Thank you, you explain robust SE excellently. You are a master teacher, thank you. Your application is questionable. Many authors, such as David Freedman defend a slightly misspecified models as still interpretable if the evidence of an effect is still clear (effect estimate is larger than the robust SE).. It is a matter of context. Whether to go for formal perfection (model SE=robust SE) or not is often just a matter of taste. You are a perfectionist. That is a complement, by the way. The larger problem with your application is the use of the wacky Box-Cox transformation. The Box-Cox is widely ridiculed for producing totally uninterpretable results. (The formula is a block box and what does a Box-Cox transformed outcome mean, anyway?). A better solution perhaps would have been a generalized linear model, using a Gamma distribute outcome, for instance, to take account for the non-normalty of the outcome. .So you chose formal perfection (insist on equality of standard errors) but then implement it in a wacky way (Box Cox rather than GLM). Your approach seems somewhat strange to this PhD statistician.... I'd be grateful to hear a response.... Thanks again! P.S> There are even more modern machine learning approaches like Random Forest that could be used on your application...
I am attempting to identify the appropriate code for calculating clustered standard errors following the execution of a regression using the multinom() function in r. I attempted the following code but consistently encountered an error: Calculate the cluster-robust variance-covariance matrix vcov_clustered
Great video(s), thank you so much! Is it possible that the "systematic" and "stochastic" components are mixed up in slide 11? Shouldn't it be the reverse?
I benefited quite a bit from this lecture. Thank you, professor King.
Very intuitive explanation,
Many thanks
Great class!
love your articulation !
Thank you, you explain robust SE excellently. You are a master teacher, thank you.
Your application is questionable. Many authors, such as David Freedman defend a slightly misspecified models as still interpretable if the evidence of an effect is still clear (effect estimate is larger than the robust SE).. It is a matter of context. Whether to go for formal perfection (model SE=robust SE) or not is often just a matter of taste. You are a perfectionist. That is a complement, by the way.
The larger problem with your application is the use of the wacky Box-Cox transformation. The Box-Cox is widely ridiculed for producing totally uninterpretable results. (The formula is a block box and what does a Box-Cox transformed outcome mean, anyway?). A better solution perhaps would have been a generalized linear model, using a Gamma distribute outcome, for instance, to take account for the non-normalty of the outcome.
.So you chose formal perfection (insist on equality of standard errors) but then implement it in a wacky way (Box Cox rather than GLM). Your approach seems somewhat strange to this PhD statistician....
I'd be grateful to hear a response....
Thanks again!
P.S> There are even more modern machine learning approaches like Random Forest that could be used on your application...
Warm greetings and thanks so much! 🌻🌻🌻 In two weeks final exam in statistics (Master / Psychology)... Now there is hope! Greetings from Germany 🙃🌲
Extremely helpful lecture on RSE.
Thanks for such a good lecture about the RSE.
I am attempting to identify the appropriate code for calculating clustered standard errors following the execution of a regression using the multinom() function in r.
I attempted the following code but consistently encountered an error:
Calculate the cluster-robust variance-covariance matrix
vcov_clustered
Amazing video: thank you very much!
Great video(s), thank you so much! Is it possible that the "systematic" and "stochastic" components are mixed up in slide 11? Shouldn't it be the reverse?
Yes, good point, thanks I'll fix in the next version. The math is right as is but the names need to be swapped.