dataprep() applies predictors.op to the treated unit but not the controls

[soodoku]

dataprep() aggregates predictors over time.predictors.prior separately for the treated unit and the controls. The treated unit uses the operator the caller supplied:

# R/dataprep.R:216
X1 <- apply(X1, 2, paste(predictors.op), na.rm = TRUE)

The controls use a hard-coded mean:

# R/dataprep.R:257
X0 <- sapply(X0, apply, 2, mean, na.rm = TRUE, simplify = TRUE)

So for any predictors.op other than "mean", X1 and X0 are built under different aggregation rules and the predictor-balance matrix that synth() optimises over is internally inconsistent.

Reproduction

Bundled basque data, predictors = "invest", predictors.op = "median", time.predictors.prior = 1964:1969:

library(Synth)
data(basque)
controls <- c(2:16, 18); prior <- 1964:1969

dp <- dataprep(foo = basque, predictors = "invest", predictors.op = "median",
               dependent = "gdpcap", unit.variable = "regionno",
               time.variable = "year", treatment.identifier = 17,
               controls.identifier = controls,
               time.predictors.prior = prior,
               time.optimize.ssr = 1960:1969, time.plot = 1955:1997)

hand <- function(u, f) f(basque$invest[basque$regionno == u & basque$year %in% prior],
                         na.rm = TRUE)

On main (1.2-0):

TREATED unit (regionno 17)
  dataprep X1        : 24.41825
  hand median        : 24.41825   <- operator honoured
  hand mean          : 24.64738

CONTROL units
  max |X0 - hand median| : 2.061043
  max |X0 - hand mean|   : 0.000000   <- controls received the mean

Asking for the median moves only the treated unit's row; every control row is byte-identical to what the "mean" call produces.

Change

One line, mirroring what line 216 already does:

     X0 <- split(X0, X0[,dim(X0)[2]])
-    X0 <- sapply(X0, apply, 2, mean, na.rm = TRUE, simplify = TRUE)
+    X0 <- sapply(X0, apply, 2, paste(predictors.op), na.rm = TRUE, simplify = TRUE)
     X0 <- as.matrix(X0[-dim(X0)[1],])

With it, max |X0 - hand median| becomes 0.000000.

Two things suggested this is an oversight rather than a deliberate asymmetry, though you would know better than I do:

1. The message printed a few lines earlier in the control branch refers to the argument by name — "We ignore (na.rm = TRUE) all missing values for predictors.op" (R/dataprep.R:252) — in the same block that hard-codes mean.
2. special.predictors operators are applied to both the treated unit and the controls (R/spec.pred.func.R), so the package currently treats the two kinds of predictor differently.

?dataprep describes predictors.op as the method used to aggregate the predictor variables, with no treated/control distinction.

Verification

• Added tests/testthat/test-dataprep-predictors-op.R: FAIL 1 / PASS 1 without this change, PASS 2 with it.
• Full suite via cd tests && NOT_CRAN=true Rscript testthat.R: FAIL 0 | SKIP 3 | PASS 187 both before and after.
• R/dataprep.R is byte-identical between CRAN 1.1-10 and main at 1.2-0, so this affects the released version too.

What I have not established

Whether the asymmetry is intended — it is possible you want control units summarised by their mean regardless of predictors.op. If so the right change is a documentation note rather than this patch, and I am happy to be told that.