Does the public sector respond to private competition? An analysis of privatization and prison performance

ABSTRACT

The competition thesis states that the introduction of competition from private-sector service providers will spur performance improvements in previously monopolistic public-sector service providers, who fear (further) delegation of their responsibilities to the private sector. This article examines the competition thesis in the context of incarceration. Using data on U.S. adult correctional facilities in 2000 and 2005, it employs a difference-in-difference strategy to compare over-time performance changes among newly competitive facilities relative to non-competitive facilities. Prison performance is measured along four dimensions (safety, order, activity, and conditions), using survey responses from prisons. The results do not show a beneficial competition effect; prisons in newly competitive states experienced performance change in ways that were statistically indistinguishable from prisons in non-competitive states. Supplemental analyses reveal this finding to be robust to most modifications to the sample and the model. A discussion considers four reasons – constitutional safeguards, professional standards, labor opposition, and non-credible threats – incarceration may be resistant to a competition effect and the implications for public policy.

KEYWORDS:

• Privatization
• prison
• punishment
• competition
• public policy

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

1. Their research was funded in part by the Corrections Corporation of American (now ‘CoreCivic’) and the Association for Private Correctional and Treatment Organizations, as revealed in the paper’s acknowledgments section.

2. The Census excludes local jails, immigration detentions facilities, privately operated facilities that are not primarily intended for state or federal inmates, juvenile facilities, military facilities, U.S. Marshals Service facilities, Bureau of Indian Affairs facilities, or hospital wings or wards dedicated to prisoners.

3. The Census does not include measures of operating cost for all facilities. Consequently, cost will not be a part of the analysis.

4. The manual matching involved two discretionary coding decisions. First, when the name of the facility stayed substantially the same and the city was unchanged, these facilities were coded as matches (n = 36 observations or 18 facilities). However, if a facility in the same city changed names and operator, it was coded as separate facilities. Second, facilities that subdivided were coded as separate facilities. Omitting ‘mismatched’ facilities from the analysis does not alter the main results below (supplemental results are available by request).

5. The analytic sample omits federal facilities (the federal system has long used private contractors and was thus not at risk of new competition), joint local-state facilities (the notion of competition is ambiguous with multiple jurisdictions), and observations from Illinois in 2005 (Illinois did not report data in 2005).

6. The data refer to private operation of facilities, irrespective of whether the operating firm owned the physical facility.

7. Facilities in stable competitive (private competition → private competition) and newly non-competitive (private competition → public monopoly) states are omitted from consideration.

8. In robustness checks presented below, this assumption is relaxed and performance is modeled as a function of contemporaneous changes in competition. This alternate specification does little to change the main results.

9. Major infractions exclude ‘minor violations such as horseplay, failure to follow sanitary or other facility regulations, and chewing gum where prohibited’ (Bureau of Justice Statistics 2009, survey question 31).

10. The causal attribution is reliant on the common trends assumption, discussed below.

11. Binary outcomes were also estimated using logistic regression, which produce coefficients on a log-odds metric (rather than probability). Some of these models could not be estimated due to lack of variance in outcomes within particular categories of facilities (Scott and Freese 2006). In all cases in which the outcome could be estimated as a linear probability model and a logistic regression model, the direction and significance of the results at a 95% confidence level were identical. Results from the logistic regressions are available upon request.

12. This period is considered ‘pre-treatment’ because any effects of private competition are assumed to take several years to manifest. Thus, performance trends in the 1995–2000 period should adequately capture trends that were unfolding in an era of public monopoly, even though the treatment group would have private competition by the end of this period. The assumption of delayed competition effects is relaxed below, where a robustness check considers contemporaneous effects of competition.

13. These models include a restricted set of control variables (prisoner population (logged); prisoner gender composition; facility functions (general adult population; reception); and facility age) due to missing data or perfect multicollinearity.

14. Because crowding may decrease as the number of available facilities increases, all other things equal, a supplemental model added a control for the number of correctional facilities in a state. The coefficient for this control variable was not statistically significant, and adding it did not alter the estimates of β₃ in any meaningful way. Results are available by request.

15. With more parameters (19) than states (14), overall tests of joint significance (F-test for OLS, χ² for negative binomial regressions) cannot be computed for models with state-clustered standard errors, and the standard errors themselves are suspect. Supplemental models used conventional standard errors instead of state-clustered standard errors. They yielded results very similar to those presented below, with only one exception; replacing clustered standard errors with conventional ones results in the inmate-on-inmate assaults becoming significant at the 95% confidence level (β₃ = −0.689, s.e. = −0.345).

16. The major disturbances model omits several control variables that were non-significant in previous analyses: inmate gender composition, facility function indicators, and facility age.

17. This finding remains statistically significant when controlling for the number of facilities in a state, which is correlated with both privatization and crowding.

Additional information

Notes on contributors

Brett C. Burkhardt

Brett C. Burkhardt is an assistant professor of sociology in the School of Public Policy at Oregon State University. His research examines the exercise of coercive power by legal and quasi-legal authorities. He is currently involved in four projects: prison privatization, race and punishment, policing mental illness, and police militarization. He holds a PhD degree in sociology from the University of Wisconsin-Madison.