Just got this interesting bit of information from a blog site dedicated to Taylor Hicks regarding predicting which contestant is most likely to be voted out :
« Taylor Interview from WadePredicting the American Idol results via busy signals - flawed methodology? Or maybe not?
So there is an interesting website called Dial Idol ( dialidol.com) that allows users to download a speed dialing application for American Idol. Assuming the application works as advertised (not verified by me) that makers attempt to predict the final results based on the number of busy signals the dialing application receives per contestant.
Clearly the assumption is that if you receive busy signals for your favorite contestant - you can be sure that others are voting for them. While I was highly skeptical that this made any sense a little investigation yields some interesting results.
What are the predictions for this week?
Higher numbers indicate the busiest lines, Taylor leads by a landslide in popularity - but can this be right?
Performer % Busy
Taylor 37
Ace 24
Chris 19
Will 17
Gedeon 14
Kevin 9
Sway 8
Elliott 7
Bucky 7
David 5
On their own website the operators say . . .
How accurate are the predictions?
That depends on the number of people running DialIdol. With minimal data during the last 6 votes of season 4 DialIdol was about 65% accurate including predicting who would win. With extensive data … well we’ll see!
If you read the following analysis you’ll see that the above statement is, well, mostly incorrect.
While we here at Gray Charles are pleased with the predicted results we were heavily skeptical that it was reliable so we contacted our friendly neighborhood statistician. Here is what he (my brother) had to say (take a deep breath).
First, a few definitions: The ‘population’ is the total number of all calls made, regular and computer dialed. In this case the population is 40 million calls. The ’sample’ is the cases selected for computer-aided dialing. In this case x (the percentage of all calls using dialing software) times the population yields the number of cases in the sample–let’s call it ‘y’, the total number of computer-dialed calls.
It sounds crazy, but the percentage, ‘x’ of the total is not critical in and of itself. It’s just the total number of cases ‘y’ in the sample that’s important. Thus if the population is really big–i.e. 40 million calls–’x’
can be really small and you can still end up with reliable results. Thus, it’s better (statistically) to sample 1% of 40 million calls than 50% of 400,000 calls. So, although ‘x’ is important in establishing the total number of calls ‘y’, I can’t give you a ‘critical’ figure of ‘x’.
Also–and this sounds crazy, too–the relative proportion yielding a positive result is important too. You’ll have more confidence in the estimate (narrower confidence limits) for an individual with a busy signal rate of about 50% than an individual with a busy signal rate of 10%. So even if they have the same number of attempted calls, each contestant’s confidence limits will be different. You’ll have to work it out for each contestant based on their busy signal rate and the number of calls placed for them.
Here’s some examples:
Candidate A: 20% busy signal, 500 calls. Confidence limits are 20% plus or minus 3.4%. Put another way, we can be 95% confident that the ‘true’ busy signal rate for Candidate A is between 16.6 and 23.3%.
Candidate B: 15% busy signal, 500 calls. Confidence limits are 15% plus or minus 3.1%. Put another way, we can be 95% confident that the ‘true’ busy signal rate for Candidate B is between 11.9 and 18.1%.
Candidate C: 5% busy signal, 500 calls. Confidence limits are 5% plus or minus 1.9%. Put another way, we can be 95% confident that the ‘true’ busy signal rate for Candidate C is between 3.1 and 6.9%.
Candidate D: 20% busy signal, 5000 calls. Confidence limits are 20% plus or minus 1.1%. Put another way, we can be 95% confident that the ‘true’ busy signal rate for Candidate A is between 18.9 and 21.1%.
Note that although candidate A and D have the same busy signal rates, the confidence limits are much narrower for candidate D. You can be reasonably sure that candidate D will receive more votes that candidate B, but not so sure that candidate A will.
Lots of assumptions here to be fair - we’ll be reporting more on the American Idol phone system structure in the upcoming days.
Bottom Line - Dial Idol maybe onto something here - at least in the early stages in predicting who stays and who goes. You have to look at margin of error to see how accurate the predictions are but for now, I think the David, Bucky, Sway and Elliott (wft? Elliott) might want to watch out.
Questions, comments, concerns? Simply freaking out?
Let me know.