- ‹ previous
- 927 of 2820
- next ›
The Vote in Iran – Revisited
Yesterday I posted a chart showing that as Iran's Interior Ministry announced election results throughout the day, the winning percentage for Mahmoud Ahmadinejad had stayed almost eerily constant. It seemed likely that in a genuine election there would have been a little more variation, so this looked like a piece of evidence that the vote count had been rigged.
It still seems likely that the vote was rigged, but the steady vote count apparently doesn't prove anything one way or the other. Nate Silver plotted the 2008 U.S. election results using waves of states in alphabetical order, and he came up with an almost dead straight line, just like the Iranian results. One of Andrew Sullivan's readers did the same with the results as announced every half hour through the night, and again the line was as straight as a laser. So this is apparently a null piece of evidence.
But now I'm curious. The Sullivan graph shows that by 7:30 pm Eastern time, when you have two data points, you could predict the final popular vote in the 2008 election with about 99% accuracy. Question: would you get the same results if you plotted the last five or six elections? If so, it means that most years we'll know with almost complete certainty who the winner is by 7:30 pm, exit polls be damned. Can this really be true?
Iran election
From what I understand, the difficulty comes from the fact that some areas that would be obviously pro-Mousavi voted for Ahmadinejad by exactly same the same margin as the final result. If that's so, it would be the same as rural Idaho voting for Obama by eaxactly the national margin...which would be highly suspect, at best. But we still need more reliable information. Bye the bye, why didn't any of the outside "experts" see this result coming?
7:30 PM EST
"If so, it means that most years we'll know with almost complete certainty who the winner is by 7:30 pm, exit polls be damned. Can this really be true?"
No, because Sullivan's reader is still using final results rather than real-time partial results. Taking the final results and randomizing them by poll-closing time still smooths out the irregularities people think they would see in results as they come in irregularly from different regions, just as alphabetizing does.
I don't know what the chart would look like if it were real-time, partial results but this still isn't it.
7:30 PM EST
"If so, it means that most years we'll know with almost complete certainty who the winner is by 7:30 pm, exit polls be damned. Can this really be true?"
No, because Sullivan's reader is still using *final* results rather than real-time partial results. Taking the final results and randomizing them by poll-closing time still smooths out the irregularities people think they would see in results as they come in irregularly from different regions, just as alphabetizing does.
I don't know what the chart would look like if it were real-time, partial results but this still isn't it.
Nope
G.C. has it right, twice. Sullivan's readers are not using real-time tallies as they came in; they're using the final vote count artificially divvied up by poll-closing time.
That's why, despite all the smoke and mirrors, the original graph still has relevance. The real-time tallies in Iran, as they came in, showed surprisingly little variance. Even Silver's model, effectively randomized by alphabetizing the final vote count, showed twice as much variation as what actually happened in Iran.
Kevin is misinterpreting the
Kevin is misinterpreting the meaning of R^2.
Looking at raw vote totals messes with our intuition, since election watchers never look at raw vote totals come in.
If you instead look at fluctuations of the actual vote-share percent(I have a graph on my website, http://StochasticDemocracy.com), then you'll see that results fluctuated within a band of about 3-4% the entire night.
For the most part, yes, you can usually tell the end-result of an election within 3-4% by the time 20% has come in. I say this as someone who constructed a real-time forecasting system for the 2008 elections.
Obviously, this doesn't mean the election wasn't stolen. I personally suspect it was, based on the reports that have been coming in from Tehran. But this particular point was meaningless. If the government would actually post some regional election returns, I'd love to do a standard data-verification analysis. (Benford's law and all)
Two data points
Like Scalia and Thomas?
I started to make a graph
I started to make a graph myself, yesterday, of the 2008 election to see if there was any validity to your Iranian election fraud argument. I quickly got frustrated trying to determine the chronological order of vote returns, though. I hadn't thought to correlate the totals alphabetically instead of chronologically, as Mr. Silver did. Kudos, mate. Sometimes the obvious bites you, or maybe I'm just getting stupid. I sure feel that way, now.
It does go to show that you can probably come up with a seemingly reasonable statistical argument to show that the earth is flat, however. Statistics in the hands of the unqualified can be very dangerous.
Sometimes a line is just a line
"...using final results rather than real-time partial results."
And if someone plotted exactly the data you suggest, and it still showed a straight line, what objection would come up next to bolster the incriminating nature of the graph?
It sure looks to me like somebody saw the original graph, thought "Hey, that line is way too straight to be real!" and hastily proceeded to use it as evidence of fraud in itself, without even knowing what similar data from a clean election looks like.
As much as some people want this graph to be the silver bullet that "proves" the election was stolen, the sad fact is that it provides no such evidence in itself. This is not to say that there isn't plenty of other evidence of fraud, just that the graph is non-informative by itself in that regard.
No, dammit. Alphabetizing
No, dammit. Alphabetizing implies that results come in utterly at random, and that each wave of results come from a representative sample of geographical areas. In fact, it's not that case that each update will include tallies from such a diverse and statistically predictive swathe of the country. At all.
The electoral college makes this phenomenon very easy to understand. Think back to 2008: as the night went on, states were called one by one, occasionally with several states being called at a time. So (and I'm making these numbers and chronologies up) you had New York called, with a 65-35 margin to Obama. Then you had DC all of New England called, with huge Obama margins everywhere. Then you had one or two southern states called, with 55-45 McCain victories. If you collected all of the results gathered so far and used them as a data point on a graph like Andrew's, you'd see a huge advantage for Obama, way huger than he actually ended up with. Now, after those states have been called, Kentucky, Alabama, Georgia, Mississippi, and Louisiana are called for McCain, and Virginia is called for Obama. If you used these results to add a new point on Andrew's graph, McCain would shoot up. And so it would go throughout the night.
The one chronological phenomenon I can actually remember from 2008 also reflects this point: the results from California, Washington, and Oregon all came in simultaneously, giving Obama huge wins and allowing all the networks to call it for him.
And we should have expected to see something like this from Iran. Maybe the first few results came in from urban areas, giving Mousavi big margins. Then the rural parts of the country started getting their tallies in, allowing Ahmadinejad to surge back. And so on. The chronology of the results' tabulation is not a random sample; it reflects geography, and geography reflects political preferences. So we should expect to see fairly substantial swings back and forth across the regression line, because it doesn't make sense to expect that every wave of results will include the same proportion of support for both candidates.
The problem, of course, is that we don't have good data about whose results came in when for Iran, and we also don't have great polling that allows us to predict which areas would have gone for which candidate. But the graph is still fishy.
Yes, or rather, mostly.
Yes, or rather, mostly. Here's the way it works, once a certain number of results come in, and they are wide-spread enough (i.e. not all in say, NYC or something) you can extrapolate the result with statistics because you have the equivalent of a Random Sample. It's why you can project results with 19-20 of the vote counted. So that's not going to change in the long run much.
This I learned during the Calderon election in Mexico when we were watching the lefty agonizingly lose at the end.
fumphis, VoteForAmerica has
fumphis,
VoteForAmerica has posted actual real-time results from November, they also show very high R^2 values. See them at http://voteforamerica.net/editorials/Comments.aspx?ArticleId=271&Article... . I also posted a round-up of simulations at http://stochasticdemocracy.blogspot.com/2009/06/iran-update.html .
I personally suspect the election was stolen, but pushing this bogus analysis doesn't help anybody.
Good Grief
It's like some of you guys have never been through an election night before.
Here, watch this compilation of MSNBC calling the results: http://www.youtube.com/watch?v=x-V2zLEzxKk&feature=related
As you'll see, up until the race was called, the data set of electoral vote totals as they came in looks like this (3 Obama - 8 McCain, 3-16, 103-34, 175-70, 195-76, 200-85, 200-124, 207-129, 207-138, 284-146). If you plot this at home, you'll see it's legitimately variable, with an R^2 of less than 0.84. Everybody who follows politics knows this intuitively, because we actually do have to wait for a while before a race can be called; the first temporal data points do not determine the election. I don't know why everyone is being so obtuse.
SouthPaw, There's a huge
SouthPaw,
There's a huge difference between the electoral vote and the popular vote. An analysis of the 2008 elections, looking at realtime data coming in from MSNBC was done at http://voteforamerica.net/editorials/Comments.aspx?ArticleId=271&Article... .
David, there is definitely a
David, there is definitely a difference, and I'm sure doing what I did with the national popular vote would show less variation.
But since you prefer it, let's look at the VFA analysis. The graphs presented are on the state level, where there's presumably not as much geographical variation as in a multiethnic nation like Iran (though, in fairness, Virginia does have perhaps three distinct political regions). Even still, their state level data is significantly more variable than what we see for Iran. The R^2 values of .9954 for KY and .9946 for VA (though objectively quite high) are still 3.3 and 3.9 times
more variable
than the data set from Iran. Similarly, Nate Silver's artificially randomized, alphabetical data set was about twice as variable as what we saw out of Iran.
For what it's worth, in the
For what it's worth, in the various posts I've read attempting to throw cold water on the Sullivan graph, I have yet to see anyone produce an analytical model that yields an R^2 value greater than .9986.
Sorry about the weird formatting above, I screwed up an html tag.
Missing the point
OF COURSE if you have a representative sample of the whole then you can predict the whole. That's Duh 101. The Kentucky samples predict the Kentucky whole. The Virginia samples predict the Virginia whole. But my understanding is that the Iranian data is being touted (by the Iranian gvt) as showing the exact same *slope* regardless of region. Using the VFA data, Kentucky, which went for McCain, obviously has a different slope than Virginia. As I say, my understanding is that the Iranian data is being reported as being perfectly and equally proportional regardless of region. That's ridiculous and indicate fraud.
The Vote in Iran
The neoCons in Iran fooled more progressives here in the USA than they did in Iran.