12 June 2018

US Prison Economy

The fascinating 'Economic Consequences of the U.S. Convict Labor System' by  Michael Poyker considers
the economic externalities of U.S. convict labor on local labor markets. Using newly collected panel data on U.S. prisons and convict-labor camps from 1886 to 1940, I show that competition from cheap prison-made goods led to higher unemployment, lower labor-force participation, and reduced wages (particularly for women) in counties that housed competing manufacturing industries. At the same time, affected industries had higher patenting rates. I find that the introduction of convict labor accounts for 16% slower growth in U.S. manufacturing wages. The introduction of convict labor also induced technical changes and innovations that account for 6% of growth in U.S. patenting in affected industries. 
Poyker notes
Convict labor is still wide-spread, not only in developing countries but also among the world’s most developed countries.  In 2005 U.S. convict-labor system employed nearly 1.4 million prisoners, among them 0.6 million worked in manufacturing (constituting 4.2% of total U.S. manufacturing employment). Prisoners work for such companies as Wal-Mart, AT&T, Victoria’s Secret, and Whole Foods, and their wages are substantially below the minimum wage, ranging from $0 to $4.90 per hour in state prisons. 
Convict labor may impose externalities on local labor markets and firms. Prison-made goods are relatively cheap. Companies that hire free labor find it harder to compete with prisons, especially in industries that rely on low-skilled labor. They face lower demand on their products, pushing down their labor demand. Excess labor moves to industries not competing with prisons and overall wages decreased. Convict labor affects firms, too. Many (predominantly labor-intensive firms) go out of business, unable to compete with prison-made goods, even when they lower wages.  Finally, those affected firms that do not close have to innovate and adopt new technology, either to decrease their production costs, or to produce higher-grade goods that do not compete with prison-made goods.  In this paper, I use a historical setting to evaluate the effect of competition with prison-made goods on firms and free workers. It is challenging to identify causal effects of convict labor in the contemporary setting, since the data on prison output are not available, and due to the embedded endogeneity problem. First, U.S. prisons are built in economically depressed counties under the assumption that they will provide jobs (e.g., guards) in the local labor market (Mattera and Khan (2001)). Second, contemporary convict-labor legislation is endogenous. For these reasons I rely on the historical setting, to identify the effects of convict labor. I digitize a dataset on U.S. convict- labor camps and prisons. Starting in the 1870s, states enacted laws that allowed convict labor, but the timing varied from state to state. Its introduction was unanticipated, both by firms and by prison wardens, who were suddenly in charge of employing prisoners within their institutions.
Moreover, as all convict-labor decisions were determined at the prison level, subsequent changes in convict-labor legislation were exogenous to the choices of individual prison wardens. In addition, I use the fact that pre-convict-labor-era prisons were built without any anticipation that they would be used to employ prisoners. In comparison with contemporary prisons, old prisons were built in populated areas with higher wages and employment, which hinders my ability to find negative effects on local labor markets. Finally, the historical setting allows me to document long-run effects of convict labor in a developed country.
To elicit the effect of prison-labor competition on the local labor market, I construct a county- decade panel data set spanning 1850 to 1950. I measure the exposure of each county to convict labor as the industry-specific value of convict-made goods in all U.S. prisons weighted by the county’s industry labor share and by the distance from those prisons to the county centroid. This imposes two central assumptions: low labor mobility across counties, and iceberg costs of trade.
I estimate the effect of exposure to convict labor on manufacturing wages, employment outcomes, and patenting rates using ordinary-least-squares specification with fixed effects. While the panel dataset allows me to account for time- and county-invariant unobserved heterogeneity and state- specific time trends, three endogeneity concerns remain. First, there is an omitted-variable bias due to the endogenous choice of industry and the amount of goods produced by prisons. Second, prisons could be strategically located to earn higher profits for their states. Third, convict labor was used in industries where local labor unions were stronger and the wage growth rate was higher (Hiller (1915)). To address these concerns, I employ an instrumental variable estimation. I use state-level variation in the timing of passage of convict-labor laws interacted with the capacity of prisons that existed before convict-labor laws were enacted to construct an instrument for the prevalence of convict labor. Prison production was determined by a prison’s warden, and the state-level legislature can be considered exogenous. Old prisons were built without any anticipation that they would be used for production of goods; their locations were determined primarily by population size and urban share of population. Thus, conditional on factors important to the location of the old prisons, the interaction of convict-labor legislation and capacities of old prisons is likely uncorrelated with wardens’ activity and possible strategic location of prisons constructed after convict-labor systems were enacted. I find that the introduction of convict labor in 1870-1886 accounts for 16% slower growth in manufacturing wages in 1880-1900, 20% smaller labor-force participation, and 17% smallermanufacturingemploymentshare.
Comparingtwocounties,oneatthe25thpercentileand the other at the 75th percentile of exposure to convict labor, the more exposed county would on average experience a 2 percentage-point larger decline in mean log annual wages in manufacturing, a 0.9-percentage-point larger fall in manufacturing employment share, and a 0.6-percentage-point larger decline in labor-force participation.
While prison labor was used in quite a few industries, most prisons were producing clothes and shoes. Apparel and shoemaking industries employed mostly women, and they were the most affected by coerced labor. Female wages decreased 3.8 times more than those of men.
I also show that convict-labor shocks affected technology adoption. Comparing two counties, one at the 25th percentile and the other at the 75th percentile of exposure to convict labor, the more exposed county would be expected to experience a 0.6-standard-deviation larger number of registered patents in industries where prisoners were employed. I calculate that the introduction of convict labor accounts for 6% of growth in U.S patenting in affected industries. Because forms of convict labor differed in the North and South, I analyze subsamples. I show that results are mainly driven by the Northeastern and Midwestern states. For the Southern states, all coefficients remain significant, while the magnitudes of all effects are smaller.
I show that the results are robust to various model specifications and ways I construct the explanatory variable. Results hold if I use exposure to convict labor, weighted only by distance to prison (i.e. disregarding industry shares). I also demonstrate that results are not entirely driven by differences between counties with and without prisons: I find that results hold within the sample of counties with prisons. Then, comparing counties with prisons to counties adjacent to counties with prisons, and to second-order adjacent counties, I find that effects of convict labor decay with distance. Also, I find no effect on manufacturing outcomes when using as a placebo convict-labor output in farming. Further, I find no significant effect of convict labor on the number of patents in industries where prisoners were not employed. Finally, I employ firm-level repeated cross-section data for 1850-1880 from Atack and Bateman (1999) to show that firms in affected industries experienced larger decreases in wages. The firm-level data also suggest a decrease in the number of firms in affected labor-intensive industries.
My results relate to three broad economic literatures. I find that the problem of convict labor is similar to the discussion of low-skilled labor competition related to trade shocks (Autor, Dorn and Hanson (2013, 2016), and Holmes and Stevens (2014)). I find that local labor-market shocks come not only from foreign competition or technological progress but can arise from internal sources. Besides, my findings relate penitentiary policies to patterns of directed technological progress (Acemo ̆glu (2002, 2007), and Autor et al. (2016)). I provide evidence in support of findings in Bloom, Draca and Van Reenen (2016) that firms increase patenting as a way to survive competition. Moreover, in contrast to these recent shocks, I estimate the long-run effects of competition coming from the convict labor system. While sociologists and criminologists thoroughly studied convict labor in the 20th century, only a few qualitative papers raised the topics of competition be- tween prison-made goods and products created by free laborers (Roback (1984), McKelvey (1934), and Wilson (1933)).
The rest of the paper is organized as follows. Section 2 reviews the existing literature and relates this paper’s contributions to it. Section 3 introduces the history of U.S. convict labor and the records of its competition with free labor. Section 4 describes the data. Section 5 presents my identification strategy and estimation results. Section 8 assesses the possible impact of the contemporary U.S. convict-labor system and concludes.