impacts of aggregation, the lack of a clear benchmark is not a major 

 drawback. It is enough to show the variability in the results without 

 indicating which set of results is somehow best. 



Since a reasonably large set of data was available to us at low 

 cost, the second approach was chosen. We estimated three models at 

 three levels of aggregation; the same observations were used in each 

 case. These examples will demonstrate that the confounding conse- 

 quences of aggregation are not idle prospects that can be ignored 

 with impunity. 



Pennsylvania provides the setting for these examples. Observa- 

 tions on more than 2,000 minor civil divisions (MCD's) were col- 

 lected. These observations were collapsed into 66 county observa- 

 tions, and these in turn were collapsed into observations on ten 

 regions (see figure l). 1 We generated regression results at each level 

 of aggregation. 



The first model attempts to identify characteristic patterns of 

 local, county, and regional tax behavior. In Pennsylvania, localities 

 are entitled to impose a number of taxes, other than real estate and 

 occupation taxes, on its residents (the state imposes a sales and 

 income tax on all residents). This right derives from the Local Tax 

 Enabling Act (Act 511) of 1965 (commonly referred to as the "Tax 

 Anything Law"). Included within the categories of taxes allowed 

 through this legislation is a per capita head tax. To account for the 

 level of 1974 per capita tax revenues we selected two variables: 

 (1) the level of these revenues in 1970; and (2) the change in earned 

 income derived taxes between 1970 and 1974. The first variable 

 provides a historical benchmark, and the second provides a measure 

 of the shift in dollars generated through the exercise of a 511 tax. 



Table 1 presents three sets of results for this simple model. Each 

 row contains the estimated constant («■), the estimated coefficient on 

 the 1970 level of per capita tax (B^, the estimated coefficient on the 



'In Pennsylvania there are 2,547 political subdivisions (excluding, counties, school 

 districts, and authorities). Data were gathered for 2,463 municipalities. The remain- 

 ing cases were eliminated from the analysis for several reasons. First, Pittsburgh and 

 Philadelphia were eliminated on the basis of their uniqueness (by far the two largest 

 metropolitan cities in the state). The elimination of Philadelphia also reduces the 

 number of counties from 67 to 66 since Philadelphia is a county-city administrative 

 unit. Second, many municipalities were eliminated because they were involved in 

 political mergers with other municipalities or because their census identification 

 numbers did not match with other sources of data. The remaining cases were deleted 

 because census data were undisclosed for these communities. The availability of a 

 data set which includes 96.7% of all municipalities, 66 counties, and affords us the 

 opportunity to use the 10 uniform regions so designated by the Pennsylvania 

 Department of Community Affairs prompted us to adopt the second approach 

 described above. Further, because of the makeup and distribution of its population, 

 Pennsylvania is often used to generalize to the country as a whole (cf. Zelinski, et ai, 

 1974). 



