PROTOCOL: Day reporting centres for reducing recidivism: A systematic review

The day reporting centre (DRC) model emerged in Great Britain in the 1960s, and was first seen in the U.S. a couple of decades later (Boyle, Ragusa, Lanterman, & Marcus, 2011; Craddock, 2000). Its arrival coincided with a wave of interest in alternative to incarceration programmes that might help limit the costs of jail and prison and reduce overcrowding, while maintaining public safety. DRCs are nonresidential facilities that provide offenders with supervision and forms of rehabilitative programming (Boyle et al., 2011). A primary goal of DRCs is to reduce recidivism. However, thus far, no systematic review has sought to establish the efficacy of these programs in doing this.


condition, or issue
The day reporting centre (DRC) model emerged in Great Britain in the 1960s, and was first seen in the U.S. a couple of decades later (Boyle, Ragusa, Lanterman, & Marcus, 2011;Craddock, 2000). Its arrival coincided with a wave of interest in alternative to incarceration programmes that might help limit the costs of jail and prison and reduce overcrowding, while maintaining public safety. DRCs are nonresidential facilities that provide offenders with supervision and forms of rehabilitative programming (Boyle et al., 2011). A primary goal of DRCs is to reduce recidivism. However, thus far, no systematic review has sought to establish the efficacy of these programs in doing this.

| The intervention
Offenders participating in DRC programs typically reside at home and report to the DRC on a regular schedule, which can be a couple of times per week or several times per day (Diggs & Pieper, 1994).
They provide a higher level of supervision compared with probation and parole (more frequent contacts between offenders and super-criminogenic environments, where temptations to commit crime are present. This would lead to reductions in offending for the duration of their involvement in the DRC. 3) Longer-term effects on offending through stabilisation:The structure provided by DRCs, over a period of time, may provide the offender a more consistent routine, allowing them to re-focus their priorities and address challenges, with longer-term effects on criminal lifestyles.

4)
Longer-term effects on offending through services and treatment: DRCs tend to incorporate service components, such as substance abuse treatment, cognitive behaviour programs, or employment and training support. These services may have direct effects on criminogenic needs, reducing offenders' propensity to reoffend.
We should also acknowledge two additional effects, the first a possible iatrogenic effect, caused potentially by offenders grouping together at DRCs and providing delinquent reinforcement to one another. Such dynamics may account for short-term iatrogenic effects on recidivism found by Boyle et al.'s (2013) experimental study of DRCs (see also Dishion, McCord, & Poulin, 1999). The second potential effect is one in which increased contact between supervision staff and offenders at DRCs may lead to increased opportunities to detect violations and recidivism.

| Why it is important to do the review
A variety of studies (including those mentioned previously) have sought to assess the effects of DRCs on offending outcomes. While there are few randomised control trials of DRCs, there is at least one: The study by Boyle et al. (2011) of DRCs catering for New Jersey parolees. There are also are a number of quasi-experimental studies that appear to use credible comparison groups along with statistical controls to assure an adequate level of internal validity (Champion, Harvey, & Schanz, 2011;Jones & Lacey, 1999;Ostermann, 2009;Solomon, 2008). For example, Solomon (2008) used a preprogram cohort to compare with defendants sentenced to a "day custody program" in Manhattan, New York, while Craddock (2000) compares probationers in Wisconsin attending a DRC with a similar group who would have been eligible, but did not participate.
However, no scholars have yet attempted to synthesise the disparate findings of these (and other) studies to form generalised conclusions about DRCs. Specifically, there is to date no known systematic review or meta-analysis of DRCs. Our proposed systematic review will help fill this gap by assessing whether DRCs appear, overall, to be effective at reducing recidivism, and-if possiblewhether effects are contingent on variations in program design and population. In doing so, it will focus on a range of offender types, including both adults and juveniles, and program elements, contingent upon the local context of the DRC program (Diggs & Pieper, 1994;Parent, 1990; also see Jones & Lacey, 1999;G. Martin, 2003;Solomon, 2008). Generally speaking, most DRC programs do not service violent or sex offenders. However, participants range through pretrial releases, violators of probation or parole, probationers, work releases, furloughs, those who have finished their prison sentence and are on mandatory supervision, early-releases close to parole eligibility, and those released due to overcrowding emergencies (McDevitt & Miliano, 1992;Parent, 1990).

| OBJECTIVES
We will perform a comprehensive review and synthesis of rigorous research available on (a) the short-term suppression effects of DRC programs on recidivism while participants are attending the DRC, and (b) the longer-term effects on recidivism during and beyond the period of participation, indicative of broader rehabilitative impacts.
The focus of the review is on programs that require offenders to make multiple appearances at a centre each week, and which may also provide rehabilitative programming. The effects of these programs are tested against a counterfactual of offenders not required to attend a DRC, in (a) comparison group(s) of offenders provided that are plausibly similar to DRC participants, or (b) where there are statistical controls that address likely selection biases. We will make separate and combined comparisons between DRC offenders and, respectively, comparison groups under community supervision. The primary outcome measure is recidivism, considered to be rearrest, reconviction, or reincarceration. We are interested in both short-term and long-term effects on recidivism.
The research questions we plan to address are: recidivism. An offender is considered to be participating in a DRC if he or she is required to make appearances at physical premises on multiple (more than one) occasions each week; he or she may also receive rehabilitative programming during these visits.

| Types of study designs
In an attempt to balance internal validity with coverage, we will include randomized controlled trials, rigorous quasi-experimental designs involving matching and premeasures and postmeasures of offending behaviour, quasi-experiments with plausibly similar comparison groups, and quasi-experiments with credible post hoc statistical controls to compensate for differences between groups. We expect, based on our current knowledge, to find several studies meeting these criteria for inclusion in the systematic review.
The comparison group(s) must be comprised of offenders who are under criminal justice supervision in the community, for example, experiencing conventional probation or parole supervision. Moreover, we will code characteristics of these comparison criminal justice sanctions, such as whether probation or parole supervision is intensive, or whether it incorporates a strong therapeutic component. This will allow us to take into account the nature of the comparison groups in our analyses.

| Types of participants
DRCs can serve both juvenile and adult offenders and clients, though we expect most DRCs to serve adult males. While most of these will likely be convicted or adjudicated, pretrial program clients will not be.

| Types of outcome measures
Eligible studies will measure recidivism in terms of new arrests, convictions, and/or incarcerations. Technical violations of probation or parole, for example failing a drug test, will be included as a separate outcome measure, though these are not of primary interest. This is because technical violations may vary simply by as a function of being supervised DRCs-where high levels of surveillance could increase the detection of a violation or new offence-even without differences in underlying offender behaviour.

| Duration of follow-up
We will examine outcomes for a minimum of 1 month for short term outcomes. For longer-term outcomes, we will examine outcomes from at least 6 months. There will be no upper limit to the follow-up duration.

| Types of settings (and timeframes)
Given that DRC programs are relatively new (they emerged in the 1960s; mid-1980s in the United States), we will consider all studies from 1960 onward for inclusion in the systematic review. Studies will not be excluded on the basis of geography. However, we will limit studies to those written in English.

| Search strategy
We will use a number of strategies to conduct an exhaustive search for literature on DRCs. The main literature search will involve keyword searches of online databases and search engines, research organisations, and government databases. We will also review reference lists in reviews of DRCs, and in all studies being considered for inclusion in the review. We will make every effort to locate unpublished material, for example through reaching out to experts in the field. A study's eligibility will be determined by first reading its title and then (if it appears relevant) the abstract, and then reviewing the full text, the latter occurring only if the study appears appropriate.
List of online databases:

| Additional outreach
Authors will contact a total of 30-60 experts in the field, primarily via e-mail. The purpose of doing so is to locate unpublished studies of the effects of DRCs on recidivism. Additionally, the authors will consult with Phyllis Schultze, an information specialist at Rutgers University.

| Search strings and keywords
The keywords below will be used for searching databases and websites are detailed in Table 1. As far as possible, searches will include elements from 1, 2, and 3.
Since different electronic databases accept different approaches, we will create combinations of terms and, where possible, using Boolean operators (e.g., AND and OR), and wildcards (*). This will require some judgement. However, where possible, these terms will be combined in a comprehensive search strategy as follows: (day OR evening OR daily) AND (report* OR program* OR cent* OR custod* OR resource* OR pre-releas* OR prereleas*) AND (effective* OR experiment* OR evaluat* OR recid* OR arrest* OR incarc* OR convict*) We will keep a record of each search, including for the keywords used and their combination, the date the search is performed, the sources consulted. For searches (e.g., through Google) that produce very large numbers of results (e.g., hundreds of thousands), attention will be given to the most relevant results that are delivered higher in the query (results will be read until no further relevant studies are identified). Other studies we are aware of involve quasi-experimental methods with varying degrees of sophistication. Solomon (2008) compared arrests of offenders sentenced to a "day custody program" (DCP; total n = 626) in Manhattan, New York to recidivism among a prior year's equivalent cohort (n = 2,231). To be eligible for DCP offenders must have had at least three prior misdemeanours, were not waiting arraignment for certain cases, have had no history of violent crime, and did not have an active hold for a warrant.

| Description of methods used in primary research
Meanwhile, Ostermann (2009)  Efforts will be made to analyse randomized control trials (RCTs) and quasi-experiments separately to account for the increased risk of selection bias in the latter type studies. However, if only one RCT is available, then studies will be compared according to whether they have RCTs or high quality quasi-experiments on the one hand (involving quasi-experiments with extensive design or statistical controls or recent cohort comparisons), or whether they are some other kind of quasi-experiments on the other.

| Criteria for determination of independent findings
Arrests, reconvictions, and reincarceration outcomes will, in the first instance, be analysed separately. Importantly, these will not necessarily be affected in the same way by DRCs, and may be a product of quite different mechanisms. This is particularly relevant to reincarceration, because this often arises in the absence of criminal behaviour, but following violation of the terms of supervision. We will continue to separate arrests reconvictions and reincarcerations given the variability in decision-making surrounding these events. If multiple studies report effect size data for different outcomes, we will code effect sizes and conduct separate meta-analyses for each of the different outcomes. Similarly, if multiple studies report multiple follow-up periods, we will conduct multiple meta-analyses, stratified by outcome (e.g., arrest, reconviction, and reincarceration) and length of follow up (e.g., 1 month, 3 months, and 6 months). We will also conduct an analysis of all studies, prioritising the most frequently occurring failure measure to maximise statistical power.
Where multiple measures are present in a single study we will choose one only, prioritising arrests because they are likely to be less T A B L E 1 Search terms to be used in systematic review query dependent on downstream court-decision-making processes than convictions, and hence a better reflection of underlying criminal behaviour.
Where multiple studies refer to the same underlying evaluation, this will be treated as a single experiment. In such cases, there may be multiple follow-up periods to choose between. For a given evaluation, we will choose the longest follow-up period for which the original cohort is still largely uncompromised by attrition.

| Details of study coding categories
Coding will be completed by the second and third authors, with the first author overseeing the coding. A full coding scheme can be found in the Appendix.

| Statistical procedures and conventions
If there are at least two eligible studies, we will conduct meta-analysis.
We will include effect sizes based on comparable outcomes in discrete studies (e.g., rearrests and reincarceration) and calculate appropriate weights. Given the dichotomous character of the recidivism events we measure we expect our effect size measures will be log odds ratios (though we will be open to other metrics should they arise).
In experimental studies that do not include statistical controls, and where raw outcomes are reported, we will include log odds ratios based on the counts or positive and negative outcomes across treatment and control groups. However, because our selection strategy of allowing quasi-experimental studies with statistical controls for inequalities in treatment and comparison groups, we will also likely need to include adjusted log odds ratios based on coefficients and standard errors from logistic regression analyses. If other forms of outcome measure are available, we will create appropriate effect size measures (see Lipsey & Wilson, 2001).
We will conduct random effects meta-analysis to calculate the overall effect size, based on the assumption that variation in the observed treatment effect in part reflects real differences in the treatment between daily reporting programs (rather than sampling variability alone). We will present forest plots to visually display the effect sizes in the meta-analysis.
We will also explore heterogeneity, for example by calculating I 2 , and Q statistics. If appropriate (and where there are ideally at least five studies per comparison group), we will conduct a moderator analysis, to take account of program variations. Moderator analysis will use analogue to the analysis of variance (ANOVA; with random effects) to assess associations across individual categorical moderator variables and effect size variance. Moderators of interest include the population served (age, risk levels, and offence types), programing provided (intensity and type of services) and reporting regime (e.g., how frequent contacts tend to be), research design (to distinguish between stronger and weaker experimental designs), and researcher and evaluator independence (i.e., whether or not the researcher is associated with, or independent from, the day reporting centre program). Regarding the design type, if there are insufficient RCTs to make a separate comparison, we will classify studies according to whether they are RCTs or high-quality quasi-experiments on the one hand (involving quasi-experiments with extensive design or statistical controls or recent cohort comparisons), or other quasi-experiments on the other.
We will also create funnel plots to assess for publication bias, and use trim and fill methods (Duval & Tweedie, 2000) to produce adjusted mean effect size estimates (Steichen, 2000) as a robustness check to the main analysis. We plan to use user-written programs and macros available for Stata to conduct all meta-analysis functions (e.g., metan, metatrim, metaf.ado).

| Treatment of qualitative research
Qualitative research studies are not included in the core review. We will, however, include relevant qualitative research in the background of our systematic review.

SOURCES OF SUPPORT
Graduate Research Assistant support will be provided by a graduate student at Fairleigh Dickinson University. The student will receive credit towards her Master's degree completion for her assistance in this review.

PRELIMINARY TIMEFRAME
Approximate date for submission of the systematic review: June 2020.

PLANS FOR UPDATING THE REVIEW
All authors will be responsible for updating the review. Updatescan be expected every 3 years following the completion of the initialsystematic review.  (1 = Nonrandomized; high likelihood of baseline differences between groups or known differences related to future recidivism) (5 = Nonrandomized design with strong evidence of initial equivalence) (7 = Randomized design with large N or small N design with matching) C18. Was attrition discussed in the report? attrep  Gill, Hyatt, and Sherman (2010).