I haven't tried it yet, but with a binary predictor it should work. For more of two categories you need dummy coding and I don't know whether lmrob builds those automatically or whether you have to do this before running the robust regression.
there is no pvalue for the significance of the over all model. Isn't that necesary in robust regression? how can you tell that the model is significant?
I think in most situations the overall significance of the model isn't really that interesting. Yes, we report it in a multiple regression, but often only in the notes to the regression table. It is quite rarely the case that one has a hypothesis about the overall model; most hypotheses are based on specific regression weights only. So for me this limitation has not been relevant in my analyses, yet.
@@RegorzStatistik I agree, however journals frecuently demand it. I founded later that some people use an ANOVA WALD test comparing the model with an only intercept y~1 vs y~1+1m+²m+3m. If anyone is interested.
I haven't used it for that, yet. But you do get regression weights as a result of the robust regression. I think you could use those for prediction purposes.
Thank you very much. Such a clear and concise video
Thank you for this lesson! Very helpful!
This helped me a lot! Thank you!
Thanks for your help, sir.
Nice! Do have a video about robust regression done with Huber-White-Hinkley (HC1) ?
Unfortunately, no.
Thanks
Thank you very much, it was realy interesting!
Thank you very much! Can I use lmrob if my predictor is categorical?
I haven't tried it yet, but with a binary predictor it should work. For more of two categories you need dummy coding and I don't know whether lmrob builds those automatically or whether you have to do this before running the robust regression.
The adjusted R squared in robust regression is negative and less than the OLS, does that mean it’s not a good fit?
I would only interpret adjusted R squared when comparing two or more models with different numers of predictors.
there is no pvalue for the significance of the over all model. Isn't that necesary in robust regression? how can you tell that the model is significant?
I think in most situations the overall significance of the model isn't really that interesting. Yes, we report it in a multiple regression, but often only in the notes to the regression table. It is quite rarely the case that one has a hypothesis about the overall model; most hypotheses are based on specific regression weights only. So for me this limitation has not been relevant in my analyses, yet.
@@RegorzStatistik I agree, however journals frecuently demand it. I founded later that some people use an ANOVA WALD test comparing the model with an only intercept y~1 vs y~1+1m+²m+3m. If anyone is interested.
how to make a prediction model from the robust regression results
I haven't used it for that, yet. But you do get regression weights as a result of the robust regression. I think you could use those for prediction purposes.
if the intercept is significant what does it imply? should the intercept be significant of insignificant?
The same as in the OLS regression: That the intercept is different from zero (which is in 99% of the cases not very interesting).