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 #105: Two words: [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 Fri Nov-25-05 01:58 PM
Response to Reply #101
105. Two words:
as per #96:

random sampling.

But to elaborate, we are asked to:

"Consider real numbers, not non-response."

If you assume random sampling (i.e. no non-response bias) then I am sure the model will do what its author says it will do. The only non-fraud explanation for the discrepancy between polls and count is some kind of bias in the poll. This is sometimes called "non-response bias" although that needs also to include "selection bias" - i.e. bias in the selection of respondents. "Non-response bias" strictly means that those who refused were different in some way from those who agreed to participate. "Selection bias" can occur if there is bias in the selection process itself. Both can happen all too easily in any survey.

If we assume that there was neither non-response bias nor sampling bias, then the model will, I am sure, reproducably give a Kerry win. I would certainly expect it to. But that is the assumption I do not share. Social scientists never do, as selection bias and non-response bias are a pretty intractable methodological problems, and, unlike "sampling error" (on which the MoE is calculated) do not reduce as sample size increases. They are also not addressed by increasing your variance estimate to allow for clustered sampling (e.g. using the DESR - Design Effect Square Root. They have nothing to do with clustered sampling, which will introduce greater variance, but should not introduce bias. The sad fact is that we cannot mathematically compute selection bias, nor non-response bias.

The only way of investigating whether what caused a poll-count discrepancy was bias in the poll or bias in the count is some kind of correlational analysis: is the bias greater in circumstances where, a priori*, you would expect it, if fraud was the factor? Or the bias greater in circumstances where, a priori, you would expect polling bias to be greater?

But we can no more investigate polling bias by ruling it out, than we can investigate fraud by ruling it out, and it is the former that our undead friend asks us to do.



*see, I can do Latin when I try!
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