[Steffen Mauch]

Jared Greathouse Thank you for your input and approach. If I interpret my results correctly, I'm not in this range, though. Am I right?

Also, would you use the -predict- command to create such a plot?

[Clyde Schechter]

Re #14: Sorry, I spoke too quickly. I was thinking about linear probability and logistic models in non-panel data, where it is easier to do that. I don't know of a good way to test the fit of an -xtlogit, fe- model, and I don't know of even a bad way to do that which could be made comparable to a test of fit of a linear probability model in panel data.

With regard to the linear probability model, while R² statistics are aggregate measures of fit, I always like to look at scatter plots of predicted vs observed values. Now, with a dichotomous outcome, the observed values are always 0 or 1. So the next best thing is to group the observations into bins based on the predicted outcome. (What I'm recommending here is very much in the spirit of the Hosmer-Lemeshow test used with logistic or probit models in non-panel data.) Then plot the observed number of 1 responses vs the predicted number of 1 responses in each bin. The bins should be of roughly equal size and the number of them should be such that each bin contains around 50-100 observations: that smooths things out reasonably well in most circumstances. By inspecting a graph like that you not only get a visual impression of how well calibrated the model is, you also can see if there are systematic errors in the model. For example, the model might systematically overpredict in one part of the range and underpredict in others. That might suggest ways to improve the model by adding variables or interaction terms if the original fit seems unsatisfactory.

[xiao wen]

Dear Mr. Schechter @Clyde Schechter,

I hope you are well. I have a question regarding the use of -margins-

Notice that the -margins- results are different for the two regressions; yet these two models are not substantively different--they are just two different ways of breaking the colinearity between treat and idcode. You have the same problem. When you use -xtreg, fe- Stata has one particular strategy for dealing with the colinearity. But the -margins- results would be different if another one were used. This is why these margins are not estimable: they are just artifacts of the particular choices made to break colinearity. In fact, you have this problem twice: once for worker_id and treat, and again for post and year. Nevertheless, note that your key outcome, the DID estimate of the treatment effect, comes out the same in both models--it is the coefficient of the interaction term in the regression output. And that is what really counts here.

in my DID analysis with panel data, where the time period ranges from months 5 to 29. I am using a Linear Probability Model (LPM), and below is my code and result:

 xtreg $y treat time $x $control i.month_indicator, fe vce(cluster id_num)
note: treat omitted because of collinearity.
note: 29.month_indicator omitted because of collinearity.

Fixed-effects (within) regression               Number of obs     =    179,538
Group variable: id_num                          Number of groups  =     11,218

R-squared:                                      Obs per group:
     Within  = 0.3352                                         min =          1
     Between = 0.2767                                         avg =       16.0
     Overall = 0.3322                                         max =         25

                                                F(33,11217)       =     312.91
corr(u_i, Xb) = -0.1940                         Prob > F          =     0.0000

                               (Std. err. adjusted for 11,218 clusters in id_num)
---------------------------------------------------------------------------------
                |               Robust
Y             | Coefficient  std. err.      t    P>|t|     [95% conf. interval]
----------------+----------------------------------------------------------------
          treat |          0  (omitted)
           time |   .1563764   .0082816    18.88   0.000      .140143    .1726098
     treat_post |  -.0973506    .004796   -20.30   0.000    -.1067516   -.0879497
                |
month_indicator |
             6  |   .0088768   .0049505     1.79   0.073    -.0008271    .0185807
             7  |   .0217028   .0048793     4.45   0.000     .0121386    .0312671
             8  |   .0273711   .0050675     5.40   0.000      .017438    .0373042
             9  |   .0265076   .0052259     5.07   0.000     .0162639    .0367512
            10  |   .0450282   .0052348     8.60   0.000      .034767    .0552893
            11  |   .0410444   .0053232     7.71   0.000     .0306099    .0514788
            12  |   .0426608   .0054737     7.79   0.000     .0319314    .0533901
            13  |   .0563141   .0055621    10.12   0.000     .0454114    .0672169
            14  |   .0549608   .0056474     9.73   0.000     .0438909    .0660306
            15  |   .0598383    .005735    10.43   0.000     .0485966      .07108
            16  |   .0663735   .0058667    11.31   0.000     .0548737    .0778733
            17  |  -.0300051     .00465    -6.45   0.000    -.0391199   -.0208902
            18  |  -.0243395   .0044541    -5.46   0.000    -.0330703   -.0156087
            19  |  -.0151883   .0044271    -3.43   0.001    -.0238661   -.0065105
            20  |  -.0157295   .0043193    -3.64   0.000     -.024196    -.007263
            21  |  -.0131291   .0041711    -3.15   0.002    -.0213052    -.004953
            22  |  -.0054799   .0041665    -1.32   0.188    -.0136469    .0026871
            23  |  -.0143556   .0039684    -3.62   0.000    -.0221344   -.0065768
            24  |  -.0190829   .0038441    -4.96   0.000     -.026618   -.0115478
            25  |  -.0056036   .0037541    -1.49   0.136    -.0129624    .0017551
            26  |  -.0031526   .0036179    -0.87   0.384    -.0102444    .0039392
            27  |  -.0006298   .0033876    -0.19   0.853      -.00727    .0060104
            28  |   .0013787   .0033443     0.41   0.680    -.0051768    .0079341
            29  |          0  (omitted)
                |
          _cons |   .5143838   .0225582    22.80   0.000     .4701657    .5586019
----------------+----------------------------------------------------------------

My Stata version: 17.0

My question is:
I would like to examine the marginal probabilities for both the treatment and control groups within each month to compare them. As you mentioned, the treatment group is omitted due to the fixed effects. Could you please advise on how I should modify the code to achieve this?

Thank you for your time and assistance.

Kind regards,
Xiao

[Clyde Schechter]

regress outcome_variable i.treat##i.post, vce(cluster id_num)
margins treat, over(month)

[xiao wen]

regress outcome_variable i.treat##i.post, vce(cluster id_num)
margins treat, over(month)

really appreciate for it!