My sister was doing some Internet research on American Idol, and during lunch, she told me about a programme that has accurately predicted the weekly results. So, I decided to reverse engineer the programme and this is, (I think), how it works...

.

**The Science of Interactive Reality TV Voting Prediction**

.

Voting, on shows like American Idol, is done via phone lines, to various call centres with toll-free number(s). Think of it like an altered Election Polling Station.

.

In a normal poling station, the voters queue up in a few lines to cast their votes, with one ballot box per line. Voters of, say Party A and Party B, are normally distributed within those few lines, and they cast their vote, regardless of affiliation, into the same ballot box. It would be difficult to predict the voting trend in the centre based on questioning a sample set of people since, the ballot is supposed to be secret. Hence, the alternate way is to "steal" one of the ballot boxes, and predict the voting trend from that box. Of course, in real life, one ballot box represents a locality and may be biased and may only reveal a trend in that locality.

.

Now, let's alter this station. This will never happen in reality, but it reflects the system of phone voting. Voters of Party A and B are made to queue in separate lines and vote into boxes designated for that party. However, to maintain secrecy, the voters line up in a pitch dark room, thus concealing the length of the line of voters of each party, and hence, preventing a foregone conclusion on the voting results. To increase efficiency, there are a few of these ballot boxes for each queue and the voter is streamed into one of the available boxes. Since there are more voters than boxes, there is a decent chance that you may end up in a short secondary queue to get to anyone of the boxes.

Now, let's alter this station. This will never happen in reality, but it reflects the system of phone voting. Voters of Party A and B are made to queue in separate lines and vote into boxes designated for that party. However, to maintain secrecy, the voters line up in a pitch dark room, thus concealing the length of the line of voters of each party, and hence, preventing a foregone conclusion on the voting results. To increase efficiency, there are a few of these ballot boxes for each queue and the voter is streamed into one of the available boxes. Since there are more voters than boxes, there is a decent chance that you may end up in a short secondary queue to get to anyone of the boxes.

.

But... a phone line is binary, i.e. it's either busy or not. And so, if there's a person at the "box" when you get there, you have to start over. In the case of this station, "stealing boxes" is useless, since they will hold no information on the voting trend. So, we have to be even more creative...

.

Any one person will not be able to tell you how many people voted along with him, since, remember, the room is pitch dark... and of course, in this case, multiple votes per person is allowed, so the number of people (even if he knew) is not representative of the voting trend, since there may be more over-zealous people on one party than the other which in turn makes asking a random group of people WHO they voted for inaccurate.

.

But, this hurdle can be exploited... now, since multiple votes are allowed, we can send in bugs. Since voting costs next to nothing and has no limitations, other than time, a statistical analysis method could be used. By sending in a large but equal number of bugs and votes for both parties, and than calculating the number of times we encounter a busy secondary line and compare it against number of attempts, we will be able to estimate the number of votes being cast per unit time. How is this done?

.

Say we have 10 voters for each party designated as bugs, and they will vote as many times humanly possible in the voting time window. Say also, that there are 10 boxes available per party station, but an unknown number of voters. The average ratio of busy:available lines at that section of time should be about 1:1 if there are 20 voters including our bugs. Conversely, if we achieve this ratio when other voters are present, we can safely assume that there about twice the number of voters compared to the boxes. However, in reality, the number of voters is far higher than the number of bugs. and there is really a lot more boxes to vote into... we have to be content with a small sample set of these boxes per test...

.

...so, in real life, we assume that the boxes we have tested are representative of the larger box population, and thus the extrapolating the number of voters we predict in our box population to the larger population, to predict the number and hence percentage of voters for one party as compared to the other party, to a decent margin of error.

.

The programme works this way. Some people will download the programme into their computers, and the programme will instruct the modem to dial the phone number to vote repeatedly. The modem will then report the number of attempts, busy lines and available lines to the database. The more busy lines there are, by commonsense, will imply that there are more voters. Given that enough people run the programme simultaneously, the programme will have sufficient data to construct the voting trend for any particular period of time. From these numbers, we will be given a ballpark estimate of the number of people voting for that party at that period of time, and from comparing the activity of the lines for both parties, we will know who will be the next American Idol...

.PS: Based on past weeks' trends, David C. is more likely to win...

## 0 Comments:

Post a Comment