Ok, Statisticians, here's a question for ya'll

[diva77]

In counties with early voting, is it possible to extrapolate from early voting results what the final results will be in that county? It would seem so-- based on the ongoing discussions of what percent audit is acceptable...hmmmmm...

Maybe this discussion has occurred already; if it has, would appreciate link. Thanks.

[valerief]

1. Not if they hack the votes early. n/t

[diva77]

2. ha! well, what if they wait a little before that happens??

[nam78_two]

3. Just out of curiosity

are any such results available? If so could you point me in the direction-thanks!
Any good news is welcome right now....

[diva77]

16. hopefully you checked out post #8

[FogerRox]

29. I know in FLorida 2004 we got lists of WHO VOTED

each day. Allows campaigns to focus GOTV efforts on who didnt vote.

[SharonAnn]

4. They don't count the votes until Election Day. Same with Absentee

ballots. Actually, they don't count them here until after the polls close.

[diva77]

6. how can you know for sure no one's counting if there are

memory cards and wireless available?

[rock]

10. "they don't count them here ...."

Ah ah ah! Stop right there!

[diva77]

15. funny!

:spray:

[salvorhardin]

5. No

You could make a projection, but you'd never know or not if you had a representative sample and thus you could not reliably predict anything about the actual vote. In fact, because the early voters are a self-selected group you can almost be assured you wouldn't.

[diva77]

7. got any info on the demographics of early voters?

[OnTheOtherHand]

8. try here for Florida 2004

http://www.reed.edu/~gronkep/docs/GronkeBishinStevensGalanes-Rosenbaum.APSA.2005.pdf

As far as I can tell, there is no "general" answer to this question -- the data aren't widely available. But Gronke has done other work on this subject, so you could keep googling....

[salvorhardin]

9. Thanks.

I thought this was just in GD, otherwise I would have left it up to you. ;-)

[OnTheOtherHand]

19. heck, no

It's not as if Febble and I own the sampling-theory territory. Step right up!

[diva77]

13. Very interesting study; many thanks for posting...b-b-but...

that being said, however, I didn't see (maybe Iglossed over detail) any accounting for the voting wares used for early voting versus the wares used on election day and then a comparison of those totals to the certified results of the election.

In some counties that have paper ballots on election day, early voting is done with touchscreens with the justification that a paper ballot cannot be offered because there are too many ballot types to do so. This seems a back door way to introduce tabulation that has "override capability" where that opportunity did not previously exist:tinfoilhat:

Also, in some counties, the results of early voting are not posted by precinct nor are they included in audits.

Yes, the study is interesting, but I'd like the researchers to add in these variables and then see what they have to say about early voting :hi:

[OnTheOtherHand]

18. quite simply, the paper wasn't about that

You asked about demographics, and this paper presents some evidence about demographics, for one state in one election. But it doesn't purport to say whether the vote count was accurate. Nor does it purport to say whether early voting is good, although the authors seem pretty skeptical.

I'm not sure what you have in mind by an "accounting for the voting wares used for early voting versus the wares used on election day and then a comparison of those totals to the certified results of the election." It is hard to conclude much from comparing the early-voting and the election-day results, for reasons that have already been mentioned. In this case, we can infer from Figure 1 that the early voting electorate probably had a Democratic skew (although it also had a non-black skew) -- but it's all pretty iffy.

Sometimes academics don't write the paper we wanted to read because they didn't think of it; sometimes because they didn't want to; sometimes because that's another paper; sometimes because they couldn't. This may be a case of "couldn't," but I can't tell yet.

EDIT TO ADD: For what it's worth, Figure 1 (very broadly) dovetails with the Gallup result cited by Awsi Dooger, that Kerry led in the early voting in Florida. But (to echo a point you made) I don't know that Florida has ever released early vote results for 2004. We could try looking at North Carolina again.

[diva77]

21. you're absolutely right/ I went off on a tangent

with wishful thinking - I wanted all the dots connected!

Thanks for info. and input you provided!

[Febble]

11. What he said

and what's more, if the early vote turned out to be "statistically significantly" different from the election day, all you could say was that something was indeed different (or probably). You couldn't infer that the way the early votes were counted was more/less reliable than the way the election day votes were counted because, as with the exit poll problem, you can't assume that those who voted early were "drawn from the same population" as those who voted on election day (i.e. just as you can't assume that those who should have, but didn't, participate in the exit poll were "drawn from the same population" as those who did).

So if you want to predict the vote, you have to get a random sample of voters, not a self-selected sample. Which is, of course, what polls try to do. But getting a truly random sample from the population you want to predict is quite extraordinarily difficult.

[diva77]

27. I appreciate you stating what's needed for a study that would pass

muster with your colleagues...I s'pose I had more of an "in vivo" study in mind that took into account things that had nothing to do with the voters. But you're absolutely right.

[Febble]

28. Certainly you can use an "in vivo" approach

which might well be considered to have greater "ecological validity". You just have to be aware of its limitations.

Ideally, to find out what causes something else, you need to manipulate a randomly allocated variable. That's what is called a true experiment. For example, if you wanted to find out whether dressing up in a gorilla suite discouraged voters for A more than voters for B from taking part in an exit poll you could randomly allocate gorilla suits to half your exit poll interviewers and have the other half dress normally. Then you'd look at the discrepancy between your poll responses and the count. The beauty is that you wouldn't need to worry about whether or not there were other factors (like fraud for instance) that affected the discrepancy, because there would be no reason to suppose it was not equally prevalent in gorilla and non-gorilla precincts. So if you found that in the gorilla precincts the responses were "signficantly" more pro-A, relative to the count, than in non-gorilla precincts, you could conclude that the gorilla suits had affected your poll.

In other words if you as the experimenter can manipulate the factor you are interested in, and if you make sure that that factor is randomly allocated, if you get a significant difference between the effect where that factor is present and where it is absent, you can be sure that it is that factor that is responsible for the difference (or fairly sure - you will be able to compute your confidence precisely).

But most of the time in social science we don't have that luxury. We have to do "in vivo" studies in which we are comparing things that are different because that's the way they are. The problem is that if we find two things that are different (say the vote proportions in pre-election votes, and the vote proportions in election day votes), because WE haven't allocated the variable, we won't be able to be sure that it was the counting that was different or the voters.

That's not an insuperable problem, although you have to be careful. If you look at enough known factors (e.g. find out the demographics of the two groups, their party ID or voting intentions) you may be able to "control" for factors likely to be causing differences between the two sets of votes, and therefore determine whether it is likely that some other factor was also operating. But even at best, you will not be able to conclude for sure that the mystery factor was what you think it is.

Famously "you cannot infer causality from a correlation". That is in fact not true, you can. But only if you manipulate one of the predictor variables. If you don't, you always have to have a caveat. You can say, possibly with a high degree of confidence, that X is associated with Y (e.g. "early voting is associated with greater Democratic vote-share"), but not why.

[Awsi Dooger]

12. I remember two early voting numbers from 2004

There was a Gallup poll in Florida indicating 30% of registered voters had already voted and Kerry led 51-43. In Iowa it was a Des Moines Register poll with similar findings, that 27% had voted early with Kerry ahead 52-41.

Those numbers came out a day or two before the election and were heavily touted on DU and elsewhere. So when the early exit polls came out, especially Florida, the early voting estimates contributed to the certainty that Kerry was the winner. Kerry, of course, lost both states.

On MyDD today there is already a post touting the early numbers of requested absentee ballots from Democrats compared to Republicans, a massive edge for our side, and concluding the Iowa gov race is therefore not really a tossup. Kos and MyDD are generally in complete agreement but I noticed Kos cautioned about making judgments about Iowa 2006, due to what happened in 2004.

[diva77]

14. do you have link for that Gallup poll info?

also, how do you get info. on nos. of requested absentee ballots?

thanks

[Awsi Dooger]

17. It's mentioned at the bottom of this USA Today article

http://www.usatoday.com/news/politicselections/nation/polls/2004-10-31-poll-x_x.htm

Both of them are listed here: http://www.talkingpointsmemo.com/archives/003885.php

On absentee ballots I'm not sure how you find state-to-state info. Yesterday's Des Moines Register had the Iowa details: http://desmoinesregister.com/apps/pbcs.dll/article?AID=/20060928/NEWS09/609280403/1056

[diva77]

23. thanks!

[BeFree]

20. Good question

But as you can see the statisticians have no good answer. But common sense tells you that when 30% of the voting public casts votes in a completely open to all election, that the resulting stats should compare favorably with the stats of other 70%.

But common sense was thrown out the window in 2004.

The most complete study I have seen was one posted on DU just after the 2004 election, and that study made it into Mark Miller's book "Fooled Again". Read it and weap.

NC: The Full Unofficial Audit: The Case for Tampering...in the NC state forum

http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=170&topic_id=2562&mesg_id=2562

[diva77]

22. from what I have seen in the counties I have looked into with

early voting, it doesn't seem like it's a good idea.

Maybe we could rename early voting "concrete results that potentially provide info. and a pool of votes needed to manipulate the results on election day." :eyes:

People in counties with early voting should, at very least,

1)demand precinct level results for on site early voting
2)demand precinct level results for absentee ballots

or maybe try this

3)lobby their Board of Supervisors to stop offering/budgeting for early voting

[Febble]

24. Well, I absolutely agree with what you say

abut what people should demand. Some of the most valuable checks could be the most basic, and providing vote totals according to where and how those votes were cast seems fundamental to transparency.

[WillYourVoteBCounted]

25. in NC and elsewhere

early voting
was about 1/4 of all ballots.
this is where the majority of our machine malfunctions occurred
you have more votes in one location - so if something goes wrong, more votes affected
your ballot is not completely secret - it is in effect an absentee ballot tied to you
you are more likely to vote on a DRE if early voting - all ballot styles must be in one location
poll workers must be better trained - officials try to put their best here
poll worker fatigue - long days, long weeks
if you are fighting paperless voting, or don't want DRE/touchscreens,please don't push early voting.
early voting increases the push for touchscreen voting.

more here
http://www.ncvoter.net/earlyvoting.html

and lots more here
http://www.ejfi.org/Voting/Voting-93.htm#marker-1476452

[diva77]

26. great info. on these - the 2nd one especially is amazing.

I'll have to let others know about it

thanks!

[OnTheOtherHand]

30. wow, the 2nd one cites both Greg Palast and John Fund!

I've never seen that before.

There's a lot to hate about Fund's book -- he's a big voter ID booster, and... well, there's a lot to hate. The guy writes for the Wall Street Journal editorial page, after all. (In case anyone is curious, Time for change posted a review here last October.) But I don't like massive absentee/early voting either. I think Fund is against it partly because many Democrats are for it, because they reason that it increases turnout and benefits Democrats. I haven't seen good evidence that it does either of those things.

Hey, can you point me to more about machine malfunctions in NC early voting?