- 19 -
first include a count of the number of departments with crosscutting jurisdiction over the policy
issues in the proposal (mean 3.3, s.d. 2.5), which runs from one to as many as fourteen. We also
include a 0-2 index that corresponds to agencies’ internal management. A proposal that details
no reorganization or personnel reform is coded with a 0 (72%). A proposal that reshapes an
existing bureaucracy or creates a new one is coded with a 1 (17%). Reorganization plans or
legislation that has bureaucratic reorganization as its main goal is coded with a 2 (11%). Our
expectation is that crosscutting issues and managerial proposals are easier to manage in a
centralized fashion, and thus positive coefficients are expected in both cases.
Methods
Since the dependent variable is ordered and categorical, we estimate a series of ordered
probit models that include the variables defined above. This will help us predict the probable
impact of a given independent variable on the extent of centralization without requiring us to
assume that the spacing between the values of the centralization index is exactly even.
Note that there are two immediate difficulties that arise in estimation. First, the
observations may not be independent from one another. For example, policy proposals from
the same president may have a higher or lower likelihood of centralization based upon the
unique characteristics of the presidents themselves. Similarly, proposals from the same year,
same Congress, or same department may have a higher or lower probability of being
centralized. To account for this, we estimated a series of models that include indicators for
presidents, year, congresses, and departments, respectively. In none of the cases could we reject
the null hypothesis that the inclusion of these indicators did not improve the fit of the model,
though they were not themselves individually statistically significant.
Convention