Violence Beyond Reason: Prohibition of Alcohol and the Decline of

Violence Beyond Reason: Prohibition of Alcohol
and the Decline of Lynching in the US South*
Hoyt Bleakley†
Emily Owens‡
This version: April 6, 2011
First version: September 2, 2010
Abstract
Within economics, there exists a tension between the rational
and more psychological approaches to explaining conflict. We
examine the effect of the prohibition of alcohol circa 1910 on
mob violence and legal executions in the US South. We
document that lynching declined markedly when a county went
dry. We suggest that this result is about alcohol's effect on social
interactions, rather than some concurrent shift in policy and
attitudes. We find both a reduction in lynching right after going
dry, as well as a downwards break in the trend. We find no
corresponding reduction in legal executions, suggesting that
reducing the supply of alcohol had a larger impact on more
impulsive uses of force against black southerners. This is
consistent with both a direct effect of alcohol on judgment and
impulsivity, and a slower adjustment via the stock of social
capital.
Keywords: mob violence, temperance, bounded rationality
JEL codes: N42; I18; K42; P48
* Preliminary; please do not cite. The authors thank Jonathan Guryan, Robert Margo, and Miriam Wasserman,
and seminar participants at the University of Chicago, Stanford, Northwestern, and Vanderbilt for helpful
comments, and Jeff Broom and Julianne Zhang for able research assistance.
† Booth School of Business, University of Chicago. Electronic mail: bleakley at chicagogsb dot edu.
‡ Department of Policy Analysis and Management, Cornell University. Electronic mail: ego5 at cornell dot edu.
I.
Introduction
…the abolition of the barroom is a blessing to the Negro second only to the abolition of
slavery. Two thirds of the mobs, lynchings, and burnings at the stake are the result of bad
whiskey drunk by bad black men and bad white men.
-Booker T. Washington (Quoted in the New York Times, March 2, 1908.)
Recent years have seen a surge in economic research on the causes of terrorism and civil
conflict. Early work in this area, reviewed by Krueger and Maleckova (2003), focused on the
impact of poverty on the propensity of individuals to engage in purposeful violence. More recent
papers by Abadie (2006), Alesina et al (1996) have highlighted the role of political inequality as
a contributor to domestic terrorism. Berman and Laitin (2008) point out that, to the extent that
individuals engaged in civil conflict are acting in the interest of a group, policies that promote of
benevolent forms of civic engagement may reduce violent attacks. While an active debate
remains about the role of economic, political, and civic factors in determining acts of targeted
violence, the existing literature has modeled the decision to engage in a terror attacks as a
rational decision.
In this paper, we provide further evidence on the determinants of civil violence by
examining the role of substance use, specifically alcohol, in a specific type of domestic terrorism.
A large literature in public health and economics has demonstrated that alcohol consumption
increases both impulsivity and aggression, contributing to a wide range of socially undesirable
outcomes. Using historical county level data on alcohol regulation, lynching, and legal
executions in the southern United States between 1890 and 1930, we show that the passage of
county ordinances prohibiting the sale of alcohol reduced the incidence of terrorist attacks on
black southerners. Indeed, the time pattern of lynching suggest that passing a “Dry” ordinance
not only immediately reduced the incidence of lynching, within county limits, but each passing
dry year further reduced the probability that a lynching would occur.
While we are unable to directly measure changes in alcohol consumption, the negative
correlation between temperance and lynching does not appear to be driven by other measures of
social and demographic change reflected in decennial census data and county voting patterns.
We argue that the reduction in terror attacks against black southerners was most plausibly caused
by the reduced alcohol consumption of white southerners.
2
We posit several distinct mechanisms to explain the timing of the observed relationship.
Our preferred explanation is twofold. First, temperance led to the immediate closing of legal
saloons, raising the price of alcohol. To the extent that alcohol is a normal good, the subsequent
decline in alcohol consumption meant that there were fewer drunken mobs to victimize black
southerners. Over time, dry ordinances may have affected social capital, as young men of both
races spent less time in saloons, and more time engaged with family members or in other social
activities. A small variant of this mechanism that fits the data is that county-level prohibitions
were implemented slowly, although historical evidence suggests that saloons were closed almost
immediately. The incidence of lynching was declining over time throughout the south, but
restricting the availability of alcohol appears to have accelerated that decline.
Our finding that prohibitions on the commercial sale of alcohol reduced the incidence of
lynching relates to multiple strands of literature in applied economics. In addition to the growing
field of research on the causes of civil conflict, our findings contribute to the large literature in
economics examines the relationship between alcohol and violent behavior (Carpenter, 2008;
Cook, 2008; Cook and Moore, 1993; Joksch and Jones, 1993). In addition, we provide further
evidence in the debate over the social impacts of the American Temperance Movement, which
culminated with the passage of the 18th amendment banning the sale of alcohol in the United
States (Owens, 2011; Cook, 2007; Dills and Miron, 2004; Miron 1999). Finally, to the extent
that lynching represents the dissatisfaction of white southerners with the rule of law, we provide
empirical evidence on the substitutability of formal and informal political institutions (Kranton,
1996; Dixit, 2003; Tabellini, 2008).
In the context of this paper, terrorist attacks against southern blacks can be thought of as
the result of southern whites choosing informal enforcement of law over the formal legal system;
by definition, lynching is an illegal murder committed by one or more persons in response to the
perceived violation of a social code or norm. While not all lynching victims would have been
convicted and executed by the state, we present historical evidence that legal execution and
lynching were to some extent substitutes during this time period.
Theoretical research on informal and formal contract enforcement has typically framed
informal contract enforcement as based in cooperation, which can be more beneficial to all
parties involved under certain conditions (when there are high levels of trust, or some cultural or
geographic closeness between parties). In large groups, where there is limited trust or restricted
3
dissipation of information, agents are willing to pay a cost to gain access to formal third party
enforcement. In the post-Reconstruction South, whites who felt victimized by black southerners
could resolve these disputes in two ways. Bringing formal legal charges against the suspect
involved waiting for a trial to take place, and there was some probability that the accused would
not be found guilty. Alternately, white southerners could resolve the dispute using informal
means- appealing to the white public. Punishment administered informally, lynching, was both
certain and swift.
Using data on all executions between 1890 and 1930, we show that reducing the supply
of alcohol had a larger effect on terrorist acts against black southerners than the use of the formal
sanctions against them. Specifically, we cannot reject the null hypothesis that legal executions
of black (or white) southerners are unrelated county dry ordinances. This null effect is consistent
with dry ordinances affecting the propensity of whites to engage in group violence, as opposed to
a change in the ability of black southerners to avoid conflicts with white southerners.
The paper proceeds as follows: In Section II, we briefly summarize the temperance
movement in the southern United States and the wave of southern lynching between 1890 and
1930. Drawing on the public health literature on alcohol and violence, we use the history of the
two phenomena to motivate a causal link between temperance and lynching. We describe our
data on county level dry laws, lynching, legal executions, and other social change in Section III. .
Section IV presents our primary results, in which we use event study methodology to examine
how lynching changes when a county goes dry. We contrast this with the absence of any strong
relationship between temperance and legal executions of blacks or whites. In Section V we
present evidence in support of our causal interpretation by including a number of measures of
other social changes that may have reduced lynching. Finally, we conclude with a brief
discussion in Section VI.
II.
Background
a. The Temperance Movement in the South
The American Temperance Movement is conventionally divided into three waves, the
first lasting from 1850 to the start of the Civil War, the second from 1880 to the 1890s, and the
third beginning in 1907 and culminating in the passage of the 18th amendment. The first
temperance wave, corresponding with the Protestant Second Great Awakening, failed to gain
traction in the south because of the affiliation of movement leaders with anti-slavery causes
4
(Szymanski 2003). The leaders of the next wave did not share the same commitment to racial
equality (Szymanski 2003), and the temperance movement was unusually successful in the
southern states. Indeed, even though national temperance organizations were bankrupted by the
panic of 1893, the number of people living in dry areas increased almost every year between
1895 and 1906, in large part due to the popularity of local dry laws in the former Confederacy
(Cherrington 1920).
Prior to Federal Prohibition, there were generally three ways in which an area could
become dry. First, the state legislature could pass a temperance law, forcing all counties to
become dry at once. Alternately, voters in individual municipalities could approve motions to
ban alcohol in that area, which would be submitted to the state legislature for approval. This
particular path to prohibition was popular in the south; between 1872 and 1874, the Alabama
state legislature approved 114 local prohibition laws. Finally, the state legislature could pass a
general local option law, pertaining to either the entire state or in a subset of areas. Local option
laws allowed local voters to decide whether to allow liquor sales, with the outcome of the local
election respected by the state. Georgia passed a general local option law in 1885, followed by
Mississippi in 1886, Kentucky in 1906, and South Carolina in 1909 (Cherrington 1920).
The small geographic scale at which these laws operated meant the residents could travel
to nearby wet areas to purchase alcohol, but the local option movement was believed by national
temperance groups to be a path towards eventual state level regulation (Krout 1925). That
expectation was not always realized; in 1907, 825 of the 994 counties in the former confederacy
were dry, but none had a state-wide law (Woodward 1994). Indeed, in 1911, a popular
referendum on statewide prohibition in Texas was defeated, even though 168 of 248 counties
were dry under local option, and only one dry county had voted to become wet since 1908
(Cherrington 1920). The disproportionate success of local dry laws compared to broader state
laws reflects the popularity of temperance movement in more rural areas; state-wide laws were
unlikely to pass without support of the urban middle class (Blocker 1976). Both local and state
dry laws were also more likely to succeed in areas with a large native-born population, protestant
population, and in counties with fewer taverns (Lewis 2008).
While the correlation between urbanization, immigration, religion and temperance is
reasonably well established, less obvious is the relationship between the temperance movement
and race relations. Black community leaders, including Henry McNeal Turner, the National
5
Association of Colored Women, and Booker T. Washington supported temperance (Litwack
1998), but so did notorious racists such as Benjamin Tillman (Cherrington 1920; Hamm 1995).
At the same time that southern county governments were increasingly outlawing the commercial
sale of alcohol, “two or three black Southerners were hanged, burned at the stake, or quietly
murdered every week” (Litwack (1998):284). To the extent that temperance laws were more
easily circumvented by people with high income (Cook 2007), dry ordinances may have
primarily reduced alcohol consumption by black southerners and poor whites. There is some
evidence that the disproportionate impact of temperance on black southerners was not
overlooked by leaders of the temperance movement.
b. Black Code, Temperance, and the Courts
Over the course of three years, the enactments of the 13th and 14th amendments elevated
the status of black southerners from property to person. The rise of lynching, like the passage of
Jim Crow laws, was, to some extent, the result of the conflicting aspirations of black Americans
and resentment by southern whites who coveted the idealized pool of bound labor and social
order to which their grandparents had access. In particular, unlike the former slaves exiting the
labor force, members of the “uppity” new generation were able to read their pay stubs and
contest unfair compensation (Higgs 1980, Litwack 1998). After federal law reduced their power
in the market sector, white southerners used their control of the non-market sector to construct an
informal legal system that came to be known as “Negro Law” (Litwack 1998), or “Black Code”
(Ransom and Sutch 1977, Higgs 1980).
The bulk of Black Code placed constraints on interactions between whites and blacks,
always benefiting the white party. For example, white employers were protected from being sued
by black employees for unpaid wages (Higgs 1980), but it was criminal for black laborers to
break employment contracts with white employers (Ransom and Stuch 1977). Social
interactions were similarly one-sided; any sexual contact with a white woman constituted a
capital offense for black men under Black Code, but white men were rarely penalized for raping
a black woman.1 Indeed, official records of legal executions reflect a heavy penalty for sexual
offenses committed by black men. Since 1976, none of the people executed in the United States
had been convicted of non-fatal crimes. However, between 1890 and 1920, 15% of executed
1
Using logic that is appalling in its attitudes towards both women and people of color, under Black Code the
“lascivious nature” of black women precluded the possibility nonconsensual sex (Litwack 1998).
6
black men were convicted of sexual offenses that did not involve murder, compared to less than
1% of white men.
The southern temperance movement was also shaded by Black Code. Alcohol
consumption has been shown to increase aggression in laboratory experiments and observational
data (Miczek et al. 1994, Cook and Moore 1993, Joksch and Jones 1993, Markowitz and
Grossman 2000, Carpenter 2008, Dobkin and Carpenter 2008), and the relationship between
alcohol consumption and violent behavior was understood in the early 20th century as well.
Intoxicated white men might squander their income and beat their wives (Cook 2007), but in the
post-Reconstruction South, the potential harm caused by drunken black men was considered
even larger. Liquor “made the negro tenfold more than ever a menace to the safety of the white
man’s home.” Public health officials reasoned that “intoxicants act more quickly and more
acutely on the brain, emotions, and passions of the colored man.” Contemporary observers
noted that “use of liquor by Negroes [...] is and has been an ugly factor in the friction and
quarrels between the races in the South.”
Fear of black violence became an argument for prohibition: “an inferior race cannot
remain peaceable … if the crime-suggesting and passion-stimulating saloon is left open freely to
all.” (p243) Limiting black consumption of alcohol was seen as a way to reduce violence overall.
After the Atlanta race riot of 1906, when thousands of whites stormed the city for four days, the
city government responded by shutting down over 80% of city saloons that served black
customers (Brundage 1993).
Consistent with the perceived disproportionate harm of alcohol consumption by blacks,
monetary penalties levied on black bootleggers were over six times the size of fines levied on
whites (Cherrington 1920). As such, the passage of dry laws provided yet another legal means
for white southerners to restrict the upward mobility of black southerners.
c. Lynching and Legal Executions in the South
Black Code allowed the public to “take the law into their own hands” when formal legal
outcomes were considered unsatisfactory (Wright 1990). Lynching was not a random act of
violence against blacks- “it was important that some sort of infraction had taken place”
(Brundage 1993 p 49). As in the legal system, punishments executed under Black Code
maintained a perverse standard of proportionality; black southerners only suspected of
committing crimes were typically hanged- burning was reserved for those who confessed
7
(Wright 1990). Speaking against the passage of a federal anti-lynching legislation in 1934, one
of the founders of the Association of Southern Women for the Prevention of Lynching,
specifically voiced concerns that “courts would be prostituted before community demands for
vengeance” (Hall 1993, pg 243).
It is plausible that some victims of lynching did in fact commit the crimes in question,
and would have been legally executed. Two key features distinguished informal and formal
executions of black southerners- conditional on some perceived offense occurring, the
probability that a black southerner would be found guilty was substantially higher and the time
from accusation to execution was also significantly lower, under the informal system.
Differing standards of proof meant that individuals adjudicated not guilty in a court of
law could still be punished by lynching. The historical record is replete with examples of crowds
killing black people who were formally acquitted: Tom Miles was hung and shot after being
acquitted in Shreveport, LA of sending a harassing note to a white girl, Richard James was hung
after being acquitted of murder (Ginzburg 1960). Henry Askew and Ed Russ, who were lynched
in Mississippi city in 1900, had simply been “in the vicinity” of a murdered white girl. George
Reed was lynched in Rome, GA, as was Walter Johnson in Princeton WV, even though the
alleged victims expressed doubt that they were the culprits (Ginzburg 1960).
Failure by the courts to act swiftly also incited mob violence; Richard Coleman (of
Maysville KY) and George Well (of Scammon, KS) were abducted from jail, and murdered
before they could stand trial (Ginzburg 1960). While many white Americans, including some
white southerners, expressed some dismay at the incidence of lynching (cite), the swift and
certain punishment of informal law was generally considered an effective way to deter the
“inappropriate” behavior of black southerners; while serving as governor of South Carolina,
Benjamin Tillman affirmed his willingness to “lead a mob in lynching a Negro who had
committed an assault upon a white woman.” Under Black Code, state courts were
disproportionately likely to convict and execute black men, but the informal legal system
allowed white southerners to punish the alleged offender while the loss associated with the crime
was still salient.
d. Temperance, Lynching, and Legal Executions
National time series evidence suggests that the temperance movement was associated
with increased violence due to the proliferation of underground markets (Dills and Miron, 1999;
8
Miron, 1999). However, recent research exploiting panel data on homicides and temperance
laws has found evidence that, in most states, dry laws were in fact associated with reductions in
the homicide rate (Owens 2011). The net negative effect of temperance on homicides, even in
the presence of a growing underground market, is due to the strong positive relationship between
alcohol consumption and aggression. Based on the historical record of the temperance
movement and conflict between southern blacks and southern whites, we suggest three
distinguishable hypotheses for the relationship between lynching and legal executions, as they
relate to the county going dry:
(1) Alcohol alters judgment, and in particular appears to increase the propensity to inflict
harm on others, as well as the perceived willingness of others to inflict harm on you
(Miczek et al., 1996). Under Black Code, white southerners were able to use both
formal and informal law to penalize black southern who they perceived as threatening
to the old social order in general and white women in particular. To the extent that
prohibition reduced alcohol consumption and perceptions of aggression by white
southern, and the probability that white southerners would behave aggressively
toward black southerners, fewer negative interactions between black and white
southerners should have occurred. As a result, the incidence of punishment, meted out
by the formal and informal legal systems, should decline in dry areas.
(2) Not only will an increase in the cost of obtaining alcohol reduce the rate of conflict
between black and white southerners, reduced alcohol consumption by southern
whites may decrease the use of informal sanctions against black southern when
conflicts did occur. Lynching occurred in the presence of discrepancy between the
judgment of the courts and the informal judgment of the mob. To the extent that
availability of alcohol altered the judgment of white southerners, dry laws may have
reduced the propensity of white southerners to engage in terrorist attacks against
black southerners after a conflict occurred.
If “cooler heads prevailed” after perceived offenses by black southerners in dry
counties, the change in the rate of legal executions is ambiguous. As previously
described, an unknown fraction of lynching victims might have been adjudicated
9
guilty and executed if the formal legal system had been used. The substitution of
formal for informal sanctions would tend to increase the number of legal executions.
However it is plausible that most victims of informal sanctions would not have been
prosecuted, even under Black Code. Whether a reduction in the propensity of white
southerners to lynching black southerners would increase the number of criminal
trials enough to offset the reduction in the number of racial disputes is an empirical
question.
(3) Dry laws that disproportionately reduce alcohol consumption by poor Southerners
should reduce conflict overall, reducing the contexts for lynching and for legal
prosecution. In this case, dry ordinances should be associated with reductions in
lynching, reductions in the legal execution of black southerners, and also plausibly a
reduction in the execution of poor white southerners, who should also be consuming
less alcohol.
III.
Data
In order to estimating the relationship between alcohol restrictions, lynching, and legal
executions, we draw on data from six sources, which we use to construct a panel data set of
county level measures of lynching, legal executions, restrictions on alcohol sales, and estimates
of racial composition, urbanization, education, and political attitudes.
Several sources of historical data on lynching exist. Our data were assembled by Tolnay
and Beck (1995), and were originally drawn from a combination of NAACP records, special
reports by the Chicago Tribune, and the Department of Records and Archives at Tuskegee
University, all checked against newspaper and other first-person accounts. For each recorded
lynching, we know the date and location of the murder, and some information on the pretexts of
the lynching. We aggregate this information by county and year to construct a panel of lynching
incidence in all counties in Alabama, Arkansas, Florida, Georgia, Kentucky, Louisiana,
Mississippi, North Carolina, Oklahoma, South Carolina, Tennessee, Texas and Virginia between
1895 and 1930. Dots in Panel A of Figure 1 represent the intensity of lynching in these states,
which large dots representing places where more lynching occurred. For sake of comparison, we
10
overlay this measure of lynching on county level estimates of the fraction of the county that was
black in 1890. Note that lynching is concentrated in the “black belt” of Louisiana, Alabama,
Arkansas and Mississippi, but also occurred in places with small black populations, like Florida
and Texas.
Our data on county level dry ordinances is drawn from Sechrist (1983). This data set
contains information on the “prohibition status” of each county in the United States between
1801 (or whenever the county was created) until 1920. The data indicate whether or not a county
was dry−whether or not alcohol could legally be sold in that county−for that year. Panel B of
Figure 2 plots the proportion of all southern counties that were dry each year between 1895 and
1930. Georgia was the only southern state that had a significant dry presence during the first
wave of prohibition; only a handful of counties across Kentucky, North Carolina and Tennessee
were dry for a few years before to 1870. The temperance movement gave little ground; after
1895, only Arkansas and Virginia had a significant number of counties convert from dry to wet
status. There are three distinct breaks in the spread of prohibition. Between 1895 and 1903, there
was a slow increase in the fraction of dry counties. In 1904 the proportion of dry counties
increased from 11 to 50%; only Alabama, Arkansas and Tennessee did not see large increases in
dry territory in this year. In 1908, counties in these three states adopted dry laws, causing the
last jump in dry coverage.
County boundaries were in flux during the early 20th century, and not all lynching
occurred in modern counties. For example, Forrest County, Mississippi was created out of Perry
County in 1908, but lynchings in 1895, 1899 and 1903 were recorded in the area that would
become Forrest. Since we know that the laws of Perry County applied in “Forrest” at these times,
we code these three lynchings as occurring in Perry. Lamar County, Mississippi was similarly
formed out of Marion County, and so pre-1904 lynching recorded as occurring in Lamar is coded
as occurring in Marion. On the other hand, lynching incidents were recorded as taking place in
George County (MS) and Houston County (AL) before those counties existed. Unlike Forrest
and Lamar, these two counties were created out of parts of multiple counties, so we exclude
those lynching events from our data.
We also exclude five Virginia lynchings, three South
Carolina lynchings, and one Alabama lynching that occurred in extinct counties for which we
have no prohibition information.
11
We measure the number of legal executions using data collected by Espy and Smykla
(2004). These data, often referred to as the “Espy File” were collected over a period of 30 years
and contain information on almost all executions that took place in the United States from 1608.2
These data are considered to be the most complete historical record of executions available. For
each execution, we observe the county in which the offender was convicted, the race of the
offender, and with some degree of certainty, we know the controlling offense for which the
individual was executed and the method of execution. For the purpose of our analysis, we
aggregate the total number of executions occurring in each year and in each county by race and
controlling offense.
In our primary sample of analysis, roughly 40 people were legally executed each year
between 1895 and 1915, a total number which fell to 30 a year after 1915. While black
southerners represented roughly 38% of the total population, 78% of people executed in southern
counties in this time period were black. Unlike lynching, the overrepresentation of black people
on death row was not a Southern phenomenon; 5% of the Northern population was black,
compared to 34% of those executed. Between 1895 and 1923, there were on average 2.4
lynchings per executed black southerner in our sample. After 1923, the prevalence of legal
execution surpassed lynching, and there were on average of 2.5 executions per lynching.
Finally, we estimate social and demographic change in these counties using Census and
Election data. For each county in each decennial census year, we observe the total population,
the fraction of the population that is black or colored, the fraction of the county that lives in a
place with more than 2,500 people, and the number of acres devoted to farmland. These data are
recorded in Haines (2004). We then estimate political attitudes using county election returns in
statewide elections, as recorded in the ICPSR United State Historical Election Returns data. For
each election year in each county, we calculate that fraction of all votes cast that went to the
either national or local affiliates of the Democratic party, the Republican party, the Populist party,
or the Prohibition party. For both the census and election data, we construct annual estimates by
linearly interpolating values between observation years.
2
The data exclude older executions of which no record remains. Louisiana and Mississippi used portable
electric chairs to conduct some executions during the 1940s and 1950s, and those data are missing as well.
12
IV.
Main Results.
IV.A. Realign the data by year going dry
To better gauge the relationship between prohibition and lynching, we realign the data
according to the year each county goes from wet to dry. (For the very few counties that have
multiple transitions, we use the last wet-to-dry transition in the sample window.) We dub this
renormalized measure of time “years since going dry”, or k. Figure 3 displays summary
statistics of the sample, decomposed by year since going dry. Panel A presents the number of
counties present according to renormalized time. The panel of counties is approximately
balanced for |k|<10 years, but the numbers start to drop as counties disappear out of the sample
window. This is because some counties had their transition to try status either very early (before
1895) or very late (close to 1919, when national prohibition took effect). But, as we saw above,
the bulk of wet-to-dry transitions happened between 1903 in 1910, which is compatible with our
sample window. As discussed above, we restricted the sample to the range shown in Figure 3
because there are so few counties that lie outside of this window. Panel B displays the fraction
of counties that are dry as a function of the re-normalized time variable. By construction, this
fraction is equal to unity for k>0. This fraction is quite close to zero for k<0, which reflects the
fact that the vast majority of counties experienced only one transition from wet to dry in the
sample window. Realigning the data in this manner allows us to sharply contrast county/years
that were wet versus those that were dry.
IV.B. Lynching outcomes relative to year going dry
Next, we examine lynching as a function of years since going dry (k). These results are
found in Figure 4, Panels A and B, which display the proportion of counties with a lynching and
the number of victims lynched per county, respectively, as a function of k. The average for each
re-normalized year is marked with a dot.3 Because we do not know at what point in the year that
prohibition took effect, we do not display the point corresponding to k=0.
In Figure 4, there appears to be both a jump down as well as a downward rotation of the
trend that coincides with the county going dry. For k<0, the fraction of counties with a lynching
is approximately 9%, while the average number of victims lynched is around 0.13. Prior to
3
Note that taking averages by k is numerically equivalent to running a regression with a full set of
dummies for k.
13
going dry, each series exhibit trends that are slight relative to the range of variation in the figures.
In the few years after going dry, however, the average lynching rates drops by between 0.01–
0.02. Further, the series after k=0 exhibit a markedly faster trend downwards.
The changes in lynching associated with going from wet to dry are both economically
and statistically significant. The jump downwards after going dry is around 20-25% of the preProhibition levels for each outcome. The break in the trend after going dry is also quite large;
essentially all of the decline in lynching over the 25 years that follow going dry is due to the
downward acceleration of the lynching trend that follows prohibition. We assess the importance
of these breaks statistically by running a regression on the data points that are displayed in
Figures 4 and 5. The specification includes a trend, k, in years relative to the year going dry; a
level shift following the wet-to-dry transition, I(k>0); a shift in the trend at the same time,
k×I(k>0); and the constant term. The results for the regression with average number of victims
lynched on the left-hand side is as follows:
Lk = -.0329
I(k > 0)
(.0166)
{.0161}
+ -.0043
k × I(k > 0)
(.0016)
{.0015}
+
.0009
k +
(.0016)
{.0014}
.1314
+ ωk ,
(.0139)
{.0164}
in which the coefficient estimates are displayed in front of the variables, and the numbers below
are standard errors. The standard errors are computed using a bootstrap on the entire
county×year data set, which takes into account the fact that this is a two-step estimator. The
standard errors reported in parentheses are computed with a standard bootstrap that iteratively
recomposes samples drawn randomly from original county×year data. In contrast, the standard
errors reported in curved brackets are computed from a block bootstrap in which states are used
as clusters when creating the new bootstrap samples. (The block bootstrap with state clusters
adjusts for serial correlation within states.) For either set of standard errors, the estimated
coefficients are significant at conventional confidence levels, except for the trend (k). Results are
similar for the proportion of counties with a lynching.4 These results confirm what was seen
4
Specifically, the estimated equation for the proportion of counties with a lynching is
14
above in the Figures: following going dry, there was a jump downwards in average lynching and
a shift to a downward trend.
These results suggest an effect of prohibition on mob violence working through two
different channels, at two different horizons. The drop in lynching in the years immediately
following prohibition is consistent with a direct effect of alcohol consumption on propensity of
white southerners to engage in violence towards black southerners. This is consistent with some
of the medical and psychological evidence that was discussed above. Note that this could also be
driven by alcohol’s impact on impulsive behavior and time-discounting. If racial conflicts
continued to occur at the same rate, but white southerners were more likely to use legal sanctions
against black southerners in dry states, we would observe a reduction in lynching, but a weak
increase in legal executions.
The trend break is consistent with multiple theories of the impact of alcohol on behavior.
Temperance might have an effect on social capital, in particular which social groups form, what
activities they choose to adopt and where people tend to congregate. It is conceivable that a
secondary, indirect effect of alcohol is that it brings together groups of people who are of such a
mindset that they are more prone to violence when they are together. Put more colloquially, the
consumption of alcohol might cause the “bad element” to get together and reinforce one
another’s antisocial tendencies. These types of groups may have been more prone to engage in
lynching, and were directly implicated in various lynchings (Gizburg, 1964). The closing of
saloons would have immediately disrupted these networks, but social groups which previous met
in the saloon may have only gradually disbanded when the county went dry.
Another mechanism that would generate slow adjustment could be the alcohol works
through a cohort rather than a time effect, perhaps because early exposure to alcohol (directly or
via one’s parents) affects one’s judgment or attitude towards violence. This cohort-based
mechanism would imply, however, that the average age of either the victims or perpetrators
L!
=
-.0168
I (k > 0)
+
-.0020
k ! I (k > 0)
+
-.0004
k
+
.0881
+ "
k
k
(.0082)
(.0010)
(.0010)
(.0075)
{.0094}
{.0009}
{.0009}
{.0099}
in which the standard errors are constructed as described in the text above, and reported coefficients are
significant at conventional confidence levels, except for the trend (k).
15
would rise over time, a supposition that is not supported by anything that we have found in the
historical record. Finally, the observed trend is also consistent with a secular downward trend in
lynching over time, which is offset in by the increased rates of racial conflict in wet states. In
this case, the observed discrete drop in the probability of lynching is driving any trend break,
rather than a separate mechanism.
IV.C. Falsification exercise using random wet-to-dry transitions
A falsification exercise suggests that these results are not simply an artifact of having
realigned the data. We randomly assign a wet-to-dry year to each county, and then estimate the
jumps and trend breaks associated with this spurious prohibition. Repeating this exercise for
numerous iterations, we ask the following question: “how often would we estimate breaks as
large as the ones we found above?” Note that this exercise is similar to one conducted by
Bertrand, Duflo, and Mullainathan (2003) in analyzing the possibilities for false positives when
estimating difference-in-difference models. To generate random years of Prohibition, we use the
empirical distribution of wet-to-dry transition years within each state in our sample. (Again, we
use the state as a cluster for this procedure for the same reasons as stated above. Our results here
are not materially affected, however, by drawing the year from the entire distribution of wet-todry years in the South, or using a uniform distribution of years across the sample window.)
The results from this exercise are found in Figure 7. Each graph displays the distribution
of point estimates for the indicated outcome (proportion with a lynching and number of lynching
victims) and the jump and trend-break coefficients. This distribution is estimated from 3000 sets
of random breakpoints. Vertical lines in each graph denote the respective coefficients estimated
above. It is noteworthy that the point estimates from above lie in the tails of the distribution
estimated from spurious breakpoints. For the average number of victims lynched (Panels A.2
and B.2), only about one time in 100 would we estimate a coefficient from the spurious
breakpoints that is more negative than what we estimated using the actual wet-to-dry transitions.
For the proportion of counties experiencing a lynching, we would estimate such a large
downward jump with the spurious breakpoints less than one time in 50, and such a large negative
trend break less than one time in ten.
IV.D. Additional sensitivity analyses
The findings above regarding the jump and trend break are not sensitive to a variety of
16
assumptions regarding the specification or sample. We discuss these results in this subsection,
although the accompanying figures are found in Appendix B, to save space. First, we obtain
similar results when controlling for fixed effects for county, fixed effects for year, state–year
dummies, or interactions of year dummies with various county characteristics (pre-1895 lynching
rate, latitude and longitude, or hookworm and malaria rates circa 1915). Appendix Figure B-1
presents graphs that were constructed using lynching data that had been residualised onto these
control variables before taking the averages over k. Second, we attain similar results for the
break at k=0 if we construct the breakpoint based on the first, rather than last, transition from wet
to dry. (This is to be expected, given that almost all counties in our sample experienced only one
wet-to-dry transition.) These results are seen in Appendix Figure B-2.B. Third, we obtain similar
breaks in the lynching data around the time of going dry if we use a sample that includes all of
the counties in the South or a sample that includes only counties that had lynching incidents
recorded prior to 1895.
V.A.
Legal executions
We examine the Espy data on legal executions at the county/year level to further explore
the relationship between alcohol, lynching, and the use of legal sanctions against black
southerners. We reviewed the reported pretexts for lynchings in the Tolnay/Beck data, and
found that approximately 85% of the pretexts could be regarded as capital crimes, such as
murder and rape.5 Recall, however, that a lynching did not come with the same burden of proof
or due process associated with the albeit-imperfect criminal trials of the period. Indeed, several
of the lynching victims were identified after the fact as being victims of mistaken identity; even
worse, in one instance, a person who attempted to intervene by pointing out that the mob had
gone after the wrong person was himself lynched. But, it is likely that some of the lynchings
motivated by ‘rape’ were in fact in response to consensual relations between a black man and a
white woman (see discussion in Section 2).
Turning back to the data, we do not find evidence of a significant jump in executions
legal executions of blacks when a county goes dry. These results are found in Figure 6, where we
5
Rather than use our contemporary notion of what constitutes a capital crime, we examined the causes for
legal executions in the Espy data as a point of comparison. The majority of executions were for murder, and the
second most common offense was rape. There were a number of smaller offenses such as arson, burglary, and horse
stealing. These pretexts were also seen in the lynching data.
17
construct average execution rates as a function of time since going dry, as we did above in
Section IV. Panel A displays the results for executions of blacks. Using bootstrapped estimates
of the same equation as above, we do not reject the null hypotheses of there being no jump and
no trend break at the year of going dry. Further, we can reject the alternative hypotheses that the
series jumps and rotates at k=0 with equal and opposite magnitude to that estimated for the
lynchings above. This latter result would seem to reject the idea that legal and extralegal
executions were substitutes. Neither do we see a jump down in legal executions of blacks, which
indicates that the decline in lynching that we related to prohibition was not induced by a decline
in conflict between black and white southerners. The lack of any significant change in the rate of
legal executions of blacks also disfavors the idea that there was some contemporaneous policy
change affecting law and order in dry counties. Indeed, the apparent lack of a relationship
between prohibition and legal executions would seem to favor the idea that lynching and legal
sanctions were imperfect substitutes, and the lynchings averted in dry counties were for pretexts
that would not have been judged capital crimes by the legal system.
Results for the legal executions of whites in relation to the time since going dry are
found in Panel B of Figure 6. We do not see evidence of a jump down in white executions
following a county going dry; indeed, there is weak evidence of a shift up instead. We cannot,
however, reject the null hypothesis of no break in this time series at the year the county goes dry.
The lack of a concurrent decrease in executions of whites suggests that the results above for
lynching are not due to a general decrease in white criminality, but rather something specific to
alcohol’s contribution to the behavior of a lynch mob.
IV.E. Difference-in-difference estimates
In Table 1, we present the results of a standard difference-in-differences analysis, where
we compare the change in the incidence of lynching over time in dry and wet counties. This
complimentary specification allows us to easily control for the other observed demographic
changes in these counties, as well as estimate heterogeneity in the relationship between
temperance and lynching. We also test the sensitivity of our event study analysis to the
functional form that we use to model lynching, as well as different samples of counties.
In the first row of table 1, we present results from a multivariate analysis of lynching and
dry ordinances. When we include only county and year fixed effects (column 1) we estimate that
18
lynching was 2.4 percentage points (se=1.1) less likely to occur in dry counties. This effect is
robust to including state-specific time trends (column 2), controls for the fraction of the county
population that was black (column 3), and the fraction of the county population engaged in
sharecropping (columns 4). When we replace the dependant variable with the number of
lynching victims, we estimate consistently estimate that there were, on average, just under 0.05
(se=0.02) lynching victims per county per year once counties went dry.
Lynching was a rare event, and we test the sensitivity of our results to variation in the
modeling specification in row 2. In the first column, we replace our county-clustered standard
errors with bootstrapped standard errors clustered at the state level. In columns two and three we
estimate mean marginal effects from a probit and conditional logit, respectively, and also find
that going dry is associated with an average 2.4 percentage point reduction in the probability of a
lynching. Turning to our continuous measure of lynching, we find that our results are more
sensitive to modeling choice- treating the number of lynching victims as an ordered logit reduces
the estimated marginal effect by over ½, but is still statistically different from zero.
In our primary sample, we include exclude counties that never had a lynching, which is
justified on the grounds that such areas are likely structurally different from counties where such
violence did occur. We examine the sensitivity of our analysis to this sample restriction in the
third row of Table 1, which re-estimates the model in column (1), row (1). First, we attempt to
limit contamination of long-run trends in lynching by focusing first on the seven years before
and after counties went dry. We lose precision, but our point estimate is the same, corresponding
to a 2.4 percentage point drop. When we exclude the years 1904 and 1906, we find no
substantive difference. Perhaps surprisingly, the relationship between alcohol availability and
lynching is driven by areas outside of the Deep South (see columns 3 and 4). Our baseline
estimate of the impact of temperance on the number of lynching victims each year is more robust
to variation in the length of our panel (columns 5 and 6). As with the binary measure, dry
ordinances had a larger impact on the number of lynching victims outside of the Deep South.
Finally, in the last row on Table 1, we test for heterogeneity in the impact of dry laws on
violence along two dimensions. To the extent that county-level dry laws were easily
circumvented by crossing the border, we might expect that the effective price of alcohol in a dry
county would be an increasing function of the fraction of neighboring counties that were also dry.
Regardless of modeling specification, we fail to find evidence that lynching was any more or less
19
likely to occur in a dry county surrounded by dry counties than in a dry county surrounded by
wet counties. We do, however, find suggestive evidence that the likelihood of any lynching
occurring and the number of lynching victims fell more in dry counties that had more lynching
prior to 1895. When we estimate this model with ordinary least squares, we find statistically
significant differences in impact of temperance on lynching in counties where this was common
before 1895. However, when we use non-linear models, which arguably better fit the data, the
heterogeneity is no longer statistically detectable.
V.
Other outcomes
In this section, we relate various other outcomes to when a county goes dry. To some
extent, the analysis here is as a specification check for our even study model above; some
variables should not jump in response to prohibition, nor would we expect them to be changing
abruptly just prior to going dry. In contrast, some of the variables might plausibly respond to the
prohibition of alcohol and the change in lynching.
V.B.
Census data on demographics and agriculture
We do not find evidence of an effect of going dry on various outcomes from the
decennial Censuses. This analysis is seen in Figure 7. First, because these outcomes are only
observed every 10 years, we check for any differences in the sample size on either side of the
k=0 year, which is seen in Panel A of Figure 7. Because a disproportionate fraction of the
transitions happen in a year ending in 4 or 8, there are some periodic spikes in the proportion of
counties that are in a census year as a function of years since going dry. However, this does not
seem to be systematically changing on either side of the transition from wet to dry. Next we
consider (in logs) total population, percent black, and percent urban in Panels B, C, and D,
respectively. None of these outcomes show signs of shifts or trend breaks near k=0. Finally, we
consider total farm acreage and average farm size (again, both in natural logs) as a function of
year since going dry. These results are seen in Figure 7, Panels E and F, respectively. Neither
outcome shows a break coincident with going dry. None of these outcomes yield statistically
significant differences if we estimate the jump/break equation as above.
20
V.C.
Voting data
Finally, we consider data from elections. Because election are not held every year, we
again check to see whether some unusual break at going dry in the proportion of counties in an
election year. This is seen in Figure 8, Panel A. Transition years are more likely to be in a year
that follows an election-year, which induces periodicity in this figure. But there is no apparent
jump or trend break in this series. Panel B of Figure 8 reports the natural log of total votes cast.
This series would seem to indicate a downward trend prior to going try followed by an upward
trend afterwards. This is consistent with lynching being associated with vote suppression. In
Figure 9, we examine vote shares by various political parties. Panel B contains the series for the
Democratic Party, which was the dominant political party in the region in that period. We also
examine vote shares for the Republican Party, Populist parties, and Prohibition parties, in Panels
D, E, and F, respectively. Vote shares for the major parties (Democratic and Republican) seem
unrelated to going dry. Vote shares for Populists were on the wane before most counties went
dry. Vote shares for Prohibition parties do seem to be affected by the county going dry, in that
there is a markedly downward shift of the trend.
VI.
Conclusions
Both modern and ancient societies are marked by incidents of organized group violence
that seem difficult to justify. Multiple theories have been put forth to explain why groups of
people engage in such acts, primarily focused on the ability of political institutions to resolve the
conflicts used as justifications for acts of often brutal aggression. In this paper, we examine the
role of an alternate type of policy intervention on civil conflict- the availability of alcohol.
Between 1895 and 1920, a successive number of counties in the southern United States passed
dry ordinances banning the commercial sale of alcohol, and in 1920 all alcohol sales were
banned by the 18th amendment. We posit a relationship between the temperance movement
based on a large literature in health economics linking alcohol consumption to both aggression
and impulsivity.
Using data on county level dry ordinances and organized violence against black
southerners, we show that local restrictions on the availability of alcohol reduced the incidence
of lynching. Not only was the approval of a dry ordinance associated a discrete drop in the
21
probability that a lynching occurred, the likelihood that a lynching occurred feel each successive
year that alcohol sales were banned. The timing of this discrete jump and trend break is highly
unlikely to have occurred at random, and is not observed in event-study analyses of demographic
changes, voting patterns, or urbanization. We compliment our even study analysis with a
difference-in-difference design, where we estimate similar reductions in the incidence of
lynching when counties go dry. This finding is robust to controls for demographic change and
political attitudes, and our modeling specification, and qualitatively robust to our choice of
sampling frame.
We do not find any correlation between the passage of a dry ordinance and legal
executions southerners of either race. Based on the historical record, lynching and legal
executions were imperfect substitutes for each other. The lack of evidence that temperance
reduced legal executions suggests that closing down saloons may have limited racial conflict, but
primarily reduced impulsive violent behavior of white southerners.
22
VII. References
Abadie, Alberto. (2006). “Poverty, Political Freedom, and the Roots of Terrorism” The
American Economic Review 96(2): 50-56.
Alesina, Alberto, Ozler, Sule, Roubini, Nouriel, and Swagel, Phillip. (1996) "Political Instability
and Economic Growth" Journal of Economic Growth. 1(2): 189-211.
Beck, EM and Tolnay SE (1990). “The Killing Fields of the Deep South: The Market for Cotton
and the Lynching of Blacks 1882-1930.” American Sociological Review,
Berman, Eli and David Laitin, 2008. “Religion, Terrorism and Public Goods:
Testing the Club Model” Journal of Public Economics. 92(10-11):
1942-1967.
Blocker, Jack S. Jr. 1976. Retreat from Reform: The Prohibition Movement in the United States
1890-1913. Westport: Greenwood Press.
Brundage, W. Fitzhugh. 1993. Lynching in the New South: Georgia and Virginia, 18801930.Chicago: University of Illinois Press.
Carpenter, C. (2008) “Heavy Alcohol Use and Crime: Evidence from Underage Drunk Driving
Laws.” Journal of Law and Economics 50(3): 539-557.
Cherrington. Ernest H. 1920. The Evolution of Prohibition in the United States of America: A
Chronological History of the Liquor Problem and the Temperance Reform in the United
States from the Earliest Settlements to the Consummation of National Prohibition.
Westerville: The American Issue Press.
Cook, Philip J. 2007. Paying the Tab. Princeton: Princeton University Press
Cook, Philip J. 2008. “A Free Lunch” Journal of Drug Policy Analysis1(1):1-5.
Cook, P., and Moore, M. (1993) "Violence Reduction through Restrictions on Alcohol
Availability." Alcohol Health & Research World 17(2): 151-156.
Dills, Angela, and Jeffrey Miron. 2004. “Alcohol Prohibition and Cirrhosis” American Law and Economics Review 62: 285-­‐317. Dixit, Avinash. 2003. “Trade Expansion and Contract Enforcement” Journal of Political Economy 111(6) 1293-­‐1317 Dobkin, C. and Carpenter, C. (2008) “The Drinking Age, Alcohol Consumption, and Crime.”
mimeo.
Espy, M. Watt, and John Ortiz Smykla. Executions in the United States, 1608-2002: The ESPY
23
File [Computer file]. 4th ICPSR ed. Compiled by M. Watt Espy and John Ortiz Smykla,
University of Alabama. Ann Arbor, MI: Inter-university Consortium for Political and
Social Research [producer and distributor], 2004. doi:10.3886/ICPSR08451
Ginzburg, Ralph. 1962. 100 Years of Lynchings. Baltimore: Black Classic Press.
Green DP, Glaser J and Rich A (1998) “From Lynching to Gay Bashing: The Elusive connection
between economic conditions and crime”. Journal of Personality and Social Psychology,.
Hamm, Richard F. 1995. Shaping the Eighteenth Amendment: Temperance Reform, Legal
Culture, and the Polity, 1880–1920. Chapel Hill: University of North Carolina Press.
Haines, Michael R., and Inter-university Consortium for Political and Social Research.
Historical, Demographic, Economic, and Social Data: The United States, 1790-2002
[Computer file]. ICPSR02896-v3. Ann Arbor, MI: Inter-university Consortium for
Political and Social Research [distributor], 2010-05-21. doi:10.3886/ICPSR02896
Joksch, H. C., and Jones, R. K. (1993). “Changes in the Drinking Age and Crime.” Journal of
Criminal Justice 21: 209-221.
Kranton, Rachek. 1996. “Reciprocal Exchange: A Self Sustaining System” American Economic Review 86(4) 830-­‐851 Krueger, Alan B. and Maleckova, Jitka.(2003) "Education, Poverty and Terrorism: Is There a
Causal Connection?" Journal of Economic Perspectives. 17(4): 119-44.
Lewis, Michael. 2008. “Access to Saloons, Wet Voter Turnout, and Statewide Prohibition
Referenda, 1907–1919” Social Science History 32(3): 373-404.
Litwack, Leon F. 1998. Trouble in Mind: Black Southerners in the Age of Jim Crow. New York:
Alfred A. Knopf.
Owens, Emily G., 2011. “Are Underground Markets Really More Violent? Evidence from Early
20th Century America” American Law and Economics Review, forthcoming
Sechrist, Robert P. Prohibition Movement in the United States, 1801-1920 [Computer file].
ICPSR version. Robert P. Sechrist [producer], 1983. Ann Arbor, MI: Inter-university
Consortium for Political and Social Research [distributor], 1999.
doi:10.3886/ICPSR08343
Sechrist, Robert P. Basic Geographic and Historic Data for Interfacing ICPSR Data Sets, 16201983 [Computer file]. ICPSR version. Baton Rouge, LA: Robert P. Sechrist, Louisiana
State University [producer], 1984. Ann Arbor, MI: Inter-university Consortium for
Political and Social Research [distributor], 2000. doi:10.3886/ICPSR08159
Tabellini, Guido. 2008. “The Scope of Cooperation: Values and Incentives” The Quarterly
Journal of Economics. 123 (3) 905-950
24
Figure 1: Summary Statistics by County
Panel A: Lynching and Black Population
Panel B: Last Transition from Wet to Dry
Notes: Panel A presents the spatial configuration of lynchings and of the black population in the South. Darker shading
indicates a higher population density among blacks in the 1890 Census data on counties. Red dots indicate the approximate
locations (the county seat of the county of occurence) of lynchings from 1895–1930. The smallest dot denotes a single lynching
in the county during 1895–1930, and the dots have areas proportional to the number of lynching victims during that period.
The data for lynching are drawn from Tolnay and Beck (1995). Information on black population density is reported by Haines
(2004). The 1900 county boundaries are from HUSCO (Earle, Otterstrom, and Heppen, 1999), and the latitude and longitude
of county seats are taken from Sechrist (1984). (The point to the south of Florida is Key West, the county seat of Monroe
County.) Panel B indicates which counties were going from wet to dry in our sample window, and during approximately which
years. The dots are placed at the county seat of each county. A greener dot indicates an earlier transition from wet to dry
status. Those places that transition from wet to dry before 1895 are not in our main sample, and are not shown on this map.
The data on county-level prohibition are drawn from Sechrist (1983). States in which no dots appear were excluded from our
sample.
Figure 2: Summary Statistics by Year
.2
Panel A: Lynching Rates
0
.05
.1
.15
Proportion of Counties With a Lynching
Average Number of Victims Lynched
1890
1900
1910
Year
1920
1930
1920
1930
0
.2
Proportion, Dry Counties
.4
.6
.8
1
Panel B: Proportion of Dry Counties
1890
1900
1910
Year
Notes: this figure presents averages by year within the sample of counties and over the years 1895–1930. The sample includes all
counties for which at least one lynching is reported during those years. The data for lynching, shown in Panel A, are drawn from
Tolnay and Beck (1995). The data on county-level prohibition, shown in Panel B, are drawn from Sechrist (1983). Averages
and proportions are unweighted.
Figure 3: Summary Statistics by Time Since Dry
200
400
Number of Counties
600
800
1000
1200
Panel A: Sample Size
−20
−10
0
10
20
30
20
30
Years since Going Dry
0
.2
Proportion of Dry Counties
.4
.6
.8
1
Panel B: Proportion of Dry Counties
−20
−10
0
10
Years since Going Dry
Notes: this figure displays characteristics of the sample as a function of the re-normalized measure of time, “Years since Going
Dry.” This variable sets time equal to zero in the most recent year in which the county transitions from wet to dry status.
The data on prohibition are drawn from Sechrist (1983). Panel A displays the number of counties in each year since going dry.
Panel B displays the proportion of counties that are dry. By construction, this proportion is equal to 1 after year zero, because
year zero is defined to be the year of the last transition from wet to dry status. Additional information on data sources and
definitions is found in Section 2 and the data appendix.
Figure 4: Lynching and Time Since Dry
0
.05
.1
.15
.2
Panel A: Average Number of Victims Lynched
−20
−10
0
10
20
30
20
30
Years since Going Dry
0
.02
.04
.06
.08
.1
Panel B: Proportion of Counties with a Lynching
−20
−10
0
10
Years since Going Dry
Notes: This figure displays, as a function of the re-normalized time variable, “Years since Going Dry,” the number of victims
reported lynched, per county (Panel A) and the proportion of counties in which a lynching occurred (Panel B). This time
variable is constructed by setting year equal to zero in the most recent year in which the county transitions from wet to dry
status. The lynching data are taken from Tolnay and Beck (1995), while the data on prohibition are drawn from Sechrist (1983).
Each year’s datapoint is marked with a dot. Because we do not know at what point in the year of the prohibition took effect,
we do not display the point corresponding to t = 0. Additional information on data sources and definitions is found in Section
2 and the data appendix.
80
60
Density
40
20
0
.01
−.04
−.02
0
Coefficient Estimates
−.002
Coefficient Estimates
0
B.1: Proportion of Counties
.002
−.006
−.004
−.002
0
Coefficient Estimates
B.2: Number of Lynchings
Panel B: Magnitude of the Break in Trend When Going Dry
−.01
Coefficient Estimates
A.2: Number of Lynchings
.002
.02
.004
Notes: These graphs display the distribution of point estimates for the shift and break in trend when using randomly generated breakpoints instead of the actual year
of transition from wet-to-dry status. The breakpoints are randomly drawn from the within-state distribution of years of the last wet-two-dry transition. The shift and
break in trend at this spurious t = 0 are estimated by fitting separate linear models on either side of the breakpoint, as in Figures ?? and ??. The distribution is
estimated from 3,000 sets of random breakpoints. The shift and trend break from the actual wet-to-dry transitions (that were shown in Figures ?? and ??) are denoted
by the vertical lines in each graph. The fractions of area to the left of the vertical lines are, from Panel A.1 and going counterclockwise, 0.017, 0.093, 0.011, and 0.007.
−.004
−.02
A.1: Proportion of Counties
Panel A: Magnitude of the Shift When Going Dry
Figure 5: Falsification Exercise: Estimates with Simulated Wet-to-Dry Transitions
Density
0
500
400
Density
300
200
100
0
40
30
Density
20
10
0
300
200
100
0
Figure 6: Legal Executions
.02
.04
.06
.08
.1
Panel A: Average Number of Executions of Blacks
−20
−10
0
10
20
30
20
30
Years since Going Dry
0
.01
.02
.03
Panel B: Average Number of Executions of Whites
−20
−10
0
10
Years since Going Dry
Notes: This figure displays statistics on legal executions blacks or whites as a function of the re-normalized time variable, “Years
since Going Dry.” This time variable is constructed by setting year equal to zero in the most recent year in which the county
transitions from wet to dry status. The data on prohibition are drawn from Sechrist (1983). The data on legal executions were
compiled by Espy and Smykla (2004), and refer to the county in which the crime took place.
Figure 7: Various Census Outcomes versus Time Since DryData
Panel B: Total Population
0
8.5
9
.1
9.5
.2
10
.3
10.5
.4
11
Panel A: Proportion in a Census Year
−20
−10
0
10
Years since Going Dry
20
30
−20
0
10
Years since Going Dry
20
30
20
30
Panel D: Percent Urban
−10
−2.5
−8
−2
−6
−1.5
−4
−1
−2
0
−.5
Panel C: Percent Black
−10
−20
−10
0
10
Years since Going Dry
20
30
−20
0
10
Years since Going Dry
Panel F: Average Farm Size
10
−5.8
10.5
−5.6
11
−5.4
11.5
−5.2
−5
12
Panel E: Total Farm Acreage
−10
−20
−10
0
10
Years since Going Dry
20
30
−20
−10
0
10
Years since Going Dry
20
30
This figure displays statistics from the decennial censuses as a function of the re-normalized time variable, “Years since Going
Dry.” This time variable is constructed by setting year equal to zero in the most recent year in which the county transitions from
wet to dry status. The data on prohibition are drawn from Sechrist (1983), while the census data are from Haines (2004). Panel
A presents the proportion of counties that are observed in a census year, and therefore for which census data are available.
Panel B presents the average of (log) total population in the census year. Panel C and D present the (log) percent of the
population that is, respectively, black or urban. Panel E displays the log of total farm acreage farm acreage in the county.
Panel F shows the log average farm size, constructed by dividing the total farm acreage by the total number of farms in a each
county.
Figure 8: Voting Variables versus Time Since Dry
Panel B: Total Votes Cast (Log)
0
7.4
.2
7.6
.4
7.8
8
.6
.8
8.2
1
8.4
Panel A: Proportion in an Election Year
−20
−10
0
10
Years since Going Dry
20
30
−10
0
10
Years since Going Dry
20
30
Panel D: Share, Republican Party
.12
.7
.14
.75
.16
.8
.18
.2
.85
.22
.9
Panel C: Share, Democratic Party
−20
−20
−10
0
10
Years since Going Dry
20
30
−10
0
10
Years since Going Dry
20
30
Panel F: Share, Prohibition Parties
0
0
.002
.05
.004
.1
.006
.15
.008
Panel E: Share, Populist Parties
−20
−20
−10
0
10
Years since Going Dry
20
30
−20
−10
0
10
Years since Going Dry
20
30
Notes: This figure displays statistics on elections as a function of the re-normalized time variable, “Years since Going Dry.”
This time variable is constructed by setting year equal to zero in the most recent year in which the county transitions from wet
to dry status. The data on prohibition are drawn from Sechrist (1983), while the elections data are drawn from ICPSR (1999).
(3)
(4)
OLS
-0.428
(0.184) **
-0.021
(0.011) **
Drop 1904,
1908
-0.026
(0.012) **
Probit
-0.024
(0.011) **
State-specific
trends
-0.023
(0.012) **
Conditional
Logit
0.004
(0.039)
-0.051
(0.024) **
Deep South
-0.026
(0.021)
Conditional
Logit
-0.024
(0.010) **
Year x pre1895
lynching
-0.023
(0.011) **
Baseline
-0.048
(0.022) **
|k|<7
-0.058
(0.029) **
Conditional
Logit
-0.135
(0.231)
-0.051
(0.024) **
OLS
-0.016
(0.029)
-0.046
(0.021) **
Panel D: Interactions
Outside of
Deep South
-0.033
(0.014) **
Panel C: Alternate Samples
OLS, cluster
on state
-0.048
(0.019) **
Panel B: Alternate Estimators
Year x %
black,
sharecrop
-0.021
(0.012) *
(6)
(7)
OLS
-0.748
(0.448) *
-0.045
(0.021) **
Drop 1904,
1908
-0.057
(0.025) **
Poisson
-0.044
(0.017) ***
State-specific
trends
-0.048
(0.023) **
Poisson
-0.106
(0.121)
-0.254
(0.081) **
Deep South
-0.043
(0.033)
Ordered Logit
-0.020
(0.009) **
Year x pre1895
lynching
-0.049
(0.022) **
-0.067
(0.034) *
Year x %
black,
sharecrop
-0.046
(0.023) **
(8)
Poisson
-0.159
(0.728)
-0.246
(0.081)
Outside of
Deep South
Outcome: number of lynching victims
Panel A: Various Controls
(5)
This table presents estimates of equation (1) in the text. All specifications contain fixed effects for county and for year, and the standard errors are
adjusted for clustering at the county level, unless noted. Coefficients for limited-dependent-variable estimators are marginal effects evaluated at the
sample mean. The outcome variable is the number of victims reported lynched, per county (Columns 1-4) and the proportion of counties in which a
lynching occurred (Columns 5-8). The lynching data are taken from Tolnay and Beck (1995), while the data on prohibition are drawn from Sechrist
(1983). Additional information on data sources and definitions is found in Section 2 and the data appendix.
Estimator
OLS
-0.014
(0.016)
Fraction Dry, Nearby
Counties, x Dry Dummy
Pre-1895 Lynching Rate x
Dry Dummy
-0.022
(0.011) **
|k|<7
-0.024
(0.017)
OLS, cluster
on state
-0.024
(0.011) **
Baseline
-0.024
(0.011) **
Dummy if county is dry
Sample restriction
Dummy if county is dry
Estimator
Dummy if county is dry
Additional controls
Dummy if county is dry
Explanatory variables:
(2)
Outcome: binary variable if lynching occurred
(1)
Table 1: Lynchings and Time Since Dry
A
Details on data construction
[TO COME]
B
Extra Figures
Appendix Figure B-1: Sensitivity Analysis I
.2
Panel A: Area and Time Fixed Effects
0
.05
.1
.15
Unconditional
County FE
Year FE
(State x Year) FE
−20
−10
0
10
20
30
Years since Going Dry
.2
Panel B: Calendar Year Fixed Effects × County Fixed Characteristics
0
.05
.1
.15
Unconditional
Pre−1895 Lynchings
Latitude and Longitude
Hookworm and Malaria
−20
−10
0
10
Years since Going Dry
Notes:
20
30
Appendix Figure B-2: Sensitivity Analysis II
.2
Panel A: Control for Interpolated Census Variables
0
.05
.1
.15
Unconditional
Percent Black
Percent Urban
Farm Size
Total Acres in Farms
−20
−10
0
10
20
30
Years since Going Dry
.2
Panel B: Alternative Samples
0
.05
.1
.15
Unconditional
First Not Last Year Going Dry
Counties with Lynchings Prior to 1895
Counties in Southern States
−20
−10
0
10
Years since Going Dry
Notes:
20
30