Question: In 2000, the US Supreme Court decided the presidential election after reviewing recounts of the ballots from some parts of Florida. A major issue was the count in Palm Beach county, where many voters claimed that the ballot design was confusing and had led to mistaken votes for Buchanan (Reform Party) that were intended for Gore (Democratic Party). We will compare the actual votes for Buchanan in Palm Beach with the predicted number of votes for Buchanan based on the other counties in Florida. Data taken from Agresti and Presnell (2002) Statistical Science, 17, 436-440.

The data are given in florida2000.xls with these columns: County: name of Florida County

Palm Beach: an indicator variable equal to 1 for Palm Beach county, 0 for all others Population: number of county residents

Black percent: percent of Black county residents Hispanic percent: percent of Hispanic county residents White percent: percent of White county residents Percent65: percent of county residents aged 65 and over

percentcollege: percent of population who had attended college Income: county average annual income in $1000s

Gore: number of county votes for Al Gore

Bush: number of county votes for George W. Bush Buchanan: number of county votes for Pat Buchanan Nader: number of county votes for Ralph Nader Total votes: number of county votes

The aim in this problem is to estimate the vote for Buchanan in Palm Beach from these county characteristics, without using the data from Palm Beach to fit or check the model.

1. Using all the data, plot the vote for Bush (horizontal axis) against the vote for Buchanan (vertical axis) [this is the plot in the class notes]. Where is Palm Beach in the plot? What does this suggest about the vote for Buchanan in Palm Beach?
   Before doing the next models, drop Palm Beach from the data set by doing the Stata command drop if county=="PALM BEACH"
2. As we did in class, regress Buchanan votes on Bush votes. Call this model A.
3. Using Model A, predict how many votes Buchanan should have received in Palm Beach. Give a prediction interval for your guess. Note that Buchanan received 3407 votes in Palm Beach county.
4. Fit Buchanan on all the predictors excluding percent votes for Gore, county name, Palm Beach indicator, and Buchanan—this is Model B.
5. Reduce Model B by dropping non-significant predictors, and call this Model C. Use your own judgment in this model reduction, not an automatic selection procedure.
6. Using Model C, predict how many votes Buchanan should have received in Palm Beach. Give a prediction interval for your guess.
7. Which model seems better, Model A or Model C? Explain.

Price: $2.99

Solution: The downloadable solution consists of 6 pages
Deliverable: Word Document