Looking for quick help with statistical terminology (1 Viewer)

[tombonneau]

So I know there is a name for it, just can't recall what it is, and I know I read it here so I'm sure one of the FBGs or forum members can help me out .......

What is the term for when you choose an arbitrary cut-off point to help support your statistical argument? Like if you say since Nov 2007 the team is 5-0 when favored by 3+ points but if you go back to September they would show as being 5-4.

Or a popular thing in fantasy circles is looking at backs who had 370 carries or whatever the previous year just because that number supports a trend in decline but if you move it to 355 it looks different.

What's that called?

Thanks!

 

[KCC]

truncating the range

 

[tombonneau]

truncating the range

No, it wasn't that. That is the act of what you're doing, yeah, but there was a specific term for trying to use that type of technique to support your point. Kind of like the concept of a straw man argument ...

 

[RamMan]

It's called "Moving It And Then it Looks Different" (or MIATILD to us insiders)

 

[Chase Stuart]

Multiple endpoints?

 

[Bob Henry]

A statistical pat-down. Of course you can opt-out.

 

[belljr]

sampling bias

 

[OPS]

selection bias

 

[twheeler]

selection bias

We have a winner.http://en.wikipedia.org/wiki/Selection_bias

 

[belljr]

selection bias

We have a winner.http://en.wikipedia.org/wiki/Selection_bias

I disagreehttp://en.wikipedia.org/wiki/Sampling_bias

 

[tombonneau]

selection bias

We have a winner.http://en.wikipedia.org/wiki/Selection_bias

Yes, I think this is exactly the phrase I was thinking of. Thanks.

 

[Judo czop!]

The Chewbacca Defense

 

[CalBear]

I think "multiple endpoints" is the correct answer. Selection bias is when your sample is biased due to how it is selected. For example, if you are running a survey about beer consumption, and you select to poll football fans about it, your survey will show more beer drinking than really exists in the general population, because the population you're selecting from is biased towards beer. (I know I am). The trick where you choose "370 carries" as the mythical cutoff point for production, because that's the number that supports your hypothesis, is properly termed "multiple endpoints"--there's no particular reason to choose 370 carries other than that the resulting sample gives you the answer you wanted to see.

Edit to add: I suppose you could say that multiple endpoints is a specific type of selection bias.

 

[radballs]

Data mining and manipulation?

 

[tombonneau]

I think "multiple endpoints" is the correct answer. Selection bias is when your sample is biased due to how it is selected. For example, if you are running a survey about beer consumption, and you select to poll football fans about it, your survey will show more beer drinking than really exists in the general population, because the population you're selecting from is biased towards beer. (I know I am). The trick where you choose "370 carries" as the mythical cutoff point for production, because that's the number that supports your hypothesis, is properly termed "multiple endpoints"--there's no particular reason to choose 370 carries other than that the resulting sample gives you the answer you wanted to see.Edit to add: I suppose you could say that multiple endpoints is a specific type of selection bias.

"Multiple endpoint selection bias." I like it. It has a certain ring of authority to it.

 

[Ignoratio Elenchi]

Data snooping.

 

[Judo czop!]

Data snooping.

I mention Chewbacca Defense then Ignoratio Elenchi appears. Your ears must have been ringing

 

[Ignoratio Elenchi]

Data snooping.

I mention Chewbacca Defense then Ignoratio Elenchi appears. Your ears must have been ringing :X

 

[DiStefano]

I don't know what it's called, but the proper response is: Bull####!