lnmor

Stata module to compute marginal odds ratios after model estimation

lnmor computes (adjusted) marginal odds ratios after logit or probit using G-computation. lnmor works by applying fractional logit to averaged counterfactual predictions from the original model.

For methods and formulas see Jann and Karlson (2023).

Installation from GitHub:

. net from https://raw.githubusercontent.com/benjann/lnmor/main/
. net install lnmor, replace

The moremata package is required. Type

. ssc install moremata, replace

to install moremata.

Main changes:

16jan2023 (version 1.1.5)
- vce(bootstrap) and vce(jackknife) did not pick up level(); this is fixed

15jan2023 (version 1.1.4)
- minor changes in how data is handled when enumerating predictions

13jan2023 (version 1.1.3)
- new subsample() option to evaluate the marginal odds ratio over a subsample
  of the data only; use this option, for example, to obtain an ATET equivalent
  of a marginal OR (by default, an ATE equivalent is computed)
- the core of the code of lnmor has been rewritten (in part, so that the subsample
  option can be supported); the code is now better organized and also faster in
  most situations
- option [no]tbal added
- lnmor did not work with fweights; this is fixed
- IFs were not updated to take account of the estimation of the mean in case of
  dx(atmean); this is fixed

10jan2023 (version 1.1.2)
- lnmor can now be specified repeatedly even if results are posted in e()
  (i.e. if option -post- or option -vce()- has been specified, or if lnmor has
  been applied after -mi estimate- or after -svy-)
- option -miopts()- added; mi options will no longer be collected from
  command line of model
- lnmor now also works after -mi estimate: svy: ...-
- -or- added to program properties

06jan2023 (version 1.1.1)
- option -notable- added
- revised implementation of support for svy

29dec2022 (version 1.1.0)
- fixed bug in display of progress dots in case of dx()

29dec2022 (version 1.0.9)
- option -constant- added
- matrix r(levels) now returned if dx(numlist) or dx(levels) is specified
- r(table) was only returned if option -post- was specified; this is fixed
- revised labeling of terms affected by dx(); revised info on dx() and delta()
  displayed in output
- revised display of progress dots

02sep2022 (version 1.0.8)
- delta(0) now computes log odds rather than log odds ratios
- minor changes to output
- minor changes to help

02sep2022 (version 1.0.7)
- options -rifgenerate()- and -ifscaling()- added
- r(sum_w) added to returns

01sep2022 (version 1.0.6)
- lnmor now requires Stata version 15 or newer
- can now type dx(levels) to report derivative at each observed level
- new option delta() causes dx() to compute discrete-change effects rather than
  derivatives; new options -centered- and -normalize- provide settings; option
  -epsilon- is now undocumented (epsilon is equivalent to
  -delta(#) centered normalize- with # = 2*exp(log(c(epsdouble))/3))

30aug2022 (version 1.0.5)
- dx() now takes derivatives analytically; can specify -epsilon- for old
  behavior

24aug2022 (version 1.0.4)
- options dx(), epsilon(), and nowarn added

23aug2022 (version 1.0.3)
- can now specify interactions such as -c.x##c.x- to analyze nonlinear effects
- some minor changes in how svyr is handled
- lnmor no longer breaks if averaged predictions have no variance

22aug2022 (version 1.0.2)
- support for application after -svy- added
- support for application after -mi estimate- added
- now returning df_r (affects p-values and CIs)
- improved formatting of progress information

19aug2022 (version 1.0.1)
- new at() option introduced; over() is discarded
- imporved parsing of varlist and added various checks and error messages

18aug2022 (version 1.0.0)
- released on GitHub