Reputation
12,883
Top tag
Next privilege 15,000 Rep.
Protect questions
Badges
1 18 54
Newest
 Enlightened
Impact
~717k people reached

1d
revised Predict with pseudo-mean factors in new data
added 351 characters in body
1d
revised Predict with pseudo-mean factors in new data
added 351 characters in body
1d
answered Predict with pseudo-mean factors in new data
1d
comment Predict with pseudo-mean factors in new data
This might be nitpicky, but it seems strange to think of this as default behavior when you are using the atmeans option.
2d
comment what do estimation techniques mean in econometrics? and how different estimators work?
@Nancy Try the Kennedy books mentioned here.
Apr
27
comment what do estimation techniques mean in econometrics? and how different estimators work?
@MatthewGunn Hayashi seems much too advanced a route to pursue in this case.
Apr
25
comment zero-inflated negative binomial in Stata
@JimCurry If that's the actual range, I would not worry. I might try adding the difficult maximization option to see if that helps.
Apr
25
comment zero-inflated negative binomial in Stata
@JimCurry Is the scale of distmag very different from the others? Sometimes that can impede convergence.
Apr
25
comment Bivariate probit model with sample selection
The only mention of something like this that I can find is a very vague statement on p. 183 of Propensity Score Analysis by Guo and Fraser: "In the early days of discussing the Heckman or Heckit models, some researchers, especially economists, assumed that λ [IMR] could be used to measure the level of selectivity effect, but this idea proved controversial and is no longer widely practiced." I still don't get the intuition why someone would do this.
Apr
25
comment Bivariate probit model with sample selection
@quirik I wouldn't, other than in the way I mentioned above. I think this discussion has strayed too far from the original question. I encourage you to start a new question rather than continuing with more here.
Apr
25
comment Bivariate probit model with sample selection
@quirik I have never seen this interpretation and it does not make sense to me. I would looks at the predicted probabilities from the first stage if I wanted to compare.
Apr
24
comment Bivariate probit model with sample selection
Yup. That a great way of doing it's
Apr
24
comment Bivariate probit model with sample selection
Almost right. The percentage part depends on how how wages are measured or transformed, say with logs. The sample average part is wrong. The IMR is a nonlinear function of variables in the selection equation, so you could evaluate it any values, not just the average ones.
Apr
24
comment Bivariate probit model with sample selection
@quirik The IMR times its coefficient is the expected error term for those working.
Apr
22
comment Bivariate probit model with sample selection
This is turning into its own question! It is much better to ask related questions as separate, new questions. You are right that this is essentially an omitted variables problem. If we observed those factors or had an estimate of them, we could fix this by putting them in. From properties of the bivariate normal distribution, we do have such an estimate, up to an unknown parameter: it is the inverse Mills ratio times the its coefficient in the wage equation. This captures the fact that the wage epsilon no longer zero in expectation since the negative epsilons with low education are hiking.
Apr
22
comment Model panel data with ONLY time-invariant variables
That's exactly what the test will tell you. Also, some people have tried to model the FE as an outcome of the time invariant regressors, but that it's not quite right since it is an estimated quantity.
Apr
22
comment How to interpret the constant in Oaxaca Blinder decomposition?
@sanne Great!. Please select the answer by clicking on the check mark on the left.
Apr
22
comment Bivariate probit model with sample selection
It's a potential outcome. Think of it as a wage offer, rather than an actual wage. Just because someone did not accept it, doesn't mean it does not exist.
Apr
21
comment Pratio and the Probit Model
I have not idea what a Pratio or delta are. Please clarify. Typically, $\Pr(Y=1 \vert X=1)=\Phi(1.4727 - 1.5794 \cdot 1)=0.457513$, where $\Phi(.)$ is the standard normal CDF.
Apr
21
revised Endogeneity issue in time series model
added 12 characters in body; edited tags