parental arrest, incarceration and the outcomes of their ... · ˙nd that only 42% of incarcerated...

Parental Arrest, Incarcerationand the Outcomes of Their Children

Stephen B. Billings∗

November 6, 2017

Abstract

Parental arrest and incarceration represents a profound and traumatic experience foralmost 3 million children in the U.S. and scholars in sociology and criminology consis-tently �nd negative impacts of parental incarceration on children across a range ofacademic and behavioral outcomes. Unfortunately, the challenge of disentangling parentalincarceration from other parenting attributes has limited causal inference in this liter-ature. The research presented here provides compelling evidence that parental arrestcoincides with negative outcomes for children, but that the incarceration of a parentmay bene�t the child. Results suggest that incarceration removes negative role modelsand leads to changes in a child’s home environment.

∗[email protected], Leeds School of Business, University of ColoradoAcknowledgements: I thank Eric Chyn, Jennifer Doleac, Rick Mans�eld, Terra McKinnish, Erik Plug

and Je�rey Zax as well as participants at the University of Virginia Public Economics Workshop, 2017University of Lund-SFI Labor/Health Economics workshop. I would also like to thank Julia Rush of theMecklenburg County Sheri�’s Department; Andy Baxter from Charlotte-Mecklenburg Schools.

1

Preliminary Draft : November 6, 2017

1. Introduction

The large number of adults incarcerated in the U.S. and the age pro�le of criminalsentails that a number of young and school age children have the traumatic experienceof a parental �gure being removed from the household through the criminal justicesystem. In fact, Western and Pettit (2010) estimate about 2.7 million children have anincarcerated parent which represents about 1 in every 28 children. Given the prevalenceof this childhood experience, a rather robust literature in sociological and criminologystudies the impacts of parental incarceration on children and consistently highlights alarge negative e�ect of parental incarceration on the life course outcomes of children(Wildeman et al. (2014) ; Wildeman et al. (2014) ; Foster and Hagan (2007); Wildeman andWestern (2010) ; Besemer et al. (2011)). Scholars �nd evidence that parental incarcerationis linked to negative academic outcomes (Cho (2009) Cho (2010) Foster and Hagan(2007) ) , greater child mortality ( Wildeman et al. (2014) ) and more antisocial behavior((Murray and Farrington 2005), Murray et al. (2012) Wildeman (2010)) among children.This research attributes these negative outcomes to factors such as reduced social capital,changes in a child’s living arrangements, �nancial loss, trauma of actual arrest andincarceration and change/loss of relationship with a parent.

Unfortunately and often recognized in this literature, an overarching identi�cation issueis that criminal parents are likely to have non-criminal negative impacts on a child’s homeenvironment and is it not clear how much direct involvement with the criminal justicesystem matters. Recent work (Dobbie et al. (2017)) using swedish administrative dataattempts to address this shortcoming by incorporating variation in parental incarcerationdue to random judge assignment.2 Dobbie et al. (2017) �nd negative e�ects of parentalincarceration of a parent’s own labor market outcomes and no e�ects on educationalattainment for children, but do �nd increased rates of teen childbearing and criminalconvictions for children of incarcerated parents. Even though these negative impactscon�rm some of the existing literature, they still contrast Bhuller et al. (2016) who userandom judge variation and �nd bene�ts of adult incarceration on own outcomes forrecidivism and labor market outcomes in neighboring Norway.3 This recent research doeshighlight the potential for long-term e�ects on children due to parental incarceration,but also heterogeniety across criminal justice systems. Additionally, existing literatureonly focuses on birth parents and does not account for the large number of single mother

2In the paper I am presenting here, the use of random judge assignment is not possible given the lack ofdata on court records.

3The authors explain their positive impacts in Norway as a function of the job training and other bene�cialprograms provided to inmates in the Norweigan prisons.

2


households with absentee birth fathers in this population of families.4

The research presented here provides two contributions to this literature. First, I disen-tangle the e�ects of parental arrest from that of parental incarceration thereby providingdirect evidence of the impacts of parental incarceration on children in the US. I primarilyfocus on short-term e�ects, but provide evidence of impacts on younger children thatextend into early adulthood.5 Second, I focus on parents living with the children at thetime of their arrests which overcomes the problem of absentee birth fathers muting anyrelationship between parental incarceration and children’s outcomes.

In order to quantify the impacts of parental arrest and incarceration on their children, Imatch rich administrative data on children, adult arrests and incarceration to provide astudent level panel of outcomes (e.g absences, school crimes, suspensions, test scoresand grade retention) through public school and directly link the timing of arrest andincarceration to annual outcomes. My data incorporates all incarcerations of individuals16 and older in Mecklenburg county (city of Charlotte, NC) jails and North Carolina stateprisons and thus includes both pretrial incarceration as well as post-trial sentencing.6

Matching of parents to children is done based on names and addresses to ensure thatparental �gures live in the same residence as children.7

In order to identify the e�ects of parental incarceration, I use the nature of arrests andincarcerations in the US criminal justice system. The typical experience of a criminalis to be arrested and then either released with a citation or summons for court. Formore serious crimes, arrestees are often arraigned in court and released upon paymentof bond or serve time until a scheduled court appearance. The more serious the crimeand more limited an individual’s �nancial resources, the more likely an individual willspend time in jail/prison and that time in jail/prison will last longer. Since I only havenon-experimental variation in parental incarceration, I incorporate student �xed e�ectsto control for average parental attributes for a child and limit results to temporal variationin parental arrests and incarceration.

4Directly determining if parents are living and regularly interacting with children is important sincethe incarceration of an absentee father or mother would have substantially di�erent impacts than aparent living with the child. This point is particularly salient given that Glaze and Maruschak (2008)�nd that only 42% of incarcerated fathers and 60% of incarcerated mothers live with children prior toincarceration.

5Short term refers to e�ects on children’s outcomes that occur within the same year or in some modelswithin 1-3 years of parental incarceration.

6I cannot directly disentangle pretrial from post-sentencing incarcerations since this data does not includecriminal court records or speci�c sentences.

7This type of matching does allow the inclusion of other relatives that live with the child such as unclesor aunts. I therefore use the de�nition of parent more broadly to include any relative providing carefor the child. I do restrict the designation of parent to the age pro�les that would allow the individualto be old enough to be a biological parent.

3


Given concerns that the timing of incarceration is endogenous to other changes inparenting behavior, I bound my estimates by drawing on the qualitative criminal justiceliterature where the incarceration of a parent typically follows traumatic and othernegative changes in a child’s household environment.8 Under this assumption, the timingof more serious criminal activity will lead to parental incarceration and also coincide withlarger non-criminal negative impacts on a child’s home environment. My estimates willbe conservative as unobserved parental attributes such as substance addiction, limited�nancial resources or violence in the home are likely to be positively correlated withcriminal arrest and incarceration which would bias any estimates toward �nding negativeimpacts of parental incarceration on children’s outcomes. Therefore, any positive impactsof parental incarceration is likely due to positive in�uences of parental incarceration onthe child such as the removal of a negative role model or allowing the other parental�gure to end or even escape an abusive or negative relationship. I provide two compellingfacts that reinforce this empirical strategy. First, I show that in aggregate, children ofparents with more arrests and more incarcerations have worse outcomes. Second, Ishow that consistent with the removal of negative bias due to omitted variables, theincremental inclusion of control variables and �xed e�ects for family environment shiftthe e�ects of parental incarceration from negative to positive.

Results highlight a positive and signi�cant bene�t from parental incarceration on behav-ioral outcomes with parental incarceration generating an impact of 0.07 standard devia-tions of an annual summary index of school behavior. Smaller positive results of 0.02standard deviations also hold for test scores, but not grade retention. Even though, Icannot provide causal estimates that the act of arresting a parent impacts children,estimates for the impact of parental arrest do provide support for the joint negativeimpact of home environment and parental arrest on the educational and behavioraloutcomes of children. In terms of longer-term impacts of parental incarceration, I �nda decrease in the probability of dropping out of high school, but limited impacts onadult crimes for teenagers with incarcerated parents. Estimated bene�ts are largest formaternal incarceration and for younger-aged children, especially in cases where themothers is a violent o�ender and the father is alcohol/drug o�ender. Results are robustto varying assumptions on the matching of parents to children, the use of subsets ofincarcerations that are only a few days or more than a month, as well limiting arrestedand incarcerations to only parents that can be veri�ed using birth records.

Even though my main models focus on short-term impacts, the fact that I �nd longer-term incarcerations or parents that are multiple o�enders generate larger behavioral

8See Braman (2004) , Turanovic et al. (2012) and Parke and Clarke-Stewart (2003) for examples orsummaries of this qualitative literature

4


e�ects speaks to potentially larger aggregate bene�ts from removing the criminal parentfrom the home. Results are also similar across type of crimes committed by parents,as well as gender of the child. I show that one response to parental incarceration isto leave the school district for the next academic year.9 I also provide evidence thatincarcerated parents are more likely to live at a di�erent address if rearrested afterrelease from incarceration. Overall, results suggest that parents involved in the criminaljustice system may provide negative role models for children’s behavior, but e�ectsare more limited for educational outcomes. Therefore, the removal of parents, that maydemonstrate or make acceptable antisocial behavior, may bene�t a high risk population ofstudents - namely children with incarcerated parents. The presence of mostly short-termbene�ts from parental incarceration could simply be a result of new household formationinitiated by the incarceration of a parent which may simply replace an incarceratedparent within another similar type of parent.

Results contribute to a broader economics literature that provides evidence that parentalenvrionment can have profound e�ects on the outcomes of children. Gertler et al. (2004)�nd that children whose parents died were twice as likely to drop out of school inIndonesia. Shocks to maternal health also generate behavorial issues among children(Mühlenweg et al. (2016) ) and poor parental health helps explain black and whiteacademic achievement gaps. Balsa (2008) shows that parental alcoholism is associatedwith worse adult labor market outcomes for children while Snow Jones et al. (1999) andChatterji and Markowitz (2001) link parental alcohol-use to worse behavioral outcomes.Doyle Jr et al. (2007) and Doyle Jr (2008) highlight negative impacts on children’s adultlabor market outcomes from foster care placement for children on the margin of beingremoved from their parent’s home. Of this literature, military deployment papers are themost related to the research presented here as they highlight the e�ect of removing aparental �gure from the household. Angrist and Johnson IV (2000) �nd parental absenceimpacts divorce rates and spousal employment and results di�er between male andfemale soldiers. Speci�cally for children, Lyle (2006) �nds a negative e�ect of about 1/10of a standard deviation in test scores for the school year of parental absence.

More consistent with the presence of family stress and negative role models, there isa more sizeable literature on the relationship between parental job loss and children’soutcomes. Parental job loss impacts children negatively through higher likelihood ofdivorce ( Doiron and Mendolia (2012) ), lower ability to pay for or attend college (Hilger

9One may be concerned about families that move during the year and thus limit one’s ability to observebehavioral outcomes. I show that incarceration does not impact if a student takes end-of-grade examswhich implies that incarcerations do not impact if a student is currently enrolled in the last few weeksof school.

5


(2016) ) , short-term academic di�culties (Ananat et al. (2011) ; Stevens and Schaller(2011)), higher rates physical injuries and mental disorders (Schaller and Zerpa (2015))and even intergenerational e�ects with children of displaced workers earning less asadults (Page et al. (2007) ; Oreopoulos et al. (2008)). Speci�cally related to parentalenvironment, Rege et al. (2011) �nd that parental job loss negatively impacts children’school performance and this e�ect is most consistent with mental distress, not incomeloss surrounding job displacement.

Consistent with this literature on the e�ects of parental environment on children, I �ndremoving negative potential role models through incarceration bene�ts children. Giventhese �ndings, policies that focus on removing negative or even promoting positive adultrole models for children may have more immediate bene�ts. These policies could rangefrom social service assistance that helps mothers move their children away from domesticabuse by a father to assisting fathers or mothers in ending parental relationships that areunhealthy to providing in-patient drug and alcohol addiction programs for parents. Thispaper is organized as follows. Section 2 highlights the administrative data incorporatedinto this study as well as details about the population of criminal parents. Section 3presents a simple empirical model of the impacts of both parental arrest and incarcerationon academic and behavioral outcomes for children. Section 4 provides the main resultsas well as a number of speci�cations that further explore heterogeneous results withrespect to both parental and children’s attributes. Section 5 concludes.

2. Data

In order to examine the relationship between parental incarceration and children’soutcomes one needs a panel dataset that links children to adult criminals. One also needsto be able to determine if a given adult criminal is present in the life of a given child. Tobegin, I incorporate four administrative databases from Charlotte-Mecklenberg County,NC which include detailed pupil records from 1999-2011; detailed arrest records from1998-2011; Mecklenburg County jail records from 1998-2011 and North Carolina StatePrison Records from 1998-2011. I limit my sample of children to those born between 1990and 2001 to allow all individuals to be observed for at least two years in public schoolrecords that overlap with arrest and incarceration records. This data is comprehensivewith student demographics on race, gender and home address, yearly end-of-grade(EOG) test scores for grades 3 through 8 in math and reading10, number of days absent,

10Test scores are standardized at the state level by grade and year.

6


days suspended from school, and the number of incidents of school crime.11 The threedatabases from the criminal justice system allow one to determine the timing of arrests,demographic information about the criminal including full name and date of birth, thehome address of the arrestee, criminal charges as well as all subsequent jail and prisonperiods of incarceration.12

The key for my analysis is to be able to determine the timing of arrests and incarcerationsfor adult criminals that are the parents of school-age children and highlight academicand behavioral outcomes for those children. The �rst main challenge is determiningthe presence of a criminal parent in the student database. Initially, one could use birthrecords to link children to parents and then link parents to arrest and incarcerationsrecords using full names and date-of-birth. My students record are uniquely identi�ed interms of full name and date-of-birth.13 The problem with this strategy is the large numberof birth records that contain missing information on fathers as well as the overall lowmatch rate between birth records and the student database. In order to later explore thistype of matching, I incorporate a database of all birth records from 1990-2001 in NorthCarolina with individual identifying information and match children’s full name anddate-of-birth to the student records. For my main estimation sample, I match about 66%of student records to birth records.14 Even with a better match rate, missing informationon fathers is problematic given the large number of male criminals as well as the factthat a birth father could be absent and thus not even live with their children whichwould weaken my ability to estimate a relationship between father’s incarceration andchildren’s outcomes.

Therefore, I turn to an alternative strategy of last name and address matching. Theadvantage of this method of matching is twofold. First, it leverages the detailed andwell populated address information available in the student and criminal justice records.Second, it allows this analysis to focus on parental �gures that live with the child thuseliminating concerns of absentee fathers. Of course, I cannot directly verify parents, soin some cases this matching may capture other relatives with the same last name livingin the home with the child. This is unlikely to be problematic for capturing parental11According to NC State Statute 115C – 288(g), any incident at school involving any violent or threats

of violent behavior, property damage, theft or drug possession must o�cially be reported to the NCschool crimes division. This statute ensures that this measure of school crime is consistently reportedacross schools and cannot be treated di�erently based on school administrators.

12Approximately 16% of arrestees have non-geocodable or missing residential address information.13I drop the approximately 50 students out of 194,163 that are not uniquely identi�ed by full name and

date-of-birth14This relatively low match rate is likely due to the time period between birth and public school, outmigra-

tion to neighboring South Carolina as well as some missing information in the birth records database.This is about 20 percentage points below Figlio et al. (2016)’s study in Florida but their data wasadministratively matched using social security numbers.

7


environment since these other relatives also provide "parental" guidance to the child.There are some issues with using address and last name to match students. One concernis the uniqueness of last name and address in a given year. This potential problem ismost likely for children with common last names living in large apartment complexes.15

In cases where student records match arrested individuals that have the same address,gender and last name but di�erent �rst names or date-of-births, I drop the studentobservation. I also drop any student living in a larger apartment complex (> 5 units).16

Approximately 5% of arrests are duplicated in terms of names and addresses and 15%of arrests are linked to students living in large apartment complexes (> 5 units). I laterprovide results where I keep all observations irrespective of duplicates or living in largeapartment complexes and results are quite similar.

Appendix Table A1 provides an example of the structure of the data and what potentialmatches look like. In this example, I observe a student John Williams for �ve years inour dataset and he lives in the same residence the entire time. In the arrest records, Ihave arrests of people with the same last name and in some cases the same address. Forall matching, I require that same last name and addresses must match between studentrecords and an arrestee’s home address at the time of booking. Incarceration recordsare easily linked to arrest records based on arrestee information and provide the timingof subsequent incarcerations. These records do not provide the exact incarceration fora given arrest, but do give the exact dates of arrest and incarceration spells. I do haveto make some assumptions on assigning a speci�c arrest to an incarceration and do sobased on the incarceration occurring closest in time immediately following a given arrestbut no more than 1 year after the arrest. Most cases involve an arrest that is immediatelyfollowed by an incarceration. For all arrests and incarcerations I aggregate to a givenschool year and separate school years based on July 1st.17

Two elements of my analysis limit concerns about overall match quality. First, identi�ca-tion is based o� of changes in parental arrest and incarcerations over time conditional onmatching an arrestee to a child. Therefore, missing addressing information for arresteeswould only serve to change the generalizeability of results to those parents with estab-lished residences and no clerical errors in criminal justice records. Additionally, there may

15Information on apartment numbers is not provided in the student records since addresses are simplyuses for school assignment in this database. Mailing address and contact information for students isnot made available to outside researchers.

16Apartments with more than 5 units tends to increase the portion of students with arrested parentsand makes one concerned about false positive matches for same last name families living in di�erentapartments.

17I use July 1st as our cuto� since I want to make sure to link all arrests and incarceration that occurjust prior and during a school year to that school year’s outcomes. Other cuto�s during the summerprovide minimal di�erence in our results.

8


be concerns about matching across individuals that move to/from apartment complexesbut since results that include/excluding individuals in large apartment complexes arequite similar this concern does impact later results. Second, I perform a back of theenvelope calculation to get a sense of the portion of likely criminal parents I am able tomatch to the student database. I estimate that between 55% and 75% of the estimatedpopulation of criminal parents are matched to student records.18

Since my analysis focuses on parsing out the e�ects of incarceration while controllingfor arrest, one would like to know the typical criminal justice experience in our dataset.When an individual is arrested they are typical booked and then depending on thenature of the o�ense, they may be released within a few hours and thus never enter thejail/prison system or held for longer periods in a county jail facility or transferred into astate prison. My data captures all stints of incarceration in jail or prison and treats almostall transfers from county jail to state prison as a single incarceration lasting the entirespan of jail and prison. For more serious crimes, arrestees are often arraigned in court andreleased upon payment of bond or serve time until a scheduled court appearance.19 Figure1 provides the breakdown of parental incarceration spells in my main dataset whichexcludes incarcerations of 2 days or less. The most frequent spell of incarceration is 1-3months with the second largest group being incarcerated for more than 2 years. About1/3 of parental incarcerations are for less than 30 days, with a number of incarcerationspells lasting 1-3 months or more than 2 years.

Even though my dataset is mostly populated by shorter term incarcerations, there maybe potential bene�ts as the removal of a parent allows for shifts in residence and changesin parental relationships as well as other public interventions. Furthermore, a releasedindividual may have been reformed while incarcerated or is under parole/other supervi-sion that instills more positive parenting behavior. The use of shorter term incarcerationsalso limits potential negative bias from more serious criminals that likely have worsenon-criminal parenting behavior. The main results will not include parents incarceratedless than 3 days as incarcerated and thus all children with an incarcerated parent hadparents in jail or prison at least 3 days. I limit the sample of incarcerations to this lengthbecause based on North Carolina statutes, local jurisdictions are required to hold anarraignment to determine release conditions including the setting of bail for an arrestedindividual within 48 hours of booking. Therefore, I am excluding individuals that are

18See Appendix for details and Appendix Figure A1 that highlights estimated match rates across di�erentyears.

19Appendix Figure A2 provides a breakdown of the typical arrest and incarceration experience of adultsin my dataset. As one can see, a little less than 1/2 of arrests lead to incarceration and 1/3 of thoseincarcerations lead to 2 days or less in jail. Even among individuals incarcerated for more than 3 days,the majority of people are incarcerated for less than 30 days.

9


released at initial arraignment and do not return to jail or prison at a later date sincethey may have had charges dropped and also spent very little time away from the home.Later results highlight that this restriction has minimal impact on my results as modelsthat include all incarcerations no matter how long they last or only incarcerations of amonth or more generate small changes in the main results.

Table 1 provides a summary of my dataset. I provide mean, standard deviations and thenumber of observations for each variable in order to provide a sense of the structureand coverage of the dataset. I do lose some observations for residence or neighborhoodfrom our 2% of student observations with missing residential information and I havefewer observations for some outcomes such as test scores which are taken in grades 3through 8 and school crimes which are not reported consistently until 2007. I also losesome observations for outcomes like changing residence and leaving the school district(Left CMS) because I need two years of data to determine these outcomes. Overall, mydata is comprised of a unbalanced panel of students for years observed in public school.Since I restrict the sample to students born between 1990 and 2001, I ensure overlap withour arrest records (1998-2011) for school age years and later incorporate a number ofcontrols and �xed e�ects to address di�erences in potential exposure to an arrested andincarcerated parent based on cohort and school year. For all students in the dataset, Iobserve 194,163 unique students and have 957,611 observations which represents anaverage of 4.9 school years per student.

Focusing on the population of all students, one can see that the population of publicschool students is racially diverse, and most public school students live in single-family orsmaller apartment buildings (5 units or less). Thirteen percent of students in our datasethave one parent that is arrested sometime during their time in the public school system.The majority of crimes that lead to a parental arrest are for relatively minor o�enses andabout 3% of students have a parent that was ever arrested and incarcerated. In general,incarceration occurs for more serious crimes, but incarceration can occur for even minoro�enses especially if the family is not able to pay for bail or if the parent has a moreserious history of problems with the criminal justice system.20 In an average school yearabout 2% of children have a Parent Arrested and 1% have a Parent Incarcerated for atleast 3 days.

Turning to my main dataset for analysis, students that have arrested parents, I observe18, 858 unique students with 26% of these students having an incarcerated parent. Formy unit of observation of each student in a given school year, the annual rate of parental20Appendix Table A3 provides the breakdown of crime types by individuals just arrested and individuals

arrested and incarcerated. These distribution do not have large di�erences, but one sees a larger shareof serious crimes such as assault and burglary and fewer minor o�enses.

10


arrest is 16% and parental incarceration is 5%. I �nd about 2/3 of parental arrests to befathers and 1/3 to be mothers and 10% of students have both parents arrested at sometime during their school years. Students with arrested parents are di�erent than otherstudents and disproportionately black, live in lower income neighborhoods and haveworse educational and behavioral outcomes.

In order to test multiple outcomes and avoid concerns about multiple hypothesis testingissues with our results, I provide most of my results using summary indices for twotypes of outcomes: educational achievement and behavioral problems in school. I createindices for annual outcomes based on student records and follow the methodology forcreating a summary index as outlined in Anderson (2008) in a re-evaluation of severalearly childhood intervention programs. The educational index is composed of threeoutcomes: reading test scores, math test scores and grade retention and is standardized tomean zero and standard deviation one. The behavior index is composed of three di�erentoutcomes: days absent from school, number of days suspended and school reportedcrimes and is also standardized to mean zero and standard deviation one. I also estimateand present results separately for individual outcomes used in the summary indexes in anumber of results.

3. Empirical Model

My main empirical model is given by Equation 1, where Yicgt is an outcome for individuali in time period t, grade g and who is in birth cohort c. The main variable of interest isa indicator for parental incarceration (ParentIncarcicgt) in year t for student i, withParentArresticgt controlling for the e�ects of parental arrest on the child. My mainmodels include individual �xed e�ects (γi) and all models include birth year by gradeby school year �xed e�ects (λcgt) as well as time-varying covariates (Xit) such as ifindividuals live in a detached (single-family) home, additional controls for parental arrestsin previous and subsequent years (years t-2, t-1, t+1, t+2) ; and neighborhood attributesfor household income, population density and unemployment. Remaining covariates aretime-invariant and thus captured by individuals �xed e�ects (γi). All models includestandard errors clustered by individual i.

Yicgt = α+ β1ParentArresticgt + β2ParentIncarcicgt (1)

+β3Xicgtδ+ γi + λcgt + εicgt

11


Identi�cation in Equation 1 results from variation over time in parental incarcerationspells for students that have at least one parent arrested during school age (7-17). Thekey coe�cient in this model is β2 and controlling for parental arrest is important fortwo issues. First, non-criminal parenting is likely correlated with the timing of arrest andvaries over time which requires an annual control for criminal activity, ParentArrestit.Annual measures of parental arrest should control for a number of factors that negativelyin�uence children’s outcomes including the trauma of parental arrest, potential job lossand �nancial problems related to �nes and legal costs. Conditional on parental arrest,β2 identi�es some of the factors that coincide with parental incarceration includingchanges in child’s living arrangements and change/loss of relationship with a parent.Loss of parental �gure and moving to a new residence could generate negative e�ects ifmoving leads to loses in social relationships for the child or if the incarcerated parentwas a positive role model. Conversely, if parental incarceration leads to the removal of anegative role model or moving to an improved household or neighborhood environment,then we may see positive e�ects on the child from parental incarceration.

The main identi�cation issue is that incarcerated parents are di�erent than parents thatare only arrested or parents with no criminal justice experience. In the simplest sense,incarcerated individuals are either more serious criminals or have more limited resourcesto incur legal costs and avoid incarceration through legal representation or payment ofbond. Figure 2 highlights the relationship between criminal parent type and a child’saverage educational and behavioral outcomes. Irrespective of outcomes, children withparents that have more arrests and/or more incarcerations have worse outcomes, lowervalues for the education index and higher value for the behavioral index. Furthermore,this relationship holds even when one conditions on individual attributes, neighborhoodattributes and �xed e�ects for cohort, grade and school year as shown in Appendix FigureA4.

Given I want to highlight the role of incarceration, the relationship highlighted in Figure2 leads to issues with any results that �nd worse outcomes for children with parentalincarceration since one cannot disentangle the e�ects of incarceration from parentalenvironment. Finding a policy that generates random variation in parental incarcerationis challenging and recent literature has used sentencing variation due to random judgeassignments (Aizer and Doyle (2015) , Mueller-Smith (2015), Dobbie et al. (2017)) togenerate credible estimates of the impacts of incarceration on a criminal’s post-releaseoutcomes. In my context using variation in sentencing by judges is not possible givendata limitations. Therefore, I bound the impact of parental incarceration by using the factthat any concerns regarding unobserved parenting quality or �nancial resources would

12


only serve to bias my estimates toward �nding a negative e�ect of incarceration. Todisentangle the incarceration e�ects from the initial e�ects of arrest, I control for arrestsand use only variation in the share of arrests that lead to incarcerations. In general,results that show any bene�ts from incarceration will be conservative in nature sinceany concerns about omitted variables would bias results toward �nding no bene�ts.

In order for this empirical strategy to work, it is important to properly control for as muchas possible about the timing of arrests and other parental attributes. The incorporationof individual �xed e�ects as well time-varying controls for arrests including some lagand lead measures of arrests should help address omitted parental quality that may varyover time. Omitted parental quality may include parental �nancial resources, violence orinstability in the family as well as negative parental role modeling. All of these factors arelikely to be positively correlated with being arrested and being arrested more frequently.

Some additional concerns arise in determining the impact of parental incarceration onchildren including sorting away from the public school system in response to parentalincarceration, sentencing and incarceration length being in�uenced by family structureand teachers or schools treating the children of parental incarceration di�erently. Onepotential impact of parental incarceration is that the remaining parent moves the childto a new residence or the child may even be relocated to other family member homes oreven a social service facility.21 This relocation may be problematic for observing childrenin subsequent school years especially if children leave the county or attend school at aprivate facility. Since later results do highlight movement away from the school systemafter both arrest and incarceration, I am limited in my ability to examine the longer-termimpacts of incarceration for children leaving the school district. Later results do highlightthat leavers and stayers have similar short-term outcomes. It may be the case that leavingthe school district may be more bene�cial to the child in terms of removing a negativerole model but this e�ect may be decreased by the cost to the child of moving furtheraway or changing to a di�erent type of school. Stayers in the school district after parentalincarceration may receive lower bene�ts from parental incarceration, but do not incurthe costs of moving.

One may be concerned about di�erential sentencing for arrested parents. Since I onlyfocus on children with arrested parents this concern is minimized, but any di�erentialsentencing may relate to more serious criminals and incarceration. In general, sentencingmay be less harsh if a judge worries about negatively impacting families through incar-

21Based on Bureau of Justice Statistics Inmate Surveys, most parents reported (84%) that at least one oftheir children was in the care of the other parent; Fifteen percent identi�ed grandparents as caregivers,6 percent other relatives and 3 percent that at least one child was in a foster home/agency. ( Glaze andMaruschak (2008) )

13


ceration while sentencing may be harsher if a judge wants to protect children. Ultimately,I cannot verify di�erential sentencing for parents, but later results do show that mymain conclusions and results occur across di�erent criminal charges as well as for theincarceration of either a mother or father.

One also may be worried that teachers or administrators are treating kids whose parentsare incarcerated di�erently and this may lead to harsher or more lenient punishmentsfor bad behavior. This type of change in teacher or administrator behavior should notin�uence test scores but may in�uence suspensions or grade retention. This concernis not directly testable, but later results provide some evidence that this concern isnot the main determinant of my results. Speci�cally, I only �nd signi�cant e�ects oneducation outcomes for state standardized test scores. Also, I �nd that behavioral resultsare consistent across all three outcomes (school suspensions, absences and school crimes)even though there is really only discretion in the use of school suspensions.

4. Main Results

As presented in Figure 2, one sees that arrested and incarcerated parents have childrenwith worse academic and behavioral outcomes. These large e�ects are on the extensivemargin of having a parent ever arrested as well as on the intensive margin as parentswith more arrests and incarceration had children with even worse outcomes. My mainregression results restrict the sample to only students that have a parent that is everarrested to focus on models that use variation in the timing of arrests and incarcerationto identify the impacts of parental incarceration.

Table 2 provides 7 models that begin with a basic model in column 1 and moves tomy preferred speci�cation in column 7. Column 1 indicates that in the year a parent isarrested, a student has a 0.04 standard deviation lower value for test scores and a 0.12standard deviation higher behavior index.22 These results are similar for incarceratedparents in column 2. Since incarceration is preceded by an arrest, remaining modelsinclude both parental arrest and incarceration. In essence, the parental incarcerationvariable indicates the additional e�ect of incarceration conditional on arrest. Models 3and 4 continue to highlight worse outcomes for children whose parents are arrested andincarceration even with Census Block Group (CBG) �xed e�ects. Model 5 introduces

22I will typically just incorporate measures of academic outcomes using a composite of EOG test scores andexclude grade retention since grade repetition consistently provides small and imprecise impacts fromparental arrest or incarceration and thus mutes the impacts of test scores. Test scores are normalizedto mean of zero and standard deviation of one based on North Carolina statewide test score for a givengrade and year.

14


family �xed e�ects where one identi�es families based on two children with a shared lastname living in the same residence in the same school year. Once I control for family �xede�ects, one sees the negative e�ects of incarceration on test scores becoming positiveand the impact of parental incarceration on the behavior index changing to a negativecoe�cient and thus a positive e�ect on children. In essence, better controlling for familyenvironment and using temporal variation in arrests and incarceration removes someof the negative e�ects captured in the incarceration variable. I further re�ne controlsfor parenting and family environment by introducing student �xed e�ects in column6 and adding controls for individual i’s parental arrests in the two years just prior andjust after time period t in column 7. The key trend to observe in the seven models inTable 2 is that as I introduce successively better controls for parental environment atthe time of arrest, results show bene�ts from parental incarceration. Column 7 �nds apositive impact of 0.024 standard deviations in test scores and 0.065 standard deviationin behavioral problems due to contemporaneous parental incarceration. These e�ectsmore than o�sets the negative e�ects on test scores and almost completely o�sets theincrease in behavioral problems that coincide with parental arrest.

Table 3 breaks down e�ects for individual outcomes and when I focus on speci�coutcomes, one sees slightly larger positive impacts of parental incarceration on readingrelative to math test scores. Grade repetition is not e�ected by parental incarceration.The �nding of no e�ect on grade retention may simply be that test scores e�ects aretoo small to in�uence the larger decision to have a child repeat a grade. Results for alloutcomes that are used to create my behavioral index show consistent bene�ts fromparental incarceration. In general, behavioral e�ects are small in magnitude with speci�coutcome e�ects ranging from about 4% to about 14% of mean annual values. Since theseare only the e�ects on outcomes for one school year, the e�ects of repeated incarcerationscould be cumulatively quite larger or could even e�ect long-term outcomes.

One way to explore the plausibility of these results is to examine the impact of time ofyear for when a parent is incarcerated. Since I begin assigning parents to the school yearof a child starting in the summer prior to school beginning, I focus on four seasons ofthe year where a parental incarceration would impact the child just prior and potentiallyincluding the entire school year. Fall and winter would only impact part of the year andspring would only impact the last few months of the school year.23 Figures 3 and 4 providethe results of my main model in Table 2 where I interact parental incarceration and arrestwith a dummy for each of the four seasons of the year. Result are quite compelling inthat the behavioral index results show a clear trend of smaller and smaller e�ects as23Summer is June, July, August; Fall is September, October and November; Winter is December, January

and February; Spring is March, April and May.

15


we shorten the time frame upon which one observes behavioral outcomes. One wouldexpect larger e�ects for summer or fall parental incarceration on behavioral outcomessimply because there are more school days for a student to be absent or get in troubleafter the parental incarceration happens. The most interest result is Figure 3 where onesees only a positive e�ect for spring incarceration. This result seems strange at �rst incomparison to the behavioral index, but makes sense when one considers that test scoresare taken during the last 3 weeks of school in the spring and thus the e�ects of parentalincarceration may generate an improved home environment in the months just prior orduring EOG test taking where students are doing in-class practice exams, homework toprepare for exams and getting adequate sleep during the week(s) of exams.

The age pro�le of children may matter for my outcomes due to the varying in�uenceof parents on di�erent age pro�les of their children. To test the e�ects for d�erentaged children, Figures 5 and 6 provide results where I interact dummies for age groupswith incarceration and arrest variables. Results for test scores are small, but indicatepositive e�ect for ages 7-8 and 9-10. The age pro�le for the behavioral outcomes is moreheterogenous with large and signi�cant e�ects for young children (ages 7-8, 9-10). E�ectsdisappear for the older groups with 15-17 years old provide imprecise and opposite e�ects.Stronger e�ects of parental incarceration on younger children is consistent with Lyle(2006) �nding larger academic e�ects for young children whose parents were subject tomilitary deployments. It may be the case that incarcerations have very di�erent e�ectson young children relative to older children. For example, young children are staying inthe house with a parental �gure and incarceration will likely just change the adults in thehome, while older kids may simply leave the home in the case of parental incarcerationand stay with friends or other families. These older kids may still attend school, but haveless supervision and support which could lead to worse outcomes.

4.1. Medium and Longer-term Impacts

Given that results so far focus on same year e�ects, I do explore if e�ects last beyonda given year as well as the impact of previous and future arrests and incarcerations oncurrent outcomes in Figures 7 and 8. These �gures provide the coe�cients and con�dencebands for a single model for each outcome based on the inclusion of variables for threeyears of lagged and lead Parent Incarcerated and Parental Arrest variables for a givenstudent and year.24 The use of lags and lead measures of arrests and incarceration are

24Given the presence of student �xed e�ects, the omitted group in this model is all arrests and incarcerationsmore than 3 years before or after our dependent variable in year t. I exclude students that exit theschool system after year t as well as students without observations prior to year t. For included students

16


informative in two ways. First, arrest likely coincide with negative parental environmentswhich may begin prior to an actual arrest as well as continue after an arrest and laggedand lead arrests will highlight this relationship. Second, lagged measures of incarcerationtest if previous year incarcerations a�ect later year outcomes and lead measures ofincarceration provide a falsi�cation test by testing in future arrests and incarcerationimpact current year outcomes. Consistent with the main results for test scores in Table 3,results in Figure 7 highlight a positive and signi�cant e�ect at time t for incarcerationsand a smaller negative e�ect for arrests. One does not see any signi�cant e�ects fromprevious year incarcerations (t-1 through t-3) and smaller and nosier estimates fromfuture incarcerations (t+1 through t+3).

Results for my Behavioral Index in Figure 8 show e�ects for both arrests and incar-cerations across time periods. Speci�cally, the positive coe�cients show the negativeoutcomes surrounding arrests on behavioral outcomes beginning in time period t− 1with the largest e�ects in time period t and impacts moving to close to zero in t + 1through t+ 3. These results indicate that a child’s behavior is worse right around thetime of parental arrest and is consistent with arrests capturing the trauma of arrestin year t as well as the worse parental environment leading up to the year of arrest.In Figure 8, one sees bene�ts from parental incarceration starting in year t − 2 andthe largest and only signi�cant e�ect in year t. E�ect sizes drop by half in year t + 1and future years are small in magnitude and imprecise. Positive impacts from previousparental incarcerations may simply be a result of parents with longer jail/prison stintsbene�ting current year outcomes.25 Results from these event study �gures consistentlyshow bene�ts of contemporaneous parental incarceration on a child’s outcomes andsome, albeit imprecise, e�ects of previous years incarcerations on current outcomes.

Since I can test the impacts of parental arrest and incarceration on some adult outcomesfor a subset of earlier cohorts, I introduce a slightly di�erent model in Table 4. Here I canno longer use individual �xed e�ects since outcomes are for young adults and limited tothe last year(s) of matriculation in public school. Therefore, I include each student onlyonce and focus on the subset of cohorts (1990-1995 birth year) where I observe schoolmatriculation for children through age 17. I limit my sample to children with parentsthat were incarcerated at some point between ages 7 and and 17 and identi�cation inthis model is based on the prevalence of arrests and incarcerations throughout early andlater childhood.26 Speci�cally, I estimate the following model (Equation 2) which allows

that may have missing value for a year included in this �gure, I simply include a missing indicator andset that lagged or lead arrest and incarceration variable to 0.

25Later results speci�cally explore the impacts of longer-term incarcerations on children’s outcomes.26I also limit my sample of students to individuals that matriculate for at least one year in high school in

CMS. Results are similar without this restriction.

17


the age when a child experienced a parental arrest or incarceration to identify my e�ectson long-term outcomes.27

Yic = α+ β1ParentArrestic ∗D(age7− 11) + β2ParentArrestic ∗D(age12− 17) (2)

+β3ParentIncarcic ∗D(age7− 11) + β4ParentIncarcic ∗D(age12− 17)

+β5Xcδ+ λc + εic

I estimate this model for my two longer-term outcomes - Dropping out of High School& Number of Arrests (age 16-18). The �rst set of results in columns 1 and 3 are based onmodels that only examine parental arrests and incarceration at ages 7-11 and columns 2and 4 examines results when I also include results for ages 12-17. Results for Droppingout of High School show that parental incarceration bene�ts younger children (ages7-11). Coe�cients indicate an increase of 38% over baseline means of 0.14. The impactson dropping out of high schools likely represents a combination of both academic andbehavioral e�ects from Table 3. Results for number of arrests are limited and less precise.The presence of no e�ects for adult arrests is consistent with the more general �ndingof shorter term impacts of parental incarceration. This result may be explained by thedynamics of household formation, where the incarceration of a parent initiates newfamily formation and an incarcerated parent may simply be replaced with another similartype of parent. I explore how households respond to parental incarceration in the nextsection.

4.2. Mobility and Sample A�rition

One of the �rst things to explore regarding the mechanisms underlying the main resultsis the impact of parental incarceration on household mobility. Table 5 provides a numberof mobility based outcomes for my main parental incarceration model. Results providesome interesting insight into how families react to the incarceration of a parent. First,one sees a small and marginally signi�cant e�ect of incarceration on moving during aschool year with parental incarceration decreasing the probability of moving by about1.4 percentage points of an average of 12 percentage points. Another signi�cant result inTable 5 is leaving the Charlotte-Mecklenburg School (CMS) District in the following year.Parental incarceration increases the probability of transferring out of a CMS public school

27Since I limit my sample to children ages 7 through 17 , I am comparing two parents that were incarceratedthat only vary in the child’s age at the time of incarceration.

18


for the next school year by 2 percent points which represents a 33% increase over the 6percent point baseline annual probability of leaving CMS. Leaving the school districtlikely indicates moving out of Mecklenburg County, but could represent transferring to aprivate school. Given the income pro�le of this population, resources would be limited forany private school attendance. One potential explanation for leaving the school district isdue to the loss of �nancial resources and/or childcare from the incarceration of a parent.This may lead families to move to other relatives that live outside the school district inorder to a�ord housing or replace childcare.

Leaving the school district is not consistent with social service intervention for childrenof incarcerated parents since social service agencies would keep children in the publicschool system. Based on previous literature, children of incarcerated parents are mostlikely living with the other parent, but some children are transferred to other relativesand only a small percent are transferred to foster homes/institutions.28

One additional concern with children moving out of CMS is that they may be leaving inthe middle of the school year and that behavioral outcomes may just mechanically belower due to observing students for only a portion of the school year. To test for thisissue, column 6 asks if a student took a required End-of-Grade (EOG) exam. EOG examsare administered in the last couple weeks of school and all students in 3rd-8th gradeare required to take the exams. In my data, 81% of student-grade observations in grades3 through 8 indicate a completed EOG exam and the main conclusion is that parentalincarceration has no impact on taking EOG exams. This result limits any concerns thatthe children of incarcerated parents are leaving the school district prior to the end ofthe school year. A �nal result of interest in column 7 of Table 5 is some evidence that anincarcerated parent is more likely to live at a new address after incarceration. Using theaddress information from future parental arrests, I examine if incarcerated parents areless likely to have a future arrest while living at the same address as given at the time ofa parent’s original arrest.29 Given that children with a incarcerated parent are not reallyany more likely to move residences within the school district and we observe arrestswithin the school district, the fact that a incarcerated parent is more likely to be at anew address suggests that an incarcerated parent is less likely to live with the child afterrelease from jail/prison.30 This result is consistent with new household formation for the28 Based on inmate surveys, Glaze and Maruschak (2008) highlight that most children (88%) of an incar-

cerated father live with the mother and only a small percentage of children are transferred to fosterhomes (2.9%). Results are quite di�erent between maternal versus paternal incarceration with maternalincarceration leading to about 2/3 of kids to live with a relative other than the father and 10.9% beingin foster care.

29This sample is limited in size to only include student-year observations with arrest and future rearrestaddress information as well as public school information in the following year.

30Since it is clearly selective who gets rearrested, this evidence is only suggestive that an incarcerated

19


child that does not include the incarcerated parent post-release.

4.3. Robustness Tests and Validating the Empirical Model

To ensure our speci�cation are robust to di�erent assumptions regarding parentalmatching, types and lengths of incarcerations as well as control variables, I present aseries of models in Table 6 that varies models along a couple of dimensions and highlightsfairly consistent results. Column 1 of Table 6 presents my main results and column 2provides result when I allow any children that are assigned multiple parents based onparental matching or live in a larger apartment complex (> 5 units) to be included inestimation.31 One can see that the number of observations increases by about 30% when Iinclude these types of matches. Column 3 drops students with duplicate parental matchesfrom column 2 but keeps all apartment complex matches. Both sets of results are similarto the main model in column 1 with a slightly smaller e�ect for behavioral outcomes,which is consistent with the inclusion of some non-parental arrestees due to multiplematches. Column 4 provides a model where I match adult arrestees to children onlybased on addresses. These models allow for other relatives, housemates and some neigh-boring units (with the same address) to be counted as parental arrests and incarcerations.Results provide small and insigni�cant impacts on behavioral outcomes and a smalland insigni�cant e�ect on test scores when we include same address individuals thathave di�erent last names. This result indicates that the arrest and incarceration of otherrelatives or co-residents does not impact test scores and behavorial outcomes.

Columns 5 and 6 of Table 6 examine the impacts of using di�erent sets of incarcerationsas a measure of parental incarceration. Irrespective of including all incarcerations (eventhose less than 3 days) or only incarcerations that last at least a month give similarresults. At �rst, one would expect that focusing on longer-term incarceration in column5 would generate stronger e�ects, but two things work against this �nding. First, parentswith longer-term incarcerations are likely more serious criminals thus creating potentialbias towards �nding a negative e�ect of incarceration. Second, treating short-termincarcerations as non-incarcerations changes comparisons in my model since it mayproduce better outcomes for those time periods assigned as simply arrests if short-termincarcerations have any bene�ts. In the end, I adopt the most conservative de�nition ofincarcerations ( at least 3 days in length) in my main model since it better mirrors thearrest and criminal process which requires bail hearings within the �rst 48 hours andthus does not assign incarcerations to any individuals that are released after arrest due

parent is less likely to live with the child post-release.31In cases with missing apartment numbers, parental matching is based on last name and property address.

20


to insu�cient evidence to prosecute or individuals held to avoid domestic con�ict or forintoxication. The remaining three columns in Table 6 test the e�ects of the inclusion ofdi�erent combinations of control variables from a more parsimonious model to the mostexpansive set of control variables. All model still include basic individual controls thatvary over time as well as individual �xed e�ects. Results are quite consistent with themain results.

Another way to test if the intuition of the main results are consistent with the idea ofbene�ts accruing to the removal of a parental �gure is to see if the type of release mattersfor my results. Table 7 provides a model for my outcomes where I separate the parentalincarceration variable into four categories of inmate release. I restrict my sample to the2005 through 2011 school years due to limitation in the availability of speci�c releasetypes for this dataset. The idea is to test if individuals released through bond, by timeserved, other types of release and longer-term sentences with a release in a subsequentschool year.32 Incarcerated parents are fairly evenly split among these release typesand other releases in this context would indicate transfers to another jurisdiction, casedismissed, a weekender program where an individual only serves time on the weekendsuntil their sentence is completed or released to the custody of another person.33 In general,these other release types would still indicate time served in jail or prison and even bondrelease could be followed by later incarceration.34 Table 7 provides some evidence thatmy main results replicated in column 1 for this shorter time period are primarily drivenby longer-term releases that occurred in subsequent school years as well as releasedafter time served. Results are somewhat noisy but are consistent with the main e�ectsfor behavioral outcomes.

Variation in release type also re�ects variation in incarceration length, which I explorein more detail in Figures 9 and 10. These Figures provide a plot of parents incarcerationcoe�cients for a version of main results in Table 3 where I replace the parents incarceratedvariable with a series of dummy that disaggregate all incarcerations by the length ofincarceration. Figures 9 shows no consistent e�ects based on length on incarceration andre�ects my general �nding of small positive e�ects of parental incarceration on sameyear test score outcomes. Figure 10 highlights a consistent e�ect of parental incarcerationdecreasing behavioral outcomes with some larger e�ects for 3-6 month incarcerations32If North Carolina, o�enders that commit crimes after October 1st, 1994 are not given early release and

parole, but rather released on post-release supervision.33Ideally, I can examine all of these unique release types, but the sample size would get quite small to

look at speci�c release program such as the weekender program.34In some cases, I am unable to see later incarceration after initial bond release due to transfer to other

jurisdictions, federal cases or incarcerations that occur after the end of study time period. These casesare likely the exception since I see all county jail and state prison incarcerations including coding uptransfers from jail to prison as a single incarceration spell.

21


and 2 or more year incarcerations. Di�erences across di�erent incarceration lengths mayre�ect the complex nature of having a parent removed from the home with the potentialfor short-term behavioral reactions as well as longer-term changes in behavior due tochanges in parental environment. The potential for short and longer-term behaviorale�ects may also explain some of the heterogeneity in the test score results in Figure 9.

4.4. Heterogeneous Impacts of Parental Incarceration

Table 8 provides some insight into the main result by testing some elements of hetero-geneity in my results. Columns 2 and 3 provide results that are interacted by gender ofparent and then gender of student. E�ects for mother are imprecise but suggest largerbene�ts from the incarceration of a mother for both test scores and behavioral outcomes.The gender of the student has little e�ect of estimated impacts. For movers in column4, there is larger bene�ts for test scores, but smaller bene�ts for behavioral outcomes.Column 5 shows almost no di�erence in e�ects on outcomes for students that left theschool district in the following year. Column 6 examines if my �ndings di�er for parentsthat are only arrested one time in our study period and column 7 for families where bothparents are arrested at some point.35 Parents that have only one arrest provide moremuted bene�ts from incarceration, while children that have both parents arrested at somepoint in time lose any bene�ts for test scores and have smaller bene�ts for behavioraloutcomes. This estimates are imprecise but are consistent with the loss of incarcerationbene�ts for families when both parents are involved in the criminal justice system. Thisis not surprising given that the incarceration of a parent typically leaves the remainingparent as the primary caregiver.

Column 8 limits my analysis to students with at least one parent incarcerated to furtherlimit variation to only households with parents that have more serious involvement inthe criminal justice system. Results in column 8 are quite similar to the main results. One�nal speci�cation that is informative but substantially limited by my ability to matchbirth records to my main sample is column 9. In column 9, I restrict the sample to onlychildren who have an arrestee parent who I can also match based on birth records.36

This double veri�cation is informative since it limits result to birth parents only as well

35I am unable to speci�cally examine years where both parents are arrested and incarcerated given thesmall sample of children and years where both parents are arrested and incarcerated (n=30).

36The small sample of parental arrests and incarceration veri�ed by birth records occurs for two reasons.First, about 66% of birth records are matchable to student records. This number is lower than the 80%in Figlio et al. (2016). This di�erence is most likely due to the fact that Figlio et al. (2016) had recordsthat were matched administratively by the state of Florida and also covered the entire state of Florida.Second, approximately 1/3 of birth records are missing information on fathers and given the largeshare of men in the criminal justice system this generates a substantially smaller sample size.

22


as provides validation that address matching is capturing similar individuals. Result forthe Education and Behavior Index outcomes are both larger in magnitude. These largerimpacts may simply be a result of the stronger impact of actual birth parents on children’soutcomes. It may be the case that birth parents are more substantial role models thanother family members or step parents and/or that imprecise matching based on addressis muting our main results. Appendix Table A2 furthers test the role of heterogeneityin families decision to move or leave the school district by di�erent family or studentattributes and results indicate move decisions do not vary by maternal incarceration,student gender or variation in the degree to which families are involved in the criminaljustice system.

Another element of heterogeneity is the gender of the child as well as di�erences betweenmaternal versus paternal incarceration. Table 9 provides results where I disaggregate myvariable for parental incarceration into speci�c e�ects by a child’s and parent gender. Ingeneral, e�ects are less precise but consistent with earlier results showing larger bene�tsfrom maternal incarceration than paternal incarceration. These e�ects do not seem tovary that much by the gender of the child. The largest and most precise estimates are fordaughters of incarcerated mothers who have imprecise positive bene�ts on test scoresof about 0.06 standard deviations and marginally signi�cant decreases in behavioraloutcomes of 0.12 standard deviations. My �nding of greater bene�ts from maternalincarceration may relate to the potential bene�ts from the removal of a more in�uentialor important role model. The importance of the mother as a role model is even morelikely given the large e�ects I �nd for younger aged children.

Table 10 shows e�ects for di�erent criminal charges at the time of parental arrest. Resultare small in magnitude and imprecise for educational outcomes but again behavioraloutcomes provide some support that more serious crimes - indexed property and violentcrimes have larger and more precise estimated bene�ts. Results are also consistent, butless precise for drug and alcohol crimes. Other crime and technical violations providesmaller and less precise e�ects. Heterogeneity by crime type is somewhat limited bythe fact that incarceration captures elements of crime seriousness which correlates withcrime type.

Combining the results for parental gender, crime type and focusing on young childrenprovides some stronger e�ects from parental incarceration. Speci�cally, Table 11 showsthe e�ects of maternal versus paternal incarceration on young children based on thecriminal charge at the time of arrest. Across these di�erent groups, the e�ects in averagetest scores are small and not precise. E�ects for behavioral outcomes are more interestingand shows that maternal incarceration has a larger e�ect on younger children than

23


paternal incarceration and represents about a 0.15 standard deviation decrease. Thesee�ects are mostly heavily concentrated in more serious crimes for mothers and thelargest and most precise e�ects are for mothers who commit violent crimes. For fathers,the largest e�ect is for drug and alcohol crimes and may represent potential bene�tsfrom a paternal incarceration in terms of recovery from addiction while in jail/prison.

5. Conclusions

The dual results for arrests and incarceration highlight the trauma and negative parentalenvironment at the time a parent is arrested, but incarceration may bene�t children andthis is likely through its ability to remove negative role models from the household. Infact, parental incarceration erases the negative e�ects of parental arrest on behavorialoutcomes in a given school year, with average e�ects of about 0.07 standard deviationsin absenteeism, suspensions and school crimes. I provide evidence that bene�ts coincidewith changing household composition after incarceration, but the exact mechanismthrough which parental incarceration alters a child’s environment cannot be directlydetermined in the data. I provide evidence that families bene�t more from the removalof a parent with longer-term incarcerations, that has more serious crime charges and isa mother. Additionally, larger bene�ts accrue to younger aged children.

These results contradicts existing literature that consistently �nds a negative relationshipbetween parental incarceration and the outcomes of children. The novelty of results hereis that I provide evidence of bene�ts due to parental incarceration in the US criminaljustice system. Results are important for policy given the substantial debates regardingthe failure of the criminal justice system to provide any bene�ts to families and thecriticism that factors that lead to parental incarceration are often transferred inter-generationally. Future research should determine the exact mechanisms through whichparental incarcerations bene�ts children and tailor policies that leverage the bene�ts ofremoving negative role models while minimizing the long-term costs of incarcerationon a parent.

References

Aizer, A. and Doyle, J. J.: 2015, Juvenile incarceration, human capital, and future crime:Evidence from randomly assigned judges, The Quarterly Journal of Economics p. qjv003.

Ananat, E. O., Gassman-Pines, A., Francis, D. V. and Gibson-Davis, C. M.: 2011, Children

24


left behind: The e�ects of statewide job loss on student achievement, Technical report,National Bureau of Economic Research.

Anderson, M. L.: 2008, Multiple inference and gender di�erences in the e�ects of earlyintervention: A reevaluation of the abecedarian, perry preschool, and early trainingprojects, Journal of the American Statistical Association 103(484).

Angrist, J. D. and Johnson IV, J. H.: 2000, E�ects of work-related absences on families:Evidence from the gulf war, ILR Review 54(1), 41–58.

Balsa, A. I.: 2008, Parental problem-drinking and adult children’s labor market outcomes,Journal of Human Resources 43(2), 454–486.

Besemer, S., Van der Geest, V., Murray, J., Bijleveld, C. C. and Farrington, D. P.: 2011, Therelationship between parental imprisonment and o�spring o�ending in england andthe netherlands, British Journal of Criminology 51(2), 413–437.

Bhuller, M., Dahl, G. B., Løken, K. V. and Mogstad, M.: 2016, Incarceration, recidivismand employment, Technical report, National Bureau of Economic Research WP .

Braman, D.: 2004, Doing time on the outside: Incarceration and family life in urban America,University of Michigan Press.

Chatterji, P. and Markowitz, S.: 2001, The impact of maternal alcohol and illicit drug use onchildren’s behavior problems: evidence from the children of the national longitudinalsurvey of youth, Journal of Health Economics 20(5), 703–731.

Cho, R. M.: 2009, Impact of maternal imprisonment on children’s probability of graderetention, Journal of Urban Economics 65(1), 11–23.

Cho, R. M.: 2010, Maternal incarceration and children’s adolescent outcomes: Timingand dosage, Social Service Review 84(2), 257–282.

Dobbie, W., Gronqvist, H., Niknami, S., Priks, M. and Palme, M.: 2017, The intergenerationale�ects of parental incarceration, Technical report, Working Paper.

Doiron, D. and Mendolia, S.: 2012, The impact of job loss on family dissolution, Journalof Population Economics 25(1), 367–398.

Doyle Jr, J. J.: 2008, Child protection and adult crime: Using investigator assignment toestimate causal e�ects of foster care, Journal of political Economy 116(4), 746–770.

Doyle Jr, J. J. et al.: 2007, Child protection and child outcomes: Measuring the e�ects offoster care, American Economic Review 97(5), 1583–1610.

Figlio, D., Karbownik, K., Roth, J., Wasserman, M. et al.: 2016, School quality and the gendergap in educational achievement, The American Economic Review 106(5), 289–295.

Foster, H. and Hagan, J.: 2007, Incarceration and intergenerational social exclusion, SocialProblems 54(4), 399–433.

25


Gertler, P., Levine, D. I. and Ames, M.: 2004, Schooling and parental death, Review ofEconomics and Statistics 86(1), 211–225.

Glaze, L. E. and Maruschak, L. M.: 2008, Parents in prison and their minor children, USDepartment of Justice, O�ce of Justice Programs Washington, DC.

Hilger, N. G.: 2016, Parental job loss and children’s long-term outcomes: Evidence from 7million fathers’ layo�s, American Economic Journal: Applied Economics 8(3), 247–283.

Lyle, D. S.: 2006, Using military deployments and job assignments to estimate the e�ectof parental absences and household relocations on children’s academic achievement,Journal of Labor Economics 24(2), 319–350.

Mueller-Smith, M.: 2015, The criminal and labor market impacts of incarceration.

Mühlenweg, A. M., Westermaier, F. G. and More�eld, B.: 2016, Parental health and childbehavior: evidence from parental health shocks, Review of Economics of the Household14(3), 577–598.

Murray, J. and Farrington, D. P.: 2005, Parental imprisonment: e�ects on boys’ antisocialbehaviour and delinquency through the life-course, Journal of Child Psychology andpsychiatry 46(12), 1269–1278.

Murray, J., Farrington, D. P. and Sekol, I.: 2012, Children’s antisocial behavior, mentalhealth, drug use, and educational performance after parental incarceration: A system-atic review and meta-analysis., Psychological Bulletin 138(2), 175.

Oreopoulos, P., Page, M. and Stevens, A. H.: 2008, The intergenerational e�ects of workerdisplacement, Journal of Labor Economics 26(3), 455–483.

Page, M., Stevens, A. H. and Lindo, J.: 2007, Parental income shocks and outcomes ofdisadvantaged youth in the united states, The problems of disadvantaged youth: Aneconomic perspective, University of Chicago Press, pp. 213–235.

Parke, R. D. and Clarke-Stewart, K. A.: 2003, The e�ects of parental incarceration onchildren, Prisoners once removed: The impact of incarceration and reentry on children,families, and communities pp. 189–232.

Rege, M., Telle, K. and Votruba, M.: 2011, Parental job loss and children’s school perfor-mance, The Review of Economic Studies p. rdr002.

Schaller, J. and Zerpa, M.: 2015, Short-run e�ects of parental job loss on child health,Technical report, National Bureau of Economic Research.

Snow Jones, A., Miller, D. J. and Salkever, D. S.: 1999, Parental use of alcohol and children’sbehavioural health: a household production analysis, Health Economics 8(8), 661–683.

Stevens, A. H. and Schaller, J.: 2011, Short-run e�ects of parental job loss on children’sacademic achievement, Economics of Education Review 30(2), 289–299.

26


Turanovic, J. J., Rodriguez, N. and Pratt, T. C.: 2012, The collateral consequences ofincarceration revisited: A qualitative analysis of the e�ects on caregivers of childrenof incarcerated parents, Criminology 50(4), 913–959.

Western, B. and Pettit, B.: 2010, Collateral costs: Incarceration’s e�ect on economicmobility, Washington, DC: The Pew Charitable Trusts .

Wildeman, C.: 2010, Paternal incarceration and children’s physically aggressive behaviors:Evidence from the fragile families and child wellbeing study, Social Forces 89(1), 285–309.

Wildeman, C., Andersen, S. H., Lee, H. and Karlson, K. B.: 2014, Parental incarcerationand child mortality in denmark, American journal of public health 104(3), 428–433.

Wildeman, C. and Western, B.: 2010, Incarceration in fragile families,The future of children20(2), 157–177.

27


Figure 1: Distribution of Parental Incarcerations20

040

060

080

010

0012

00F

requ

ency

< 5

days

5−10

day

s

11−1

9 da

ys

20−2

9 da

ys

1−3

mon

ths

3−6

mon

ths

6−12

mon

ths

12−1

8 m

onth

s

18 m

.− 2

year

s

> 2

year

s

Parent − Incarceration Length (days)

28


Figure 2: Average Index Outcomes by Criminal Parent Type

A. Education IndexNo Arrests

1−2 Arrests

1−2 Arrests/1 Incarc

3+ Arrests

3+ Arrests Serious

3+ Arrests/2+ Incarc

3+ Arrests/2+ Incarc/Serious

−.3 −.2 −.1 0 .1

B. Behavior IndexNo Arrests

1−2 Arrests


3+ Arrests

3+ Arrests Serious



0 .1 .2 .3 .4 .5

Notes This �gure depicts mean outcomes by the di�erent types of criminal parents.

Figure 3: Interaction with Season - Test Scores

−.1

−.0

8−

.06

−.0

4−

.02

0.0

2.0

4.0

6.0

8.1

Summer Fall Winter SpringSeasons

ArrestsIncarcerations

This �gure provides the results of the main estimation model with student �xed e�ects, parental arrestvariables and other controls included in Table 3 and interacts arrest and incarceration variables withdummies for the time of year a parent is initially arrested or sent to jail/prison.

29


Figure 4: Interaction with Season - Behavior Index

−.2

−.1

6−

.12

−.0

8−

.04

0.0

4.0

8.1

2.1

6.2

Summer Fall Winter SpringSeasons


This �gure provides the results of the main estimation model with student �xed e�ects, parental arrestvariables and other controls included in Table 3 and interacts arrest and incarceration variables withdummies for the time of year a parent is initially arrested or sent to jail/prison.

30


Figure 5: Interaction with Age - Test Scores

−.1

−.0

8−

.06

−.0

4−

.02

0.0

2.0

4.0

6.0

8.1

7−8 9−10 11−12 13−14School Age


This �gure provides the results of the main estimation model with student �xed e�ects, parental arrestvariables and other controls included in Table 3 and interacts arrests/incarceration variables with dummiesfor di�erent ages of a student based on the student-year observation.

31


Figure 6: Interaction with Age - Behavior Index

−.2

4−

.2−

.16−

.12−

.08−

.04

0.0

4.0

8.1

2.1

6.2

.24

7−8 9−10 11−12 13−14 15−17School Age


This �gure provides the results of the main estimation model with student �xed e�ects, parental arrestvariables and other controls included in Table 3 and interacts arrests/incarceration variables with dummiesfor di�erent ages of a student based on the student-year observation

32


Figure 7: Coe�cient Plot over Time - Tests Scores

−.0

8−

.06

−.0

4−

.02

0.0

2.0

4.0

6.0

8

Est

imat

ed E

ffect

s ov

er T

ime

t−3 t−2 t−1 t t+1 t+2 t+3


This �gure provide the coe�cients and con�dence bands for a single model for each outcome based on theinclusion of variables for three years of lagged and lead Parent Incarcerated and Parental Arrest variablesfor a given student and year. Estimation incorporates my main model with student �xed e�ects and othercontrols included in Table 3.

33


Figure 8: Coe�cient Plot over Time - Behavior Index

−.1

2−

.1−

.08

−.0

6−

.04

−.0

20

.02

.04

.06

.08

.1

Est

imat

ed E

ffect

s ov

er T

ime

t−3 t−2 t−1 t t+1 t+2 t+3


This �gure provide the coe�cients and con�dence bands for a single model for each outcome based on theinclusion of variables for three years of lagged and lead Parent Incarcerated and Parental Arrest variablesfor a given student and year. Estimation incorporates my main model with student �xed e�ects and othercontrols included in Table 3.

34


Figure 9: Interaction with Days Incarcerated - Test Scores

−.3

−.2

−.1

0.1

.2.3

< 10

day

s

10−2

0 da

ys

21−3

0 da

ys

1−3

mon

ths

3−6

mon

ths

6−12

mon

ths

1−2

year

s

2+ ye

ars

Days Incarcerated

This �gure provides the results of the main estimation model with student �xed e�ects, parental arrestvariables and other controls included in Table 3 where I disaggregate the incarceration variable intodi�erent intervals of total days incarcerated for the parental incarceration for student i in time period tand plot the coe�cients.

35


Figure 10: Interaction with Days Incarcerated - Behavior Index

−.4

−.3

−.2

−.1

0.1

.2.3

< 10

day

s

10−2

0 da

ys

21−3

0 da

ys

1−3

mon

ths

3−6

mon

ths

6−12

mon

ths

1−2

year

s

2+ ye

ars

Days Incarcerated

This �gure provides the results of the main estimation model with student �xed e�ects, parental arrestvariables and other controls included in Table 3 where I disaggregate the incarceration variable intodi�erent intervals of total days incarcerated for the parental incarceration for student i in time period tand plot the coe�cients.

36


Table 1: Summary Statistics - Individuals

All Students Students w/ ParentsEver Arrested

Mean Std Dev N Mean Std Dev N

Student AttributesMale 0.51 (0.50) 957,611 0.51 (0.50) 122,747Black 0.43 (0.50) 957,611 0.66 (0.48) 122,747Hispanic 0.12 (0.33) 957,611 0.08 (0.28) 122,747Non-Apartment Complex Home 0.76 (0.43) 938,802 0.74 (0.44) 120,629CBG Pop Density (000s/sq mile) 2.25 (1.87) 938,802 2.62 (2.01) 120,629CBG Median HH Income ($000s) 56.83 (24.78) 938,802 45.99 (20.09) 120,629CBG Percent Unemployed 3.37 (3.20) 938,802 4.26 (3.89) 120,629Number of Unique Students 1.00 (0.00) 194,163 1.00 (0.00) 18,858

Parent Ever Arrested 0.13 (0.33) 938,802 1.00 (0.00) 120,629Parent Ever Incarcerated 0.03 (0.18) 938,802 0.26 (0.44) 120,629Mother Ever Arrested 0.05 (0.22) 938,802 0.38 (0.49) 120,629Father Ever Arrested 0.09 (0.28) 938,802 0.69 (0.46) 120,629Parent Arrested 0.02 (0.14) 938,802 0.16 (0.36) 120,629Parent Incarcerated 0.01 (0.08) 938,802 0.05 (0.21) 120,629

Outcome VariablesEducation Index 0.00 (1.00) 824,864 –0.17 (1.14) 110,582Read Test Score 0.01 (1.00) 539,058 –0.35 (0.94) 73,022Math Test Score 0.01 (1.00) 553,412 –0.36 (0.90) 75,513Repeat Grade Next Year 0.04 (0.20) 768,698 0.06 (0.24) 104,665Behavior Index 0.00 (1.00) 957,611 0.29 (1.35) 122,747Days Absent 7.65 (9.78) 957,611 10.21 (12.71) 122,747Days Suspended from School 0.97 (3.98) 957,611 1.87 (5.71) 122,747Total School Reported Crimes 0.51 (1.69) 456,681 0.93 (2.32) 54,299Changed Residence Next Year 0.18 (0.38) 740,042 0.27 (0.44) 100,975Next Year Left CMS 0.10 (0.29) 836,707 0.06 (0.24) 109,706

Means and standard deviations are reported above.The data sample consists of an unbalanced panel of students observed during the 1998/1999-2010/2011

school years. I restrict the sample to only individuals born between 1990-2001 that attend a publicschool in Mecklenburg County, NC between 1999-2011.

37


Table 2: Regression - Incarceration Models(1) (2) (3) (4) (5) (6) (7)

Average Test ScoresParent Arrested –0.039*** –0.026*** –0.024*** –0.005 –0.008 –0.011*

(0.009) (0.009) (0.009) (0.006) (0.005) (0.006)Parent Incarcerated –0.068*** –0.045** –0.035* 0.023** 0.022** 0.024**

(0.018) (0.019) (0.019) (0.012) (0.011) (0.011)

Observations 74,652 74,652 74,652 74,652 74,652 74,652 74,652

Behavior IndexParent Arrested 0.119*** 0.114*** 0.109*** 0.061*** 0.063*** 0.072***

(0.012) (0.012) (0.012) (0.011) (0.012) (0.013)Parent Incarcerated 0.116*** 0.015 0.009 –0.046* –0.050* –0.066**

(0.024) (0.026) (0.026) (0.026) (0.027) (0.026)

Observations 120,629 120,629 120,629 120,629 120,629 120,629 120,629Basic Controls X X X X X X XCBG FE - - - X - - -Family FE - - - - X - -Individual FE - - - - - X XAdditional Controls for Arrests - - - - - - X

Basic controls include individual covariates for race, gender, stand-alone residence home as well asneighborhood attributes based on Census Block Group 2000 measures of population density, medianhousehold income and percent unemployed. All models also include birth year by age by school year�xed e�ects. * p < 0.1, ** p < 0.05, *** p < 0.01. Standard errors robust to arbitrary correlation byindividual student. Additional Controls for Arrests includes dummies for arrests within each year for2 years prior through 2 years post a given student-year observation.

The data sample consists of an unbalanced panel of students observed during the 1998/1999-2010/2011school years. I restrict the sample to only individuals born between 1990-2001 that attend a publicschool in Mecklenburg County, NC between 1999-2011.

I restrict the sample to students that had a parental �gure (by birth or marriage) that was arrested for atleast one crime between 1998 and 2011 while living at the same residence as the student. Details onmatching parents to children is provided in the Data Section.

38


Table 3: Incarceration - All Outcomes(1) (2) (3) (4)

AverageTest

Scores

ReadTest

Score

MathTest

Score

RepeatGrade

Parent Arrested –0.011* –0.018** –0.008 0.002(0.006) (0.008) (0.007) (0.003)

Parent Incarcerated 0.024** 0.031** 0.022* 0.001(0.011) (0.014) (0.013) (0.006)

Dep. Var. (mean) -0.36 -0.35 -0.37 0.06Observations 74,652 71,980 74,433 102,914

BehaviorIndex

DaysAbsent

DaysSuspended

SchoolCrimes

Parent Arrested 0.072*** 0.886*** 0.137** 0.015(0.013) (0.125) (0.058) (0.049)

Parent Incarcerated –0.066** –0.435* –0.254** –0.123*(0.026) (0.255) (0.120) (0.074)

Dep. Var. (mean) 0.29 10.21 1.87 0.93Observations 120,629 120,629 120,629 53,394



All regressions include individual i �xed e�ects, birth year by grade by school year �xed e�ects, controlsfor arrests within a 2 years window, neighborhood controls for income, unemployment and popula-tion density and an indicator for stand-alone homes. * p < 0.1, ** p < 0.05, *** p < 0.01. Standard errorsrobust to arbitrary correlation by individual student.

39


Table 4: Incarceration - Adult Outcomes

HS Dropout HS DropoutNumber of

Arrests(age 16-18)

Number ofArrests

(age 16-18)

Parents Arrested (age 7-11) 0.047* 0.048* 0.071 0.069(0.026) (0.026) (0.134) (0.135)

Parents Incarcerated (age 7-11) –0.053** –0.055** –0.019 –0.019(0.025) (0.026) (0.154) (0.164)

Parents Arrested (age 12-17) –0.013 0.048(0.029) (0.198)

Parents Incarcerated (age 12-17) 0.004 –0.043(0.030) (0.203)

Dep. Var. (mean) 0.14 0.14 0.67 0.67Observations 2,310 2,310 2,310 2,310Basic Controls X X X XCBG FE X X X X

For these models, I limit my sample to student born between 1990-1995 and only include one observationper student given the outcomes of dropping out of HS at ages 17 or 18 as well as the number of arrestat ages 16-18. I cannot extend to later adult outcomes given data limitations and individual �xede�ects are not possible given that I include each student only once.

40


Table 5: Incarceration - Other Outcomes(1) (2) (3) (4) (5) (6) (7)

MovedNextYear

MovedDuringSchoolYear

LeftCMS

Next Year

LeftCMS, But

ReturnLaterYears

SwitchSchools

Non-Transition

Years

TakeEOGExam

ParentRearrest

NewAddress

Parent Arrested 0.020*** 0.011** –0.011*** –0.000 0.013** 0.010***(0.005) (0.004) (0.003) (0.001) (0.005) (0.003)

Parent Incarcerated –0.005 –0.014* 0.020*** 0.000 –0.008 –0.009 0.064***(0.010) (0.008) (0.005) (0.002) (0.010) (0.006) (0.013)

Dep. Var. (mean) 0.27 0.12 0.06 0.01 0.20 0.82 0.49Observations 100,975 120,542 107,868 107,868 96,192 87,467 7,378



All regressions include birth year by grade by school year �xed e�ects, controls for arrests within a2 years window, neighborhood controls for income, unemployment and population density and anindicator for stand alone homes. Columns 1 through 6 includes individual �xed e�ects while column7 does not contain individual �xed e�ects given that the sample is limited in size to only student-years with a parental arrest and a subsequent parental rearrest where both arrests contain geocodableaddress information. * p < 0.1, ** p < 0.05, *** p < 0.01. Standard errors robust to arbitrary correlationby individual student.

41


Tabl

e6:

Robu

stne

ssCh

ecks

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Mai

n

Incl

ude

Mul

tiple

Pare

ntM

atch

es&

All

Apt

s.

Incl

ude

All

Apt

s

Pare

ntM

atch

Addr

ess

Onl

y

Incl

ude

All

Inca

rcs

Incl

ude

Onl

yA

llIn

carc

s>3

0da

ys

Indi

vCo

ntro

ls+

coho

rtby

grad

eby

year

FEso

nly

(8)+

CBG

FEs

(9)+

Scho

olby

Grad

eFE

s

Aver

age

Test

Scor

esPa

rent

Arr

este

d–0

.011*

–0.01

1**

–0.01

4***

–0.00

3–0

.010

–0.00

7–0

.011*

–0.01

2**

–0.01

1*(0

.006)

(0.00

5)(0

.006)

(0.00

4)(0

.007)

(0.00

6)(0

.006)

(0.00

6)(0

.006)

Pare

ntIn

carc

erat

ed0.0

24**

0.018

**0.0

25**

*0.0

100.0

120.0

110.0

27**

0.028

***

0.025

**(0

.011)

(0.00

9)(0

.009)

(0.00

7)(0

.009)

(0.01

4)(0

.011)

(0.01

1)(0

.010)

Obs

erva

tions

74,65

210

0,612

95,85

814

7,480

74,80

974

,174

74,65

274

,652

74,65

2

Beha

vior

Inde

xPa

rent

Arr

este

d0.0

72**

*0.0

60**

*0.0

68**

*0.0

33**

*0.0

79**

*0.0

66**

*0.0

69**

*0.0

69**

*0.0

69**

*(0

.013)

(0.01

1)(0

.012)

(0.00

9)(0

.014)

(0.01

3)(0

.013)

(0.01

3)(0

.013)

Pare

ntIn

carc

erat

ed–0

.066*

*–0

.048*

*–0

.065*

**–0

.013

–0.06

7***

–0.06

5*–0

.074*

**–0

.075*

**–0

.060*

*(0

.026)

(0.02

1)(0

.022)

(0.01

5)(0

.022)

(0.03

6)(0

.026)

(0.02

6)(0

.025)

Obs

erva

tions

120,6

2916

4,341

156,5

5524

5,743

120,8

4511

9,794

120,6

2912

0,629

120,6

29

The

data

sam

ple

cons

istso

fan

unba

lanc

edpa

nelo

fstu

dent

sobs

erve

ddu

ring

the

1998

/199

9-20

10/2

011

scho

olye

ars.

Ires

tric

tthe

sam

ple

toon

lyin

divi

dual

sbor

nbe

twee

n19

90-2

001

that

atte

nda

publ

icsc

hool

inM

eckl

enbu

rgCo

unty

,NC

betw

een

1999

-201

1.Ir

estri

ctth

esa

mpl

eto

stud

ents

that

had

apa

rent

al�g

ure

(by

birth

orm

arria

ge)t

hatw

asar

rest

edfo

ratl

east

one

crim

ebe

twee

n19

98an

d20

11w

hile

livin

gat

the

sam

ere

siden

ceas

the

stud

ent.

Det

ails

onm

atch

ing

pare

ntst

och

ildre

nis

prov

ided

inth

eD

ata

Sect

ion.

Allr

egre

ssio

nsin

clude

indi

vidu

ali�

xed

e�ec

ts,b

irth

year

bygr

adeb

ysc

hool

year

�xed

e�ec

ts,c

ontro

lsfo

rarr

ests

with

ina

2ye

arsw

indo

w,ne

ighb

orho

odco

ntro

lsfo

rinc

ome,

unem

ploy

men

tand

popu

latio

nde

nsity

and

anin

dica

torf

orst

and-

alon

eho

mes

.*p

<0.1

,**p

<0.0

5,**

*p<

0.01.

Stan

dard

erro

rsro

bust

toar

bitr

ary

corr

elat

ion

byin

divi

dual

stud

ent.

42


Table 7: Incarceration - Heterogeneity by Release Type(1) (2) (3) (4) (5)

AverageTest

Scores

AverageTest

Scores

ReadTest

Score

MathTest

Score

RepeatGrade

Parent Incarcerated (current year) 0.021(0.014)

Incarcerated Bond Release 0.029 0.052** 0.007 0.000(0.019) (0.024) (0.024) (0.010)

Incarcerated Time Served/Parole –0.010 –0.012 –0.005 –0.009(0.023) (0.030) (0.028) (0.013)

Incarcerated Other Release 0.014 0.038 –0.002 0.013(0.019) (0.026) (0.023) (0.011)

Incarcerated Release Future Year 0.036 0.075* –0.008 0.005(0.030) (0.038) (0.039) (0.017)

Dep. Var. (mean) -0.36 -0.36 -0.34 -0.36 0.08Observations 49,684 49,684 49,327 49,554 70,083

BehaviorIndex

BehaviorIndex

DaysAbsent

DaysSuspended

SchoolCrimes

Parent Incarcerated (current year) –0.086***(0.032)

Incarcerated Bond Release 0.003 0.041 0.010 –0.002(0.042) (0.445) (0.206) (0.094)

Incarcerated Time Served/Parole –0.065 –0.755 –0.294 –0.009(0.052) (0.529) (0.223) (0.123)

Incarcerated Other Release –0.041 0.156 –0.172 –0.134(0.042) (0.464) (0.196) (0.087)

Incarcerated Release Future Year –0.132* –1.171* –0.489 –0.277(0.076) (0.697) (0.323) (0.238)

Dep. Var. (mean) 0.37 0.37 10.84 2.23 0.93Observations 86,990 86,990 86,990 86,990 53,469

Release type are distributed across incarcerations with 28% released with a bond; 22% released withtime served/parole; 32% released for other reasons (transferred to another jurisdiction, case dismissed,weekender program - serve time on the weekends only, released to custody of another person, etc.);18% are released in a subsequent school year. We restrict models to the shorter time window of 2005-2011 due to the availability of speci�c information on the type of incarceration release which beganin 2005.




43

Preliminary Draft : November 6, 2017Ta

ble

8:Re

gres

sion

-Het

erog

enei

ty(1

)(2

)(3

)(4

)(5

)(6

)(7

)(8

)(9

)

Mai

nM

othe

rM

ale

Mov

edLe

ftCM

S

One

/Tw

oTi

me

Arr

est

Both

Pare

nts

Ever

Arr

est

Onl

yFa

mili

esw

/Inc

arc

Pare

nts

Onl

yBR

Veri�

ed

Aver

age

Test

Scor

esPa

rent

Arr

este

d–0

.011*

–0.01

4**

–0.01

6**

–0.01

0–0

.011*

–0.00

5–0

.009

–0.01

8–0

.021*

*(0

.006)

(0.00

7)(0

.008)

(0.00

7)(0

.006)

(0.00

7)(0

.006)

(0.01

3)(0

.009)

Pare

ntIn

carc

erat

ed0.0

24**

0.023

*0.0

30**

0.012

0.026

**0.0

26**

0.022

**0.0

33**

0.044

**(0

.011)

(0.01

2)(0

.014)

(0.01

4)(0

.012)

(0.01

3)(0

.011)

(0.01

5)(0

.018)

Pare

ntal

Arr

ests

*[co

l]0.0

090.0

110.0

000.0

03–0

.017*

–0.12

1(0

.010)

(0.01

1)(0

.010)

(0.01

5)(0

.009)

(0.08

0)Pa

rent

alIn

carc

*[co

l]0.0

15–0

.011

0.035

*–0

.009

–0.00

1–0

.095

(0.02

5)(0

.021)

(0.02

0)(0

.027)

(0.01

7)(0

.136)

Obs

erva

tions

74,65

274

,652

74,65

274

,652

74,65

274

,651

68,95

619

,400

31,14

9

Beha

vior

Inde

xPa

rent

Arr

este

d0.0

72**

*0.0

56**

*0.0

71**

*0.0

58**

*0.0

71**

*0.0

050.0

95**

*0.0

72**

0.068

***

(0.01

3)(0

.014)

(0.01

5)(0

.017)

(0.01

4)(0

.014)

(0.01

5)(0

.031)

(0.02

0)Pa

rent

Inca

rcer

ated

–0.06

6**

–0.03

8–0

.071*

*–0

.069*

*–0

.069*

*–0

.100*

**–0

.054*

*–0

.069*

–0.10

9**

(0.02

6)(0

.029)

(0.03

2)(0

.035)

(0.02

9)(0

.030)

(0.02

8)(0

.035)

(0.04

3)Pa

rent

alA

rres

ts*[

col]

0.041

0.003

0.033

0.011

0.196

***

–0.39

4***

(0.02

5)(0

.024)

(0.02

2)(0

.037)

(0.02

6)(0

.078)

Pare

ntal

Inca

rc*[

col]

–0.09

50.0

100.0

300.0

170.0

630.0

19(0

.068)

(0.05

3)(0

.050)

(0.07

3)(0

.056)

(0.14

2)

Obs

erva

tions

120,6

2912

0,629

120,6

2912

0,629

120,6

2912

0,610

107,8

6831

,538

49,01

3

Inte

ract

ion

term

sare

de�n

edby

thev

aria

blen

amed

inth

ecol

umn

head

ings

and

fora

llin

tera

ctio

nsre

sults

thec

olum

nhe

adin

gsre

pres

entd

umm

yva

riabl

esin

tera

cted

with

the

pare

ntal

arre

stan

din

carc

erat

ion

varia

bles

.Th

eda

tasa

mpl

eco

nsist

sofa

nun

bala

nced

pane

lofs

tude

ntso

bser

ved

durin

gth

e19

98/1

999-

2010

/201

1sc

hool

year

s.Ir

estr

ictt

hesa

mpl

eto

only

indi

vidu

alsb

orn

betw

een

1990

-200

1th

atat

tend

apu

blic

scho

olin

Mec

klen

burg

Coun

ty,N

Cbe

twee

n19

99-2

011.

Ires

trict

the

sam

ple

tost

uden

tsth

atha

da

pare

ntal

�gur

e(b

ybi

rthor

mar

riage

)tha

twas

arre

sted

fora

tlea

ston

ecr

ime

betw

een

1998

and

2011

whi

leliv

ing

atth

esa

me

resid

ence

asth

est

uden

t.D

etai

lson

mat

chin

gpa

rent

sto

child

ren

ispr

ovid

edin

the

Dat

aSe

ctio

n.Al

lreg

ress

ions

inclu

dein

divi

dual

i�xe

de�

ects

,birt

hye

arby

grad

eby

scho

olye

ar�x

ede�

ects

,con

trols

fora

rres

tsw

ithin

a2

year

swin

dow,

neig

hbor

hood

cont

rols

fori

ncom

e,un

empl

oym

enta

ndpo

pula

tion

dens

ityan

dan

indi

cato

rfor

stan

d-al

one

hom

es.*

p<

0.1,*

*p<

0.05,

***p

<0.0

1.St

anda

rder

rors

robu

stto

arbi

trar

yco

rrel

atio

nby

indi

vidu

alst

uden

t.

44


Table 9: Incarceration - Heterogeneity by Gender(1) (2) (3) (4)

AverageTest

Scores

ReadTest

Score

MathTest

Score

RepeatGrade

Father Incarcerated * Daughter 0.030** 0.045** 0.021 –0.000(0.015) (0.019) (0.018) (0.008)

Mother Incarcerated * Daughter 0.020 0.035 0.023 –0.010(0.028) (0.038) (0.033) (0.016)

Father Incarcerated * Son 0.012 0.017 0.010 0.006(0.017) (0.021) (0.019) (0.009)

Mother Incarcerated * Son 0.061* 0.041 0.069* –0.011(0.034) (0.047) (0.039) (0.017)


BehaviorIndex

DaysAbsent

DaysSuspended

SchoolCrimes

Father Incarcerated * Daughter –0.058* –0.499 –0.115 –0.162*(0.034) (0.335) (0.153) (0.094)

Mother Incarcerated * Daughter –0.126* –0.978 –0.606** 0.139(0.068) (0.765) (0.241) (0.168)

Father Incarcerated * Son –0.057 –0.396 –0.255 –0.202*(0.042) (0.391) (0.196) (0.111)

Mother Incarcerated * Son –0.116 –0.199 –0.586 0.034(0.103) (0.996) (0.475) (0.258)





45


Table 10: Incarceration - Heterogeneity by Crime Type(1) (2) (3) (4)

AverageTest

Scores

ReadTest

Score

MathTest

Score

RepeatGrade

Incarcerated Property Crime 0.019 0.024 0.025 –0.002(0.023) (0.028) (0.026) (0.012)

Incarcerated Violent Crime 0.032* 0.029 0.024 –0.002(0.018) (0.023) (0.021) (0.010)

Incarcerated Drug/Alc Crime 0.002 0.009 0.014 0.011(0.021) (0.028) (0.025) (0.011)

Incarcerated Other Crime 0.023 0.017 0.037 0.005(0.019) (0.025) (0.023) (0.010)

Incarcerated Technical Violation 0.011 0.027 –0.017 0.005(0.031) (0.042) (0.038) (0.018)


BehaviorIndex

DaysAbsent

DaysSuspended

SchoolCrimes

Incarcerated Property Crime –0.065 –0.290 –0.390* –0.280*(0.056) (0.585) (0.228) (0.153)

Incarcerated Violent Crime –0.070* –0.735* –0.228 –0.222*(0.043) (0.396) (0.195) (0.133)

Incarcerated Drug/Alc Crime –0.064 –0.665 –0.264 0.127(0.067) (0.556) (0.319) (0.176)

Incarcerated Other Crime –0.016 0.292 –0.214 –0.097(0.045) (0.478) (0.199) (0.116)

Incarcerated Technical Violation –0.023 0.479 –0.166 –0.088(0.067) (0.705) (0.296) (0.144)





46


Table 11: Incarceration - Heterogeneity for Young Children(1) (2) (3) (4)

AverageTest

Scores

AverageTest

Scores

BehaviorIndex

BehaviorIndex

Any Parent Incarc. 0.021* 0.022* –0.051* –0.053*(0.011) (0.011) (0.028) (0.027)

Mom Incarc. & Young Child 0.025 –0.149**(0.026) (0.069)

Mom Incarc. & Young Child & Vio. Crime –0.013 –0.178*(0.040) (0.091)

Mom Incarc. & Young Child & Prop. Crime 0.033 –0.155(0.043) (0.113)

Mom Incarc. & Young Child & Other. Crime 0.007 0.082(0.068) (0.139)

Mom Incarc. & Young Child & Drug/Alc Crime 0.056 –0.149(0.050) (0.124)

Any Parent Incarc. 0.019 0.021* –0.037 –0.039(0.013) (0.012) (0.033) (0.031)

Dad Incarc. & Young Child 0.013 –0.087**(0.018) (0.040)

Dad Incarc. & Young Child & Vio. Crime 0.027 –0.066(0.024) (0.056)

Dad Incarc. & Young Child & Prop. Crime 0.016 –0.040(0.033) (0.068)

Dad Incarc. & Young Child & Other. Crime 0.010 –0.052(0.028) (0.058)

Dad Incarc. & Young Child & Drug/Alc Crime –0.037 –0.192***(0.031) (0.065)

Observations 74,652 74,652 120,629 120,629

This table provides results for mothers versus fathers and di�erent crime types for younger aged children(age 7-11) at the time of parental arrest and incarceration.




47


A. Appendix

A.1. Parental Matching �ality

In order to get a sense of the portion of likely criminal parents I am able to match to thestudent database, I created Appendix Figure A1. Appendix Figure A1 provides matchrates of arrest and student records for each year relative to the estimated populationof children with arrested parents and highlights that I am able to match about 75% ofthe estimated population to the student records.37 The dotted line provides the share ofstudents that had a parent arrested in a given year. To create the solid line, I estimate thepopulation of arrestees with children using Census data from the American CommunitySurvey for the study area of Mecklenburg County, NC. The main assumption is thatadult arrestees have similar number of children as the overall population. I assume that16.8% of households have children age 6-17 and multiply this times the population ofadult arrestees of parental age (age 15-42 for women; 16-48 for men) from the arrestrecords. I then divided this estimate of parental arrestees by the number of students inthe population of student records. I conduct this for each year of overlapping studentand arrest records 1999-2011 and present this share in this �gure as the dotted line. Onewould not expect address matching to capture anything close to 100% of the estimatedpopulation because of the large prevalence of absentee fathers in this population ofincarcerated parents. Furthermore, the estimated population may even be too low ifparents involved in the criminal justice system have above average number of children.

Table A1: Parental MatchingStudent RecordsFirst Name Last Name Student Address School Year Arrest Incarc.John Williams 200 E 43rd Street 2001 0 0John Williams 200 E 43rd Street 2002 0 0John Williams 200 E 43rd Street 2003 1 1John Williams 200 E 43rd Street 2004 0 0John Williams 200 E 43rd Street 2005 1 0

Arrest RecordsFirst Name Last Name Address at Arrest Arrest Year Incarc. YearSam Williams 200 E 43rd Street 2003 2003Sam Williams 200 E 43rd Street 2007John Williams 100 N Broadway Ave. 2004 2004Mary Williams 200 E 43rd Street 2005

37This calculation is based on including children matched to more than one arrestee as well as children inlarge apartment complexes. Excluding these types of matches would bring the average closer to 55%.

48


Figure A1: Quality of Parental Arrest Match to Student Records0

12

34

56

% P

aren

ts A

rres

ted

2000 2005 2010Year

Estimated Population Match Sample

This �gure provides my sample match rates of arrest and student records for each year relative to theestimated population of children with criminal parents.

A.2. Additional Figures and Tables

49


Figure A2: Incarceration Experience

Arrest

< 3 Days in Jail 38%

No Incarceration54%

Incarceration46%

30 days – 1 year47%

<30 days35%

1 Incarceration50%

2+ Incarcerations/ Bond 12%

> 1 year18%

30 days – 1 year31%

< 30 days66%

> 1 year3%

This �gure provides the breakdown of arrest and incarceration experiences for individuals in the arrestand incarceration records for Mecklenburg County, NC.

50


Figure A3: Distribution of Crimes0

.1.2

.3.4

Sh

are

of

Crim

es

Assau

lt

Burglar

y

Dru

gs/A

lcoh

ol

Fraud

/For

gery

Kidna

pping/

Hom

icide

Larc

eny/The

ft

Minor

Rob

bery

Sex O

ffens

e

Wea

pons

Only Arrest Arrest & Incarc

This �gure provide the distribution of crime categories used in criminal charges across two groups:individuals only arrested and individuals arrested and incarcerated. We consolidate crimes into broadercategories with minor crimes mostly composed of crimes such as vandalism, disorderly conduct or othero�enses that do not harm individuals and are typically more public nuisance crimes.

Figure A4: Conditional Average Index Outcomes by Criminal Parent Type

A. Education IndexNo Arrests

1−2 Arrests


3+ Arrests

3+ Arrests Serious



−.15 −.1 −.05 0

B. Behavior IndexNo Arrests

1−2 Arrests


3+ Arrests

3+ Arrests Serious



0 .1 .2 .3 .4

Notes This �gure depicts mean outcomes by the di�erent types of criminal parents conditional on studentdemographics (e.g. race, gender, etc.), neighborhood attributes and �xed e�ects for birth cohort, grade andschool year.

51


Tabl

eA

2:Re

gres

sion

-Het

erog

enei

ty2

(1)

(2)

(3)

(4)

(5)

(6)

(7)

Mai

nM

othe

rM

ale

One

/Tw

oTi

me

Arr

est

Both

Pare

nts

Ever

Arr

est

Onl

yFa

mili

esw

/Inc

arc

Pare

nts

Onl

yBR

Veri�

ed

Mov

edAd

dres

sin

CMS

Pare

ntA

rres

ted

0.026

***

0.023

***

0.025

***

0.025

***

0.023

***

0.020

0.041

***

(0.00

5)(0

.006)

(0.00

7)(0

.007)

(0.00

6)(0

.013)

(0.00

8)Pa

rent

Inca

rcer

ated

–0.02

2**

–0.02

5**

–0.02

3–0

.025*

–0.01

8–0

.017

–0.03

7**

(0.01

0)(0

.011)

(0.01

4)(0

.013)

(0.01

1)(0

.015)

(0.01

7)Pa

rent

alA

rres

ts*[

col]

0.009

0.003

0.003

0.020

(0.01

0)(0

.010)

(0.01

0)(0

.015)

Pare

ntal

Inca

rc*[

col]

0.028

0.003

0.012

–0.02

3(0

.024)

(0.02

0)(0

.020)

(0.02

8)

Obs

erva

tions

120,6

1012

0,610

120,6

1012

0,610

120,6

1031

,533

49,01

0

Left

CMS

Pare

ntA

rres

ted

–0.01

1***

–0.01

0***

–0.01

5***

–0.00

5–0

.011*

**–0

.007

–0.01

4***

(0.00

3)(0

.003)

(0.00

4)(0

.003)

(0.00

3)(0

.005)

(0.00

4)Pa

rent

Inca

rcer

ated

0.020

***

0.017

***

0.015

**0.0

18**

*0.0

20**

*0.0

16**

*0.0

15*

(0.00

5)(0

.005)

(0.00

7)(0

.006)

(0.00

5)(0

.006)

(0.00

8)Pa

rent

alA

rres

ts*[

col]

–0.00

20.0

07–0

.015*

**0.0

00(0

.005)

(0.00

5)(0

.005)

(0.00

7)Pa

rent

alIn

carc

*[co

l]0.0

140.0

10–0

.002

0.002

(0.01

1)(0

.010)

(0.01

0)(0

.012)

Obs

erva

tions

107,8

6810

7,868

107,8

6810

7,868

107,8

6828

,282

43,74

0

Inte

ract

ion

term

sare

de�n

edby

thev

aria

blen

amed

inth

ecol

umn

head

ings

and

fora

llin

tera

ctio

nsre

sults

thec

olum

nhe

adin

gsre

pres

entd

umm

yva

riabl

esin

tera

cted

with

the

pare

ntal

arre

stan

din

carc

erat

ion

varia

bles

.Th

eda

tasa

mpl

eco

nsist

sofa

nun

bala

nced

pane

lofs

tude

ntso

bser

ved

durin

gth

e19

98/1

999-

2010

/201

1sc

hool

year

s.Ir

estr

ictt

hesa

mpl

eto

only

indi

vidu

alsb

orn

betw

een

1990

-200

1th

atat

tend

apu

blic

scho

olin

Mec

klen

burg

Coun

ty,N

Cbe

twee

n19

99-2

011.

Ires

trict

the

sam

ple

tost

uden

tsth

atha

da

pare

ntal

�gur

e(b

ybi

rthor

mar

riage

)tha

twas

arre

sted

fora

tlea

ston

ecr

ime

betw

een

1998

and

2011

whi

leliv

ing

atth

esa

me

resid

ence

asth

est

uden

t.D

etai

lson

mat

chin

gpa

rent

sto

child

ren

ispr

ovid

edin

the

Dat

aSe

ctio

n.Al

lreg

ress

ions

inclu

dein

divi

dual

i�xe

de�

ects

,birt

hye

arby

grad

eby

scho

olye

ar�x

ede�

ects

,con

trols

fora

rres

tsw

ithin

a2

year

swin

dow,

neig

hbor

hood

cont

rols

fori

ncom

e,un

empl

oym

enta

ndpo

pula

tion

dens

ityan

dan

indi

cato

rfor

stan

d-al

one

hom

es.*

p<

0.1,*

*p<

0.05,

***p

<0.0

1.St

anda

rder

rors

robu

stto

arbi

trar

yco

rrel

atio

nby

indi

vidu

alst

uden

t.

52

parental arrest, incarceration and the outcomes of their ... · ˙nd that only 42% of incarcerated...

Documents