Quantifying Evidence for Outlierness
The data are Democratic and Republican vote counts, by (a) absentee ballot and (b) voting machine, for 22 elections in Philadelphia's senatorial districts between 1982 and 1993.
A data frame with 22 observations on the following 2 variables.
Year of election
a factor with levels
Number of absentee ballots indicating a vote for the Democratic candidate
Number of absentee ballots indicating a vote for the Republican candidate
Number of machine-counted ballots indicating a vote for the Democratic candidate
Number of machine-coutned ballots indicating a vote for the Republican candidate
Percentage of absentee ballots indicating a vote for the Democratic candidate
Percentage of machine-counted ballots indicating a vote for the Democratic candidate
a factor taking on the value
"yes"for the disputed election and
"no"for all other elections
In a special election to fill a Pennsylvania State Senate seat in 1993, the Democrat, William Stinson, received 19,127 machine–counted votes and the Republican, Bruce Marks, received 19,691. In addition, there were 1,391 absentee ballots for Stinson and 366 absentee ballots for Marks, so that the total tally showed Stinson the winner by 461 votes. The large disparity between the machine–counted and absentee votes, and the resulting reversal of the outcome due to the absentee ballots caused some concern about possible illegal influence on the absentee votes. To see whether the discrepancy in absentee votes was larger than could be explained by chance, an econometrician considered the data given in this data frame (read from a graph in The New York Times, 11 April 1994).
Ramsey, F.L. and Schafer, D.W. (2013). The Statistical Sleuth: A Course in Methods of Data Analysis (3rd ed), Cengage Learning.
Ashenfelter, O (1994). Report on Expected Absentee Ballots. Department of Economics, Princeton University. See also Simon Jackman (2011). pscl: Classes and Methods for R Developed in the Political Science Computational Laboratory, Stanford University. Department of Political Science, Stanford University. Stanford, California. R package version 1.03.10. http://pscl.stanford.edu/