What’s the easiest way to remember the following terms? heteroskedasticity autocorrelation multicollinearity
Just remember 2 of the entities of Linear Regression: 1. Variance of Residual Error (A) 2. Independent Variables (B) The 3 violations (heteroskedasticy, serial correlation and multicollinearity) are violations involving combinations of these 2 entities. And what could be the possible combinations? Easy! AB, AA and BB, right? 1. AB - When Variances of error term are correlated with Independent Variables, it is Heteroskedasticity. 2. AA - When Variances of Error terms are related to each other, it is serial correlation. 3. BB - And when Independent variables are related to each other, it is Multicollinearity.
From an old post: heterodastictiy - detect with bush pagan, correct with white h bp w = high blood pressure in white men autocorrelation - detect with durbin watson correct with hansen ac dw h = air conditioner, dishwasher heater (all appliances) Multicollineality - detect f test significant, t test insignificant - correct with drop X variable mc fs ti drop x = mc hammer’s fist dropping X (that’s a bit strained but the best i could come up with.
heres what i do on the hetero one. I’m WHITE and HETEROsexual and there is a cute asian girl named CHI at the office that i’d like to hook up with HETEROskedasticiy WHITE - standard errors CHIsquare test
i love this forum, too funny guys
I always remember Heteroskedasticity as “hetero white boys please” Hetero = Heterokedasticity White = White corrects Boys Please = breusch pagan test However, this doesn’t include the definition like rus1bus.
I would refrain from using the: autocorrelation - detect with durbin watson correct with hansen ac dw h = air conditioner, dishwasher heater (all appliances) Since I believe the DW test is invalid for autocorrelations. Can anyone confirm this? - I don’t have my book. I seem to remember: The DW test is for Serial Correlation not auto correlation - when b1 is a lagged value of of the dependent variable a t-test of the autocorrelations of the error term is needed. There’s an EOC question on this in the time series reading - the only reason I remember this is b/c it really pissed me off when I got it wrong. That’s when I went back and memorized this - somehow it has stuck untill now. Hopefully I still have this right - does anybody know for sure?
agree with this - when using an AR model, believe you cannot use the DW test, must do individual t tests
Nice. Really nice. Really really nice.
Bump for you guys/ladies. Good stuff here.