Reply #25: I jump in when I can ... but I have my own little nest of eggs [View All]

[Iceburg]

that I must tend to. I'll pm you in the next few days regarding some anomalies and correlations wrt Cuyahoga that I need a 2nd opinion on.
Back to the problem...

You: "How do you get more ballots than registrations.?"

Me:
One way is switching parties. The registrations in this problem set are part registrations not regristrations to vote. I believe in another post on another topic (How many Bush/Republicans from 2000 voted for Kerry?), I indirectly cited a huge shift from voters who voted for independents (or voted for Bush in 2000) that went with Kerry in 2004 (re: Bush loses ground in his Top 10 Cities (of 2000) in Cuyahoga County)

In most of the work/analysis that I do on Cuyahoga, I exclude the precincts that have less than 50 ballots cast (unless the underying reason for the low numbers is turnout). That said, I don't ignore them completely but I wouldn't use them in lets say a regression model. The fact that precincts exist with only 4 registered voters is of more importance to me, not whether Bush got 00% of 100% in that precinct. The answer to -- Why would you establish a precinct for 4 regsitered voters, and assign it two voting machines (the minimum machine assigment in Cuyahoga) -- I call'em faux polls for now. Thus, I would tend to want to exclude them and all the precincts that have 0 ballots cast (about 20 or so) in this current exercise.

Re: Normalcy, uniformity etc of subgroups within a county. TIA, you should know me better than that ... I did the histograms before I posed the question.

RE:should we estimate each county separately ?...
You:
I say do both.
1) Do the differences in the counties offer clues? They might.
2) Does an exhaustive data set of all th counties show us the big picture and effectively eliminate individual the outlier concern? I believe it does.

Good questions, but I'm afraid I will beg out ... I must get back to my work in quantifying the number of votes mistabulated in Cuyahoga.
Good luck.