MLE model

[ALKEBSEE RADWAN]

Hello Everyone,

I actually am trying to perform the ML estimates model to explore the distribution of my data.
Frankly I do not know how to apply it. Also, I don't know the code for this.

Here is my main model

xtreg PEG GP Size TobinQ Lev RoA ListAge restatement disacc csr SOE Top1 INST Board Indep Dual i.year, fe

PEG is the dependent variable, GP is the independent variable.

Kindly, help me out to get the codes for MLE model.
Thanks in advance

[Maarten Buis]

For a linear fixed effects model, you typically don't want the maximum likelihood estimates. My suspicion is that you don't want that either and instead got confused by the terminology in this field. That is perfectly normal, as the terminology in this field is very confusing, because different disciplines used the same terms for different things: Sometimes multilevel (sometimes abbreviated with ML, just to be extra confusing) is used synonymously for random effects. In some fields the term fixed effects is used for the not random coefficients of the explanatory variables in a random effects model. Which makes sense: it is not random, so it is fixed. However, it drives people who are used to fixed effects models crazy if you refer to the coefficients of a random effects model as fixed effects. In short, as soon as you move outside your own discipline the terminology is a complete mess.

So, given that we can no longer rely on terminology to accurately convey what you want, can you describe what you want?

[ALKEBSEE RADWAN]

For a linear fixed effects model, you typically don't want the maximum likelihood estimates. My suspicion is that you don't want that either and instead got confused by the terminology in this field. That is perfectly normal, as the terminology in this field is very confusing, because different disciplines used the same terms for different things: Sometimes multilevel (sometimes abbreviated with ML, just to be extra confusing) is used synonymously for random effects. In some fields the term fixed effects is used for the not random coefficients of the explanatory variables in a random effects model. Which makes sense: it is not random, so it is fixed. However, it drives people who are used to fixed effects models crazy if you refer to the coefficients of a random effects model as fixed effects. In short, as soon as you move outside your own discipline the terminology is a complete mess.

So, given that we can no longer rely on terminology to accurately convey what you want, can you describe what you want?

Thank you very much Maarten Buis,
Acually I have performed FE model (panel data) the reviewer said you have to justify why you used FE model by performing Huasman test and ML test. He also, mentioned a spicific paper to be followed.
See below


I need to apply this type of test to my model above

I know you may not understand me but if you need further information please let me know
thanks again

[Maarten Buis]

For a Hausman test you can just use the command named hausman. For a situation like yours you would typically compare a random and fixed effects model.

I don't understand what ML test you want to perform. ML test is very general, it could mean pretty much anything. The table you showed says it is a set of ADF tests, which I interpret as a set of Augmented Dickey-Fuller test. You can do those with dfuller

[ALKEBSEE RADWAN]

For a Hausman test you can just use the command named hausman. For a situation like yours you would typically compare a random and fixed effects model.

I don't understand what ML test you want to perform. ML test is very general, it could mean pretty much anything. The table you showed says it is a set of ADF tests, which I interpret as a set of Augmented Dickey-Fuller test. You can do those with dfuller

Thank you very much Dr. Maarten
I actually found a code for ADF test

xtunitroot fisher  Size , dfuller drift lags(2) demean

I just don't understand which value I should report .
Also, in the pic above the author reports results with 0 individual intercept and individual intercept. Could you please tell me the following
which value I have to report?
How to report the two types of results (with 0 and individual intercept and individual intercep)?

thanks again
Best,

[Maarten Buis]

I don't know, maybe someone else on this list?

[ALKEBSEE RADWAN]

Thank you very much
I hope someone will help

[Jeff Wooldridge]

Because MLE is an estimation method and not a "model," I agree with Maarten that there's probably confusion between "multi-level" and "maximum likelihood." If the referee said to use an "ML" model it was almost certainly multi-level. Plus, that's implied in the table. It looks to me like the table is effectively comparing pooled OLS -- where the intercept is the same across units -- and random effects, where the intercept varies with units. Maybe MLE under normality was used to estimate the two models; that's impossible to know. But here's how I would do it:

reg y x1 ... xK i.year vce(cluster id)
xtreg y x1 ... xK i.year, re vce(cluster id)

Because the data are macro data, it's crying out for a fixed effects estimation: replace "re" with "fe." But I still need more information to make more suggestions.

[ALKEBSEE RADWAN]

Because MLE is an estimation method and not a "model," I agree with Maarten that there's probably confusion between "multi-level" and "maximum likelihood." If the referee said to use an "ML" model it was almost certainly multi-level. Plus, that's implied in the table. It looks to me like the table is effectively comparing pooled OLS -- where the intercept is the same across units -- and random effects, where the intercept varies with units. Maybe MLE under normality was used to estimate the two models; that's impossible to know. But here's how I would do it:

reg y x1 ... xK i.year vce(cluster id)
xtreg y x1 ... xK i.year, re vce(cluster id)

Because the data are macro data, it's crying out for a fixed effects estimation: replace "re" with "fe." But I still need more information to make more suggestions.

Thank you very much prof. Jeff
I will share with you the exact words the referee said

According to Asteriou and Hall (2007), there are three different methods for estimating Simple linear panel data models: The common constant method, the Fixed effects method, and the Random effects method. To decide which type of method is suitable for our data set, we go with two tests: Hausman Test and the redundant fixed effect LM test. Testing to pool, fixed or random, are not shown that the test has been suggested that the date is statistically proven as not to be pooled fixed and random It is recommended that the authors run the Hausman Test and redundant fixed effect LM test to decide whether to go with the random effect method fixed effect method or Common constant method of the panel regression model and display results after unit root test results.

Kindly read and give your suggestions
Best,