The Impact of Family Income on Child Achievement: Evidence from the Earned Income Tax Credit

econstor www.econstor.eu

Der Open-Access-Publikationsserver der ZBW – Leibniz-Informationszentrum Wirtschaft The Open Access Publication Server of the ZBW – Leibniz Information Centre for Economics

Nutzungsbedingungen: Die ZBW räumt Ihnen als Nutzerin/Nutzer das unentgeltliche, räumlich unbeschränkte und zeitlich auf die Dauer des Schutzrechts beschränkte einfache Recht ein, das ausgewählte Werk im Rahmen der unter → http://www.econstor.eu/dspace/Nutzungsbedingungen nachzulesenden vollständigen Nutzungsbedingungen zu vervielfältigen, mit denen die Nutzerin/der Nutzer sich durch die erste Nutzung einverstanden erklärt.

Terms of use: The ZBW grants you, the user, the non-exclusive right to use the selected work free of charge, territorially unrestricted and within the time limit of the term of the property rights according to the terms specified at → http://www.econstor.eu/dspace/Nutzungsbedingungen By the first use of the selected work the user agrees and declares to comply with these terms of use.

zbw Leibniz-Informationszentrum WirtschaftLeibniz Information Centre for Economics

Akee, Randall K. Q.; Copeland, William; Keeler, Gordon; Angold, Adrian; Costello, E. Jane

Working Paper

Parents' incomes and children's outcomes: a quasi- experiment

IZA Discussion Papers, No. 3520

Provided in Cooperation with: Institute for the Study of Labor (IZA)

Suggested Citation: Akee, Randall K. Q.; Copeland, William; Keeler, Gordon; Angold, Adrian; Costello, E. Jane (2008) : Parents' incomes and children's outcomes: a quasi-experiment, IZA Discussion Papers, No. 3520, http://nbn-resolving.de/urn:nbn:de:101:1-20080605150

This Version is available at: http://hdl.handle.net/10419/34900

IZA DP No. 3520

Parents' Incomes and Children's Outcomes: A Quasi-Experiment

Randall K. Q. Akee William Copeland Gordon Keeler Adrian Angold E. Jane Costello

D I

S C

U S

S I

O N

P

A P

E R

S

E R

I E

S

Forschungsinstitut zur Zukunft der Arbeit Institute for the Study of Labor

May 2008

Parents’ Incomes and Children’s Outcomes: A Quasi-Experiment

Randall K. Q. Akee IZA

William Copeland Duke University

Gordon Keeler

Duke University

Adrian Angold Duke University

E. Jane Costello

Duke University

Discussion Paper No. 3520 May 2008

IZA

P.O. Box 7240 53072 Bonn

Germany

Phone: +49-228-3894-0 Fax: +49-228-3894-180

E-mail: [email protected]

Any opinions expressed here are those of the author(s) and not those of IZA. Research published in this series may include views on policy, but the institute itself takes no institutional policy positions. The Institute for the Study of Labor (IZA) in Bonn is a local and virtual international research center and a place of communication between science, politics and business. IZA is an independent nonprofit organization supported by Deutsche Post World Net. The center is associated with the University of Bonn and offers a stimulating research environment through its international network, workshops and conferences, data service, project support, research visits and doctoral program. IZA engages in (i) original and internationally competitive research in all fields of labor economics, (ii) development of policy concepts, and (iii) dissemination of research results and concepts to the interested public. IZA Discussion Papers often represent preliminary work and are circulated to encourage discussion. Citation of such a paper should account for its provisional character. A revised version may be available directly from the author.

IZA Discussion Paper No. 3520 May 2008

ABSTRACT

Parents’ Incomes and Children’s Outcomes: A Quasi-Experiment*

Identifying the effect of parental incomes on child outcomes is difficult due to the correlation of unobserved ability, education levels and income. Previous research has relied on the use of instrumental variables to identify the effect of a change in household income on the young adult outcomes of the household’s children. In this research, we examine the role that an exogenous increase in household incomes due to a government transfer unrelated to household characteristics plays in the long run outcomes for children in affected households. We find that children who are in households affected by the cash transfer program have higher levels of education in their young adulthood and a lower incidence of criminality for minor offenses. These effects differ by initial household poverty status as is expected. Second, we explore two possible mechanisms through which this exogenous increase in household income affects the long run outcomes of children – parental time (quantity) and parental quality. Parental quality and child interactions show a marked improvement while changes in parental time with child does not appear to matter. JEL Classification: J24, O12, H23 Keywords: cash transfer programs, quasi-experiment, educational attainment, criminality,

difference-in-differences, panel data Corresponding author: Randall Akee IZA P.O. Box 7240 D-53072 Bonn Germany E-mail: [email protected]

* We are grateful to participants at the University of Gottingen Development Seminar, IZA Brown Bag Luncheon, University of Hawaii Economics Department Seminar, Oxford University Center for the Study of African Economies Development Seminar, University of Nottingham Economic Development Seminar, University of Bristol and Chris Avery, Sonia Bhalotra, Lorenzo Cappellari, Ana Cardoso, Deborah Cobb-Clark, Eric Edmonds, Gary Fields, Ira Gang, Andrea Ichino, Lakshmi Iyer, David Jaeger, Erin Krupka, Mark Rosenzweig, Uwe Sunde, Kostas Tatsiramos, Arthur van Soest, Mutlu Yuksel, Anzelika Zaiceva and Zhong Zhao for helpful discussions and valued input. Any remaining errors, omissions or oversights are ours alone.

1 Introduction

Household conditions and characteristics certainly play a role in determining the outcomes of children. The strength and nature of that role has been an important research area for social scientists. One characteristic is of special im- portance for economists �household incomes. Does having more money in the household produce better child outcomes over time? Alternatively, does growing up in poverty produce worse outcomes for children? It is exceedingly di¢ cult to answer these questions because household incomes are not exogenously given. Income depends crucially on parental characteristics, both observed and un- observed. Therefore, simply observing that children from high (low) income families tend to have positive (negative) educational, income and employment outcomes in young adulthood tells us little about the actual causation. Par- ents transmit to their genetic o¤spring some of their innate abilities and the observed correlation between parental incomes and child outcomes later in life may simply re�ect this intergenerational transfer and not the e¤ect of income per se. Researchers have sought to overcome this endogeneity problem by using a

number of instrumental variables and �xed e¤ects techniques that attempt to isolate the di¤erence in household incomes that are not due to parental charac- teristics or ability. Using father�s union and occupational status as instruments for income, Shea (2000) �nds that income has no e¤ect on child outcomes while Chevalier et al. (2005) �nds that permanent income matters in children�s edu- cational attainment. Maurin (2002) uses grandparent socioeconomic status as a predictor of parental incomes which is then used to explain a child�s perfor- mance in early education. He �nds that a child is much less likely to be held back in school the higher the household income. Loken (2007) uses the Norwe- gian oil boom of the 1970�s and 1980�s, which only a¤ected a few regions of the country, as an instrument for increases in household income that is unrelated to parental characteristics. She �nds that there is no e¤ect of family income on child educational attainment. Mayer (1997) uses household assets and child support payments as measures of household income (these are taken to be less closely related to parental characteristics) and she �nds that income has a pos- itive and signi�cant e¤ect on educational attainment and wages. Blau (1999) uses child �xed e¤ects in the NLSY data and �nds that parental income (at least the transitory component) does not a¤ect child test scores. Previous research has found con�icting results with regard to the e¤ect of

household income on the young adult outcomes of household children. None of those studies have been able to identify a truly exogenous income shock at the household level. Our approach attempts to overcome the standard household income endogeneity problem in a direct manner - we observe households where incomes are increased exogenously and permanently through a governmental transfer program without regard to parental human capital, ability or other household characteristics. In our study, the increase in incomes is community- wide. We follow children that reside in households with and without exogenously increased incomes. The children are sampled in three age cohorts. The youngest

2

children reside as minors in households with higher incomes for a longer period of time than the oldest children in this study. We compare outcomes from the youngest age cohort to the oldest age cohort to determine the e¤ect of residing in a household with exogenously higher incomes. The children from households without additional household income serve as a control for any changes in local labor market opportunities that may have arisen between the age cohorts. Our study uses data from the Great Smoky Mountains Study of Youth

(GSMS). In this longitudinal study of child mental health in rural North Car- olina, both American Indian and non-Indian children were sampled. Halfway through the data collection, a casino opened on the Eastern Cherokee reserva- tion. A portion of the pro�ts from this new business operation is distributed every six months on an equalized, per capita basis to all adult tribal mem- bers regardless of employment status, income or other household characteristics. Non-Indian households are not eligible for these cash disbursements. Figure 1 provides a clear depiction of the change in household incomes over the �rst eight survey waves of our study. A marked increase is noted in the number of households with incomes above $30,000 for the treatment (American Indian) households after the disbursement of casino payments in 1997.1 No long-run change is observed for non-Indians households. We �nd that children who reside the longest in households with exogenously

increased incomes tend to do better later in life on several outcome measures. The children in these households are more likely to have graduated from high school by age 19; by age 21 the children from the poorest households have al- most an additional year of schooling. Additionally, we �nd, using administrative records on criminal arrests, that these same children have statistically signi�- cantly lower incidence of criminal behavior for minor o¤enses. These children also self-report that they have a lower probability of having dealt drugs than children from households una¤ected by the additional income. As expected, the poorest households in the survey experience the largest

gains in terms of child outcomes. Separating the data according to prior poverty status, we �nd that results are driven primarily by changes in the poorer house- holds. There are numerous mechanisms that may translate higher household in-

comes intobetter childoutcomes. Weexplore twopotentialmechanism: parental quality and parental quantity. The additional income may allow the poorer households to substitute away from full-time employment towards part-time employment thus allowing for more child care. This does not appear to happen in our data; parents do not reduce their working time and we �nd some evi- dence that they may actually intensify their labor e¤orts. On the other hand, we �nd that parental interactions and experience with the children in the af- fected households tends to improve dramatically. Both child and parent report improved behavioral e¤ects and parent-child interactions relative to una¤ected

1We use the percentage of households by group (American Indian vs. non-Indian) that have household incomes greater than $30,000. This corresponds to the median value of non- Indian households in the survey wave 3 which was just prior to the opening of the casino.

3

households. We observe that parent behavior, similar to those of the child, tend to improve with regard to criminality and drug use.2 Previous research has found a direct relationship between poverty and parenting ability (McLeod, 1993; Sampson, 1994; Ennis, 2000) and we con�rm this result in our research. There is at least some indication that one of the mechanisms responsible for translating higher household incomes into better child outcomes is through in- creased parental quality; while parenting time does not appear to have been an important causal factor. The next section describes the data from the Great Smoky Mountains Study

of Youth and our empirical methods. Section III provides our estimation results. We explore some potential mechanisms which may play a role in translating increased incomes into better child outcomes in Section IV. Section V concludes.

2 The Great Smoky Mountains Study of Youth, Empirical Methods and Data Description

The Great Smoky Mountains Study of Youth (GSMS) is a longitudinal survey of 1420 children aged 9, 11 and 13 years at intake that were recruited from 11 counties in western North Carolina. The children were selected from a popula- tion of approximately 20,000 school-aged children using an accelerated cohort design.3 American Indian children from the Eastern Band of Cherokee Indians were over sampled for this data collection e¤ort, survey weights are used in the child outcome regressions that follow. The federal reservation is situated in two of the 11 counties within the study. The initial survey contained 350 In- dian children and 1070 non-Indian children. Proportional weights were assigned according to the probability of selection into the study; therefore, the data is representative of the population. Attrition and non-response rates were found to be equal across ethnic and income groups. The survey began in 1993 and has followed these three cohorts of children

annually up to the age of 16 and then re-interviewed them at ages 19 and 21. Additional survey waves are scheduled for these children when they turn 24 and 25 years old. Both parents and children were interviewed separately up until the child was 16 years old; interviews after that were only conducted with the child alone. After the fourth wave of the study, a casino was opened on the Eastern

Cherokee reservation. The casino is owned and operated by the tribal govern- ment. A portion of the pro�ts are distributed on a per capita basis to all adult tribal members.4 Disbursements are made every six months and have occurred

2Similar results were found in the Moving to Opportunity program (Kling, et. al, 2007; Kling, et. al, 2005). In this case, low-income households were given the means to move into lower poverty neighborhoods. Incidence of mental illness decreased for parents and youth. Additionally, in previous research utilizing the GSMSY data, Costello, et al. (2003) found decreased mental illness for children from households that were lifted out of poverty as a result of the casino income.

3See Costello, et. al (1996) for a thorough description of the original survey methodology. 4All adult tribal members received these per capita disbursements. If there were any

4

since 1996. The average annual amount per person has been approximately $6000. This income is subject to the federal income tax requirements.

2.1 Empirical Speci�cations

We compare young adult outcomes for children that resided for a total of six years as minors in households with increased incomes to children who resided for just two years as minors in households with increased incomes. Essentially the older children (initially 13 year old cohort) serve as a control group for the younger children (initially 9 year old cohort) in this study. This speci�cation allows us to compare the e¤ect of four additional years of higher household incomes on young adult outcomes for these children. The data also contains a middle age cohort (initially 11 year old age cohort) which allows us to test whether or not two additional years of higher household incomes have an e¤ect on young adult outcomes. The size of the exogenous increase in household incomes can take on two

di¤erent values depending upon the number of American Indian parents in each household. It is possible for there to be 0,1 or 2 American Indian parents in each household.5 Clearly households with two American Indian parents will have double the amount of exogenous income than households with only a single American Indian parent. Households without an American Indian parent serve as a control household. These control households are also representative of the entire income distribution �both rich and poor households are represented in this group. We employ a methodology based on a di¤erence-in-di¤erences methodology.

In our case, we employ the two youngest age cohort variables (Age 9 and Age 11) which function as the �after-treatment� cases and the oldest age cohort (Age 13) functions as the �before-treatment�case. The number of American Indian parents in the household (NumParents) serves to distinguish between the control and test groups. We treat the number of parents as a continuous variable and we therefore have two interaction variables which are of interest. The equation below details the speci�cation:

(1) Yi = � + �1 � Age9i + �2 � Age11i + � � NumParentsi + 1 � Age9i � NumParentsi + 2 �Age11i �NumParentsi +Xi�� + �i

In the equation above, Y is the outcome variable of interest for the child at ages 19 or 21. We will examine educational attainment, high school completion

non-compliers (American Indian parents that either did not receive or refused the additional income) then any estimates found here would be an under estimate of the true e¤ects of additional income.

5In some cases, the biological parent does not live in the same household. In these cases, while the child is not necessarily living in a household with the additional income, he or she still has a parent with exogenously increased income. The inclusion of these households should actually reduce the e¤ect of household incomes on child outcomes if there is no direct e¤ect of the additional income for non-resident parents on their children. We have excluded these households and �nd that in general while the sample size is reduced and standard errors increase, the results tend to hold for most of the reported outcomes.

5

variables and measures of criminal arrests at various ages. In the equation above, the Age9 and Age11 variables indicate whether or not the child is drawn from the initially age 9 or age 11 cohorts respectively �the age 13 cohort is the omitted category in this regression. The variable NumParents indicates the number of American Indian parents in that child�s household. The two coe¢ cients of interest for this research are and , which measure the e¤ect of receiving the casino disbursements and being in either the age 9 or age 11 cohorts relative to the 13 year old cohort and not receiving any household casino disbursements. The vector X controls household conditions prior to the opening of the casino and includes household poverty status, average household income over the four years, the sex of the child, the race of the child and education levels of both parents. Survey weights are employed in all of these di¤erence in di¤erence regressions. Appendix I provides robustness tests, where possible, for the outcome variables described above. Additionally, robustness tests are provided for changes in parental outcomes as well. Identi�cation of equation 1 relies on the fact that the di¤erent age cohorts

of children were randomly sampled within American Indian and non-Indian groupings. The next section provides evidence for this fact and also indicates that the two groups of households (American Indian and non-Indian) faced similar conditions in the labor market and with regard to social conditions. It is also important to note that there were no new health or educational programs which were created immediately after the advent of casino disbursements by the tribal government. This is important in establishing the fact that time variant characteristics that were related only to American Indians (such as tribally- funded anti-crime programs or tutoring programs) are not the causal factor here. In later years new programs have been developed, but for the crucial period in which these children were minors in their parents�households, there is little evidence of any new programs. An additional important point is that the e¤ect of this new industry, casino gambling, may have a rather large e¤ect on the demand for labor in the local labor market. This increase in demand may a¤ect the long-run aspirations, discount rates and human capital investment for children in the community. I control explicitly for this by using distance of the household to the casino. Using global positioning data (GPS) I compute a distance measure and �nd that inclusion of this measure does not diminish the e¤ects reported in later tables. The distance measure is meant to capture the increased likelihood of a household which is located in close proximity to be a¤ected through the labor demand e¤ects than a household located further away from the casino. Given the panel nature of the data, we are also able to utilize individual

�xed e¤ects for one of the outcome variables �child�s school attendance. This educational measure is meaningful at various points throughout the child�s life, not just at young adulthood as is the case with the other educational attainment measures. Therefore, we employ a �xed e¤ects regression for the number of days a child is present at school in the last three months prior to the interview. The regression is given of the form:

6

(2) Yit = X0it� +�0 +�i + �it

In this regression, �i is the individual �xed e¤ect and X is the vector of control variables, including whether the individual child, i, belongs to a house- hold that is eligible for casino payments. This indicator variable is always zero for households without American Indian parents; for households with Ameri- can Indian parents the variable is zero for the �rst four survey waves and then take the value of one thereafter. We employ a similar model when testing for changes in parental employment status, drug use, arrest, and relationship with their children in the second half of the paper which investigates the mechanisms through which additional household income a¤ects young adult child outcomes.

2.2 Data Description

Table 1 provides the means for the data used in this analysis by the type of household. The �rst panel provides the variables used primarily in the di¤er- ence in di¤erence regressions, while the second panel provides the data used in the �xed-e¤ects regressions. In panel A, the �rst set of columns provides the means and standard deviations for the households with at least one American Indian parent and the second panel contains the means and standard devia- tions for households that do not have any American Indian parents. It is worth noting that children from households with at least one American Indian parent have statistically signi�cantly di¤erent educational attainment on average as compared to children from households with no American Indian parents.6 On all measures, children from the �rst type of household have lower recorded ed- ucational attainment or completion. Interestingly, there is almost no di¤erence in the drug or alcohol use between children from these two types of households at age 21. This result stands in stark contrast to the results for their parents. Fathers from households with at least one American Indian parent have almost twice the incidence of drug and alcohol abuse (9% versus 5%) of households with no American Indian parents. The next group of variables indicates the distribution among the di¤erent age

cohorts and the number of American Indian parents. There is a slightly higher proportion of children found in the 9 year old age cohort for the American Indian parent household than for the non-Indian parent household �but this di¤erence is not statistically signi�cant. The second age cohort is much closer in number distribution between the two types of households. The number of American Indian parents and the interaction terms di¤er between the two household types by design. The third set of variables provides a look at the household conditions prior to

the opening of the casino for both groups of children. There are level di¤erences between all of the initial household conditions except for the gender distribution for children from both types of households. Children from households with at

6The other races in this data set are White and African-American. The African-American children make up less than 6% of the total observations; therefore, using Non-Indians refers to these two other groups but Whites make up the highest proportion of that group.

7

least one American Indian parent are much more likely to be American Indian than from the households with no American Indian parents; however, there are a few cases where American Indian children reside in households without their biological American Indian parents. The parental education variables, unlike the education measures for the child, are given in categories not in years. The value for the �rst parent (3.95) from a household with at least one American Indian parent corresponds to approximately a high school diploma.7 While statistically di¤erent, the actual di¤erence in years of educational content is very small on average. The second parent�s educational level di¤ers on average between the categories of �some high school�and �GED or high school equivalency�for the two types of households. Finally, the last two variables provide insight into the economic conditions of the households. On average, households with at least one American Indian parent have spent at least one year in poverty in the �rst three years of the study while the �gure is 0.66 years for the households with no American Indian parents. Income is also given in categories and the value of 4.58 corresponds to an annual income between $15,001 and $20,000. For households with no American Indian parents, the average household income value of 6.65 falls in the $25,001 to $30,000 annual income category. The �nal set of variables in this panel provide the criminal activity of the

sample children. These data are gathered independently from the GSMS data. Searches of public databases in the North Carolina Administrative O¢ ce of the Courts produced these data. All counties in North Carolina are covered by these data including arrests made on the American Indian reservation. Arrests after the 16th birthday fall under the jurisdiction of the adult criminal justice system. Arrest records were found for juvenile arrests with the permission of the juvenile court judges. We have classi�ed the arrest records into three broad categories: minor arrests which includes arrests for disorderly conduct, trespassing and shoplifting; moderate arrests which are primarily property crimes that do not involve serious harm to a person such as simple assault, felony larceny and drug- related o¤enses; violent arrests which include sexual assault, armed robbery and assault with deadly weapons. The �rst set of variables reports whether a child has committed any crime in the years indicated. The categories are not cumulative and are independent of one another. Therefore, we see that a child from a household with at least one American Indian parent had a 10 percent chance of committing any type of crime (minor, moderate, violent) between the ages 16-17. A child from an American Indian household had a 17% chance of committing any type of crime between the ages 18-19. The next set of variables measures whether a child has committed any crime by age 21 by arrest category. The �rst variable indicates that a child from a household with at least one American Indian parent had a 25 percent chance of having committed a minor crime by age 21, while the same �gure for a household with no American Indian parents was 29 percent. Interestingly, children from American Indian households are less likely to have been arrested for all crimes across the board

7Category 3 is a GED or high school equivalency; Category 4 indicates having a high school diploma in this data; Category 5 indicates some post-high school training or vocational education.

8

and statistically sign�cantly less (at the 10% level) for moderate crimes by age 21. The �nal variable is found within the GSMS survey and indicates the child�s self-reported drug dealing behavior at each survey wave. The mean of this variable indicates that children from both types of household report having dealt drugs in 6 percent of the time. Panel B of Table 1 provides the data used primarily in the �xed-e¤ects

regressions for changes in parental behavior. The �rst variable gives the number of days the child was present in school in the last quarter. This question is asked at every survey wave while the child is less than 18 years old. There is no statistically signi�cant di¤erence between children in the two types of households. The next set of variables provides characteristics of the mother at each stage

over the survey time period. This variable is coded 1 for individuals who are in the labor force (working outside of the home) and 0 otherwise. There is no statistically signi�cant di¤erence between the labor force participation of mothers by household type. Labor force attachment is a categorical variable which measures (on a scale of 0 to 4) an individual�s degree of involvement in the labor force: a zero indicates full time employment, one indicates part- time employment, two indicates currently unemployed, three indicates work only in the home, while four indicates no work whatsoever (student, retired or disabled). For mothers it does not appear that there is any di¤erence in the attachment to the labor force. For mothers who are working, they tend to be less than full-time employed. The next variable indicates whether an individual has no drug alcohol problems (coded 0), a single problem (coded 1) or a combination of the two (coded 2). Mothers from households with at least one American Indian parent have a statistically signi�cantly higher incidence of these types of problems. Arrest status is simply an indicator variable for whether the mother was arrested since the last survey wave. Once again there is a statistically signi�cant di¤erence here. The child supervision variable measures the adequacy of parental supervision of their child. There are three options here: a zero indicates that the parent has age appropriate supervision or control over the child; the next option indicates that the parent does not have adequate control at least once a week; the �nal option indicates that the parent does not have adequate control at least �fty percent of the time or more. The �nal variable is a measure of the percentage of parent-child activities and interactions that are categorized as enjoyable by the child at each survey wave. The three options possible here are: a zero indicates that at least 75% of all activities are enjoyable; the next option indicates that between 25% and 74% of all activities are a source of tension, worry or disinterest to the child; the �nal option indicates that less than 25 % of all activities with the parent are enjoyable to the child. We observe that there is no statistically signi�cant di¤erence between household types for these last two variables. The results for fathers are presented in the next section. There is a statistically signi�cant di¤erence for fathers by type of household for labor force participation, labor force attachment, drug or alcohol problems and arrest status. Table2presentsacomparisonof these initialhouseholdcharacteristicsbyage

9

cohort for each of the two types of households. This table provides information on the suitability of the third age cohorts to serve as controls for the two other age cohorts in this study. In this table, t-statistics are presented for a test of a meandi¤erencebetweenthe indicatedagecohorts foragivenvariable. In the top panel of Table 2 we show the di¤erences in age cohorts for households that have no American Indian parents. There are statistically signi�cant di¤erences in the number of American Indian children in these households for cohorts 2 and 3 (age 11 and age 13 initially) and cohorts 1 and 3 (age 9 and age 13 initially). The di¤erence is driven by the relatively large amount of American Indian children in the third age cohort (7%). There is no di¤erence in the gender distribution for any of the three cohorts. We do �nd a statistically signi�cant di¤erence between cohorts 1 and 3 in the �rst parent�s educational attainment. This di¤erence, while statistically signi�cant, is not large in absolute magnitude (4.5 vs. 5). In qualitative terms the di¤erence is having a high school diploma versus having completed some post-high school (non-college) training. We observe little di¤erence in education levels for the second parent by age cohorts. We do �nd statistically signi�cant di¤erences for household income levels for cohorts 1 and 2 as well as for cohorts 1 and 3. The mean di¤erence between income categories is very small here 0.7 and 0.6 for each respectively. Each income category represents a step of $5,000 each. Therefore, the di¤erence represented here on average is between $3,000 - $3,500 per year. The bottom part of Table 2 provides a similar analysis for the households

with at least one American Indian parent. There appears to be very little dif- ferences between these age cohorts. The one statistically signi�cant di¤erence is found for the second parent�s educational level for cohorts 2 and 3. The mean values for each cohort is 1.7 and 2.7 which indicates a qualitative di¤erence of �some high school�and �GED or high school equivalency�. In sum, it appears that the data are reasonably similar across age cohorts for both types of house- holds. While there are some statistically signi�cant di¤erences, the magnitude of these di¤erences for most variables is in fact quite small. Finally, we provide some evidence on the similarity of the time trends of

the two types of households in the time period prior to the opening of the casino. It is not, of course, possible to show how the unobserved heterogeneity e¤ect evolves over time for the two types of households; however we do show that the households have similar trends in a number of dimensions. Figure 1 provides the trend in household incomes for the two types of households and we have already noted that there is a signi�cant di¤erence after the opening of the casino. However, prior to the opening of the casino, the growth in the percentage of households with incomes greater than $30,000 was similar between the two groups. Figures 2 and 3 show the changes in the unemployment rate for mothers and fathers respectively. Both �gures indicate that unemployment was generally decreasing and consistent for both household types. Figure 4 shows the di¤erence in reported incidence of alcohol or drug abuse problems for the second parent (reported by the �rst parent).8 The distance between the two

8We take the report of the �rst parent about the second parent�s drug and alcohol abuse

10

time trends decreases slightly between periods 1 and 2, but then is a relatively constant distance between waves 2 and 3. Taken together these �gures indicate that the two types of households, while

di¤ering in levels, appear to be equally a¤ected by the same social conditions, macroeconomic conditions and labor market experiences. The Eastern Chero- kee reservation is located in the middle of the eleven counties surveyed in this research. There is little evidence to support that the two household types are a¤ected di¤erently by changes at the local level in the period prior to the casino opening. Additionally, testing between the nature of household types across time, it

appears that there is not statistically signi�cant di¤erence in the composition of households across time. Appendix Table 1 provides t-tests of di¤erences between in marital status for the household types after the casino begins operations. The additional casino funds does not appear to a¤ect the marital status of couples included in this data. This �nding indicates that the casino payments are not creating incentives for the dissolution or the creation of new partnerships which may directly a¤ect the young adult outcome of children.

3 The E¤ects of Exogenous Change in Income on Young Adult Educational Attainment and Criminal Behavior

In this section, we present the results from the di¤erence-in-di¤erence regression described in equation 1 and the �xed-e¤ects regression described in equation 2. All of the results control for robust standard errors and employ survey weights. Where the outcome variables are indicator variables, we use a probit speci�ca- tion and report marginal coe¢ cients. For continuous outcome variables, such as years of education, we use a simple ordinary least squares regression for our analysis.9

3.1 Education Outcome Variables

Table 3 presents the results from regressions for the educational outcome vari- ables. The �rst column presents the regression of years of completed child�s education at age 21 on the level and interaction variables previously described. The two interaction variables presented in the �rst two rows indicate that there is a positive, but not statistically signi�cant, e¤ect of residing in a household

to be more accurate than the self-reported information about the �rst parent�s own drug and alcohol problems. There is reason to suspect that there would be problems with a self-reported measure of drug and alcohol abuse, but less so with regard to the other parent.

9In the following regressions, the sample sizes vary primarily because of missing information in the outcome variables. We take advantage of the maximum number of observations possible for each outcome variable and do not restrict our analysis to a smaller subset. Reducing the sample size does not appear to a¤ect the sign or magnitude of results, however, the standard errors do increase somewhat, as expected.

11

with exogenously increased incomes for six or four years relative to just two years. The other variables of interest in the regression are the parental educa- tion variables which are positive and statistically signi�cant as expected. The average household income variable is also positive and statistically signi�cant in this and the other two regressions as well. Column two presents the probability of a child being a high school graduate by age 19. The marginal coe¢ cient on the �rst interaction variable indicates that the e¤ect of having four more years of exogenously increased household income increases a child�s probability of �nishing high school by age 19 by almost 15 percent. The second interaction coe¢ cient is positive, but smaller in absolute magnitude and not statistically signi�cant. The third column outcome variable measures whether an individual has a high school diploma or a general equivalency degree. The �rst interaction coe¢ cient is once again positive but it only reaches statistical signi�cance at the 10% level.

3.2 Educational Outcome by Previous Poverty Status

We now investigate whether the exogenous increase in incomes has di¤ering im- pact by the prior poverty status of households. Table 4 presents the same analy- sis as Table 3, except that the sample has been divided according to whether the household has ever previously been in poverty prior to casino operation.10 We �nd in these �rst three regressions, for households previously in poverty, that the coe¢ cient on the �rst interaction term is always statistically signi�cant at the 5 % level and larger in magnitude than in Table 3. The coe¢ cient for the interaction variable for the years of education regression triples in size and im- plies that the treatment of four additional years of exogenously increased income increases educational attainment at age 21 by almost a full year (0.9 years).11

The coe¢ cient in the two high school graduation and GED regressions increases in magnitude and are highly statistically signi�cant. The next set of columns present the results from the subsample of households that were never previ- ously in poverty in the �rst three survey waves. None of the coe¢ cients on the interaction variables are statistically signi�cant. These results explain why the results for the full sample yielded statistically insigni�cant results for the years of education regression �the additional household income does not have a noticeable e¤ect in households not previously in poverty.

3.3 Education Outcome by Child Gender

We disaggregate the data in this section by the child�s gender in order to in- vestigate whether the additional household income has di¤erential impacts for boys or girls. Table 5 presents the same analysis for the educational outcome

10Using the US poverty levels adjusted for household size. 11Future survey waves will collect data on educational attainment when the children are 24

and 25 years old. This will allow for an additional look at the educational attainment as well as college completion rates.

12

variables. In the �rst set of columns, the sample is restricted to male chil- dren and the next set of columns present only the female children�s regressions. Examining the years of education regressions for each gender, it appears that females are reaping the most bene�ts from the exogenous increase in household income; the coe¢ cient is three times as large as that for the males and reaches statistical signi�cance at the 10% level.12 The same holds for the probability of high school graduation by age 19; females have a 21 % higher probability of �nishing high school on time when they reside in households with four more years of exogenously increased incomes. Interestingly, the reverse is true for the high school diploma or GED regressions. Males have a higher probability of receiving this type of educational attainment when they come from households with increased household incomes than the female children. A secondary check on a child�s educational achievement is a simple measure

of school attendance. Given that there is data on the number of days present in school in the past three months at each interview wave as reported by the primary parent, we can investigate whether this additional income a¤ects school attendance rates throughout childhood. We remove all time-invariant household characteristics (both observed and unobserved) and control for the time-varying characteristics directly in our �xed-e¤ects regression. Table 6 presents these �xed-e¤ects results; in the �rst column we regress the number of days present in school in the last three months on the household�s casino payment eligibility, household income, parental ages, child�s age and the number of children less than six years old in the household. The results indicate that casino payment eligibility increases school attendance by almost two and half days per quarter. Dividing the data once again by households that previously were in poverty we �nd that the e¤ect almost doubles in size: children from households with this additional income are present at school for almost four additional days than their untreated counterparts. The e¤ect persists, albeit smaller in magnitude and statistical signi�cance, for the households that previously were not in poverty. Overall, the additional household income appears to have a very strong e¤ect on the child�s school attendance.

3.4 Criminal Behavior during Young Adulthood

Table 7 examines the criminal behavior of all of the sample children. Adminis- trative data has been merged with the GSMS data at the individual level with information on the number and nature of each crime for all of the survey chil- dren. We classi�ed the arrests into three broad categories of minor, moderate and violent o¤enses. Additionally, information about when the arrests occurred allows us to identify the ages (16-21) of arrests for each person. The di¤erence in di¤erence regressions in Table 7 Panel A indicates that

children from households that receive casino payments are 22% less likely to

12Others have found that increasing household incomes in developing countries can have a di¤erential impact on children depending upon their gender; di¤erent household responsibil- ities along gender lines imply that additional income will change the composition of work or duties for the household children. See for instance Chen (2006).

13

have been arrested at ages 16-17 than their untreated counterparts. Examining the e¤ect on criminality in later years, speci�cally ages 18 -21, the additional household income has no direct e¤ect on criminal arrests for either �rst age cohort or the second age cohort. This result is somewhat puzzling but may be due to the fact that the children are no longer under their parents direct control after age 18. Therefore, the diversion in criminal behavior and arrests appears to be directly related to the child�s minor status.

3.5 Criminal Behavior During Young Adulthood by Gen- der and Previous Poverty Status

The second panel of Table 7 restricts the analysis to males and the reduction in criminal arrests at ages 16-17 are due primarily to the changes of the boys. The coe¢ cient on the �rst interaction term is larger in magnitude than in the full sample which includes the females and highly statistically signi�cant. Similar analysis for the girls alone results in insigni�cant results and is not reported here. We �nd that males from households with higher incomes at older ages, 18-21, do not di¤er systematically in their criminal arrests from households without the additional income. A �nal restriction, provided in the third panel of Table 7, divides the data by previous household poverty status and �nds that the children from previously poor households were the ones that had the largest reduction in criminal arrests at age 16 and 17. We �nd no results for children who come from households that were never previously in poverty; this held true for all the older ages as well (18-21).

3.6 Criminal Behavior During Young Adulthood by Type of Crime

Table 8 presents the e¤ect of additional household income on the child�s crimi- nal behavior by the type of crime committed. The �rst panel indicates that the reduction in criminal behavior occurs only in minor crimes. By age 21, a child who resided in a household with the additional casino income has a 16% lower probability of having ever committed a minor crime than a similar child from an untreated household. Further regressions that examined the e¤ect of addi- tional household income on the number of crimes (by category) did not yield signi�cant results. This indicates that the additional income a¤ected whether an individual entered into criminal behavior but not on the number of crimes once they had entered into criminality. Conducting a separate analysis for males alone, provided in the second panel of Table 8, we �nd that the results hold up for minor crimes, if slightly diminished in signi�cance, and become rather strong for moderate crimes. Consequently, we observe that women have a coe¢ cient that is positive for moderate crimes, but not statistically signi�cant at even the 15% level. The third panel of Table 8 shows thatpriorhousehold poverty status matters.

However, the direction di¤ers from that of previous results in that the household which were never in poverty are the ones that exhibit the largest changes. We

14

�nd that children from households which receive casino payments and were never previously in poverty had lower levels of minor crimes by age 21. The deterrent e¤ect of additional household income on crime appears to matter most for the previously wealthier households in our sample. A �nal measure of child criminal behavior is provided in Table 9. The

child�s self-reported drug dealing activities are regressed on the same set of explanatory variables used in the previous regression. The �rst interaction term indicates that children from households with exogenously increased incomes are 7% less likely to have reported dealing drugs at all in their youth. Restricting this to households that were previously in poverty, we �nd that the poorer households are driving the main results - the coe¢ cient changes very little but loses statistical signi�cance at the 5% level. Households that previously were not in poverty do not yield any statistically signi�cant results.

4 Potential Mechanisms

The previous section provided evidence that the exogenous increase in house- hold income has positively a¤ected young adult outcomes for children from these households. The results indicate that children from households with additional income have better educational attainment and reduced criminal behavior. In this section, we discuss a few of the potential mechanisms that may be con- tributing to the observed changes in child outcomes. There are several potential explanations for why increased incomes may af-

fect theyoungadult childoutcomes. Onepotential anddirect explanation is that the additional household income is used to purchase better quality educational inputs. Unfortunately, the data does not contain consumption or expenditure data.

4.1 Parental Labor Force Attachment

A second potential explanation is that parents use their additional income to substitute away from full time employment and into more childrearing. We have information on both parents�labor force attachment for each interview wave. Because we have panel data with regard to the parental labor force at- tachment, we employ a �xed-e¤ects regression model for mother�s and father�s labor force attachment. In the �rst panel of Table 10 we regress mother�s labor force attachment on whether the household was eligible for casino dis- bursements, a lag of household income, number of children less than six years old in the household and mother�s age. The outcome variable is categorical data for full time employment, part time employment, unemployment, house- hold employment and out of the labor force status. A negative coe¢ cient on the casino eligibility variable indicates that the additional household income increases the labor force attachment of the mother. The coe¢ cient is negative

15

and statistically signi�cant at the 10% level in the second regression. This indi- cates that mothers are actually increasing their labor force attachment given the additional income. Creating a binary variable for labor force participation and running a quasi �xed-e¤ect probit regression (Wooldridge, 2005), we �nd that there is no overall change in mother�s labor force participation given the addi- tional household income. Therefore, the additional household income does not appear large enough to a¤ect the mother�s labor force participation; however, it appears to allow mothers to increase their labor force attachment. While it is not possible to determine exactly what is driving this process, potentially the additional income allows mothers to a¤ord child care so that they may work more at the margin. Nevertheless, the simple explanation that mothers are us- ing the additional income to substitute away from work into leisure or childcare is not supported in our data. While fathers appear to have outcomes that di¤er from the mothers, they,

too, do not appear to make any signi�cant changes to their labor force attach- ment at either the extensive or intensive margins. The coe¢ cients on the casino income variable are positive, but it loses statistical signi�cance once father�s age is controlled. Similar to the result found for the mothers, when we examine fa- ther�s labor force participation using a quasi �xed-e¤ects probit regression we �nd that the result is highly insigni�cant.

4.2 Parental Behavior and Quality Measures

A third explanation is that parental quality improves with additional income. Increased household incomes may translate into lower levels of household stress and disruption. There is existing research that indicates poverty decreases parental quality. McLeod et. al. (1993) �nd using NLSY data that currently poor mothers are more likely to spank their children and are less responsive to child needs. They also �nd that the persistence of poverty increases the direct internalization symptoms in children. Sampson et. al. (1994) �nd that poverty decreases adult stability and good decision-making. Ennis et al. (2000) have found that poverty can adversely a¤ect mental health and depression among parents. Conger (1994) �nds direct evidence that not having su¢ cient income produces stresses on individual parents.

4.2.1 Parental Arrests

Additional information is available with regard to the two parents�arrests since the last interview at each survey wave. Table 11 indicates that fathers have a reduced probability of being arrested when they come from households with casino payments. This e¤ect is intensi�ed for the households that were previ- ously in poverty, however the sample size falls dramatically and is not shown here.

16

4.2.2 Parental Supervision of Children

The two preceding tables appear to indicate that parents are engaging in less destructive behavior as a result of the increased incomes. This improvement in parental behavior and choices also tends to spill over into parent-child interac- tions and supervision. Contained in the GSMS data is a parental supervision variable which asks the parent at each interview wave the percentage of time they know their child�s whereabouts and activities. In Table 12 we conduct a �xed-e¤ects regression of the mother and father�s reported supervision of their child on the household�s eligibility for casino payments, the child�s age, household income, parental ages and the number of children below age six in the household. The negative coe¢ cient on the casino disbursement indicates an improvement in parental supervision (the outcome variable is actually the inadequacy of parental supervision) for those households receiving additional income. This is conditional on child�s age and the other covariates as well as the time invariant family �xed-e¤ects. Our results indicate that fathers have an improvement in supervision of their child with the additional incomes as well. The third column reports the joint outcome of parental supervision on casino disbursements; the results indicate that there�s an increase in overall supervision for their children.

4.2.3 Parent-Child Interactions

Finally, Table 13 presents a direct measure of parental quality as reported by the child. Previous parental behavioral information was provided by the parent at all survey waves. The variable we consider here measures the amount of negative interactions between the child and parent from the child�s perspective. In both cases, the estimated coe¢ cient is negative which indicates an improve- ment in parent �child interactions. The results indicate that there is a large improvement for the relationship between the child and the mother and that this improvement is statistically signi�cant. The results are not statistically signi�cant with regard to the father, while the estimated coe¢ cient is of the same sign as the mother. Overall, the results indicate that parents in households with additional in-

comes make better choices in their personal behavior with regard to drug and alcohol abuse and criminal behavior. They do not appear to make signi�cant changes in their labor force participation e¤orts. Children report better rela- tionships overtime in the households with additional income and parents report better supervision of their children over time in these same households. While there are many potential causal mechanisms at work here, it is instructive to learn that parental time is not responsible for the observed changes in child out- comes. Parental quality and interactions with their children appears to be an important candidate for explaining how additional household income translates into better child outcomes.

17

5 Discussion and Conclusion

Our results indicate that changes in a household�s permanent income can have permanent e¤ects. The e¤ect on children continues on into young adulthood in our sample. We have seen that an exogenous treatment of increasing incomes tends to improve the overall child outcomes in terms of educational attainment at ages 19 and 21 and reduced criminal behavior at ages 16 and 17. Given the unique design of the research, we are able to control for several important confounding factors that might otherwise be the cause of the observed changes. We have been able to control for cohort di¤erences by using a control group of non-treated households in our sample. Additionally, the comparison between the age 9 and age 13 cohorts provides us with the counterfactual observations of a household where incomes were unchanged for a shorter period of time (6 years versus 2 years). We have also explored a couple of the potential mechanisms that transform

additional household income into better child outcomes. While it is not possible in this analysis to de�nitively identify the true causal mechanism responsible for the improvement in young adult outcomes, we have been able to identify a few changes in parental behavior (parental quality) that is suggestive of a mechanism. Parents have a better overall relationship with their children after the additional household income is introduced as evidenced by responses from both the parent and child. Additionally, parents appear to have less problems over time once the exogenous income is introduced: we see that fathers are less likely to be arrested themselves over time. On the other hand, we do not have much evidence that the additional income is by parents to make a dramatic shift from labor force participation towards more child care (parental quantity). While our data is not perfect, it appears that neither mothers nor fathers are leaving the labor force because of the additional household income. There is some evidence, in fact, that mothers may be increasing their labor force attachment (work intensity) but we do not have actual hours of work here, so our results once again do not provide conclusive evidence. More research that focuses on the mechanisms that translate household incomes into child well-being is certainly needed. It is important to note the di¤erences from this research and previous ef-

forts. The program described here di¤ers in at least two dimensions: size and duration. The size of the casino payments is large relative to other income augmentation programs and certainly with regard to other quasi-experimental policies. The additional $5-10,000 dollars per year represents anywhere from 1/4 to 1/3 of many of these household�s incomes. Second, this casino dis- bursement program has no foreseeable end date. While it is contingent upon successful and continued operations of the casino, there has been no indication that there would be a change in the program or that pro�ts have decreased over time. Therefore, people treat these changes in their income as permanent and spend accordingly. These two e¤ects are probably responsible for the large e¤ects found in this research which are not often evident in studies with smaller amounts and temporary income changes.

18

Future work will allow us to explore the e¤ect of this additional income on the geographic mobility of the children. The casino payments are not limited by geographic proximity to the Eastern Cherokee reservation. Therefore, in fu- ture work we anticipate evaluating how this additional income has increased the geographic distribution of these children from American Indian households- indi- viduals may move out of state and they will still be eligible for casino payments. In future survey waves we shall also have additional employment information for the children at ages 24 and 25 which will allow us to explore whether they di¤erentially enter into di¤erent occupations and industries and any resulting wage di¤erentials.

6 Appendix I

In this appendix, we discuss a few of the robustness tests that we conduct to investigate whether these observed outcomes were already prevalent in the data prior to the advent of the casino payments. With regard to the children�s outcomes, it is unfortunately not possible to run placebo tests on all of the outcomes variables as many of them have little meaning at earlier ages. For instance, high school completion rates will be uniformly zero in the period prior to the casino operations as the children are all below the ages of 15 in this time period. . Therefore, it is not possible to create a placebo test in these cases. We can, however, investigate whether there is any di¤erence in number of days of school attendance in the previous three months in the period prior to the casino operations given the panel nature of the data. In Appendix Table 2 we restrict analysis to the four survey waves prior to the casino opening and create a false treatment that occurs in wave 3 and 4 only; we �nd that there is no e¤ect on children�s schooling attendance in this �xed e¤ects regression. Previously in Table 6 we established that the additional household income increased school attendance by an average of 2.5 days. The second model in Appendix Table 2 provides the di¤erence in di¤erence regression for educational attainment at age 16. There appears to be no di¤erence for the two interaction variables in this regression. However, school attendance is compulsory up to age 16 in North Carolina and therefore, we are not clear whether �nding no result is due to there being no actual di¤erence in behavior or because of the e¤ect of the law.

Additionally, the arrest data that we have collected on the children in this survey is for ages 16-21, therefore, it is not possible to create a placebo test here in the period prior to the casino operations. Although we have no data on arrests at this early age it would also most likely be a very low probability event and any resulting tests would be expected to have very low power in any case. Examining parental changes in the period prior to casino disbursements is

a little more straightforward. Appendix Table 3 provides outcomes for parents when the period is restricted to the �rst four survey waves as above. A placebo

19

treatment is created for waves 3 and 4 as we did for the school attendance rates previously. In column one and two, we show that there is no statistical di¤erence in the labor force attachment outcome variable for parents prior to the casino disbursements and operations. The next two sets of columns provides the same analysis for mother�s and father�s activities with the child. We see once again that there is no statistically signi�cant coe¢ cient on the casino disbursement eligibility variable. The last two columns provides the placebo test for mothers� and fathers�supervision of their children. While the coe¢ cients on casino disbursement eligibility is statistically signi�cant in both cases it is in positive - indicating that prior to the casino operations mothers and fathers in these households actually had a lower level of supervision of their children. In Table 12 we have previously shown that the exact opposite occurs once the additional household income arrives from the casino disbursements; parents have better supervision of their children when their incomes are exogenously increased.

References

[1] Behrman, Jere R. and Mark R. Rosenzweig, �Does Increasing Women�s SchoolingRaise theSchoolingof theNextGeneration,�AmericanEconomic Review, V. 92 no. 1 (2002), 323 �334.

[2] Behrman, Jere R. and Susan W. Parker and Petra E. Todd, �Long-Term Impacts of the Oportunidades Conditional Cash Transfer Program on Rural Youth in Mexico,�Discussion Paper Nr 122, Ibero-America Institute for Economic Research, October (2005).

[3] Black, Sandra E. and Paul Devereux and Kjell Salvanes, �Why the Apple Doesn�t Fall Far: Understanding Intergenerational Transmission of Human Capital,�American Economic Review, V. 95 no. 1 (2005), 437 �449.

[4] Blanden, Jo and Paul Gregg, �Family Income and Educational Attainment: A Review of Approaches and Evidence for Britain,�CMPO Working Paper Series No. 04/101 (2004).

[5] Blank, Rebecca M., �Selecting Among Anti-Poverty Policies: Can an Economist be Both Critical and Caring?�Review of Social Economy, V. LXI No. 4 (December 2003), 447-470.

[6] Chen, Joyce. �Migration and Imperfect Monitoring: Implications for Intra- Household Allocation, �American Economic Review: Papers and Proceed- ings, (2006).

[7] Chevalier, Arnaud and Colm Harmon and Vincent O�Sullivan and Ian Walker, �The Impact of Parental Income and Education of the Schooling of Their Children,�IZA Discussion Paper No. 1496 (2005).

20

[8] Conger, R, and X. Ge, G. Elder, F. Lorenz, R Simons, �Economic stress, coercive family process, and developmental problems of adolescents,�Child Development, V. 65 (1994), 541-561.

[9] Costello EJ, Angold, A., Burns, B. , Stangl, Dalene, Tweed, Dan, Erkanli, Alaattin, Worthman, Carol, "The Great Smoky Mountains Study of Youth: Goals, Design, Methods, and the Prevalence of DSM-III-R Disorders," Archives of General Psychiatry, V.53(12) (December 1996), 1129-1136.

[10] Costello EJ, Compton S, Keeler G, & Angold A., �Relationships Between Poverty and Psychopathology: A Natural Experiment,� Journal of the American Medical Association, V. 290 (2003), 2023-2029.

[11] Ennis, N and S. Hobfoll, K Schroder, �Money doesn�t talk it swears: How economic stress and resistance resources impact inner-city women�s depres- sive mood.�American Journal of Community Psychology, V. 28 (2000) , 149-173.

[12] Epps, Sylvia R. and Aletha C. Huston, �E¤ects of a Poverty Intervention Policy Demonstration on Parenting and Child Behavior: A Test of the Direction of E¤ects,�Social Science Quarterly, V. 88, No. 2 (June 2007), 344-365.

[13] Hoynes, Hilary W. and Marianne E. Page and Ann Hu¤Stevens, �Poverty in America: Trends and Explanations,�Journal of Economic Perspectives, V. 20 No. 1 (Winter 2006), 47-68.

[14] Kling, Je¤rey R. and Jens Ludwig and Lawrence F. Katz, �Neighborhood E¤ects on Crime for Female and Male Youth: Evidence from a Randomized Housing Voucher Experiment,�Quarterly Journal of Economics, V. (2005), 87-128.

[15] Kling, Je¤rey R. and Jens Ludwig and Lawrence F. Katz, �Experimental Analysis of Neighborhood E¤ects,�Econometrica, V. 75 No. 1 (January 2007), 83-119.

[16] Loken, Katrine, �Family Income and Children�s Education: Using the Nor- wegian Oil Boom as a Natural Experiment,�Unpublished Manuscript, De- partment of Economics, University of Bergen, Norway (2007).

[17] Maurin, Eric, �The Impact of Parental Income on Early Schooling Tran- sitions. A Re-examination using data over three generations,�Journal of Public Economics, V. 85 (2002), 301-332.

[18] McLeod, J. and M. Shanahan, �Poverty, Parenting, and Children�s Mental Health,�American Sociological Review, V. 58 (1993), June: 351-366.

[19] Plug, Erik and Wim Vijverberg, �Schooling, Family Background, and Adoption: Is it Nature of Is it Nurture?�Journal of Political Economy, V. 111 no. 3 (2003), 611-641.

21

[20] Plug, Erik, �Estimating the E¤ect of Mother�s Schooling on Children�s Schooling Using a Sample of Adoptees,�American Economic Review, V. 94 no. 1 (2004), 358 �368.

[21] Sacerdote, Bruce, �The Nature and Nurture of Economic Outcomes,� American Economic Review, V. 92 no. 2, (2002), 344 �348.

[22] Sampson, R. and J. Laub, �Urban poverty and the family context of delin- quency: a new look at structure and process in a classic study,� Child Development, V. 65 (1994), 523-540.

[23] Shea, John, �Does parents�money matter?�Journal of Public Economics, V. 77 (2000), 155-184.

[24] Wooldridge, Je¤rey M., "Simple Solutions to the Initial Conditions Prob- lem in Dynamic, Nonlinear Panel Data Models with Unobserved Hetero- geneity," Journal of Applied Econometrics, V. 20 (2005), 39-54.The Har- vard Bibliography System

22

Variable Mean Mean

T-Statistics for Difference in Group Means

Education Variables Years of Education 11.21 11.96 -4.10 High School Graduation Probability at age 19 0.62 0.69 -2.12 Received a GED or Graduated from High School at age 19 0.76 0.82 -2.26

Age, Parents and Interaction Variables Age Cohort Initially 9 Year Olds 0.39 0.35 1.26 Age Cohort Initially 11 Year Olds 0.33 0.34 -0.51 Age Cohort Initially 13 Year Olds ref. ref.

Number of American Indian Parents 1.34 0.00 20.63 Interaction Age 9 Cohort x Number of American Indian Parents 0.52 0.00 17.98

Interaction Age 11 Cohort x Number of American Indian Parents 0.45 0.00 79.58

Household Characteristics Male Child Indicator 0.52 0.53 -0.29 Parent 1's Educational Level 3.95 4.92 -5.63 Parent 2's Educational Level 2.15 3.06 -4.11 Average Years Household in Poverty over initial 3 years 1.40 0.66 9.60 Average Household Income (by category) for first 3 years 4.58 6.65 -8.79

Crime Variables Any Crime Ages 16-17 0.10 0.14 -1.72 Any Crime Ages 18-19 0.17 0.22 -1.81 Any Crime Ages 20-21 0.16 0.15 0.28

Any Minor Crime by Age 21 0.25 0.29 -1.10 Any Moderate Crime by Age 21 0.09 0.14 -1.79 Any Violent Crime by Age 21 0.04 0.05 -0.86

Ever Dealt Drugs by Age 21 0.06 0.06 -0.47

Note: Sample size is 1060 observations for all three age cohorts when they are 21 years of age

Table 1: Mean Values for Variables

At least one AI Parent Household

No AI Parent Household

Panel A: Difference in Difference Regressions

Variable Mean Mean

T-Statistics for Difference in Group Means

Total Observations

Education Variable Days Present at School in Last Quarter 39.64 39.15 1.27 3317

Mother's Characteristics Labor Force Participation Rate 0.88 0.87 1.14 6780 Labor Force Attachment 0.76 0.78 -0.61 6780 Drug or Alcohol Problem 0.24 0.13 8.66 5333 Arrest Status 0.12 0.06 7.51 5333 Supervision of Child 0.08 0.10 -0.79 5758 Activities spent with Child 0.21 0.20 0.97 6673

Father's Characteristics Labor Force Participation Rate 0.90 0.93 -3.95 4161 Labor Force Attachment 0.59 0.33 6.63 4161 Drug or Alcohol Problem 0.08 0.05 2.75 3316 Arrest Status 0.27 0.13 9.18 3309 Supervision of Child 0.05 0.06 -0.41 5758 Activities spent with Child 0.18 0.15 1.30 3829

Note: Sample size differs across these variables due to missing information.

At least one AI Parent Household

No AI Parent Household

Panel B: Fixed Effect Regressions Table 1: Mean Values for Variables, cont.

Difference Between Cohort 1 and 2

Difference Between Cohort 2 and 3

Difference Between Cohort 1 and 3

Number of American Indian Parents American Indian Indicator -1.43 -2.00 -3.35 Male Child Indicator -0.93 1.84 0.95 Parent 1's Educational Level -1.08 -1.39 -2.36 Parent 2's Educational Level -0.47 -0.09 -0.53 Household Income -2.47 0.36 -2.04 Note: Each cell provides t-statistics for a test of difference in means

Difference Between Cohort 1 and 2

Difference Between Cohort 2 and 3

Difference Between Cohort 1 and 3

Number of American Indian Parents -0.49 1.29 0.84 American Indian Indicator -1.89 1.86 0.04 Male Child Indicator -0.56 0.05 -0.46 Parent 1's Educational Level -0.29 0.51 0.27 Parent 2's Educational Level 1.05 -2.56 -1.50 Household Income 0.34 -1.60 -1.29

Table 2: Differences by Age Cohort and American Indian Parent Status

Households with No American Indian Parent

Households with at least one American Indian Parent

Independent Variables Coeff. Std Error Marg. Eff. Std Error Marg. Eff. Std Error Interaction 1: Age Cohort 1 x Number of American Indian Parents

0.332 0.477 0.149*** 0.072 0.080* 0.046

Interaction 2: Age Cohort 2 x Number of American Indian Parents

-0.039 0.347 0.029 0.065 0.022 0.040

Age Cohort 1 (9 yo) -0.143 0.291 -0.009 0.060 -0.005 0.039 Age Cohort 2 (11 yo) 0.256 0.275 < 0.001 0.055 -0.007 0.037 Number of American Indian Parents in Household -0.111 0.426 -0.122* 0.066 -0.092** 0.039 American Indian -0.400 0.480 0.053 0.060 0.046 0.034 Sex -0.547** 0.227 -0.118*** 0.043 -0.070** 0.030 Parent 1 Education 0.221*** 0.052 0.013 0.009 0.014** 0.006 Parent 2 Education 0.080* 0.043 0.020*** 0.008 0.022*** 0.005 HH in Poverty Indicator Variable -0.183 0.165 -0.054* 0.028 -0.033* 0.018 Average HH Income 0.163 0.051 0.020** 0.010 0.010 0.007 Constant 10.298*** 0.509 Observations 1044 1059 1059 Wald Chi-Squared (15) 33.33 95.81 94.5 Pseudo R2 0.2731 0.1645 0.1936 Note: *** indicates coefficient statistically significant at the 1% level, ** at the 5% level and * at the 10% level.

Table 3: Education Variables

Years of Education, Age 21

Probability of HS Grad, Age 19

Prob of HS Grad/GED, Age 19

Independent Variables Coeff. Std Error Coeff. Std Error Marg. Eff. Std Error Marg. Eff. Std Error Marg. Eff. Std Error Marg. Eff. Std Error Interaction 1: Age Cohort 1 x Number of American Indian Parents

0.903* 0.462 0.336*** 0.133 0.239** 0.115 -0.111 0.787 0.124 0.086 0.040 0.053

Interaction 2: Age Cohort 2 x Number of American Indian Parents

0.169 0.462 0.190 0.137 0.112 0.115 -0.240 0.509 0.006 0.071 0.001 0.045

Age Cohort 1 (9 yo) -0.612 0.532 -0.130 0.126 -0.050 0.121 0.096 0.360 0.037 0.049 0.012 0.031 Age Cohort 2 (11 yo) 0.299 0.526 -0.156 0.126 -0.021 0.121 0.263 0.322 0.048 0.045 0.003 0.029

Number of American Indian Parents in Household -0.659* 0.402 -0.451*** 0.130 -0.348*** 0.108 0.309 0.603 0.010 0.074 -0.001 0.046 American Indian -0.038 0.423 0.237** 0.104 0.201** 0.082 -0.906 0.696 -0.054 0.090 -0.030 0.061 Sex -0.528 0.366 -0.097 0.092 -0.070 0.087 -0.528* 0.283 -0.099** 0.040 -0.063** 0.026 Parent 1 Education 0.177** 0.088 0.011 0.023 0.037* 0.022 0.243*** 0.062 0.011 0.008 0.008 0.005 Parent 2 Education 0.053 0.074 0.009 0.024 0.065*** 0.020 0.085* 0.049 0.018*** 0.006 0.012*** 0.004 Average HH Income 0.263** 0.133 0.066* 0.034 0.033 0.032 0.133** 0.060 0.007 0.008 0.003 0.005 Constant 9.956*** 0.852 10.332*** 0.583 Number 437 443 443 607 616 616 Wald Chi-Squared (15) 6.03 28.07 36.94 11.49 45.53 32.6 Pseudo R2 0.146 0.065 0.115 0.204 0.111 0.121

Years of Education, Age 21

Probability of HS Grad, Age 19

Prob of HS Grad/GED, Age 19

Probability of HS Grad, Age 19

Prob of HS Grad/GED, Age 19

Table 4: Education Variables by Poverty Status

Household Previously in Poverty Household Not Previously in Poverty

Years of Education, Age 21

Independent Variables Coeff. Std Error Coeff. Std Error Marg. EffStd Error Marg. Eff. Std Error Marg. EfStd Error Marg. Eff. Std Error Interaction 1: Age Cohort 1 x Number of American Indian Parents

0.315 0.452 0.137 0.105 0.158** 0.069 0.992* 0.580 0.218*** 0.082 0.061 0.043

Interaction 2: Age Cohort 2 x Number of American Indian Parents

0.181 0.438 0.040 0.097 0.092 0.065 0.066 0.468 0.055 0.077 -0.016 0.039

Age Cohort 1 (9 yo) 0.054 0.415 0.066 0.097 0.035 0.064 -0.361 0.408 -0.097 0.070 -0.048 0.046 Age Cohort 2 (11 yo) -0.035 0.416 0.053 0.086 -0.027 0.065 0.354 0.370 -0.059 0.067 -0.005 0.034 Number of American Indian Parents in Household -0.071 0.411 -0.069 0.092 -0.113 0.057 -0.531 0.592 -0.159** 0.080 -0.044 0.036 American Indian -0.326 0.439 0.001 0.099 0.006 0.062 -0.372 0.715 0.048 0.060 0.025 0.029 Parent 1 Education 0.174** 0.084 0.009 0.016 0.003 0.011 0.255*** 0.064 0.016* 0.010 0.020*** 0.007 Parent 2 Education 0.113* 0.059 0.013 0.012 0.030*** 0.009 0.064 0.059 0.026*** 0.008 0.014*** 0.004

HH in Poverty Indicator Variable 0.062 0.238 0.005 0.046 0.003 0.032 -0.371 0.227 -0.091*** 0.032 -0.045*** 0.016 Average HH Income 0.277*** 0.083 0.0455*** 0.016 0.027** 0.011 0.079 0.062 -0.004 0.012 -0.002 0.006 Constant 8.91*** 0.730 10.923*** 0.655 Number 547 552 552 497 507 507 Wald Chi-Squared (15) 17.910 38.750 48.160 23.360 72.160 88.980 Pseudo R2 0.265 0.125 0.166 0.284 0.220 0.264 Note: *** indicates coefficient statistically significant at the 1% level, ** at the 5% level and * at the 10% level.

Table 5: Education Variables by Child's Gender

Male Child Female Child

Years of Education, Age 21

Probability of HS Grad, Age 19

Prob of HS Grad/GED, Age

19 Years of Education,

Age 21

Probability of HS Grad, Age

19 Prob of HS

Grad/GED, Age 19

Independent Variables Coeff. Std Error Coeff. Std Error Coeff. Std Error Household Eligible for Casino Disbursement 2.442** 1.124 3.85** 1.914 2.485* 1.419 Household Income -0.298** 0.145 0.142 0.266 -0.421** 0.174 Primary Parent's Age -0.181* 0.093 -0.212 0.266 -0.159 0.125 Secondary Parent's Age -0.056 0.068 -0.193 0.114 0.058 0.085 Age of Child 0.105 0.174 -0.768** 0.346 0.283 0.209 Number of Children Less than 6 years old 0.448 0.604 1.157 0.896 -0.547 0.853 Constant 49.373*** 3.416 64.825*** 5.945 43.034*** 4.469

Number of obs 3317 1120 2197 Number of groups 1110 444 666 Wald chi2(7) 2.550 3.95 2.04 Prob > chi2 0.0183 0.0007 0.0571 Note: *** indicates coefficient statistically significant at the 1% level, ** at the 5% level and * at the 10% level.

Number of Days Present Within the Last 3 Months if Household Never in Poverty

Table 6: Child School Attendance

Number of Days Present Within the Last 3

Months

Number of Days Present Within the Last 3 Months if

Household Previously in Poverty

Independent Variables Marg Coeff Std Error Marg Coeff Std Error Marg Coeff Std Error Interaction 1: Age Cohort 1 x Number of American Indian Parents

-0.226*** 0.076 -0.047 0.073 0.052 0.075

Interaction 2: Age Cohort 2 x Number of American Indian Parents

-0.108* 0.062 -0.020 0.069 0.008 0.062

Age Cohort 1 (9 yo) 0.077* 0.043 -0.018 0.053 -0.072** 0.034 Age Cohort 2 (11 yo) -0.017 0.037 -0.042 0.049 -0.059* 0.034 Number of American Indian Parents in Household 0.125 0.090 -0.050 0.064 0.090 0.078 American Indian -0.051 0.049 -0.019 0.058 -0.092*** 0.030 Sex 0.069** 0.029 0.149*** 0.040 0.110*** 0.032 Parent 1 Education -0.003 0.006 -0.001 0.009 0.000 0.007 Parent 2 Education -0.005 0.005 -0.005 0.007 -0.001 0.006 HH in Poverty Indicator Variable 0.004 0.017 0.013 0.026 0.014 0.018 Average HH Income -0.006 0.006 -0.004 0.009 -0.004 0.007

Number of obs 1092 1060 1044 F( 11, 1032) 43.87 25.75 29.76 Prob > F 0 0.0071 0.0017

R-squared 0.0763 0.0534 0.0709 Note: *** indicates coefficient statistically significant at the 1% level, ** at the 5% level and * at the 10% level.

Independent Variables Marg Coeff Std Error Marg Coeff Std Error Marg Coeff Std Error Interaction 1: Age Cohort 1 x Number of American Indian Parents -0.251*** 0.089 -0.028 0.112 -0.055 0.092 Interaction 2: Age Cohort 2 x Number of American Indian Parents -0.169* 0.093 -0.056 0.106 -0.133 0.109 Age Cohort 1 (9 yo) 0.107 0.071 0.054 0.093 -0.088 0.063 Age Cohort 2 (11 yo) -0.036 0.057 0.009 0.083 -0.017 0.071 Number of American Indian Parents in Household 0.222* 0.128 -0.042 0.092 0.155 0.129 American Indian -0.111* 0.058 -0.052 0.096 -0.114 0.072 Sex Parent 1 Education 0.006 0.011 0.014 0.015 0.003 0.013 Parent 2 Education -0.007 0.008 -0.015 0.013 0.004 0.010 HH in Poverty Indicator Variable 0.021 0.027 0.002 0.043 0.037 0.033 Average HH Income -0.004 0.010 -0.013 0.015 -0.007 0.011

Number of obs 587 553 547 F( 11, 1032) 20.37 11.37 11.96 Prob > F 0.026 0.330 0.287

R-squared 0.062 0.031 0.035 Note: *** indicates coefficient statistically significant at the 1% level, ** at the 5% level and * at the 10% level.

Committed Any Crime if Male, Age 16-17

Committed Any Crime if Male, Age 18-19

Committed Any Crime if Male, Age 20-21

Table 7: Any Criminal Behavior of the Child in Young Adulthood

Committed Any Crime, Age 16-17

Committed Any Crime, Age 18-19

Committed Any Crime, Age 20-21

Independent Variables Marg Coeff Std Error Interaction 1: Age Cohort 1 x Number of American Indian Parents -0.244*** 0.084 Interaction 2: Age Cohort 2 x Number of American Indian Parents -0.060 0.066 Age Cohort 1 (9 yo) 0.259*** 0.078 Age Cohort 2 (11 yo) 0.044 0.055 Number of American Indian Parents in Household 0.071 0.071 American Indian 0.004 0.065 Sex 0.091* 0.052 Parent 1 Education -0.011 0.013 Parent 2 Education -0.001 0.014 HH in Poverty Indicator Variable -0.011 0.033 Average HH Income -0.011 0.020

Number of obs 465 F( 11, 1032) 27.98 Prob > F 0.003

R-squared 0.138 Note: *** indicates coefficient statistically significant at the 1% level, ** at the 5% level and * at the 10% level.

Committed Any Crime if Household Previously in

Poverty, Age 16-17

Table 7 cont: Any Criminal Behavior of the Child in Young Adulthood

Independent Variables Marg. Coeff Std Error Marg. Coeff Std Error Marg. Coeff Std Error Interaction 1: Age Cohort 1 x Number of American Indian Parents

-0.167* 0.086 -0.006 0.064 -0.002 0.013 Interaction 2: Age Cohort 2 x Number of American Indian Parents

-0.072 0.084 -0.024 0.047 -0.009 0.014 Age Cohort 1 (9 yo) -0.059 0.056 -0.016 0.026 0.000 0.010 Age Cohort 2 (11 yo) -0.096* 0.054 -0.043** 0.022 0.014 0.013 Number of American Indian Parents in Household 0.095 0.092 0.118* 0.067 -0.003 0.009 American Indian -0.099 0.067 -0.074*** 0.015 0.003 0.010 Sex 0.179 0.044 0.072*** 0.024 0.047*** 0.014 Parent 1 Education -0.001 0.010 0.002 0.006 0.003 0.002 Parent 2 Education -0.012 0.008 -0.006 0.004 -0.001 0.002 HH in Poverty Indicator Variable 0.059* 0.029 0.013 0.013 -0.001 0.004 Average HH Income 0.008 0.010 -0.005 0.005 -0.004*** 0.002

Number of obs 1044.000 1044 1044 Wald chi2(10) 40.150 46.81 55.160 Prob > chi2 0.000 0.000 0.000 Pseudo R2 0.073 0.096 0.164 Note: *** indicates coefficient statistically significant at the 1% level, ** at the 5% level and * at the 10% level.

Independent Variables Marg. Coeff Std Error Marg. Coeff Std Error Marg. Coeff Std Error Interaction 1: Age Cohort 1 x Number of American Indian Parents

-0.185 0.115 -0.162** 0.075 -0.032 0.047 Interaction 2: Age Cohort 2 x Number of American Indian Parents

-0.162 0.135 -0.132* 0.080 -0.054 0.046 Age Cohort 1 (9 yo) -0.061 0.090 -0.038 0.052 0.017 0.035 Age Cohort 2 (11 yo) -0.095 0.088 -0.051 0.048 0.061 0.050 Number of American Indian Parents in Household 0.209 0.142 0.161 0.122 -0.009 0.034 American Indian -0.210** 0.103 -0.104** 0.046 0.001 0.031 Sex Parent 1 Education 0.010 0.016 0.005 0.010 0.010 0.006 Parent 2 Education -0.020 0.014 -0.005 0.008 -0.004 0.005 HH in Poverty Indicator Variable 0.083* 0.046 0.028 0.027 -0.005 0.014 Average HH Income 0.011 0.016 -0.009 0.009 -0.012* 0.007

Number of obs 547 547 547 Wald chi2(10) 14.83 26.93 24.96 Prob > chi2 0.138 0.003 0.005 Pseudo R2 0.040 0.063 0.085 Note: *** indicates coefficient statistically significant at the 1% level, ** at the 5% level and * at the 10% level.

Table 8: Likelihood of Committing Crime by Type at Any Time in Late Teen Years (16-21)

Ever Committed A Minor Crime by Age 21

Ever Committed A Moderate Crime by Age

21

Ever Committed A Violent Crime by Age

21

Ever Committed A Minor Crime by Age 21 if Male

Moderate Crime by Age 21 if Male

Violent Crime by Age 21 if Male

Independent Variables Marg. Coeff Std Error Marg. Coeff Std Error Interaction 1: Age Cohort 1 x Number of American Indian Parents

-0.146 0.111 -0.231* 0.136 Interaction 2: Age Cohort 2 x Number of American Indian Parents

-0.020 0.111 -0.118 0.119 Age Cohort 1 (9 yo) 0.137 0.116 -0.129** 0.056 Age Cohort 2 (11 yo) -0.017 0.117 -0.119** 0.055 Number of American Indian Parents in Household 0.085 0.095 0.114 0.116 American Indian -0.142* 0.084 -0.033 0.108 Sex 0.252*** 0.085 0.150* 0.050 Parent 1 Education 0.009 0.025 -0.005 0.010 Parent 2 Education -0.014 0.020 -0.013 0.008 HH in Poverty Indicator Variable 0.016 0.065 Average HH Income 0.037 0.037 0.013 0.010

Number of obs 437 607 Wald chi2(10) 18.18 29.32 Prob > chi2 0.078 0.001 Pseudo R2 0.078 0.077 Note: *** indicates coefficient statistically significant at the 1% level, ** at the 5% level and * at the 10% level.

Ever Committed A Minor Crime by Age 21 if

Household Previously in Poverty

Ever Committed A Minor Crime by Age 21 if Household Never

Previously in Poverty

Table 8 cont: Likelihood of Committing Crime by Type at Any Time in Late Teen Years (16-21)

Independent Variables Coeff. Std Error Coeff. Std Error Interaction 1: Age Cohort 1 x Number of American Indian Parents -0.072** 0.033 -0.073 0.055 Interaction 2: Age Cohort 2 x Number of American Indian Parents -0.010 0.020 -0.013 0.040 Age Cohort 1 (9 yo) 0.003 0.016 -0.005 0.034 Age Cohort 2 (11 yo) 0.025 0.018 0.048 0.038 Number of American Indian Parents in Household -0.013 0.018 -0.021 0.036 American Indian 0.036 0.029 0.031 0.039 Sex 0.066*** 0.012 0.075*** 0.022 Parent 1 Education 0.005** 0.003 0.005 0.006 Parent 2 Education -0.003 0.002 -0.004 0.006 HH in Poverty Indicator Variable -0.004 0.007 -0.013 0.015 Average HH Income -0.007** 0.003 -0.007 0.010

Number of obs 1044 437.000 Wald chi2(10) 59.90 18.660 Prob > chi2 0.000 0.068 Pseudo R2 0.125 0.097

Table 9: Likelihood of Dealing Drugs

Ever Dealt Drugs by Age 21

Ever Dealt Drugs by Age 21 if Household Previously in

Poverty

Independent VariablesCoeff. Std Error Coeff. Std Error Marg. Coeff Std Error Household Eligible for Casino Disbursement -0.171** 0.074 -0.153* 0.079 0.069 0.196 Lag of Household Income -0.013 0.011 -0.010 0.011 0.021 0.028 Number of Children Less than 6 years old 0.086* 0.045 0.077* 0.046 0.031 0.096 Mother's Age -0.010 0.008 0.012 0.018

Mother's Initial Labor Force Status 0.964*** 0.145 Constant 0.817*** 0.073 1.170*** 0.311 0.117 0.358

Number of obs 3567 3487 Number of obs 3318 Number of groups 1145 1140 Number of group 1076 Wald chi2(7) 4.01 3.350 Wald chi2(9) 343.31 Prob > chi2 0.0074 0.0095 Prob > chi2 0

Note: *** indicates coefficient statistically significant at the 1% level, ** at the 5% level and * at the 10% leve.

Independent VariablesCoeff. Std Error Coeff. Std Error Marg. Coeff Std Error Household Eligible for Casino Disbursement

0.121* 0.068 0.077 0.071 -0.011 0.384 Lag of Household Income -0.008 0.009 -0.008 0.010 0.072 0.072 Number of Children Less than 6 years old 0.024 0.051 0.030 0.051 -0.235 0.285 Father's Age 0.005 0.007 -0.102** 0.044

Father's Initial Labor Force Status 2.093** 0.582 Constant 0.358*** 0.074 0.134 0.277 0.423 0.765

Number of obs 2227 2177 Number of obs 1988 Number of groups 729 723 Number of grou 643 Wald chi2(7) 1.22 0.740 Wald chi2(9) 105.95 Prob > chi2 0.3023 0.5676 Prob > chi2 0.00

Note: *** indicates coefficient statistically significant at the 1% level, ** at the 5% level and * at the 10% level.

Note: Probit regression includes the mean over all time periods for the following variables: household eligibility for casino, mother's age, the lag of household income, number of children below age 6

Mother's Labor Force Attachment

Table 10: Parents' Employment Status Changes

Mother's Labor Force Participation

Father's Labor Force ParticipationFather's Labor Force Attachment

Note: Probit regression includes the mean over all time periods for the following variables: household eligibility for casino, mother's age, the lag of household income, number of children below age 6

Independent Variables Marg. Coeff Std Error Marg. Coeff Std Error Household Eligible for Casino Disbursement -0.181 0.210 -0.550** 0.260 Mother's Age -0.036 0.022 Father's Age -0.043 0.032 Household Income 0.026 0.035 -0.137** 0.064 Labor Force Status - Mother -0.105** 0.052 Labor Force Status - Father 0.053 0.143 Number of Children Less than 6 years old 0.160 0.103 -0.044** 0.020 Initial Arrest Status for Mother 0.821*** 0.127 Initial Arrest Status for Father 0.699*** 0.113

Number of obs 3473 2158 Number of groups 1135 721 Wald chi2(7) 523.46 458.200 Prob > chi2 0 0 Note: The first regressions includes means for all of the independent variables over all time periods: household eligibility for casino, mother's age, household income, labor force status of the mother, number of children below age 6; the second regression includes : household eligibility for casino, father's age, household income, labor force status of the father, number of children below age 6. This is a random effects probit specification as suggested by Wooldridge (2005).

Table 11: Parent's Arrest Since the Last Interview

Mother Arrest Since Last Interview

Father Arrest Since Last Interview

Independent Variables Coeff. Std Error Coeff. Std Error Coeff. Std Error Household Eligible for Casino Disbursement

-0.105*** 0.041 -0.154*** 0.056 -0.277** 0.125

Household Income 0.011* 0.006 0.010 0.008 0.020 0.019 Mother's Age -0.006 0.011 0.102 0.098 Father's Age 0.017 0.021 0.012 0.020 Age of Child 0.004 0.007 -0.005 0.008 -0.008 0.020 Number of Children Less than 6 years old 0.021*** 0.009 0.038*** 0.011 0.069*** 0.027 Labor Force Status - Mother 0.027 0.023 0.002 0.033 Labor Force Status - Father 0.067* 0.040 0.038 0.047 Constant -0.397** 0.182 -0.295 0.250 -1.024* 0.621

Number of obs 3802 2365 2025.0 Number of groups 1163 745 637 Wald chi2(7) 5.74 5.260 3.5 Prob > chi2 0 0 0.0 Note: *** indicates coefficient statistically significant at the 1% level, ** at the 5% level and * at the 10% level.

Mother's Supervision Father's Supervision Parental Supervision

Table 12: Parental Supervision

Independent Variables Coeff. Std Error Coeff. Std Error Household Eligible for Casino Disbursement

-0.089* 0.046 -0.054 0.057

Household Income 0.012** 0.006 0.003 0.007 Number of Children Less than 6 years old 0.033 0.026 -0.007 0.041 Mother's Age 0.004 0.007 0.009 0.009 Age of Child 0.013 0.009 0.012 0.012 Constant -0.243 0.195 -0.427 0.264

Number of obs 3910 2448 Number of groups 1172 760 Wald chi2(7) 3.73 2.310 Prob > chi2 0.0023 0.0422

Table 13: Activities With Parent

Activities With Mother Activities With Father

Note: *** indicates coefficient statistically significant at the 1% level, ** at the 5% level and * at the 10% level.

Figure 1: Household Income By American Indian Parent Status in Waves 1-3

0 10 20 30 40 50 60 70 80 90

100

1993 1994 1995 1996 1997 1998 1999 2000 Data Waves

P er

ce nt

ag e

of H

ou se

ho ld

s A

bo ve

$ 30

,0 00

Household with no AI Parents

Household with at least one AI Parent

Figure 2: Mother's Unemployment Incidence by Waves 1-3

0

5

10

15

20

25

30

1993 1994 1995 Data Waves

P er

ce nt

ag e

Households with no AI Parents

Households with at least one AI Parent

Figure 3: Father's Unemployment Incidence by Waves 1-3

0

5

10

15

20

25

30

1993 1994 1995 Data Waves

P er

ce nt

ag e

Households with no AI Parents

Households with at least one AI Parent

Figure 4: Second Parent's Reported Drug and Alcohol Incidence by Data Waves as Reported by

First Parent

0

5

10

15

20

25

30

1993 1994 1995 Data Waves

P er

ce nt

ag e

Households with no AI Parent Households with at least one AI Parent

Comparison Ages: 12/13 with 14 14 with 15 15 with 16 12/13 with 16

Age Group 1 -0.377 -0.898 -0.513 0.270

Age Group 2 1.400 0.520 -0.794 0.000

Age Group 3 -0.530 0.522 0.444 -0.545

Age Group 1 0.000 -0.650 -0.145 1.040

Age Group 2 0.140 -0.146 -0.146 0.044

Age Group 3 0.000 -0.629 -1.002 -0.480A m

er ic

an

In di

an

H ou

se ho

ld N

on -I

nd ia

n H

ou se

ho ld

Appendix Table 1: Comparison of Marital Status of Parents Across Time by Age Cohort and Household Type

Note: Reported figures are t-ratios for difference in the mean value of whether the child's parents are currently married at each survey wave. Ages 12 or 13 are used as not every age group was surveyed at ages 12 and 13, therefore, we combine those years for comparison.

Independent Variables Coeff. Std Error Independent Variables Coeff. Std Error Household Eligible for Casino Disbursement

-0.919 1.081

Interaction 1: Age Cohort 1 x Number of American Indian Parents

-0.012 0.130

Household Income -0.448** 0.207

Interaction 2: Age Cohort 2 x Number of American Indian Parents

-0.062 0.139 Mother's Age -0.200 0.131 Age Cohort 1 (9 yo) 0.148* 0.080 Father's Age -0.027 0.089 Age Cohort 2 (11 yo) 0.203*** 0.078

Child's Age -0.411 0.346 Number of American Indian Parents in Household 0.069 0.109

Number of Children Less than 6 years old 0.481 0.853 American Indian 0.034 0.141 Constant 56.827*** 5.879 Sex -0.072 0.062

Parent 1 Education -0.009 0.013 Parent 2 Education 0.004 0.011 HH in Poverty Indicator Variable -0.081** 0.044 Average HH Income 0.016 0.015 Constant 8.992 0.144

Number of obs 2372 Number of obs 1064 Number of groups 1062 F( 11, 1052) 2.67 F(6,1304) 2.92 Prob > F 0.0022 Prob > F 0.0079 R-squared 0.0655

Note: *** indicates coefficient statistically significant at the 1% level, ** at the 5% level and * at the 10% level.

Appendix Table 2: Placebo Tests on Children's Outcomes

Days in School in the Previous Quarter

Educational Attainment at age 16

Note: The first regression is conducted with a child fixed effect and is restricted to only the first four survey waves, with a placebo treatment introduced in waves 3 and 4. The second regression restricts analysis to age 16 for all of the children, which is five years earlier than the analysis presented in the main part of the paper; compulsory schooling laws may play a role as ages 7-16 are compulsory in North Carolina.

Independent Variables Coeff. Std Error Coeff. Std Error Coeff. Std Error Coeff. Std Error Coeff. Std Error Coeff. Std Error Household Eligible for Casino Disbursement -0.092 0.072 -0.027 0.066 0.058 0.051 -0.039 0.056 0.078** 0.039 0.131*** 0.050 Household Income 0.003 0.015 0.000 0.014 -0.003 0.010 -0.009 0.012 0.002 0.008 0.015 0.011 Number of Children Less than 6 years old 0.034 0.058 0.145** 0.074 -0.003 0.038 -0.002 0.060 0.010 0.029 -0.003 0.054 Mother's Age -0.022* 0.012 -0.003 0.009 -0.003 0.007 Father's Age 0.004 0.011 0.004 0.010 -0.005 0.009 Child's Age 0.003 0.017 0.041** 0.020 0.024* 0.013 0.036** 0.018 Labor Force Participation Mother 0.018 0.013 Labor Force Participation Father -0.012 0.024 Constant 1.594*** 0.466 0.159 0.441 0.246 0.303 -0.454 0.365 -0.165 0.245 -0.320 0.326

Number of obs 2655 1644 2727 1697 2721 1685 Number of groups 1114 699 1135 724 1135 718 F(6,1304) 1.69 1.03 0.36 1.39 2.59 4.12 Prob > F 0.1495 0.3885 0.8743 0.2258 0.0169 0.0004

Note: *** indicates coefficient statistically significant at the 1% level, ** at the 5%

Mother's Activities Father's Activities Father's SupervisionMother's Supervision

Note: All regressions contain a household fixed effect and is restricted to only the first four survey waves, with a placebo treatment introduced in waves 3 and 4.

Mother's Labor Force Father's Labor Force

Appendix Table 3: Placebo Tests on Parental Behaviors