Do we need to know the formula for “R squared adj”? And do we have to do a BP test? I know we need to know what they mean, but last year we needed to know the calculations. I’m trying to throw out as many note cards from last year as possible!
I don’t have my book with me for quant, but check the LOS. It should tell you whether you need to calculate or just discuss/explain.
know all the formulas. they were tested.
- Adjsuted Rsq – looks like the LOS says “distinguish between… and interpret … Rsq and adjusted Rsq” - doesn’t actually say “calculate” . But the flormula is not as complex as it looks - it looks messy and I would never remember it if I had to actually write it out on a page. But it’s not that bad — let’s say: n = 60 Rsq = .63 k = say 5 so the df numbers are: n-1 = 59 (= the total df on the ANOVA table) then n-(k+1) = 54 (= the df for Error) so you write down the df numbers: 59 and 54. So how do you adjust Rsq of 0.63 to take account of the different df for k=5? since Rsq is 0.63 we know that the adjusted Rsq is going to be LESS than the simple Rsq if k = more than 1. Just take 1 - Rsq and get 0.37. Then multply this by 59/54 (ie the ratio of two df numbers above) = 40.4. Take this away from 1 = 0.596 = Adjusted Rsq. that’s it. you can do it without remembering the formula - because you’re just adjusting 1-Rsq for the ratio of the df’s, then taking it away from 1 again. sanity check - if you accidentally do it the wrong way (ie multiply by 54/59 instead of 59/54) by mistake, you know it’s wrong because it comes out more than 0.63. - so just flip it around. the whole thing takes about 10 seconds, including the sanity check (ps i’ve never done math in my life beyond high school - that’s just my little way of simplfying it for my simple non-math brain) 2. BP test - looks like the LOS doesn’t actually mention Breusch-Pagan by name, but talks about knowing the types of hetero, effects of hetero, impact, etc. So I guess like all the hypothesis test stuff: - know what test to use to test what - know what the rule is (reject, don’t reject and what that tells you) - how to calc the test stat = n * Rsq. but here it’s the Rsq of the 2nd regression of the sq errors - can’t imagine you having to actually calculate this - probably just given to in a question. - know what to do with it - again probably just given the chi value and asked what it means. (it if test stat > critical chi val --> hetero is significant) - know how to correct (white) - maybe even a red herring like 1-tailed or 2-tailed test? hope this helps…
.