bradley4ster: hey Jamie are you busy ?
cowzendux: Hey there. Just got out of the shower.
cowzendux: What can I do for you?
bradley4ster: had a small question about the stats reading
cowzendux: no problem.
bradley4ster: In your notes and also in the book you discuss the non-associative properties of regression... ab does not equal ba
bradley4ster: what are the consequences/benifits of this scenario
cowzendux: yes, that's true. 
cowzendux: in practical terms it isn't a huge deal
cowzendux: the regression lines are usually similar, although they are not actually the same.
cowzendux: After all, they are both trying to fit the relation between the same pair of variables.
bradley4ster: right.. and that is what seemed pecular
cowzendux: It is also the case that the significance level of the two equations is always the same too.
cowzendux: If one is significant then the other will also be significant.
bradley4ster: ahhh i did not know this 
bradley4ster: another quick question...
bradley4ster: what is the purpose of centering your variable (making the deviation scores)
cowzendux: You've probably read this already, but the source of the difference is that least squares regression specifically tries to minimize the difference between the line and the data, but specifically the difference in the vertical direction.
bradley4ster: yes.. it is making more sense now
cowzendux: because that is how we define our residuals (I'll get to your second question in a moment).
bradley4ster: k
cowzendux: so if you think about the fact that we're not talking about minimizing the shortest distance between each data point and the line, but specifically the *vertical* distance between each point and the line, I think it makes sense that you might get slightly different equations depending on whether X or Y is your DV.
bradley4ster: yes that makes sense.
cowzendux: Deviation scores are necessary for certain types of regression analyses, specifically those involving interaction or polynomial terms.
cowzendux: I'm not sure of the context in which they're discussed in chapter 2
cowzendux: so i should ask you about that before I go much further
bradley4ster: ahha!! ... the context is slipping myy mind as well.  However, I have done that for both Rosanna and Helen and had no rationale for what I was doing.
bradley4ster: but Rosanna only had me do it for interaction terms
bradley4ster: I wish I could remember from the chapter but it has been I while sense I wrote these thoughts down 
cowzendux: If you don't center your IVs then including an interaction term makes it so that the coefficients on the IVs don't actually test the main effects
cowzendux: i'll cover that in detail when we get to interaction terms. the logic is actually kinda cool.
cowzendux: someone had to be insightful to figure that out :)
bradley4ster: ahh. ok.. I will be counting the days?
bradley4ster: :)
cowzendux: yes. i'm sure you can hardly wait.
bradley4ster: And I am also not sure how imperative it is that we fully understand regression to the mean but it was a little unclear in the text
cowzendux: i don't think it's terribly important for this class.
cowzendux: it's a useful concept in terms of methodology, but not really statistics.
bradley4ster: i have used the concept in presentations and class and such ...however after reading about it i am now thinking that I did not know what i talking about
bradley4ster: k
bradley4ster: well I will not worry to much about that then ..
cowzendux: the idea is that if you choose a group of people that are specifically high on a variable
cowzendux: for example, taking a sample of people who score high on an iq test
cowzendux: chances are that the random influences for those people were more likely to have inflated their scores rather than depressed them.
cowzendux: simply because the more positive the net influence of random factors, the more likely they'll get a score high enough for your group
cowzendux: this is important because if you tested those same people again, those random influences would need to be "rerolled" if you know what i mean
cowzendux: by definition they are random, and so you would not expect them to stay the same.
bradley4ster: ahh and so then those random influences should not occur a second time thus reducing their score toward the mean? 
cowzendux: yes, exactly.
bradley4ster: i gottcha now
cowzendux: and that's regression toward the mean
bradley4ster: I guess I understood the concept lightly .. but now I feel like I can use it with a little more confidence
cowzendux: excellent. 
cowzendux: i hadn't recalled that they talked about that in the book.
bradley4ster: yeah .. in chapter 2 this is a whole section devoted to it 
bradley4ster: titled : Regression to the mean :)
cowzendux: well, that's certainly straightforward :)
bradley4ster: k... well I appreciate this ... it helps a lot
cowzendux: no problem. anything else I can help with?
bradley4ster: nope that covers it.
cowzendux: Ok. Hey, I just had an idea. Would you mind if I saved this chat log and then posted it on the website? That might be a fun thing to do - keep track of any chats I have and post them for people to read if they want.
bradley4ster: nope.  I would not mind at all.  
bradley4ster: I would actually love to read others 
bradley4ster: it may help elucidate some concepts