Appreciate the great explanation! I have a follow-up question: why does REML not give a proper "likelihood" that we are not allowed to do likelihood ratio comparison with? Is it due to the definition of likelihood ratio tests where we have to use the traditional likelihood values without correction on degree of freedoms?
The theory of LRT applies to ML estimates. REML are not ML, so there is no theory (as far as I know) supporting model comparisons. I am not sure what SAS does and whether LRT tables are based on ML or REML statistic when REML estimates are requested. If in doubt, do estimates with REML and if model comparisons are needed, re-estimate the model with ML. This is what I do with Stata. (I think that Stata does not allow comparing REML estimated models using LRT, but I might remember wrong.) Because retuning the estimates is simple to do, I have not bothered searching the literature for any studies about comparing REML estimated models with nested model tests.
Great sir,....but can you make a short example of data and show much better ways slowly and step and step next time so that people from non mathematics background also understand it nicely..but it is a good video..
Ajay, I am not sure what the purpose of such video would be. I mean, if you want to know why REML estimation is used I do not think that you can simplify much more than what I do here. If you are interested in knowing how to apply them, it is just a feature that you turn on in your statistical software.
@@mronkko Hi Mikko by simple video i mean can you show how from a sample data set variance is estimated using REML method...not just analysing the last result.
@@ajaydhungana1921 I am not sure what you mean. Do you want to see 1) what do I click to turn REML estimation on in a statistical software or 2) or do you want to seen the equations that are used and an explanation of how they differ from ML equations?
I do not understand the question. As far as I know, the error variance is estimated by the variance of the residuals corrected for degrees of freedom. I am not aware of any other estimation techniques for OLS regression.
Thank you so much!! Great explanation and incredibly helpful!
Glad you enjoyed it!
Amazing video!!!
Glad you liked it!!
Spot on video. Very helpful to me in my understanding REML rationale in my longitudinal data analysis graduate course in biostatistics.
Well good that it helped.
Thank you very much, very helpful!
You're welcome!
Great video
You are welcome!
Nice, thx!
You are welcome
Thanks on your great explanation 👍
You are welcome
Appreciate the great explanation! I have a follow-up question: why does REML not give a proper "likelihood" that we are not allowed to do likelihood ratio comparison with? Is it due to the definition of likelihood ratio tests where we have to use the traditional likelihood values without correction on degree of freedoms?
Yes, that is the readon.
Wondering why REML can't be compared with LRT? Does it mean the LRT table from proc mixed of SAS is meaningless?
The theory of LRT applies to ML estimates. REML are not ML, so there is no theory (as far as I know) supporting model comparisons. I am not sure what SAS does and whether LRT tables are based on ML or REML statistic when REML estimates are requested. If in doubt, do estimates with REML and if model comparisons are needed, re-estimate the model with ML. This is what I do with Stata. (I think that Stata does not allow comparing REML estimated models using LRT, but I might remember wrong.) Because retuning the estimates is simple to do, I have not bothered searching the literature for any studies about comparing REML estimated models with nested model tests.
Great sir,....but can you make a short example of data and show much better ways slowly and step and step next time so that people from non mathematics background also understand it nicely..but it is a good video..
Ajay, I am not sure what the purpose of such video would be. I mean, if you want to know why REML estimation is used I do not think that you can simplify much more than what I do here. If you are interested in knowing how to apply them, it is just a feature that you turn on in your statistical software.
@@mronkko Hi Mikko by simple video i mean can you show how from a sample data set variance is estimated using REML method...not just analysing the last result.
@@ajaydhungana1921 I am not sure what you mean. Do you want to see 1) what do I click to turn REML estimation on in a statistical software or 2) or do you want to seen the equations that are used and an explanation of how they differ from ML equations?
Salam sir .
How different methods of estimation of error variance of simple linear regression model?
I do not understand the question. As far as I know, the error variance is estimated by the variance of the residuals corrected for degrees of freedom. I am not aware of any other estimation techniques for OLS regression.
Amazing video!
You are welcome