Indian gaming and Tribal Revenue Allocation Plans: Socio-economic determinants of policy adoption

Abstract

As the Indian gaming industry has experienced unprecedented growth over the past two decades, tribes have pursued different paths regarding the utilization of gaming revenues within parameters established by the Indian Gaming Regulatory Act. Since 1993, more than 100 tribes have received approval through the Department of the Interior to distribute revenues directly to tribal members through per capita payments governed by a Tribal Revenue Allocation Plan (RAP). This paper improves our understanding of nations with payment plans by exploring whether socio-economic tribal features are associated with the successful adoption of a RAP. We find that tribes who gained approval of a RAP in the 1990s have higher per capita incomes, while also having smaller populations and lower levels of educational attainment. Population is the strongest predictor of RAP adoptions in both the 1990s and 2000s, with the impact of other tribal features being less meaningful in explaining adoption in the second decade.

Notes

1 The BIA provided a list of nations with a RAP and the date of Interior Department approval via email correspondence (October 15, 2009). This data is available upon request.

2 There are eight tribes where population is missing in 1990 and four others in 2000. We include these cases in the subsequent analysis as valid values and decade permit.

3 We also evaluated the impact of unemployment and family poverty on adoptions, both correlated noticeably with per capita income. The results are similar but attenuated when either is employed in lieu of income but the equations are unstable when one or both are included with income. Our assessment also considered several other tribal characteristics including urbanization, percent non-Indian, percent of individuals living in crowded homes, child poverty, and high school dropout rate, which were either correlated with one of the other independent variables or unrelated to adoption.

4 Thus, for tribes adopting gaming in the 1990s, the first decade in our analysis, we subtract the year they signed a gaming compact from 2000. For example, a tribe signing a compact in 1993 we code as having legalized Class III gaming regardless of whether they have an operating casino or not for seven years in the first decade of our analysis. For the second decade, we subtract the compact year from 2008, the last year of available RAP data.

5 The best example of this concerns per capita income, where three tribes have per capita incomes greater than $50,000 in 2000, which was considerably higher than their incomes in 1990. When excluding these cases, the standard deviation decreases from $14,606 to $5448 in 2000. All three tribes adopted Class III gaming in the 1990s, and two adopted a RAP in the same decade, while the third adopted one the next decade. Although the first two do not pose a problem because we drop them when examining the second wave of RAP adoptions, the remaining case is an influential outlier, though it has minimal impact on the multivariate results. On a more general level, we are sensitive to the possibility of a few cases skewing our results but detect no problems in this regard. Although it is not possible to verify, this case suggests that some nations have had more than one plan; it is difficult to imagine a tribe with a six figure per capita income, which is true of this case in 2000, without making payments to their members.

6 Upon examining the residuals for the models we discuss herein, we report robust standard errors in light of the non-random nature of the error terms at higher levels of the independent variables. To evaluate the predictive power and overall fit of the different models, we use the percent of cases correctly classified as well as the likelihood ratio and Wald test. Variation explained, the idea behind R², is not relevant in logistic regression, though a large number of pseudo measures are available. When considering six options for the results in Table 2, for instance, the values range between 0.10 and 0.76, with an average of 0.23. We report Cragg and Uhler's R² statistic, which falls around the average value of the six options.

7 Sixteen tribes that adopted Class III gaming in the late 1990s had adopted RAPs prior to signing a compact in our dataset, indicating these tribes adopted RAPs while operating Class II casinos. All but four of these tribes are in California, where there was a difficult and prolonged legal battle surrounding casino-style gaming throughout the decade that eventually led to the adoption of Class III gaming in 1999. The inclusion of a control variable for California tribes in the multivariate models was insignificant. Dropping these 16 cases, another option examined, results in the loss of roughly one-fourth of RAP adopters in our analysis and greatly affects the results.

8 It is for these reasons we do not pool the data for analysis. Another argument against pooling is due to differences in RAP regulations that changed at the beginning of the second decade that may have appreciable effects on both the pursuit and approval of payment plans. The estimation of a pooled model, with a dummy variable for gaming decade, generates results largely consistent with what we report but obscures the differential findings by decade.

9 Given the measurement scales and distributional skewness, we convert per capita income to 1,000s and population to the base-10 logarithm.

10 A test of multicollinearity for both models reveals little reason for concern; the tolerance measures range from 0.84 to 0.94 and 0.68 to 0.92, respectively. Given the non-random nature of the data based on population and not a sample, a test of statistical significance is not appropriate but helps to detect relationships suggesting greater impacts. Furthermore, while three of our variables have expected relationships with RAP adoption, which would imply a one-tail test of significance, we use a more stringent two-tail test, making it harder to reject the null.

11 Given the lackluster performance of the year variable in Table 2 and the grouping of tribes by decade of gaming's adoption, we exclude the number of compact years.

12 Part of the reason why several of the important variables failed to achieve statistical significance in the logistic regression models is the small number of cases for both sets of analyses when censoring the data.