[Luca Uberti]

I see. That explains it. Thanks a lot!

Best,

Luca

[Luca Uberti]

Many thanks. All clear!

Best,

Luca

[pritika dua]

Can the regression result and its marginal effects give absolutely same results???
This is my example
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
name: <unnamed>
log: C:\Users\abc\Desktop\pca\Untitled.smcl
log type: smcl
opened on: 31 Jul 2017, 21:32:21

. xttobit index L.Notice CPCB CSRcommittee MNC Parent Union Iso9001 Export L.award AH L.PATENTFILED sqage logsales COACT i.STATEID ,ll(0)

Obtaining starting values for full model:

Iteration 0: log likelihood = 626.50199
Iteration 1: log likelihood = 682.66397
Iteration 2: log likelihood = 707.58351
Iteration 3: log likelihood = 710.07847
Iteration 4: log likelihood = 710.1235
Iteration 5: log likelihood = 710.12352

Fitting full model:

Iteration 0: log likelihood = 245.40268
Iteration 1: log likelihood = 301.42609
Iteration 2: log likelihood = 307.58363
Iteration 3: log likelihood = 307.76688
Iteration 4: log likelihood = 307.76731
Iteration 5: log likelihood = 307.76731 (backed up)
Iteration 6: log likelihood = 307.76806
Iteration 7: log likelihood = 307.76814
Iteration 8: log likelihood = 307.76818
Iteration 9: log likelihood = 307.76818

Random-effects tobit regression Number of obs = 741
Group variable: companynum Number of groups = 151

Random effects u_i ~ Gaussian Obs per group: min = 2
avg = 4.9
max = 5

Integration method: mvaghermite Integration points = 12

Wald chi2(31) = 330.07
Log likelihood = 307.76818 Prob > chi2 = 0.0000

------------------------------------------------------------------------------
index | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
Notice |
L1. | .0148768 .0107438 1.38 0.166 -.0061806 .0359342
|
CPCB | .0570734 .0187271 3.05 0.002 .020369 .0937779
CSRcommittee | .007154 .012725 0.56 0.574 -.0177865 .0320945
MNC | .0719511 .0440901 1.63 0.103 -.0144639 .1583662
Parent | .2333815 .0304461 7.67 0.000 .1737082 .2930548
Union | .065177 .0199763 3.26 0.001 .0260241 .1043299
Iso9001 | .114971 .0426138 2.70 0.007 .0314495 .1984924
Export | .0154074 .0284589 0.54 0.588 -.0403711 .0711858
|
award |
L1. | .0318221 .0124943 2.55 0.011 .0073337 .0563106
|
AH | .0046308 .022849 0.20 0.839 -.0401525 .0494141
|
PATENTFILED |
L1. | -.0002222 .0001323 -1.68 0.093 -.0004816 .0000372
|
sqage | -5.23e-06 4.87e-06 -1.08 0.282 -.0000148 4.31e-06
logsales | .0927621 .0181345 5.12 0.000 .0572192 .128305
COACT | .0302906 .0091358 3.32 0.001 .0123848 .0481964
|
STATEID |
2 | .1126832 .0572142 1.97 0.049 .0005454 .2248211
3 | .0310708 .0980633 0.32 0.751 -.1611297 .2232713
4 | .0514212 .0801664 0.64 0.521 -.1057022 .2085445
5 | .0838364 .0919361 0.91 0.362 -.0963551 .2640279
6 | -1.039833 56.02809 -0.02 0.985 -110.8529 108.7732
7 | -.005597 .0878369 -0.06 0.949 -.1777541 .1665601
8 | .3041142 .1527436 1.99 0.046 .0047422 .6034862
9 | -.028072 .0852568 -0.33 0.742 -.1951723 .1390284
10 | -1.00849 293.668 -0.00 0.997 -576.5873 574.5703
11 | -.8904127 277.92 -0.00 0.997 -545.6036 543.8227
12 | .0718574 .0847519 0.85 0.397 -.0942533 .237968
13 | .1042805 .1547154 0.67 0.500 -.1989561 .407517
14 | .2275617 .1153982 1.97 0.049 .0013855 .453738
15 | .1777647 .2115276 0.84 0.401 -.2368219 .5923513
16 | .457966 .216701 2.11 0.035 .0332398 .8826921
17 | .3761832 .1556118 2.42 0.016 .0711898 .6811767
18 | .2484007 .2149322 1.16 0.248 -.1728588 .6696601
|
_cons | -.6155456 .0819813 -7.51 0.000 -.776226 -.4548652
-------------+----------------------------------------------------------------
/sigma_u | .2002977 .0157608 12.71 0.000 .1694071 .2311884
/sigma_e | .0807691 .0029926 26.99 0.000 .0749038 .0866344
-------------+----------------------------------------------------------------
rho | .8601362 .0215064 .8136494 .8980356
------------------------------------------------------------------------------

Observation summary: 262 left-censored observations
479 uncensored observations
0 right-censored observations

. argins, dydx(_all)
unrecognized command: argins
r(199);

. margins, dydx(_all)

Average marginal effects Number of obs = 741
Model VCE : OIM

Expression : Linear prediction, predict()
dy/dx w.r.t. : L.Notice CPCB CSRcommittee MNC Parent Union Iso9001 Export L.award AH L.PATENTFILED sqage logsales COACT 2.STATEID 3.STATEID 4.STATEID 5.STATEID 6.STATEID 7.STATEID
8.STATEID 9.STATEID 10.STATEID 11.STATEID 12.STATEID 13.STATEID 14.STATEID 15.STATEID 16.STATEID 17.STATEID 18.STATEID

------------------------------------------------------------------------------
| Delta-method
| dy/dx Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
Notice |
L1. | .0148768 .0107438 1.38 0.166 -.0061806 .0359342
|
CPCB | .0570734 .0187271 3.05 0.002 .020369 .0937779
CSRcommittee | .007154 .012725 0.56 0.574 -.0177865 .0320945
MNC | .0719511 .0440901 1.63 0.103 -.0144639 .1583662
Parent | .2333815 .0304461 7.67 0.000 .1737082 .2930548
Union | .065177 .0199763 3.26 0.001 .0260241 .1043299
Iso9001 | .114971 .0426138 2.70 0.007 .0314495 .1984924
Export | .0154074 .0284589 0.54 0.588 -.0403711 .0711858
|
award |
L1. | .0318221 .0124943 2.55 0.011 .0073337 .0563106
|
AH | .0046308 .022849 0.20 0.839 -.0401525 .0494141
|
PATENTFILED |
L1. | -.0002222 .0001323 -1.68 0.093 -.0004816 .0000372
|
sqage | -5.23e-06 4.87e-06 -1.08 0.282 -.0000148 4.31e-06
logsales | .0927621 .0181345 5.12 0.000 .0572192 .128305
COACT | .0302906 .0091358 3.32 0.001 .0123848 .0481964
|
STATEID |
2 | .1126832 .0572142 1.97 0.049 .0005454 .2248211
3 | .0310708 .0980633 0.32 0.751 -.1611297 .2232713
4 | .0514212 .0801664 0.64 0.521 -.1057022 .2085445
5 | .0838364 .0919361 0.91 0.362 -.0963551 .2640279
6 | -1.039833 56.02809 -0.02 0.985 -110.8529 108.7732
7 | -.005597 .0878369 -0.06 0.949 -.1777541 .1665602
8 | .3041142 .1527436 1.99 0.046 .0047422 .6034862
9 | -.028072 .0852568 -0.33 0.742 -.1951723 .1390284
10 | -1.00849 293.668 -0.00 0.997 -576.5873 574.5703
11 | -.8904127 277.92 -0.00 0.997 -545.6036 543.8227
12 | .0718574 .0847519 0.85 0.397 -.0942533 .237968
13 | .1042805 .1547154 0.67 0.500 -.1989561 .407517
14 | .2275617 .1153982 1.97 0.049 .0013855 .453738
15 | .1777647 .2115276 0.84 0.401 -.2368219 .5923513
16 | .457966 .216701 2.11 0.035 .0332398 .8826921
17 | .3761832 .1556118 2.42 0.016 .0711898 .6811767
18 | .2484007 .2149322 1.16 0.248 -.1728588 .6696601
------------------------------------------------------------------------------
Note: dy/dx for factor levels is the discrete change from the base level.

. log close
name: <unnamed>
log: C:\Users\abc\Desktop\pca\Untitled.smcl
log type: smcl
closed on: 31 Jul 2017, 21:34:52
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Kindly guide on the same

[Francesco Baraldi]

Dear Professor Silva,
I read this thread because I have issues like Luca’s ones, but I would need some further clarifications. I ran xtpoisson with RE and vce cluster (before I ran hausman test to check whether using FE or RE), however after having read your replies I decided to try to run a pooled poisson with vce cluster. The results, both coefficients and p-values, are the same, while the AIC of the pooled poisson is slightly smaller. I also tested the goodness of fit of the pooled poisson and it seems to fit quite well. My questions are three:

1. Is it normal to have the same results?
2. If it is ok, are you suggesting using the pooled poisson instead of the xtpoisson, re as I understood from one of your replies?
3. Could you kindly provide some references for the choice of pooled poisson over xtpoisson, re?

Thank you in advance,
All the best

[Joao Santos Silva]

Dear Francesco Baraldi

It is difficult to comment without seeing your results, but I would not use a Poisson RE estimator. The beauty of Poisson regression (with or without FE) is its robustness, but that is lost if you use RE (unless T is very large). So, I would stick to either Poisson regression or Poisson regression with FE.

Best wishes,

Joao

[Francesco Baraldi]

Dear Professor Silva,
Thank you for your reply. I decided for fe, it ultimately seems more consistent. However, I have another question: could you kindly provide some references for margins after xtpoisson, fe? As I understood from your past replies, looking at margins after xtpoisson, fe is not useful, but I cannot grasp why.
Thank you again
Best wishes,
Francesco

[Joao Santos Silva]

Dear Francesco Baraldi,

Please see (especially the penultimate slide)

http://repec.org/usug2016/santos_uksug16.pdf

Best wishes,

Joao

[Osiris Parcero]

Dear Grace & Luca,

If you are using a fixed effects estimator, -margins- will not give you any interesting results. Also, you cannot do predictions. The reason is simple: all of these depend on fixed effects and we do not have consistent estimates for them. Setting the FE at some value is meaningless because the result will depend on the way you measure your regressors. For example, if in the model in #1 we replace age with year of birth the results of -margins- and predictions will change.

In short, do not do it! In the Poisson case the coefficients have a natural interpretation and that should be enough.

All the best,

Joao

Dear Joao,
In your comment #4, you advise not to calculate the marginal effect when using Poisson regression. And then you add: “In the Poisson case the coefficients have a natural interpretation and that should be enough.”
Thank you so much for making us realize this fundamental issue.
What if there are interaction terms and one is interested in finding the (marginal) effect of one individual variable? The effect of an individual variable can only be obtained by using the marginal effects.
For instance, in the regression y = a + b x1 + c x2 + d x1 x2, we want to know the (marginal) effect of x1.
In my case, I am interested in a fixed-effects regression. So, is it right to say that in this case, we may have to give up using a fixed-effects Poisson regression and instead use a linear fixed effects regression?
Osiris

[Joao Santos Silva]

Dear Osiris Parcero,

What the FE Poisson regression gives you in this case is the elasticity with respect to x1, which is varies with x2. I would say that this has a meaningful interpretation, and therefore you do not need to compute a partial effect.

Best wishes,

Joao

[Osiris Parcero]

Thank you, Joao, for your prompt response.
Yes, getting the elasticity would be good enough. Continuing with the same example I provided in my previous post, i.e.,
y = a + b x₁ + c x₂ + d x₁ x₂
Let’s say the estimated regression coefficients are as follows: b=0.5 and d=–0.3, and both are statistically significant. In this case, is it right to calculate the elasticity of y with respect to x₁, for x₂=1.1, in the following way?
Elasticity = 0.5 – 0.3 1.1 = 0.5 – 0.33 = 0.17
Also, how can I estimate whether this point estimation of the elasticity (0.17) is significant or not?
Thank you again, Joao, and apologies for my delayed answer. I thought I set my Stata account for the response to be announced via email, but apparently, I did not do it properly.
Osiris

[Joao Santos Silva]

Dear Osiris Parcero,

That is correct and you can use lincom to compute the standard error.

Best wishes,

Joao

[Osiris Parcero]

Thank you, Joao,
I now see that using lincom solves the problem of calculating the significance of the elasticity.
Please allow me another question in relation to the same fixed-effects regression with interaction terms:
y = a + b x₁ + c x₂ + d x₁ x₂
Would the elasticity still have a meaningful interpretation if the variable x₁ is a dummy variable (having 0s and 1s)?
Clearly, a significant negative or positive elasticity will be meaningful. But beyond that, I am not sure how to interpret the elasticity of y with respect to x₁ in this case because the percentage change from 0 to 1 is equal to infinity.
Thank you again,
Osiris

[Joao Santos Silva]

Dear Osiris Parcero,

I need to correct What I said above. The coefficient is an elasticity when the regressor is in logs, otherwise it is a semi-elasticity. So, if x1 is a binary variable or any variable in levels, the coefficient is a semi-elasticity, and it is an elasticity if the is in logs. Sorry about the oversight.

Best wishes,

Joao

[Osiris Parcero]

Thank you Joao, for the clarification and all the previous answers. They very much helped my understanding, so now I can move on.
Best wishes,
Osiris

[Steven YH Chen]

Dear Joao Santos Silva:

I have a question about the post estimation after xtpoisson with the option of robust i(id) fe. Is anything we can run for post estimation after that?

Thank you so much!