A Brief Moment in the Sun: The Racialized(Re)Construction of Punishment in the AmericanSouth*

Soumyajit Mazumder†Harvard Universitysmazumder@g.harvard.edu

Date last updated: September 30, 2019

Why did policing and incarceration emerge as an institution in the South in light of its oth-erwise “hollow state?” I explore this question in context of the 19th and early 20th century U.S.South–a region which many point to as the crucible of punishment in America. I argue that out-side intervention by the federal government during Reconstruction helped lay the groundworkfor the carceral state. By empowering African Americans without fundamentally changing thesocial structure of Southern society, Reconstruction generated incentives for Southern elites toinvest in repressive state institutions like incarceration and the police to maintain the existingsocial order. I test the argument by assembling a new panel dataset from individual-level ad-ministrative records from Georgia in addition to data on black wealth, black office holding, andthe police force. A difference-in-differences identification strategy demonstrates support for theargument: counties with greater exposure to Reconstruction had higher rates of incarceration es-pecially against blacks after the end of Reconstruction. Additional results on black wealth, blackoffice-holding, and the police force provide suggestive evidence in line with the theory. This arti-cle verifies key arguments going back to W.E.B. Du Bois in addition to emphasizing the centralityof race and coercion in the evolution of American politics.

*I thank Matt Blackwell, Jacob Brown, Dan Carpenter, Chris Chaky, Stephen Chaudoin, Laurel Eckhouse,Ryan Enos, Alex Gourevitch, Allison Harris, Elizabeth Hinton, Jennifer Hochschild, Kaneesha Johnson, MarkoKlasnja, Jason Lyall, Chris Muller, Michael Olson, Jon Rogowski, Ariel White, and Steven White for incisivefeedback and discussions on the manuscript. I would like to thank the Institute for Quantitative Social Scienceat Harvard University for funding early stage data collection for the project. I would also like to thank seminaraudiences atHarvard and EmoryUniversity as well as theAPSA 2019Meeting andAPSA 2019Class and Inequal-ity Poster Session for excellent comments and discussions of the project. The author acknowledges support bythe James M. and Cathleen D. Stone Ph.D. Scholar Fellowship from the Multidisciplinary Program in Inequality& Social Policy at Harvard University. All errors are the author’s own.

†Ph.D. Candidate, Department of Government, Harvard University, web: http://smazumder.me

1

The slave went free; stood a brief moment in the sun;then moved back again toward slavery.

W.E.B. Du Bois

introduction

Maintaining social order is one of the most fundamental activities that states engage in. In

nearly all societies, states attempt to maintain social order coercively through the use of pun-

ishment (Foucault 1995). In the U.S. context and especially in the South, direct state involve-

ment in punishment remained limited up until the Civil War.1 For instance, data from Geor-

gia in Figure 1 shows that incarceration via state imprisonment rapidly grew following the

end of the Civil War and Reconstruction–a trend well-documented throughout the South by

historians of the time period (Ayers 1984; Mancini 1996; Lichtenstein 1996; Blackmon 2008).

When viewed in tandemwith theU.S.’s lack of substantive state institutions in the 19th century

(Skowronek 1982), this sudden growth in incarceration generates the following puzzle: why

did policing and incarceration emerge as an institution in the South in light of its otherwise

“hollow state?”

Starting from the fundamental premise of the centrality of punishment for understanding

the state, this paper goes back in time to the antebellum and post-bellum U.S. South–a region

that many point to as exemplifying many of the qualities of the modern carceral state such

as private corporate involvement, high racial disparities, and periods of rapid expansion–as a

window into understanding the origins of policing in the United States (Blackmon 2008). In

this respect, I refer to this early penal system as the proto-carceral state. A prominent set of

extant research on the topic emphasizes the importance of incarceration–particularly, convict

leasing–as a way to help modernize the South through the development of mining and rail-1Institutions such as chattel slavery in the ante-bellum South relied on vigorous private provision of pun-

ishment and coercion on the plantation. The state’s involvement in this process was largely around recapturingescaped slaves. This piece focuses on the transition of the provision of punishment on the plantation to thestate-led provision of punishment.

2

1820 1840 1860 1880 19000

30

60

90

1810 1820 1830 1840 1850 1860

0.0

2.5

5.0

7.5

10.0

Year

Tota

l Con

vict

Pop

ulat

ion

(Per

100

,000

)

Source: Georgia Convict Ledger and Biennial Report of the Penitentiary

Figure 1: Growth of Incarceration over Time in Georgia, 1817-1911

road infrastructure (Lichtenstein 1996). This set of explanations is insufficient on its own to

explain why African Americans were disproportionately incarcerated and why some counties

had high levels of incarceration while others did not since there were relatively few convict

camps. Moreover, the rise of the carceral state cannot alone be explained as a simple repro-

duction of slavery (Blackmon 2008): incarceration itself was nowhere near the same scale as

chattel slavery. Even more, the institutionalization of policing is even more puzzling within

the context of the South given its long-reliance on an “honor culture” to enforce social order

(Nisbett and Cohen 1996). These features then suggest existing explanations for the rise of

the proto-carceral system are incomplete.

I build on work across history and the social sciences to develop a political framework for

3

understanding variation in the rise of the proto-carceral state (Du Bois 1901b, 2014; Weaver

2007; Acemoglu and Robinson 2008). This framework emphasizes how sudden resistance

by and empowerment of marginalized groups motivate the development of a repressive state

apparatus by elites. The historical turn in this piece necessitates a re-characterization of incar-

ceration and policing during this time period. Drawing on work within the political violence

literature in comparative politics, I conceptualize incarceration during this time period as a

form of state repression meant to suppress the political, economic, and social freedoms of

a targeted population (Davenport 2007).2 By reconceptualizing incarceration as a form of

political violence, it becomes clear why elites would make costly investments in this system:

repression via incarceration is meant to inspire fear among marginalized groups and demo-

bilize them from challenging the existing status quo that puts them into coercive economic

and social relationships with elites (Lerman and Weaver 2010, 2014; Young 2019; White 2019).

I argue that the temporary and incomplete empowerment of a marginalized group gener-

ates incentives for the elite to engage in repression so as tomaintain the existing socio-political

hierarchy. By incomplete empowerment, I mean the material empowerment of a group with-

out the fundamental alteration of the fundamental social structures and hierarchies in a so-

ciety. Applying the framework to the U.S. South during and after the Reconstruction period,

I argue that the incomplete empowerment of African Americans by the federal government

created incentives for white elites to use incarceration as a means to exert social control over

this threatening population. This frameworkmakes two key predictions: (1) areaswith greater

exposure to Reconstruction should, after its end, be more likely to incarcerate their popula-

tions, and (2) that the impact of Reconstruction should be increasing in the proportion of

the population that benefited from it (African Americans). By not fundamentally leveling2Davenport (2007) uses a definition of state repression from Goldstein (2001), who focused on the case of

the United States. Wright (1973) also makes this argument in discussing the role of the prison in Americansociety. The concepts and study of political violence and state repression no longer appear the “mainstream”American politics canon. I return to this point in the conclusion.

4

racialized land-labor power structures in the South, Reconstruction generated incentives for

white elites to set up a highly repressive state apparatus to maintain their dominant status in

the social order.3 Though not a reproduction of slavery itself, I argue that the proto-carceral

state served to maintain highly racialized and coercive relationships between white planter

elites and African Americans.4

To test these predictions, I assemble a new panel dataset from individual-level hand-

written records from the Georgia Convict Ledger from 1817-1868, individual-level data from

administrative records in the Georgia Biennial Report from 1880-1882, and county aggregates

from 1897-1911. In total, this data spans 137 counties, 94 years, and over 15,000 prisoners.

Using a difference-in-differences (DID) design leveraging the sharp removal of troops from

Georgia once the state was re-admitted to the Union in 1870, I find evidence of a strong link

between backlash against Reconstruction and the rise of the proto-carceral state. Moreover,

the impact of Reconstruction is concentrated in areas with high shares of enslaved popula-

tions in 1860. Using data on prisoner race, I show that these results are concentrated among

individuals of color rather than whites, ruling out an alternative explanation rooted in general

rising crime trends across both groups. In addition, the results remain robust to relaxing the

identification assumptions through inclusion of county-specific linear and quadratic trends

in addition to being insensitive to any one outlier county.

To test the hypothesizedmechanism–thatReconstruction incompletely empoweredAfrican

Americans thereby engendering incentives for elites to invest in carceral capacity, I collect

data originally assembled by Du Bois (1901a) on black property holding in Georgia in addi-

tion to data on black office-holding recorded by Foner (1996).5 Further results suggest that3Thus, an important scope condition of the argument is the nature of empowerment. Where the empow-

erment does not fundamentally alter social structure, the theory predicts an investment in repressive state in-stitutions by elites. In situations where social structures do become level, the theory does not have explanatorypower.

4It is in this respect, how incarceration became “slavery by another name” (Blackmon 2008).5By carceral capacity, I mean inputs into policing such as the number of police, establishment of prisons or

labor camps, and equipment (Gilmore 2007).

5

Reconstruction was effective in increasing African Americans’ economic empowerment as

proxied by property wealth. This increase in black wealth is inconsistent with the alternative

explanation that blacks were engaging in more crime given the vast research showing that

improved material status reduces crime (Becker 1968). In line with the argument, I also show

that Reconstruction did not fundamentally alter the land-labor social structure in Georgia:

areas with greater exposure did not see a substantial increase in black landholding. This lack

of a fundamental change in social structure and hierarchy translated into modest gains at

the office as proxied by black office-holding (Foner 1996, 2011). I also show using data from

the U.S. Federal Census that greater exposure to Reconstruction also led to a rise in the size

of the police force. These additional results are consistent with the argument of this paper

that Reconstruction helped to empower African Americans, which generated incentives for

white elites to invest in coercive state infrastructure. Overall, the results paint a clear picture:

state-led punishment emerged as a mechanism to control a population that could threaten

the power of white elites (Du Bois 2014).

This study makes several contributions to literatures involving incarceration, race, and

American political development. First, this piece speaks directly to scholars interested in

the development of the carceral state. While much of this literature focuses on explaining

the growth of incarceration (Gottschalk 2006; Weaver 2007; Miller 2008; Murakawa 2014;

Gottschalk 2016; Hinton 2017), this piece departs from this literature by studyingwhy policing

and incarceration emerged as institutions to begin with. Building on recent work by Muller

(2018) on Georgia’s convict lease system, I emphasize the way in which economic structures

are ensconced in broader political transformations that may generate incentives for elites to

use the state’s coercive power to hold onto racial hierarchies. This analysis shows striking

similarities with the frontlash thesis offered by Weaver (2007) on the modern carceral state:

the Southern proto-carceral state operated in a way that was designed to subjugate a status-

threatening population. This study builds on and contributes to existing studies of incarcer-

6

ation in the United States by highlighting the tragedy of external empowerment–one of many

factors that drive changes in punishment throughout American history (Smith 2004; Lerman

and Weaver 2010; Jacobs and Jackson 2010; Walker 2014; Alexander 2012; Fortner 2013; Eck-

house 2016; Enns 2016; Pfaff 2017; Forman, Jr. 2017; Beim, Clark, and Lauderdale 2019; Harris

2019; Harvey and Yntiso 2019).

Additionally, this study pushes forward the literature on race and ethnic politics in the

United States. While much of the literature on racial politics emphasizes the role of public

opinion (Gilens 1996; Kinder and Sanders 1996; Valentino and Sears 2005; Tesler and Sears

2010), geography (Key 1949; Enos 2017; Cook, Logan, and Parman, Forthcoming; Reny and

Newman 2018), or electoral politics in shaping black-white relations (Valentino, Hutchings,

and White 2002; Frymer 2010), this study joins and extends the literatures looking at the

independent effect of the state in crystallizing black-white political, social, and economic in-

equality (Frymer 2005; Katznelson 2005; King and Smith 2005; Francis 2014; White, Nathan,

and Faller 2015; Soss and Weaver 2017; Michener 2018; White 2019). By taking the analytical

lens back in time, this study shows and reinforces the idea that the state was used and con-

tinues to be used to stratify societies by race and class (Soss and Weaver 2017). Moreover,

the framework and results presented in this article have serious implications for contempo-

rary debates around racial conflict. Outside efforts to empower marginalized populations

can backfire when these marginalized groups are not able to institutionalize their gains. As

a whole, the upshot of this piece is that studies of racial politics are incomplete insomuch as

the state (whether federal, state, or local) is absent in these accounts.

A renewed stream of American political development work seeks to understand the puz-

zles of democratization in the South and its role in the larger national polity (Mickey 2015;

Acharya, Blackwell, and Sen 2018a; Caughey 2018; Bateman, Katznelson, and Lapinski 2018).

For instance, work by Gibson (2013) and Mickey (2015) shows that the South can generally

be characterized as an “authoritarian enclave” where subnational politics essentially operate

7

as a single-party autocracy. This study enriches this observation by focusing on a particu-

lar dimension of state power–punishment–in subjugating subaltern populations. Viewed in

tandem with studies on Southern paternalism (Alston and Ferrie 1993), white supremacist

institutions were either implicitly or explicitly private-public partnerships. Additionally, this

helps push forward the recent historiographical and empirical literature on the successes and

failures of Reconstruction (Foner 2011). While this study and others show that Reconstruction

did indeed empower African Americans economically (Logan 2018; Rogowski 2018; Chacon

and Jensen 2018; Suryanarayan and White 2019), it also shows that this empowerment came

at the cost of black lives (Stewart and Kitchens 2018). Moreover this study lies in contrast

to Bensel (1990) who argues that Reconstruction represents a failed opportunity at state-

building; instead, the argument and results presented here suggest that Reconstruction ac-

tually helped pave the way for coercive state-building precisely because it was an “unfinished

revolution (Foner 2011).

Finally, this study also contributes to deepening our understanding of violence and re-

pression in America. While many existing studies place primary focus on the use of lynching

as an informal method of whites’ subordination of the African American population (Beck

andTolnay 1990; Francis 2014; Smangs 2016;Weaver 2019), the scale of the proto-carceral state

dwarfed the scale of lynching in the period right after Reconstruction. Data from Georgia,

for instance, puts the number of black lynching victims at approximately 30 while there were

more than 800 black prisoners on hand during this same time period. A contribution of this

study, then, is to highlight and better understand the logic of the first substantive investment

in coercive capacity through incarceration. More broadly, the sheer magnitude of the state-

driven coercion relative to informal systems of subordination demands much further study

by the discipline (Soss and Weaver 2017).6

6Obviously, both incarceration and lynching during this time period are two forms of violence productionthat deserve further study given their enormous importance to the process of American political development(MingFrancis2014).

8

framework: racialized coercion and state power

A large branch of existingwork generally relates the use of formalized punishment as a tool for

white elites to control the subaltern black population (Du Bois 1901b; Myers 1990; Mancini

1996; Du Bois 2014).7 In this section, I distill this literature into several analytical building

blocks to develop a framework for understanding race, punishment, and state power. To

summarize, I argue that the subaltern population’s outside options and salience threaten the

in-group’s positions of economic, political, and cultural status. In response to these features,

elites in societies with sufficient capacity to engage in formalized violence through punish-

ment will do so. This theory essentially places the state as a technology of racial hierarchy

production.

Before I outline the variables emphasized in this article, it is important to address the fol-

lowing question: what does punishment do? Ostensibly, punishment is a means to address

some sort of transgression against a written set of laws that all citizens of a society must abide

by. Yet, two factors transform punishment beyond a technocratic policy designed to correct

or deter certain proscribed behaviors: its uneven application and the way in which punish-

ment is ensconced in social systems. As Holland (2016) shows, the uneven application and

enforcement of laws generates distributive consequences, which can go on to empower some

groups and take away power from others. This feature of punishment is one way in which

formal punishment becomes politicized. Additionally, punishment itself is embedded in so-

cial rules that proscribe some behaviors as “deviant” and others as not. This feature allows

punishment to become not only a way to create order, but also to exert social control (Black

1983). Thus, punishment is a way in which the state exercises coercive power in ways that can

be and often are politicized along group boundaries.

The state’s ability to use repression, often times through incarcerating members of “de-7Throughout the rest of the text, I refer to African Americans as the subaltern population.

9

viant” populations, is a key tactic towardmaintaining power. A large literature in comparative

politics highlights how the police are used in authoritarian societies as a means to exert con-

trol over their populations (Davenport 2007; Greitens 2016). Of course, authoritarian elites

do not jail the entire status threatening subaltern populations. Instead, they opt to some small

fraction of the group as a means to generate a sense of fear among the entire group (Young

2019).

What are the incentives for elites to engage in this sort of repression? I argue that one

such incentive that would drive elites to use formal punishment to control subaltern popula-

tions is the role of outside options–a feature noted in political economy theories of coercion

and exploitation (Chwe 1990; Wright 2000; Acemoglu and Wolitzky 2011). Minority power

is most threatening in situations where that group has the highest outside options. The logic

is that elites can hold power and extract more from groups when they are able to reduce the

value of all the alternative relationships that a minority group might enter into. For instance,

when an individual can credibly exit a relationship with a potentially coercive elite by moving

away or entering into some other occupation or social relationship, then the elite is able to

extract less from that individual. This suggests that in places where the subaltern group be-

comesmaterially empowered (i.e, his/her outside options become higher in value), then elites

should endogenously use the tools they have available in order to reduce the subaltern group’s

outside options. In contexts where there is an incomplete transformation in the fundamental

power structure between elites and the subaltern, the empowerment of the subaltern, even in

amaterial and “real” sense, can lead the elite to dis-empower that group via state coercion and

violence (Acemoglu and Robinson 2008). Formalized punishment and incarceration, then,

provide tools through which white elites can use the state to “lock-up” individuals who pose

a threat to their status and power. This logic, which I formalize in the Online Appendix, can

help to explain variation in the intensity of incarceration.

In this article, I focus on how Reconstruction helped to empower African Americans

10

thereby engendering incentives for elites to use repression. A primary goal of Reconstruction

was to fundamentally alter the race-class relations of the South by empoweringAfricanAmer-

icans to gain land and property (Foner 2007, 2011; Du Bois 2014). With this goal in mind, the

militaries used a whole host of coercive methods to push for this transition in the Southern

social order: “Military oversight of the law changed the dynamics of Southern life; magis-

trates who once held vast powers to regulate their neighborhoods (and whose decisions were

rarely tested by appeal to higher courts) suddenly could be called to answer to nearbymilitary

authorities who could overrule and even depose them” (Downs 2015, p. 30). In addition to

taking control of local institutions, the military played a large role in overseeing elections in

the face of electoral violence. Where the Freedmen’s Bureau could not perform its functions

as a result of staffing issues, the military stepped in to fill these gaps (Downs 2015). While it is

well established that Reconstruction itself did not fundamentally transform Southern society

as many of its more radical proponents wished (Vallely 2004; Foner 2011; Du Bois 2014), re-

cent research on the short to medium-term impact of Reconstruction provides evidence that

it did at least increase black literacy (Stewart and Kitchens 2018), voter turnout (Rogowski

2018), and fiscal capacity (Chacon and Jensen 2018; Suryanarayan and White 2019). Blacks’

material empowerment as a result of the enforcement of Reconstruction suggests that white

elites should have an incentive to use the state to re-exert social control over this population

after the end of its enforcement.

Without a doubt, the end of slavery and the newfound sense of freedom by former slaves

as a result of Reconstruction brought a real sense of empowerment to them. Foner (2011, p.

79) writes,

Blacks relished opportunities to flaunt their liberation from the innumerable reg-

ulations, significant and trivial, associated with slavery. Freedmen held mass

meetings and religious services unrestrained bywhite surveillance, acquired dogs,

11

guns, and liquor (all barred to them under slavery), and refused to yield the side-

walks to whites. They dressed as they pleased, black women sometimes wearing

gaudy finery, carrying parasols, and replacing the slave kerchief with colorful hats

and veils. In the summer of 1865, Charleston saw freedmen occupying ‘some of

the best residences,’ and promenading on King Street ‘arrayed in silks and satins

of all the colors of the rainbow,’ while black schoolchildren sang “John Brown’s

Body’ within ear-shot of Calhoun’s tomb.’ Rural whites complained of ‘insolence’

and ‘insubordination’ among the freedmen, by which they meant any departure

from the deference and obedience expected under slavery. On the Bradford plan-

tation in Florida, one untoward incident followed another. First, the family cook

told Mrs. Bradford ‘if she want any dinner she kin cook it herself.’ Then the for-

mer slaves went off to a meeting with Northern soldiers to discuss ‘our freedom.’

Such displays of power from the ballotbox to day-to-day life threatened the status that

whites had at the time. Most of the former slaves tended to stay near the plantations where

they were previously enslaved: “In fact, a majority of freedmen did not abandon their home

plantations in 1865, and those who did generally traveled only a few miles” (Foner 2011, p.

81). The presence of such a newly empowered group, then, generated a sense of resentment

by whites. This suggests that the impact of Reconstruction should be increasing the salience

of the newly freed black population. Additionally, the white elite cannot costlessly engage

in coercion against all groups. At a psychological level, engaging in coercion against other

individuals who share the same ethnic/racial identity group as the elite might be cognitively

costly. This becomes less costly as the social distance between the elite’s ethnic identity group

and the subaltern’s ethnic identity group increases (Tajfel and Turner 1986; Enos 2017). From

a strategic perspective, white elites also needed to maintain cross-class cohesion in its project

of white supremacy (Du Bois 2014; Merritt 2017). By punishing and violently coercing one’s

12

own group, this could threaten intra-group cohesion–a point that Du Bois notes throughout

his writings (Du Bois 1901b, 2014). These features of class-race stratified societies can help to

explain why punishment would become racialized in this way. As a result, we should expect

the impact of external empowerment (Reconstruction) on incarceration to be increasing in

the share of the subaltern (formerly enslaved) population.

The theory, then, makes two key testable predictions:

1. H1: Punishment should be increasing in the intensity of Reconstruction.

2. H2: The impact of Reconstruction should be increasing in the relative size of the en-

slaved population.

The following section provides a brief overview of the case used in this study. Then, I

describe the data used to test the argument and present empirical tests of themain hypotheses,

assumptions, and causal mechanisms.

background: formal punishment in a violent society

Many refer to the 19th century U.S. South as a violent society (Ayers 1984; Nisbett and Cohen

1996; Roth 2009). Slavery, for instance, was an institution predicated on the threat and use of

violence against African Americans in order to extract labor from them. Beyond this type of

violence, the South also has a particular relationship with informal and formal punishment.

During the antebellum period, the Southern penal institutions essentially had two different

tracks–one for slaves and one for whites. For slaves, transgressions were nearly exclusively

dealt with on the plantation since slaves were viewed as expensive property. For whites, there

was a system of police, courts, and juries to deal with white criminality–a phenomenon driven

by the way in which slavery generated a sense of destitution for poor, landless whites (Merritt

2017).

13

Following the Civil War and emancipation, punishment on the plantation turned into a

vast formalized system known as convict lease. Noting this shift, Du Bois (2014, p. 415) writes,

“This penitentiary system began to characterize the whole South. In Georgia, at the outbreak

of the Civil War, there were about 200 white felons confined at Milledgeville. There were no

Negro convicts, since under the discipline of slavery, Negroes were punished on the planta-

tion. The white convicts were released to fight in the Confederate armies. The whole criminal

system came to be used as amethod of keeping Negroes at work and intimidating them. Con-

sequently there began to be a demand for jails and penitentiaries beyond the natural demand

due to the rise of crime.” This new system, which allowed private companies and individuals

to hire out incarcerated individuals, represented a reproduction of slavery in nature (forced

labor), but also a departure from it in its scale and industrial nature. With the imposition of

this new system following the end of Reconstruction, the incarcerated population rose over

ten fold to over 3,000 prisoners–the vast majority of whom were African American.8 The

first recorded lessee was leased on May 11th, 1869, but by April 1st, 1879, all of the prisoners

in Georgia were leased out to private corporations (Taylor 1942b). In this sense, one might

think of the convict lease system as one of America’s earliest examples of a private-public

partnership.

Much of this historiography around the rise of the proto-carceral state in the South likens

it to a form of neo-slavery (Oshinsky 1996; Blackmon 2008). Though obviously incarceration

was not on the same scale as slavery, its nature resembled many of the features of slavery. For

instance, the convict lease system forced incarcerated individuals–who were mainly African

American–to work long hours, often under brutal conditions. Taylor (1942b) recounts a se-

vere, but not irregular instance of punishment:

According to the testimony, “Directly the Captain called the negro out and8Statistics based on author’s tabulations. I provide more details and descriptive statistics on the data later in

the article.

14

whipped him. He whipped him a while and put him back on the barrel andmade

him work for a few minutes and then took him off of the barrel and called two

negroes and made them turn the negro across a barrel and hold him down while

he whipped him again ; and after he turned the negro loose, the negro staggered

off to one side and fell across a lumber pile there, and laid there for a while.” He

died shortly afterwards. An inquest was held, with the coroner’s jury consisting

of the guards of the camp. Witnesses testified that the Negro man had become

overheated and had drunk a great deal of water while in that condition. The camp

physician stated that in his opinion the death was caused by Negro’s drinking too

much water; and that was the verdict reached by the jury.

This form of coercion shared a marked continuity with the forms of violence on the ante-

bellum plantation.

As many historians note, much of the convict labor pool was used for projects aimed

at industrializing the South such as building railways and road infrastructure (Zimmerman

1951; Lichtenstein 1996; Worger 2004). The lease system presented a perfect constellation of

political interests where Southern industrialists and racists were able to find common ground

over this policy. Despite many known reports of the brutality of the system, politicians held

onto this system as a journalist documents, “The state treasury has just lost a quarter of a

million dollars through prohibition, and the complete abolition of the lease system would

take as much or more annually from the state treasury. Some of the legislators are coming

to the conclusion that while morality is a very good thing, it don’t go very well with a low

tax rate” (Taylor 1942a). Both the ways in which incarceration allowed for the re-subjugation

of a newly freed population and to augment Southern state budgets helped maintain the the

survival of this system even in the face of harsh criticism by a variety of groups. On top of all

of this violence done against African Americans, the cash coming in through the lease system

15

helped to finance the white public education system (Taylor 1942b). Violence done against

black bodies helped to nurture white ones. At its core, the consolidation of the convict lease

system eventually turned Georgia into a state of prisoners without any prisons.9

That the rise of this system had little to do with overall crime trends seems to be an agreed

upon point by the historical record. Governor James M. Smith of Georgia in 1872 discusses

this issue directly in referring to the rise of incarceration in Georgia: “...this marked increase

in the number of convicts is not due to any augmentation of crime in the South, but is believed

to be the result entirely of a more rigid and proper enforcement of the laws” (quoted in Ayers

(1984)). Thus, crime alone seems to be insufficient to understanding the rise of the proto-

carceral state.10

As the system expanded, it began to face opposition from an alliance of religious groups

and white labor within the context of reports of corruption in the system. In Georgia, the

suffragist Rebecca Latimer Felton lobbied politicians quite fiercely and coordinated religious

groups to demonstrate their opposition to the system largely because this system seemed too

harsh and dehumanizing (Zimmerman 1951). Ironically, Rebecca Latimer Felton also owned

slaves before the war and was noted for being an ardent supporter of lynching. In tandem

with this moral crusade, labor unions opposed the system since they viewed the largely black

convict labor system as competing with white labor. Eventually, Georgia formally abolished

the practice of convict lease in 1908 with the last lease expiring on April 1st, 1909.

The theory developed in this paper argues that at least part of the rise of this system can be

traced back to the way inwhich Reconstruction incompletely empoweredAfricanAmericans.

The following sections outline the data and empirical strategy used to test this argument.9LeFlouria (2015) documents how the proto-carceral state even broke deep norms around gender roles in

the way in which the system forced women to take on traditionally “masculine” economic roles.10It could be the case that such a statement is insincere. Further results on themechanisms, however, provide

evidence inconsistent with this crime explanation.

16

data and descriptive statistics: evidence from georgia

To test the argument, I build a new dataset combining individual-level and county-level ad-

ministrative records on over 15,000 prisoners over the time period 1817-1911 from Georgia. I

then go onto describe the data and show basic descriptive trends and statistics. Figures A1-5

and A1-6 in the Online Appendix show that Georgia represents a typical case in the South

thus allaying worries about external validity.11

Data

The core research design leverages over-time change within counties to assess the empirical

validity of the aforementioned hypotheses. I assemble pre-period data (1817-1865) from the

Georgia Register of Convicts, which records fairly granular information on the identity of the

incarcerated individual. These include the prisoner’s age, name, county of sentencing, crime,

skin color, eye color, and height. From the skin color, I deduce the race of the prisoner de-

pending onwhether theywere described as being “fair” or “sallow” versus some other descrip-

tion. While skin color does not itself equate with ethnicity, it is the best available information

to ascertain the prisoner’s race. For the post-period, I use the first available Biennial Report

of the Principal Keeper of the Georgia Penitentiary, which covers prisoners entering through

the Georgia State Penitentiary System between October, 1880 to October, 1882.12 The Bien-

nial Report records and classifies prisoners based on their race such as “white”, “negro”, and

“mulatto.” In addition, I collect data on individuals also discharged and pardoned during this

period from the same report so as to capture all individuals who had some contact with the

Georgia Penitentiary System between 1880-1882. 13 Finally, I collect additional post-period

data from the Annual Report of the Principal Keeper of the Georgia Penitentiary for the years11See also Aronow and Samii (2016) on the ways in which regression with larger samples do not recovermore

“representative” causal effects than quasi-experimental methods.12Muller (2018) uses data from this report in a recently published article.13I use the term “colored” when presenting and analyzing the data since that is the term used at the time.

17

1897, 1902, 1905, 1907, and 1911. This data only contains aggregate counts of total convicts for

each county and does not contain data on counts by race.

A key concern regarding bias in these records is the potential for systematic undercount-

ing by race and county. In the pre-period, the key concern is that this data does not pick

up individuals incarcerated for petty crimes. Given police and administrative discretion over

this issue, one would expect potential systematic undercounting of African Americans if they

were disproportionately targeted by the whims of the police bureaucracy. Thus, the results

would under-estimate the “true” effect. In the post-period, a key concern is that officials

might not record everyone. This does not seem to be an issue given the convict lease incen-

tive system. Officials needed to record and keep track of all prisoners in order to properly

account and lease out prisoners. Additionally, the research design uses within-county varia-

tion thereby addressing some of these differential measurement error issues that are constant

within county. Therefore, these potential biases do not seem to be a prodigious issue for the

empirical analysis.

Descriptive Statistics

Having assembled this data, one can instantly see the explosion in incarceration in Georgia.

Figure 2 plots the total incarcerated population over time separately by race. Prior to emanci-

pation, two features are clear. First, the overall scale of the proto-carceral state is quite small

in magnitude. Second, whites seem to be the main incarcerated population–a fact consistent

with qualitative accounts (Merritt 2017). Following the Civil War and emancipation, though,

it is clear that incarceration rises dramatically in Georgia. This is a pattern consistent with the

rest of the South as well (Ayers 1984). What is also clear from the post-period is that African

Americans become an increasingly large fraction of the incarcerated population–a fact that

would be prima facie evidence consistent with the historical record and theory.

Was the rise in African American incarceration following the war racially “biased?” Fig-

18

1820 1840 1860 18800

1000

2000

3000

1820 1840 1860

0

50

100

150

200

250

Year

Tota

l Con

vict

Pop

ulat

ion

Race All Colored

Source: Georgia Convict Ledger

Figure 2: Total Incarcerated Population by Race, 1817-1882

Note: Bottom panel zooms in on period right before Emancipation.

ure 3 plots the number of whites and African Americans incarcerated scaled by each group’s

respective populations in 1880. Relative to their populations, people of color were more than

10 times likely to be incarcerated relative to whites. This, again, is consistent with historian’s

characterization of the Southern penal system across the region (Ayers 1984).

research design: continuous difference-in-differences

Unit of Analysis: County

For the analysis, I use the county as the main unit of analysis. The county is the appropriate

unit during this time for several reasons. First, law enforcement and courts in Georgia were

19

0

50

100

150

200

colored whiteRace

Coun

t of P

rison

ers

per 1

00,0

00 P

opul

atio

n (w

ithin

race

)

Blacks more than 10 times likely to be incarcerated relative to whites.

Massive Racial Bias in Post-Bellum Incarceration Rates in Georgia...

Source: Georgia Biennial Prison Reports (1880-1882) and 1880 Decennial Census.

Figure 3:

organized along county boundaries. Essentially, decisions about who to incarcerate were

made from county-to-county. Second, counties are generally the most comparable unit in

Georgia because it is the lowest level of aggregation for which the ledgers report the sentenc-

ing location. Thus, I proceed to use the county as the main unit of analysis.

Dependent Variables

The main dependent variable that I use for the empirical analysis is the total number of con-

victs in a given county in a given year scaled by the total population in 1860. I scale by the total

population in order to make sure that the dependent variable is comparable across counties,

but use 1860 so as to avoid any issues of post-treatment conditioning (Acharya, Blackwell, and

20

Sen 2016a; Montgomery, Nyhan, and Torres 2018). I use the total convicts for the main analy-

sis instead of the natural logarithm since the parallel trends seems to best hold visually in the

total number (see Figure 4). To ensure that my results are robust to alternative transforma-

tions, I also show that the results hold when using the natural logarithm transformation. To

establish the racialization of punishment in addition to discounting alternative explanations

about crime waves, I also compute alternative dependent variables of the scaled white and

black convict populations.

Independent Variables

Data for the main independent variable in this study–Reconstruction intensity–comes from

Downs and Nesbit (2015)’s Mapping Occupation Project to compute the average number of

troops stationed in a given county in Georgia during Reconstruction. I then scale this by the

1860 population and take the natural logarithm. This variable captures the degree to which

the federal government exerted effort at empowering the newly freed population as per Hy-

pothesis 2.

In addition, I consider a number of other alternative explanations. First, I consider the

potential for the sudden enfranchisement of African Americans in the South to confound the

impact of Reconstruction. To measure this, I compute the county-level proportion enslaved

from the 1860 Federal Census county-level aggregates (Haines). Second, I also consider the

role of the destruction caused by the CivilWar. It could be the case that areas were targeted for

Reconstruction in order to rebuild areas that suffered significant infrastructural damage. To

measure this, I create an indicator for whether a county in Georgia intersected with Sherman’s

March to the Sea where General William Tecumseh Sherman razed counties that his troops

crossed on their way from Atlanta to Savannah, Georgia in 1865.14

14I thank James Feigenbaum for sharing data on the location of Sherman’sMarch. For more on the economicdestruction and recovery in response to Sherman’s March, see Feigenbaum, Lee, and Mezzanotti.

21

Measuring the Mechanism: Black Wealth

Thekeymechanismhighlighted in this paper is the role of blacks’ outside options. Tomeasure

this, I rely on data compiled byW.E.B. Du Bois on blackwealth inGeorgia from 1875-1900 and

presented in the Bureau of Labor’s Bulletin (Du Bois 1901a). This data records the total value

of all property held by African Americans in Georgia consistently in 5-year periods from 1875

to 1900. Despite its incredible richness, this data has yet to be used by political scientists (to

the best of the author’s knowledge). For the analysis, I compute the natural logarithm of the

total wealth of African Americans in a given county in each period.

Measuring the Mechanism: Black Officeholding

To measure the degree to which African Americans were politically empowered, I collect

data on black officeholders during and after the Reconstruction period from Foner (1996). I

addition to measuring the descriptive representation of African Americans during this time

period, black officeholders alsomattered substantively in terms of financing public goods pro-

vision (Logan 2018). This data records the county associated with the officeholder. For the

analysis, I create a binary variable for whether the county had any black officeholders.15

Measuring the Mechanism: Police Force Size

Finally, an implication of the theory is that there should be more investments in carceral ca-

pacity in response to African American’s empowerment during Reconstruction. To measure

this, I collect data from the 1850 and 1880 full count U.S. Decennial Census and digitized

by the Minnesota Population Center (Ruggles et al. 2019). From this, I create a count of the

number of individuals employed as police per 10,000 population in 1850 as a measure of state

carceral capacity.16

15The vast majority of counties only had one black elected official if they had any.16I also show that the results are robust to using the log and absolute level measures.

22

Identification Strategy

A simple cross-sectional comparison between counties with high exposure to Reconstruction

versus low exposure to Reconstruction could be driven any number of pre-existing differences

either observable or unobservable to the researcher. It could be the case, for instance, that

certain counties just have a punitive culture and that this is correlatedwith both the treatments

as well as the amount of incarceration. To eliminate these threats to inference, I rely on a

difference-in-differences (DID) framework to recover the causal effect of the key variables of

interest. This relies on comparing changes in the treatment intensity within unit over time.

While many have pointed out deficiencies of this approach when the treatment time period

is staggered (Goodman-Bacon 2018) or when units can select into timing of treatment (Imai

and Kim 2019), the setup in this paper links most closely to the classical DID framework in

which the units are all forced into treatment at the same time following the end of the Civil

War/abolition (1865) and the end of federal occupation in Georgia (1871). Thus, all units are

forced into treatment with the intensity of the treatment based on intensity of the enforcement

of Reconstruction. These types of continuous DID strategies are common in the economics

literature (Duflo 2001; Bailey 2006; Bleakley 2007).

Within this framework, I estimate equations of the following form:

Yc,t = β1Post-1871t ∗ Log(Avg. Troops)c,1870 + πXc,t + ΩCc + ΓTt + ϵc,t (1)

The main outcome of interest is represented by Yc,t in a given county c in time t. To mea-

sure this, I use the number of incarcerated individuals from a county c in time t divided by the

pre-treatment population in 1860 per 100,000. The results are also robust to a log specification

to limit the influence of outliers. The main causal parameter of interest is β, which represents

the impact of each treatment variable on Yc,t. The terms ΩCc and ΓTt represent county and

23

time fixed effects to eliminate unit-level time invariant confounders and to flexibly control for

global time trends. Additionally, the term πXc,t represents time-varying confounders such

as the impact of slavery post-emancipation and Sherman’s March post-Civil War. The term ϵ

is a stochastic error term. Finally, I estimate Equation 1 using Ordinary Least Squares (OLS)

and cluster standard errors at the county.

0

5

10

15

20

1820 1830 1840 1850 1860 1870

Year

Mea

n Co

nvic

ts p

er 1

00,0

00

Reconstruction Any Reconstruction No Reconstruction

Source: Prison—Inmate Administration—Central Register of Convicts, 1817–1976. Series 21/3/27. Georgia State Archives, Morrow, Georgia. Gregory P. Downs, Mapping Occupation Troop Locations Dataset, 2015

Figure 4: Pre-Trends: Reconstruction

A key implication of the identification assumption is that the treated and untreated coun-

ties should be evolving on parallel trends absent the actual treatment. To visually interrogate

this assumption, I create a dichotomous treatment variable for whether a county had any fed-

eral troops. I then plot the average of the incarceration rate (y-axis) over time (x-axis) sepa-

rately by each treatment group in Figure 4 by Reconstruction-status. This plot shows that both

24

the treated and untreated groups seemed to have parallel trends prior to Reconstruction and

actually very little level differences. Thus, these graphical results provide suggestive evidence

in support of the DID identifying assumption.17

Results

Table 1: Main Difference-in-Differences Estimates: Georgia 1817-1882

Num. Convicts per 100,000(1) (2) (3)

Log(Troops per Thousand) 40.127∗∗ 40.040∗ 44.461∗(20.467) (20.634) (22.733)

Main Effect Reconstruction Yes No NoCounty FE No Yes YesYear FE Yes Yes YesTime-Varying Covariates No No YesNumber of Clusters 137 137 137N 7,989 7,989 7,699∗p < .1; ∗∗p < .05; ∗∗∗p < .01Standard errors clustered by county in parentheses.

First, I begin by presenting the main DID results in Table 1. Column 1 begins by pre-

senting the interaction between Reconstruction and post-1871 controlling for the main effect

of Reconstruction and year fixed effects. Next, Column 2 presents the results of a county

fixed effects model and coefficient remains essentially unchanged. These results suggest that

Reconstruction did indeed increase the size of the incarcerated population.

What do these results mean in substantive terms? Going from a county with no fed-

eral presence to one with the median amount (∆ ≈ Log(3 troops per 1,000)) increased the

number of convicts per 100,000 by about 40. This is approximately a 25% effect above the17I also present additional evidence in FigureA1-4 that Reconstruction exposure is balanced on an exhaustive

set of observable characteristics prior to the Civil War.

25

control-group mean in the post-period. In short, these effects are quite large especially given

the overall explosive growth in the scale of the Georgia incarceration rates.

I also show that these results are robust to controlling for twoprimary alternative explanations–

slavery and Civil War destruction–in Column 3. Encouragingly, the coefficient remains pos-

itive and the magnitude actually increases. This results, then, indicates that other events that

also occurred during this time period are insufficient to explain the patterns found here.

Next, I assess the degree to which these effects persisted using an event study framework.

Here, I estimate lags and leads of the Reconstruction treatment with a regression of the fol-

lowing form:

Yc,t =1905∑

t=1818

(βt1 [Yeart] · Log(Avg. Troops)c,1870

)+ ΩCc + ΓTt + ϵc,t (2)

This not only allows me to estimate the degree to which the effects persist over time, but

also allows me to more formally analyze whether there were pre-existing trends in the data

prior to the end of Reconstruction. I plot the results of this event study analysis in Figure 5

below where each point represents an estimated coefficient of the Reconstruction exposure

measure interacted with a year fixed effect.18

18The baseline year is 1817.

26

Post-Reconstruction

0

25

50

75

1820 1840 1860 1880 1900Year

Estim

ated

Eff

ect

Effect of Reconstruction on incarceration persists over 30 years...

Light shaded region represents 95% confidence intervals. Darker regions represent 90% confidence intervals.

Figure 5: Event Study Estimates

This set of results shows that the impact of Reconstruction seems to have been persistent

over time. Thus, not only did exposure to Reconstruction shape incarceration trends in the

short-run, but also over 30 years after the end of Reconstruction.

27

Num Non-White Convicts per 100,000

Num White Convicts per 100,000

0 10 20 30 40Estimated Effect

Inca

rcer

atio

n Ra

te b

y Ra

ceEffect concentrated among Non-Whites...

Note: Models estimated using county and year fixed effects.

Figure 6: Effect on White versus “Colored” Incarceration Rate

Thick shaded bars represent 90% confidence intervals while light bars represent 95% confi-dence intervals.

Though there is strong evidence showing a causal relationship between slavery or Recon-

struction and incarceration, it might just reflect a general increase in crime. This alternative

explanation potentially has little to do with social control, but instead could reflect crime

waves that took hold in these areas due to changing opportunity-cost incentives. While there

is unfortunately no systematic crime data from this time period at the county-level, the dataset

I assembled does have information on race of the incarcerated individual. If these results are

driven by general crime trends, then one should expect these effects to be symmetric between

whites and non-whites. Using this test, I re-estimate the results from Column 2 of Table 1

and estimate it separately for the white and non-white incarceration rates in Figure 6 shows

28

that the effects are concentrated among African American/colored individuals rather than

whites–a pattern consistent with the argument advanced in this piece and inconsistent with

a general crime channel. Together, the results presented are consistent with the argument

advanced in this piece.19

In addition to the results presented in the paper, I also run a number of robustness checks

(details can be found in the Online Appendix). First, one might be worried about the poten-

tial for military presence during Reconstruction to be driven by factors such as the presence

of racial conflict over the course of Reconstruction. To address this, I demonstrate that the re-

sults are robust to using the location of troops right at the end of theCivilWar–variation that is

plausibly exogenous with respect to later on racial conflict since these locations are primarily

determined by on-the-ground battlefield conditions. Second, a major concern with panel de-

signs is the presence of secular trends by unit (Stephens, Jr. and Yang 2014). I directly address

this issue by estimating models that include county specific linear and quadratic trends. The

point estimates remain unchanged and statistically significant even when estimating these

more demanding specifications. Next, I show that the results remain robust to an alternative

identification strategy that conditions on both county and year fixed effects in addition to the

lagged dependent variable. Then, I show that the results are not driven any one county. I

also show that the results hold when just subsetting to crimes with more discretionary poten-

tial such as property crimes (Muller 2018). Overall, these results provide robust evidence in

support of the argument.19Given the lack of racial breakdowns after 1880, this analysis only uses data from 1817-1880.

29

-40

0

40

80

0.00 0.25 0.50 0.75Moderator: Prop. Enslaved (1860)M

argi

nal E

ffec

t of L

og(T

roop

s pe

r Tho

usan

d) o

n N

umbe

r of N

on-W

hite

Con

vict

s pe

r 100

,000 Effect of Reconstruction on incarceration increasing in intensity of slavery...

Figure 7: Effect Modification of Reconstruction by Intensity of Slavery

Error bars represent 95% confidence intervals of the binned regression estimate. Shaded re-gions represent the 95% confidence intervals of the marginal effect across the support of pro-portion population enslaved in 1860.

Another implication of the argument is that the impact of Reconstruction should only

matter in places with large shares of newly empowered populations. This implies that the ef-

fect of Reconstruction should be increasing in the share of the subaltern population. To test

this, I estimate the impact of exposure to Reconstruction as it varies with the proportion pop-

ulation enslaved in 1860. Figure 7 provides evidence consistent with this hypothesis. While

Reconstruction seems to have had little or even a negative impact on incarceration in areas

with little exposure to slavery, the effect seems to be particularly concentrated in areas with

greater than half the population enslaved.

30

Economic Land

-0.04 0.00 0.04 0.08 0.12 -0.04 0.00 0.04 0.08 0.12

Log(Acres Owned by Blacks per Capita)

Log(Acres Owned by Blacks)

Log(Black Wealth per Capita)

Log(Black Wealth)

Estimated Effect

Out

com

e

but Not Control over Land

Reconstruction Increased Black Wealth

Note: Covariates include 1860 total population, percentage enslaved, rail access, canal access, and wealth.

Figure 8:

Dark shaded bars represent 90% confidence intervals while light bars represent 95% confi-dence intervals.

One major concern with interpreting this interaction is that slavery is not randomly as-

signed. To address this, I also compute the county-level suitability for growing cotton as

a more exogenous measure of a county’s reliance on slavery (Acharya, Blackwell, and Sen

2016b). Results presented in the Online Appendix show that the interaction effect is robust

to using cotton suitability instead of the proportion enslaved thereby supporting a causal in-

terpretation of the effect modification results.

31

Mechanism: The Role of (Incomplete) Black Economic Empowerment

The theoretical framework posits that white elites will bemost likely to use the state to exercise

control over African Americans in areas where African Americans have increased power to

exit coercive relationships with white elites. To test this, I estimate the impact of Reconstruc-

tion on black empowerment as proxied by their wealth as well as landholding. Unfortunately,

data on black wealth does not exist in a readily attainable format prior to Reconstruction so I

am limited to showing simple cross-sectional evidence over time. In this respect, the analysis

here should only be viewed as suggestive. I run analyses regressing the total black wealth in a

county against the measure of Reconstruction exposure used throughout the study control-

ling linearly for the log of the total population, percentage slave population, rail access, canal

access, and overall wealth in 1860. Figure 8 plots the estimated coefficient on Reconstruction

intensity on the total and per capita measure of wealth and land holding. The results provide

suggestive evidence consistent with the theoretical mechanism. Areas with more exposure to

Reconstruction havewealthier black populations, but they did not fundamentally gain control

of land–the root of political power during the period. Additionally, these results are incon-

sistent with an alternative explanation around the role of black criminality since increases in

material status are known to decrease criminal activity (Becker 1968).20 Thus, these results

provide additional empirical support for the theoretical framework advanced in this project

in addition to existing work on historical incarceration (Muller 2018).

Mechanism: The Role of Black Political Empowerment

In addition to looking at the role of black economic empowerment, I also investigate the

impact of Reconstruction on black political empowerment. To do so, I collect data on the20In the Appendix, I also show that there is little support for the notion that Reconstruction made African

Americans at the bottom of the economic distribution worse off. Thus, it is unlikely that these results are drivenby economic incentives to engage in crime.

32

0.00

0.04

0.08

0.12

Bivariate CovariatesModel

Est

imat

ed E

ffect

Reconstruction Increased Black Officeholding...

Note: Covariates include 1860 total population, percentage enslaved, rail access, canal access, and wealth.

Figure 9:

Thick shaded bars represent 90% confidence intervals while thin bands represent 95% confi-dence intervals.

number of black political office holders in a given county following Reconstruction from

Foner (1996)’s directory of black officeholders. The original directory contains detailed bi-

ographical data on all known black/mixed-race elected officials in the post-Reconstruction

South. I collect and aggregate this data to the county-level for Georgia and re-estimate simple

cross-sectional models looking at the impact of Reconstruction exposure on whether a given

county had any black office holders. These results, shown in Figure 9 demonstrate supportive

evidence for the idea that backlash may have occurred because African Americans were not

able to substantially increase their hold over Southern political institutions. On average, areas

withmore exposure to Reconstruction were only slightlymore likely to elect a black politician

33

0

1

2

1850 1860 1870 1880Year

Avg.

Num

ber o

f Pol

ice

per 1

0,00

0

Reconstruction Any Reconstruction Any Reconstruction (Counterfactual) No Reconstruction

Figure 10:

Thick shaded bars represent 90% confidence intervals while thin bands represent 95% confi-dence intervals.

relative to those with less exposure. Thus, these results are again consistent with the results

on land ownership: African Americans were not able to gain control over fundamental eco-

nomic and political institutions thereby creating the space for political backlash by southern

elites.

Mechanism: The Rise of the Southern Police Force

Finally, I assess the role of Reconstruction in increasing the state’s carceral capacity using

data on the county-level growth of the police force from 1850 to 1880.21 To do so, I visually21Data comes from the full population decennial census.

34

demonstrate within a DID framework the growth of the police force comparing trends in the

police force size per 10,000 population between areas with any military presence and those

with none. Figure 10 demonstrates that Reconstruction nearly tripled the size of the police

force in Georgia. These results hold when estimated within a regression framework as well.

Thus, these results demonstrate that Southern elites seemed to have significantly invested in

Georgia’s coercive capacity as proxied by the size of the police.

conclusion

What explains the rise of the proto-carceral state? I explore this question within the context

of Georgia during the 19th century–a state that many historians view as being a paragon of

Southern punishment during the time period. Building off of the work of many scholars, I

synthesize existing arguments into a parsimonious theoretical framework to understand the

ways in which newly empowered subaltern populationsmight only experience such power for

a brief moment in time when a white, wealthy elite maintains control over the state’s coercive

capacity. Using newly assembled panel data on over 130 counties across the time period 1817-

1911, I show that the external (yet incomplete) empowerment of African Americans led to an

increase in incarceration. This effect is concentrated solely among African Americans and not

whites–a pattern consistent with a political account of the rise of this system and inconsistent

with theories stress the role of general crime waves. In addition, data assembled from Bureau

of Labor reports on black wealth in Georgia and data on black elected politicians provide

suggestive evidence consistent with the causalmechanisms. I also show that areas with greater

exposure to Reconstruction also saw a larger increase the size of the police force consistent

with the argument. In short, this project lends systematic empirical credence to arguments

about punishment and state power that go as far back as Du Bois (1901b) yet remain highly

relevant today.

35

While this study uses Georgia as a window into studying the origins of the proto-carceral

state, it remains to be determined the degree to which the lessons learned form this case can

be applied to understand the rest of the United States. Particularly, the South’s history of racial

repression by whites might make the incentives outlined in this article particularly acute rela-

tive to other regions (Acharya, Blackwell, and Sen 2018a). In this respect, the theory may not

apply to societies that lack such racialized conflict. Future research would benefit from focus-

ing on the degree to which racial conflict per se animates the creation of repressive political

institutions like prisons and policing.

The theory and data collected here open up many new avenues of inquiry into the role

of incarceration in the political development of the United States. A natural question that

comes out of this study is whether these early, local incarceration patterns persist until today.

Furthermore, we might ask ourselves under to what degree the policing of African American

communities in the past continues the shape the lives of members of this community today.

Interpreted against the backdrop of other recent work in American politics, which urges

us to more carefully consider the role of repression, violence, and coercion (Soss and Weaver

2017), this study suggests that students of American politics ought to more thoroughly inter-

rogate the way in which state-led repression has so fundamentally shaped themarginalization

of groups based on their identities. For instance, we might ask ourselves to what degree did

the rise of other class-based subaltern movements such as labor suffer from the same sorts of

repressive carceral strategies in the North?22 Not only was repression a strategy the state pur-

sued against the labor movement, but also civil rights organizations following World War II.

Lynchings of African Americans and immigrants, often done under the tacit or explicit con-

sent of the state, served as a way to reinforce the social bonds needed to maintain a repressive

Jim Crow system (Smangs 2016). Even today, the revelation of systematic extraction from

African Americans in Ferguson via the police suggests the continued relevance of state-led22I thank Alex Gourevitch for raising this inquiry.

36

violence for the lives of those on the periphery.

Incarceration, when politically motivated, should be viewed analytically as an act of po-

litical violence. If we take this idea seriously, then existing analytical and theoretical tools

we use to study American politics, could be deficient to understanding cause, process, and

consequence of incarceration as political violence.23 Fortunately, students of comparative

politics have much to contribute to our understanding given their long emphasis on the mi-

cro, meso, and macro-structures that shape political violence (Tilly 2003; Wilkinson 2004;

Kalyvas 2006; Weinstein 2006; Wood 2012; Fujii 2013; Cohen 2016). Thus, this study joins

recent work such as Davenport (2015) and Weaver (2019) in using concepts from the compar-

ative literature on political violence to better understand how repression works in the United

States. Clearly, the historical and contemporary politics of the United States are rife with ex-

amples of political violence; as such, our research ought to take more seriously its occurrence

(and non-occurrence) and how the politics of American repression intersect with the politics

of identity.

23There are, however, some recent studies that attempt to understand the causes and consequences of violenceproduction in American history (Kalmoe 2018; Obert 2018; Hall, Huff, and Kuriwaki, Forthcoming)

37

References

Acemoglu, Daron, and James A. Robinson. 2008. “Persistence of Power, Elites, and Institu-

tions.” American Economic Review 98 (1): 267–293.

Acemoglu, Daron, and AlexanderWolitzky. 2011. “The Economics of Labor Coercion.” Econo-

metrica 79 (2): 555–600.

Acharya,Avidit,MatthewBlackwell, andMaya Sen. 2018a.DeepRoots:HowSlavery Still Shapes

Southern Politics. Princeton University Press.

— . 2016a. “Explaining Causal Findings Without Bias: Detecting and Assessing Direct Ef-

fects.” American Political Science Review 110 (3): 512–529.

— . 2018b. “Explaining Preferences fromBehavior: A Cognitive Dissonance Approach.” Jour-

nal of Politics 80 (2): 400–411.

— . 2016b. “The Political Legacy of American Slavery.”The Journal of Politics 78 (3): 621–641.

Alexander, Michelle. 2012.TheNew Jim Crow: Mass Incarceration in the Age of Colorblindness.

The New Press.

Alston, Lee J, and Joseph P Ferrie. 1993. “Paternalism in Agricultural Labor Contracts in the

U.S. South: Implications for the Growth of the Welfare State.” American Economic Review

83 (4): 852–876.

Aronow, Peter, and Cyrus Samii. 2016. “Does Regression Produce Representative Estimates of

Causal Effects?” American Journal of Political Science 60 (1): 250–267.

38

Ayers, Edward L. 1984. Vengeance and Justice: Crime and Punishment in the 19th-Century

American South. Oxford University Press.

Bailey, Martha J. 2006. “More Power to the Pill: The Impact of Contraceptive Freedom on

Women’s Life Cycle Labor Supply.” Quarterly Journal of Economics 121 (1): 289–320.

Bateman, David A., Ira Katznelson, and John S. Lapinski. 2018. Southern Nation: Congress

and White Supremacy after Reconstruction. Princeton University Press / the Russell Sage

Foundation.

Beck, E M, and Stewart E Tolnay. 1990. “The Killing Fields of the Deep South: The Market

for Cotton and the Lynching of Blacks, 1882-1930.” American Sociological Review 55 (4):

526–539.

Becker, Gary S. 1968. “Crime and Punishment: An Economic Approach.” Journal of Political

Economy 76:169–217.

Bensel, Richard Franklin. 1990. Yankee Leviathan: The Origins of Central State Authority in

America, 1859-1877. Cambridge University Press.

Black, Donald. 1983. “Crime as Social Control.” American Sociological Review 48 (1): 34–45.

Blackmon, Douglas A. 2008. Slavery by Another Name: The Re-Enslavement of Black Ameri-

cans from the Civil War to World War II. Anchor Books.

Bleakley, Hoyt. 2007. “Disease and Development: Evidence from Hookworm Eradication in

the American South.” Quarterly Journal of Economics 122 (1): 73–117.

Caughey, Devin. 2018.TheUnsolid South: Mass Politics and National Representation in a One-

Party Enclave. Princeton University Press.

Chacon, Mario, and Jeffrey Jensen. 2018. “De Facto Power, Democracy, and Taxation: Evi-

dence from Military Occupation during Reconstruction.” Working Paper.

39

Chwe, Michael Suk-Young. 1990. “Why Were Workers Whipped? Pain in a Principal-Agent

Model.” Economic Journal: 1109–1121.

Cohen, Dara Kay. 2016. Rape During Civil War. Cornell University Press.

Cook, Lisa D., TrevonD. Logan, and JohnM. Parman. Forthcoming. “Segregation and South-

ern Lynching.” Social Science History.

Davenport, Christian. 2015.How Social Movements Die: Repression and Demobilization of the

Republic of New Africa. Cambridge University Press.

— . 2007. “State Repression and Political Order.” Annual Review of Political Science 10:1–23.

Downs, Gregory P. 2015.After Appomattox:Military Occupation and the Ends ofWar. Harvard

University Press.

Downs, Gregory P., and ScottNesbit. 2015.MappingOccupation: Force, Freedom, and theArmy

in Reconstruction. http://mappingoccupation.org.

Du Bois, W.E.B. 2014. Black Reconstruction in America (The Oxford W. E. B. Du Bois): An

Essay Toward a History of the Part Which Black Folk Played in the Attempt to Reconstruct

Democracy in America, 1860-1880. Oxford University Press.

— . 1901a. The Negro Landholder of Georgia. Tech. rep. Department of Labor.

— . 1901b. “TheSpawnof Slavery:TheConvict-Lease System in the South.”Missionary Review

of the World 14:737–745.

Duflo, Esther. 2001. “Schooling and Labor Market Consequences of School Construction in

Indonesia: Evidence from an Unusual Policy Experiment.” American Economic Review 91

(4): 795–813.

Eckhouse, Laurel. 2016. “Descriptive Representation and Political Power: Explaining Racial

Inequalities in Policing.” Working Paper.

40

Enns, Peter K. 2016. Incarceration Nation: How the United States Became the Most Punitive

Democracy in the World. Cambridge University Press.

Enos, Ryan D. 2017. The Space between Us: Social Geography and Politics. Cambridge Univer-

sity Press.

Feigenbaum, James J., James Lee, and FilippoMezzanotti. “Capital Destruction and Economic

Growth: The Effects of Sherman’s March, 1850–1920.” NBER Working Paper No. 25392.

%5Curl%7Bhttps://www.nber.org/papers/w25392%7D.

Foner, Eric. 1996. Freedom’s Lawmakers: A Directory of Black Officeholders during Reconstruc-

tion. Louisiana State University Press.

— . 2007. Nothing but Freedom. Louisiana State University Press.

— . 2011. Reconstruction: America’s Unfinished Revolution, 1863-1877. Harper / Row.

Forman, Jr., James. 2017. Locking Up Our Own: Crime and Punishment in Black America. Far-

rar, Straus / Giroux.

Fortner, Michael Javen. 2013. “The Carceral State and the Crucible of Black Politics: AnUrban

History of the Rockefeller Drug Laws.” Studies in American Political Development 27:14–

35.

Foucault, Michel. 1995. Discipline & Punish: The Birth of the Prison. 2nd. Vintage Books.

Francis, Megan Ming. 2014. Civil Rights and the Making of the Modern American State. Cam-

bridge University Press.

Frymer, Paul. 2005. “Racism Revised: Courts, Labor Law, and the Institutional Construction

of Racial Animus.” American Political Science Review 99 (3): 373–387.

— . 2010. Uneasy Alliances: Race and Party Competition in America. Princeton University

Press.

41

Fujii, Lee Ann. 2013. “The Puzzle of Extra-Lethal Violence.” Perspectives on Politics 11 (2): 410–

426.

Gibson, Edward L. 2013. Boundary Control: Subnational Authoritarianism in Federal Democ-

racies. Cambridge University Press.

Gilens, Martin. 1996. ““Race Coding” and White Opposition to Welfare.” American Political

Science Review 90 (3): 593–604.

Gilmore, Ruth Wilson. 2007. Golden Gulag: Prisons, Surplus, Crisis, and Opposition in Glob-

alizing California. University of California Press.

Goldstein, Robert Justin. 2001. Political Repression in Modern America: From 1870 to 1976.

University of Illinois Press.

Goodman-Bacon,Andrew. 2018. “Difference-in-DifferenceswithVariation inTreatmentTim-

ing.” NBER Working Paper No. 25018.

Gottschalk, Marie. 2016. Caught: The Prison State and the Lockdown of American Politics.

Princeton University Press.

— . 2006. The Prisons and the Gallows: The Politics of Mass Incarceration in America. Cam-

bridge University Press.

Greitens, Sheena Chestnut. 2016. Dictators and their Secret Police: Coercive Institutions and

State Violence. Cambridge University Press.

Haines, Michael R. Historical, Demographic, Economic, and Social Data: The United States,

1790-2002. Inter-university Consortium for Political and Social Research. https://www.

icpsr.umich.edu/icpsrweb/ICPSR/studies/2896.

Hall, Andrew B., Connor Huff, and Shiro Kuriwaki. Forthcoming. “Wealth, Slave Ownership,

and Fighting for the Confederacy: An Empirical Study of the American Civil War.” Amer-

ican Political Science Review.

42

Harris, Allison P. 2019. “Can Racial Diversity Among Judges Affect Sentencing Outcomes?”

Working Paper. %5Curl%7Bhttps://www.allisonpharris.com/uploads/1/0/7/

3/107342067/harris_divsent-4-2019.pdf%7D.

Harvey, Anna, and Sidak Yntiso. 2019. “Racial Bias in Criminal Appeals.” Working Paper.

%5Curl % 7Bhttps : / / s18798 . pcdn . co / annaharvey / wp - content / uploads /

sites/6417/2019/07/Racial_Bias_in_Criminal_Appeals.pdf%7D.

Hinton, Elizabeth. 2017. From the War on Poverty to the War on Crime: The Making of Mass

Incarceration in America. Harvard University Press.

Holland, Alisha C. 2016. “Forbearance.” American Political Science Review 110 (2): 232–246.

Imai, Kosuke, and In Song Kim. 2019. “When ShouldWeUse Fixed Effects RegressionModels

for Causal Inference with Longitudinal Data?” American Journal of Political Science.

Jacobs, David, and Aubrey L. Jackson. 2010. “On the Politics of Imprisonments: A Review of

Systematic Findings.” Annual Review of Law and Social Science 6:129–149.

Kalmoe, Nathan P. 2018. “Legitimizing Partisan Violence in Public Opinion: Historical and

Experimental Evidence of Bias in State Violence Views.” Working Paper.

Kalyvas, Stathis N. 2006. The Logic of Violence in Civil Wars. Cambridge University Press.

Katznelson, Ira. 2005. When Affirmative Action was White: An Untold History of Racial In-

equality in Twentieth-Century America. W. W. Norton & Company.

Key, V O. 1949. Southern Politics in State and Nation. University of Tennessee Press.

Kinder, Donald R, and LynnMSanders. 1996.Divided by Color: Racial Politics andDemocratic

Ideals. University of Chicago Press.

King, Desmond S, and Rogers M Smith. 2005. “Racial Orders in American Political Develop-

ment.” American Political Science Review 99 (1): 75–92.

43

LeFlouria, Talitha L. 2015. Chained in Silence: Black Women and Convict Labor in the New

South. University of North Carolina Press.

Lerman, Amy E., and Vesla M. Weaver. 2014. Arresting Citizenship: The Democratic Conse-

quences of American Crime Control. University of Chicago Press.

— . 2010. “Political Consequences of the Carceral State.” American Political Science Review

104 (4): 817–833.

Lichtenstein, Alex. 1996. Twice theWork of Free Labor:The Political Economy of Convict Labor

in the New South. Verso.

Logan, Trevon D. 2018. “Do Black Politicians Matter?” Working Paper. http://www.nber.

org/papers/w24190.

Mancini, Matthew J. 1996. One Dies, Get Another: Convict Leasing in the American South,

1866-1928. University of South Carolina Press.

Merritt, Keri Leigh. 2017. Masterless Men: Poor Whites and Slavery in the Antebellum South.

Cambridge University Press.

Michener, Jamila D. 2018. Fragmented Democracy:Medicaid, Federalism, andUnequal Politics.

Cambridge University Press.

Mickey, Robert. 2015. Paths Out of Dixie: The Democratization of Authoritarian Enclaves in

America’s Deep South, 1944-1972. Princeton University Press.

Miller, Lisa L. 2008. The Perils of Federalism: Race, Poverty, and the Politics of Crime Control.

Oxford University Press.

Montgomery, Jacob M., Brendan Nyhan, and Michelle Torres. 2018. “How Conditioning on

Posttreatment Variables Can Ruin Your Experiment and What to Do about It.” American

Journal of Political Science 62 (3): 760–775.

44

Muller, Christopher. 2018. “Freedom andConvict Leasing in the Postbellum South.”American

Journal of Sociology 124 (2): 367–405.

Murakawa, Naomi. 2014.TheFirst Civil Right: How Liberals Built Prison America. OxfordUni-

versity Press.

Myers, Martha A. 1990. “Black Threat and Incarceration in Postbellum Georgia.” Social Forces

69 (2): 373–393.

Nisbett, Richard E, and Dov Cohen. 1996. Culture of Honor: The Psychology of Violence in the

South. Westview Press.

Obert, Jonathan. 2018.The Six-Shooter State: Public and Private Violence in American Politics.

Cambridge University Press.

Oshinsky, David M. 1996. Worse than Slavery: Parchman Farm and the Ordeal of Jim Crow

Justice. The Free Press.

Pfaff, John. 2017. Locked In: The True Causes of Mass Incarceration and How to Achieve Real

Reform. Basic Books.

Reny, Tyler T., and Benjamin J. Newman. 2018. “Protecting the Right to Discriminate: The

Second Great Migration and Racial Threat in the American West.” American Political Sci-

ence Review 112 (4): 1104–1110.

Rogowski, Jon C. 2018. “Reconstruction and the State: The Political and Economic Conse-

quences of the Freedmen’s Bureau.” Working Paper.

Roth, Randolph. 2009. American Homicide. Belknap Press.

Ruggles, Steven, et al. 2019. IPUMS USA: Version 9.0. Tech. rep. University of Minnesota.

%5Curl%7Bhttps://doi.org/10.18128/D010.V9.0%7D.

Skowronek, Stephen. 1982. Building a New American State:The Expansion of National Admin-

istrative Capacities, 1877-1920. Cambridge University Press.

45

Smangs, Mattias. 2016. “Doing Violence, Making Race: Southern Lynching and White Racial

Group Formation.” American Journal of Sociology 121 (5): 1329–1374.

Smith, Kevin B. 2004. “The Politics of Punishment: Evaluating Political Explanations of In-

carceration Rates.” Journal of Politics 66 (3): 925–938.

Soss, Joe, andVeslaWeaver. 2017. “Police AreOurGovernment: Politics, Political Science, and

the Policing of Race-Class Subjugated Communities.” Annual Review of Political Science

20:565–591.

Stephens, Jr., Melvin, and Duo-Yan Yang. 2014. “Compulsory Education and the Benefits of

Schooling.” American Economic Review 104 (6): 1777–1792.

Stewart, Megan A., and Karin E. Kitchens. 2018. “Statebuilding, Social Order and Violence:

Evidence from U.S. Reconstruction.” Working Paper.

Suryanarayan, Pavithra, and Steven White. 2019. “Slavery, Reconstruction, and Bureaucratic

Capacity in the American South.” Working Paper.

Tajfel, Henri, and JohnC. Turner. 1986. “Psychology of IntergroupRelations.” Chap.The Social

Identity Theory of Intergroup Behavior, ed. by Stephen Worchel and William G. Austin,

7–24. Nelson Hall.

Taylor, A. Elizabeth. 1942a. “The Abolition of the Convict Lease System in Georgia.” Georgia

Historical Quarterly 26 (3/4): 273–287.

— . 1942b. “The Origin and Development of the Convict Lease System in Georgia.” Georgia

Historical Quarterly 26 (2): 113–128.

Tesler, Michael, and David O. Sears. 2010. Obama’s Race: The 2008 Election and the Dream of

a Post-Racial America. University of Chicago Press.

Tilly, Charles. 2003. The Politics of Collective Violence. Cambridge University Press.

46

Valentino, Nicholas A., Vincent L. Hutchings, and Ismail K. White. 2002. “Cues that Matter:

How Political Ads Prime Racial Attitudes during Campaigns.” American Political Science

Review 96 (1): 76–90.

Valentino, Nicholas A., and David O. Sears. 2005. “Old Times There Are Not Forgotten: Race

and Partisan Realignment in the Contemporary South.” American Journal of Political Sci-

ence 49 (3): 672–688.

Vallely, Richard M. 2004. The Two Reconstructions: The Struggle for Black Enfranchisement.

University of Chicago Press.

Walker, Hannah L. 2014. “Extending the Effects of the Carceral State: Proximal Contact, Po-

litical Participation, and Race.” Political Research Quarterly 67 (4): 809–822.

Weaver, Michael. 2019. “‘Judge Lynch’ in the Court of Public Opinion: Publicity and the De-

legitimation of Lynching.” American Political Science Review.

Weaver, Vesla M. 2007. “Frontlash: Race and the Development of Punitive Crime Policy.”

Studies in American Political Development 21 (2): 230–265.

Weinstein, Jeremy M. 2006. Inside Rebellion: The Politics of Insurgent Violence. Cambridge

University Press.

White, Ariel. 2019. “MisdemeanorDisenfranchisement?TheDemobilizing Effects of Brief Jail

Spells on Potential Voters.” American Political Science Review.

White, Ariel R., Noah L. Nathan, and Julie K. Faller. 2015. “What Do I Need to Vote? Bu-

reaucratic Discretion and Discrimination by Local Election Officials.” American Political

Science Review 109 (1): 129–142.

Wilkinson, Steven I. 2004.Votes and Violence: Electoral Competition and Ethnic Riots in India.

Cambridge University Press.

47

Wood, Elisabeth Jean. 2012. Insurgent Collective Action and Civil War in El Salvador. Cam-

bridge University Press.

Worger, William H. 2004. “Convict Labour, Industrialists and the State in the US South and

South Africa, 1870-1930.” Journal of Southern African Studies 30 (1): 63–86.

Wright, Erik Olin. 2000. “Class, Exploitation, and Economic Rents: Reflections on Sorensen’s

‘Sounder Basis’.” American Journal of Sociology 105 (6): 1559–1571.

— . 1973.ThePolitics of Punishment: ACritical Analysis of Prisons inAmerica. HarperColophon

Books.

Young, Lauren E. 2019. “The Psychology of State Repression: Fear and Dissent Decisions in

Zimbabwe.” American Political Science Review 113 (1): 140–155.

Zimmerman, Jane. 1951. “The Penal Reform Movement in the South during the Progressive

Era, 1890-1917.” Journal of Southern History 17 (4): 462–492.

Online Appendix

A0

Contents

Contents A0

Formalizing the Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A1

Data Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A9

Convict Records . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A12

Black Land Values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A14

Balance Check of Pre-Treatment Baseline Observables . . . . . . . . . . . . . . . A15

External Validity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A16

Additional Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . A19

Ruling Out Crime as an Alternative Explanation . . . . . . . . . . . . . . . . . . A25

Robustness Checks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A28

A0

A1

formalizing the framework

Baseline Model Setup and Predictions

I build a static game theoretic model of a potentially repressive society inspired by features

identified by the historical work on Southern society following the Civil War as described

above. This model abstracts away from many features of the specific coalitions and bureau-

cratic features of Southern society so as to elucidate the underlying political economy and

social identity incentives that would influence why Southern elites decided to invest in in-

carceration as a strategy of social control. The model delivers two key predictions, which I

test using newly collected data: (1) that raising the value of the subaltern group’s “outside op-

tions” induces more repression and (2) greater social distance between the elite group and the

subaltern generates more repression.

The society is composed of representative actors from two groups–an elite E and labor

L.24 While I assume that E is unified in terms of types, I allow L to be differentiated by

its social distance from E represented by the term ∆. Larger values of ∆ indicate less social

distance between E and L. This variable captures, in the context of the U.S. South, the social

distance between whites and blacks as guided by race. In doing so, this model parsimoniously

captures situations of class-conflict conditioned by ethnic and racial identity concerns.

Initially, E offers a wage w ∈ R+ to L. Next, L chooses to accept or reject E’s offer.

If L accepts the offer, the game ends and players receive their utilities UE = Π − w and

UL = u(w). In this case, E receives some utility from output under a “free” labor pool Π

minus the amount of wagesw thatE must pay toL. L receives utility u(w) from the contract

with E.25

24I assume a representative actor from each group, which abstracts away from issues like coordination or col-lective action. Since these are not features that are central to the phenomenon I wish to explain, this assumptionis reasonable in this situation.

25These contracts are not “free” in the sense that they are made in the shadow of repression as described inthe next stage of the game.

A2

When L rejects the offer, E can decide to use repression r ∈ R+ against L to forcibly

extract from L. Using repression, however, comes at a cost c(r) = ∆2r2 that is convex and

increasing in the closeness ofE andL in social terms.26 These costs might come from the fact

that repression requires concentrating government resources in “unproductive” ways in ad-

dition to the way in which repression reduces the supply of labor available for other activities.

I allow the cost to vary by ∆ to capture both the psychological costs of engaging in violence

against members of one’s own group and the potential costs that E might play in terms of

maintaining a cross-class coalition among members of their own racial/ethnic group. I al-

low repression to forcibly generate output at the rate β ∈ R+ thus generating benefits for E

equal toΠβr.27 This captures both a general feature of coercion as being “productive”/output-

enhancing as per Acemoglu and Wolitzky (2011) in addition to capturing the feature of the

Southern prison system (and many other carceral systems around the world) that force pris-

oners to work. After deciding how much repression to engage in, actors receive their utilities

UE = Πβr − ∆2r2 and UL = u(g)− r. In this situation where L decides not to accept, they

receive an outside wage equal to g. This outside-wage/option can be thought of as the value

of other careers outside of picking or growing cotton for the planter elite.

To summarize, the sequence of play is as follows:

1. E offers a wage w ∈ R+ to L.

2. L can accept or reject this offer.

• If L accepts, the game ends and each group receives utilities.

3. IfL rejects, thenE chooses the amount of repression r ∈ R+ to engage in. Utilities are

realized.26While the model in this paper parameterizes the cost function c(r) using quadratic loss, the general com-

parative statics hold so long as the cost function is convex and differentiable.27Values ofβ ∈ [0, 1] imply a relatively lower rate of return on repression relative to the ”free-market” output.

A3

Solving the game for Subgame Perfect Nash Equilibria (SPNE) via backwards induction

yields the following results:

Proposition 1: Positive shocks to outside options will induce more repression for sufficiently

high values of β.

This proposition captures the tradeoff that L makes when weighing the benefits of “exit-

ing” and pursuing their outside options against the costs of repression that they might face

when doing so. From the perspective of E, repression is only credible when the benefits are

sufficiently high. So long as the value of g is sufficiently high, L still stands to benefit from

pursuing the outside option. As described above, African American ex-slaves pursued many

ways to “exit” from their previously coercive relationships with the planter elite despite the

risks that this would entail. In this sense, embracing a “brief moment in the sun” entailed the

risk of going back into slave-like relationships through state repression.

While the above model makes the assumption of deterministic costs, I relax this assump-

tion in the Online Appendix and show that the basic results hold when considering the po-

tential to be repressed as a costly lottery with risk-neutral preferences. I also consider an ex-

tension where the value of the outside option g is probabilistically determined and the basic

results hold.

Proposition 2: Greater social distance between groups induces more repression.

The intuition behind this result is straightforward. Repression, as described above, re-

quires the elite to engage in actions that might raise the risk of revolt, breaking one’s ethnic

coalition, or simple psychological costs of doing harm to member’s of one’s social identity

group. As these groups get farther apart in social distance (ie, the identity group of the labor

pool is farther apart from the identity group of the elite), the cost of engaging in repression

by E against L becomes lower. Thus, increases in social distance increases repression.28

28Repression itself can also endogenously increase social distance. See Acharya, Blackwell, and Sen (2018b)for a formal model of this process.

A4

Moreover, the impact of social distance should be increasing in the value of the outside

option. This is because for low values of g, repression is less likely to happen in equilibrium.

As g increases however, repression increases at a rate inversely proportional to social distance.

The theory, then, makes several testable predictions:

1. H1: Punishment should be increasing in the size of the subaltern group where the size

proxies for the salience of ∆.

2. H2: Punishment should be increasing in the value of the subaltern group’s outside op-

tions g.

3. H3: The impact of group size should be especially salient where the group has a higher

outside option.

Model Proofs and Extensions

Proof. Proof of Proposition 1: The Effect of Outside Options

Given the Subgame Perfect Equilibrium Concept used in this paper, I solve the game via

backwards induction. Starting at the terminal node whereE must choose a level of repression

r ∈ R+. At this node, E’s utility function is UE = Πβr − ∆2r2.

Taking the derivative with respect to r, we get the following first-order condition:

∂UE

∂r= Πβ −∆r (3)

Setting this equal to zero, we can solve for the optimal level of repression r∗:

r∗ =Πβ

∆(4)

We can now substitute E’s expression for the optimal level of repression to characterize

L’s “exit” constraint:

A5

u(g) ≥ u(w) +Πβ

∆(5)

Here, L chooses to “exit” and pursue their outside options if and only if their utility from

the outside option is greater than the utility they would receive from entering into “wage”

labor plus the cost he would face from repression.

From this constraint, we can see that positive shocks to g will induce more repression.

A6

Proof. Proof of Proposition 2: The Effect of Social Distance

Consider E’s expression for the optimal level of repression r∗. From this equation, it is

clear to see that an increase in ∆ (closer social distance between the E and L) reduces the

optimal level of repression. This is because the derivative of r∗ with respect to ∆ is strictly

negative:

∂r∗

∂∆=

−Πβ

∆2< 0 (6)

A7

Proof. Risky Lottery of Labor’s Outside Option

Instead of considering a deterministic cost of repression for labor, we can reinterpret that

term as an “expected” utility as follows:

UL = (1− p)U(g)− pc(r) (7)

Where p ∈ (0, 1] represents the probability of getting caught. So long as U(g) is still

sufficiently high, L will take the risky lottery.

Now we can re-write L’s ”exit” constraint as:

(1− p)U(g)− pr∗ ≥ u(w) (8)

=⇒ U(g) ≥ u(w) + pr∗

1− p(9)

A8

Proof. Probabilistic Outside Option

Assume we have a mass ofLi which with outside option gi ∼ F (.)with mean µ and F (.)

symmetric around µ.. This allows us to rewrite Li’s constraint as

U(gi) ≥ u(w) + pr∗

1− p(10)

This generates a cutpoint where Li with sufficiently high draws of gi will decide to exit.

All else equal, increasing the average µ increases the proportion of individuals that E will

attempt to repress. Those under the cut-point are offered “free”-market wages.

A9

data appendix

• Incarceration (pre-Reconstruction):

– Source: Georgia Register of Convicts. Accessed through Ancestry.com.

– Level of analysis: Individual

• Incarceration (1880-1882):

– Source: Biennial Report of the Principal Keeper of the Georgia Penitentiary, 1880-

1882

–

– Level of analysis: Individual

• Incarceration (1897-1911):

– Source: Annual Report of the Principal Keeper of the Georgia Penitentiary

– Level of analysis: County

• Population (1810-1910):

– Source: Haines, Michael R., and Inter-university Consortium for Political and So-

cial Research. Historical, Demographic, Economic, and Social Data: The United

States, 1790-2002. Ann Arbor, MI: Inter-university Consortium for Political and

Social Research [distributor], 2010-05-21.

– Level of analysis: County

• Average number of troops during Reconstruction

– Source: Gregory P. Downs and Scott Nesbit, Mapping Occupation: Force, Freedom,

and the Army in Reconstruction

A10

– Level of analysis: County

• Proportion population enslaved (1860):

– Source: Haines, Michael R., and Inter-university Consortium for Political and So-

cial Research. Historical, Demographic, Economic, and Social Data: The United

States, 1790-2002. Ann Arbor, MI: Inter-university Consortium for Political and

Social Research [distributor], 2010-05-21.

– Level of analysis: County

• Sherman’s March Locations

– Source: US War Department (1901): The War of the Rebellion: a Compilation of

the Official Records of the Union and Confederate Armies. US Government Printing

Office, Thousand Oaks, CA.

– Level of analysis: County

• Freedmen’s Bureau locations:

– Source: https://www.google.com/maps/d/u/0/viewer?ie=UTF8&oe=UTF8&

msa=0&mid=11kGBhLi6gJ9jvXqF9I2X47uHAgk&ll=32.47155077708247%2C-88.

26405249999999&z=5

– Level of analysis: County

• Total Black Property Wealth (1875-1900):

– Source: Du Bois, W.E.B. 1901. The Negro Landholder of Georgia. Department of

Labor.

– Level of analysis: County

• Black Officeholding:

A11

– Source: Foner, Eric. 1996. Freedom’s Lawmakers: A Directory of Black Officehold-

ers during Reconstruction. Louisiana State Press.

– Level of analysis: Individual

• Police Force Size:

– Source: Steven Ruggles, Sarah Flood, Ronald Goeken, Josiah Grover, Erin Meyer,

Jose Pacas, and Matthew Sobek. IPUMS USA: Version 9.0 [dataset]. Minneapolis,

MN: IPUMS, 2019.

– Level of analysis: Individual

A12

convict records

Figure A1-1: Sample from the Georgia Convict Register, 1817-1868

A13

Figure A1-2: Sample from the Biennial Report of the Georgia Penitentiary, 1880-1868

A14

black land values

Figure A1-3: Sample from the Bureau of Labor Bulletin on Black Wealth

A15

balance check of pre-treatment baseline observables

The lack of level differences in the pre-treatment period suggests two things: (1) that the

difference-in-differences design essentially leans on cross-sectional variation for estimating

causal effects and (2) places that experienced Reconstruction did not seem to differ on the

baseline outcome variable. To assuage concerns over cross-sectional imbalances, I perform

an omnibus balance check regressing the Reconstruction exposure variable on an exhaustive

set of observable, pre-treatment covariates. These include demographics such as population,

urbanization, and slave intensity; economic variables such as manufacturing intensity and

wealth; social variables such as church counts; and climactic variables such as suitability for

growing different crops. The test here suggest that Reconstruction exposure is balanced on

these pre-treatment characteristics thereby supporting the causal interpretation of the results.

A16

Cotton Suitability

Log (Acres Improved)

Log (Mfg. Estab.)

Log (Mfg. Output)

Log (Personal Estate)

Log (Real Estate)

Log (Total Pop.)

Log(Churches)

Maize Suitability

Pct. Free Blacks

Pct. Slave

Pct. Urban

Rail Access

Rice Suitability

Sugarcane Suitability

Tobacco Suitability

Water Access

Wheat Suitability

-2 -1 0 1

Estimated Coefficient

Var

iabl

e (1

860)

Balance on Pre-Treatment Observables

Omnibus F-test p-value = 0.175

Figure A1-4: Balance Check

external validity

Though the study leans on the case of Georgia for both analytical and empirical reasons, Geor-

gia is emblematic of the Former Confederacy. The following figures demonstrate that this is

the case using census data from 1860. Thus, there are reasons to believe that this case is in-

structive of broader patterns in the South.

A17

Pct. Enslaved Pct. Free (of Blacks)

0 40 80 0 40 80Alabama

Arkansas

Florida

Georgia

Louisiana

Mississippi

North Carolina

South Carolina

Tennessee

Texas

Virginia

value

stat

e_na

me

Figure A1-5: Distribution of Demographic Variables Across the Former Confederacy

A18

Log(Real Estate Wealth) NA

Log(Mfg. Establishments) Log(Personal Estate Wealth) Log(Population)

0 5 10 15 0 5 10 15

0 5 10 15Alabama

Arkansas

Florida

Georgia

Louisiana

Mississippi

North Carolina

South Carolina

Tennessee

Texas

Virginia

Alabama

Arkansas

Florida

Georgia

Louisiana

Mississippi

North Carolina

South Carolina

Tennessee

Texas

Virginia

value

stat

e_na

me

Figure A1-6: Distribution of Economic Variables Across the Former Confederacy

A19

additional descriptive statistics

0.00

0.02

0.04

0.06

0 20 40 60 80

Age (in years)

dens

ity

Race white colored

Age Distribution by Race

Figure A1-7: Age Distributions of Convicts in 1880-1882 by Race

A20

Other Property Violent

Distribution of Charges by Type: Georgia, 1880-1882

One box = one prisoner

Figure A1-8: Distribution of Crimes by Type in 1880-1882

A21

31°N

32°N

33°N

34°N

35°N

86°W 85°W 84°W 83°W 82°W 81°W

0 250 500 750Change in Num. Convicts per 100,000

Change in Num. Convicts per 100,000 between 1860-1880

Figure A1-9: Map of Convict Population Growth in Georgia

A22

Before Reconstruction Post-Reconstruction

0 1 2 3 0 1 2 3

0

100

200

300

400

Log(Troops per Thousand)

Num

. Con

vict

s pe

r 100

,000

Binned scatter plot created using the binsreg package.

Figure A1-10: Binned Scatter plot of raw relationship.

Bins computing using optimal binning procedure developed in Cattaneo, Crump, Farrell andFeng (2019).

A23

Before Reconstruction Post-Reconstruction

0 2 4 0 2 4

0

500

1000

1500

2000

2500

Log(Troops per Thousand)

Num

. Con

vict

s pe

r 100

,000

Figure A1-11: Raw scatter plot of relationship.

A24

31°N

32°N

33°N

34°N

35°N

86°W 85°W 84°W 83°W 82°W 81°W

0 1 2 3 4 5Log(Total Troops per 1000)

Geographic Distribution of Reconstruction Exposure

Figure A1-12: Map of Reconstruction Exposure Variable.

A25

ruling out crime as an alternative explanation

A key alternative mechanism for the findings is the potential for Reconstruction to lead to

more incarceration by disrupting economic conditions and leading to more incentives to en-

gage in crime. While the results on black wealth presented in the main text of the manuscript

run counter to this on average, it may be the case that some African Americans are actually

made worse off as a result of Reconstruction and therefore have more incentives to engage

in crime. To address this issue, I rely on data from the 1880 U.S. Census full count file for

the state of Georgia. From this dataset, I can compute the Log(Occupation Score) for each

individual i in county c in Georgia. This variable, commonly used as a proxy for economic

status in the economics and sociology literatures, represents the log of the income of a person

in a given occupation in thousands of 1950 dollars.

Using this data, I address the alternative explanation that the results may driven by incen-

tives to engage in crime in two ways. First, in Figure A1-13 I show that African Americans

and whites have remarkably similar occupation scores in 1880. Thus, any racial differences in

the economic incentives to engage in crime are minimal just descriptively.

A26

Female Male

0.0 0.5 1.0 1.5 0.0 0.5 1.0 1.5

Black

White

Race

Log(

Occ

upat

ion

Scor

e)Racial differences in incomes small relative to differences in incarceration

Source: 1880 U.S. Census Note: Occupation scores are defined as the median income of an individual in a given occupation in hundreds of 1950 dollars.

Figure A1-13: Racial Differences in Occupation Scores, 1880

Next, I estimate more formally the impact of Reconstruction on the distribution of occu-

pation scores for African Americans. Formally, I estimate equations of the following form:

1

(Log(Occupation Score)i,c,b > Y ∗

)= βLog(Avg. Troops)c,1870+λBirth Yearb+ΩXc+ϵi,c,b

The term Y ∗ is some value across the empirical support of Log(Occupation Score)i,c,b for

African Americans in 1880. I vary Y ∗ across the support of occupation scores and re-estimate

the above equation. This in essence represents the continuous version of a quantile regression

thus allowing me to assess distributional effects. I also control for age fixed effects as well pre-

treatment 1860 county-level covariates such as population, wealth, transportation access, and

A27

proportion enslaved. The results, shown in Figure A1-14 demonstrate that Reconstruction

seems to have essentially raised the middle of the distribution without harming the lower

end. In essence, all African Americans seem to have been made better off by Reconstruction

thereby lending even less support to the alternative mechanism that African Americans were

more likely to engage in crime as a result of some having been made worse off.

0.00

0.02

0.04

0.06

0 1 2 3 4Log(Occscore)

Est

imat

ed E

ffect

Note: Models estimated using OLS controlling for 1860 total population, prop. enslaved, manufacturing, and wealth.

Figure A1-14: Distributional Impacts of Reconstruction on Black Occupation Scores

A28

robustness checks

Robustness to Using Measure of Troops at End of Civil War

Over the course of Reconstruction in Georgia, one might be concerned that exposure to the

military might be driven by factors such as the amount of racial conflict or violence thus con-

founding the relationship between Reconstruction and incarceration. To address this con-

cern, I instead use a measure of troop presence right after the end of the Civil War in 1865 as

themain explanatory variables. The location of troops at the end of the CivilWar is a function

of local battlefield considerations, which are arguably exogenous with respect to later on in-

carceration patterns. Using this measure, I re-analyze the main results and demonstrate that

they remain robust.

Table A1-1: Robustness to Using Troops Located Right after End of Civil War

Num. Convicts per 100,000(1) (2) (3)

Log(Troops per Thousand) 49.644∗∗ 48.314∗∗ 51.192∗∗(22.577) (22.382) (23.467)

Main Effect Reconstruction Yes No NoCounty FE No Yes YesYear FE Yes Yes YesTime-Varying Covariates No No YesNumber of Clusters 137 137 137N 7,989 7,989 7,699∗p < .1; ∗∗p < .05; ∗∗∗p < .01Standard errors clustered by county in parentheses.

A29

Num Non-White Convicts per 100,000

Num White Convicts per 100,000

0 10 20 30 40 50Estimated Effect

Inca

rcer

atio

n Ra

te b

y Ra

ceEffect concentrated among Non-Whites...

Note: Models estimated using county and year fixed effects.

A30

Post-Reconstruction

0

30

60

90

1820 1840 1860 1880 1900Year

Estim

ated

Eff

ect

Effect of Reconstruction on incarceration persists over 30 years...

Light shaded region represents 95% confidence intervals. Darker regions represent 90% confidence intervals.

Robustness to Inclusion of Freedmen’s Bureaus

One potential concern with the analysis is that the results may instead be driven simply by

the location of Freedmen’s Bureaus offices instead of the actual military enforcement of Re-

construction. To address this, I geolocate Freedmen’s Bureau locations onto 1880 Georgia

county boundaries. Then, I re-estimate the main difference-in-differences models with only

Freedmen’s Bureau status interacted with the post 1871 variable and simultaneously with the

main Reconstruction intensity variable used in the main analysis. The results indicate that

the impact of military occupation is unaffected by the location of Freedmen’s Bureaus.

A31

Table A1-2: Difference-in-Differences Estimates (Freedmen’s Bureau): Georgia 1817-1882

Num. Convicts per 100,000(1) (2)

Any Freedmen’s Bureau Office 34.163 −3.933(43.541) (32.203)

Log(Troops per Thousand) 40.531∗∗(18.865)

County FE YES YESYear FE YES YESNumber of Clusters 137 137N 7,989 7,989∗p < .1; ∗∗p < .05; ∗∗∗p < .01Standard errors clustered by county in parentheses.

A32

Interaction with Freedmen’s Bureau Location

The argument highlights the role of empowerment in shaping the incentives for state repres-

sion via incarceration. To further bolster this argument, I investigate whether the military

enforcement of Reconstruction had an interactive effect with the supply of state bureaucratic

resources through the Freedmen’s Bureau. The results suggest that there is indeed a posi-

tive interactive effect. The effects of military exposure seem to be greatest in areas with the

presence of a Freedmen’s Bureau.

0

50

100

150

Any Freedmen's Bureau Office No Freedmen's Bureau OfficeFreedmen's Bureau Presence

Est

imat

ed E

ffect

Concentrated in Areas with Freedmen's Bureau Presence

Impact of Military Exposure

A33

County-specific Linear Trends

Table A1-3: Robustness to County Linear Trends

Num. Convicts per 100,000(1) (2)

Log(Troops per Thousand) 36.368∗ 36.966∗(18.739) (19.668)

County FE YES YESYear FE YES YESTime-Varying Covariates NO YESNumber of Clusters 132 132N 7,989 7,699∗p < .1; ∗∗p < .05; ∗∗∗p < .01Standard errors clustered by county in parentheses.

A34

County-specific Quadratic Trends

Table A1-4: Robustness to County Quadratic Trends

Num. Convicts per 100,000(1) (2)

Log(Troops per Thousand) 25.613∗∗∗ 25.549∗∗∗(9.792) (9.790)

County FE YES YESYear FE YES YESTime-Varying Covariates NO YESNumber of Clusters 132 132N 7,989 7,699∗p < .1; ∗∗p < .05; ∗∗∗p < .01Standard errors clustered by county in parentheses.

A35

Lagged Dependent Variable

Table A1-5: Robustness to Lagged Dependent Variable

Num. Convicts per 100,000(1) (2) (3)

Log(Troops per Thousand) 7.420∗∗ 8.119∗∗ 9.214∗∗(3.486) (3.814) (4.429)

Main Effect Reconstruction Yes No NoCounty FE No Yes YesYear FE YES YES YESTime-Varying Covariates No No YesLagged DV YES YES YESNumber of Clusters 132 132 132N 7,852 7,852 7,567∗p < .1; ∗∗p < .05; ∗∗∗p < .01Standard errors clustered by county in parentheses.

A36

Log Specification

Table A1-6: Robustness to Log Transformation of Total Convicts per 100,000

Log(Num. Convicts per 100,000)(1) (2) (3)

Log(Troops per Thousand) 0.087 0.086 0.074(0.059) (0.059) (0.064)

Main Effect Reconstruction Yes No NoCounty FE No Yes YesYear FE YES YES YESTime-Varying Covariates No No YesNumber of Clusters 132 132 132N 7,989 7,989 7,699∗p < .1; ∗∗p < .05; ∗∗∗p < .01Standard errors clustered by county in parentheses.

A37

Interaction with Cotton Suitability

Table A1-7: Effect Moderation by Cotton Suitability

Num. Non-White Convicts per 100,000(1) (2) (3)

Log(Troops per Thousand)*Post −43.767∗∗ −43.740∗∗ −36.830∗∗(17.862) (17.992) (17.848)

Log(Troops per Thousand) 0.124(0.314)

Log(Troops per Thousand)*Post*Cotton Suitability 0.014∗∗∗ 0.014∗∗∗ 0.013∗∗∗(0.003) (0.003) (0.003)

Log(Troops per Thousand)*Cotton Suitability 0.303(0.852)

Main Effect Reconstruction Yes No NoCounty FE No Yes YesYear FE YES YES YESNumber of Clusters 132 132 132N 7,308 7,308 7,043∗p < .1; ∗∗p < .05; ∗∗∗p < .01Standard errors clustered by county in parentheses.

A38

Impact on Black Migration

Table A1-8: Estimated Impact on Migration

Log(Non-White Population)(1) (2) (3)

Log(Troops per Thousand) 0.093∗ −0.051 −0.008(0.053) (0.057) (0.051)

Trends No Linear QuadraticCounty FE Yes Yes YesYear FE Yes Yes YesTime-Varying Covariates Yes Yes YesNumber of Clusters 137 137 137N 7,854 7,854 7,854∗p < .1; ∗∗p < .05; ∗∗∗p < .01Standard errors clustered by county in parentheses.

A39

Impact on Property Crimes

0

10

20

30

Covariates No CovariatesModel

Est

imat

ed E

ffect

Effect of Reconstruction among Property Crimes

Note: Covariates include 1860 total population, percentage enslaved, wealth, rail access, and canal access.

Figure A1-15: Impact Concentrated among Discretionary Crimes

A40

Results Not Driven by Any One County

0

20

40

60

80

1331

313

107

1329

913

155

1320

913

315

1306

913

191

1317

513

291

1318

713

047

1324

313

235

1308

913

091

1328

513

019

1301

513

271

1306

713

211

1313

313

059

1311

113

255

1312

313

227

1305

513

049

1310

313

301

1327

513

295

1324

113

149

1312

913

093

1313

913

001

1330

313

151

1323

913

229

1306

313

137

1319

713

263

1302

913

317

1325

113

065

1316

713

249

1321

713

119

1317

913

077

1319

913

007

1315

913

027

1306

113

233

1310

113

273

1305

313

193

1303

513

033

1317

713

221

1315

713

245

1303

713

231

1325

913

115

1314

113

009

1309

913

297

1315

313

215

1307

913

163

1329

313

269

1330

713

169

1304

513

145

1308

313

087

1303

913

207

1323

713

057

1311

713

319

1310

513

195

1326

513

125

1303

113

147

1304

113

181

1320

513

051

1313

513

201

1318

913

223

1332

113

073

1324

713

143

1308

513

113

1321

913

203

1328

913

011

1328

113

311

1321

313

097

1309

513

267

1330

513

261

1312

713

185

1307

113

021

1312

1

Fips Code of County Dropped

Est

imat

ed E

ffect

Robustness to Sequentially Dropping Counties.

Note: Models estimated using county and year fixed effects.

Figure A1-16: Results Robust to Sequentially Dropping Counties from Dataset