is an alternative version of White estimator and intended to compute heteroskedasticity and
serial correlation consistent standard errors for fixed effects models. Note that Levin, Lin, and
Chu’s unit root test for panel data (Levin, Lin and Chu, 2002) indicates that our dependent
variables are stationary.
5
Findings
We begin by testing the effects of Black and Latino population size on per capita welfare
expenditures in the states. Table 2 reports the estimated effects of the racial environments
when only the percentages of Black and Latino residents are included in the model (Column
1), when Black and Latino group size is interacted with our indicator of voter registration on
election day (Column 2), when the interactions between Black and Latino group size and the
income ratios are included (Column 3), and when Black and Latino group size is interacted
with the share of Black and Latino representatives in the state legislature (Column 4). Table 3
presents estimates using the same specification for per capita education expenditures. The
threat or backlash hypothesis suggests that the percentage of Blacks and Latinos with a state is
negatively related to each spending measure. Our conditional hypotheses lead us to expect that
the interactions between Black and Latino group size and the income ratios will be positive.
Likewise, the interaction between Black and Latino group size and the level of descriptive
representation is hypothesized to have a positive effect, while the the interaction of Black and
Latino group size with voter registration may potentially be either positive or negative.
[Tables 2 and 3 Here]
As expected, Column 1 in Tables 2 and
3 reports that the percentage of Blacks and
Latinos within a state has a negative impact on welfare and education expenditures. Except
for the effect of Black population size on education expenditures, the estimates are statistically
discernible from zero. As minority group size increases, state governments allocate less money
to welfare and education programs.
Is the relationship between the racial/ethnic context and policy outputs conditional on
the degree of minority electoral influence? Tables 2 and 3 also present evidence to address
this question. After the interaction terms are included in the models, most of the coefficients
5
We included a trend variable in the test.
10
Convention