Jurisdiction over Indian Country in Historical Context

At the start of colonization and settlement in North America, American-Indian tribes were viewed as legitimate political entities that could negotiate with European colonial powers through treaties. Continuing that tradition, the United States signed a number of agreements with tribes during the period between the founding of the nation and 1871, when the practice of signing treaties with tribes was explicitly banned by Congress (Reference Richland and DeerRichland & Deer 2004: 59). Many of these treaties dealt with jurisdiction over crimes committed by Indians against non-Indians and vice versa. A number of treaties explicitly recognized “the power of tribes to exercise criminal jurisdiction over non-Indian citizens of the Unites States” (Reference LeonhardLeonhard 2012: 7). From 1817 onward, however, a number of treaties limited tribal jurisdiction over non-Indians (Reference LeonhardLeonhard 2012: 8–9). Notably, in the 1830s, the legal status of Indian tribes changed from one of complete sovereignty to one of “dependent nations with limited sovereignty and the right to occupy and use their original land” (Reference Richland and DeerRichland & Deer 2004: 58). The period between 1835 and 1887 witnessed the continuous removal of American-Indian tribes from the Eastern territories and the creation of Indian reservations.

The establishment of Indian reservations gave rise to the notion of “Indian country,” which provides the basis for understanding jurisdiction in Indian Law. Indian country refers to

After 1871, Congress dealt with Indian affairs by adopting new legislation instead of signing treaties. The Major Crimes Act of 1885 delineated federal jurisdiction over serious crimes in Indian country (Reference ClintonClinton 1976). This Act represented a “significant incursion on tribal sovereignty in a manner inconsistent with the development of federal Indian criminal law during the treaty period” (Reference LeonhardLeonhard 2012: 12). As such, the Act foreshadowed the assimilationist agenda of the Termination Period (see below).

Enactment of PL280

In 1953, Congress enacted PL280 which first shifted jurisdiction over criminal and civil matters in Indian country from the federal government to state governments and, second, expanded “the reach of nontribal law enforcement and criminal justice on reservations” (Reference Goldberg, Champagne and SingletonGoldberg, Champagne, & Singleton 2008: 3). This jurisdictional change was mandatory for five states: California, Minnesota, Nebraska, Oregon, and Wisconsin.Footnote ² Importantly, the change of jurisdiction occurred without tribal consent, despite debates in Congress about the issue of consent prior to 1952:

In addition to being imposed on the “mandatory” states, PL280 allowed all other states to voluntarily assume jurisdiction in Indian country. Between 1953 and 1968, a number of “optional” states adopted PL280. These states fall into two categories: states with disclaimers in their constitutions limiting state jurisdiction over Indian country and states without disclaimers limiting state jurisdiction. The latter could adopt PL280 in a straightforward manner. The former, however, needed to amend their constitutions in order to adopt PL280.Footnote ³ Surprisingly, a number of states with disclaimers did not amend their constitutions before adopting PL280. The lack of amendments resulted in lawsuits challenging the jurisdictional transfer (Reference Gardner and Pecos MeltonGardner & Pecos Melton 2004). In 1979, however, the Supreme Court ruled in Washington v. Yakima Indian Nation that disclaimer states are not required to amend their constitutions in order to assume jurisdiction.

In 1968, because of a significant backlash against the original content of PL280, an amendment (under the Indian Civil Rights Act) was passed requiring states to obtain tribal consent before adopting PL280. This amendment also allowed states to initiate retrocession of jurisdiction back to the federal government (Reference GoldbergGoldberg 2010: 1046). “Although Indian nations were not given control over this process, they have been in a position to lobby their state legislatures to support this ‘retrocession’ of jurisdiction” (ibid.: 1046–47). The process of lobbying with the state for retrocession has presented many political obstacles and has been quite “formidable” for the tribes (Reference Goldberg and ChampagneGoldberg & Champagne 2006: 723). To date, retrocessions have occurred in several tribes and no tribe consented to adopting state jurisdiction under PL280 after the 1968 amendments (Reference GoldbergGoldberg 2010: 1047).

Table A1 in the Online AppendixFootnote ⁴ presents a detailed list of all mandatory and optional states that adopted PL280, the year of adoption, and, if applicable, the year of retrocession.

Political Context of PL280

According to official government documents, an important motivation for Congress in considering the promulgation of PL280 was weak law enforcement apparatus on reservations (H.R. Report No. 848). A careful inspection of the law's well-documented political context, however, reveals that it is unlikely that states were indeed selected to enact PL280 solely or primarily because of lawlessness. Political considerations reflecting the general spirit of the era seem to have been at least as important.

In fact, at the beginning of the lawmakers' deliberations, the focus was solely on the state of California where tribes had specifically expressed willingness to assimilate and to do away with the Bureau of Indian Affairs and with federal jurisdiction (Reference HerzbergHerzberg 1978: 155). Tribes that were strong enough to effectively lobby against PL280 were not included in the initial set of areas mandated to adopt the law (Reference Goldberg, Champagne and SingletonGoldberg, Champagne, & Singleton 2008: 3). Three notable exceptions to PL280 adoption in the “mandatory” states—the Red Lake Reservation in Minnesota, the Menominee Reservation in Wisconsin, and the Warm Springs Reservation in Oregon—successfully lobbied against the law and, consequently, were not subject to state jurisdiction.Footnote ⁵

Adoption of PL280 also required the consent of state and local authorities (see H.R. Report No. 848). For example, “in Nevada, authorities of some counties have indicated their willingness to accept jurisdiction, others opposed it and still others stated they would accept jurisdiction only with an accompanying Federal subsidy” (H.R. Report No. 848: 7). This, in essence, undermines the argument that states were chosen primarily because of lawlessness.

The historical context of the law's passage and the surrounding debate also reveals that a key underlying consideration for the selection of areas to which the new law would apply was inherently political. As suggested by Reference Jiménez and SongJiménez and Song (1998: 1662), “[n]o interpretation of PL280 can be persuasive unless it both acknowledges and is reconciled with the fact that it was enacted during a time in American history that is now referred to as the ‘Termination Period’.”

The 1940–1962 Termination Period “was marked by a distinctive philosophy and accompanied by legislation designed to promote the termination of Indian tribes” (ibid.). During this period, Congress passed legislative acts “promoting … the assimilation of the Indians into the mainstream of society … the termination of federal supervision over Indian affairs … and … the eventual relinquishment of federal control over Indian affairs to state and local authorities” (ibid.: 1662–63).

PL280 was enacted at the height of the Termination Period, “just two weeks after House Concurrent Resolution 108” (ibid., fn 204) had “crystallized the Termination Period's goals into official congressional policy and set the assimilation process in motion” (ibid.: 1663). As a consequence, “Public Law 280 contains a strong assimilationist bent and there [exists] language in the statute's legislative history that could support an assimilationist agenda” (ibid.: 1664). One senator, for example, argued that “Public Law 280 was appropriate because Indians had ‘reached a state of acculturation and development’ allowing for a smooth transition into society at large” (Reference TwettenTwetten 2000: 1323). All else equal, therefore, PL280 was more likely mandated in areas where tribes were deemed “ready for complete freedom from Federal supervision and wardship” (Reference Jiménez and SongJiménez & Song 1998: fn 206). We draw on these historical facts when devising our empirical strategy to estimate PL280's impact.

Data and Variables

To investigate the impact of PL280 on crime, we construct a county-level cross-sectional dataset. The data are drawn from datasets of the Inter-University Consortium for Political and Social Research (ICPSR), which are in turn based on the 1980 and 1950 census data. In conducting part of our analysis, we also match the 1980 census data with 1930 data obtained from a 1937 census report on the American-Indian population.

We restrict the sample to counties that contain an American-Indian reservation and in which American-Indian population constitutes at least 5 percent of the total county population in the year 1980. The choice of the 5 percent threshold reflects the following tradeoff. On the one hand, restricting our analysis to counties with a larger share of American-Indian population significantly reduces the size of our sample. On the other hand, inclusion of counties with a smaller share of American-Indian population increases the size of our sample, but dilutes the effect of PL280, which governs legal matters on American-Indian reservations. Below, we clarify that our results are robust to varying this threshold level.

To determine which counties contain reservations, we use the 1990 Census, which breaks down American-Indian population for each reservation in the United States by county, along with a U.S. Geological Survey map overlaying reservations and counties. These considerations, together with the availability of data for the variables used in our regressions, result in a sample of 78 counties, listed in Table A6 in the Online Appendix.

We use two dependent variables: the number of all crimes and the number of property crimes for the year 1981. As already noted, 1981 is the earliest year after the period of PL280 adoption for which we have sufficiently extensive data on crime and other explanatory variables.Footnote ⁸ The year 1981 also marks the period before the emergence of organized tribal law enforcement agencies and casinos, both of which could blur the impact of PL280.

As with any crime data, there are caveats about the accuracy of these data. On the one hand, the collection of crime data by law enforcement agencies in Indian country has not been systematic and has been relatively more incomplete in non-PL280 than in PL280 areas (Reference Goldberg, Champagne and SingletonGoldberg, Champagne, & Singleton 2008: 22). Yet, on the other hand, American-Indian residents in PL280 areas have, in comparison to American-Indian residents in non-PL280 areas, been known to systematically underreport crime due to lack of trust in county and state law enforcement agencies. According to a recent comprehensive analysis of crime in Indian country, “[t]ribal members report crimes about equally to tribal police and federal-BIA police, but significantly fewer are willing to report crimes to Public Law 280 state or county police” (ibid.: 318). Therefore, while our (county-level) data likely underestimate the level of crime in both PL280 and non-PL280 areas, it is not clear whether the data underestimate crime levels in the PL280 areas to a greater or to a lesser extent than in the non-PL280 areas. Thus, any estimate of the impact of PL280 on crime could either underestimate or overestimate the law's true effect on the occurrence of crime and should be treated with caution.

Our focal explanatory variables are indicator variables describing a county's PL280 status. We discuss these and our control variables as we introduce different empirical specifications below. Table A2 in the Online Appendix summarizes our variables and sources of data. Table A3 presents summary statistics.

OLS

We first examine the association between a county's PL280 status and crime by estimating the following model with OLS:

(1) $c_i=\alpha +{\beta }_1\cdot MandPL280_i+{\beta }_2\cdot OptPL280_i+{X^{\prime }}_i\gamma +u_i\text{}.$

In specification (1), c_i is either the logged number of all crimes or logged number of property crimes in county i in the year 1981. The process of adoption of PL280 in “mandatory” states differed from that in “optional” states. Optional states adopted the law at various times between 1953 and 1968 and asserted state jurisdiction to a varying extent, that is, not necessarily over all criminal and civil matters (see Table A1). We thus allow for the impact of PL280 to vary depending on whether a county lies in a “mandatory” or in an “optional” state and, if a county lies in an optional state, according to the extent to which PL280 applied in that state. MandPL280_i is an indicator variable equal to 1 if county i adopted PL280 and the county lies in a “mandatory” state; and 0 otherwise. OptPL280_i is an indicator variable equal to 1 if county i adopted PL280 and the county lies in one of “optional” states in which state jurisdiction granted by PL280 applied broadly to criminal and civil matters (i.e., Florida, Montana, Nevada, and Washington, but not Arizona, South Dakota, or Idaho; see Table A1); and 0 otherwise. Our findings are robust to an alternative coding of the OptPL280_i variable. Table A6 in the Online Appendix provides a detailed coding of counties' PL280 status.

X_i is a vector of control variables, which includes a county's total population in 1980, local government's expenditures on police in 1977, and three additional variables: number of persons that completed at least a high school degree, number of unemployed, and median family income. It is possible that PL280, much like other legislation affecting the lives of American Indians, influenced educational and labor market outcomes, and thus family incomes (see, e.g., Reference GrossGross 1979). To avoid the bias associated with inclusion of controls that could themselves be outcome variables (Reference Angrist and PischkeAngrist & Pischke 2009: 64–66), we therefore measure these three additional control variables in 1950, before the PL280 “treatment” took place. All of the control variables in vector X_i enter the regression in logged form. u_i is the error term.

In all of the regressions, we base our inference on heteroscedasticity-robust standard errors, clustered at the state level. Clustering of error terms at the state level allows for the plausible nonzero correlation between error terms for counties from the same state.

Table 1 presents the results from estimating model (1). Both the coefficient on MandPL280_i and the coefficient on OptPL280_i are positive and statistically significant. Under a causal interpretation, the magnitude of the effect of PL280 status on crime is noteworthy. Based on the estimates in columns (4) and (8), which feature specifications with a full set of controls, holding all else equal, mandatory PL280 status on average increased the volume of all crimes by nearly 130 percent (column (4)) and the volume of property crimes by about 140 percent (column (8)). Ceteris paribus, optional PL280 status on average, increased the volume of all crimes by more than 60 percent (column (4)) and the volume of property crimes by more than 35 percent (column (8)).

Table 1. OLS Estimates for Whole Sample, Cross Section

Notes: The table reports results based on OLS regressions. Dependent variable is log of number of all crimes in 1981 for regressions in panel A, and log of number of property crimes in 1981 for regressions in panel B. The 1950 controls (all logged) are: persons completed high school in 1950, unemployed in 1950, and median family Income in 1950. Heteroscedasticity-robust standard errors clustered at state level in parentheses. ^*, ^**, and ^*** indicate significance at the 10, 5, and 1% levels, respectively.

The differential impact of mandatory versus optional PL280 status on crime is statistically significant in regressions in columns (4) and (8), respectively. This resonates with the fact that, in contrast to the adoption of PL280 in mandatory states, the implementation of PL280 in optional states was an outcome of a political process at the state level, which plausibly better incorporated local knowledge and more adequately reflected local resources and needs, with a less adverse societal impact as a consequence.

To meaningfully attribute causal interpretation to OLS estimates β ₁ and β ₂ in specification (1), an area's PL280 status would have to be exogenous and, therefore, uncorrelated with the error term. This is a strong and, we argue, very likely inaccurate assumption. As discussed earlier, historical evidence suggests that the initial choice of areas mandated to adopt PL280 was based on careful deliberation about the perceived extent of American Indians' “readiness” for further socioeconomic integration, the willingness of state and local authorities to assume jurisdiction over Indian country, and control of lawlessness. In the following section, we therefore propose an instrumental variable approach to estimate the effect of PL280 on crime.

Instrumental Variable Approach

As argued earlier, PL280 was passed in 1953 at the height of the Termination Period in support of an assimilationist agenda (Reference Jiménez and SongJiménez & Song 1998). The transfer of jurisdiction from the federal to state courts was mandatory for California, Minnesota, Nebraska, Oregon, and Wisconsin (see Table A1 for exceptions). These areas constitute our “treatment” group. Optional states adopted PL280 years later at the initiative of state legislatures rather than Congress—and, hence, as a result of a different political dynamics—and to a varying extent. To obtain a clear “comparison” group vis-à-vis the areas initially assigned to adopt PL280, we therefore exclude from our sample the counties lying in those “optional” states in which state jurisdiction applied broadly to criminal and civil matters (i.e., “optional” PL280 counties in Florida, Montana, Nevada, and Washington for which the OptPL280_i dummy, defined above, takes on the value 1) and estimate the following model:

(2) $c_i=\alpha +\beta \cdot MandPL280_i+{X^{\prime }}_i\gamma +u_i\text{}.$

As in specification (1), MandPL280_i in (2) is an indicator variable equal to 1 if county i adopted PL280 and the county lies in a “mandatory” state; and 0 otherwise. The vector of controls X_i is also the same as in specification (1).

Our instrumentation strategy draws directly on the historical and political context surrounding the enactment of PL280. As a source of exogenous variation in counties' mandatory PL280 status, we suggest a variable measured at the county level approximately two decades before the law's enactment: the 1930 ratio of full-blooded American Indians to all American Indians. To this end, we match the 1980 census data used in the previous section with data obtained from a 1937 census report on the American-Indian population. The 1937 report, inter alia, contains 1930 county-level data on the number of full-blooded American Indians and the total number of American Indians in a county.Footnote ⁹

We expect a county's mandatory PL280 status to be negatively correlated with the 1930 ratio of full-blooded American Indians to all American Indians. Deliberating on the course of actions taken during the Termination Period, Reference WatkinsWatkins (1957), for example, argues that “[e]xperience developed in carrying out the legislation adopted by the … Congress for freedom from special federal control over Indians … has shown that other factors being equal, the … rather well assimilated tribes … appear the more likely subjects for prompt release from federal controls” (ibid.: 52). In fact, when laying out his plan for the withdrawal of Federal supervision, just a few years before the passage of PL280, the Acting Commissioner of Indian Affairs William Zimmerman argued that the key criterion to determine whether a tribe was ready for “freedom from Federal supervision” would be “the degree of individual acculturation” (Reference HerzbergHerzberg 1977: 303). Moreover, Zimmerman's first criterion (the degree of tribal acculturation) was based upon his assessment of “such factors as the admixture of white blood [emphasis added], the percentage of illiteracy, the business ability of the tribe, their acceptance of white institutions, and their acceptance by the whites in their community” (ibid.: 307).

All else equal, at the dawn of its enactment in 1953, PL280 was therefore more likely mandated in areas where tribes were deemed “ready for complete freedom from Federal supervision and wardship” as “Indians had reached a state of acculturation and development that made them amenable to these changes” (Reference Jiménez and SongJiménez & Song 1998: fn 206).

At the same time, the 1930 ratio of full-blooded American Indians to all American Indians should not affect the level of crime in 1980 through any channel other than PL280. This conjecture is particularly plausible since we control for the 1950 level of human capital (number of persons who completed at least a high school degree), number of unemployed, and median family income—variables that could have perhaps been affected by the extent of American Indians' integration into the mainstream society and, at the same time, put a given geographic area on a specific crime path. We argue, therefore, that our instrumental variable for mandatory PL280 status plausibly satisfies the exclusion restriction.

In addition to instrumenting for the mandatory PL280 status, we address the likely endogeneity of another variable, which enters our model (2) as a control: local government expenditures on police. The level of police expenditures in an area reflects actual and anticipated extent of crime, and is therefore not exogenous to the volume of crime. We suggest that the level of local government expenditures on highways represents a source of exogenous variation in local government expenditures on police. Expenditures on highways should not directly affect crime, thus satisfying the exclusion restriction. At the same time, a local government's expenditures on highways are highly correlated with its expenditures on police, rendering our instrument relevant.

As a benchmark, the left column in each of the panels of Table 2 presents OLS estimates of model (2) using the same sample of observations as used in the IV (2SLS) estimation discussed below. The implied magnitude of the effect of mandatory PL280 status on crime is very similar to that reported in the previous section (see columns (4) and (8) of Table 1).

Table 2. OLS and IV (2SLS) Estimates for Sample Excluding Counties for which Optional PL280 = 1, Cross Section

Notes: The table reports results based on OLS and IV (2SLS) regressions for the sample excluding counties for which optional PL280 = 1 (see Table A6 in the Online Appendix). Dependent variable is log of number of all crimes in 1981 for regressions in panel A, and log of number of property crimes in 1981 for regressions in panel B. The 1950 controls (all logged) are: persons completed high school in 1950, unemployed in 1950, and median family income in 1950. The endogenous regressors are mandatory PL280 and log of local government expenditure on police in 1977. Heteroscedasticity-robust standard errors clustered at state level in parentheses. ^*, ^** and ^*** indicate significance at the 10, 5, and 1% levels, respectively.

The right column in the two panels of Table 2 presents the IV (2SLS) estimates when we instrument for both a county's mandatory PL280 status and local government expenditures on police. The estimate of β is statistically significant when using either log of all crimes or log of property crimes as the outcome variable, and is more than double the size of the OLS estimates reported in the left column. Under a causal interpretation, the effect of mandatory PL280 status on the volume of all crimes and property crimes is therefore large in magnitude. Turning to other coefficients reported in the IV (2SLS) column of Table 2, the number of all crimes and the number of property crimes, as expected, both increase with total population in 1980 and both decrease with local government expenditures on police in 1977.

The first-stage regressions (see Table 3) indicate that, as conjectured, the likelihood that the geographic area was mandated to adopt PL280 is negatively and statistically significantly associated with the 1930 ratio of full-blooded American Indians to all American Indians. Similarly, local government expenditures on police in 1977 are positively and statistically significantly associated with the local government expenditures on highways in the same year. The Kleibergen-Paap rk Wald F-statistic of 4.75 suggests that our instruments are, as a group, reasonably strong in a statistical sense: For two endogenous regressors and two excluded instruments, the Stock–Yogo critical value obtained based on the assumption of i.i.d. errors is 4.58 for 15 percent maximum test size.

Table 3. First-Stage Regressions for IV (2SLS) Regressions in Table 2

Notes: The table reports first-stage regressions for the IV (2SLS) regressions in Table 2. The Kleibergen-Paap rk Wald F-statistic is a robust analog of the Stock and Yogo (Reference Stock, Yogo, Andrews and Stock2005) Wald F-statistic in the presence of heteroscedasticity, autocorrelation, or clustering (see Baum, Schaffer, and Stillman Reference Baum, Schaffer and Stillman2007). According to Stock and Yogo (Reference Stock, Yogo, Andrews and Stock2005), the instruments are considered weak if the conventional 5% Wald test based on 2SLS statistics has an actual size that could exceed a certain threshold (maximum test size). Under weak instruments, the conventional Wald test thus rejects the null hypothesis that the coefficients of endogenous regressors are equal to zero too frequently (see Baum, Schaffer, and Stillman Reference Baum, Schaffer and Stillman2007). For two endogenous regressors and two excluded instruments, the Stock–Yogo critical values obtained under the assumption of i.i.d. errors are 7.03 (for 10% maximum test size), 4.58 (for 15% maximum test size), 3.95 (for 20% maximum test size), and 3.63 (for 25% maximum test size). The 1950 controls are as listed in notes under Tables 1 and 2. Heteroscedasticity-robust standard errors clustered at state level in parentheses. ^*, ^** and ^*** indicate significance at the 10, 5, and 1% levels, respectively.

The increase in the magnitude of the estimate of β when moving from OLS to IV (2SLS) estimation (see Table 2) suggests a downward bias in the OLS estimates of the impact of mandatory PL280 status on crime. The downward bias in OLS results is consistent with the endogenous nature of PL280 as documented by the debates surrounding the law's enactment. In particular, Congressional records indicate that the five mandatory states were, inter alia, selected because the respective state and local officials had agreed to the implementation of PL280; conversely, states where politicians and local representatives objected to assuming jurisdiction over Indian country without a federal support (e.g., Nevada, South Dakota, Montana) were excluded from the list of mandatory states (see H.R. Report No. 848). Since a state's readiness to adopt PL280 is indicative of the authorities' ability and willingness to provide effective law enforcement, which in turn reduces the incidence of crime, a failure to address the issue of selection of mandatory PL280 areas underestimates the law's impact on crime.

Sensitivity Analysis Results

To investigate the sensitivity of our results, we conducted several robustness checks. We summarize our findings.Footnote ¹⁰

First, we explored the consequences of an alternative coding for the OptPL280_i dummy variable such that this variable takes on the value 1 if county i adopted PL280 and the county lies in any of the “optional” states, regardless of the extent of state jurisdiction granted by PL280; and 0 otherwise. Unsurprisingly, the coefficient on the OptPL280_i dummy in our OLS regressions on the whole sample becomes statistically insignificant. In contrast, the adverse effect of mandatory PL280 status on crime is fully robust to this recoding: The coefficient on MandPL280_i remains statistically significant and positive in all regressions presented in earlier sections.

Second, we reran all regressions by gradually increasing the threshold on the share of American-Indian reservation population from 5 to 15 percent. While sample size decreases notably as a consequence, the effect of PL280 on crime is still positive, statistically significant, and large in magnitude.

Third, we examined the robustness of our results to the inclusion of Oklahoma counties in our sample and the exclusion of the Menominee County, Wisconsin, from our sample. Oklahoma, where reservations have an ambiguous legal status, never adopted PL280 (Reference TinkerTinker 2011). The Menominee county of Wisconsin contains the Menominee Reservation, which found itself both terminated and reinstated during the period of our study (see Table A1). Our results are fully robust to these changes in the sample.

Fourth, we checked the sensitivity of our results to the inclusion of total urban population in 1980 as an additional control variable, which could plausibly determine the prevalence of crime. Our results about PL280's impact on crime remain unchanged. The coefficient on 1980 total urban population is statistically insignificant in all regressions.

Finally, we analyzed the sensitivity of our IV (2SLS) results by replacing, as well as combining, the 1930 ratio of full-blooded American Indians to all American Indians, our chosen instrument for mandatory PL280 status, with two other plausible instrumental variables available in the 1937 census report: the 1930 ratio of American Indians who do not speak English to all American Indians, and the 1930 ratio of illiterate American Indians to all American Indians. Our findings on PL280's impact on crime remain unchanged as a result.

Data

To examine PL280's impact on economic development, we construct a panel dataset of counties using each decennial census from 1950 to 1980, the time period before and after PL280's enactment, and prior to the emergence of casinos (which could obscure the effect of PL280). Much like in the case of constructing the cross section, we employ the electronic versions of county-level data publicly available through ICPSR.

Our outcome of interest in this section is the level of economic development, which PL280 likely impacted less directly than the occurrence of crime studied earlier. The panel structure of the data increases the size of our sample. Our sample-censoring threshold on the percentage of American-Indian population is therefore naturally set higher in this section than in our earlier cross-sectional analysis of crime. Specifically, we restrict our sample to counties that had reported at least 15 percent American-Indian population in the year 1980 and which contain an American-Indian reservation. (Below, we clarify that our results are quite robust to varying this threshold level.) These considerations, together with the availability of data for variables used in our analysis, give rise to a balanced panel of 35 counties for each of the four census years (see Table A7 in the Online Appendix).

Our dependent variable, measuring the level of economic development, is median family income. Median family income is the only consistently reported income-related variable in the Census across the four decades (i.e., in years 1950, 1960, 1970, and 1980) covered by our panel.

As already noted, the degree to which state jurisdiction was extended under PL280 varied across the “optional” states. Accordingly, our focal explanatory variable, PL280, is an indicator variable that takes on the value of 1 if, in a given year, PL280-granted state jurisdiction in a county applied broadly to criminal and civil matters, and 0 otherwise. (Below, we explain that our results are fully robust to an alternative coding of the PL280 dummy.) This variable, therefore, conveys information both about the year of adoption of PL280 and, if applicable, the year of retrocession. Specifically, the value of 0 for a given county in a specific year could indicate either that the state in which the county lies has not (yet) adopted PL280 or, alternatively, that the county retroceded after the 1968 amendments to the law (see Table A7).

We also control for the level of human capital (measured by percent of people with at least a high school degree), which plausibly affects the level of economic development. Table A4 in the Online Appendix describes our variables. Table A5 presents summary statistics.

Pooled OLS

We first explore the association between a county's PL280 status and median family income using the following specification:

(3) $y_{it}=\alpha +\beta \cdot PL280_{it}+\gamma \cdot X_{it}+{\lambda }_t+u_{it}\text{},$

where y_it is the log of median family income in county i in time period t, PL280_it is a dummy equal to 1 if PL280 applies broadly to county i in time period t, and 0 otherwise, X_it is a vector of controls, and λ_t is the year fixed effect, implemented in the estimation with the inclusion of year dummies. Year dummies, inter alia, absorb the changes in the aggregate price level during the period of our study. This allows us to interpret the effect of right-hand-side variables in (3) on median family income as a real, rather than nominal, effect (see Reference WooldridgeWooldridge 2009: 448). α is the regression constant and u_it is the error term. The coefficient of interest is β.

We estimate equation (3) using pooled OLS. For all of the regression models we estimate, we base our inference on cluster robust standard errors (see, e.g., Reference Cameron and TrivediCameron & Trivedi 2005; Reference WooldridgeWooldridge 2003). To allow for the conceivable nonzero correlation between error terms for counties from the same state, we cluster errors at the state level.

Columns (1) and (2) in Table 4 report the results. When we control for year fixed effects only (column (1)), the estimate of β indicates a positive, albeit statistically insignificant, association between a county's PL280 status and income. Additionally controlling for the level of human capital (column (2)) renders the coefficient on PL280_it negative; the estimate of β, however, is still statistically insignificant.

Table 4. Pooled OLS and FE Estimates, Panel

Notes: The table reports regression results based on pooled OLS (columns (1) and (2)) and fixed effect (columns (3) and (4)) estimation. The dependent variable is log of median family income. Heteroscedasticity-robust standard errors clustered at state level in parentheses. ^*, ^**, and ^***indicate significance at the 10, 5, and 1% levels, respectively.

Fixed Effects Specifications

To be able to give a causal interpretation to coefficient β in model (3), the error term u_it must be uncorrelated with PL280_it and X_it. This is a strong assumption that is unlikely to hold given the law's context. As a result, an estimate of β based on pooled OLS estimation is likely to be biased. To better address the endogenous nature of a county's PL280 status, we posit instead the following empirical model:

(4) $y_{it}=\beta \cdot PL280_{it}+\gamma \cdot X_{it}+{\lambda }_t+{\mu }_i+{\epsilon }_{it}\text{},$

where y_it, PL280_it, X_it, and λ_t are as defined in specification (3).

In specification (4), we split the error term into two components: μ_i and ε_it. μ_i is the time-invariant component of the error term. μ_i captures county-level, time-invariant (“fixed”) factors that we do not directly control for, such as, for example, the availability of natural resources and other geography-based determinants of economic development (Reference Anderson and ParkerAnderson & Parker 2008), historical determinants of reservations formation (Reference Anderson and LueckAnderson & Lueck 1992; Reference DippelDippel 2011), tribal culture and governance mechanisms (Reference Akee, Jorgensen and SundeAkee, Jorgensen, & Sunde 2012; Reference Cornell and KaltCornell & Kalt 2000; Reference Pickering and BiolsiPickering 2004), as well as the constituency's—local authorities' and tribes'—willingness to adopt PL280. Under this fixed effects specification, consistent estimates of parameters are obtained even when the time-invariant, county-level unobserved heterogeneity μ_i is correlated with the right-hand-side variables in equation (4). ε_it is the time-varying component of the error, assumed to be uncorrelated with right-hand-side variables in equation (4). (We relax this assumption below.)

We estimate model (4) using the least-squares-dummy-variable estimator with a full set of county dummies, an approach equivalent to using the fixed effects “within” estimator (see, e.g., Reference WooldridgeWooldridge 2002). Column (3) of Table 4 reports the results. In contrast to the results based on the naïve pooled OLS (columns (1) and (2)), the coefficient on PL280_it is now both negative and statistically significant. Under a causal interpretation, the estimate of β in column (3) suggests that, ceteris paribus, PL280 status on average decreased median family income in a county by nearly 16 percent.

Finally, to address the possibility that changes in unobserved county-level characteristics are correlated with both PL280 status and median family income, we estimate a model with county-specific time effects. The inclusion of a full set of county-year dummies is not possible since doing so would eliminate degrees of freedom. Instead, we posit the following specification with county-specific linear time trend μ_it:

(5) $y_{it}=\beta \cdot PL280_{it}+\gamma \cdot X_{it}+{\lambda }_t+{\mu }_i+{\mu }_{i}t+{\epsilon }_{it}\text{}.$

Column (4) of Table 4 reports the results from estimation of model (5). The estimate of β remains statistically significant, negative, and in terms of magnitude, virtually identical to the estimate obtained based on specification (4) (see column (3)).

General-Method-of-Moments Instrumental Variables (GMM-IV) Approach

In an attempt to estimate the effect of PL280 on income, the fixed effects specifications (4) and (5) control for time-invariant county-level heterogeneity as well as the level of human capital proxied by the percent of population that completed high school. Specifications (4) and (5), however, do not allow for the possibility of correlation between the right-hand-side variables, in particular PL280_it, and the time-varying component of the error term ε_it. Yet, PL280 status could plausibly be endogenous to some aspect of time-varying unobserved heterogeneity which, given available data, we cannot directly control for. This could raise concerns about possible bias of our estimates reported in columns (3) and (4) of Table 4.

Similarly, specifications (4) and (5) do not allow for persistence in the evolution of median family income over time. To the extent that the choice of the PL280 status could have been systematically related to past levels of economic development, controlling for median family income in the previous period further alleviates any bias due to endogenous selection of areas subject to PL280.

We therefore specify the following dynamic model:

(6) $y_{it}=\rho \cdot y_{i,t-1}+\beta \cdot PL280_{it}+\gamma \cdot X_{it}+{\lambda }_t+{\mu }_i+{\epsilon }_{it}\text{},$

where y _i,t−1 is the lagged value of logged median family income. All other components of specification (6) match those featured in specification (4).

The presence of lagged dependent variable among the set of explanatory variables in (6), and the possibility that PL280_it is correlated not only with μ_i but also with ε_it, violates the assumption of strict exogeneity of regressors. Lack of strict exogeneity of regressors renders the fixed effects estimator biased and inconsistent (see, e.g., Reference NickellNickell 1981). To obtain consistent estimates of the coefficients, in particular β, we estimate model (6) using the GMM-IV approach proposed by Reference Arellano and BondArellano and Bond (1991).

To apply the GMM-IV approach, we first eliminate the county fixed effects (μ_is) by first-differencing expression (6):

(7) $\Delta y_{it}=\rho \cdot \Delta y_{i,t-1}+\beta \cdot \Delta PL280_{it}+\gamma \cdot \Delta X_{it}+\Delta {\lambda }_t+\Delta {\epsilon }_{it}\text{}.$

While expression (7) is free of county fixed effects, Δy _i,t−1 is, by construction, correlated with Δε_it (because y _i,t−1 is correlated with ε _i,t−1). ΔPL280_it and ΔX_it are also correlated with Δε_it if a county's PL280 status and level of human capital, respectively, are contemporaneously correlated with the error term ε_it in equation (6) because of some unobserved time-varying, county-level heterogeneity or even simultaneity. As long as the PL280 status and the level of human capital are uncorrelated with the future realizations of the error term (i.e., they are “weakly exogenous”), the second and deeper lags of Γ_it ≡ (y_it, PL280_it, X_it) are uncorrelated with Δε_it, and thus all available as instruments.Footnote ¹¹

Under the assumptions of no serial correlation in the error term ε_it, the Reference Arellano and BondArellano and Bond (1991) “difference” GMM-IV estimator thus relies on the following moment conditions: E[Γ _i,t-s(ε_it−ε _i,t−1)] = 0 for s ≥ 2. Because our panel consists of four cross sections only, the implied number of available internal instruments is necessarily small. Our analysis, therefore, does not suffer from the bias-inducing “instrument proliferation problem” (Reference RoodmanRoodman 2009).

Table 5 reports our results based on the GMM-IV approach.Footnote ¹² The coefficient on the PL280 indicator variable is negative and statistically significant in both specifications in Table 5. We do not find evidence on persistence of income: the coefficient on the lagged dependent variable is statistically highly insignificant (see column (1)). The Hansen test of overidentifying restrictions is indicative of the appropriateness of our instruments in both specifications in Table 5. When computed, the (Arellano–Bond) test of serial correlation further supports our instrumental variables approach (see column (2)).

Table 5. GMM-IV Estimates, Panel

Notes: The dependent variable is log of median family income. Columns (1) and (2) report results using the Arellano and Bond (Reference Arellano and Bond1991) difference GMM estimation. The instrument set for the regression in column (1) is based on the second and deeper lags of logged median family income, PL280, and percent completed high school, and differenced year dummies. The instrument set for the regression in column (2) is based on the second and deeper lags of PL280 and percent completed high school, and differenced year dummies. The heteroscedasticity-robust standard errors reported in parenthesis are clustered at the state level and calculated using the Windmeijer (Reference Windmeijer2005) correction. ^*, ^** and ^*** indicate significance at the 10, 5, and 1% levels, respectively.

^a Some of the potential instruments were dropped because of collinearity.

^b The null hypothesis is that the instruments are not correlated with residuals.

^c The null hypothesis is that the errors in the first-difference regression exhibit no second-order autocorrelation.

In sum, even when allowing for correlation between a county's PL280 status and the time-varying component of the error term (see above), we find that PL280 status decreases median family income. Based on the estimates reported in Table 5, under a causal interpretation, PL280 status ceteris paribus decreased median family income on average between 34 percent (column (2)) and 36 percent (column (1)).

Our findings, thus, do not resonate with the conjecture that PL280 spurred economic development (Reference Anderson and ParkerAnderson and Parker 2008). This, of course, does not preclude the possibility that the increased involvement of state courts in civil matters under PL280 had some beneficial effects (see, e.g., Reference CooksonCookson 2012; Reference ParkerParker 2012). Based on our analysis, which, in contrast to Anderson and Reference Anderson and ParkerParker's (2008) work, carefully addresses the endogenous nature of PL280, the adverse effects of PL280 through other channels, in particular an increase in crime and lawlessness, seem to have outweighed any positive effects of the law on economic development associated with the transfer of civil jurisdiction to state courts, at least in the period up to 1980 (prior to emergence of casinos).

Sensitivity Analysis Results

We again performed several robustness checks of our results. First, because the degree to which state jurisdiction over criminal and civil cases in “optional” state was extended under PL280 was uneven, we coded an alternative PL280 indicator variable which takes on the value of 1 whenever state jurisdiction granted under PL280 applied to at least some kind of criminal and civil cases (see Table A1).Footnote ¹³ Our results based on this alternating coding of the PL280 indicator variable are quantitatively very similar and qualitatively identical to the ones reported in the sections above.

Second, to allow for the possibility that the impact of PL280 was different in mandatory versus optional states, we examined additional specifications where we split our PL280 dummy variable into two dummy variables: mandatory PL280 and optional PL280. Mandatory PL280 equals 1 if a county in a given time period was under PL280 status in a mandatory PL280 state (see Table A1), and 0 otherwise. Optional PL280 equals 1 if in a given time period a county was under PL280 status in an optional PL280 state that applied PL280 broadly, and 0 otherwise. The pattern of statistical significance of the coefficients on mandatory PL280 and optional PL280 dummies varied across specifications and estimation methods. Moreover, the coefficients on mandatory PL280 and optional PL280 were never statistically significantly different from each other. Thus, while the association between PL280 and median family income is robustly negative and statistically significant in all preferred specifications, we do not find robust evidence that in the period up to 1980 the adverse impact of PL280 on the level of economic development was different in mandatory versus optional PL280 areas.

Third, we reran all regressions for different levels of the threshold on the share of American-Indian reservation population. Under the 20 percent threshold, the effect of PL280 remains highly statistically significant and of an order of magnitude similar to that noted in the previous sections. Under the 10 percent threshold, the effect of PL280 is, as expected, somewhat diluted, but remains negative and statistically significant.

Finally, much like in the sensitivity analysis described in our examination of PL280's impact on crime, we examined the robustness of our results to the inclusion of Oklahoma counties in our sample and the exclusion of the Menominee County, Wisconsin, from our sample. Our results on the impact of PL280 on the level of economic development are virtually unaffected by these changes in the sample.