GLM extends regression by adding a link function and distribution of the dependent variable conditional on the predicted mean. The chosen distribution should match the population distribution. If not, is not chosen correctly. I hope this help.
@@mronkko So if i have an outcome which is a proportion, a common idea which strikes is the application of beta straight up ! Do you suggest that i should check this somehow before fitting, if yes then how. And also, how do you suggest that whether the distribution which we are assuming as beta actually fits it or if we have outcome as counts then how should we check before fitting a distribution using Chi-Square?
@@amanrastogi5184 I would go for Logistic QML instead of beta, unless you can justify beta distribution based on theoretical grounds. Logistic QML is more generally consistent. Beta can give you an advantage in efficiency, but I am not sure how large that advantage would be so I would not bother with considering beta.
![Lil Uzi Vert - Chill Bae [Official Music Video]](http://i.ytimg.com/vi/D7F42F_JM3U/mqdefault.jpg)
Thank you Mikko, this helped a lot !
Glad it helped!
"We need the correct specification of the conditional mean" can you explain this line to me. it will be helpful.
It means that you have all the relevant variables in the model (i.e. no endogeneity) and the functional form is correctly specified.
Sir, What do you mean by correctly specifying the distribution of outcome to apply beta regression? Thank You
GLM extends regression by adding a link function and distribution of the dependent variable conditional on the predicted mean. The chosen distribution should match the population distribution. If not, is not chosen correctly. I hope this help.
@@mronkko So if i have an outcome which is a proportion, a common idea which strikes is the application of beta straight up ! Do you suggest that i should check this somehow before fitting, if yes then how. And also, how do you suggest that whether the distribution which we are assuming as beta actually fits it or if we have outcome as counts then how should we check before fitting a distribution using Chi-Square?
@@amanrastogi5184 I would go for Logistic QML instead of beta, unless you can justify beta distribution based on theoretical grounds. Logistic QML is more generally consistent. Beta can give you an advantage in efficiency, but I am not sure how large that advantage would be so I would not bother with considering beta.
Sir, can i ask you what is difference from quasi poisson regression and quasi likelihood regression?
🙏
Nothing. Or alternatively we can think that qml refers to also other regression type models like quasi logit and not specifically to Poisson QML.