There is a longstanding scholarly interest in measuring and understanding disparities in sentencing, i.e. differences in the severity of sanctions imposed by judges for similar criminal cases. However, far less is known about disparities arising after sentences are issued by trial judges, at the stage of execution of sanctions. Particularly important and salient are the conditions under which prison sentences are executed, as incarceration is the most severe and coercive type of sanction. In practice, the prison conditions that convicts experience while incarcerated can noticeably differ from one facility to the next, in terms of available activities, services and overall harshness, even within the same judicial system: some prisons are security-oriented while others offer generous rehabilitation services, some prisons are highly overcrowded while others work under-capacity, some prisons are plagued by violence and hostility while others are more peaceful and humane, etc.
Such disparities in prison conditions between similar offenders can represent a great source of inequality and prejudice that is not only unfair, but that can generate frustration and delegitimation of the justice system in the eyes of offenders and the general public. Poor prison conditions can also lead to degrading and inhumane treatments that violate national laws and the European Convention on Human Rights. In addition, because they determine the level of rehabilitation and specific deterrence that incarceration exerts on individuals, prison conditions have a direct causal effect on inmates’ future outcomes, such as reoffending and employment (overall, harsher treatment generally leads to worse post-release outcomes, as found by Chen and Shapiro, 2007; Drago et al., 2011; Gaes and Camp, 2009; Mastrobuoni and Terlizzese, 2022; Tobon, 2022). It is thus important to be able to measure prison conditions as objectively as possible, both at the macro level of a criminal justice system and at the micro level of each prison facility within a given country.
In this paper, we present a new database, the “Prison Conditions in France” database (PCiF Database), and its related “Prison Conditions in France” Index (PCiF Index), that seek to document the quality and diversity of prison conditions for the universe of all 187 currently active French prisons. This initiative, part of a broader effort to analyze disparities in criminal justice in France, is useful for several reasons. First, it aggregates reliable statistical information that is currently spread between several sources, on some of the key characteristics of all French prisons. Such data is informative to researchers but also to a general audience of citizens, journalists, as well as judges and prison administration officials who often lack information on the characteristics of prison facilities and the differences that can exist between neighboring prisons. Second, the database can also prove useful for researchers studying crime and deterrence as a way to open the black box of prison, analyze between-prison disparities and distinguish between “de jure” sentencing (sentence type and length decided by judges) and “de facto” experiences of sentencing (in terms of disutility of prison time, rehabilitation and reentry preparedness, potential criminal capital accumulation, etc.).
To our knowledge, this is the first attempt to characterize the key aspects of prison conditions in all facilities of a given country. Prior work in this area either propose richer information but for a very limited set of prisons (e.g. the very detailed “Prison Conditions Monitoring Index” was only computed for six prisons in Bulgaria) or use a comparative approach to evaluate rules and practices regarding national-level prison conditions across countries (e.g. the European Prison Observatory, the Prison Life Index). Other studies like Coretti et al. (2023) have access to a large range of prison-level characteristics from the Prison Administration but do not specifically provide a measurement of prison conditions.
The remainder of the paper is organized as follows. Section 1 presents the database and the variables that we select and collect. Section 2 presents the index that we propose as an overall measure of the quality of prison conditions, based on rankings between prisons. Section 3 concludes with potential applications of the data.
1-The Database: Selection and Collection of Variables
Our Prison Conditions database seeks to provide information on key dimensions of prison conditions for all 187 currently active prisons in France (as of May 2023). In order to select which characteristics to include or not in our Prison Conditions database among all potential variables, we use six cumulative criteria listed in Table 1.
Table 1. Criteria for inclusion of variables
To summarize, each variable considered for inclusion in the PCiF database has to be relevant for prison conditions, available, well-measured, varying across prisons but rather stable over time, and not redundant with other variables. These criteria exclude possibly important variables that can’t be measured properly as of now, such as health services or access to rehabilitation programs for example.
For the current (first) version of the PCiF database, we include 9 prison-level characteristics that meet our criteria, capturing the following aspects of prison conditions: year of construction, geographic isolation, size, crowding rate, guard workload, access to in-prison work, complaints to the Prison Inspectorate, suicide among inmates, and family visiting rooms. The variables, their sources and calculations, are described in Table 2. Details about each source are provided in Appendix (Table A1).
Table 2. List of variables included in our Prison Conditions Database
In order to validate our selection of variables and the resulting Index, we also collect data on two additional variables that provide external signals of poor prison conditions. The two variables, listed in Table 3, measure whether a prison facility has ever been convicted for unfit or inhumane prison conditions by the justice system (French administrative courts or the European Court of Human Rights), and whether the Prison Inspectorate (CGLPL) has ever issued emergency recommendations against a prison facility.
Table 3. List of additional variables used to validate our Prison Conditions Index
1.3-Revisions, updates and additions
The Prison Conditions Database is designed to be revised and updated on a yearly basis to keep track of changes over time (e.g. in terms of prison crowding) as well as to benefit from future improvements in measurement quality. We also intend to make additions as new variables meet our criteria for inclusion (see Table 1), in particular once they become available with high quality at relatively low cost. Such changes in the future may for example include the revision of our variable proxying access to in-prison work (currently proxied by the per-prisoner area dedicated to workshops) thanks to the future implementation of a new information system by the Prison Administration; or the inclusion of a new variable capturing the overall level of tensions and violence within prisons based on a topic-level analysis of complaints to CGLPL.
2-The Index: From Raw Data to the Prison Conditions Index
In order to synthesize the characteristics collected in the database, we propose a measure of the overall quality of prison conditions within each facility, called the Prison Conditions Index (PCiF index). This index, from 0 to 100, seeks to aggregate in a simple manner information on all the k continuous variables collected in the database (K=8, i.e. all variables listed in Table 2 except for the binary “Family visiting rooms”).
The index is designed as a relative measure of the quality of prison conditions, based on between-prisons comparisons. It is computed as the unweighted average of each prisons’ ranks, also measured from 0 to 100, for all height variables included in the index:
To obtain ranks for a given variable, all prison facilities are initially sorted from “worst” to “best”: the worst facility, e.g. the most overcrowded prison, is ranked 1st, the second-worst is ranked 2nd, etc. Then, the ranks are stretched to range from 0 to 100 for each variable k, using the following formula:
To handle ties, we assign the same rank to prisons sharing the same value for variable k and the next prison in the ranking is assigned a rank that corresponds to the number of prisons that are worse. For two variables, suicide and complaints, there are ties at the top (several prisons with no suicide and no complaint), hence we rescale ranks to obtain a consistent 0-100 scale.
Once the ranks are assigned for each of the eight continuous variables, we compute the unweighted average to obtain an overall index, ranging potentially from 0 to 100. A prison with an index of 0 (100) would correspond to a prison that is the worst (respectively the best) of all 187 prisons in each dimension.
By computing an unweighted average, we make the agnostic assumption that all height dimensions are equally important in measuring prison conditions. However, opinions may differ on which aspects of prison life are most important, so the index can easily be modified to assign different weights to each variable k.
Either weighted or unweighted, the PCiF index allows easy comparisons across all prisons, as well as between prisons from the same region, same type, same period of construction, etc.
As an illustration, Table 4 reports the mean index of prison conditions for the main types of facilities. Unsurprisingly, jails (dedicated to pre-trial detention and short prison sentences) obtain the lowest scores on average but show substantial heterogeneity. Conversely, juvenile prisons (dedicated to prisoners under 18 years old) offer the best prison conditions according to our index.
Table 4. Mean of PCiF Index by type of prison facility
As a validation exercise, we compare the PCiF index for prison facilities which have ever versus never been convicted by courts in cases of unfit or inhumane conditions. The index is significantly lower for convicted prisons than other facilities (-8 points). Similarly, the index is lower in prisons that received emergency recommendations by the Prison Inspectorate (-5 points). These differences tend to survive in multivariate regressions after controlling for prison type and size, thus confirming that our PCiF index captures relevant disparities in prison conditions also detected by external institutions.
3-Conclusion and potential applications
The Prison Conditions in France database corresponds to one of the very first attempts to collect data on the universe of prison facilities in a given country, in order to produce quantitative measures of prison conditions.
We argue that building and granting access to such data, and its associated index of prison conditions, can prove useful for different actors: legal practitioners (judges, prosecutors, lawyers) to make better-informed decisions, journalists and civil-society organizations to document prison conditions and their disparities, and researchers to study how stakeholders integrate prison conditions in their behaviors.
At least two research questions are worth exploring with this data. First, how do prison conditions affect the chances of successful reentry among prisoners? And which dimension of prison life are most critical in preventing or fueling recidivism? Second, how do judges incorporate (or fail to incorporate) information on prison conditions to make their decisions? And do they even have reliable assessments of the prison conditions in local facilities?
The current version of the dataset, denoted PCiF-v1, is hosted on the Nakala repository (DOI: 10.34847/nkl.fb9c58uv). It is made freely available under CC BY-NC 4.0 license, starting from April 1, 2024 onward. Future versions shall be distributed under similar conditions.
Chen, M. K. and Shapiro, J. M. (2007). Do harsher prison conditions reduce recidivism? A discontinuity-based approach. American Law and Economics Review, 9(1), 1–29.
Coretti, S., Fedeli, S., and Santoni, M. (2023). Assessing the ethics of prison policies to ensure human rights compliance: Suicides and self-inflicted critical events in Italian prisons. European Journal of Political Economy, under press.
Drago, F., Galbiati, R. and Vertova, P. (2011). Prison Conditions and Recidivism. American Law and Economics Review, 13(1), 103–130.
Gaes, G. G. and Camp, S. D. (2009). Unintended consequences: experimental evidence for the criminogenic effect of prison security level placement on post-release recidivism. Journal of Experimental Criminology, 5(2), 139–162.
Mastrobuoni, G. and Terlizzese, D. (2022). Leave the door open? Prison conditions and recidivism. American Economic Journal: Applied Economics, 14(4), 200–233.
Spohn, C. (2009). Sentencing disparity and discrimination, in How do judges decide? The search for fairness and justice in punishment. SAGE Publications.
Tobon, S. (2022). Do better prisons reduce recidivism? Evidence from a prison construction program. Review of Economics and Statistics, 104, 1256–1272.
Table A1. Details on data sources
 Sentencing disparities can occur across courts, across judges, as well as across defendants within judges based on individual characteristics like gender and race, or even depending on external events like weather or sports results (see Spohn (2009) for an introduction).
 The “Prison Conditions Monitoring Index” was developped in 2015 by researchers of the Center for the Study of Democracy https://csd.bg/fileadmin/user_upload/publications_library/files/22285.pdf
 The Prison Life Index is currently under construction by the specialized media Prison Insider, with a coverage of approximately 50 countries: https://www.prison-insider.com/en/comparer/prison-life-index
 Correti et al. (2023) use detailed administrative data on all 188 Italian prisons, from 2016 to 2021, to study the determinants of suicide and self-harm at the prison level.
 By restricted-access we mean data that were obtained through agreements with institutions which usually do not publicly release their data. This is the case for data on all referrals by prisoners to the Prison Inspectorate (CGLPL) and for data on prisoner suicides collected by a specialized non-profit (OIP).
 This implies that we can collect data either automatically from existing files or by hand after reading easily accessible documents (such as CGLPL’s prison visit reports). This criteria excludes variables that could only be obtained by conducting local visits or in-prison surveys for example.
 An alternative might be to build an absolute index, but it is practically very difficult to imagine how to set the absolute standard that prisons should meet on each of the selected variables: an occupancy rate of 100% or 50% or even 0%? An infinite space for prison work?
 As an example, imagine that the two worst prisons in terms of overcrowding share the same overcrowding rate. They will both get assigned a rank of 1. The next-worst facility will therefore get a rank of 3.
 Rescaling uses the following formula: (rank - minimal-rank) / (maximal-rank - minimal-rank) * 100