|
|
|
|
 |
| Author |
 Topic  |
|
|
jrh7675
USA
521 Posts |
 Posted - 03/04/2006 : 3:59:25 PM
|
Week w/2 (to 1) Idols Comments
Of 23 Idols to leave the show, 20 of them were within a very reasonable statistical margin of error.
Obviously measuring the busy signal works.
See you next season!
Congratulations Taylor!
Week w/3 (to 2) Idols Comments
Of 22 Idols to leave the show, 19 of them were within a very reasonable statistical margin of error. This means that considering statistical ties DialIdol predicted 19/22 Idols who left the show. In any given week if there is a statistical tie DialIdol predicts multiple people to go home and if just one of them goes home then DialIdol is right. This week for example Katharine & Elliot were tied for last place and since Elliott went home DialIdol was right - but even if Kat went home DialIdol would have been right.
Week w/4 (to 3) Idols Comments
Of 21 Idols to leave the show, 18 of them were within a very reasonable statistical margin of error.
If you didn't believe DialIdols predictions this week then you weren't the only one! I really didn't believe it either. Still wish it wasn't true! 
Week w/5 (to 4) Idols Comments
Of 20 Idols to leave the show, 17 of them were within a very reasonable statistical margin of error.
Week w/6 (to 5) Idols Comments
Of 19 Idols to leave the show, 16 of them were within a very reasonable statistical margin of error.
Pretty interesting results show eh? Kellie was statistically tied with Paris so technically DialIdol got it right again last night but admittedly it's more fun when the bottom person on DI is the bottom person on AI.
Now as far as the rest of them go. Technically there is a way to arrange the DialIdol ranks to match the results from the show last night - except for Katharine. Don't believe me? Look at again at their DialIdol Rank on the predictions page. Considering statistical ties DialIdol correctly predicted 5/6 of the Idols positions. What's a bit eerie to me is that 5/6 = 83% which was DialIdol's accuracy rate going into this week.
Also of note - the one DialIdol got wrong, Katharine - got more calls and more votes than anyone else on DialIdol by far.
Week w/7 (to 6) Idols Comments
Of 18 Idols to leave the show, 15 of them were within a very reasonable statistical margin of error.
Well DialIdol's accuracy continues to surprise me. Sure it's not 100% accurate but it is exceeding my wildest expectations!
Ace was sent home last night and Paris was sent BACK last night but that does NOT mean she came in third. The truth is we don't know the precise order Chris/Paris finished in - unless you trust DialIdol a little too much. 
Week w/8 (to 7) Idols Comments
Of 17 Idols to leave the show, 14 of them were within a very reasonable statistical margin of error.
Well if nothing else DialIdol correctly predicated that DialIdol wouldn't always be right!
Obviously everyone will have their own interpretation over how accurate DialIdol was and that makes the world go around. Of course DialIdol's spin is that within the margin of error DialIdol got it right. Ace/Bucky were separated by two tenths of one percent. This is a virtual tie not to mention a statistical tie. Also, Elliot was even within that margin of the bottom three.
Week w/9 (to 8) Idols Comments
Of 16 Idols to leave the show, 13 of them were within a very reasonable statistical margin of error.
Wow!
Well there are a lot of reasons DialIdol is more accurate this week but let’s not discount luck and coincidence just yet. The other primary reason would be the fact that the scoring system has been tweaked to account for things that were throwing DialIdol off, primarily population weight and late voting.
I’d like to thank everyone who participated in this thread: http://www.dialidol.com/asp/forums/topic.asp?TOPIC_ID=78&whichpage=1 for helping get the algorithm tweaked for week 9.
Finally let me set some expectations here. I do not expect the predictions to be this accurate every week! I will be surprised and pleased if DialIdol can maintain the 80% accuracy rate it’s got now. The predictions primary purpose - by far - is to help you direct votes to your favorite in the most need. I encourage you to watch the results show because it is the only way you will ever know how your favorites are doing and it’s also the only way to figure out how DialIdol is doing!
Finally - thanks to everyone for helping to make this site so popular!
Week w/10 (to 9) Idols Comments
Of 15 Idols to leave the show, 12 of them were within a very reasonable statiscal margin of error.
Well it's hard to be wrong when you predict 70% of the contestants will get the boot! I am not thrilled to take credit for "getting it right" this week but statistically speaking DialIdol did get it right. I like the new scoring system better and I think we'll stick with it. Next week I do expect to adjust the equation slightly to compenstate for population differences.
All in all I think everything made sense this week. Simon mentioned that based on the performances anyone could go and I believe that was reflected in DialIdol's stats.
This week also helps prove a point I keep harping on - if you want to know who is going to get voted off you have to watch the show. DialIdol is not meant to replace the results show - in fact DialIdol adds more drama and suspense by adding the element of "is the busy signal an accurate measure?" question to the drama of the results show.
Week w/11 (to 10) Idols Comments
Of 14 Idols to leave the show, 11 of them were within a very reasonable statiscal margin of error (2%).
Well the good news, in my opinion, is that the person with the least talent left the show. More good news is that this week helps prove a point I've been trying to make - if you want to know who is going to get voted off you have to watch the show. DialIdol is not meant to replace the results show - in fact DialIdol adds more drama and suspense by adding the element of "is the busy signal an accurate measure?" question to the drama of the results show. While DialIdol didn't get it right this week I am thrilled that it continues to be used as a tool to help people vote more intelligently while sparing thir finger tips the pain of the redial button.
Prediction Method Change Effective 3/18/2006
In order to further comply with margin of error (MOE) I have revised predictions to calculate based on plus/minus the MOE whereas before I was simply calculating on plus MOE. This makes DialIdol look better so to be fair I bumped the MOE down 1% for all weeks to 2%.
Week w/12 (to 11) Idols Comments
Of 13 Idols to leave the show, 11 of them were within a very reasonable statiscal margin of error (2%).
For all the fuss about DialIdol this week it turns out it wasn't very accurate. Sure the general location of Idols seems right but you still had to tune in to see that Melissa got the boot - not Lisa, Ace or Kevin. Nonetheless people seem impressed that DialIdol put Lisa & Ace in the bottom three and that they were indeed in the bottom three.
Overall I'm quite pleased with the number of calls placed by DialIdol this week (almost 18,000) but a little disappointed in the results.
Comments below reflect the previous scoring system of plus the MOE - not plus/minus.
Week w/16 (to 12) Idols Comments
Of 12 Idols to leave the show, 10 of them were within a very reasonable statiscal margin of error (3%).
As many people suspected (myself included) Lisa Tucker was pegged incorrectly but everything else fell in line. 75% this week. Not too shabby. It's a little comfort to note that Ayla wasn't too far behind and she was in the bottom three (as far as DialIdol goes).
Surprisingly DialIdol got the guys right and the girls 1/2 right even with all the problems we had Wednesday night (when the guys voted).
Week w/20 (to 16) Idols Comments
Of 8 Idols to leave the show, 7 of them were within a very reasonable statiscal margin of error (3%). The one that wasn’t missed the margin by like .1% and if that weren’t enough it was on the first night this year with a very low call volume. 7/8 is 88% accurate which is pretty darn good for such low volume.
And if you look at the data further you’ll see it’s even closer than that. For example this week all four girls were within 1.9% of each other - amazingly close. For the guys DialIdol nailed David as the low vote getter dead on and the second lowest vote getters were seperated by 1.3%. If you look at the previous weeks stats you see the same story except for the first night where Stevie was an “anomally”.
Being accurate within 1.9% & 1.3% is pretty f’ing good if I do say so myself. Take into consideration that most Presidential polls have margins of error of 3%.
Now I am not saying it will be like this every week - no way, no how… I am simply “spinning” the results we’ve got to show how they have been accurate so far. Part of the reason for doing this is to actually validate whether “measuring the busy signal” works or not and so far I think it’s pretty accurate.
Jim |
Edited by - jrh7675 on 05/25/2006 05:50:32 AM |
|
|
mojoidol
41 Posts |
Posted - 03/27/2006 : 01:41:36 AM
|
Hi: The first time I visited Dial Idol the site was predicting LT and Ace in the bottom 3. I could hardly believe it so I was extremely impressed when that prediction came true. The following week the bottom fell out as DI predicted Elliott to be booted and he was not even in the bottom 3.
I know the "What went wrong" question has been asked, but I have not read a satifactory answer yet. If you don't know how the Elliot selection was so off, I am wondering how DI can pick the bottom 3 with any kind of confidence?
One other note, I talked to a friend in Fla. and when I told him that DI had Elliot at the bottom he found that strange because he said he was unable to get through to Elliot on voting night as often as he was getting through to Kevin. This flies in the face of your poll. There seems to be a major flaw that needs fixing. Good luck in nailing your picks for this coming week. |
 |
|
|
|
|
TypoII
USA
1 Posts |
Posted - 03/29/2006 : 7:31:17 PM
|
quote:
mojoidol wrote:
Hi: The first time I visited Dial Idol the site was predicting LT and Ace in the bottom 3. I could hardly believe it so I was extremely impressed when that prediction came true. The following week the bottom fell out as DI predicted Elliott to be booted and he was not even in the bottom 3.
I know the "What went wrong" question has been asked, but I have not read a satifactory answer yet. If you don't know how the Elliot selection was so off, I am wondering how DI can pick the bottom 3 with any kind of confidence?
One other note, I talked to a friend in Fla. and when I told him that DI had Elliot at the bottom he found that strange because he said he was unable to get through to Elliot on voting night as often as he was getting through to Kevin. This flies in the face of your poll. There seems to be a major flaw that needs fixing. Good luck in nailing your picks for this coming week.
I once voted for Fantasia to try and save her before things started to look up for her. I did 1,000 text messages from a Blackberry (7290 I think) in just under ten minutes. I got a sore thumb and a huge bill but it's possible to really, really turn up the voting with text messages. Many cell-phone companies are starting to offer unlimited text messaging for a flat rate.
I'm betting texts equal or even exceed calls for AI voting. The dialidol statistics might be perfectly accurate but the text messaging slants the results.
Also: I don't know how the 866 system works but does Verizon capture the AI votes at the regional switch or at one central location. My guess is it's regional so Ellito might be getting bukoo votes on the east coast while there is essentially no busy signal on the west coast (I don't know where DI places calls from but something like this is sure to be slanting results as well...)
Regardless: I run a large AI pool and this is a GREAT resource.
Good job DI! |
|
|
|
|
|
lks
27 Posts |
Posted - 03/30/2006 : 01:02:17 AM
|
You wore out your thumb needlessly. I am guessing the only way you amassed that many text messages was to send one VOTE message to multiple people, with all of them being Fantasia's number. (Otherwise you couldn't physically send that many in two hours.)
Well, guess what? Fantasia got credit for ONE vote for each time you actually sent the text message, regardless of how many times her number was included in the TO message.
Also, phone votes FAR outnumber text messages, plus there is no signficcant difference in the percentage of text to phone votes that any contestant gets. Basically, if, say, Ace gets 70% of his votes from phone votes and 30% from text messages, then every contestant gets about the same ratio. |
|
|
|
|
|
angryfan
45 Posts |
Posted - 03/30/2006 : 11:27:13 AM
|
Where is your proof re: text message votes vs phone votes? I think you are just blowing smoke out of your own *** to try to make it seem more reliable than it is.
As for text messages - you seem not to know how they work. Once you type in and send a message, all you have to do is resend it like you would when hitting 'redial' and contrary to your biased opinion, this -does- count as a vote each time, and not one vote in total. |
|
|
|
|
|
lks
27 Posts |
Posted - 04/05/2006 : 02:04:50 AM
|
quote:
Originally posted by jrh7675
Week w/10 (to 9) Idols Comments
Of 15 Idols to leave the show, 12 of them were within a very reasonable statiscal margin of error.
Well it's hard to be wrong when you predict 70% of the contestants will get the boot! I am not thrilled to take credit for "getting it right" this week but statistically speaking DialIdol did get it right. I like the new scoring system better and I think we'll stick with it. Next week I do expect to adjust the equation slightly to compenstate for population differences.Jim
Yawnnn...Once again, this is NOT true..MOE does NOT mean that if the 9th place is within the MOE of hte 8th pace, and the 8th place is within the MOE of hte 7th place, and the 7th place is within the MOE of hte 6th place, and the 6th place is within the MOE of the 5th place that the 9th place is within the MOE of the 5th place..
Here is how statistucs REALLY works.
If the last place is say, 19.2%, and your arbitrary MOE is 3%, then ONLY those with a busy percentage/score of 22.2% or lower are within the MOE and therefore are statistically the same.
It does NOT matter if the second to last is, say 21%, and the third to the last is 23% and the 4th to last is 25%...
By your faulty logic, the one with 25% is within the MOE . But in REALITY, it is not.
Also...I don't know who told you to use +/- for hte MOE, but that does NOT mean that if the MOE is 3%, and two people are within 5.5% that they are within the MOE. You CANNOT SUBTRACT the MOE from one, and add it to another.
My god! Did you not even take HIGH SCHOOL level statistics!
I posted this in another thread, but I'll do it again.
Here is how Margin of Error works . (By hte way, MOE is a CALULABLE number, not a guess that someone thinks it is within.)
Let's say that the MOE is 3%, and Ace's lines are busy 20% of the time. What this means is that the range is 17-23% Statistics would tell us that there is a 65% chance that the TRUE value falls within this range. (And that there is 95% chance that it falls within 14-26%.) At the extremes...17% and 23%, there is exactly a 17.5% chance of it occurring. (100%-65%)/2
With me so far? So...there would be a 17.5% chance that Ace's number is actually as high as 23%.
Now let's say that Bucky's was 26%. This would make his 65% range be 23% and 29%. And at the extreme of 23%, it would be exactly 17.5% likely.
So, for BOTH of these extremes to be the case, both 17.5% probabilities must be true. Or...the chances of that happening are 0.175 times 0.175.
Simple math shows that result to be 0.03625, or 3.6%
So...what you are claiming to be within aa reasonable MOE (subtracting the arbirtary margin from one and adding it to the other) wil actually only occur 36 times out of 1000.
This is NOT reasonable.
I honestly can't tell if you guy are all really ignorant about how MOE's and statistics work, or if you are just ignoring reality so that the predictions SEEM better. But the fact remains that if you truly apply the correct method, the accuracy rate of the predictions drops CONSIDERABLY.
That is why when you analyze it based on MOE, you take plus OR minus. Not plus AND minus. |
Edited by - lks on 04/05/2006 02:54:28 AM |
|
|
|
|
lks
27 Posts |
Posted - 04/05/2006 : 02:09:14 AM
|
quote:
Originally posted by angryfan
Where is your proof re: text message votes vs phone votes? I think you are just blowing smoke out of your own *** to try to make it seem more reliable than it is.
As for text messages - you seem not to know how they work. Once you type in and send a message, all you have to do is resend it like you would when hitting 'redial' and contrary to your biased opinion, this -does- count as a vote each time, and not one vote in total.
It still must dial. And you can't dial that many times (over 1000) in two hours. Even with just pushing RESEND. |
|
|
|
foobar
Suriname
67 Posts |
Posted - 04/06/2006 : 07:30:54 AM
|
You've increased the sample size a LOT in recent weeks. The error size will decrease with the sqrt of the sample size. So, with the roughly fold increase in sampling in the last 4 weeks, you decrease the standard error from about 1% to about 0.37%...doubling sample size again gets your standard error, ballpark, to about 0.25%.
I don't think it is going to get much more accurate than it is now. Most of the gains were in the algorithm, it will get slightly more accurate with even more sampling. I also agree, you will not get it this accurate every week. But, if something is off by more than 1-2% in busy rate, it will look very suspicious, like kicking Kevin off a few weeks ago...
Bravo on a great job!
Uh Huh |
|
|
|
fingers
14 Posts |
Posted - 05/06/2006 : 09:56:15 AM
|
User foobar has made an excellent discovery about bad data being allowed through for Katharine (see the "Elliott not Katharine will be in the bottom 2" thread).
I strongly recommend adding a check at the end of each timezone shift to throw out the data from clients that are sending in clearly bad data. Reports of 100% no-busy should clearly be rejected. And now that there are multiple phone lines available, 100% busy reports are probably also suspect. |
|
|
|
Sunkist
1 Posts |
Posted - 05/24/2006 : 4:35:38 PM
|
lks is correct in saying that someone with a predicted score of 15 and someone with a predicted score of 21 are not statistically "tied" if both have a MOE of 3. However, lks' analysis of what MOE actually means and how it is applied is wrong.
In the case of dialidol, MOE refers to the value where the lower bound of a 99 percent confidence interval is Predicted value minus MOE, and the upper bound of a 99 percent confidence interval is predicted value plus MOE. That means if someone's predicted score is 20 and the MOE is 3, we can say with 99 PERCENT CERTAINTY that his actual score is somewhere between 17 and 23 (***uming normal distribution). We know this because of the formula used in http://www.dialidol.com/asp/faq/faq.asp#What_is_a_Margin_of_error.
(The value of 2.58 that he plugged into the equation results in a MOE that corresponds with a confidence interval of 99 percent, you can look this value up in statistical tables.)
I believe lks is confusing MOE as it is used here with Standard Deviation. They are two different things as the margin of error on dial idol corresponds with 2.58 times the standard deviation. |
|
|
| |
Topic |
|
|
|
| DialIdol.com Forums |
© www.DialIdol.com |
|
|
|
|
|
|
DialIdol.com is NOT affiliated in any way with FOX TV ®, AMERICAN IDOL ®, or FremantleMedia North America, Inc.
|
Contact
DialIdol
| Links
| Terms Of Service
| Privacy
Copyright © 2010. James Hellriegel Jr, all rights reserved.
|
|
|
|
Topic Locked
