|
|
|
|
 |
| Author |
 Topic  |
|
jrh7675
USA
520 Posts |
 Posted - 03/01/2006 : 10:58:49 PM
|
(Rules modified last modified on : 4/8/6.)
The DialIdol score is an algorithm that "measures the busy signal". Every three minutes the busy percentage is calculated for each Idol then weighted according to the time zones submitting data. Each intervals score is added up then averaged out to produce the DialIdol score. Using this method versus the original method of just calculating the actual busy percentage eliminates problems related to call volume and population influence.
For example:
First hour (Eastern & Central voting):
Elliot : 1000 total calls, 300 of which are busy
Taylor : 5000 total calls, 2500 of which are busy
Second hour (Everyone except Eastern & Central voting):
Elliot : 5000 total calls, 200 of which are busy
Taylor : 1000 total calls, 200 of which are busy
Elliot: 12.14% ( ((300/1000) * .78) + ((200/5000) * .22) ) / 2
Taylor: 21.7% ( ((2500/5000) * .78) + ((200/1000) * .22) ) / 2
Time zone scoring weight:
Eastern = .38
Central = .4
Mountain = .06
Pacific = .1525
Alaskan = .0025
Hawaiian = .005
An Idol will be predicted to be voted off if they are within a margin of error (MOE) of the highest ranked Idol predicted to be voted off. The MOE will be calculated in both directions (plus/minus). For example consider the following fictional busy percentage rank:
8) Kellie Pickler 9.0125
9) Stevie Scott 6.0606
10) Brenna Gethers 5.8851
11) Becky O'Donohue 2.9851 (base)
12) Heather Cox 0
In this case two Idols are going home and DialIdol scores Becky & Heather as leaving. Since Becky is higher her score is the base score. The margin of error is 3% so her score could be as high as 5.9851, therefore anyone with a busy percentage minus the MOE less than 5.9851 is predicted to be voted off. Brenna's 5.8851 minus the MOE equals 2.8851, clearly within the MOE. Stevie's 6.0606 minus the MOE equals 3.0606 also within the MOE so she will be included in the prediction. Kellie's 9.0125 minus the MOE equals 6.0125 which is greater than Becky's 5.9851 MOE adjusted score so she will not be predicted to be voted off.
In weeks where only one Idol gets a ticket home, all the stats are calculated from the last place contestant.
The margin of error (MOE) will be calculated per Idol, per week using the following formula. MOE = 2.58 * sqr(((p)*(1-p))/n) where p = each Idols DialIdol score (percent) and n = the actual score for each Idol. 2.58 is used for polls with a 99% confidence which is what DI is using.
Despite all this mumbo jumbo I still maintain the DialIdol is not scientific and not a true representation of the actual American Idol score. There are many reasons for this including but not limited to the fact that there are many issues with the ways modems detect busy signals, the fact that DialIdol does not consider text message votes and a host of other statistical factors including DialIdol's inability to consider age, sex, gender, financial position and many other factors that prevent the sample DialIdol uses from truely representing the actual voting pool.
Jim |
Edited by - jrh7675 on 04/08/2006 7:52:56 PM |
|
|
Samaria_3
USA
1 Posts |
Posted - 03/03/2006 : 8:09:23 PM
|
quote:
Originally posted by jrh7675
OK... So here is where I will comment on prediction stats. Check back here for updates to the "rules" DialIdol will play by.
The first rule I am making about how accurate DialIdol can be is this:
1) An Idol is predicted to be voted off is he/she is within five percentage points of the lowest ranked Idol. That could mean that DialIdol predicts more people getting voted off than can actually be voted off in that week.
This is fair and any statistical guru will tell you that. Polls & surveys have margins of error even when they are scientific - DialIdol is no different and is NOT scientific.
Jim
I'm really confused right now because I don't know what this site is, I kind of just got here because someone gave me a link...but you seem to know a lot. Do you think Will Makar will make the top 12? |
 |
|
|
|
|
jrh7675
USA
520 Posts |
Posted - 03/03/2006 : 10:51:43 PM
|
| Lol. OK well welcome to DialIdol... I don't know about Will. We shall see but I have a feeling he will. |
|
|
|
|
|
clevpetp
13 Posts |
Posted - 03/08/2006 : 12:18:52 PM
|
In your above fictitious scenario, you have to take into account that though Stevie had 6.06%, with a +/- 3%, her range is 3.06%-9.06%. Likewise with Becky, her range is 0%-5.98%. Statistically, Stevie was well within range of second bottom. Looks like both Brenna and Becky varianced high that week and Stevie was on the lower end of her variance. Of course this is fictitious so no harm.. no foul.
If you look at the real data from week 24, given a +/- 3% the ranges should be:
Kellie Pickler 43.83-49.83
Paris Bennett 30.89-36.89
Katharine McPhee 28.25-34.25
Lisa Tucker 24.86-31.86
Ayla Brown 15.86-21.86
Mandisa 15.86-21.86
Kinnik Sky 6.43-12.43
Melissa McGhee 3.06-9.06
Stevie Scott 3.06-9.06
Brenna Gethers 0-5.98
Becky O'Donohue 0-5.98
Heather Cox 0-3
So in this scenario the bottom 5 were all within range of each other. Though there is an anomoly in that Stevie should not have gotten the boot as she is statistically not within range of Heather. But I think that with it being the first week, there may have have been a higher margin. As we get more pollers, that margin of error will get smaller.
|
|
|
|
|
|
jrh7675
USA
520 Posts |
Posted - 03/08/2006 : 5:57:50 PM
|
You know you bring up a good point! While Stevie may not be within range of Heather she is statiscally tied with Becky & Brenna so in theory she was within the margin of error.
I am still sticking with the existing formula though because it's too hard to re-code.  |
|
|
|
|
|
scdawg
USA
1 Posts |
Posted - 03/09/2006 : 08:46:34 AM
|
could you please explain your raw numbers to me? It seems that the raw numbers predict a totally different outcome than your busy dialing numbers. WHy is this and which one is more accurate? |
|
|
|
|
jrh7675
USA
520 Posts |
Posted - 03/12/2006 : 6:54:43 PM
|
Raw numbers are the number of calls/busy signals/votes DI registered. Since DI lets people vote for anyone they want - different Idols will get more or less busy signals/votes. This is not a far method to compare Idols...
Instead DI takes the busy percent - the percent of a single Idols calls that were busy and rates each other against it.
Soooo... If Taylor got 1000 calls and 600 were busy - his busy percent is 60. Also, if Kellie got 100 calls and 60 were busy her busy percent is 60% as well. DI thinks it's more fair to judge by busy percent but gives you the raw numbers for fun.
Jim |
|
|
|
lks
27 Posts |
Posted - 03/14/2006 : 11:09:27 PM
|
quote:
Originally posted by clevpetp
In your above fictitious scenario, you have to take into account that though Stevie had 6.06%, with a +/- 3%, her range is 3.06%-9.06%. Likewise with Becky, her range is 0%-5.98%. Statistically, Stevie was well within range of second bottom. Looks like both Brenna and Becky varianced high that week and Stevie was on the lower end of her variance. Of course this is fictitious so no harm.. no foul.
This is not how statistics works. Otherwise, you could basically go all the way up the line and say "Well, the 8th place is within the MOE of the 7th place. And the 7th place is within the MOE of the 6th place. So technically speaking, the 8th place is within the MOE of 6th place. And 5th place is within the MOE of 6th. So really, 5th-8th are all up in the air."
IT just doesn't work that way.
Also, a 3% MOE does NOT mean that their busy signal percetnages is within 3%.
Have you guys ever even TAKEN a statistics class? |
|
|
|
gumbo
4 Posts |
|
|
faithdrool
2 Posts |
Posted - 03/15/2006 : 4:45:01 PM
|
that website addresses exactly the issue i posted on another thread. that, because the voting traffic is much higher right after the voting period starts, you have to make some kind of distinction for when the dial was made.
One issue that neither dialidol nor above site addresses is that regional differences in each contestant's fan base. It is possible that high percentage of Lisa fans, for example, are from Southern California and Cingular network there can withstand a LOT MORE calls than say, some small city south of Georgia. If this were the case, each busy signal made for Lisa would represent a lot more callers compared to a busy signal received while voting a contestant within Cingular network where there are fewere towers and, so fewer calls saturate the network. |
|
|
|
flygirl1964
USA
1 Posts |
Posted - 03/21/2006 : 11:16:37 PM
|
Ow my head hurts!! Statistics was soooo long ago!
Not sure if this can REALLY predict who is ahead, but it is a VERY interesting concept. When I am dialing, if I get through for a favorite a few times in a row, I automatically think, "Oh no, they are in danger! I'd better call more for them!"
I also vote for people who may attract the same target market as the contestant I want voted off. Yes, I took statistics as part of my marketing major... 
Of course I vote for my favorite singers a gazillion times!
You all may be interested to know that although Mandisa appears at the top of the list on this site, that I (East Coast) called her number between 10:30 and 11 and I got through 9 or 10 times IN A ROW. I was very surprised! Most of the night other than that, it has been nearly impossible to get through for the 'Diva!
This website is, as I said, very interesting. Bear in mind what the moderators are saying: It's not exact, it is just a way of predicting (by how busy someone's number is) whether or not they will be eliminated.
Of course, there is a statistical percentage to be taken into effect, which means that although Elliott is at the "bottom" right now, that Kevin is within a few percentage points of him and could also be the one to go.
QUESTIONS FOR THE WEBSITE MODERATOR:
Do you have a percentage of success (prediction of who will be voted off as to who actually will) week to week? (past success?)
How many phones do you use to compile the busy signal data?
Just curious.
Have a good one! Interested to see how accurate this is, but hoping Elliott will not go as he has a fabulous voice!
|
|
|
|
lks
27 Posts |
Posted - 03/22/2006 : 05:24:20 AM
|
| By the way...there is no such thing as a "margin of error" for a NON-SCIENTIFIC survey! The phrase margin of error ONLY applies to a real statistical analysis. |
|
|
|
lks
27 Posts |
Posted - 04/05/2006 : 02:39:49 AM
|
quote:
Originally posted by jrh7675
(Rules modified last modified on : 3/18/6.)
An Idol will be predicted to be voted off if they are within a margin of error (MOE) of the highest ranked Idol predicted to be voted off. The MOE will be calculated in both directions (plus/minus). For example consider the following fictional busy percentage rank:
8) Kellie Pickler 9.0125
9) Stevie Scott 6.0606
10) Brenna Gethers 5.8851
11) Becky O'Donohue 2.9851 (base)
12) Heather Cox 0
In this case two Idols are going home and DialIdol scores Becky & Heather as leaving. Since Becky is higher her score is the base score. The margin of error is 3% so her score could be as high as 5.9851, therefore anyone with a busy percentage minus the MOE less than 5.9851 is predicted to be voted off. Brenna's 5.8851 minus the MOE equals 2.8851, clearly within the MOE. Stevie's 6.0606 minus the MOE equals 3.0606 also within the MOE so she will be included in the prediction. Kellie's 9.0125 minus the MOE equals 6.0125 which is greater than Becky's 5.9851 MOE adjusted score so she will not be predicted to be voted off.
In weeks where only one Idol gets a ticket home, all the stats are calculated from the last place contestant.
The margin of error will be calculated each week and may vary from week to week and even from guys to girls (if they are separated). I will try to use a 3% margin of error when possible but 5% is more reasonable to me.
This is fair and any statistical guru will tell you that. Polls & surveys have margins of error even when they are scientific - DialIdol is no different and is NOT scientific.
Jim
Actually, polls and surveys have Margin of Errors ONLY when they are scientific. Non-scientific surveys cannot by definition have an MOE.
MOE's are not arbitrary guesses as to what the error might be. It is something that is CALCULATED based on a formula.
Since this obviously is a simple fact lost on you and everyone on this board...let me explain how the Margin of Error works:
In a statisical sample, the true results will fall within the MOE of the predicted result 95% of hte time. They wil fall within 2 MOEs 98% of the time.
Now..let's throw all logic out the window and pretend that MOE's applied to this site. Let's say that the MOE is 3% and Ace had a 19.5% busy percentage.
That would mean that there is a 95% chance that the TRUE percentage of busy signals is betwee, 16.5% and 22.5%. There is about a 2.5 percent chance that his true busy percentage is 22.5 or great, and a 25% percent chance that it is lower than 16.5
Now, let's just look at the extremes. At 22.5%, there is EXACTLY a 2.5% chance
So, let's move on to someone higher. Let's say that Bucky's line is busy 25.4% of the time. His 95% probability range is 2.4% through 28.4% However, there is only a 2.5% chance that it is as low as 22.4
In other words, for Bucky to be truly lower than Ace, it would have to be 2.5% squared. Or to put it mathematically .025 x .025.
What this would mean is that using your method, you are claiming something is within a reasonable MOE when there is only a 0.0625% chance of it happening. That is SIX times out of TEN THOUSAND.
In othr words, WAY out of reasonable range. |
Edited by - lks on 04/05/2006 02:41:23 AM |
|
|
|
lks
27 Posts |
Posted - 04/05/2006 : 02:57:26 AM
|
quote:
This is fair and any statistical guru will tell you that. Polls & surveys have margins of error even when they are scientific - DialIdol is no different and is NOT scientific.
Jim
Actually, polls and surveys have Margin of Errors ONLY when they are scientific. Non-scientific surveys cannot by definition have an MOE.
MOE's are not arbitrary guesses as to what the error might be. It is something that is CALCULATED based on a formula.
Since this obviously is a simple fact lost on you and everyone on this board...let me explain how the Margin of Error works:
In a statisical sample, the true results will fall within the MOE of the predicted result 65% of the time. They wil fall within 2 MOEs 95% of the time.
Now..let's throw all logic out the window and pretend that MOE's applied to this site. Let's say that the MOE is 3% and Ace had a 19.5% busy percentage.
That would mean that there is a 65% chance that the TRUE percentage of busy signals is betwee, 16.5% and 22.5%. There is about a 17.5 percent chance that his true busy percentage is 22.5 or great, and a 17.5% percent chance that it is lower than 16.5
Now, let's just look at the extremes. At 22.5%, there is a 17.5% chance
So, let's move on to someone higher. Let's say that Bucky's line is busy 25.4% of the time. His 65% probability range is 2.4% through 28.4% However, there is only a 17.5% chance that it is as low as 22.4
In other words, for Bucky to be truly lower than Ace, it would have to be 17.5% squared. Or to put it mathematically .175 x .175.
What this would mean is that using your method, you are claiming something is within a reasonable MOE when there is only a 3.6% chance of it happening. This is not reasonable.
It is plus OR minus the MOE. Not plus AND minus.
[/quote] |
|
|
|
Atlantix2000
USA
103 Posts |
Posted - 04/05/2006 : 08:12:44 AM
|
Yo, lks, stop responding to weeks old posts and pointing out that people are doing things wrong. If you catch up to the recent posts, you'll see Jim has adopted a calculated MOE. Sure DialIdol is still not scientific, but it's meant as a fun project for him, and is still very much a work in progress. He's made many improvements and will probably continue to do so as long as he's having fun. There's no need to be so harsh.
--A2K |
|
|
|
kevinjmoore
USA
1 Posts |
Posted - 04/05/2006 : 9:55:10 PM
|
Nobody here knows anything about statistics. Margin of Error is not the same as standard error. +/- 1 standard error in the sampling distribution of sample means should include 68.3% of the data in a normal distribution according to the Empirical Rule. +/- 2 standard errors should include 95.5% of the data in a normal distribution and +/- 3 standard errors includes 99.7% of the data in a normal distribution. Margin of error refers to the total amount of error on each side of the mean +/- margin of error = # of standard errors times (x) the size of the standard error.
Kevin J Moore |
|
|
|
Topic |
|
|
|
| DialIdol.com Forums |
© www.DialIdol.com |
|
|
|
|
|
|
DialIdol.com is NOT affiliated in any way with FOX TV ®, SO YOU THINK YOU CAN DANCE ®, or FremantleMedia North America, Inc.
|
Contact
DialIdol
| Links
| Terms Of Service
| Privacy
Copyright © 2009. James Hellriegel Jr, all rights reserved.
|
|
|
|
Topic Locked
