kallend 2,182 #1 February 12, 2007 www.washingtonpost.com/wp-dyn/content/article/2007/02/12/AR2007021200626.html... The only sure way to survive a canopy collision is not to have one. Quote Share this post Link to post Share on other sites
StreetScooby 5 #2 February 12, 2007 I'd love to read about the statistics used in this paper. How do you "refine" statistical results such that a reasonably orthogonal collection of independent variables are determined?We are all engines of karma Quote Share this post Link to post Share on other sites
SpeedRacer 1 #3 February 12, 2007 You use duct tape. Speed Racer -------------------------------------------------- Quote Share this post Link to post Share on other sites
rehmwa 2 #4 February 12, 2007 QuoteHow do you "refine" statistical results such that a reasonably orthogonal collection of independent variables are determined? Guidleline: When Duct tape is not available, then I'd ensure that VIF < 5 when possible A metric, called the Variance Inflation Factor (VIF), calculates the degree of multicollinearity. VIF = 1/(1-Ri^2) Ri^2 is the R^2 value obtained when Xi is regressed against the other X's in the (proposed) model A large VIF implies that at least on variable is redundant VIF > 10 (Ri^2 > 0.9): high degree of multicollinearity - cause for serious concern VIF > 5 (Ri^2 > 0.8): moderate degree of multicollinearity That should give a nap to some people. ... Driving is a one dimensional activity - a monkey can do it - being proud of your driving abilities is like being proud of being able to put on pants Quote Share this post Link to post Share on other sites
StreetScooby 5 #5 February 12, 2007 Quote A metric, called the Variance Inflation Factor (VIF), calculates the degree of multicollinearity. VIF = 1/(1-Ri^2) Ri^2 is the R^2 value obtained when Xi is regressed against the other X's in the (proposed) model A large VIF implies that at least on variable is redundant VIF > 10 (Ri^2 > 0.9): high degree of multicollinearity - cause for serious concern VIF > 5 (Ri^2 > 0.8): moderate degree of multicollinearity Noice! I've always wandered about that. Well put, and thank you!We are all engines of karma Quote Share this post Link to post Share on other sites
kallend 2,182 #6 February 12, 2007 QuoteQuoteHow do you "refine" statistical results such that a reasonably orthogonal collection of independent variables are determined? Guidleline: When Duct tape is not available, then I'd ensure that VIF < 5 when possible A metric, called the Variance Inflation Factor (VIF), calculates the degree of multicollinearity. VIF = 1/(1-Ri^2) Ri^2 is the R^2 value obtained when Xi is regressed against the other X's in the (proposed) model A large VIF implies that at least on variable is redundant VIF > 10 (Ri^2 > 0.9): high degree of multicollinearity - cause for serious concern VIF > 5 (Ri^2 > 0.8): moderate degree of multicollinearity That should give a nap to some people. I'm always impressed by the breadth of knowledge available on this forum.... The only sure way to survive a canopy collision is not to have one. Quote Share this post Link to post Share on other sites
rehmwa 2 #7 February 12, 2007 QuoteI'm always impressed by the breadth of knowledge available on this forum. 1 - leave it up to statisticians to invert a simple correlation activity of each of the input parameters and call it another name. Just to make it seem harder than it really is. I really hate statistics for that. There are a bunch of really simple things they do that they then add a bunch of unnecessary mathematical manipulation........ job security I guess. 2 - I think duct tape is really, really, neat. ... Driving is a one dimensional activity - a monkey can do it - being proud of your driving abilities is like being proud of being able to put on pants Quote Share this post Link to post Share on other sites
crwtom 0 #8 February 12, 2007 Quote VIF > 5 (Ri^2 > 0.8): moderate degree of multicollinearity even after you're through with any multicubical and what-not regressions analysis the old adage still stands Correlation is Not Causation A correlation between napping and longlivety can be generated by a common cause (which can be anything from career/job choice to a physiological disposition that makes you more likely to nap) without there being any direct benefit of napping for your health or life expectancy. Cheers, T ******************************************************************* Fear causes hesitation, and hesitation will cause your worst fears to come true Quote Share this post Link to post Share on other sites
willard 0 #9 February 12, 2007 QuoteQuoteI'm always impressed by the breadth of knowledge available on this forum. 1 - leave it up to statisticians to invert a simple correlation activity of each of the input parameters and call it another name. Just to make it seem harder than it really is. I really hate statistics for that. There are a bunch of really simple things they do that they then add a bunch of unnecessary mathematical manipulation........ job security I guess. 2 - I think duct tape is really, really, neat. I always have duct tape nearby when taking a nap. You never know when somebody is going to disturd that nap and needs duct taped to a folding chair. Quote Share this post Link to post Share on other sites
rehmwa 2 #10 February 13, 2007 Quote even after you're through with any multicubical and what-not regressions analysis the old adage still stands Correlation is Not Causation A correlation between napping and longlivety can be generated by a common cause causation needs to be validated through experiment if not apparent (via previous experiment) - regression and correlation are different tools and the coefficients are determined with different assumptions. Strangely enough, most popular stats s/w packages out there don't have a separate correlation function, only regression "longlivety" is an interesting word ... Driving is a one dimensional activity - a monkey can do it - being proud of your driving abilities is like being proud of being able to put on pants Quote Share this post Link to post Share on other sites
