This case study is about exploring anomalies in the 2000 Presidential Election in which George W. Bush won the election but lost the popular vote. Importantly, he won the state of Florida and all its electoral votes in spite of claims of certain "issues" and irregularities that arose and the margin of 537 overall votes! The study explores different anomalies and attempts to see if they really are anomalies and to quantify the effect of one particular anomaly.
Complaints arose on election day about the layout of Palm Beach
County's (PBC) ballot. Specifically, a "butterfly" ballot
which looks like
caused problems.
By law, the order of the parties corresponds to the
order of the election in the previous election (state-wide).
By this prescription, Gore should have been second, but
because of the two page layout, Buchanan was second.
"Many" claimed to have voted erroneously.
This ballot was unique to Palm Beach county.
This project involves trying to determine if, and provide evidence for, the problem is real and then quantify what the likely effect was on the number of votes Buchanan received. The student is supposed to think about how to go about finding an effect and evidence to support their conclusion. Much of the investigation is complicated by the very different numbers of votes in the different states, counties and precincts. We have to take this into account. We use EDA to look for the effect and compare the results from PBC with the rest of Florida and the rest of the country. We identify other outliers and try to understand these.
We use regression to develop a model for the proportion of votes for Buchanan using votes for republicans, census data and previous election data. This is a GLM and involves over-dispersion and also issues of robustness due to the different counts and proportions.
We also use maps and precinct level data to compare what should be similar voting patterns. And we use the comparison between absentee and election day voters and also party affiliation within ballots to further conclude that there was an effect.
The specific questions are:
What can we explore? We have voting data from the entire nation at the state and county level that gives vote count and percentage for each candidate including the different minority candidates. Can we detect the anomaly from this?
When we look at "obvious" plots of gore versus bush, gore versus buchanan, bush versus buchanan the "outliers" are not evident. We immediately have issues with counties and states of different sizes and comparing total counts and also comparing percentages. We have to take account of different variances based on different electorate size and different proportions. This introduces the students to variation in different ways.
We also have election data from the previous elections, i.e. 1994 and 1996. We contrast these with the 2000 results.
We also have census data that gives county- and state-level data on education levels, income levels, ethnicity, ... We can try to compare Palm Beach County to other counties in the country. We do this via visual displays.
Next we develop regression models for predicting the county-level Buchanan vote based on the votes for Bush and previous election results and demographic variables. We do this for each state, aggregating states with few counties, e.g. Hawaii, Rhode Island, Washington DC.
The regression models need to take account of the different proportions and county sizes. An overdispersedWe also have precinct level data for Florida. We draw a map of the Buchanan vote percentage by precinct and contrast Palm Beach County with neighboring counties. We also look at other variables such as income, education & ethnicity to illustrate that some neighboring counties are very similar in some characteristics but different in others and that the voting patterns tend to flip.
The absentee ballot used for Palm Beach County was not a butterfly ballot but rather a regular, sequential ballot. We compare the results from this sub-group to see if there is a "ballot effect". Can we determine if the absentee voters are substantively different in other characteristics?
We clearly don't have information about how an individual voted. We do however have individual ballots and can look at how an individual voted for different candidates and parties for different positions, e.g. president, senate, congress, local.
Because we used this in our summer school, the data are available directly as R objects via an R package and so the students have to do no data input. However, we can have them collect this information from various sources.