stata2r

R package that can translate a subset of Stata commands used for data preparation to R. This will not be usable to convert complete Stata analyses to R (use AI for full translation of some Stata code you have). It only translates selected data preparation steps.

It will be used as part of the repbox / metareg pipeline that shall facilitate systematic methodological meta studies based on actual reproduction package.

A pilot example for such a meta study is my paper:

From Replications to Revelations: Heteroskedasticity-Robust Inference

But eveything is work in progress.

Architectural Scope: What stata2r Does (and Doesn't) Do

stata2r is strictly a Data Manipulation Translator.

• It translates commands that structurally or mathematically alter the data set (e.g., generate, replace, collapse, merge, keep, drop, xi, egen).

Generat no e() or r() values

• It does not translate commands for side effects beyond data manipulation, i.e. to generate e() or r() variables.

How it is handled instead: metaregBase natively executes the Stata script and writes any needed e() or r() values directly to disk (e.g., as .dta or .txt files). repboxDRF injects R code to load these exact values into stata2r_env immediately before they are needed. stata2r simply maps the expression e(sample) to stata2r_env$e_sample, blindly trusting that the environment has been correctly populated by the orchestrator.

Why: If the pipeline initiates a reproduction from a cached intermediate data file (Safe Caching), prior regression commands are skipped. If stata2r relies on tracking its own running state, it will crash when asked to recall an e(sample) from a skipped command.

NO Regression Translation

• stata2r should never attempt to simulate Stata's regress or similar command.

• This will be done elsewhere in the repbox pipeline

• Results like e(sample) from a regression are pre-populated by repboxDRF.

• The predict command, which modifies the data set but relies on a previous regression will not be translated by stata2r.