You are viewing an obsolete version of the DU website which is no longer supported by the Administrators. Visit The New DU.
Democratic Underground Latest Greatest Lobby Journals Search Options Help Login
Google

Reply #24: Numbers tell everything [View All]

Printer-friendly format Printer-friendly format
Printer-friendly format Email this thread to a friend
Printer-friendly format Bookmark this thread
This topic is archived.
Home » Discuss » Topic Forums » Election Reform Donate to DU
Febble Donating Member (1000+ posts) Send PM | Profile | Ignore Mon May-02-05 06:29 AM
Response to Reply #22
24. Numbers tell everything
Exactly. It is numbers I am looking at.

To be even more exact, I am looking at the Within Precinct Error numbers. This tells us how wrong the predictions were for each precinct, after what E-M considered appropriate weighting. And they were massively wrong. By which I mean they massively didn't match the count.

It is impossible that this could have happened "by chance". Which means something very precise also. We can calculate the "sampling error" which depends on the number of people sampled and the proportion of each who said they voted for each candidate. The error here is way way way outside sampling error. So we know the prediction didn't match the count.

But when two things don't match, the numbers don't tell you which was wrong. What you have to look at is correlations.

Here is a testable fraud hypothesis: if the mismatch was due to fraud, and fraud was more common in precincts where electronic voting was used, then you would expect the error to be lower where there were paper ballots. Hypothesis supported. However, there is a problem - paper ballots precincts were small in number and mainly rural. There's a confound. But it's perhaps suggestive. Needs further probing.

Here is a testable rBr hypothesis: if the mismatch was due to Bush voters being less likely than Kerry voters to participate, then you'd expect the mismatch to be greater in precincts where it would be easier for Bush voters to "escape" - large precincts, inexperienced interviewers, precincts where the pollsters were a long way from the precinct. Hypothesis supported. However, there are several problems. One is the WPE itself which needs fixing, and I've fixed it. Another is that if Bush voters were less responsive than Kerry voters, response rates should be lower in high Bush precincts.

So here's another testable hypothesis: if the mismatch is due to reluctant Bush responders, response rates should be lower in high Bush precincts. Hyoothesis not supported. There is no difference in response rates (no significant difference). However, there is a problem here - we do not know the variance in response rates. So we cannot tell whether the reason we failed to support the hypothesis is because there was no effect or because there was insufficient statistical power. It is a question that needs addressing. It is one we should be seeking answers to. But we do not have the data, so the null is still retained. We have not proved that the response rates in high Bush precincts were higher than in high Kerry precincts (which might suggest fraud), but we have not disproved it either.

You are right. Numbers tell the story. If they can be heard over the noise. And it is very noisy data.




Printer Friendly | Permalink |  | Top
 

Home » Discuss » Topic Forums » Election Reform Donate to DU

Powered by DCForum+ Version 1.1 Copyright 1997-2002 DCScripts.com
Software has been extensively modified by the DU administrators


Important Notices: By participating on this discussion board, visitors agree to abide by the rules outlined on our Rules page. Messages posted on the Democratic Underground Discussion Forums are the opinions of the individuals who post them, and do not necessarily represent the opinions of Democratic Underground, LLC.

Home  |  Discussion Forums  |  Journals |  Store  |  Donate

About DU  |  Contact Us  |  Privacy Policy

Got a message for Democratic Underground? Click here to send us a message.

© 2001 - 2011 Democratic Underground, LLC