Robustness Check for Logistic Coefficients

[Mitch Lingo]

Does anyone know of a logistic coefficient robustness/sensitivity check for downloading into Stata or has an online calculator? Ken Frank and his colleagues have KonFound-It!, but the check for logistic regression is unpublished and still in beta.

Thanks,

Mitch

[ericmelse]

Possibly the module calibrationbelt, developed by Giovanni Nattino, Stanley Lemeshow, Gary Phillips, Stefano Finazzi and Guido Bertolini, is of use to you:

ssc install calibrationbelt
* note that some example data files can be downloaded as well, for that use:
net sj 17-4 gr0071

Note that there also is a paper in The Stata Journal: Assessing the calibration of dichotomous outcome models with the calibration belt (but still behind a paywall). For more papers, see the help file:

h calibrationbelt

[Maarten Buis]

See ssc desc scenreg

[Mitch Lingo]

Thank you both!

[Magashi Joseph]

Dear Ben Jann et al

Is there any one with idea on how I can get robust binary and multinomial logistic regression using roblogit command?
I have tried to install the command but could not be found and it is not recognized as a STATA command. Kindly assist me.

The link for the commnd
https://www.stata.com/meeting/uk17/s...uk17_Jann2.pdf

With regards,
Joseph

[kenneth frank]

Mitch, I appreciate the interest. We are working on publications now. The application will be to medical research with dichotomous outcomes (see the Blog at https://www.konfound-it.org/). In a COVID example from an early RCT of HCQ, if only one treatment recovery were switched to treatment and not recovered the result would not longer be statistically significant.

The app for dichotomous outcomes has 2 flavors, one for user entered 2x2 table and another for entering a logistic regression coefficient, its standard error, the sample size, number of covariates and number of treatment cases.

For the 2x2 table it is easy to verify the resulting chi-square is just above p=.05 after switching a certain number of treatment successes to treatment failures. We will implement other test statistics and thresholds at some point. For the mode where the user enters the coefficient, etc., the trick is we have to convert that into an implied 2x2 table and then switch cases. Sometimes the implied table doesn't work in unusual circumstances (I think especially small n). But if the implied 2x2 table looks right, you can again easily verify the p value after cases have been switched from treatment success to treatment failure. There is an R macro now for this. working on STATA.
Check out:
http://konfound-it.com or the Blog https://www.konfound-it.org/

Ken

[kenneth frank]

we have now published our paper on this:
https://papers.ssrn.com/sol3/papers....act_id=3607967
*Lin, Q., *Maroulis, S. J.,, and *Mueller, A. S., Xu, R., Rosenberg, J.M., Hayter, C. S., Mahmoud, R.A., Kolak, M., Dietz, T., Zhang, L. (accepted for publication). “Hypothetical case replacement can be used to quantify the robustness of trial results.” Journal of Clinical Epidemiology. *equal first authors, listed alphabetically. DOI:https://doi.org/10.1016/j.jclinepi.2021.01.025

it is for 2x2 tables, but the app at http://konfound-it.com can convert a logistic regression coefficient and it's standard error into an implied 2x2 table and then proceed.