(08-25-2022 04:59 PM)bullet Wrote: (08-25-2022 02:40 PM)Yosef181 Wrote: (08-25-2022 02:28 PM)bullet Wrote: (08-25-2022 02:19 PM)Yosef181 Wrote: (08-25-2022 02:09 PM)bullet Wrote: Read the article. Look at the graphs. There's a clear dropoff for even Alabama and Ohio St. the more games you add.
Do you not understand how "Average" is calculated?
Do you not understand the concept of "lies, damn lies and statistics?" You clearly don't understand the article and may not have even read it. Taking one school's best 27 games and comparing it to someone else's 37 games gives you a meaningless result.
There's no "27 best games" against "someone else's 37". One school has 27 data records. Another has 37 data records. That's the complete dataset. It's beyond ridiculous to assume that if you added 10 App games, they would automatically be the 10 lowest App games. That's speculation, and has no basis in the reality of this dataset.
Data "lies" when bias and tampering is introduced. One example of that is to intentionally leave out 10 records you don't like. One way to lessen the effect of bias and tampering is to look at the most complete dataset possible.
The actual, real conclusion, based on the complete dataset: Appalachian State averages more viewers (0.587mm) than East Carolina (0.547mm).
Could that change if more App games were on TV? The only way to find out is to put more App games on TV.
Read the article. You might then understand. As pointed out above, if you use Vanderbilt's 2 data points in 2018, they average better than Miami (FL) does on their 11. All schools have declining curves on their ratings as shown in Miami's curve earlier in the thread.
Of course they do, the curves are showing the games in order of highest viewership to lowest. It's impossible to show anything BUT a decline. The assumption you're making is that the more games you add the lower the viewership becomes, but that's not correct. That would be the case if all the games were in the same season and you plotted them by order of when they happened and it showed a decline, but that's not at all what the data is doing. Dates are irrelevant and it's simply most to least. Ohio State could have a game tomorrow that's their highest rated game. More likely it'll fall somewhere in the middle of their curve though, but it's not a case where adding an additional game will make the curve decline more.
The issue here is that what you're doing is assuming that an additional 10 games for App State would be at the end of the curve as opposed to in the middle of the curve, or even at the beginning of the curve (meaning: any 10 additional games could be their highest viewed one, their lowest viewed one, or somewhere else in the middle).
This is statistical analysis, and therefore you need decent sample sizes to extrapolate what games outside of the given data set will look like. A sample of 2 is not big enough (same with a sample of 6 from my own UMass). A sample of 27 compared to a sample of 37 is much more reasonable. It's not perfect, yeah, but it does a pretty good job at showing a trend.
He is completely correct that you can't just take the lowest 10 and remove them from the East Carolina dataset for a comparison. What you would need to do is randomly remove 10 of them if you want the same number of games to compare against. It
should end up showing about what the average is showing now because of how probability works, but it could also make the average higher or lower (but it won't change much if it was truly randomly selected).