Regression To The Mean

[Bramblethorn]

It doesn't suggest to me that those ones with all the votes and the good scores are not genuinely popular stories.

I don't suggest they're not.

Being on lists makes them attract both extra bad-faith votes as well as extra good-faith votes. Being good (or at least popular) stories is what got them on a list in the first place.

I agree, but I'm not sure which part of my comment you're responding to here.

 

[SimonDoom]

One that did drop from July to August was Colleen Ch 3, which dropped by 0.01 (everything dropped by 0.01). It's got 13.7k ratings now; 13457 on July 10 and 13534 on August 10. My calculation was that it got 77 ratings in that period averaging 3.10; my numbers could be slightly off. That assumes a movement of a full 0.01 rather than 0.006 and rounding. After dropping to 4.85, it received another 190 ratings in the next four months (August 11-December 11) and remained at 4.85. That's just about 48 ratings each month, so a hair over half as many as it received in the July-August period. I'm doing this off memory; I closed the spreadsheet earlier.

How can you know, or even assume, this? You don't know that the story took a full .01 step down. It might have gone from 4.8601 to 4.8599 (I don't know how the site rounds things off). If it ticked just a hair down the "rounding up cutoff" then those 77 votes might have averaged 4.8, or something like that. Or am I missing something?

 

[YmaOHyd]

How can you know, or even assume, this? You don't know that the story took a full .01 step down. It might have gone from 4.8601 to 4.8599 (I don't know how the site rounds things off). If it ticked just a hair down the "rounding up cutoff" then those 77 votes might have averaged 4.8, or something like that. Or am I missing something?

I can't know that, and like everyone else I don't know how the site rounds. What I can say is that between July 10 and August 10, sixteen of nineteen stories in T/I had their displayed rating drop by at least 0.01 and zero of nineteen stories had their displayed rating increase. Even if that one only dropped by the smallest amount statistically possible to cause the displayed number to change, some of the other fifteen didn't -- and if they all received ratings sufficient to cause them to hit a displayed rating of 4.85, then the pace of ratings dropped off (as in the case with Colleen), that would be suggestive of scripted behavior on its own.

However, if the real score on that story was 4.860 on July 10, and the real score on August 10 was 4.854999, the 77 votes in that period had to have averaged 3.98 to cause that change, which is still way out of phase with the first 13500 ratings. And maybe the real scores are 4.8550 on July 10 and 4.8549 on August 10! But that would be a hell of a coincidence considering the hundreds of other stories that have dropped down to 4.85 in a similar period.

 

[Bramblethorn]

How can you know, or even assume, this? You don't know that the story took a full .01 step down. It might have gone from 4.8601 to 4.8599 (I don't know how the site rounds things off).

They round to the nearest .01, with scores that are exactly halfway rounded up.