ArticlePDF Available

Trigger Warning: The Causal Impact of Gun Ownership on Suicide

Authors:

Abstract

With a growing debate over tighter firearm regulations, we consider an important social consequence of increased firearm access: increased firearm suicides. Using data from the federal criminal background check system, we consider the impact of firearm ownership on firearm suicide rates. To deal with concerns of identification, we instrument for firearm background checks with state-year-level Google search intensity for phrases that reflect fear of future gun shortages and learning about the constitutional rights of firearm owners. We find that an increase in firearm ownership has a sizable and statistically significant impact on firearm suicide rates. A 10% increase in firearm ownership increases firearm suicide rates by approximately 3%, which is five times larger than non-instrumented estimates. Furthermore, we find no effect of gun ownership on non-firearm suicide rates, suggesting our findings are not simply capturing a suicide method substitution effect. The results are consistent with a variety of validity and robustness tests. Our results make clear the link between firearm ownership and firearm suicide rates, which have increased dramatically over the last decade.
Trigger Warning: The Causal Impact of Gun Ownership on
Suicide
David C. VittAlexander F. McQuoid§
Charles MooreStephen Sawyer k
July 28, 2018
Abstract
With a growing debate over tighter firearm regulations, we consider an important social
consequence of increased firearm access: increased firearm suicides. Using data from the
federal criminal background check system, we consider the impact of firearm ownership on
firearm suicide rates. To deal with concerns of identification, we instrument for firearm
background checks with state-year level Google search intensity for phrases that reflect fear
of future gun shortages and learning about the constitutional rights of firearm owners. We
find that an increase in firearm ownership has a sizable and statistically significant impact on
firearm suicide rates. A 10% increase in firearm ownership increases firearm suicide rates by
approximately 3%, which is five times larger than non-instrumented estimates. Furthermore,
we find no effect of gun ownership on non-firearm suicide rates, suggesting our findings are
not simply capturing a suicide method substitution effect. The results are consistent with
a variety of validity and robustness tests. Our results make clear the link between firearm
ownership and firearm suicide rates, which have increased dramatically over the last decade.
JEL classification: I18; I31; K4;
Keywords: Suicide; Gun Ownership; Google Trends;
We would like to thank the discussants and participants at AEA session 202-A of the 2018 ASSA Meetings
along with anonymous referees for helpful feedback on this paper.
The views expressed in this paper are those of the authors and should not be attributed to the United States
Naval Academy or Department of Defense. This research did not receive any specific grant from funding agencies
in the public, commercial, or not-for-profit sectors.
Economics Department, Farmingdale State College SUNY, 2350 Broadhollow Rd, Farmingdale, NY 11735,
vittd (at) farmingdale (dot) edu.
§Economics Department, United States Naval Academy, 589 McNair Road, Annapolis, MD 21402, mcquoid (at)
usna (dot) edu
Economics Department, United States Naval Academy, 589 McNair Road, Annapolis, MD 21402, m164458 (at)
usna (dot) edu
kEconomics Department, United States Naval Academy, 589 McNair Road, Annapolis, MD 21402, m165808 (at)
usna (dot) edu
1
1 Introduction
In 2014, 21,334 suicides were committed by the use of firearms in the U.S. This represents a 28%
increase since the start of the century. Mortality records from the CDC suggest suicides by firearm
are nearly twice as prevalent as homicides by firearm, and that half of all suicides are committed
via firearm. The social cost of suicide is staggering. A public health crisis this large necessitates
further consideration of policy options that can reduce the social and economic consequences of
suicide.
To understand the determinants of this public health crisis, we focus on the impact of firearm
availability on suicides. Figure 1 shows the evolution of firearm background checks in the US along
with the evolution of suicides by firearm over the last 15 years. First, note the downward trend
in firearm suicides in the early 2000s, broken by a sudden sharp increase in suicides. This sharp
increase coincides with a spike in firearm background checks.1
There is a danger in drawing sweeping conclusions from simple charts like those in Figure 1.
For example, a rise in depression could cause both an increase in firearm sales and an increase in
suicides, generating a spurious correlation. Or perhaps an increase in suicidal tendencies leads to
a similarly large increase in gun purchases. Alternatively, the evolution of these two phenomena
may be completely unrelated and only chance led to similar patterns.
Nonetheless, the patterns are interesting and suggestive, and in the present work, we seek to
disentangle and uncover any causal relationships if present. We proceed by looking for a plausible
source of exogenous variation in gun sales and gun ownership unrelated to issues of mental health
and suicide. Such a source of exogenous variation will allow us to make causal statements about
the impact of increased firearm sales on suicides by firearm.
Our identification strategy uses concern regarding future access to guns as an instrument
for present gun sales. That is, if agents are worried about access to guns in the future, they
will intertemporally substitute future gun ownership with present gun ownership. To capture
1Kentucky is an outlier in the data as they started additional background checks of concealed carry permit
holders in 2006, regardless of whether the permit holder was attempting to acquire a new gun at the time. As
Figure 1 shows, excluding Kentucky has no effect on the overall national trend.
2
variation in this concern, we use Google Trends search intensity for key words related to the
second amendment and gun bans. Agents searching for these terms are likely concerned about
future access to gun ownership, but not directly contemplating suicide. In response to this concern
about future access, agents are likely to buy guns today. Empirical evidence of this mechanism is
presented in Depetris-Chauvin (2015). With this intertemporal substitution in mind, we consider
the impact of moving future gun sales into the present on the incidence of suicides by firearm.
Using state-level variation in gun sales predicted by Google Trends search intensity for future
gun restrictions, we estimate the impact of these plausibly exogenous additional guns on suicide
rates. Our first stage results suggest that intertemporal gun ownership concerns do predict current
gun sales well, and these estimates are robust to additional socio-economic controls.
The second stage results imply that additional guns do have a statistically significant and
economically meaningful effect on suicide by firearm rates. Our estimates imply that a 10% increase
in gun sales leads to an approximate 3% increase in suicides by firearm. The identification strategy
deployed here is important as instrumental variable estimates are more than 5 times larger than
non-instrumented models, underscoring the importance of selection bias in the present context.
To test the robustness of our results and methodological approach, we consider a variety of
alternative specifications. Importantly, when we consider the impact of instrumented gun sales on
non-firearm suicides, we find no statistically significant effect. This finding is key for two reasons.
First, our approach would predict no effect on non-firearm suicides as the chain of events leads from
additional access to guns to additional suicides by firearm because of greater access to firearms.
The lack of a statistically significant finding thus supports the implied mechanism. Second, access
to guns could simply substitute suicide by alternative means to suicide via firearm. This would
imply an increase in suicides by firearm and an equal decrease in suicide by other means. Our
lack of such a finding implies these additional suicides would likely not have occurred otherwise,
representing a true social cost of additional firearms.
We test the validity of our instrument by replicating the exercise using Google Trends searches
for the 27th Amendment as an instrument. Our concern is that, if our instruments really do
3
measure a chain of events starting with fear of future gun shortages and ending with intertemporal
substitution of firearm purchases, then our results should not spuriously replicate with other
Google Trends search phrases. To test our identification claim, we instrument for gun sales using
searches for the 27th amendment, and find no statistically significant effects, providing additional
support for our instruments.
The results presented here suggest that gun ownership has significant social costs in the form
of additional suicides. The notable run-up in gun ownership over the last decade has coincided
with a startling increase in suicides. Our findings suggest these processes are not unrelated, and
that addition gun ownership leads to increased suicide by firearm rates.
The rest of the paper proceeds as follows. Section 2 discusses related literature, while Section
3 introduces the data. Section 4 lays out the empirical approach, discusses results, and presents a
series of robustness checks. Section 5 concludes.
2 Related Literature
Focus on the determinants of suicide has gained new-found interest from economists recently, as
the increase in suicide rates has become more pronounced over the last decade. Case and Deaton
(2015) find that changes in self-reported measures of well-being are poor predictors of changes in
suicide rates. However, they do find that physical pain is a strong predictor of suicide in many
contexts. Daly et al. (2011), however, find that happier areas tend to have higher suicide rates,
while Daly et al. (2013) provides evidence that interpersonal income comparisons may influence
suicide decisions. With increased focus on the importance of social factors, the economics literature
is moving away from an purely individual framework of suicide determination.
Perhaps due to data limitations, or perhaps on account of the stigma surrounding topics like
suicide and gun ownership, there have been limited attempts at establishing a causal impact of
gun ownership on firearm suicide rates. The previous empirical literature on the topic, much of
which comes from research in medicine or sociology, is based on exploring partial correlations as
4
a necessary first step towards understanding causality. While there is an abundance of empirical
papers on the topic of suicide, there are fewer economic theory papers on the topic. One possible
explanation may be the inappropriateness of a rational framework for understanding suicides. We
proceed by summarizing the theory underpinning our economic understanding of suicide thus far,
and then follow up with a synopsis of the econometric investigations into the relationship between
gun ownership and suicide.
An economic theory of suicide was first proposed in Hamermesh and Soss (1974), who built a
mathematical model to show that, given a cost of maintaining oneself through the aging process,
there is a point where the (marginal) cost of living exceeds the (marginal) benefit of each year of
life. The implication of this marginal analysis is that a rational agent would take the appropriate
suicidal measures at the point where the marginal cost of living exceeds the marginal benefit.
An updated approach to this rational framework is presented in Marcotte (2003), who proposes
an innovation that allows for suicide attempts to affect both the maintenance cost of living and
the probability of reaching next year, conditional on making it to this year. The model presented
predicts that suicide attempts increase in likelihood when, conditional on survival, suicide attempts
positively affects income either through direct income transfers or through enabling the attempter
to improve their life and productivity. They provide evidence for these mechanisms by appeal to
the National Comorbidity survey, where they find individuals with (unsuccessful) suicide attempts
have higher income than their peers with common suicidal ideation but lacking a suicide attempt.
Becker and Posner (2004) introduces greater uncertainty over the life cycle of an agent when
considering rational utility-maximizes behavior for unhappy individuals. The framework provides
valuable corrections and extensions to the Hamermesh and Soss (1974) optimizing approach, al-
lowing for greater testable predictions of the rationality theory of suicide.
Fundamentally, the criticism of Becker and Posner (2004), Marcotte (2003), and Hamermesh
and Soss (1974) are the same. These models approach an issue that is inappropriate for the rational
choice framework. The “suicide contemplating agent” lacks the ability required to precisely and
accurately calculate both the monetary and non-monetary gains from years of life to come, while
5
simultaneously overestimating the costs of maintenance as a result of their current emotional state
of being. There are many reasons to believe that an economic agent contemplating suicide has
behavior that may not fit the homo economicus paradigm. For example, it is possible that mental
illness is cognitively taxing in such a way that the agent’s judgment is clouded, prohibiting that
agent from forming a complete ordering of preferences regarding future states of the world. Such
agents would be unable to weigh the gains and losses appropriately.
Additional economic approaches to suicide have focused on decision theory modeling, but with
less demanding rationality requirements. Cutler et al. (2001) study the startling rise of youth
suicides over the last half of the 20th century, and consider alternative theoretical frameworks to
explain this rise, especially in light of the declining rates at the time for other age groups. Their
preferred theoretical interpretation for the youth results focuses on a signaling theory of suicide
attempts, since most youth suicide attempts fail, where a suicide attempt is interpret as a signal
for help. They also find evidence that supports a contagion view of suicide, which relies on social
pressures plus variability in emotions for youths.
Especially noteworthy for our purposes here, Cutler et al. (2001) consider an instrumental
view of suicide where access to easier suicide might increase the rate of suicide. The data they
considered does not allow for a clean test of this instrumental theory of suicide view, leaving an
open question we consider more rigorously here.
Seiden (1977) found that many suicides appear to be the result of impulsive behavior, where
individuals who take their own lives often do so when confronting a severe, but temporary crisis.
Simon et al. (2002) found that, among people who made near-lethal suicide attempts, 24% took
less than five minutes between the decision to kill themselves and the actual attempt, and 70%
took less than one hour. Rich et al. (1986) found that at-risk teenagers are more likely to act
impulsively in suicidal ideation, and are more likely to be affected by the means at hand.
The impulsive decision-making process of suicide is also addressed in the literature by studies
of survivors. Chapdelaine et al. (1991) found that, in cases of men who survived a self-inflicted
gunshot wound, subsequent suicide attempts were uncommon. Peterson et al. (1985) found that,
6
of self-inflicted gunshot wounds which were considered fatal without emergency medical treatment,
none of the 30 subjects studied had written a suicide note, and more than half reported having
suicidal thoughts for less than 24 hours. Furthermore, within two years of follow-up, none had
attempted suicide or died.
If suicide attempts are not strictly rational, then opportunity and method may have significant
effects on suicide rates. Using data from Canada, Chapdelaine et al. (1991) find that 92% of gun
attempts resulted in death, compared to 78% by carbon monoxide or hanging, 67% by drowning,
and 23% by drug overdose. Hemenway et al. (1995) found that 21% of gun owners store a gun
both loaded and unlocked, and that in 14% of gun-owning homes with children, a gun is stored
both loaded and unlocked. If guns are more plentiful and available, and if suicides are not a purely
rational decision, then the increased availability of guns could lead to more suicide attempts and
suicide deaths given the higher firearm success rate.
The substitutability of suicide method may be an important factor in understanding suicide
patterns. Under the rational suicide framework, substituting one method for another would depend
on the relative opportunity cost of each method. If access to a gun is made more difficult, the
agent would move to the next best suicide method. Assuming the cost of the next best method
was not significantly greater, the rational framework would predict only minor changes in suicide
attempts as a result of slightly higher method costs. On the other hand, if suicide attempts are
often impulsive, then easier access to firearms could lead to greater firearm suicides without a
commensurate decline in non-firearm suicides.
Fischer et al. (1993) find that there is an imperfect substitutability among methods of sui-
cide. They find that restricting access to a frequently used means of suicide such as firearms can
reduce total completed suicides by altering the composition of suicides to less effective methods
and because alternative methods are less socially acceptable, thus decreasing the probability of
being used. Related research has found that factors other than intent matter with respect to the
completion of a suicide attempt. Seiden and Spence (1984) analyze data from suicide patterns at
the Golden Gate Bridge and the Oakland Bay Bridge, and find that availability, suggestion, and
7
symbolic factors affect the choice of suicide method and location.
Correlational evidence of the relationship between gun ownership and suicide rates is much
more prevalent in the literature from the medical profession than it is for economics. Using a
matched pairs research design, Kellermann et al. (1992) find that keeping a firearm in the home
was strongly associated with an increased risk for suicide, estimating an adjusted odds ratio of
4.8. A matched pairs identification strategy depends upon the belief that after matching on
certain observable characteristics, all other differences are randomly distributed. In this study,
authors matched on sex, race, age, and neighborhood. However, unobservable characteristics such
as mental health could easily lead to households acquiring more guns and being more likely to
commit suicide.
Furthermore, their investigation is conducted only in two counties, both selected for being
large and being at opposite “extremes” of racial composition, which leaves concerns regarding
external validity. Nonetheless, the approximation of a more credible research design to tease out
the treatment effect in Kellermann et al. (1992) is an improvement over correlative studies that
use regional or international cross-sectional variation, such as those in Kaplan and Geling (1998),
Markush and Bartolucci (1984), or Molina and Duarte (2006).
In addition to providing a thorough literature review with a sociological lens, Kposowa et al.
(2016) uses a random effects strategy to estimate the association between ease of access to firearms
and both overall suicide rates and firearm suicide rates. Their results find a positive correlation
between ease of access to firearms and firearm suicide rates, and a negative correlation between
strictness of gun laws and firearm suicide rates. We note that their strategy incorporates a reason-
able set of control variables, which we use to inform our strategy regarding appropriate controls.
Our results build on their contribution in many ways. We relax the stringent identification require-
ments associated with random effects estimators by instead employing a fixed effects instrumental
variable (FEIV) estimator that fully allows for correlation between time invariant unobservables
and our variables of interest. To adequately use the FEIV estimator, we construct a panel of all
states from 2004 to 2013. Furthermore, we develop an instrumental variable strategy to deal with
8
endogeneity concerns surrounding variation in gun ownership that gives us confidence the partial
correlation we estimate represents a causal relationship.
In the empirical economics literature, the focus on guns and suicide was hampered since firearm
suicide was actually used as a proxy for gun ownership (see Cook and Ludwig (2006)). The closest
paper to our approach here is Lang (2013), who uses NICS background checks as a proxy for gun
ownership and studies the correlation between this proxy and suicide rates using panel data. The
attempt to deal with endogeneity in that paper focuses on youth suicides under the assumption
that youth are not able to legally purchase guns. However, as discussed previously the challenges
to identification are severe and sample disaggregation is unlikely to eliminate all identification
concerns related to measurement error, simultaneity, and omitted variables. Our use of Google
trends data to proxy for intertemporal gun ownership substitution provides a credible path to
identification.
The relationship between gun ownership and crime has received more attention in the economics
literature, with rigorous debate surrounding the hypothesis that guns increase/decrease crime.
The evolution of this literature is summarized in Aneja et al. (2011). While the topic is different,
many of the econometric challenges are the same. From this econometric discussion, we focus our
attention on panel data that allows us to control for state and time trends as well as confounding
covariates. However, whereas the “more guns, more/less crime” literature struggles to deal with
the endogeneity of crime and Right-to-Carry (RTC) laws, our identification strategy provides
plausible exogenous variation in gun ownership through intertemporal substitution based on lack
of future access to guns.
Our empirical strategy will rely on instruments that provide time varying measures of consumer
interest in various topics. Choi and Varian (2012) gives a thorough overview of the search volume
index (SVI). They show the utility of SVI as a time varying measure of consumer preferences
and interests for predicting and forecasting economic statistics like motor vehicle sales, home
sales, unemployment claims, and tourism. The field of finance has recently taken interest in the
search volume index as an improvement for covariates formerly proxied on account of limited data
9
availability. As an example, Da et al. (2011) use Google Trends data to measure investor attention.
They find evidence that increases in the search volume index for stock tickers correlate highly with
increases in stock prices and eventual reversals of the high prices. In finance research, Vlastakis
and Markellos (2012) and Da et al. (2011) use search volume for a stock ticker to proxy demand
for information about the company for interested investors; Vitt (2017a) uses search volume for
phrases that reflect learning about Bitcoin to proxy speculative demand.Vitt (2017b) uses Google
Trends to instrument for e-commerce use intensity with search intensity for various keywords like
“porn” and “cat videos”, and we borrow aspects of this strategy to confirm robustness of the
approach. Our intent is to use Google Trends data to measure demand for information that might
lead an agent to intertemporally substitute a future gun purchase into a present gun purchase.
3 Data
The first step in our analysis is to define a measure of gun ownership. Given that there is no
standardized database that directly tracks ownership of guns, we rely on the National Instant
Criminal Background Check System (NICS), maintained by the Federal Bureau of Investigations
since 1998. This system is used by firearms vendors to determine the worthiness of a prospective
firearm buyer. Prior to completing the sale, a call to the FBI or other designated agency is
conducted to ensure the customer is not prohibited from purchasing a firearm, and this is recorded
as a NICS check.2This metric does not represent the total gun ownership or the number of firearms
in a given state, but it does proxy changes in the stock of gun owners as well as changes in the
accumulation of firearms in the state, while also capturing changes in intent to own a firearm.
Thus, we can think of the NICS metric as being a measure of gun ownership that is observed
with some measurement error that will necessitate an instrumental variable approach to isolate
the exogenous variation in gun ownership.
The validity of background checks as a proxy for gun ownership is discussed in Lang (2013).
2Prohibitions include people convicted of a crime punishable by imprisonment for at least one year, people who
have been documented as addicted to controlled substances, and people who have been adjudicated as mentally
defective, among others.
10
The literature has considered other proxies for firearm ownership at the national and census level
using the General Social Survey, but this source is inappropriate for lower levels of observation
such as the state in a given year. The CDC Behavioral Risk Factor Surveillance System also
collects some data on firearm ownership, but not with sufficient granular information to be useful
at the state-year level. Duggan (2001) uses subscriptions to Guns & Ammo as a proxy for firearm
ownership, with limited success. Although there are concerns with non-compliance, private gun
purchases, and transfers across state lines, Lang (2013) shows that NICS background checks are
comparable to alternative measures of gun ownership at the national and census level, and thus
likely to be useful proxies at lower levels of aggregation over time.3
Our data on suicides by firearm come from the Center for Disease Control’s mortality records.
In particular, we use the Public-Use files for Multiple-Cause-of-Death records, which is available
from 1999 until 2014. These files are drawn from all death certificates files in the United States
in a given year. Causes of death are classified according to the International Classification of
Disease 10th edition (ICD-10) standards. We focus on Intentional Self Harm codes, which we
further distinguish between intentional self-harm with a firearm and all other intentional self-
harm deaths. The use of the CDC suicide data has been used in a variety of economic contexts
recently, ranging from international trade (Pierce and Schott (2016)) to pain epidemics (Case and
Deaton (2015)), among others.
These records track all deaths and report not only the cause of death, but demographic variables
of interest such as race, sex, and age. We use this mortality data to construct the total number of
suicide deaths in a state over time, as well as to partition suicide deaths in firearm suicides and
non-firearm suicides. Doing so allows us to investigate whether an increase in gun ownership has
a substitution effect on the method of suicide, or if gun ownership increases suicide rates at the
margin by enabling those already considering it to more easily make a rash decision and commit
suicide. We present the average firearm suicide rate and average firearm background check rate
nationally over time in Figure 1.
3For a recent survey of gun ownership acquisition without a background check, see Miller et al. (2017).
11
Search intensity data is available on the Google Trends page.4Search intensity can be refined
by geographic unit (countries, states, MSAs, etc.) and over time from 2004 until present. The
search volume index is reported on a monthly basis as the total monthly query volume for the
particular keywords or phrases as a fraction of the total number of search queries in the geographic
area that month. Google then normalizes the maximum query share for the time period of interest
to 100. We aggregate these monthly values to an annual search volume index by averaging over
the months in the year.
We believe that spikes in gun purchases today are largely stemming from an aversion to ex-
pected future firearm shortages. It should be the case that news about future gun restrictions
prompts risk averse consumers to substitute away from future personal security purchases in fa-
vor of purchasing that equipment today. This mechanism, evidence of which is documented in
Depetris-Chauvin (2015), drives our decision to include several keywords from Google Trends as
instruments in order to measure consumer concerns about future firearm shortages. We start with
state level search intensity for the phrase“second amendment”, which captures concern about con-
stitutional rights associated with firearm ownership. To further supplement intertemporal demand
for firearm ownership, we include a second instrument based on state level search intensity for “gun
bans”. Concern about future gun bans would be a significant motivator to substitute future gun
ownership for present gun ownership.
To get a sense of the variation in our instruments over time, we present the state level variation
in the search intensity for these phrases in Figure 2 for selected states. In this graph, we see that
there is significant variation in the search volume index for these phrases within states, as well as
across-state differences in the search intensity relative to the peak search intensity for the state.
Additional state and year controls were collected to control for additional social and economic
factors that may be correlated with our instruments. If these factors are also correlated with suicide
rates, failure to include these would bias our instrumented estimates. As will be shown below,
inclusion of additional controls does not statistically affect our IV estimates, but we present results
4www.google.com/trends or trends.google.com
12
with and without additional controls for completeness. Demographic data such as population
estimates, median income, the percentage of the population between 18-24, the percentage of the
population that is African-American, and the veteran population are sourced from the U.S. Census
Bureau. Data on the state unemployment rate is sourced from the U.S. Bureau of Labor Statistics.
Crime data is taken from the FBI’s Uniform Crime Reporting (UCR) program, while prison
population statistics come from the Bureau of Justice Statistics.5In a series of robustness checks,
we use proxies for mental health from the CDC’s annual Behavioral Risk Factor Surveillance
Survey (BRFSS). Our complete sample covers all 50 states from 2004 until 2013.6Summary
statistics for all variables are reported in Table 1.
4 Empirical Analysis
4.1 Specifications and Results
First, to get a sense of the relationship between firearm suicide rates and guns, we consider an
empirical strategy without instruments. Fixed effect (FE) estimates are presented in Table 2. The
first observation is that a naive approach that did not fully exploit panel data information would
conclude that gun background checks are positively associated with firearm suicide rates, although
the effect is small. Column (1) uses fixed effects and a linear time trend, while columns (2) and (3)
include additional covariates, and column (4) weights the observations by state population size.
The estimated effect is small: for a 10% increase in gun checks, firearm suicide rates increase by
0.6%.
This initial exercise provides some evidence consistent with the trends presented in Figure
1. Serious concerns about selection bias, measurement error, and simultaneity, however, suggest
5We include these on account of how they seem to be plausible sources for differences in relative well-being,
which is in turn an important determinant of suicide rates as explored in Daly et al. (2013).
6Our primary sample has 500 observations, but we lose some observations due to missing Google Trends data
and missing socio-economic controls. For the Google Trends data, four states (Delaware, Rhode Island, Vermont,
and Wyoming) are missing data for ”gun ban” searches because the intensity was not above the threshold Google
Trends sets for reporting. 8 other states are missing a single observation, mostly from 2004 or 2005.
13
extreme caution in interpreting the results in Table 2 as causal.
For a causal estimation and interpretation, we proceed following our instrumental variable
estimation strategy. Our preferred specification estimates the elasticity of firearm suicide rates
with respect to gun background checks per capita, while controlling for additional factors that may
influence firearm suicide rates and be correlated with our instrument. Our preferred specification
is:
ln(Firearm Suicide Ratest) = β0+β1ˆ
ln(Gun Background Checks Per Capitast )
+Controlsstβ
+µs+γ t +εst
(1)
where the dependent variable is the log number of firearm suicides per 100,000 population in
state sfor year t. Our variable of interest, ˆ
ln(Gun Background Checks Per Capitast ), is the log
predicted number of NICS background checks per capita in the state for the given year from
the first stage results. Included in Controlsst are time-varying state characteristics that could be
correlated with our instrument and also correlated with firearm suicides In Eq. (1), µsrepresents a
state fixed effect that accounts for any time-invariant determinants of suicide, such as differences
in cultures towards suicide across geographic borders as explored in Neumayer (2003). Time
trends are included to account for unobserved national forces driving suicide rates. To address
the concerns regarding undersized standard errors when shocks may be correlated at a geographic
level, as in Bertrand et al. (2004), we cluster observations at the state level as we view this as the
most appropriate level for likely correlations in shocks to suicide, although results are robust to
alternative assumptions regarding the distribution of the disturbances.
Given concerns regarding our proxy for gun ownership via background checks, a classic mea-
surement error problem, and to address (omitted) time varying confounding factors, we adopt an
instrumental variables strategy. We instrument for gun background checks using Google search
intensity for “second amendment” , and in robustness checks, “gun bans”. To isolate exogenous
14
variation in gun ownership, our first stage specification is as follows:
ln(Gun Background Checks Per Capitast ) = π0+π1ln(second amendment search intensityst)
+π2ln(gun ban search intensityst )
+Controlsstπ
+µs+γ t +vst
(2)
Our identification strategy relies on the idea that, through intertemporal substitution, con-
sumers respond to a fear of future firearm shortages (perhaps due to anticipation of regulatory
changes) by substituting away from future firearm purchases in favor of firearm purchases in the
current period.
Furthermore, we argue that these search phrases, which measure the expectation of future
difficulty to attain firearms and learning about constitutional rights respectively, influence suicide
only through increasing the stock of guns today. An increase in the stock of guns today increases
the ease of making an impulsive decision to kill oneself with a gun. It seems unlikely that searches
for these keywords would be direct determinants of suicide themselves. This exclusion restriction
is akin to saying that the fear of future gun shortages in and of itself is not the reason people are
committing suicide. To account for possible correlations between search intensity and other factor
that may influence suicide rates, we include additional economic and demographic controls in our
IV estimation.
Before turning to the first stage results, we consider some graphical anecdotal evidence of the
relationship between the instruments and our proxy of gun ownership. Figure 3 shows state-level
partial correlations between Google search intensity for “Second Amendment” and NICS back-
ground checks, after controlling for state population and linear time trends. The figure shows
significant heterogeneity across states, which is important for the implementation of the IV strat-
egy. Although the evidence is anecdotal, it provides initial empirical support for our approach,
which we formally and systematically confirm by estimating Eq. (2).
From Table 3 we note that our hypothesis regarding the relationship between expectations of
15
future gun shortages, learning about gun owner’s rights, and gun background checks is supported
by the significance of “gun ban” and “second amendment” search intensity. For our primary
instrument, the second amendment search intensity is highly significant (0.1% in columns (1)-
(4)), and the excluded instrument F is above 10, suggesting the approach does not suffer from
a weak instruments problem. Column (5) considers search intensity for gun ban, with similar
individual statistical significance and passes the weak instrument test. Column (6) includes both
search intensities, with second amendment search intensity significant at 1% and gun ban search
intensity significant at 10%. The specification also passes the weak instrument test. Our first
stage results strongly suggest that each of our instruments generates significant variation in the
number of gun background checks within the state over time. Given the statistical value of the
first stage, we now move to a full estimation of the impact of gun background checks on firearm
suicide rates.
Armed with an empirical strategy suited to address measurement error, endogeneity, and
omitted variable bias problems, we are able to paint a clearer picture of the relationship between
gun ownership and suicide. First, consider Table 4, which presents IV estimation. Column (1)
includes state fixed effects and linear time trends, and estimates an elasticity of 0.38. The effect is
highly significant, and implies that a 10% increase in gun ownership results in a 3.8% increase in
firearm suicides. This estimate is notable because it is nearly six times larger than the equivalent
non-instrumented estimate in Table 2, underscoring the importance of a credible identification
strategy and confirming concerns that measurement error, endogeneity, and omitted variables are
biasing the estimates in Table 2.
When additional time varying controls are added in columns (2) and (3) to address inde-
pendence concerns regarding the instrument, the point estimate is slightly smaller, although not
statistically distinguishable, and continue to be highly significant. The estimates are again 5-6
times larger than the equivalent estimates in Table 2. Column (4) weights the regression by the
population of the state, resulting in a point estimate that is about 50% smaller. Unweighted
regressions treat all suicide rates equally so that changes in the suicide rate in California are
16
given equal weight as changes in suicide rates in Delaware. However, while the estimated effect is
smaller, the same 50% reduction is present in Table 2, so that the estimated effect is still roughly
5 times larger when our instrument is utilized. The lower weighted regression estimate suggests
that smaller states (in terms of population) have higher elasticities on average, although this might
be partially driven by the fact that smaller population states have noisier suicide rates since an
additional suicide has a larger impact on the suicide rate than in larger population states. The
weighted regression results imply that a 10% increase in background gun checks results in a 2.3%
increase in firearm suicides.
4.2 Robustness Checks
To consider the robustness of our underlying identification, we next consider the search intensity
for gun bans, an alternative approach to capturing a desire for intertemporal gun ownership
substitution. In Table (4), the estimates presented in column (5) use only search intensity for
“gun ban” as an instrument, and are very similar to column (3) with an estimated elasticity of
0.26. In column (6), we include both search phrases as instrument to proxy for intertemporal gun
ownership substitution, and find that for a 10% increase in gun background checks, firearm suicide
rates increase by 2.9%, which is 5 times larger than the non-instrumented estimate.
The results from Table 4 strongly support the view that gun ownership, as proxied by gun
background checks, results in higher firearm suicide rates. Furthermore, an identification strategy
that accounts for endogeneity, measurement error, and omitted variables is important. Failure
to account for selection bias would result in estimates that understate the true causal impact by
a factor of five. Social costs of gun ownership in terms of the impact of suicide rates are thus
significantly understated without a clear identification strategy.
To further consider the validity of our approach and hypothesized mechanism, we consider
the effect of guns sales on non-firearm suicide rates. The mechanism we have in mind is that an
increase in the abundance of guns translates directly into more opportunities for someone to make
a rash decision and kill themselves with one. If our mechanism is valid, and more gun sales lead
17
to more firearm suicides that would not have otherwise occurred, then we would not expect to
find an effect of gun background checks on non-firearm suicide rates.
Any effect, positive or negative, could cast doubt on our approach. If we find a positive effect
on non-firearm suicides, this would imply that whatever effect we are picking up, it is a more
general affect related to suicides, and therefore our hypothesized causal mechanism of more guns
leading to more gun suicides would be questionable. Alternatively, if we find a negative effect
on non-firearm suicides, this would imply that more guns may shift the composition of suicide
method without impacting the fundamental forces driving suicide attempts.7
To test this placebo hypothesis of no relationship between gun ownership and non-firearm
suicide rates, we run a regression similar to Eq. (1) with non-firearm suicide rates as the dependent
variable. Fixed effects instrumental variable results from this placebo specification appear in
Table 5, providing strong evidence that our instrument is not capturing some suicide specific
force unrelated to firearms. We find no evidence that gun background checks impacts non-firearm
suicides. In our preferred specification (column (3)), the estimated elasticity is essentially zero.
Perhaps not unsurprisingly given our identification strategy, the OLS (not reported) and IV point
estimates are nearly identical, -0.024 compared to -0.016. We thus rule out either story related to
general suicide trends or suicide method substitution effects, while providing additional support
for our hypothesized mechanism.
To test the validity of our instruments, we show that our selection of instruments is meaningful
in the sense that if we used the search intensity for another constitutional amendment as an
instrument, our results should fail to replicate spuriously. If we use a different constitutional
amendment search intensity, and find an effect on suicide rates, this would imply that our primary
results are likely spurious and should be discounted. For this purpose, we choose to collect state-
year level search intensity for the phrase “27th amendment”. The 27th amendment, our most
recent, prevents congress from passing any law that would increase or decrease the salary of
7It is, however, possible that by substituting a firearm suicide attempt for a non-firearm suicide attempt may
increase successful suicides if firearm suicides attempts are on average more likely to result in successful suicides,
as in Chapdelaine et al. (1991).
18
congress members until the beginning of the next term. While we see no reason for searches for
such an amendment to be correlated with gun sales, it may be correlated with general unhappiness
with social and political institutions. If this general unhappiness is symptomatic of some deeper
unhappiness, it may be related to suicide rates.
For this validity test, we run a first stage similar to Eq.(2), using only 27th amendment search
intensity as an instrument. Results from this first stage appear in column (1) of Table 6. The
first stage results show no apparent relationship between 27th amendment search intensity and
gun background checks. The estimated coefficient on search intensity of the 27th amendment is
essentially zero and highly insignificant, while the excluded instrument F statistics is less than
2, consistent with a weak instrument. Unlike second amendment and gun ban search intensity,
we find no evidence that 27th amendment search intensity is a good predictor of gun background
checks. Nonetheless, we use this first stage result to regress suicide rates on predicted gun back-
ground checks, and present the results in column (2). Gun background checks, when instrumented
with search intensity for the 27th amendment, are statistically unrelated to firearm suicide rates.
The fact that not just any Google search phrase gives significant results lends credibility to the
instruments selected and our hypothesized mechanism.
To further demonstrate the validity of our instruments, we present limited information maxi-
mum likelihood (LIML) estimates for a validity check similar to the Cruz and Moreira (2005) check
of Angrist and Krueger (1991). In column (3) of Table (6) we use LIML to estimate the effect of
interest and compare it to the FE estimates in Table 2 and the FEIV estimates in Table 4. In
light of Stock et al. (2002), since the F statistic of the excluded instrument test is sufficiently high,
any bias introduced by weak instruments is minimal. If our instruments were weak and we were
committing a Type I error regarding the existence of weak instruments, then the LIML estimates
will be close to the OLS estimates. Since the LIML estimates in column (3) of Table 6 are close to
our preferred FEIV specification replicated in column (4) of Table 6, we find additional support
for the validity and strength of our search phrase intensity instruments.
For another robustness check, we try to eliminate alternative stories and mechanisms driving
19
the results. First, we consider whether Google search intensity may just be capturing general
Internet usage or even specific Internet usage that may be correlated with anti-social behaviors
associated with suicide risk. For example, perhaps people who are suicidal feel isolated from their
local communities and retreat to virtual communities. These people may spend a lot of time
at home on the Internet, and while on the Internet, they may search for a variety of phrases,
including constitutional amendments. This would result in a biased overestimation of the impact
of gun ownership on suicides, since our instrument would be correlated with people who were
predisposed to committee suicide. Alternatively, robust online communities may reduce the sense
of isolation, resulting in happier individuals who are less likely to commit suicide. This would
result in a biased underestimation of the impact of gun ownership on suicide, since our instrument
would be correlated with people who were less likely to commit suicide as a result of their online
participation. From our identification perspective, our concern is that our Google trends search
intensity is correlated with general online usage intensity, which could be correlated with suicide
risk. To separate our proxy for intertemporal gun ownership substitution from general Internet
usage, we include a covariate of Internet use intensity.
Ideally, we would pick a search phrase that would capture as wide a cross-section of the state’s
Internet users as possible. A 2015 study by Internet accountability and filtering firm CovenantEyes
found that 30% of all Internet traffic is pornography. For this reason, we include Google Trends
data on the state level search intensity for pornography as an additional control for Internet access
and intensity. Results from adding this Internet use intensity variable as a control are found in
column 1 of Table 7. Our point estimate on instrumented gun background checks is essentially
unchanged, while we do find a weakly significant negative effect on our Internet usage proxy. These
findings together imply that our instrument was not biased correlated with general Internet usage,
and that Internet access does seem to lower suicide rates, consistent with the view that online
communities provide valuable social connections.
Next, we consider alternative measures of mental health. In our primary specifications, we
include Google trends searches for psychiatry. Here we consider an alternative measure based
20
on the CDC’s annual Behavioral Risk Factor Surveillance System (BRFSS). We calculate state
specific measures of self-reported mental health. Based on survey response data, we calculate the
percent of survey respondents in a state in a given year reporting having had at least twenty “not
good” days of mental health in the last month, proxying for extreme unhappiness. In column (2),
we find no statistically significant difference in the estimated effect of gun background checks on
firearm suicide rates. When we include both this alternative measure of mental health and our
measure of general Internet usage intensity, the results are unchanged.
For a further check of model robustness, we consider an alternative time trend assumption with
a random trend model, where we estimate a trend coefficient specific to each state. The random
trend model is included to address the concerns in Aneja et al. (2011), namely we may not want
to impose that the trend in suicide rates from a vacation state like Florida is the same as the
trend in suicide rates for more isolated states like Wyoming or Montana. In column (4) of Table
7, we include state specific linear trends with no noticeable effect on our point estimate. There
is disagreement about the appropriateness of state specific time trends within the guns and crime
literature, but this appears not to matter in our context as our results are robust to adding this
flexibility. 8
We believe it is also necessary to demonstrate that data from Kentucky are not driving our
results. As discussed in Lang (2013), in 2006, Kentucky began performing background checks
on concealed carry permit holders, who were typically exempt from NICS background checks
when purchasing firearms. The NICS measures from Kentucky thus may not be an equivalent
proxy for gun ownership as in other states, although this would not impact the identification
unless Kentucky’s background check of concealed carry permit holders was correlated with our
instrument. To confirm that our results are not being driven by this outlying behavior, we drop
Kentucky from the sample, and estimate the model in column (5). The results are unchanged.
Our penultimate robustness check is to consider whether our results hold when we split the
sample by sex. Summary statistics for female firearm suicide rates and male firearm suicide rates
8See Aneja et al. (2011)for a thorough discussion of the literature and the disagreement over this assumption.
21
can be found in Table 1. Note that the average male firearm suicide rate is more than six times the
average female firearm suicide rate, which is consistent with previous findings of gender differences
in suicide as in Case and Deaton (2015), Callanan and Davis (2012), Denning et al. (2000), and
others. Table 8 estimates Eq. (1) with female and male firearm suicide rates in columns (1)-(2)
and (4)-(5) respectively. While columns (1) and (4) present unweighted estimates, columns (2) and
(5) present estimates weighted by the female and male population respectively. Columns (3) and
(6) check for suicide method substitution by replacing the dependent variable with non-firearm
suicide rates in each sex. The totality of evidence in these columns reinforces our result: all else
equal, an increase in firearm background checks per capita leads to higher firearm suicide rates
in either sex with no corresponding reduction in non-firearm suicide rates. We note the point
estimate for the elasticity of firearm suicide with respect to NICS checks is larger for females than
for males, which is intuitively reasonable given that the average firearm suicide rate for females
is an order of magnitude smaller than the average for males. We leave precise estimation and
hypothesis testing of differential effects by sex as a fruitful area of future work.
Lastly, we consider alternative error term assumptions to test the validity of our inference. In
Table 9, we estimate our preferred second stage specification, given in Eq. (1), with a sequence of
different methods for estimating the standard errors of our point estimates. This sequences begins
with conventional standard errors, then robust standard errors a la White (1980), regional clusters,
jackknife, and bootstrap standard errors. In each case, statistical significance varies between 5%
and 0.1%, confirming the inferential resilience of the model.
5 Conclusion
Understanding the relationship between firearm ownership and suicides by firearm is an important
social policy question given the increasingly large social cost of suicide in the U.S. A lack of rigorous
empirical evidence previously hindered painting a clear picture of the causal relationship between
the two. In this paper, we attempt to make this causal impact clear: an increase in the number of
22
firearms available today means increased opportunities for costly impulsive decisions like suicide
by firearm.
We utilize an instrumental variables approach in order to address the measurement error in
current measures of gun ownership, and omitted variable problems inherent in models of suicide
determination. With this strategy, we find that an increase in the number of gun background
checks within a state indeed causes a significant and sizable increase in the rate of firearm suicides
within that state. To show that these suicides deaths are truly suicides that would have not
otherwise occurred, we also show that increased firearm access does not induce a substitution
across the various methods of committing suicide.
Our estimates yield a natural way to estimate the economic impact of firearm regulation
policies. Consider a modest policy that would reduce handgun ownership by 20%. This would
be met with a reduction in the firearm suicide rate of approximately 6%. Given that there were
21,000 firearm suicides in 2014, this reduction in ownership would result in nearly 1,300 fewer
firearm suicides each year.
The social costs of gun ownership have mostly focused on the impact on crime, and in particular
violent crime. While the relationship between gun ownership and firearm suicides has been ob-
served in the data, confidence in the underlying relationship was limited by the correlative studies
used to estimate the relationship. Our fixed effects instrumental variable estimates are nearly five
times larger than the fixed effect estimates, suggesting that the true costs of firearm ownership
have been significantly understated. Our results here suggest a meaningful causal relationship
between firearm ownership and firearm suicide. Preventable suicides impose enormous costs on
society, and the expanding prevalence of gun ownership has contributed notably to these increased
costs. Future public policy should carefully weigh the benefits and costs of gun ownership, and
the impact on suicide should be included in that cost-benefit analysis.
23
References
Aneja, Abhay, John J. Donohue, and Alexandria Zhang, “The Impact of Right-to-Carry
Laws and the NRC Report: Lessons for the Empirical Evaluation of Law and Policy,” American
Law and Economics Review, 2011, 13 (2), 565–631.
Angrist, Joshua D. and Alan B. Krueger, “Does Compulsory School Attendance Affect
Schooling and Earnings?,” The Quarterly Journal of Economics, 1991, 106 (4), 979–1014.
Becker, Gary S and Richard A Posner, “Suicide: An economic approach,” University of
Chicago, 2004.
Bertrand, M, E Duflo, and S Mullainathan, “How Much Should We Trust Differences-In-
Differences Estimates?,” The Quarterly Journal of Economics, 2004, 119 (1), 249–275.
Callanan, Valerie J and Mark S Davis, “Gender differences in suicide methods,” Social
psychiatry and psychiatric epidemiology, 2012, 47 (6), 857–869.
Case, Anne and Angus Deaton, “Suicide, age, and wellbeing: an empirical investigation,” in
“Insights in the Economics of Aging,” University of Chicago Press, 2015.
Chapdelaine, Antoine, Esther Samson, MD Kimberley, and L Viau, “Firearm-related
injuries in Canada: issues for prevention.,” CMAJ: Canadian Medical Association Journal, 1991,
145 (10), 1217.
Choi, Hyunyoung and Hal Varian, “Predicting the Present with Google Trends,” Economic
Record, 2012, 88 (SUPPL.1), 2–9.
Cook, Philip J. and Jens Ludwig, “The social costs of gun ownership,” Journal of Public
Economics, 2006, 90 (1–2), 379 – 391.
Cruz, Luiz M and Marcelo J Moreira, “On the validity of econometric techniques with
weak instruments inference on returns to education using compulsory school attendance laws,”
Journal of Human Resources, 2005, 40 (2), 393–410.
Cutler, David M, Edward L Glaeser, and Karen E Norberg, “Explaining the rise in youth
suicide,” in “Risky behavior among youths: An economic analysis,” University of Chicago Press,
2001, pp. 219–270.
Da, Zhi, Joseph Engelberg, and Pengjie Gao, “In Search of Attention,” The Journal of
Finance, 2011, 66 (5), 1461–1499.
Daly, Mary C., Andrew J. Oswald, Daniel Wilson, and Stephen Wu, “Dark contrasts:
The paradox of high rates of suicide in happy places,” Journal of Economic Behavior and
Organization, 2011, 80 (3), 435–442.
Daly, Mary C, Daniel J Wilson, and Norman J Johnson, “Relative Status and Well-Being:
Evidence from U.S. Suicide Deaths,” Review of Economics, 2013, 95 (5).
24
Denning, Diane G, Yeates Conwell, Deborah King, and Chris Cox, “Method choice,
intent, and gender in completed suicide,” Suicide and Life-Threatening Behavior, 2000, 30 (3),
282–288.
Depetris-Chauvin, Emilio, “Fear of Obama: An empirical study of the demand for guns and
the U.S. 2008 presidential election,” Journal of Public Economics, 2015, 130, 66–79.
Duggan, Mark, “More guns, more crime,” Journal of political Economy, 2001, 109 (5), 1086–
1114.
Fischer, Ellen P, George W Comstock, Mary A Monk, and David J Sencer, “Charac-
teristics of completed suicides: implications of differences among methods,” Suicide and Life-
Threatening Behavior, 1993, 23 (2), 91–100.
Hamermesh, Daniel S. and Neal M. Soss, “An Economic Theory of Suicide,” Journal of
Political Economy, 1974, 82 (1), 83.
Hemenway, David, Sara J Solnick, and Deborah R Azrael, “Firearm training and storage,”
Jama, 1995, 273 (1), 46–50.
Kaplan, Mark S. and Olga Geling, “Firearm suicides and homicides in the United States:
Regional variations and patterns of gun ownership,” Social Science and Medicine, 1998, 46 (9),
1227–1233.
Kellermann, Arthur L., Frederick P. Rivara, Grant Somes, Donald T. Reay, Jerry
Francisco, Joyce Gillentine Banton, Janice Prodzinski, Corinne Fligner, and Bela B.
Hackman, “Suicide in the Home in Relation to Gun Ownership,” New England Journal of
Medicine, 1992, 327 (7), 467–472.
Kposowa, Augustine, David Hamilton, and Katy Wang, “Impact of Firearm Availability
and Gun Regulation on State Suicide Rates,” Suicide and Life-Threatening Behavior, 2016, 46
(6), 678–696.
Lang, Matthew, “Firearm background checks and suicide,” The Economic Journal, 2013, 123
(573), 1085–1099.
Marcotte, Dave E, “The Economics of Suicide, Revisited,” Southern Economic Journal, 2003,
69 (3), 628–643.
Markush, R. E. and A. A. Bartolucci, “Firearms and suicide in the United States,” American
Journal of Public Health, 1984, 74 (2), 123–127.
Miller, Matthew, Lisa Hepburn, and Deborah Azrael, “Firearm Acquisition Without
Background Checks Results of a National Survey,” Annals of internal medicine, 2017.
Molina, Jos´e Alberto and Rosa Duarte, “Risk determinants of suicide attempts among
adolescents,” American Journal of Economics and Sociology, 2006, 65 (2), 407–434.
25
Neumayer, Eric, “Are Socioeconomic Factors Valid Determinants of Suicide? Controlling for
National Cultures of Suicide With Fixed-Effects Estimation,” Cross-Cultural Research, 2003, 37
(3), 307–329.
Peterson, Linda G, McKim Peterson, Gregory J O’Shanick, and Alan Swann, “Self-
inflicted gunshot wounds: Lethality of method versus intent.,” The American Journal of Psy-
chiatry, 1985.
Pierce, Justin R and Peter K Schott, “Trade liberalization and mortality: Evidence from US
counties,” Technical Report, National Bureau of Economic Research 2016.
Rich, Charles L, Deborah Young, and Richard C Fowler, “San Diego suicide study: I.
Young vs old subjects,” Archives of general psychiatry, 1986, 43 (6), 577–582.
Seiden, Richard H, “Suicide prevention: a public health/public policy approach,” OMEGA-
Journal of Death and Dying, 1977, 8(3), 267–276.
and Mary Spence, “A Tale of Two Bridges: Comparative Suicide Incidence on the Golden
Gate and San Francisco-Oakland Bay Bridges.,” Omega: Journal of Death and Dying, 1984, 14
(3), 201–9.
Simon, Thomas R, Alan C Swann, Kenneth E Powell, Lloyd B Potter, Marcie jo Kres-
now, and Patrick W O’Carroll, “Characteristics of impulsive suicide attempts and at-
tempters,” Suicide and Life-Threatening Behavior, 2002, 32 (s1), 49–59.
Stock, James H, Jonathan H Wright, and Motohiro Yogo, “A survey of weak instruments
and weak identification in generalized method of moments,” Journal of Business & Economic
Statistics, 2002, 20 (4), 518–529.
Vitt, David C, “Does fiat-to-Bitcoin exchange activity lead to increased user- to-user Bitcoin
transaction activity?,” Academy of Economics and Finance Journal, 2017, 8, 71–76.
Vitt, David C., “Estimating the Impact of E-Commerce on Retail Exit and Entry Using Google
Trends,” Working Paper, 2017.
Vlastakis, Nikolaos and Raphael N. Markellos, “Information demand and stock market
volatility,” Journal of Banking and Finance, 2012, 36 (6), 1808–1821.
White, Halbert, “A heteroskedasticity-consistent covariance matrix estimator and a direct test
for heteroskedasticity,” Econometrica, 1980, 48 (4), 817–838.
26
6 Appendix
6.1 Figures
(Starting on next page)
27
6.5 7 7.5 8
Firearm suicides per 100,000 population
2000 2005 2010 2015
Year
Full Sample Excluding Kentucky
.03 .04 .05 .06 .07 .08
Firearm background checks per capita
2000 2005 2010 2015
Year
Full Sample Excluding Kentucky
Figure 1: Variation in the average firearm suicide rate across states in every year is presented in the left panel, and variation in
the average number of firearm background checks across states in every year is presented in the right panel.
28
0 10 20 30 40
Gun Ban Search Intensity
2004 2006 2008 2010 2012 2014
Year
Alabama Dist. of Columbia
Nebraska
0 20 40 60 80
Second Amendment Search Intensity
2004 2006 2008 2010 2012 2014
Year
California Delaware
Nevada
Figure 2: Variation in Google search intensity for phrases “gun ban” and “second amendment” across and within select states
over time
29
Bin Range (bin count)
.8 - 1 (1)
.6 - .8 (10)
.4 - .6 (16)
.2 - .4 (7)
0 - .2 (7)
-.2 - 0 (3)
-.4 - -.2 (4)
-.6 - -.4 (1)
Figure 3: State-level Partial Correlations between Google Search Intensity for 2nd Amendment and NICS Background Checks
30
6.2 Tables
Table 1: Summary Statistics
Variable Mean Std. Dev. Min. Max. N
Firearm Suicide Rate 7.32 3.01 1.37 17.70 500
Female Firearm Suicide Rate 1.84 1.00 0.06 6.17 500
Male Firearm Suicide Rate 12.86 4.80 2.54 2.77 500
Non-Firearm Suicide Rate 6.31 1.56 3.43 12.50 500
All Suicide Rate 13.63 3.69 6.20 29.67 500
Population (millions) 6.09 6.71 0.51 3.83 500
Unemployment Rate 6.33 2.22 2.50 13.80 500
Median Income 54229.68 8349.57 37173 78632 500
Violent Crime Rate 384.69 158.88 87.70 789.90 500
% Veteran Population 0.083 0.014 0.045 0.11 500
% Young 9.36 0.90 7.22 13.52 500
% Black 10.18 9.23 0.00 36.46 500
% Prison 0.004 0.002 0.002 0.009 500
Gun Ban Search Intensity 8.87 6.76 0 38.83 500
Second Amendment Search Intensity 18.47 9.31 0 65.75 500
27th Amendment Search Intensity 16.64 13.48 0 90.17 410
Psychiatrist Search Intensity 46.13 13.79 8.83 83.17 500
20 or More Days “Not Good Mental Health” Share 0.08 0.02 0.04 0.13 499
31
Table 2: Fixed Effects Regression of Firearm Suicide Rates on Background Checks and Controls
(1) (2) (3) (4)
Log Firearm Suicide Rate Log Firearm Suicide Rate Log Firearm Suicide Rate Log Firearm Suicide Rate
Log Gun Background Checks Per Capita 0.0612∗∗∗ 0.0630∗∗∗ 0.0597∗∗∗ 0.0405∗∗
(0.0200) (0.0215) (0.0215) (0.0170)
Log Population -0.351 -0.362 -0.176
(0.251) (0.307) (0.223)
Unemployment Rate -0.0000367 0.000134
(0.00251) (0.00211)
Median Income -0.00000318 -0.00000300
(0.00000279) (0.00000194)
Violent Crime Rate -0.000101 -0.000100
(0.000135) (0.0000995)
% Veteran Population 3.5133.415∗∗
(1.969) (1.492)
% Young 0.00664 -0.000238
(0.00537) (0.00506)
% Black -0.00141 -0.00609
(0.0139) (0.00898)
% Prison 12.33 28.86∗∗
(20.40) (13.31)
Psychiatrist Search Intensity 0.000652 0.000369
(0.000468) (0.000274)
Instruments None None None None
Std. Error State Cluster State Cluster State Cluster State Cluster
Weight Unweighted Unweighted Unweighted Weighted by Population
Observations 451 451 451 451
Standard errors in parentheses. All models presented in this table include both state fixed effects and a linear time trend.
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
32
Table 3: First Stage Regression of Background Checks on Instruments and Controls
(1) (2) (3) (4) (5) (6)
Log Gun Background Checks Per Capita Log Gun Background Checks Per Capita Log Gun Background Checks Per Capita Log Gun Background Checks Per Capita Log Gun Background Checks Per Capita Log Gun Background Checks Per Capita
Log Second Amendment Search Intensity 0.111∗∗∗∗ 0.113∗∗∗∗ 0.118∗∗∗∗ 0.142∗∗∗∗ 0.0831∗∗∗
(0.0288) (0.0263) (0.0245) (0.0220) (0.0298)
Log Gun Ban Search Intensity 0.0572∗∗∗∗ 0.0321
(0.0143) (0.0171)
Log Population 0.677 0.784 0.207 0.774 0.850
(1.185) (1.022) (0.881) (1.000) (0.998)
Unemployment Rate -0.0136 -0.0214∗∗∗∗ -0.0113 -0.0117
(0.00895) (0.00615) (0.00848) (0.00842)
Median Income -0.00000875∗∗ -0.00000700-0.00000847∗∗ -0.00000853∗∗
(0.00000390) (0.00000403) (0.00000379) (0.00000386)
Violent Crime Rate 0.000426 0.0000232 0.000268 0.000416
(0.000281) (0.000363) (0.000278) (0.000279)
% Veteran Population -0.0687 -0.275 -1.107 -0.920
(2.799) (4.268) (3.006) (2.822)
% Young -0.00986 -0.0112 -0.00802 -0.00908
(0.0175) (0.0132) (0.0176) (0.0176)
% Black -0.00221 -0.00738 0.00377 -0.00335
(0.0231) (0.0233) (0.0214) (0.0218)
% Prison -39.62 37.37 -35.96 -29.53
(51.06) (43.27) (47.70) (49.38)
Psychiatrist Search Intensity 0.0000476 -0.000483 0.000660 0.000202
(0.00102) (0.000795) (0.000994) (0.00104)
Excluded Instruments F 14.41 17.85 22.26 39.61 15.30 12.89
Std. Error State Cluster State Cluster State Cluster State Cluster State Cluster State Cluster
Weight Unweighted Unweighted Unweighted Weighted by Population Unweighted Unweighted
Observations 451 451 451 451 451 451
Standard errors in parentheses. All models presented in this table include state fixed effects and a linear time trend.
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
33
Table 4: FEIV Regression of Firearm Suicide Rates on Background Checks and Controls
(1) (2) (3) (4) (5) (6)
Log Firearm Suicide Rate Log Firearm Suicide Rate Log Firearm Suicide Rate Log Firearm Suicide Rate Log Firearm Suicide Rate Log Firearm Suicide Rate
Log Gun Background Checks Per Capita 0.382∗∗∗ 0.367∗∗∗ 0.314∗∗ 0.226∗∗∗ 0.257∗∗ 0.291∗∗
(0.140) (0.131) (0.136) (0.0831) (0.125) (0.118)
Log Population -0.514 -0.491 -0.162 -0.462-0.479
(0.353) (0.305) (0.292) (0.272) (0.289)
Unemployment Rate 0.00409 0.00459 0.00317 0.00372
(0.00372) (0.00355) (0.00318) (0.00336)
Median Income -0.000000900 -0.00000146 -0.00000141 -0.00000111
(0.00000312) (0.00000220) (0.00000311) (0.00000308)
Violent Crime Rate -0.000133 -0.0000429 -0.000126 -0.000130
(0.000132) (0.000110) (0.000129) (0.000131)
% Veteran Population 3.271 3.060∗∗ 3.325 3.293
(2.267) (1.543) (2.185) (2.232)
% Young 0.00891 0.00124 0.00840 0.00870
(0.00643) (0.00556) (0.00594) (0.00620)
% Black -0.00518 -0.00911 -0.00434 -0.00484
(0.0162) (0.00909) (0.0157) (0.0160)
% Prison 30.15 32.43∗∗ 26.16 28.54
(26.36) (13.86) (24.97) (25.45)
Psychiatrist Search Intensity 0.000458 0.000457 0.000501 0.000475
(0.000559) (0.000369) (0.000527) (0.000545)
Instruments 2nd Amendment 2nd Amendment 2nd Amendment 2nd Amendment Gun Ban 2nd Amendment, Gun Ban
Excluded Instruments F 14.41 17.85 22.26 39.61 15.30 12.89
Std. Error State Cluster State Cluster State Cluster State Cluster State Cluster State Cluster
Weight Unweighted Unweighted Unweighted Weighted by Population Unweighted Unweighted
Observations 451 451 451 451 451 451
Standard errors in parentheses. All models presented in this table include state fixed effects and linear time trends.
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
34
Table 5: Second Stage FEIV Regression of Non-Firearm Suicide Rates on Per Capita Gun Background Checks w/ Search
Intensity Instruments and Controls
(1) (2) (3) (4) (5)
Log Non-Firearm Suicide Rate Log Non-Firearm Suicide Rate Log Non-Firearm Suicide Rate Log Non-Firearm Suicide Rate Log Non-Firearm Suicide Rate
Log Gun Background Checks Per Capita -0.0917 -0.104 -0.0155 0.0250 -0.123
(0.139) (0.134) (0.140) (0.0830) (0.152)
Log Population -0.404 -0.395 -0.464 -0.340
(0.379) (0.346) (0.294) (0.409)
Unemployment Rate 0.00543 0.005450.00368
(0.00453) (0.00287) (0.00454)
Median Income -0.00000289 -0.00000287 -0.00000386
(0.00000339) (0.00000180) (0.00000363)
Violent Crime Rate 0.000476∗∗∗∗ 0.000422∗∗∗∗ 0.000489∗∗∗∗
(0.000132) (0.000123) (0.000143)
% Veteran Population 0.957 1.150 1.059
(2.490) (1.705) (2.462)
% Young 0.00463 0.00989∗∗ 0.00367
(0.00703) (0.00478) (0.00722)
% Black -0.00656 -0.0227∗∗ -0.00496
(0.0149) (0.0108) (0.0154)
% Prison -26.19 -0.668 -33.73
(23.38) (22.66) (26.28)
Psychiatrist Search Intensity 0.00000112 -0.000210 0.0000834
(0.000395) (0.000345) (0.000413)
Instruments 2nd Amendment 2nd Amendment 2nd Amendment 2nd Amendment 2nd Amendment, Gun Ban
Excluded Instruments F 14.41 17.85 22.26 39.61 12.89
Std. Error State Cluster State Cluster State Cluster State Cluster State Cluster
Weight Unweighted Unweighted Unweighted Weighted by Population Unweighted
Observations 451 451 451 451 451
Standard errors in parentheses. All models presented in this table include state fixed effects and a linear time trend.
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
35
Table 6: Instrument Validity Tests
(1) (2) (3) (4)
Log Gun Background
Checks Per Capita Log Firearm Suicide Rate Log Firearm Suicide Rate Log Firearm Suicide Rate
Log 27th Amendment Search Intensity 0.0222
(0.0205)
Log Gun Background Checks Per Capita 0.355 0.294∗∗ 0.291∗∗
(0.498) (0.120) (0.118)
Log Population 0.428 -0.490 -0.481-0.479
(0.957) (0.337) (0.291) (0.289)
Unemployment Rate -0.0166 0.00393 0.00376 0.00372
(0.0101) (0.0104) (0.00338) (0.00336)
Median Income -0.0000115∗∗ 0.000000612 -0.00000108 -0.00000111
(0.00000502) (0.00000665) (0.00000309) (0.00000308)
Violent Crime Rate 0.000273 -0.000113 -0.000131 -0.000130
(0.000378) (0.000219) (0.000131) (0.000131)
% Veteran Population 3.932 1.943 3.291 3.293
(5.148) (2.524) (2.237) (2.232)
% Young -0.00236 -0.000805 0.00872 0.00870
(0.0191) (0.00796) (0.00623) (0.00620)
% Black 0.00681 -0.00476 -0.00488 -0.00484
(0.0263) (0.0143) (0.0160) (0.0160)
% Prison -49.80 37.14 28.72 28.54
(63.75) (41.15) (25.55) (25.45)
Psychiatrist Search Intensity -0.000135 0.000715 0.000473 0.000475
(0.00149) (0.000605) (0.000546) (0.000545)
Estimation Method FE FEIV LIML FEIV
Instruments First Stage 27th Amendment Gun Ban, 2nd Amendment Gun Ban, 2nd Amendment
Excluded Instruments F 1.175 12.89 12.89
Std. Error State Cluster State Cluster State Cluster State Cluster
Weight Unweighted Unweighted Unweighted Unweighted
Observations 389 389 451 451
Standard errors in parentheses. Column (4) is a duplicate of our preferred specification in Table (4).
All models presented in this table include state fixed effects and linear time trends.
“FE” denotes fixed effects, “FEIV” denotes fixed effects instrumental variable, “LIML” denotes limited information maximum likeliho od.
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
36
Table 7: Model Robustness Tests
(1) (2) (3) (4) (5)
Log Firearm Suicide Rate Log Firearm Suicide Rate Log Firearm Suicide Rate Log Firearm Suicide Rate Log Firearm Suicide Rate
Log Gun Background Checks Per Capita 0.278** 0.273** 0.259** 0.253*** 0.275**
(0.115) (0.127) (0.124) (0.0968) (0.114)
Log Population -0.448 -0.475* -0.447 -0.0380 -0.461
(0.281) (0.288) (0.279) (0.686) (0.284)
Unemployment Rate -0.000299 0.00354 -0.000261 0.00561 0.00132
(0.00376) (0.00334) (0.00397) (0.00388) (0.00343)
Median Income -0.00000123 -0.00000143 -0.00000157 -0.00000346 -0.000000659
(0.00000300) (0.00000338) (0.00000330) (0.00000276) (0.00000303)
Violent Crime Rate -0.000149 -0.000136 -0.000157 -0.000209 -0.000131
(0.000131) (0.000127) (0.000127) (0.000198) (0.000132)
% Veteran Population 3.122 3.861 3.697 2.883 3.135
(2.186) (2.434) (2.395) (2.503) (2.206)
% Young 0.00862 0.00811 0.00797 0.0142** 0.0100
(0.00611) (0.00606) (0.00594) (0.00642) (0.00636)
% Black -0.00577 -0.00443 -0.00524 0.0183 -0.00520
(0.0159) (0.0152) (0.0152) (0.0228) (0.0159)
% Prison 26.21 23.55 21.04 14.64 31.39
(24.92) (27.18) (26.52) (36.18) (26.93)
Psychiatrist Search Intensity 0.000531 0.000470 0.000410
(0.000531) (0.000565) (0.000533)
Lesbian Porn Search Intensity -0.000616* -0.000591 -0.000114 -0.000621*
(0.000350) (0.000367) (0.000379) (0.000354)
20 or More Days ”Not Very Good” Share 1.617 1.645
(1.333) (1.305)
Instruments 2nd Amendment, Gun Ban 2nd Amendment, Gun Ban 2nd Amendment, Gun Ban 2nd Amendment, Gun Ban 2nd Amendment, Gun Ban
Excluded Instruments F 12.92 11.33 11.30 10.82 12.67
Std. Error State Cluster State Cluster State Cluster State Cluster State Cluster
Weight Unweighted Unweighted Unweighted Unweighted Unweighted (Dropped Kentucky)
Observations 451 450 450 451 442
Standard errors in parentheses. All models presented in this table include state fixed effects and linear time trends. Column 4 includes a state specific linear time trend.
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
37
Table 8: IV and FEIV Regressions of Log Firearm Suicide Rates on Log Gun Background Checks Disagregated By Sex
(1) (2) (3) (4) (5) (6)
Log
Female Firearm
Suicide Rate
Log
Female Firearm
Suicide Rate
Log
Female Non-Firearm
Suicide Rate
Log
Male Firearm
Suicide Rate
Log
Male Firearm
Suicide Rate
Log
Male Non-Firearm
Suicide Rate
lnNICSpc 0.396 0.437∗∗ 0.0270 0.289∗∗ 0.183∗∗ -0.0550
(0.345) (0.216) (0.100) (0.137) (0.0689) (0.0884)
Log Population -0.482 -0.706 -0.608 -0.481 -0.122 -0.399
(0.618) (0.499) (0.467) (0.310) (0.321) (0.329)
Unemployment Rate -0.000556 0.00178 0.000383 0.00474 0.00484 0.00556
(0.00985) (0.0102) (0.00446) (0.00331) (0.00328) (0.00305)
Median Income -6.19e-08 -0.00000848 -0.00000549-0.000000701 -0.000000641 -0.00000242
(0.00000969) (0.00000519) (0.00000312) (0.00000398) (0.00000221) (0.00000185)
Violent Crime Rate 0.000378 0.000261 0.000416∗∗ -0.000146 -0.0000553 0.000413∗∗∗
(0.000589) (0.000308) (0.000160) (0.000158) (0.000108) (0.000132)
% Veteran Population 0.347 1.339 2.644 3.541 3.1310.843
(6.060) (4.174) (2.562) (2.958) (1.669) (1.658)
% Young -0.000630 -0.0358 0.0113 0.00932 0.00380 0.00854
(0.0220) (0.0239) (0.00804) (0.00708) (0.00655) (0.00596)
% Black -0.0136 -0.0266 -0.0104 -0.00350 -0.00929 -0.0271∗∗
(0.0274) (0.0362) (0.0245) (0.0177) (0.00877) (0.0104)
% Prison 57.04 42.37 -5.673 21.23 27.880.157
(59.49) (32.59) (45.48) (29.11) (15.00) (19.15)
Psychiatrist Search Intensity 0.000377 0.000447 0.000307 0.000602 0.000471 -0.000374
(0.00164) (0.00104) (0.000668) (0.000613) (0.000378) (0.000414)
Instruments 2nd Amendment, Gun Ban 2nd Amendment, Gun Ban 2nd Amendment, Gun Ban 2nd Amendment, Gun Ban 2nd Amendment, Gun Ban 2nd Amendment, Gun Ban
Excluded Instruments F 12.89 30.48 30.48 12.89 30.48 30.48
Std. Error State Cluster State Cluster State Cluster State Cluster State Cluster State Cluster
Weight Unweighted Weighted by Population Weighted by Population Unweighted Weighted by Population Weighted by Population
Observations 451 451 451 451 451 451
Standard errors in parentheses.
All models in this table estimated with state fixed effects and a linear time trend.
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
38
Table 9: FEIV Regressions of Log Firearm Suicide Rates on Log Gun Background Checks w/ Differing Standard Error As-
sumptions
(1) (2) (3) (4) (5)
Log Firearm Suicide Rate Log Firearm Suicide Rate Log Firearm Suicide Rate Log Firearm Suicide Rate Log Firearm Suicide Rate
Log Gun Background Checks Per Capita 0.291*** 0.291** 0.291**** 0.291** 0.291**
(0.111) (0.113) (0.0448) (0.129) (0.136)
Log Population -0.479* -0.479* -0.479 -0.479 -0.479
(0.272) (0.282) (0.375) (0.322) (0.375)
Unemployment Rate 0.00372 0.00372 0.00372 0.00372 0.00372
(0.00383) (0.00341) (0.00410) (0.00341) (0.00378)
Median Income -0.00000111 -0.00000111 -0.00000111 -0.00000111 -0.00000111
(0.00000226) (0.00000255) (0.00000111) (0.00000345) (0.00000328)
Violent Crime Rate -0.000130 -0.000130 -0.000130 -0.000130 -0.000130
(0.000135) (0.000133) (0.0000918) (0.000149) (0.000150)
% Veteran Population 3.293** 3.293 3.293 3.293 3.293
(1.654) (2.044) (2.832) (2.678) (2.388)
% Young 0.00870 0.00870 0.00870* 0.00870 0.00870
(0.00607) (0.00654) (0.00522) (0.00621) (0.00728)
% Black -0.00484 -0.00484 -0.00484 -0.00484 -0.00484
(0.0139) (0.0136) (0.0156) (0.0178) (0.0187)
% Prison 28.54 28.54 28.54* 28.54 28.54
(22.00) (22.50) (14.58) (28.25) (27.87)
Psychiatrist Search Intensity 0.000475 0.000475 0.000475 0.000475 0.000475
(0.000541) (0.000545) (0.000407) (0.000559) (0.000589)
Std. Error Conventional Huber-White Region Cluster Jackknife Bootstrap
Instruments 2nd Amendment, Gun Ban 2nd Amendment, Gun Ban 2nd Amendment, Gun Ban 2nd Amendment, Gun Ban 2nd Amendment, Gun Ban
Observations 451 451 451 451 451
Standard errors in parentheses. All models presented in this table include state fixed effects and linear time trends.
p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01, ∗∗∗∗ p < 0.001
39
... Similarly, Lang (2013) finds that an increase in the rate of firearm background check rates in a state is associated with higher youth firearm suicide rates. Vitt et al. (2018) instrument firearm background check data with Google searches about gun policy and conclude that firearm suicide rates increase when firearm ownership increases. Using a reduction in military-issued guns in private households in Switzerland, Balestra (2018) shows that firearm suicides decrease when there are fewer firearms in circulation. ...
... Our analysis uses Google Trends in a similar manner as Levine and McKnight (2017). Tefft (2011), Stephens-Davidowitz (2014), Kearney and Levine (2015), Vitt et al. (2018), and Berger, Chen, and Frey (2018) are just a few of many recent papers that incorporate Google Trends data directly into their empirical analysis. 9 The "gun for home" value reported in June 2020 is 82, since the May index is 64 and the June index is 100. ...
... Since November 1998, the Federal Bureau of Investigation (FBI) has published the number of monthly firearm background checks for every state. While this measure does not reflect firearm sales oneto-one, it has been shown to be a suitable proxy for firearm purchases (Lang 2013) and used as both an independent (Lang 2013(Lang , 2016Levine and McKnight 2017;Vitt et al. 2018) and a dependent variable (Depetris-Chauvin 2015; Levine and McKnight 2017;Brock and Routon 2020) in a number of studies. The FBI updates the data at the start of every month, allowing us to use data up to August 31, 2020. ...
... Azrael et al. (2004) and Kleck (2004) evaluated a range of firearm proxies including concealed carry permits, NRA memberships, crime and arrest data, unintentional firearm deaths, outdoor magazine subscriptions and federal firearm licensees. Lang (2013), Briggs and Tabarrok (2014) and Vitt et al. (2018) use FBI background checks as a proxy for prevalence. Kovandzic et al. (2011;2013) use outdoor sports magazines subscriptions, percentage of those voting Republican in the 1988 presidential election, and numbers of military veterans as instruments for their proxy FSS. 5 A variety of earlier papers have studied firearm license data. ...
... Most purchasers complete their purchase shortly after the FBI background check, suggesting that FBI background check data reliably indicate firearm purchase intention. Consequently, recent literature has used FBI background checks to proxy for firearm prevalence (e.g., Briggs and Tabarrok 2014;Lang 2013;Vitt et al. 2018). Still, FBI background check data have some weaknesses. ...
Article
Full-text available
Product acquisition policies define legal markets. Policy evaluations require data but prevalence data are not always available. We introduce Legal Firearm Prevalence (LFP), a direct behavioral measure based on the population of firearm licensees in Massachusetts, and argue that it can help evaluate firearm sales and usage restrictions. LFP is not directly measurable in most firearm markets, so we test candidate proxies for LFP in several common research designs, finding that firearm acquisitions are the best proxy in every research design tested. We update the classic study of guns and crime by Cook and Ludwig (2006), finding that choosing an invalid proxy can lead to false research conclusions. We recommend systematic collection and reporting of firearm acquisition data to improve firearm research and inform firearm policy.
... In fact, previous studies have used this specific indicator of firearm background checks by NICS extensively (e.g. Depetris-Chauvin, 2015;Lang, 2013Lang, , 2016Levine & McKnight, 2017;Studdert et al., 2017;Vitt et al., 2018). We acquired the number of firearm background checks conducted each month, for the following historical events and months, accessed via the NICS. ...
Article
News outlets ran stories suggesting that firearm purchases in the United States might have increased during the onset of the Coronavirus pandemic. Such claims were made because gun stores were deemed essential businesses at the onset of the pandemic. However, there is no scientific evidence to validate this claim. We tested whether intentions to own a firearm actually increased at an unprecedented rate, by comparing the rate of increase in firearm checks (a conservative estimate of intentions to obtain a firearm) at the onset of the pandemic with the same time period in previous years as well as with significant events in recent American history. We defined the month of February as the onset of the Coronavirus pandemic in the United States because this was the month in which (a) the pandemic caught wider national attention, (b) the first official presidential address relevant to the Coronavirus was made, and (c) the CDC initiated its first measures to stop the spread of the virus. Understanding why (inclination toward) firearm ownership increases during times of national crises can help researchers and gun policy makers better understand the psychological needs driving firearm ownership, and potentially improve gun regulations and gun policies for the future.
... These results also establish the fallacious character of an argument commonly made in the suicide literature, which runs like this: (1) gun levels have a significant positive association with the firearms suicide rate, and (2) have no significant association with the non-firearms suicide; therefore (3) gun levels must have a significant positive association with the total suicide rates. Authors making this argument typically report the association of gun rates with firearms suicide rates and non-firearms suicide rates, but do not report the association of gun rates with total suicide rates (e.g., Lester, 1991, p. 188;Killias, 1993;Lahti, Keranen, Hakko, Riala, & Rasanen, 2014;Tait & Carpenter, 2010, p. 96;Vitt, McQuoid, Moore, & Sawyer, 2017). As the results in Table 3 indicate, this argument is fallacious. ...
Article
Full-text available
Objective To estimate the cross-national association between suicide rates and gun ownership rates Method The association is estimated using the largest sample of nations (n = 194) ever employed for this purpose. Three different measures of national gun ownership rates are related to total suicide rates, firearms suicide rates, and non-firearms suicide rates. Results Although gun ownership rates have a significant positive association with the rate of firearms suicide, they are unrelated to the total suicide rate. Conclusions Consistent with the results of most prior macro-level studies, cross-national data indicate that levels of gun availability appear to affect how many people choose shooting as their method of suicide, but do not affect how many people kill themselves.
... Google Trends has become an important data source for studies in public health surveillance generally [16] and for gun violence research in particular. For example, in past studies, Google Trends was used to assess the effect of mass shooting incidents on public interest in gun control [17], to approximate gun ownership [18], and to predict gun purchases [19] and firearm injuries [20]. In this study, we used Google Trends to assess gun preparation amid the COVID-19 pandemic. ...
Article
Full-text available
Background: Background: National emergencies have increased gun preparation (i.e., purchasing new guns or removing guns from storage) in the past, and these gun actions have, in turn, effected increases in firearm injuries and death. Objective: Objective: We assess the extent to which interest in preparing guns has increased amid the COVID-19 Pandemic using data from Google searches related to purchasing and cleaning guns. Methods: Methods: We fit an Autoregressive Integrated Moving Average (ARIMA) model over Google search data from January 2004 up until the week that President Trump declared COVID-19 an emergency. We use this model to forecast Google search volumes, creating a counterfactual of the number of gun preparation searches we would expect if not for COVID-19, and report observed deviations from this counterfactual. Results: Results: Google searches related to preparing guns have surged to unprecedented levels, approximately 40% higher than previously reported spikes following the Sandy Hook, CT and Parkland, FL shootings and 158% (95%CI 73-270) greater than would be expected if not for COVID-19. In absolute terms, there were approximately 2.1 million searches related to gun preparation over just 34 days. States severely affected by COVID-19 appear to have among the greatest increases in searches. Conclusions: Conclusions: Our results corroborate media reports that gun purchases are increasing amid the Pandemic and provide more precise geographic and temporal trends. Policy makers should invest in disseminating evidence-based educational tools about gun risks and safety procedures to avert a collateral public health crisis. Clinicaltrial:
... (Choi and Varian 2012) survey the system and its many advantages, and demonstrate its utility in predicting motor vehicle sales and unemployment claims. Other specific examples of using Google Trends include creating real time uncertainty indices (Castelnuovo and Tran 2017), predicting cinema sales (Hand and Judge 2012), as an instrument for firearm sales (Vitt et al. 2018), explaining interest in cryptocurrency (Yelowitz et al. 2015;Vitt 2017b), and investigating the impact of e-commerce on retail employment (Vitt 2017a). ...
Article
Full-text available
I address the degree to which variation in exposure to e-commerce is associated with establishment entry and exit in the retail industry at the county level. To measure exposure to e-commerce, I rely on within-state variation in relative search frequency for the phrase “amazon prime” as reported by Google Trends. To generate exogenous variation in this e-commerce exposure measure, I use within state variation in the relative search frequency for “porn” and “cat videos”. Fixed effects instrumental variable estimates suggest at least 10 of the 27 retail industry groups experience net exit with increasing e-commerce exposure, while at least 6 experience net entry. To address endogeneity concerns about my instruments, particularly that they are driven by a notion of “hipster-ness”, I conduct a robustness check to show that my results fail to replicate in consideration of a strategy to tease out this identification threat.
Article
In light of the ongoing debate over tighter firearm regulations, this paper considers the relationship between gun prevalence and suicide. I exploit a reform in Switzerland that reduced the prevalence of military-issued guns in private households. In Switzerland, military service is compulsory for men, and military-issued guns account for nearly half of the total number of firearms available. The results show that the firearm suicide rate decreases by 9% for a reduction in gun prevalence of 1000 guns per 100,000 inhabitants. The elasticity of gun suicides with respect to firearm prevalence is +0.48, but converges towards zero for low levels of gun prevalence. The overall suicide rate is negatively and significantly related to firearm prevalence, which indicates that non-gun methods of suicide are not perfect replacements for firearms.
Article
Full-text available
I address the degree to which variation in exposure to e-commerce is associated with establishment entry and exit in the retail industry at the county level. To measure exposure to e-commerce, I rely on within-state variation in relative search frequency for the phrase “amazon prime” as reported by Google Trends. To generate exogenous variation in this e-commerce exposure measure, I use within state variation in the relative search frequency for “porn” and “cat videos”. Fixed effects instrumental variable estimates suggest at least 10 of the 27 retail industry groups experience net exit with increasing e-commerce exposure, while at least 6 experience net entry. To address endogeneity concerns about my instruments, particularly that they are driven by a notion of “hipster-ness”, I conduct a robustness check to show that my results fail to replicate in consideration of a strategy to tease out this identification threat.
Article
Full-text available
We investigate the impact of a large economic shock on mortality. We find that counties more exposed to a plausibly exogenous trade liberalization exhibit higher rates of suicide and related causes of death, concentrated among whites, especially white males. These trends are consistent with our finding that more-exposed counties experience relative declines in manufacturing employment, a sector in which whites and males are over-represented. We also examine other causes of death that might be related to labor market disruption and find both positive and negative relationships. More-exposed counties, for example, exhibit lower rates of fatal heart attacks.
Article
Background: In 1994, 40% of U.S. gun owners who had recently acquired a firearm did so without a background check. No contemporary estimates exist. Objective: To estimate the proportion of current U.S. gun owners who acquired their most recent firearm without a background check, by time since and manner of acquisition, for the nation as a whole and separately in states with and without legislation regulating private sales. Design: Probability-based online survey. Setting: United States, 2015. Participants: 1613 adult gun owners. Measurements: Current gun owners were asked where and when they acquired their last firearm; if they purchased the firearm; and whether, as part of that acquisition, they had a background check (or were asked to show a firearm license or permit). Results: 22% (95% CI, 16% to 27%) of gun owners who reported obtaining their most recent firearm within the previous 2 years reported doing so without a background check. For firearms purchased privately within the previous 2 years (that is, other than from a store or pawnshop, including sales between individuals in person, online, or at gun shows), 50% (CI, 35% to 65%) were obtained without a background check. This percentage was 26% (CI, 5% to 47%) for owners residing in states regulating private firearm sales and 57% (CI, 40% to 75%) for those living in states without regulations on private firearm sales. Limitation: Potential inaccuracies due to recall and social desirability bias. Conclusion: 22% of current U.S. gun owners who acquired a firearm within the past 2 years did so without a background check. Although this represents a smaller proportion of gun owners obtaining firearms without background checks than in the past, millions of U.S. adults continue to acquire guns without background checks, especially in states that do not regulate private firearm sales. Primary funding source: Fund for a Safer Future and the Joyce Foundation.
Article
Past studies on suicide have investigated the association of firearm ownership and suicide risk in the United States. The aim of the present study was to build on previous work by examining the impact of firearm storage practices and the strictness of firearm regulation on suicide rates at the state level. Data were compiled from primarily three sources. Suicide and firearm ownership information was obtained from the Centers for Disease Control and Prevention. Strictness of handgun regulation was derived from figures available at the Law Center to Prevent Violence, and controls were taken from the US Bureau of the Census. Mixed models were fitted to the data. Household firearm ownership was strongly associated with both suicide by all mechanisms, and firearm suicide. Storage practices had especially elevated consequences on suicide rates. Percent with loaded guns and gun readiness increased suicide rates, and strictness of gun regulation reduced suicide rates. Ready access to firearms can make a difference between life and death. Loaded and unlocked firearms within reach become risk factors for fatal outcomes from suicidal behavior. Future research might want to examine ways of obtaining more recent data on individual firearm ownership. This study proposes several policy recommendations for suicide prevention.
Article
We evaluate Angrist and Krueger (1991) and Bound, Jaeger, and Baker (1995) by constructing reliable confidence regions around the 2SLS and LIML estimators for returns-to-schooling regardless of the quality of the instruments. The results indicate that the returns-to-schooling were between 8 and 25 percent in 1970 and between 4 and 14 percent in 1980. Although the estimates are less accurate than previously thought, most specifications by Angrist and Krueger (1991) are informative for returns-to-schooling. In particular, concern about the reliability of the model with 178 instruments is unfounded despite the low first-stage F-statistic. Finally, we briefly discuss bias-adjustment of estimators and pretesting procedures as solutions to the weak-instrument problem.
Article
The Golden Gate Bridge is currently the number one suicide site in the world. In contrast, the San Francisco-Oakland Bay Bridge, which was completed six months earlier, is located less than six miles away and is the same height, has had substantially fewer suicides. The purpose of this study is to conduct an epidemiological investigation to describe and analyze the differential suicide patterns from these two structures. Using official records available for the years from 1937 through 1979, the data were analyzed with regard to the respective contributions of availability, suggestion, and psychological/symbolic factors as they enter into the choice of suicide method and location.
Article
Using monthly data constructed from futures markets on presidential election outcomes and a novel proxy for firearm purchases, this paper analyzes the reponse of the demand for guns to the likelihood of Barack Obama being elected in 2008. Point estimate suggests the existence of a large Obama effect on the demand for guns. This political effect is larger than the effect associated with the worsening economic conditions. This paper presents robust empirical evidence supporting the hypothesis that the unprecedented increase in the demand for guns was partially driven by fears of a future Obama gun-control policy. Conversely, the evidence for a racial prejudice motivation is less conclusive. Furthermore, this paper argues that the Obama effect did not represent a short-lived intertemporal substitution effect, and that it permanently affected the stock of guns in circulation. Finally, states that had the largest increases in the demand for guns during the 2008 election race experienced significant changes in certain categories of crime relative to other states following Obama s election. In particular, those states were 20 percent more likely to experience a shooting event where at least three people were killed.
Article
Organized suicide prevention efforts usually have been linked to clinical models that use either individual or small group approaches. This paper explores the use of a public health/public model to prevent suicide. Various methods of suicide are analyzed using the concepts of availability and lethality to illustrate means by which they might be modified through the public health/public policy approach of community action and legislative change.