0.210) (0.073) (0.004) (0.005)
N 246646 246646 99506 246646
Non Hisp Off | Hisp Victim
Seasonal Emp. Share -0.018 0.009 0.008∗∗∗ 0.003
(0.036) (0.014) (0.002) (0.003)
N 246646 246646 50623 246646
Hisp Off | Hisp Victim
Seasonal Emp. Share 0.113 0.065∗ 0.009∗∗ 0.012∗∗∗
(0.087) (0.035) (0.004) (0.004)
N 246646 246646 51257 246646
Note: based on data from the National Incident-Based Reporting System (NIBRS) from 2000-
2016. The type of crime for the dependent variable is given in the panel title, and the specific
transformation of the type of crime is given in the column header. The “All Crimes” includes
crimes for which no arrest is made, while all other panels are based only on crimes that include
an arrest. Each regression includes month-year and county-year fixed effects. Standard errors
clustered by county are given in parentheses. *, **, *** represent significance at 10%, 5% and 1%,
respectively.
labor-intensive as the harvest of FVH crops and is not grown in large quantity in counties with
high FVH employment. Thus, we should not expect corn production activities to attract many
migrant workers or correlate with the timing and location of labor-intensive FVH activities.
30
Seasonal Agricultural Activity and Crime
Corn harvest therefore provides variation in farm revenue while holding (roughly) constant
the migrant farm worker labor share (which does increase during harvest season but at about
one tenth of the rate as for FVH sectors). If increased agricultural revenue translates to broad
economic improvements and enhanced local labor market opportunities, we expect to find a
negative crime effect of corn farming. However, it is also possible that economic spillovers
originate from farm worker income, and in this case we should not expect to find significant
crime effects of corn farming. The results are shown in Table A4. The estimates are all
negative but statistically insignificantly different from zero. This suggests that the economic
effects of seasonal agricultural activity on reduced crime are specifically associated with the
enhanced employment effects of labor-intensive crops rather than seasonal increases in farm
revenues alone.
Table 4: Crimes Results: Corn Employment Shares
Crime Rate Ln(Crime Rate) Ln(Crime Count)
Property Crimes
Seasonal Corn Emp. Share -2.665 -0.006 -0.005
(2.377) (0.012) (0.012)
N 838332 814987 814987
Violent Crimes
Seasonal Corn Emp. Share -1.564 -0.019 -0.019
(1.027) (0.015) (0.015)
N 814020 765667 765667
Notes: Based on county-by-month data from Uniform Crime Reports (UCR)
and the Quarterly Census of Employment and Wages (QCEW) for the years
1990-2016. The table reports estimates when using seasonal corn employment
labor force shares as the explanatory variable of interest. The dependent
variable is given in the column header. Each regression includes month-
year and county-year fixed effects. Standard errors clustered by county are
given in parentheses. *, **, *** represent significance at 10%, 5% and 1%,
respectively.
31
Seasonal Agricultural Activity and Crime
Discussion
It is interesting to compare our estimates to those from analyses of energy booms. (James
and Smith 2017) find that the shale energy boom increased rates of property crimes as well
as some violent crimes (aggravated assault) by roughly 10%-20%. Consonant results are
documented by Gourley and Madonia (2018) and Komarek (2018) in their analyses of the
Colorado and Pennsylvania shale booms, respectively. Why do our results differ from these
related estimates? The answer may be that energy booms appear to attract, rather than
create, criminal activity. In fact, Street (2018) finds that the North Dakota oil boom decreased
the crime rate among residents that lived in the state prior to the boom (consistent with the
idea that improved economic conditions raise the opportunity cost of committing crimes)
whereas James and Smith (2017) find that the shale boom had especially large (positive)
effects on aggregate criminal activity in North Dakota. Considered jointly, the observed rise
in criminal activity in boom towns appears to reflect the inward migration of criminally-prone
individuals. Our results are consistent with previous findings that show that economic activity
can reduce criminal activity among non-migrant residents (Street 2018) and the literature that
finds that foreign-born immigrants are no more likely than natural-born citizens to commit
crimes on average (Reid et al. 2005; Butcher and Piehl 2007; Stowell et al. 2009; Wadsworth
2010; Bianchi, Buonanno, and Pinotti 2012; Bell, Fasani, and Machin 2013; Chalfin 2014;
Baker 2015).
32
Seasonal Agricultural Activity and Crime
Robustness
We perform several robustness checks and present the results in the online supplementary
appendix. The appendix includes complete explanations and discussions of each robustness
check, but we briefly summarize them below.
First, we adjust the labor force denominator to account for seasonal farm workers who
might remain in the local region but drop out of the labor force. We use data from the NAWS
to estimate the percentage of seasonal farmworkers who are settled in a single location and
the mean share of weeks each year that they do not work in any labor sector, and we adjust
the labor force denominator accordingly. Results are qualitatively similar to the main results.
Second, one might be concerned that employees of FLCs travel to distant farms and do not
necessarily work in the county of their employer’s address. Thus, we compute the number of
FLC workers in each county using the county share of state expenses for contract labor in
the Agricultural Census and multiply this share by the total number of FLC workers in the
state each month. We use this alternative measure of FLC workers to compute total number
of seasonal workers for each observation. Third, we examine the effects of H-2A agricultural
guest worker employment shares on crime rates. Fourth, we repeat our analysis when dropping
counties with any suppressed FVH employment data. Fifth, we repeat our analysis using the
inverse hyperbolic sine transformation in place of the natural log. Sixth, we repeat our analysis
dropping observations from the election year 2016. Seventh, we repeat our analysis weighting
for the total labor force using weighted least squares. Eighth, we repeat our semi-parametric
specification using only counties that rank among the top 5% for average employment in corn
33
Seasonal Agricultural Activity and Crime
production as the control group. Our baseline conclusions prove to be quite robust.
Next, in the appendix we investigate whether Hispanics are more or less likely than non-
Hispanics to report being victim to different types of crimes according to the National Crime
Victimization Survey. Our findings show that Hispanics who were victim to violent crimes (or
attempts) were 1.2 percentage points more likely to report the crime than non-Hispanic vic-
tims. However, Hispanics who were victims to personal thefts were 4.1 percentage points less
likely to report the crime than non-Hispanic victims. Both of these differences are statistically
significant, though qualitatively quite small, and we find no statistically significant difference
in the probability of reporting burglary or motor vehicle theft. If seasonal farm workers are
more or less likely to report crimes than other residents, this could be problematic for causal
identification in our analysis. Nevertheless, the findings from the NIBRS analysis in table
A3 show that seasonal farm labor is associated with a significant decrease in non-Hispanic
crime victimization. Since few non-Hispanics are seasonal farmworkers, this result cannot be
explained by possible differences in the probability that farmworkers report crimes, thus vali-
dating the credibility of our main findings even though there is a possibility of relatively small
measurement error. Nevertheless, we note that crimes committed against farmworkers, who
may have significant obstacles to press charges against offenders29 are of utmost importance,
even if these crimes mostly go undetected by the communities who host seasonal farmworkers
and cannot be quantified in available data.
29see for example Soriano (2020).
34
Seasonal Agricultural Activity and Crime
Conclusion
We estimate the effect of labor-intensive seasonal agricultural activity on crime, and to the
best of our knowledge, we are the first to do so. Our analysis is motivated, in part, by
the observation that many Americans think immigrants—and undocumented immigrants in
particular—are more likely to commit crimes than natural-born citizens, and that many Amer-
icans associate seasonal farm labor with drudgery30 and crime more generally.
We observe both criminal and labor-intensive agricultural activity—measured as the sea-
sonal FVH labor share—by month and U.S. county. The richness of our data allows us to
leverage seasonal variation in agricultural activity while controlling for any unobserved factors
that are fixed within a county in a given year. We find that increased seasonal FVH labor
employment reduces the property crime rate, and has little or no effect on the violent crime
rate. We also find that it reduces both the rate of crime and of victimization among the non-
Hispanic population (our proxy for the non-migrant farm worker population). Taken together,
our results are consistent with idea that agricultural activity improves local economic condi-
tions, which we provide evidence of, and that this reduces the incentive to commit property
crimes (Lin 2008; Freedman and Owens 2016; Carr and Packham 2019; Watson, Guettabi, and
Reimer 2019). Contrary to potential fears, we find that labor intensive agricultural activity
is not associated with increased violent or property crimes and thus, concerns to the contrary
are largely unwarranted.
One caveat to our analyses and conclusion is that we do not know the extent to which
crimes committed against migrant farm workers go unreported. As such, it is possible that
30See for example, (Friendly, Murrow, and Lowe 1960; Steinbeck 1939)
35
Seasonal Agricultural Activity and Crime
seasonal agricultural activity is associated with crimes that are undetected in our data. For
example, there is anecdotal evidence suggesting that female farm workers are often victims of
sexual assault, and the problem may be worse for migrant female workers (Soriano 2020). We
hope that our work will help motivate others to explore this issue further.
36
Seasonal Agricultural Activity and Crime
References
Baker, S. 2015. “Effects of immigrant legalization on crime.” American Economic Review
105:210–213.
Bell, B., F. Fasani, and S. Machin. 2013. “Crime and Immigration: Evidence from Large
Immigrant Waves.” The Review of Economics and Statistics 95:1278–1290.
Bianchi, M., P. Buonanno, and P. Pinotti. 2012. “Do Immigrants Cause Crime?” Journal of
the European Economic Association 10:1318–1347.
Blakeslee, D., and R. Fishman. 2018. “Weather shocks, agriculture, and crime evidence from
India.” Journal of Human Resources 53:750–782.
Borjas, G.J., J. Grogger, and G.H. Hanson. 2010. “Immigration and the economic status of
African-American men.” Economica 77:255–282.
Butcher, K.F., and A.M. Piehl. 2007. “Why are Immigrants’ Incarceration Rates so Low? Ev-
idence on Selective Immigration, Deterrence, and Deportation.” Working Paper No. 13229,
National Bureau of Economic Research, July.
Cadena, B.C., and B.K. Kovak. 2016. “Immigrants equilibrate local labor markets: Evidence
from the Great Recession.” American Economic Journal: Applied Economics 8:257–90.
Campaniello, N., and E. Gavrilova. 2018. “Uncovering the gender participation gap in crime.”
European Economic Review 109:289–304.
Carr, J.B., and A. Packham. 2019. “SNAP benefits and crime: Evidence from changing dis-
bursement schedules.” Review of Economics and Statistics 101:310–325.
Chalfin, A. 2014. “What is the contribution of Mexican immigration to US crime rates?
Evidence from rainfall shocks in Mexico.” American Law and Economics Review 16:220–
268.
Department of Labor, Employment & Training Administration. 2017. “National Agricultural
Workers Survey.”
Fan, M., S. Gabbard, A. Alves Pena, and J.M. Perloff. 2015. “Why Do Fewer Agricultural
Workers Migrate Now?” American Journal of Agricultural Economics 97:665–679.
Foley, C. 2011. “Welfare payments and crime.” The Review of Economics and Statistics 93:97–
112.
Foster, A.D., and M.R. Rosenzweig. 2004. “Agricultural Productivity Growth, Rural Economic
Diversity, and Economic Reforms: India, 1970-2000.” Economic Devolopment and Cultural
Change 52:509–542.
37
Seasonal Agricultural Activity and Crime
Freedman, M., and E. Owens. 2018. “Immigration, employment opportunities, and criminal
activity.” American Economic Journal: Economic Policy 10:117–151.
—. 2016. “Your friends and neighbors: Localized economic development and criminal activity.”
Review of Economics and Statistics 98:233–253.
Freudenburg, W.R. 1986. “The density of acquaintanceship: an overlooked variable in com-
munity research?” American Journal of Sociology 92:27–63.
Friendly, F., E.R. Murrow, and D. Lowe. 1960. “Harvest of Shame.” CBS Documentary.
Gould, E.D., B.A. Weinberg, and D.B. Mustard. 2002. “Crime rates and local labor market
opportunities in the United States: 1979-1997.” Review of Economics and Statistics 84:45–
61.
Gourley, P., and G. Madonia. 2018. “Resource booms and crime: Evidence from oil and gas
production in Colorado.” Resource and Energy Economics 54:37–52.
Hornbeck, R., and P. Keskin. 2015. “Does Agriculture Generate Local Economic Spillovers?
Short-Run and Long-Run Evidence from the Ogallala Aquifer.” American Economic Jour-
nal: Economic Policy 7:192–213.
James, A., and B. Smith. 2017. “There will be blood: Crime rates in shale-rich U.S. counties.”
Journal of Environmental Economics and Management 84:125 – 152.
Kelly, M. 2000. “Inequality and crime.” Review of economics and Statistics 82:530–539.
Komarek, T. 2018. “Crime and natural resource booms: Evidence from unconventional natural
gas production.” The Annals of Regional Science 61:113–137.
Lin, M.J. 2008. “Does unemployment increase crime? Evidence from US data 1974-2000.”
Journal of Human Resources 43:413–436.
Lochner, L., and E. Moretti. 2004. “The effect of education on crime: Evidence from prison
inmates, arrests, and self-reports.” American economic review 94:155–189.
Martin, P. 2017. “The H-2A Farm Guestworker Program Is Expanding Rapidly: Here Are the
Numbers You Need to Know”.” Working paper, Economic Policy Institute.
Martin, P., M. Fix, and J.E. Taylor. 2006. The New Rural Poverty . Washington, D.C.: The
Urban Institute Press.
Mohan, G. 2017. “Desired for their Labor, Rejected as Neighbors. Farmworkers in California
Face Hostile Communities.” Los Angeles Times , pp. .
Moyce, S.C., and M. Schenker. 2018. “Migrant workers and their occupational health and
safety.” Annual review of public health 39:351–365.
38
Seasonal Agricultural Activity and Crime
Patterson, P.E. 2015. “Southcentral Idaho: Magic Valley: Russet Burbank Potatoes with
Fumigation: Production and Storage Costs.” 2015 Costs and Returns Estimate No. EBB3-
Po3-15, University of Idaho, College of Agricultural and Life Sciences.
Reid, L.W., H.E. Weiss, R.M. Adelman, and C. Jaret. 2005. “The immigration–crime rela-
tionship: Evidence across US metropolitan areas.” Social science research 34:757–780.
Roberts, N. 2019. “Undocumented Immigrants Quietly Pay Billions into So-
cial Security and Receive No Benefits.” Marketplace, January 28, pp. ,
https://www.marketplace.org/2019/01/28/undocumented-immigrants-quietly-pay-billions-
social-security-and-receive-no/. Accessed March 18, 2021.
Roka, F.M., S. Simnitt, and D. Farnsworth. 2017. “Pre-employment costs associated with
H-2A agricultural workers and the effects of the “60-minute rule”.” International Food and
Agribusiness Management Review 20:335–346.
Rural Migration News. 2020. “A Tale of Two States: Farm Labor in CA and FL.”
Sampson, R.J., and W.B. Groves. 1989. “Community structure and crime: Testing social-
disorganization theory.” American journal of sociology 94:774–802.
Soriano, S. 2020. “The Rape Crisis among California’s Farm Workers.” Capitol Weekly: Cov-
ering California Government and Politics January 9.
Spenkuch, J. 2014. “Understanding the impact of immigration on crime.” American Law and
Economics Review 16:177–219.
Steinbeck, J. 1939. The Grapes of Wrath. New York: Penguin Books.
Stowell, J., S. Messner, K. McGeever, and L. Raffalovich. 2009. “Immigration and the recent
violent crime drop in the United States: A pooled, cross-sectional time-series analysis of
metropolitan areas.” Criminology 47:889–928.
Street, B. 2018. “The impact of economic opportunity on criminal behavior: Evidence from
the fracking boom.” Working Paper , pp. .
Taylor, J.E., and D. Charlton. 2018. The Farm Labor Problem: A Global Perspective. Elsevier
Academic Press.
Taylor, J.E., and D. Thilmany. 1993. “Worker Turnover, Farm Labor Contractors, and IRCA’s
Impact on the California Farm Labor Market.” American Journal of Agricultural Economics
75:350–360.
Wadsworth, T. 2010. “Is immigration responsible for the crime drop? An assessment of the
influence of immigration on changes in violent crime between 1990 and 2000.” Social Science
Quarterly 91:531–553.
39
Seasonal Agricultural Activity and Crime
Wallis, E. 2019. “Immigrant Female Farmworkers in California Need Greater Protection
Against Sexual Harrasment.” CSW Policy Briefs 2019.
Watson, B., M. Guettabi, and M. Reimer. 2019. “Universal cash and crime.” Review of Eco-
nomics and Statistics , pp. 1–45.
Weber, J.G., C. Wall, J. Brown, and T. Hertz. 2015. “Crop Prices, Agricultural Revenues,
and the Rural Economy1.” Applied Economic Perspectives and Policy 37:459–476.
Zahniser, S., J.E. Taylor, T. Hertz, and D. Charlton. 2018. “Farm Labor Markets in the United
States and Mexico Poze Challenges for U.S. Agriculture.” Working paper, U.S. Department
of Agriculture, Economic Research Service, Nov.
40
Seasonal Agricultural Activity and Crime
Appendix
Descriptive Statistics from the National Agricultural Workers Survey
We summarize some of the notable characteristics of FVH workers interviewed in the National
Agricultural Workers Survey (NAWS), which is designed to be nationally and regionally rep-
resentative of crop workers in the United States. The NAWS a random-sample survey that
interviews crop workers at their place of work and asks questions about income, hours of work,
migration, job mobility, and immigration status, among other questions.
Poverty rates are high among FVH workers. According to the NAWS, 71.1 percent of
FVH workers in 2016 who reported working the previous year had income $25,000 in 2015,
and 33.4 percent had income below $15,000. Figure A6 shows the income distribution of
workers interviewed in 2016 who reported working the previous year.
Figure A1: Income of fruit vegetable, and horticultural workers (2015)
Many FVH workers do not work year-round. FVH workers reported that they did not
41
Seasonal Agricultural Activity and Crime
work (in any job, whether farm or non-farm) for an average of 10.6 weeks of the year in
2016. Figure A7 shows mean weeks per year that FVH workers report that they did not work
between 1990-2016. From 1990-2016, the mean weeks not worked each year is 9.8.
Figure A2: Mean weeks per year that fruit vegetable, and horticultural workers did not work
Some FVH workers migrate from farm to farm (follow-the-crop workers), and some migrate
back and forth between their place of work in the United States and another location that
they call their home, often in Mexico (shuttle migrants). Panel (a) of figure A8 shows the
migratory (i.e. either follow-the-crop or shuttle migration) share of FVH farm workforce each
year from 1990-2016. There is a clear downward trend in the migratory share of FVH workers.
Fan et al. (2015) conclude that the steep decline in the share of crop workers who migrate
since 1998 can be attributed to both demographic changes in the workforce (which they find
are responsible for about one-third of the decline) and structural changes in the U.S. and
Mexican economies (which they find are responsible for about two-thirds of the decline).
Panel (b) of figure A8 plots the share of FVH workers who are foreign-born each year.
From 1990-2016, 73 percent of FVH workers were born in Mexico, 20.1 percent were born in
42
Seasonal Agricultural Activity and Crime
the United States, and 3.8 percent were born in Central America. Most FVH workers are not
native English speakers. From 1990-2016, 79.9 percent of NAWS respondents said that their
most comfortable language was Spanish while 17.3 percent responded that English was their
most comfortable language.
Panel (c) of figure A8 plots the share of FVH workers who have no nuclear family members
accompanying them. There is a clear downward trend in the share of unaccompanied farm
workers from a peak over 65 percent in the late 1990s to roughly 40 percent in 2016.
Robustness Checks
Labor Force Denominator Adjustment
As discussed in the data section, a key component of our identification strategy is the use of
changes in labor force as a proxy for changes in population. However, if seasonal agricultural
workers who are settled in the same county where they perform seasonal work do not work at
all for some share of the year, they may not be counted in the labor force in some months when
they are still residing in the county. Crime rates for those months will be artificially inflated,
since the denominator will be artificially small. To address this we obtain rough estimates
of how much our denominator is spuriously reduced by this issue through the NAWS, which
asks respondents whether they are settled in their place of employment and how many weeks
they did not work in the previous year. We use these to create region-by-year31 estimates of
the percentage of the FVH workforce that is settled and average weeks per year not working.
For each observation we calculate an adjusted labor force estimate using the following
31NAWS reports data at the regional level. There are six NAWS regions, and the NAWS is intended to be
regionally and nationally representative.
43
Seasonal Agricultural Activity and Crime
(a) Migratory (b) Foreign-born
(c) Do not live with nuclear family
Figure A3: Characteristics of fruit vegetable, and horticultural workers
Notes: Panel (a) gives the share of FVH workers who are migratory (either follow-the-crop or
shuttle migration). Panel (b) gives the share of FVH workers born outside the United States.
Panel (c) gives the share of FVH workers who do not live with their nuclear family. Data
taken from the National Agricultural Workers Survey (NAWS).
44
Seasonal Agricultural Activity and Crime
equation:
Adjusted LFimy = LFimy + (max seasiy − current seasimy) ∗ pct settledry ∗ (weeks nwry/52)
(3)
LFimy is the original labor force estimate provided by the BLS in county i, month m, and year
y. max seasiy − current seasimy is our estimate of seasonal workers who are not currently
working their seasonal jobs. We make the assumption that the total number of individual
seasonal farm workers within a county-year is the number of seasonal workers in the peak
month for that year (this is max seasiy). We subtract from max seasiy the number of seasonal
workers in month m year y, so the difference is seasonal workers not in their seasonal jobs.
pct settledry is the NAWS region-by-year estimate of the percentage of FVH workers who are
settled in the place of their seasonal employment and report not working at least one week
of the year. weeks nwry/52 is the NAWS region-by-year average number of weeks spent not
working at all among settled FVH workers divided by 52. We thus add to the original BLS
labor force an estimate of the number of seasonal workers who are currently in the same
county but not working, as these are the workers who will be under-counted in the original
labor force figure.
We rerun our baseline regressions from Table 2 with crime rates and seasonal worker
share calculated with these adjusted labor force estimates.32 These results are shown in
Table A5. Similar to our main findings, we find a statistically significant negative association
32The crime count variable is unchanged by this adjustment, but we still rerun the count regressions on the
adjusted seasonal farm labor share variable.
45
Seasonal Agricultural Activity and Crime
between seasonal farm labor shares and property crime rates in panel A,33 though the point
estimate is slightly smaller in magnitude, as expected. We find a statistically significant
positive association with logged violent crime counts, but no significant associations with
violent crime rates, again similarly to our baseline results.
Table A1: Property & Violent Crime Results, Adjusting for Seasonal Workers Settled but not Working
A. Property Crime Results
(1) (2) (3)
Property Crime Rate Ln(Property Crime Rate) Ln(Property Crime Count)
Adjusted Seasonal emp. share -4.553∗∗∗ -0.012∗∗∗ -0.005
(1.553) (0.004) (0.003)
N 837,084 813,747 813,747
B. Violent Crime Results
(1) (2) (3)
Violent Crime Rate Ln(Violent Crime Rate) Ln(Violent Crime Count)
Adjusted Seasonal emp. share 0.370 -0.001 0.006∗∗∗
(0.507) (0.002) (0.001)
N 812,772 764,427 764,427
Notes: Based on county-by-month data from Uniform Crime Reports (UCR) and the Quarterly Census of Employment and Wages
(QCEW) for the years 1990-2016. The table reports estimates when adjusting the measure of labor force as described in the text. The
dependent variable is given in the column header. Each regression includes month-year and county-year fixed effects. Standard errors
clustered by county are given in parentheses. *, **, *** represent significance at 10%, 5% and 1%, respectively.
Contract Labor
In 2012, workers hired through Farm Labor Contractors (FLCs) accounted for 19 percent of
total labor expenditures, 38 percent of all fruit and nut labor expenditures, and 31 percent
of vegetable and melon labor expenditures (Zahniser et al. 2018). FLCs are intermediaries
who hire farm workers directly and contract labor to farms. FLCs may reduce the costs of
recruiting workers directly, especially if there are substantial frictions in the labor markets
that prevent the matching of workers to farms during peak seasonal labor demands. Growers
33Note that the sample sizes are slightly smaller in this table compared to our baseline regressions, due to
the exclusion of Alaska and Hawaii in the NAWS.
46
Seasonal Agricultural Activity and Crime
may also prefer to hire workers through FLCs to manage and mitigate risks associated with
hiring unauthorized immigrants. Taylor and Thilmany (1993) find suggestive evidence that
FLCs may be willing to take on more risk than farm employers since they can more easily hide
from immigration enforcement. FLCs are constantly transporting workers from one location
to another and it is relatively easy for them to close their business and reopen under a new
name.
Although the QCEW records the number of FLC employees per county each month, FLC
employees may not work or reside at the address of the FLC. Surveys conducted with FLCs
in Florida indicate that USDA Department of Labor regulations implemented in 2012 to limit
the transport of H-2A workers to within 60 miles of their housing severely restricted the
movement and profitability of FLCs who hired H-2A workers (Roka, Simnitt, and Farnsworth
2017). To account for potential measurement error in the number of seasonal farm employees
located in each county, we construct an alternative measure in which we impute the num-
ber of contracted workers by county-year using data on contracted labor expenses from the
Agricultural Censuses, which we have for every five years from 1987-2017.
For a given census year, we calculate each county’s share of its state’s total contract labor
expense. Since the agricultural census is every five years, we impute expense shares for missing
years by linear interpolation. We then find for each month the total number of QCEW FLC
employees for a state, and assign each county a share of these employees according to its share
of contract labor expense. We then include this alternative measure of contract laborers in
our estimate of total seasonal agricultural laborers, rather than the county-level FLC counts
from QCEW. This method assumes that the contract employees work in the same state as the
47
Seasonal Agricultural Activity and Crime
address of the FLC, but this is subject to relatively small measurement error. The correlation
between the QCEW figure and our alternative measure is .95.
Unsurprisingly, given this high correlation, the results using the alternative measure shown
in Table A6 are very similar to our baseline results.34
Table A2: Property and Violent Crime Results, Alternative Measure of Contract Workers
A. Property Crime Results
(1) (2) (3)
Property Crime Rate Ln(Property Crime Rate) Ln(Property Crime Count)
Seasonal emp. share -5.018∗∗∗ -0.013∗∗∗ -0.004
(1.347) (0.003) (0.003)
N 834552 811371 811371
B. Violent Crime Results
(1) (2) (3)
Violent Crime Rate Ln(Violent Crime Rate) Ln(Violent Crime Count)
Seasonal emp. share -0.122 -0.004∗∗ 0.005∗∗∗
(0.408) (0.002) (0.001)
N 810636 762751 762751
Notes: Based on county-by-month data from Uniform Crime Reports (UCR) and the Quarterly Census of Employment and
Wages (QCEW) for the years 1990-2016. The table reports estimates when using seasonal labor share constructed with the
alternative measure of contract workers, as described in the text. The dependent variable is given in the column header. Each
regression includes month-year and county-year fixed effects. Standard errors clustered by county are given in parentheses.
*, **, *** represent significance at 10%, 5% and 1%, respectively.
H-2A Workers
In recent years, agricultural employers have increasingly hired farm workers through the H-2A
agricultural guest worker program. Employers can apply for H-2A visas prior to the season
when workers are needed.
Some, but not all, states report H-2A workers in the QCEW. All H-2A workers are ex-
empt from the federal FUTA tax, which supports unemployment insurance administration.
However, states have differing policies regarding whether employers are required to report
34Note that sample sizes are slightly smaller in these regressions due to some missing counties in the Agri-
cultural Census.
48
Seasonal Agricultural Activity and Crime
employment and earnings and pay Unemployment Insurance taxes for H-2A workers. For
example, California and Washington require employers to report employment and earnings of
H-2A workers in the QCEW while Florida does not.35
To our knowledge, there is no database identifying which states require employers to
report employment and earnings of H-2A workers in the QCEW. Nevertheless, H-2A workers
represent a small share of total seasonal farm workers–only an estimated 7 percent of the crop
workforce in 2016 even though H-2A jobs had increased 160 percent from 2006-2016 (Martin
2017). Consequently, we expect measurement error arising from omitted H-2A workers in some
states to have little impact on our main findings. Nevertheless, we conduct two robustness
checks related to H-2A employment.
In the first robustness check, we limit our sample only to counties located in California
and Washington where we know that H-2A workers are included in the QCEW. Results are
qualitatively similar to our main specifications. Panel A of Table A7 shows a statistically
significant negative association between farm labor share and logged property crime rates in
the limited geographic sample. Panel B shows a statistically significant negative association
between farm labor share and log violent crime rate but a statistically significant positive
association between farm labor share and log violent crime count. This suggests that the
number of violent crimes rises with increased farm labor share in California and Washington,
but by less than the proportional increase in population.
Second, we measure the effects of H-2A share of the labor force on county crime rates.
There are several reasons H-2A workers may affect crime rates differently than other seasonal
35Florida is already omitted from our analysis due to irregular UCR data (See the data section of the paper
appearing in the AJAE ).
49
Seasonal Agricultural Activity and Crime
Table A3: Property & Violent Crime Results, California and Washington Only
A. Property Crime Results
(1) (2) (3)
Property Crime Rate Ln(Property Crime Rate) Ln(Property Crime Count)
Seasonal emp. share -2.136 -0.005∗∗ 0.004
(1.944) (0.002) (0.003)
N 31404 31334 31334
B. Violent Crime Results
(1) (2) (3)
Violent Crime Rate Ln(Violent Crime Rate) Ln(Violent Crime Count)
Seasonal emp. share -0.656 -0.003∗ 0.006∗∗∗
(0.459) (0.002) (0.001)
N 31344 31075 31075
Notes: Based on county-by-month data from Uniform Crime Reports (UCR) and the Quarterly Census of Employment
and Wages (QCEW) for the years 1990-2016. The table reports estimates when limiting the sample to California and
Washington. The dependent variable is given in the column header. Each regression includes month-year and county-year
fixed effects. Standard errors clustered by county are given in parentheses. *, **, *** represent significance at 10%, 5% and
1%, respectively.
farm workers. First, all H-2A workers have temporary legal work visas. Incentives to commit
crime and potential consequences may differ across work status, particularly if H-2A workers
want to have their work visas renewed the following year. Second, H-2A workers are not
legally able to remain in the United States after their visa expires, so H-2A workers may
have lower social accountability in the community compared to other seasonal farm workers.
Third, employers are required by law to pay for the transport of H-2A workers from and to
their country of origin and provide worker housing. Consequently, economic incentives to
commit crimes may differ for H-2A workers.
The results from estimating Equation (1) using the monthly H-2A worker share of the
labor force are shown in Table A8. Note that these regressions only cover the years 2008-2016
since H-2A visa counts are only available starting in 2008, so these results are not directly
comparable to our main results in Table 2. These estimates are generally smaller in magnitude
than in Table 2, and are not statistically significant. Hence we find no evidence that H-2A
50
Seasonal Agricultural Activity and Crime
employment shares are associated with increased crime, but do not find decreases in rates of
crime as in our main results.
Table A4: Property & Violent Crime Results, H2A Workers
A. Property Crime Results
(1) (2) (3)
Property Crime Rate Ln(Property Crime Rate) Ln(Property Crime Count)
H2A emp. share -0.113 -0.002 -0.002
(1.237) (0.002) (0.002)
N 297528 288361 288361
B. Violent Crime Results
(1) (2) (3)
Violent Crime Rate Ln(Violent Crime Rate) Ln(Violent Crime Count)
H2A emp. share 1.180 0.001 0.001
(1.130) (0.002) (0.002)
N 291540 275527 275527
Notes: Based on county-by-month data from Uniform Crime Reports (UCR) and H-2A data from the Department of
Labor, Office of Foreign Labor Certification (OFLC) for the years 2008-2016. The table reports estimates when using
the share of H-2A workers in the labor force. The dependent variable is given in the column header. Each regression
includes month-year and county-year fixed effects. Standard errors clustered by county are given in parentheses. *, **,
*** represent significance at 10%, 5% and 1%, respectively.
Employment Suppression Robustness Check
One caveat to this study is that the QCEW suppresses employment data at the county-
industry-year level in cases where firms could be identifiable, which typically means cases
where there are a small number of firms. Our measure of seasonal agricultural labor share will
be too low in cases where any of our twelve seasonal sectors have suppressed employment data.
However, almost by definition, employment suppression overwhelmingly occurs in cases with a
very small number of firms in the sector. Across observations in our main sample, the average
number of firms in a seasonal agricultural sector that is suppressed is 2.6 (the number of firms
is still provided for suppressed sectors), while the average number for non-suppressed (and
non-zero) sectors is 25.9. Therefore this issue should typically only cause under-measurement
51
Seasonal Agricultural Activity and Crime
of seasonal sectoral shares in cases where the sectoral employment is quite low (but non-zero).
The exception is cases where a single firm employs a very large number of people and is
suppressed, though for this phenomenon to cause bias in our estimates it would have to be
somehow related with seasonal crime rates. For these reasons we do not see the suppression
issue as a significant threat to the validity of our estimates.
Nevertheless, we perform a highly conservative robustness check in which we drop all
observations where any one of our 12 seasonal sectors is suppressed. Because these are often
small sectors, and are also fairly common even outside of major agricultural regions, this strict
condition drops roughly 40% of county-year combinations in the sample. Even so, the results
shown in Table A9 are qualitatively similar to the full sample.
Table A5: Property & Violent Crime Results, Observations with Suppressed Agricultural Employment
Data Excluded
A. Property Crime Results
(1) (2) (3)
Property Crime Rate Ln(Property Crime Rate) Ln(Property Crime Count)
Seasonal emp. share -5.938∗∗∗ -0.011∗∗∗ -0.003
(1.731) (0.003) (0.003)
N 521824 502648 502648
B. Violent Crime Results
(1) (2) (3)
Violent Crime Rate Ln(Violent Crime Rate) Ln(Violent Crime Count)
Seasonal emp. share -0.064 -0.003 0.005∗∗∗
(0.469) (0.002) (0.001)
N 500182 460222 460222
Notes: Based on county-by-month data from Uniform Crime Reports (UCR) and the Quarterly Census of Employment
and Wages (QCEW) for the years 1990-2016. The table reports estimates when dropping observations with suppressed
agricultural employment data. The dependent variable is given in the column header. Each regression includes month-year
and county-year fixed effects. Standard errors clustered by county are given in parentheses. *, **, *** represent significance
at 10%, 5% and 1%, respectively.
52
Seasonal Agricultural Activity and Crime
Inverse Hyperbolic Sine
When analyzing crime counts, we use a natural log transformation to yield a percentage
change interpretation and to prevent high-population counties from dominating the results,
but this causes zero-crime observations to be dropped. This results in roughly 3% and 6%
of observations being dropped for property and violent crimes, respectively. To address, this
in Table A10 we report the results from using an Inverse Hyperbolic Sine transformation,
which is defined at zero and converges to a natural log at larger values of the transformed
variable. The results for both propeerty and violent crimes are similar to our baseline logged
specification (see Table 2).
Table A6: Crime Results: Inverse Hyperbolic Sine Transformation
IHS(Crime Rate) IHS(Crime Count)
Property Crimes
Seasonal Emp. Share -0.013*** -0.004
(0.003) (0.003)
N 838332 838332
Violent Crimes
Seasonal Emp. Share -0.007*** 0.004***
(0.002) (0.001)
N 814020 814020
Notes: Based on county-by-month data from Uniform Crime Reports (UCR)
and the Quarterly Census of Employment and Wages (QCEW) for the years
1990-2016. The table reports estimates when using inverse hyperbolic sine
transformations for property and violent crime counts. The dependent vari-
able is given in the column header. Each regression includes month-year and
county-year fixed effects. Standard errors clustered by county are given in
parentheses. *, **, *** represent significance at 10%, 5% and 1%, respec-
tively.
Excluding 2016
Figure 4 showed that the effect of seasonal labor share on violent crime jumped upwards
in 2016, a year when immigration was a controversial issue due to the presidential election.
53
Seasonal Agricultural Activity and Crime
To check that results are not driven by this possible “election year” effect, we repeat our
main specifications excluding 2016 observations. The results shown in Table A11 are virtually
unchanged.
Table A7: Crime Results: Excluding 2016
Crime Rate Ln(Crime Rate) Ln(Crime Count)
Property Crimes
Seasonal Emp. Share -4.949∗∗∗ -0.012∗∗∗ -0.004
(1.295) (0.003) (0.003)
N 805212 783001 783001
Violent Crimes
Seasonal Emp. Share -0.298 -0.004∗∗ 0.004∗∗∗
(0.404) (0.002) (0.001)
N 781596 735011 735011
Notes: The table reports estimates when excluding 2016 observations. The
dependent variable is given in the column header. Each regression includes
month-year and county-year fixed effects. Standard errors clustered by county
are given in parentheses. *, **, *** represent significance at 10%, 5% and
1%, respectively.
Weighting by Labor Force
Since many counties with high agricultural labor share have low populations, in Table A12 we
check whether our main results are driven by these low-population counties by using Weighted
Least Squares weighting by total labor force. In this case, the negative effect magnitudes for
property crimes are actually larger, and the effect for natural log of property crime counts
is significant at a 10% level (this estimate is insignificant in our main specification). For
violent crimes, the effect on the log of violent crime rate is similar in magnitude but no longer
significant, while the other results are qualitatively similar.
54
Seasonal Agricultural Activity and Crime
Table A8: Crime Results: Weighted by Labor Force
Crime Rate Ln(Crime Rate) Ln(Crime Count)
Property Crimes
Seasonal Emp. Share -9.609∗∗∗ -0.016∗∗∗ -0.010∗∗
(2.783) (0.004) (0.004)
N 838332 814987 814987
Violent Crimes
Seasonal Emp. Share -0.372 -0.003 0.003∗∗
(0.471) (0.002) (0.001)
N 814020 765667 765667
Notes: Based on county-by-month data from Uniform Crime Reports (UCR)
and the Quarterly Census of Employment and Wages (QCEW) for the years
1990-2016. The table reports Weighted Least Squares estimates weighting
by labor force participants. The dependent variable is given in the column
header. Each regression includes month-year and county-year fixed effects.
Standard errors clustered by county are given in parentheses. *, **, ***
represent significance at 10%, 5% and 1%, respectively.
Corn Producer Control Counties
We next use corn farming to construct an alternative control group in our semi-parametric
specification. In our main specification we are including all counties in the United States
(subject to the restrictions discussed in Data and Methods sections). To limit the comparison
to other agriculture-intensive counties, we find the top 5% of counties by average corn employ-
ment during the sample period, and repeat our semi-parametric specification with only the
original treatment group and corn-producing counties included in the sample. These results
are shown in figure A9 and are qualitatively similar to the main results from Figure 3.
Evidence on Share of Crimes Reported to Police
The UCR data record only crimes that are reported to the police. Increases in population
through seasonal migration may affect crime rates through changes in the actual number of
per capita crimes committed or through changes in rates of reporting crime given that a crime
55
Seasonal Agricultural Activity and Crime
Property Crime Rate Ln(Property crime rate) Ln(Property crime count)
20 .05 .05
0
0 0
−.05
−20 −.05
−.1
−40 −.1
−.15
−60 −.2 −.15
−5 −4 −3 −2 −1 0 1 2 3 4 5 6 −5 −4 −3 −2 −1 0 1 2 3 4 5 6 −5 −4 −3 −2 −1 0 1 2 3 4 5 6
Months since peak Months since peak Months since peak
Violent crime rate Ln(Violent crime rate) Ln(Violent crime count)
20 .05 .05
15
0
10
−.05 0
5
0 −.1
−5
−.15 −.05
−5 −4 −3 −2 −1 0 1 2 3 4 5 6 −5 −4 −3 −2 −1 0 1 2 3 4 5 6 −5 −4 −3 −2 −1 0 1 2 3 4 5 6
Months since peak Months since peak Months since peak
Figure A4: Property & violent crime, corn-producing counties control group
Notes: Based on county-by-month data from Uniform Crime Reports (UCR) and the Quarterly Census of Employment and
Wages (QCEW) for the years 1990-2016. The graph plots coefficients and 95% confidence intervals from estimating Equation
(2) when using the top 5% of corn-producing counties as controls. The dependent variables are indicated in the figure headers.
Standard errors are clustered by county.
was committed. Non-citizens or unauthorized immigrants may be less likely to report crimes
if they feel less trustful of the police. Since a greater share of seasonal farm workers are
unauthorized immigrants and non-citizens than the U.S. population at large, this may be an
important mechanism in the effects of migrant farm labor shares on local crime rates.
We compare differences in the probability that Hispanics and non-Hispanics report violent
crimes,36 personal theft,37 burglary, and motor vehicle theft38 to the police from 1992-2016
using the National Crime Victimization Survey (NCVS). The NCVS is an annual survey
36Violent crimes include rape, attempted rape, sexual attack with serious or minor assault, completed
robbery with injury from assault, aggravated assault, unwanted sexual contact, and verbal threat of rape or
assault.
37Personal theft includes completed or attempted purse snatching, completed pocket picking, and completed
or attempted personal larceny.
38We include only incidents of completed burglary and motor vehicle theft.
56
Coefficient Coefficient
Coefficient Coefficient
Coefficient Coefficient
Seasonal Agricultural Activity and Crime
conducted by the Bureau of Justice Statistics with a nationally representative sample of indi-
viduals. The primary advantage of the NCVS is that it collects information on crimes that are
not reported to the police, so we can evaluate reporting rates for different demographics. The
NCVS collects information on respondents’ ethnicity, but it does not contain any information
on immigration status and geographic identifiers are available only at the regional level in the
public-use data.
We present percentage of crimes reported by type of crime for Hispanics and non-Hispanics
for the years 1992-2016 in Table A13. Hispanics are more likely to report violent crimes to the
police and less likely to report incidents of personal theft, and these differences are statistically
significant (reporting rates for burglary and vehicle theft are not significantly different). It
is possible that due to this issue our estimates of effects of seasonal agricultural activity on
violent crimes are biased upwards, while estimates of property crimes are biased downwards.
However, the differences in reporting rates are qualitatively quite small.
Table A9: National Crime Victimization Survey Comparison of Means
Share of Incidents Reported Share of Incidents Reported Difference
Crime Type if Hispanic if Non-Hispanic (Hisp. − Non-Hisp.) Observations
Violent Crime or Attempt 0.456 0.444 0.012* 51,366
(0.007) (0.003) (0.007)
Personal Theft 0.260 0.301 -0.041*** 74,501
(0.005) (0.002) (0.005)
Burglary 0.533 0.551 -0.018 29,015
(0.009) (0.003) (0.010)
Motor Vehicle Theft 0.931 0.913 0.018 6,964
(0.009) (0.004) (0.010)
Notes: Based on the National Crime Victimization Survey (NCVS) for the years 1992-2016. Standard errors are reported in
parentheses. Differences and p-values are derived from the linear regression of the variable of interest as the dependent variable on
a binary variable equal to 1 if the respondent was Hispanic and 0 otherwise.
57