Reproduce Selected 'Stata' Expression and Missing-Value Semantics

A lightweight compatibility layer for researchers translating 'Stata' data-management code into R. Provides vectorised helpers that reproduce selected 'Stata' expression semantics where they differ from base R, most importantly "Stata's" rule that numeric missing values are larger than every number: relational comparisons, inlist(), inrange(), cond(), missing(), and the 'egen' row functions rowtotal(), rowmean(), rowmiss() and rownonmiss(). Only the ordinary missing value is fully supported; "Stata's" extended missing values are detected and rejected wherever they would change a result.


statacompat

A lightweight R package that reproduces selected Stata expression and missing-value semantics in R.

When Stata data-management code is translated into R, an expression can look identical in both languages and still return different results. Most of these differences come from missing values. statacompat provides a small set of vectorised helpers, all prefixed stata_, that behave like their Stata counterparts and work inside ordinary dplyr::mutate() calls.

It is meant for researchers who work in both Stata and R, including those replicating Stata results in R or migrating .do files.

Why base R and Stata disagree

In Stata, a numeric missing value (.) is larger than every number. Relational and logical expressions therefore never return missing. In R, any comparison with NA returns NA.

gen high_income = income > 10000000
income <- c(5000000, 25000000, NA)

income > 10000000                # naive translation
#> [1] FALSE  TRUE    NA

stata_gt(income, 10000000)       # what Stata computes
#> [1] FALSE  TRUE  TRUE

In Stata, the missing income is coded as high income. That is often not what the analyst intended, and the usual Stata safeguard is income > 10000000 & income < .. The point is that statacompat gives the same answer as the original Stata code, including the cases where that code is questionable, so a translation can be checked line by line:

stata_gt(income, 10000000) & stata_lt(income, NA)   # Stata: ... & income < .
#> [1] FALSE  TRUE FALSE

What is in version 1.0

Stata statacompat Note
==, !=, >, >=, <, <= stata_eq(), stata_ne(), stata_gt(), stata_ge(), stata_lt(), stata_le() Missing is greater than all numbers. Strings are compared by bytes. Never returns NA.
inlist(z, a, b, ...) stata_inlist(x, ...) Missing is an ordinary member. Candidates are evaluated row by row.
inrange(z, a, b) stata_inrange(x, lower, upper) Inclusive. Missing x gives FALSE. A missing bound means unbounded.
cond(x, a, b[, c]) stata_cond(condition, yes, no, missing) A missing numeric condition counts as true.
missing(x1, ...) stata_missing(...) For strings, "" is missing but " " is not.
egen rowtotal() / rowmean() / rowmiss() / rownonmiss() stata_rowtotal(), stata_rowmean(), stata_rowmiss(), stata_rownonmiss() The missing and strok options are supported.

Comparisons and predicates return logical vectors. Use as.integer() when you need a 0/1 indicator variable.

age <- c(12, 15, 64, 65, NA)
age >= 15 & age <= 64              # base R
#> [1] FALSE  TRUE  TRUE FALSE    NA
stata_inrange(age, 15, 64)         # Stata inrange(age, 15, 64)
#> [1] FALSE  TRUE  TRUE FALSE FALSE
stata_inrange(age, 15, NA)         # 15 and older, missing excluded
#> [1] FALSE  TRUE  TRUE  TRUE FALSE

kbli <- c(1, 4, NA)
stata_inlist(kbli, 1, 2, 3)
#> [1]  TRUE FALSE FALSE
stata_inlist(kbli, 1, NA)          # Stata inlist(kbli, 1, .)
#> [1]  TRUE FALSE  TRUE

r1 <- c(1, 1, NA); r2 <- c(2, NA, NA); r3 <- c(3, 3, NA)
rowMeans(cbind(r1, r2, r3), na.rm = TRUE)
#> [1]   2   2 NaN
stata_rowmean(r1, r2, r3)
#> [1]  2  2 NA
stata_rowtotal(r1, r2, r3)                  # egen rowtotal()
#> [1] 6 4 0
stata_rowtotal(r1, r2, r3, missing = TRUE)  # egen rowtotal(), missing
#> [1]  6  4 NA

wage <- c(0, 1500, NA)
ifelse(wage, "paid", "unpaid")
#> [1] "unpaid" "paid"   NA
stata_cond(wage, "paid", "unpaid")          # Stata cond(wage, "paid", "unpaid")
#> [1] "unpaid" "paid"   "paid"

String comparisons in base R depend on the session locale. Stata compares bytes. In an Indonesian or English UTF-8 locale:

x <- c("apel", "Bayam", "")
x > "B"                # depends on the locale
#> [1] FALSE  TRUE FALSE
stata_gt(x, "B")       # byte order, as in Stata, in every locale
#> [1]  TRUE  TRUE FALSE

In a pipeline

library(dplyr)

dta <- data.frame(
  kbli = c(1, 5, NA, 3),
  age  = c(14, 30, 70, NA),
  r10a = c(1, 2, NA, 2),
  r10b = c(2, 2, NA, 1)
)

dta |>
  mutate(
    sector      = stata_inlist(kbli, 1, 2, 3),
    working_age = stata_inrange(age, 15, 64),
    atwork      = stata_cond(stata_eq(r10a, 1) | stata_eq(r10b, 1), 1, 0)
  )
#>   kbli age r10a r10b sector working_age atwork
#> 1    1  14    1    2   TRUE       FALSE      1
#> 2    5  30    2    2  FALSE        TRUE      0
#> 3   NA  70   NA   NA  FALSE       FALSE      0
#> 4    3  NA    2    1   TRUE       FALSE      1

The row functions take vectors or data frames, so dplyr::pick() works: stata_rowtotal(pick(r1:r5)).

Where the rules come from

  • Every rule is taken from Stata's manuals: [U] 12.2.1, [U] 13.2.3–13.2.4, [U] 12.4.2, [FN] Programming functions and [D] egen.
  • A few cases are not spelled out in the manuals: inlist() with missing values, rowmiss() with string variables, and the storage type egen uses when none is given. The help pages describe what statacompat does in these cases and state that the behaviour is inferred from documented rules.
  • The expected values in the test suite are written from the manuals; they are not computed by the package.

statacompat does not claim full Stata compatibility. It reproduces the functions listed above, within the limits below.

Limitations and non-goals

  • Extended missing values (.a–.z, read by haven::read_dta() as tagged NAs) are only partly supported. A result that depends on telling one missing value from another, such as .a == ., raises an error. All other results match Stata. To treat every missing value as ., convert first with x[is.na(x)] <- NA_real_.
  • NA_character_ has no Stata equivalent, so string comparisons reject it. Stata's missing string is "".
  • Storage precision. egen stores its result as float unless a type is given, and float rounds integers above 16,777,216. statacompat always computes in double precision, so its results correspond to egen double .... This matters for large Rupiah totals.
  • Arithmetic is not emulated. In Stata 1/0 gives ., but in R it gives Inf. statacompat treats Inf as an ordinary number.
  • Types. Factors are rejected. Dates can be compared only with dates of the same class, because Stata's date origin (1960) differs from R's (1970).
  • Out of scope: translating .do files, recode, _n/_N, bysort, egen group(), estimation commands, and aliases for Stata commands.

Installation

From a local copy of the source:

# local
# install.packages("devtools")
devtools::install("path/to/statacompat")

or from github:

# github
# install.packages("remotes")
remotes::install_github("aripurwantosp/statacompat", build_vignettes = TRUE)

statacompat has no runtime dependencies beyond base R (R >= 4.1).

Learn more

  • vignette("stata-and-r-semantics", package = "statacompat"): When Stata and R Disagree: Expressions and Missing Values.
  • ?stata_comparison, ?stata_inrange, ?stata_cond, ?stata_row*.

AI-Assisted Development

Generative AI tools, particularly Claude by Anthropic, were used to assist in the development of this package, including code drafting, debugging, refactoring, documentation, and test development. All AI-assisted outputs were reviewed, evaluated, and validated by the package author.

Reference manual

It appears you don't have a PDF plugin for this browser. You can click here to download the reference manual.

install.packages("statacompat")

1.0 by Ari Purwanto Sarwo Prasojo, 12 hours ago


https://github.com/aripurwantosp/statacompat


Report a bug at https://github.com/aripurwantosp/statacompat/issues


Browse source code at https://github.com/cran/statacompat


Authors: Ari Purwanto Sarwo Prasojo [aut, cre] (ORCID:


Documentation:   PDF Manual  


MIT + file LICENSE license


Suggests dplyr, haven, knitr, rmarkdown, testthat, withr


See at CRAN