Faked high speed votes - An inside look at the AOL poll

Featured in this news article->:  American Idol: Are the AOL Poll Results Reliable?  http://blogcritics.org/archives/2005/05/23/141841.php

BEST WISHES AND LUCK TO Bo and Carrie!

Update: Monday May 23rd Bo/Carrie poll analysis here http://home.comcast.net/~pollinfo/pollfileupdate.html

The site has been revised to include new examples and a new way of looking at figures.  However I'd like to rewrite the figures/examples portions below so that they are easier to follow.  I will try to do this as soon as time permits.

Update:  Well tonight is result the final result night of AI4 and I want to say that I truly feel bad for all the people that were misled by the AOL poll.  I feel bad for all of the news writers and people that thought the poll was correct in predicting Bo as winner.  Fortunately I am fans of both of them and will be enjoying at least 2 new CD's. :-)

For those who are bothered by AOL's poll being spammed by high speed computers with fake repeat votes, I have something that you may find interesting. For those who didn't know, just think back to all the times the poll looked goofy and you wondered why.  Or the time when your favorite was in the bottom on AI, but the AOL poll said they were #1 . I know that polls don't decide the outcome, but I also know that some people like to see them. I will briefly present a way to read AOL's internal poll values and see a more accurate picture. I will also cover a huge mistake on AOL's part, and how to compensate for that.  I think Bo deserves to win as much as Carrie does, but I don't like the idea of people being mislead. Especially since it's the largest poll and it's broadcast to aol users desktops.  This is not about having a favorite or bashing. I think it is simply ridiculous that a small number of computers can completely reverse the positions of contestants on these polls.  Another real world example of this is that EBay and Yahoo for example both need what is called word verification.  This is where they make the user type a warped looking word on the screen on occasion and usually when creating accounts.  The reason they do this is to prevent computers from controlling either Yahoo or EBay under automatic control.  With word verification, a person would enter a word they see on the screen with each vote they make for a favorite in a poll.  It would be slower to enter repeat votes, but it would allow the person to vote as many times as they liked.  While it may be more effort to put a vote in, typing for example, "Cat" wouldn't be that hard to do.  Also, it would prevent any computers from voting automatically because a real person would have to read the on-screen graphical word.  This way a person could vote 20 or 30 times, but not 100,000 like a computer could do.  Hopefully AOL will come up with a better system than this.

When I found out how to read aol's internal values, it was really surprising to see what was going on. I know it's not possible, but I wish everyone would be able to see what I found out.   The aol poll can be reached at this site:  http://tv.channel.aol.com/americanidol

To understand this, you need to know that "authorized", or "authenticated" AOL votes are counted as voters randomly cast their votes in a normal way by clicking with a mouse. "Non authorized" votes are counted by AOL when too many repeat votes come from the same computer.  AOL should be able to spot repeat votes because it's possible to know if the same IP address (read same computer) keeps voting or not. However, due to a mistake in judgment, the poll doesn't correctly discard repeat votes!  In a 2nd scenario The AuthVote symbol described below may differentiate between AOL and non AOL voters.  If this is true the poll can be far more skewed because it would mean AOL has NO ability to detect bad or repeat votes. I will bring up both scenarios. Actually the poll looks like an attempt to recreate the feel of repeat phone voting by enthusiastic fans. It was a great idea, but has a serious flaw as you will soon find out.

There is nothing wrong with people voting multiple times because it averages out. However there is a sneaky way to set up computers to vote for favorites automatically. You might cast 1, 10, 100 votes and that is fair, right?  But what if you found out than someone could cast around 10,000 votes while having lunch?  Or that they cast 80,000 votes while sleeping?   This is what has been happening to polls that do not have the proper security or protection. By using this method of auto voting, a computer can vote endlessly 24 hours a day, all week, 24/7. Even while the person sleeps, or goes out to lunch it votes. This adds up to a majority of the poll votes coming from a small number of high speed computers. Because of this, the results can become skewed.  As few as probably two to ten people have been able to control the aol poll by flooding it with automatic votes. The way I determined that is too long to explain, but it has to do with watching the internal numbers change and measuring how fast the votes came in for different contestants.   Considering scenario one, a large amount of non authorized (non authenticated) votes were sent one contestant.  This is the cause for the poll to show exaggerated effects.  One hint was that it never slowed down for a particular contestant, even in the middle of the night!  The stream of 7,000 votes per hour remained constant no matter what time it was.  

At roughly 7,000 votes per hour, the poll can be fed over a million votes from a single computer in a week. Four or five of computers makes it that much worse. The big flaw/problem with this voting method is obvious in that VERY FEW people know how to do it!  For people that are ready to bash me, or try to correct my math, please understand that I did this as a favor. And also because I want a fair voting system. I have been watching AOL's internal values for over 7 weeks now and I know what I am talking about. I will make an attempt to prove I am not making this up near the end.

When you read numbers below, you will see the words "Reported" and "Actual". "Reported" means what people see on the poll screen after voting. "Actual" is the percentage of authorized votes based on 683,583 TOTAL authorized votes. These figures are taken directly from AOL's poll file. It is NOT shown on screen during voting without taking the correct steps to display them. Below in orange is actual data from the poll the night before Vonzell went home.  To see this weeks example with Bo and Carrie, go to the update link above.

-----------------------------------------------------------------------------------------------------------

Results as of 11:38p.m. EST on wed night, April 18th 2005.

Bo 245,234 authorized votes. 4,002,637 non authorized votes. Reported 60%, Actual - 35.87%

Carrie 328,532 authorized votes, 1,806,606 non authorized votes. Reported 30%, Actual - 48.06%

Vonzell 109,757 authorized votes, 539,750 non authorized votes. Reported 9%, Actual - 16.05%

-----------------------------------------------------------------------------------------------------------

Looking at the figures show that it's lopsided in an unnatural way on Bo's figures in the non authorized votes.  However, if  the 2nd scenario is considered then things become worse.  I've heard it mentioned that AOL may categorize votes according to whether they come from inside AOL (it's customers) or outside.  If this is true the meaning of that variable changes, however it shows that AOL isn't trying to track repeat votes at all.  In this case making spamming much easier.  In a way this makes sense, since it doesn't reflect this in the final total.  Let's consider this option for a moment.  Looking again at the figures I invite you to notice that only 6% of Bo's votes would come from AOL (authorized or authenticated) and 94% come from outside.  This would be fine except that 15% of Carrie's votes come from AOL.  This shows that roughly 3 times as many aol voters prefer Carrie over Bo.  But outside of Aol, it's the reverse.  This is very lopsided.

Below is another example from 5/25/05 6:04a.m. EST. Vonzell had gone home at this point.

----------------------------------------------------------------------------------------------------------

 5/25/05 6:04a.m.. EST.  

Bo 352,477 authorized votes. 2,535,189 non authorized votes. Reported 61%, Actual - 30%

Carrie 810,898 authorized votes, 1,033,092 non authorized votes. Reported 39%, Actual - 69%

---------------------------------------------------------------------------------------------------------------------------------

Bo would have only have 12% votes from inside AOL and Carrie would have a whopping 44% of her votes coming from inside AOL.  It's rather strange that nearly a million AOL voters prefer Carrie, yet no one on AOL seems to like Bo.  Looking at the Actual, 69% of AOL voters prefer Carrie over Bo! So Either way, something is very off with these numbers.

Yet another example is below, 5/25/05 4:45p.m. EST (Finale result show night).  Here Bo and Carrie are 50/50.  Mysteriously, their scores have become about even within 24 hours of the story breaking news.  Perhaps Carrie fans voted harder, but when you look at the votes from inside AOL in comparison to Carrie's they are obviously disproportionate. Only 31% of Bo's votes came from AOL users while 64.9% of Carrie's voters came from AOL.  So this is saying 2/3rds of Bo's supporters are outside of AOL, and 2/3rds of Carrie's are inside of AOL.  Of course to think that Bo or Carrie's supporters only come from AOL customers or non AOL customers is not really an acceptable answer. 

---------------------------------------------------------------------------------------------------------------------------------

5/25/05 4:45p.m. EST (Finale result show night before the show)

Bo 1,501,249 authorized votes. 3,592,809 non authorized votes. Reported 50%, Actual - 31%

Carrie 3,271,193 authorized votes, 1,767,063 non authorized votes. Reported 50%, Actual - 68.9%

---------------------------------------------------------------------------------------------------------------------------------

Whether AOL records authorized votes, or doesn't, this is not reflected in the outcome of the poll.  So it's easy to see that an automated process has been used to feed votes to the poll.

In watching in the past weeks, the poll  restarts on Wed of every week.   It's at this point that high speed computer voters start voting for their favorite automatically.  The idea of poll spamming doesn't stop with AOL either.  It's worth mentioning that other polls including the one on Fox's main web site has also been spammed using the same technique.  One poll allowed people to decide what single the favorites would sing on the CD that was to be released.  It was from a choice of three songs.  To think that the single could be chosen by ONE single person is a rather unsettling thought.  Because of the simplicity of Fox's design, it doesn't include a time delay.  It's harder to get around AOL's time delay (sliding the graphics of the finalist as it sorts takes 8 seconds), but with a simple trick that time delay can be cut to a fraction of a second.  Fox's site has no delays, and it can be spammed 8 times faster...

Here is a very brief description of how to find the AOL secret file for anyone technically inclined.  AOL stores this file on each voters computer as they vote. It's a special hidden file. For the technically inclined, go to this directory on your computer, "C:\Documents and Settings\XYZ\Local Settings\Temporary Internet Files\Content.IE5". Change XYZ to be the name of your computer. In XP, click START and read the name near the top. That's the computer name. Open "My computer" and paste this directory into the address bar. Do advanced search and look for hidden files that match "XmlPoll". The file will look like this: XmlPoll[1].xml. The vote window must be left open.  If it does not show up, cast 2 votes and then go back and search again.  This will cause the javascript/flash combination to store the numbers.  Note different versions of windows may not be the same. Click on that file and you should see the information printed below. How I found this out would take a long time to explain, so I will skip it.

**************************************************************

Inside AOL poll file 

Here is a copy of the XmlPoll.xml File described above. Note Vonzell is near the bottom and a new copy would read zero since she went home. All other contestants with zero's have already gone home. Ignore complex symbols and look for obvious things like authVote, nonAuthVote etc. See above explanation.  If you consider the 2nd scenario, simply look at the fact that authVote vs nonAuthVote between Carrie, Vonzell and Bo are disproportionate.  Meaning that people from auth don't vote the same way as people with nonAuth.  Either way, these numbers indicate problems.  

11:38p.m. EST Wed, May 18th, 2005 results.

<?xml version="1.0" standalone="yes" ?>

- <Poll pID="55286" pType="0" vID="55234" vStatus="0" qPos="1" buttonText="" country="us" lang="en" brand="aol" tID="0" backtovote="true" resultText="Total votes:" isLast="1">

- <Question isLast="1" qID="98701" qType="0" totalVotes="7032567" qText="Who should win American Idol?">

- <Answer aID="427602" aWeight="0" aText="Anthony">

<baseVote>0</baseVote>

<authVote>0</authVote>

<nonAuthVote>0</nonAuthVote>

<allVotes>0</allVotes>

<percentage>0</percentage>

</Answer>

- <Answer aID="427603" aWeight="0" aText="Anwar">

<baseVote>0</baseVote>

<authVote>0</authVote>

<nonAuthVote>0</nonAuthVote>

<allVotes>0</allVotes>

<percentage>0</percentage>

</Answer>

- <Answer aID="427604" aWeight="0" aText="Bo">

<baseVote>0</baseVote>

<authVote>245234</authVote>

<nonAuthVote>4002637</nonAuthVote>

<allVotes>4247871</allVotes>

<percentage>60</percentage>

</Answer>

- <Answer aID="427605" aWeight="0" aText="Carrie">

<baseVote>0</baseVote>

<authVote>328532</authVote>

<nonAuthVote>1806606</nonAuthVote>

<allVotes>2135138</allVotes>

<percentage>30</percentage>

</Answer>

- <Answer aID="427606" aWeight="0" aText="Constantine">

<baseVote>0</baseVote>

<authVote>0</authVote>

<nonAuthVote>30</nonAuthVote>

<allVotes>30</allVotes>

<percentage>0</percentage>

</Answer>

- <Answer aID="427607" aWeight="0" aText="Jessica">

<baseVote>0</baseVote>

<authVote>0</authVote>

<nonAuthVote>11</nonAuthVote>

<allVotes>11</allVotes>

<percentage>0</percentage>

</Answer>

- <Answer aID="427608" aWeight="0" aText="Lindsey">

<baseVote>0</baseVote>

<authVote>0</authVote>

<nonAuthVote>10</nonAuthVote>

<allVotes>10</allVotes>

<percentage>0</percentage>

</Answer>

- <Answer aID="427609" aWeight="0" aText="Mikalah">

<baseVote>0</baseVote>

<authVote>0</authVote>

<nonAuthVote>0</nonAuthVote>

<allVotes>0</allVotes>

<percentage>0</percentage>

</Answer>

- <Answer aID="427610" aWeight="0" aText="Nadia">

<baseVote>0</baseVote>

<authVote>0</authVote>

<nonAuthVote>0</nonAuthVote>

<allVotes>0</allVotes>

<percentage>0</percentage>

</Answer>

- <Answer aID="427611" aWeight="0" aText="Nikko">

<baseVote>0</baseVote>

<authVote>0</authVote>

<nonAuthVote>0</nonAuthVote>

<allVotes>0</allVotes>

<percentage>0</percentage>

</Answer>

- <Answer aID="427612" aWeight="0" aText="Scott">

<baseVote>0</baseVote>

<authVote>0</authVote>

<nonAuthVote>0</nonAuthVote>

<allVotes>0</allVotes>

<percentage>0</percentage>

</Answer>

- <Answer aID="427613" aWeight="0" aText="Vonzell">

<baseVote>0</baseVote>

<authVote>109757</authVote>

<nonAuthVote>539750</nonAuthVote>

<allVotes>649507</allVotes>

<percentage>9</percentage>

</Answer>

</Question>

</Poll>

                                                                                     You are visitor