Yuku free message boards

Wow, lots of great things to discuss.  This is a good start to analyzing how we use Edgic.

Yes, OTT + CP should usually equal CP in my estimation. 

CTTAndrewF: I see what you're saying, and I get it, and I've said it before, but I disagree.  It shouldn't just be a formula - but both quantitative and qualitative approaches to Edgic are helpful.  Triangulation!  It would be foolish to only look at the "numbers" so to speak - but also foolish to ignore the data when we already have and continue to collect it.  I fully, fully encourage a textual analysis of the episodes as well, however, and recognize that the variables that we do collect and rate are limited and flawed - but we have them.  Statistically, some do correlate well with winning (as I've learned), and that's information to put in our evidence bucket.

zeev14: Yay!  I so agree on the Chase/Holly/Jane P (obviously), and also, I am enthusiastic about your SPSS stat Edgic boner. 

Also to redcoffee and astroline and whomever else is interested: I've finally compiled most Edgic ratings into an SPSS/Excel file.  I haven't put in the first few seasons because of what redcoffee mentioned about the Edgic ratings being different back then.  Also, sometimes I left out the first 6 players or so to be eliminated from the game.  But it's close to completion.  I'm glad to make what I have available to anyone who wants to flesh out the rest of the chart or work with what we have so far.  As a result, Phoenix1269 and Hippochinfat - I'm far enough along that those files will no longer be necessary! - thank, though.

Great post, redcoffee - well thought out. 

The way I've been using factor analysis is to throw all the variables into a stats blender, so to speak, pull out the components as separate variables, and then run regressions with the component variables as predictors for the dependent binary variable of "Winner".  Maybe not the best approach, but an approach to try out.  I've been getting a few variables that works well with male winners - but becomes statistically insignificant with female winners.  My variables need some fine-tuning, tho.  Currently, it likes Chase and Brenda most this season.  But this could easily overlook a female winner.

The variables I have been using are:

Rating-Episode[x]: ordinal (tho I know it's not really ordinal - I was just reluctant to do categorical)
Visibility-Episode[x]: scale
Tone-Episode[x]: scale (-2 for NN to 2 for PP, with M as 0 because what else to do?)

Then I just started throwing in some random shit:

Tonal Presence-Episode[x]: scale (with the absolute value of the previous variable, with M as 1)
Total number of OTT(/CP/MOR/UTR/INV) up thru Episode 6: scale
Total visibility thru Episode 6
Total tonal presence thru Episode 6
Variance in Visibility thru Ep. 6
Variance in Ratings thru Ep. 6
Variance in Tonal Presence thru Ep. 6
Total tonal Neutrality*Total MOR thru ep 6

...and so on, until I ran out of variables to make up.

You're exactly right about using cumulative variables, tho, redcoffee - I was just thinking I should've done that myself. 

Also I was considering having a variable for positive tone and a variable for negative tone, and M just being a point in both columns. 

The missing data is an issue - but I'm just letting SPSS figure that out thus far and not even thinking about it.  Also I know some variables were conceived of differently then and now, but I'm also just shrugging that off at this point.

I think it's a great idea to have multiple variables for the dependent variable.  One just being "did they win" but the idea to have one reflecting types of wins is a good one.  I also have a variable just for whether someone was in the finals, but it isn't working out so far for me yet.  But it's there.

Either way, you've got a lot of great ideas, and I'm sure some of the rest of you do too.  Just let me know how I can best share the data I have compiled and you can rearrange it and add new variables however you like.  My head has gotten a little spinny from my SPSS-in' so far.