"initial values not feasible" error with the Probit Heckman model

[Paolo Maldini]

I am trying to run a Probit Heckman model following the Stata guideline here, and I keep running into this error message:
"initial values not feasible
last estimates not found"

But I have no issues running the same data with both regress and Heckman commands, and I would appreciate any support on this.

Here is a data example:
dataex avg_sentiment opin_leader collective_action economic_demand_complaint count
output:
```
----------------------- copy starting from the next line -----------------------

* Example generated by -dataex-. For more info, type help dataex
clear
input double avg_sentiment float opin_leader byte(collective_action economic_demand_complaint) int count
-.6666666666666666 0 0 0 12
-.6666666666666666 0 1 1 12
-.6666666666666666 0 0 0 12
-.6666666666666666 0 0 0 12
-.6818181818181819 1 0 0 22
-.6818181818181819 1 0 1 22
-.6818181818181819 1 0 1 22
-.6818181818181819 1 0 1 22
-.6818181818181819 1 0 1 22
-.6818181818181819 1 0 1 22
-.6818181818181819 1 0 0 22
-.6818181818181819 1 0 1 22
-.6818181818181819 1 0 1 22
-.6818181818181819 1 0 1 22
-.2307692307692307 1 0 1 26
-.2307692307692307 1 0 0 26
-.2307692307692307 1 0 0 26
-.2307692307692307 1 0 1 26
-.2307692307692307 1 0 0 26
-.2307692307692307 1 0 0 26
-.2307692307692307 1 0 1 26
end

------------------ copy up to and including the previous line ------------------
```

[Paolo Maldini]

I have posted this on Stackoverflow as well.

[FernandoRios]

Hi Paolo,
there isnt really enough information to figure out the problem.
What the command is telling you is that it couldn't find any combinations for your betas, sigmas and correlations that provided values for the loglikelihood.
It can be due to many factors including missing data, incorrectly coded data, using strings instead of numerical variables, etc.
It could also be that the data itself does not lend itself to be modeled under the bivariate normal assumption of Heckman probit.
F

[Paolo Maldini]

Hi Paolo,
there isnt really enough information to figure out the problem.
What the command is telling you is that it couldn't find any combinations for your betas, sigmas and correlations that provided values for the loglikelihood.
It can be due to many factors including missing data, incorrectly coded data, using strings instead of numerical variables, etc.
It could also be that the data itself does not lend itself to be modeled under the bivariate normal assumption of Heckman probit.
F

Thanks and apologies, this is the code that produces the error:
```
heckprob avg_sentiment opin_leader collective_action , select(economic_demand_complaint=opin_leader count)
```

[Andrew Musau]

heckprobit is for fitting a probit model with sample selection, so you need a binary outcome. For a continuous outcome, see

help heckman

[Paolo Maldini]

heckprobit is for fitting a probit model with sample selection, so you need a binary outcome. For a continuous outcome, see

help heckman

Thanks Andrew for the support!