change in earned income tax (B 2 ), the value of R 2 adjusted for 

 degrees of freedom, and the sample size (n). Values of the t statistic 

 appear in parentheses. Each column contains a regression result at 

 the MCD level, the county level, and the regional level respectively. 

 This format will be followed in later tables. The same basic data is 

 used for each of the three models; minor civil division data is aggre- 

 gated into its respective county units and the regional data is 

 derived by aggregating the appropriate county units. The observa- 

 tion of the county level is the mean (X) of the MCD level data; 

 likewise the regional mean is an average of its constituent county 

 means. 



Table 1. Summary of Regressions at Three Levels 

 of Aggregation for Equation 1. 



♦Significant at .001. 



'Numbers in parentheses are simple t statistics. 



The coefficient B Y is positive, less than one, and significant in 

 each of the three regressions. The coefficient B 2 behaves a little more 

 peculiarly. In all cases it is negative; in the first two it is fairly small 

 and in the third it is quite large. At the regional level our best 

 estimate is that for every dollar increase in earned income taxes, per 

 capita tax collections fall by $1.43. However, only at the MCD level 

 is B 2 found to be significantly different than zero. Looking at either 

 the county or regional results (only), it is clear that the hypothesis, 

 B 2 = 0, could not be rejected. In theory this is the same B 2 that exists 

 at the MCD level or at the "community" level. Yet at the MCD level it 

 is clear that we would reject the hypothesis that B 2 = 0. The issue is 

 not whether one conclusion or the other is more correct; our point is 

 simply that our conclusion depends in part on the level of aggrega- 

 tion that we select. Even when we adjust for the loss of degrees of 



