interpretation of margins from probit model

[Enrico Azzini]

Hi, I obtained the following results usin margins on a probit model

Regression model to measure the impact on log_wages
-----------------------------
                         1   
-----------------------------
working                      
Other Migrants      -0.359***
                   (0.000)   

Refugees            -0.553*  
                   (0.015)   

ETAM                0.0633***
                   (0.000)   

ETAM X ETAM      -0.000450***
                   (0.000)   

Elementary edu~n   -0.0426   
                   (0.486)   

Middle school ~n     0.236**
                   (0.002)   

Diploma 2-3 ye~s     0.453***
                   (0.000)   

Diploma 4-5 ye~s     0.618***
                   (0.000)   

Degree               0.849***
                   (0.000)   

Female              -0.239***
                   (0.000)   

Married=1            0.286***
                   (0.000)   

Constant            -1.099***
                   (0.000)   
-----------------------------
_                            
*employee                    
                             

*fulltime                    
                             
-----------------------------
N                   256533   
adj. R-sq                    
Cluster errors         yes   
Country of ori~t       yes   
Region fixed e~t       yes   
Other controls         yes   
-----------------------------
p-values in parentheses
+ p<0.10, * p<0.05, ** p<0.01, *** p<0.001

If I'm correct I can say that individuals in Other migrants group have a probability to work which is lower by 0.359. However I'm not sure if this is equavalent to say that other migrant have 36% chance in less to be in work than baseline group. Can you please help me to clear my mind?
Thank you for your attention.

[Andrew Musau]

If your outcome is \(y_i = 1\) if individual \(i\) is employed and 0 if unemployed and these are average marginal effects obtained from

probit...
margins, dydx(*)

then the interpretation is that other migrants' probability of being unemployed is 36 percentage points higher compared to the reference group.

[Richard Williams]

What were the commands given? This doesn’t look like the kind of output you would get from margins. Are you sure this isn’t the probit output? Log wages would be a curious name for a dichotomous dependent variable.

[Mohieddine Rahmouni]

Is your dependent variable binary (working 0/1)?

[Enrico Azzini]

sorry log_wage is the title from another regression, I forgot to change the title. The dependent variable is working which is a dummy 0 for unemployed and 1 viceversa.

[Richard Williams]

Ok, but I would still like to see the actual commands. This looks more like the output you would get from probit than from margins. The constant term seems odd. And, if you have done the commands right, you shouldn’t be getting margins for the squared term. My guess is you are reporting the wrong results given your question.

[Mohieddine Rahmouni]

probit y x1 x1 ....

margins, dydx(*)

the output is without constant.

[Enrico Azzini]

You rigth, indeed I open another question to ask about how should I export resutls with esttab, this results are not in margins.
Sorry and thank you for the warning
here the correct outcome

-----------------------------
                         1   
-----------------------------
Other Migrants     -0.0380***
                   (0.000)   

Refugees           -0.0258   
                   (0.591)   

*employee                    
                             

*fulltime                    
                             

Elementary edu~n  -0.00868   
                   (0.673)   

Middle school ~n   -0.0953***
                   (0.000)   

Diploma 2-3 ye~s    -0.135***
                   (0.000)   

Diploma 4-5 ye~s    -0.154***
                   (0.000)   

Degree              -0.133***
                   (0.000)   

Female              0.0295***
                   (0.000)   

Married=1          -0.0258***
                   (0.000)   
-----------------------------
N                   170232   
adj. R-sq                    
Cluster errors               
Country of ori~t             
Region fixed e~t             
Other controls               
-----------------------------
p-values in parentheses
+ p<0.10, * p<0.05, ** p<0.01, *** p<0.001

[Richard Williams]

I'm not sure what you did, but with margins you can add the -post- option , which will post margins and their VCE as estimation results, The estout documentation says

As of Stata 11, the use of mfx is no longer suggested, since mfx has been superseded by margins. Results from margins can directly be tabulated by estout as long as the post option is specified with margins. Alternatively, you may add results from margins to an existing model using estadd margins or estpost margins. See http://repec.sowi.unibe.ch/stata/est...ients.html#002 for an example on tabulating results from margins.

I'd STILL like to see the actual commands, as well as the output from margins rather than the reformatted output from esttab. As it stands, I can't tell if factor variable notation has been used correctly. I also can't tell what options you specified for margins, e.g. did you use asobserved (the default), atmeans, or what.

Assuming it is all correct, on average Other Migrants are 3.8 percentage points less likely to be employed than whatever the reference category is for the variable.