jlavicka / ei_project Goto Github PK

The R script takes as its input ethnic population totals and party vote totals from each political unit. Ecological inference estimates, in terms of percentages of each ethnic group's vote toward each political party, are then exported with additional descriptive information to an .xlsx file format.

The following indices are in reference to the imported data frame. They each are to be written as a vector of integers.

Object alpha contains the column indices of each political party.

Object bravo contains the column indices of each main ethnic group.

Object charlie contains the column index or indices of the ethnic group(s) labelled Other.

Object delta contains the column index or indices of the ethnic group(s) labelled Unknown.

The ethnic groups used to compose objects charlie and delta are compiled at the user's discretion. If no group applies, either or both objects can be left NULL. Objects alpha and bravo must be filled.

Additionally, a string value representing the desired output file name is needed for object xray.

The sum of objects bravo,charlie, and delta is used as the total population of each unit. Percentages of each party and ethnic group are computed in relation to this total population figure. The number of non-voters in each unit are computed as the difference between the total population and the total number of valid votes.

From theeiPack R package, ei.MD.bayes() is used to complete the analysis. This function implements a hierarchical Multinomial-Dirichlet model for ecological inference in RxC tables. Draws from the posterior are obtained with a Metropolis-within-Gibbs algorithm. More information on the methodology can be found in the package's R manual included in the repository.

The code now accounts for mismatched political units. Previously, the total population must have been greater than the total number of valid votes for each political unit in order to be included. Mismatched units are now retotaled as a percentage of the valid votes with the assumption of complete voter-turnout, then resubmitted into the analysis. Realigned data are still exported with the results in a separate Excel tab for further inspection.

There are still a few additional bugs that need fixing. If you come across any of these issues please let me know.