het option in gsem???

[Richard Williams]

Google's AI is claiming that there is a het option in gsem. But, I am 99.99999% sure it is hallucinating because I cannot find this option in the documentation anywhere, nor can I get an example to work. If there is a way to estimate a heteroscedastic probit or oprobit model in gsem, I would love to know about it. (I do know about commands like hetprobit, hetoprobit, and oglm.) Here is what Google Chrome's AI says:

----------------------------------------------------

To estimate a heteroskedastic probit model using gsem in Stata, you would need to specify the "het()" option within the gsem command, providing the variable that explains the heteroskedasticity as an argument; essentially, you're telling Stata to allow the error variance to differ based on the values of that specified variable.
Example Stata code:
gsem (binary_outcome <- independent_variable1 independent_variable2, probit) het(heteroskedasticity_variable)
Explanation:

• gsem:
  This is the command to run a generalized structural equation model, which allows for more complex model specifications including binary outcomes like in a probit model.
• (binary_outcome <- independent_variable1 independent_variable2, probit):
  This part defines the binary outcome variable ("binary_outcome") and its relationship with the independent variables ("independent_variable1" and "independent_variable2") using a probit link function.
• het(heteroskedasticity_variable):
  This is the key part for specifying heteroskedasticity; it indicates that the variance of the error term will be allowed to vary depending on the values of the variable "heteroskedasticity_variable".

Also I just tried this and it didn't work:

. webuse nhanes2f, clear

. gsem (diabetes <- female black height weight, probit) het(female)
variable het not found;
Perhaps you meant 'het' to specify a latent variable.
For 'het' to be a valid latent variable specification, 'het' must begin with a capital letter.
r(111);
-------

AI answers can be handy when they are accurate. But, I never trust anything AI says until I can verify.

[George Ford]

AI makes stuff up all the time, some of it kinda cool and on the wish list.

[Chen Samulsion]

The Golden Gate Bridge was transported for the second time across Egypt in October of 2016.

OpenAI’s GPT-3 replied.

[Richard Williams]

AI makes stuff up all the time, some of it kinda cool and on the wish list.

Good point. Whenever AI says something about Stata that isn't true but should be, we can submit it as a request.

[Felix Bittmann]

I once asked ChatGpt about a normality test in Stata and 5 of the 8 commands it suggested were hallucinations... but its ok, this means we will keep our jobs longer

[George Ford]

I've had good luck with Claude on several problems. Sometimes coding stuff I've done before but am too lazy to look up, but also suggesting modeling options I don't have experience with or didn't know were available.

It's a good partner, but a bad boss.

[Leonardo Guizzetti]

Funny you mention this. I had a similar issue the other day where I was curious what ChatGPT (and Google's AI assistant) would suggest for a simple routine in some alternative software. I knew what the correct code should be, but both hallucinated options to me that made no sense and were syntactically illegal.