Page 153 - Contributed Paper Session (CPS) - Volume 3

CPS1979 Francisco N. de los R.
Investigating dissimilarity in spatial area data
using Bayesian Inference: The case of voter
participation in the Philippine National and Local
Elections of 2016
Francisco N. de los Reyes
School of Statistics, University of the Philippines Diliman Quezon City Philippines

Abstract
A commonly studied characteristic of area data is the assessment of similarity
(or absence thereof) among neighboring areal units. However, most
methodologies do not measure uncertainties which are likely outcomes of
sampling variation and do not consider spatial autocorrelation. This paper
explores the ability of Bayesian modeling to address the said situations. It
attempts to apply this modeling technique to the voting participation statistics
in the Philippine National and Local Elections of 2016.

Keywords
conditional autoregressive (CAR); proximity matrix; dissimilarity; voter turnout

1. Introduction
Many inquiries in statistics are interested in determining heterogeneity in
some population. Dissimilarity is one such measure. It is the extent to which
two or more groups are integrated or isolated. The most popular metric is the
Dissimilarity Index. However, the Dissimilarity Index has the following
inadequacies in spatial data: it does not measure uncertainties which could
potentially be a result of random sampling variation, and it does not consider
spatial autocorrelation which could be present in the data.
This paper aims at detecting dissimilarity in a specific spatial area data:
voter participation. In the Philippines, voter turnout is intuitively spatially
autocorrelated. There are strong bailiwicks in various corridors in Philippine
geography: Northern Luzon is one, Bicol region is another. There is also a
strong solid vote in Panay and Negros Islands, another in Cebu and then the
Davao region. Voter turnout in nearby barangays (the Philippine basic
geopolitical unit) tends to be similar. The same may be opined for larger units
like cities and municipalities and even up to the level of the province or region.
This paper shall first present a classical method in establishing dissimilarity.
However, in consideration of the spatial nature of voter turnout, a Bayesian
model will be used to introduce smoothing in the presence of spatial
autocorrelation.

142 | I S I W S C 2 0 1 9