Abstract
Linked article:
1. Background for the review
With rates of imprisonment growing rapidly in many Western industrialized countries (Walmsley, 2005), the effects of parental imprisonment on children is an issue of increasing social concern. In many countries, it is not known how exactly how many children have parents in prison. However, national inmate surveys in the United States show that 1.5 million children had a parent in State or Federal prison in 1999 (2.1 percent of the nation's children under eighteen), over half a million more than in 1991 (Mumola, 2000).
Four key criminological theories suggest that parental imprisonment might cause an increase in child antisocial and criminal behavior (Hagan & Dinovitzer, 1999; Murray & Farrington, in press). First, social bonding theory suggests that parental imprisonment might harm children because parent-child separation disrupts children's attachment relations. Second, strain theory suggests that the loss of family income and other negative life events after parental imprisonment might cause offending behavior. Third, social control theory suggests that parental imprisonment might cause delinquency via reduced quality of care and supervision. And fourth, labeling theory suggests that social stigma and official bias following parental imprisonment may cause an increased probability of being charged and convicted for criminal behavior. These processes of attachment disruption, strain, poor quality child care, and stigma are also associated with adverse mental health outcomes for children (Garber, 2000; Harrington, 2002; Hinshaw & Cicchetti, 2000; Klein & Pine, 2002). Hence parental imprisonment might cause antisocial and criminal behavior and mental health problems for children.
Two alternative theories suggest that parental imprisonment does not cause negative outcomes for children. First, even if children of prisoners have worse outcomes than their peers, this may be because of parental criminality and disadvantage prior to imprisonment, not because parental imprisonment itself is causal. Second, imprisonment of an abusive or antisocial parent might actually decrease children's likelihood of developing behavior problems because it removes a disruptive and antisocial influence from their lives. Existing evidence needs to be synthesized to evaluate these competing hypotheses.
Parental imprisonment is not an intervention as typically studied in Campbell systematic reviews. It is not a deliberately implemented program aimed at reducing or preventing child delinquency. Instead it is a policy for convicted adults that may have unintended consequences for their children. Moreover, it has not been evaluated in randomized experiments, as have other criminal justice interventions. Instead, studies have been observational, using matched control groups and statistical balancing techniques to investigate possible effects on children. We think that, like other risk factors in criminology, parental imprisonment should be studied by investigating what is known about prediction and causation, and then investigating moderators of its effects. Thus, three key questions for research are:
Does parental imprisonment predict negative outcomes for children (is it a risk factor)? Might parental imprisonment cause adverse outcomes for children (might it be a causal risk factor)? What moderates the relationship between parental imprisonment and child outcomes?
The implications for policy are as follows. If parental imprisonment is a predictor of negative life outcomes, children of prisoners may be recognized as a vulnerable population in need of support. If parental imprisonment is a cause of negative child outcomes, a range of sentencing changes and prison and social policies may be implemented to prevent its harmful effects (Murray & Farrington, 2006). By identifying moderators of prison effects, children most at risk of harm by parental imprisonment can be identified for more intensive support. By investigating moderators, resilience processes for children might also be identified. This could suggest specific kinds of support to increase children's resilience after parental imprisonment.
Four existing reviews summarize research on the effects of parental imprisonment on children. Johnston (1995) reviewed 17 studies of children of prisoners and offenders. She reported that the rate of “behavioral problems” among children of prisoners and offenders ranged from 19% to 69% in these studies. However, the association between parental imprisonment and child outcomes was not reviewed, and study quality was not assessed in the review. Also, Johnston did not distinguish between studies of parental imprisonment and studies of parental criminality. Parental antisocial behavior is an important risk factor for children's own delinquency (Lipsey & Derzon, 1998). A key issue is whether parental imprisonment has additional effects on children.
Hagan and Dinovitzer (1999) discussed theories that might link parental imprisonment and adverse child outcomes, and reviewed some of the empirical studies on this topic. They concluded that little is known about the effects of parental imprisonment on children, but speculated that the consequences may be severe, especially when mothers are imprisoned.
Murray (2005) reviewed the effects of parental imprisonment on children reported in 43 empirical studies of diverse methodological quality. He found a range of problem behavior reported in the literature including: depression, anxiety, withdrawal, regression, clinging behavior, sleep problems, eating problems, hyperactivity, aggressive behavior, running away, truancy, poor school grades and delinquency. However, few studies used appropriate comparison groups, well-validated measures, or longitudinal follow-up to draw firm conclusions about the risks associated with parental imprisonment or its possible causal effects.
Murray and Farrington (in press) conducted the most thorough review to date of the effects of parental imprisonment on children. They only included studies with comparison groups and sufficient information to calculate an effect size. They found eleven studies that met their inclusion criteria. They separated findings according to whether studies used (1) control groups selected to be representative of the general population, (2) matched control groups, or (3) samples in which children of prisoners and controls were both recruited from courts or clinics. Based on five studies using general population samples, they found that parental imprisonment was associated with about three times the odds of child antisocial-delinquent behavior and about twice the odds of mental health problems (compared to no parental imprisonment).
The findings were mixed on whether parental imprisonment predicted adverse outcomes independently of background risks, and hence whether it might have a causal effect on children. Five studies used reasonably representative samples of prisoners’ children and controls to estimate the effects of parental imprisonment on children independently of background risks. Three studies found an effect of parental imprisonment on child antisocial behavior independently of background risks (Stanton, 1980; Murray and Farrington, 2005; Huebner and Gustafson, 2007), but two did not find an independent effect (Bor, McGee, and Fagan, 2004; Murray, Janson, and Farrington, 2007). Only one study examined the effects of parental imprisonment on child mental health using suitable controls (Murray and Farrington 2008). In this study, the association between parental imprisonment and poor mental health outcomes remained after controlling for parental criminality and child, family and socioeconomic risks. Murray and Farrington (in press) speculated that parental imprisonment might cause child antisocial behavior and mental health problems, but argued that firm causal conclusions could not be drawn from existing evidence.
Murray and Farrington (in press) also reviewed possible moderators of the effects of parental imprisonment on children, but they found very few studies that analyzed how the effects of parental imprisonment differed according to child, family or social factors. Theory and existing evidence suggested that parental imprisonment might have more harmful effects for girls, children of ethnic minority, younger children, children experiencing maternal imprisonment (as opposed to paternal imprisonment), and children whose parents are imprisoned for longer periods or in more punitive social contexts.
We are very clear that tight causal conclusions cannot be drawn from observational studies that are typical of this area of research. However, a distinction can be drawn between observational studies that simply assess the zero-order association between parental imprisonment and child outcomes, and studies that control for important confounding factors. Studies that control for important confounding factors provide some evidence for considering possible causal effects. While conclusions about causal effects must be very tentative based on such observational evidence, it is important to extract and summarize the best evidence available.
In summary, according to existing reviews, most existing studies of parental imprisonment are mainly small-scale and of poor methodological quality, with low internal validity, and often lacking a comparison group. An important issue is whether it is worth conducting a systematic review when it appears there are few high quality studies. We think it is worthwhile for the following reasons. First, a systematic review may uncover high quality studies that were not found using less thorough searching methods. Second, more high quality studies may take a long time to appear, and policy-makers need interim evidence on which to consider their decisions. Third, if a systematic review demonstrates that high quality studies are lacking, this could encourage a new generation of primary research. Hence, even though it seems there are few high quality studies of parental imprisonment, it is still worth conducting the proposed review.
2. Objectives of the review
The two main aims for the review are to assess evidence on whether parental imprisonment is a risk factor for negative child outcomes, and to assess evidence on the causal effects of parental imprisonment on children. A third aim of the review is to investigate how associations between parental imprisonment and child outcomes differ according to study features and participant characteristics (moderators).
Whether or not parental imprisonment is a risk factor for child outcomes will be investigated by aggregating zero-order associations from studies comparing children of prisoners and children from the general population. The possible causal effects of parental imprisonment will be investigated by aggregating findings from studies using matched control group designs, or statistical controls for critical covariates.
The main moderators we hope to investigate are: maternal versus paternal imprisonment, length of parental imprisonment, child sex, child age at parental imprisonment, short versus long term outcomes, the type of delinquent outcome (juvenile antisocial behavior, adult antisocial behavior, juvenile crime, adult crime), the type of mental health outcome (anxiety, depression, or general internalizing problems), and country of study. We will also analyze methodological factors as possible moderators (such as type of study design, study methodological quality and publication type).
Based on the studies we have collected so far, further moderator analyses will not be possible. However, if enough studies are retrieved with results on other variables that might increase child resilience after parental imprisonment, we will include them in our analyses. For example, moderators such as social support, quality of parenting, family income, and frequency of child-parent contact during imprisonment could be investigated.
The proposed systematic review will extend the narrative review of Murray and Farrington (in press) on these topics in three main ways. First, the systematic review will use a more thorough search strategy to locate all relevant studies. Second, studies will be more systematically evaluated regarding their methodological quality. Third, meta-analysis will be used to synthesize the findings across the studies and to investigate moderators.
In the proposed review, parental imprisonment is defined as any kind of custodial confinement of a parent, apart from being held in police cells. For example, imprisonment can refer to confinement in jails or prisons (state or federal) in the United States, and open or closed prisons (local or training) in the United Kingdom.
The three child outcomes that will be investigated are antisocial behavior, criminal behavior and mental health problems. In this review, antisocial behavior is a broad term referring to behavior that violates social norms and the rights of others but does not necessarily break the law. Criminal behavior refers to behavior that violates the penal code and could result in criminal conviction. Mental health problems refer to depression, anxiety, and general internalizing problems.
3. Methodology
3.1 Criteria for inclusion of studies in the review
Studies meeting all the following six criteria will be included in the review. (Note that additional criteria will be used to select studies for sub-analyses of particular review questions; see section 3.6.)
The study must include children of prisoners (“experimental group”) and at least one group of children without imprisoned parents (a “control group”). In the experimental group, parental imprisonment must occur between the child's birth and eighteenth birthday. The study must include a measure of children's antisocial behavior, crime or mental health. Parental imprisonment must have occurred before the child outcome. Studies will be excluded if the time-ordering of parental imprisonment and the child outcome is ambiguous, for example if the child outcome measure refers to the period before parental imprisonment. The study must use the same measure of child outcome for the experimental group and the control group. Numerical information: Effect sizes must be reported, or there must be enough numerical information to calculate effect sizes. Studies that do not meet this eligibility criterion but report the direction of findings (whether or not parental imprisonment was deemed to be associated with worse child outcomes) will be described in a separate table.
Publication. Both published and unpublished studies will be included.
Country of origin. Studies may be conducted in any country and may be reported in English, German, Dutch, French, Spanish, Swedish, Danish, or Norwegian. We are not aware of studies reported in other languages, and anticipate that colleagues will help interpret results reported in these languages for the review.
Timeframe. Studies may be published between 1960 and 2008. Existing reviews suggest that most studies of children of prisoners were conducted since 1980. A few studies were conducted in the 1970's (e.g., Moerk, 1973, and Sack, Seidler & Thomas, 1976) and we are aware of only one study that was conducted in the 1960's (Friedman & Esselstyn, 1965). However, these early studies were nearly all exploratory in nature, and would not meet eligibility requirements for this review. Therefore, we will set 1960 as the lower limit for our searches.
3.2 Search strategies
Initial Library. An initial set of 150 full reports about children of prisoners were retrieved for the Murray and Farrington (in press) review. These will be used as an initial library for the proposed systematic review.
Electronic databases will be searched for the years 1960 to 2008. We will search the following electronic databases using the following keywords:
Prison*/ Jail*/ Penitentiary/ Imprison*/ Incarcerat*/ Detention
AND
Child*/ Son*/ Daughter*/ Parent*/ Mother*/ Father*
AND
Antisocial*/ Delinquen*/ Crim*/ Offend*/ Violen*/ Aggressi*/ Mental health/ Mental Illness/ Internaliz*/ Depress*/ Anxiety/ Anxious
List of databases
Australian Criminology Database (CINCH) Australian Institute of Criminology Library (J V Barry Library) Bibliography of Nordic Criminology Cochrane Controlled Trials Register C2-SPECTR Criminal Justice Abstracts Dissertation Abstracts Education-line Embase ERIC Google (with extra search terms: research/ evaluation/ study/ pdf) Google Scholar Ingenta JSTOR MEDLINE Newton (University of Cambridge Library book search) National Criminal Justice Reference Service (NCJRS) National Institute of Corrections Information Center (NICIC) Prison Health Research Network PsychInfo/Psychlit Sage Family Studies Abstracts Sociological Abstracts Social Sciences Citation Index (SSCI) - Web of Science System for Information on Grey Literature in Europe (SIGLE)
We will search reference lists of review articles and primary studies.
To identify other studies, we will contact international colleagues in England, the United States, Canada, Australia, New Zealand, Germany, France, Norway, Sweden, Denmark, and the Netherlands.
Large-scale longitudinal studies might include results relevant to this review but not refer to parental imprisonment in report titles or abstracts. We will contact directors of leading longitudinal studies in child development to inquire whether they have existing results on parental imprisonment that we have not found.
3.3 Description of methods used in the primary research
Most studies compare children of prisoners and a control group either during parental imprisonment or shortly after it (e.g., Stanton, 1980; Trice & Brewster, 2004). Only a few studies have examined outcomes in adulthood (e.g., Huebner & Gustafson, 2007; Kandel et al., 1988; Murray & Farrington, 2005; Murray, Janson, & Farrington, 2007). Although a few randomized experiments have been conducted on the effects of imprisonment on prisoners (Killias & Villetaz, 2008), according to existing reviews, no study has used randomized assignment to investigate effects of parental imprisonment on children.
Some studies compare children of prisoners with children of the same age selected from the general population (e.g., Bor, McGee, & Fagan, 2004; Huebner & Gustafson, 2007). These studies provide estimates of the zero-order risk associated with parental imprisonment for children. Other studies use matched control groups to try to test the causal effects of parental imprisonment on children. For example, Stanton (1980) compared children of imprisoned mothers and children of mothers on probation to control for background maternal criminality and social disadvantage (although the two groups were not equivalent on these dimensions). In several studies, statistical control techniques were used to try to disentangle the effects of parental imprisonment from background disadvantage (e.g., Bor et al., 2004; Huebner & Gustafson, 2007; Murray & Farrington, 2005; Murray et al., 2007).
In the absence of randomized experiments, ideally studies should compare children of prisoners and controls on change in outcomes from before to after parental imprisonment, while balancing the two groups on critical covariates. We are aware of only two ongoing studies of this type, and hope to find more in our searches. Assuming that the studies we locate in our searches have similar designs to those described above, only very speculative conclusions about causal effects can be drawn from existing research.
3.4 Criteria for determination of independent findings
Some studies report multiple findings based on a single sample. For example, a study might report on more than one child outcome (e.g., both crime and mental health) or might use multiple measures for a single outcome (e.g., official convictions and self-reported criminal behavior). All findings will be coded and entered into the data file. Findings from the same study may sometimes be included in more than one meta-analysis. For example, from one study, findings on antisocial behavior, crime and mental health may be included in separate meta-analyses on these topics. However, because effect sizes should be independent within a meta-analysis, only one measure will be included from each sample in each meta-analysis. Within each meta-analysis, one measure will be selected from each study, using the following criteria:
Measures should be tested for reliability and validity and have good psychometric properties (as defined on the Cambridge Quality Checklist; see Appendix B, and Murray, Farrington, and Eisner, forthcoming). Short and long term outcomes will be analyzed separately. For meta-analyses of short-term outcomes (assessed up to 2 years after initial imprisonment), the first measure of child adjustment will be selected. For long-term outcomes (assessed at least 2 years after imprisonment), the last measure will be selected (so long as attrition since the previous measure is not over 10%). Any remaining multiple effect sizes (e.g., multiple measures taken at the same point in time) will be averaged and the average will serve as the effect size for the meta-analysis.
3.5 Details of study coding categories
We will code the following topics at least:
Author(s) Date of research Date of publication(s) Country and location of research Age of children (Experimental, Control) Gender composition (E, C) Ethnic composition (E, C) Initial sample size (E, C) Final sample size (E, C) Response rate/ Attrition rate Maternal/ Paternal imprisonment Length of parental imprisonment Control condition (e.g., children with parents on probation) Follow-up time period Measures of each outcome (caregiver ratings, self-report questionnaires, teacher ratings, school records, criminal records, psychiatric records) Methodological quality Effect size data
A draft coding schedule is shown in Appendix D. At least 10 studies will be coded by two persons in order to measure reliability and to agree coding criteria.
3.6 Statistical procedures and conventions
We will pool study results in meta-analyses. Nearly all the studies we have collected so far reported outcomes as proportions or percentages (see Appendix A). The measure of effect size to be used in analyses is the log odds ratio (LOR), calculated by comparing the percentage of cases among children of prisoners and controls, and taking the natural logarithm of the odds ratio. However, final results will be converted back to odds ratios for ease of interpretation.
Where scores and standard deviations are reported, the standardized mean difference (d) will be calculated and then converted into LOR using the equation: LOR = d / .5513 (see Lipsey & Wilson, 2001, p. 202). Where correlations (r) are reported, d will be estimated (see Lipsey & Wilson, 2001, p. 199), and then d will be converted into LOR. If marginal cell proportions are reported as well as r, the LOR will be calculated directly from r (see Lipsey & Wilson, p. 197). Where LOR can be computed from other statistics, this will be done using the conversion formulae in Lipsey and Wilson (2001, Appendix B).
Analyses will proceed in three stages, as described below. Each set of analyses will be repeated for the three child outcomes: antisocial behavior, crime and mental health.
The first set of analyses will investigate if parental imprisonment is a risk factor for child outcomes. To do this, we will pool all zero-order effect sizes comparing children of prisoners and children from the general population. Weighted mean effect size measures and their confidence intervals will be calculated using the procedures described in Lipsey and Wilson (2001). The key hypothesis to be tested is whether the weighted mean LOR is significantly different to zero. We will investigate heterogeneity in effect sizes using the Q statistic. We will fit a random effects model if the Q statistic reveals heterogeneity, or if the number of effect sizes (or typical number of study participants) is small. Otherwise we will fit a fixed effects model. The second set of analyses will investigate whether parental imprisonment might be a causal risk factor for child outcomes. We will do this by separately pooling effect sizes that are adjusted for critical covariates of the outcome (critical covariates are listed in Appendix C). Three types of study will be included in these analyses:
Studies that compare children of prisoners and a control group matched on critical covariates Studies that compare children of prisoners and a control group matched on propensity scores (where propensity scores are calculated based on critical covariates) Studies that use regression analyses or analyses of covariance to control for critical covariates From these studies, weighted mean effect size measures (adjusted for covariates) will be calculated using the procedures described in Lipsey and Wilson (2001). Covariate-adjusted means and unstandardized B's from OLS regression will be converted to d using the unadjusted pooled standard deviation (see Lipsey & Wilson, p. 180), and then d will be converted to LOR. The key hypothesis to be tested is whether the weighted mean LOR (adjusted for critical covariates) is significantly different to zero. We are aware that aggregating effect sizes based on studies that controlled for different covariates is problematic, and we aim to treat the covariates as a moderator variable (in separate analyses), providing that we identify enough studies. We will also investigate possible causal effects of parental imprisonment by separately analyzing studies that control for pretest scores of child outcomes (gain scores). The standardized mean difference in change from before to after parental imprisonment (d) will be calculated using the pooled standard deviation at posttest, and then d will be converted into a LOR. It will be tested whether the mean LOR (adjusted for pretest child outcome measures) is significantly different to zero. As in the first set of analyses, if the number of effect sizes is small, or effect sizes are based on very few study participants, or the Q statistic is significant, we will fit a random effects model. Otherwise we will fit a fixed effects model. If there are many covariate-adjusted study findings that cannot be converted to the common metric of LOR, we will list the direction of findings reported in these studies (whether parental imprisonment predicted adverse child outcomes after controlling for critical covariates or pretest scores). In a third set of analyses, we will investigate possible moderators of the relationship between parental imprisonment and child outcomes. Analyses of moderators will be repeated for both zero-order effect sizes and covariate-adjusted effect sizes. We will investigate associations between effect sizes and study design features (e.g., the precise technique used to control for covariates), publication type (whether or not the study was published), methodological quality (as coded on the Cambridge Quality Checklist; see Appendix B), and other characteristics of studies (e.g., the age and sex of participants, whether the father or mother was imprisoned, operational definitions, measurement of key concepts, and the covariates that were controlled). Then we will carry out meta-regression analyses (Lipsey & Wilson, 2001) to investigate independent influences on effect sizes. Effect sizes will be coded separately for distinct sub-groups within studies (e.g., for boys and girls) to conduct these analyses.
All eligible studies will be coded on paper, and then entered into SPSS. SPSS macros developed by David Wilson (Lipsey & Wilson, 2001) will be used to perform the meta-analyses.
3.7 Treatment of qualitative research
The review will be focused on quantitative studies but will include qualitative information (e.g., on prison or social context) when this is helpful in discussing explanations for findings or conflicting results.
4. Time frame
A substantial amount of work has already been done in searching and retrieving studies of parental imprisonment. Over 150 full-text articles have already been retrieved. We still need to conduct a more comprehensive search for studies, code the studies, run new analyses, and write a report conforming to Campbell Collaborative guidelines. We aim to submit the first draft of the report to the Campbell Collaboration within six months of the protocol being approved.
5. Plans for updating the review
We plan to update the review every three years. The lead reviewer will take responsibility for arranging this.
Footnotes
6. Acknowledgements
We anticipate help conducting the searches from Ivana Sekol, PhD researcher, Cambridge University, and Anne-Marie Klint Jørgensen, NC2 Librarian. Liz Ayre, Eurochips, will help with translating results from studies reported in French. Jukka Savolainen, Visiting Associate Professor, University of Minnesota, and Senior Research Analyst, National Research Institute of Legal Policy, Finland, may also join the review research team.
7. Statement concerning conflict of interest
The only possible (minor) conflict of interest is that we have conducted two studies of the effects of parental imprisonment on children, and we have argued in a recent review that parental imprisonment is associated with adverse outcomes for children (Murray & Farrington, in press). The two studies that we conducted had different findings regarding the causal effects of parental imprisonment on children: one suggested parental imprisonment might have causal effects (Murray & Farrington, 2005, 2008); the other suggested parental imprisonment had no causal effects (Murray et al., 2007). Hence we are not tied by these results to either conclusion. Also, neither of us stands to benefit in any way from any source that has any interest in the conclusions of the review. We are not aware of any personal, political, academic, or financial conflicts that might influence our judgment.