Antecedents of subjective severity of detention and perceived procedural justice

ABSTRACT

In this paper, we study what factors contribute to the extent that detained individuals (a) perceive their time in detention as severe and (b) perceive their treatment by prison staff as procedurally just. More specifically, the aim of the study is to examine the antecedents of subjective severity of detention (SSD) and perceived procedural justice (PPJ) with the aim to identify individual and situational characteristics that contribute to such perceptions. Our analyses were based on data from the Prison Project (n = 1430), which includes detailed information on measures of SSD, and PPJ among Dutch males held in Dutch penitentiary institutions. Based on their SSD and PPJ scores, detained individuals were classified as belonging to one of four subgroups (reference group, high SSD, high PPJ, or high both). Using a large set of background variables, we found that older age, a less elaborate criminal history, no daily drug use before arrest, not having experienced any victimization by prison staff, and the personality traits of neuroticism, conscientiousness, and agreeableness were the most relevant antecedents for ‘high both’ subgroup membership.

KEYWORDS:

• Procedural justice
• sanction severity
• detainees
• prison
• antecedents

Acknowledgements

We thank Dr. Wouter Steenbeek (NSCR) for intellectual discussion on confounder selection and etiological modelling. We thank Prof. Dr. Stijn Ruiter (Utrecht University, NSCR) and Prof. Dr. Henk Elffers (NSCR) for comments on our manuscript.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data Availability Statement

Due to the nature of this research, participants of this study did not agree for their data to be shared publicly, so supporting data is not available.

This work was supported by the Dutch Research Council (NWO): [Grant Number NWO: 406.18.RB.011].

Correction Statement

This article has been republished with minor changes. These changes do not impact the academic content of the article.

Notes

1 Readers should be aware that data at T1 were collected between 2010 and 2011. Overall the prison population decreased by 22.5% from 2010 to 2021 (Aebi et al., 2022). The percentage males in detention was 93.9% in 2010 and 95.3% in 2021 and the mean age of the prison population was 34.5 and 37.4 respectively for 2010 and 2021. A comparable percentage of foreign males were incarcerated in 2010 compared to 2021(19.8% and 23.3%, respectively; Aebi & Delgrande, 2013; Aebi et al., 2022).

2 We conducted a complete case analysis/ listwise deletion. The missingness pattern was found to be completely at random for the majority of variables, except for ethnicity, age at first arrest, criminal network and the proportion female officers. We conducted additional sensitivity analyses using multiple imputation to test how our results may be influenced by missingness patterns.

3 Participants were included in the analyses if they had a valid value on at least 50% of the items.

4 Participants were included in the analyses if they had a valid value on at least 50% of the items.

5 The Robust Comparative Fit Index indicates that our assumed model for SSD fits the data better than a model assuming no relationship (CFI >.95); for PPJ, the CFI is just below this cut-off (CFI = .94). Additionally, the robust RMSEA is well below the cut-off of .06 for SSD (RMSEA = .04), which indicates a good model fit. For PPJ it slightly exceeds this cut-off (RMSEA = .09). Moreover, the standardized root mean square residual of the SSD and PPJ scale is below the cut-off of .08 (PPJ = .04; SSD = .01), which indicates an adequate fit. Using the Kaiser criterion (eigenvalue > 1), the results indicated a one-factor solution for both SSD and PPJ. All items contributed significantly to their assumed underlying factor. For SSD, the standardized R² ranged from .44 to .72. For PPJ, the standardized R² indicated that the explained variance per item ranged from .39 to .68, whereby for both scales, more than 50% of the items showed a standardized R² above .5. Analyses can be requested from the first author.

6 Participants were included in the analyses if they had a valid value on at least 50% of the items.

7 This criminal network indicator was based on a question that asked participants to write down a maximum of five names of people with whom they had discussed criminal activities, knowledge, and skills related to crimes during six months prior to arrest.

8 It is important to note is that a large fraction (49%) of the days detained individuals spent in correctional institutions in 2010 in the Netherlands was spend in a pre-trial detention facilities (Aebi & Delgrande, 2013). Moreover, 71% of the individuals in detention are detained shorter than three months and 25% of the individuals are released directly after having received their verdict (DJI, 2013).

9 The variant of the item ‘Prison staff treat everyone equally’ was not included when PPJ by the police was assessed.

10 Participants were included in the analyses if they had a valid value on at least 50% of the items.

11 Overall, the model fit indexes indicate adequate model fit for PPJ at arrest (CFI  = .95; RMSEA = .09; SRMR = .03). Using the Kaiser criterion (eigenvalue > 1), the results indicated a one-factor solution for PPJ at arrest. All items contributed significantly to their assumed underlying factor. The standardized R² per item ranged from .54 to .69.

12 On request of an anonymous reviewer we repeated the analyses for the continuous measure of SSD and PPJ (this time using linear regression). While these additional analyses answer slightly different research questions we think that the results presented in the appendix (Table A2 and A3) add valuable information to the field of prison research.

13 Since the respondents in our study are clustered within detention facilities and detention units we examined whether we should take this nesting into account. We, therefore, followed the steps outlined by Sommet and Morselli (2017) and arrived at the conclusion that a one-level multinomial logistic regression is sufficient to adequately model our data, because the intra class correlation coefficient was close to zero, indicating that the independence of residuals and the design effect was below two.

14 Confounder selection relies on information from theory and prior research. While we carefully selected the confounders based on such information, model misspecifications can never be fully excluded. On the request of an anonymous reviewer, we added the results of the model that contains all variables to the appendix (Table A1).

15 We did five imputations using logistic regression to impute ethnicity, education, employment, education, homelessness, children and both victimization variables; polytomous logistic regression to impute type of offense and unit type; proportional odds modelling to impute the outcome measures; the remaining variables were imputed using predictive mean matching. Analyses can be requested from the first author.