Merger of SOV and Registration Precincts to Census Geography

Mergers of the Statement of Vote (SOV) to Registration Precincts (Reg) are done as follows:

1. Geocoding of Registered Voter files

   All registered voter files are geocoded against the TIGER files. As geocoding puts the census geography on an individual address and this address also has a registration precinct on it, this allows the creation of a registration precinct to census geography equivalency file. If every address on the registered voter file could be placed in census geography, then a complete equivalency table could be built up. Unfortunately, this is not possible. Thus, two further steps are necessary.

2. Geographical Representations of Registration Precincts

   The other method of obtaining precinct to census equivalencies is through mapping the registration precincts onto the census geography directly. One can always (assuming one has the precinct maps) create an equivalency this way. For the 1992 General elections, the geographical mapping program utilized required the use of whole blocks in making assignments to the precincts.

3. Assignment of blocks to Registration Precincts

   This allows the assigning of blocks to precincts independent of TIGER or geographical representation, and is useful primarily when the geographical representation (which was required to follow block boundaries) is not an accurate representation of the actual boundaries. This splitting is then handled by a statistical assignment procedure (see below).

4. Balancing (assignment of split census blocks)

   The primary difficulty is when a precinct splits a census block into two or more sections, as it is then indeterminate how many registered voters live in each section. This can be handled either by geographical estimation or statistical estimation (statistical is used in this process). The statistical procedure is designed to allocate registered voters which have been left unassigned to census geography by geo-coding the blocks in such a manner as to equate expected registration with actual registration (the expected registration is also an estimate). This problem is formulated as a linear programming problem and is run through multiple iterations to achieve the final result.

5. Merger of Registration data to Census Geography

   The RB##AD%%.TYR file is a precinct to block conversion file constructed by the methods described in 1, 2, 3 and 4. The precincts here are RR type precincts. For registered voters assigned to a particular block through geocoding, the derived registration data is assigned directly to that block.

   For registered voters assigned through the balancing procedure, a straight breakdown of the derived registration data proportional to the number of registered voters assigned through the balancing procedure is made. This algorithm could be improved upon by conditioning on the characteristics of the individuals in that block.

6. Merger of Statement of Vote Data to Census Geography

   The RB##AD%%.TYR files are merged to the level of the final consolidation precincts using the SR##AD%%.TYR file. SOV data is then merged to block using this merged file. A straight proportional merge is made using as a breakdown the proportion of voters assigned to each block. Note that a more accurate methodology would be to calculate the estimated proportion of each type of voter in each block voting for a particular race and adjusting by this percent. The calculation of these proportions is a difficult theoretical problem which we believe we have solved, but this solution, if indeed it is a solution, has not been tested at the level necessary to implement it.

   Absentee precincts are not in general merged to the block level unless the absentee precinct results are reported at the level of the registration precinct (counties such as San Francisco and Monterey are reported that way, for example). Thus, areas (primarily in rural, sparsely populated areas) where the election results are collapsed into a larger absentee precinct (usually at the level of the ballot group) will not have any election results reported for them. The number of these areas is relatively small.