Bootstrapping after gsem

[Yun Dong Yeo]

Hello

I am trying to test for a mediation effect using the gsem and boostrapping.
I am consistently running into a problem where the bootstrapping returns a series of 'x's with the message:
insufficient observations to compute bootstrap standard errors
no results will be saved

Looking at related posts, I think there must be something with my code, but I cannot figure out what the problem is.
Below are my codes.

If anyone can provide wisdom on this matter, it would be greatly appreciated.
Many thanks in advance.

global key_control2 "L2.ln_at L2.ln_firmage L2.tobinq L2.sgrowth L2.unabsorbedslack L2.exploratoryratio L2.comp2_w L2.ln_total_env_viol_amount L2.toxic"

capture program drop boot1
program boot1, rclass
gsem (L1.ln_invest2 <- L2.tran_risk_pro_w_std $key_control2 i.gvkey i.year)(tobinq <- L1.ln_invest2 L2.tran_risk_pro_w_std $key_control2 i.gvkey i.year) if sample==1
return scalar IV_indirect = (_b[tobinq:L1.ln_invest2]*_b[L1.ln_invest2:L2.tran_risk_pro_w_std])
return scalar IV_direct = (_b[tobinq:L2.tran_risk_pro_w_std])
return scalar IV_total = (_b[tobinq:L2.tran_risk_pro_w_std]+_b[tobinq:L1.ln_invest2]*_b[L1.ln_invest2:L2.tran_risk_pro_w_std])
return scalar Mediator_direct = (_b[tobinq:L1.ln_invest2])
end
bootstrap r(IV_indirect) r(IV_direct) r(IV_total) r(Mediator_direct), seed(12345) reps(500) nodrop: boot1 estat boot, bc percentile

[Felix Bittmann]

I think the problem is the if-condition in your sem command. I suggest you first create a dataset that only contains the relevant sample and then use this sample for bootstrapping. And make sure that your program runs fine before using it in combination with bootstrapping.

[Yun Dong Yeo]

Thank you for your suggestion Felix.
I've tried following your direction, but the problem still persists, unfortunately.

[Felix Bittmann]

And then program runs fine and reports all return values as intended when you run it on the actual data before the bootstrap?

[Yun Dong Yeo]

I apologize for the late response.
Yes, the program runs fine before the bootstrap. The problem arises with the bootstrap command. Still trying to figure it out.

[Felix Bittmann]

Without the data I can only guess. Maybe it has to do with missing values or something or your sample selection, as stated above. I suggest the following. Take a bootstrap resample manually and then run the program again. By doing so it might be easier to spot the problem.

Like:

(Define program)
boot1  // Test run
bsample  // Take bootstrap sample from your dataset
boot1