Identifying the Dynamic Effects of Income Inequality on Crime
| Published date | 01 August 2020 |
| DOI | http://doi.org/10.1111/obes.12359 |
| Author | Bebonchu Atems |
| Date | 01 August 2020 |
751
©2020 The Department of Economics, University of Oxford and JohnWiley & Sons Ltd.
OXFORD BULLETIN OF ECONOMICSAND STATISTICS, 82, 4 (2020) 0305–9049
doi: 10.1111/obes.12359
Identifying the Dynamic Effects of Income Inequality
on Crime
Bebonchu Atems
Economics and Financial Studies, David D. Reh School of Business, Clarkson University,
8 Clarkson Avenue, Potsdam, NY 13699, USA. (e-mail: batems@clarkson.edu)
Abstract
What happens to crime after an increase in income inequality? The microeconomics litera-
ture that attempts to answer this question often employsidentification strategies that exploit
external sources of variation that provide quasi-experiments to identify causal effects. In
contrast, this paper tackles this question by using structural vector autoregressions (SVAR),
a methodology typically employed in modern empirical macroeconomics to identify and
estimate dynamic causal effects of exogenous shocks. Unlike the macroeconomic SVAR
models that are often applied to time-series data, we exploit the time series and cross-
sectional dimensions of our data, leading to the estimation of panel SVAR models. Using
U.S. state-level data for the period 1960–2015, our results indicate that structural shocks
to inequality increase both violent and property crime. Variance decomposition analyses
show that inequality has little explanatory power for movements in crime.
I. Introduction
Although economic and sociological theories predict that income inequality increases
crime, empirical support for this link remains as elusive as ever. Fajnzylber et al. (2002a,b),
for example, report positive and significant relationships between income inequality and
violent crime in a panel of countries. Based on municipal-level data for Brazil, Scorzafave
and Soares (2009) estimate an elasticity of pecuniary crimes relative to inequality of 1.46.
Demombynes and ¨
Ozler (2005), who analyze data on a cross-section of South African
neighbourhoods, also find that inequality is associated with property and violent crimes.
Recently, Enamorado et al. (2016) examine the impact of inequality on crime rates during
Mexico’s drug war, reporting that a one-point increase in the Gini coefficient raises drug-
related homicides per 100,000 inhabitants by more than 36%. Kelly (2000), and Choe
(2008), using U.S. state-level data, find that income inequality has a strong and robust
impact on several measures of violent crime but no effect on property crime. In contrast,
Chintrakarn and Herzer (2012) provide evidence of a negativecor relation between inequa-
lity and crime; Brush (2007) provides mixed evidence, with inequality having a posi-
tive link with crime in a cross-section of U.S. counties, but a negative association in his
JEL Classification numbers: K42.
752 Bulletin
time-series analysis; while Doyle et al. (1999), and Neumayer (2005) find no significant
link between inequality and violent crime.
Underlying this mixed empirical evidence is a serious endogeneity problem that arises
from measurement error, unobserved heterogeneity, and the simultaneous determination
of crime and income inequality. Income inequality is affected by some of the same
factors that drive individuals’decisions to engage in criminal activity, making it necessary
to control for reverse causality and simultaneity.As pointed out by Enamorado et al. (2016),
while inequality may increase crime, an increase in crime mayalso affect inequality by, for
instance, prompting higher income individuals to move out of areas ridden by a high inci-
dence of criminal activity.As well, suppose state and local governments attempt to reduce
crime by encouraging education.1The resulting increase in education, in turn decreases
inequality, causing the estimate of the coefficient on inequality in an OLS regression of
crime on inequality to be positivelybiased.2Evidently, cause and effect are not well defined
in regressions of crime on income inequality, to the extent that these studies fail to address
this simultaneity bias. Another important caveat that cannot be overlooked, especially in
cross-country studies, is the bias that results from omitted variables and measurement error.
Cross-country crime and inequality statistics are often susceptible to measurement error,
given the difficulty of obtaining reliable and comparabledata across countries. In addition,
even after controlling for a number of observed factors that driveboth inequality and crime,
unobserved heterogeneity remains a problem that plagues much of the empirical literature.
Recent empirical papers have resorted to differences-in-differences or instrumental
variables (IV) estimation methods to as a way around these endogeneity problems and to
estimate the causal effect of inequality on crime. Enamorado et al. (2016) use the predicted
income distribution of Mexico’s municipalities based on the initial income distribution of
the municipality and the national patterns of income growth to construct an instrumen-
tal variable for the observed Gini coefficient. Intuitively, their strategy measures change
in municipal income inequality not affected by local factors, such as crime, but by na-
tional trends. Hence, the instrument captures movements in a municipality’s inequality
that are exogenous with respect to crime in that municipality. To address the issue of
unobserved heterogeneity, Enamorado et al. (2016) include state-year fixed effects in all
regressions. Additionally, they limit their analysisto Mexico’s municipalities to circumvent
international data quality and comparability problems. Similarly, Choe (2008) attempts to
sidestep problems associated with unobserved heterogeneity and measurement error by us-
ing U.S. state-level panel data from 1994 to 2004, and deal with simultaneity by using the
Arellano and Bond (1991) Generalized Method of Moments (GMM) estimator. Scorzafave
and Soares (2009) use data for Sao Paulo’s municipalities, and address concerns with
endogeneity by using lagged regressors.
This article takes a fresh look at the question of the empirical link between income
inequality and crime. We address the aforementioned endogeneity problems by relating
U.S. state-level crime rates to measures of exogenous shocks to income inequality using
structural vector autoregressive (SVAR)models, an approach typically employed in modern
1See e.g. Bell et al. (2016), Lochner (2004), Lochner and Moretti (2004), Machin et al., (2011, 2012) for literature
on the crime-reducing effects of education.
2See e.g. Sylwester (2002), and Abdullah et al. (2015) for papers that document a negative impact of education
on inequality.
©2020 The Department of Economics, University of Oxford and JohnWiley & Sons Ltd
Dynamic effects of inequality on crime 753
empirical macroeconomics to identify and estimate dynamic causal effects of exogenous
shocks. Our proposed VAR methodologyoffers several advantages. First, like VAR models
in the macroeconomics literature, our panel SVAR approach helps us to identify the linearly
unpredictable movements in income inequality, and allows for reverse causality. Specifi-
cally, identification restrictions are imposed on theVAR model so that movements in state
income inequality that are exogenous with respect to state crime rates can be recovered
trivially. Consistent with the broader macroeconomics literature, we refer to these linearly
unpredictable movements in inequality as ‘shocks’ or ‘innovations’ to income inequality.
Because these shocks or innovations are exogenous with respect to crime, their impacts
on crime represent causal effects. Second, the panelVAR methodology allows us to trace
out the dynamic effects of these shocks to inequality on crime through the use of impulse
response functions. Tracing out the dynamic responses of crime to shocks to inequality
is quite important, as it potentially explains the wide range of estimates documented in
the literature. Third, our empirical framework enables us to quantify how much of the
variability in crime rates, on average, is accounted for by shocks to income inequality, as
opposed to other shocks. We do this through the use of forecast error variance decom-
positions. Fourth, unlike VAR models applied to time series data, our panel VAR model
also exploits the cross-sectional dimension of our data, thereby allowing us to control for
state-specific unobserved heterogeneity. Fifth, with annual data for the period 1960–2015
for the fifty U.S. states and D.C. (56 obser vations per state), a traditional time-series VAR
model is likely to be inadequate to uncover the effects of inequality on crime for each
state.3Combining the cross-sectional and time-series aspects of the data yields a sample
that exceeds 2,800 observations, which is reasonably large to obtain reliable estimates of
the effects of inequality on crime. Sixth, while we make no claims that our data are free
of measurement error, our use of U.S. state-level data mitigates some of the measurement
error problems associated with cross-country studies, namely data quality and compara-
bility. Given the difficulty of finding valid instruments, this panel VAR approach provides
an ideal alternative to estimate the causal impact of inequality on crime.4
Our baseline empirical results imply that structural innovations to income inequality
increase aggregate violent and property crime rates. While the contemporaneous responses
of both aggregate crime measures are statistically indistinguishable from zero, subsequent
responses display positive and statistically significant hump-shaped patterns, with respec-
tive peak effects of 0.4% and 0.3% following a 1% point increase in the Gini index.
Decomposition of variance analyses reveal that shocks to inequality account for negligible
proportions of the variability of crime rates. In no instance do shocks to income inequal-
ity explain more than 3% of variance of crime. Our main findings are largely robust to
alternative identification restrictions, and different measures of violent crime and property
crime.
The remainder of the paper is organized as follows. The next section presents the data
and discusses some stylized facts concerning the relationship between inequality and crime.
3Washington, D.C. is drastically different than any state in the sample. Also, local governance in D.C. began in
1975. Results (not shown to save space) do not change whenD.C. is dropped from the analysis.
4Similar panel VAR methods have been employed by e.g. Love and Zicchino (2006), Fort et al. (2013), Atems
and Jones (2015), Atems (2018, 2019a,b), Matvos,Ser u and Silva(2018), and Caballero, Fern ´andezand Park (2019).
©2020 The Department of Economics, University of Oxford and JohnWiley & Sons Ltd
To continue reading
Request your trial