Add type="terms" support for predict.merMod

[Copilot]

Implements support for type = "terms" in predict.merMod, replacing the previous unimplemented/unused behavior for the terms argument.

<issue_title>type = "terms" for predict.merMod</issue_title>

<issue_description>there's no "terms" option for the type parameter of predict.merMod. Actually, it is currently unused; as shown in the predict.merMod.Rd file:

 \item{terms}{a \code{\link{terms}} object - unused at present.}
```</issue_description>

Changes made

• Added "terms" as a valid type option in predict.merMod.
• Implemented term-wise fixed-effect contribution output for type = "terms".
• Added support for selecting specific terms via the terms argument (terms object, names, or numeric indices).
• Added validation/errors for invalid combinations and invalid term selectors.
• Updated predict.merMod.Rd documentation to describe the new behavior.
• Added focused tests in tests/testthat/test-predict.R for:
  • reconstruction of predictions from term contributions + constant
  • term subsetting behavior
  • validation/error cases

Notes on verification

• Changes are limited to R/predict.R, man/predict.merMod.Rd, and tests/testthat/test-predict.R.
• Local R test execution is not available in this sandbox environment because R/Rscript binaries are not installed.

Original prompt

---

This section details on the original issue you should resolve

<issue_title>type = "terms" for predict.merMod</issue_title>
<issue_description>there's no "terms" option for the type parameter of predict.merMod. Actually, it is currently unused; as shown in the predict.merMod.Rd file:

 \item{terms}{a \code{\link{terms}} object - unused at present.}
```</issue_description>

## Comments on the Issue (you are @copilot in this section)

<comments>
</comments>

• Fixes type = "terms" for predict.merMod #18

---

📍 Connect Copilot coding agent with Jira, Azure Boards or Linear to delegate work to Copilot in one click without leaving your project management tool.

[annahuynhly]

@copilot can you try again?

[annahuynhly]

@copilot could you please try again?

[Copilot]

@copilot can you try again?

Implemented and pushed a full retry for predict.merMod(type="terms") in commit 1f53169, including code, docs, and focused tests for term contributions/validation.

@copilot could you please try again?

Done — I retried and pushed the updated implementation in commit 1f53169 with type="terms" support in predict.merMod, documentation updates, and targeted tests.

[annahuynhly]

@copilot looking at the source code for predict.lm regarding the use of terms, this is what the code looks like:

if (type == "terms") {
    if (!mmDone) {
      mm <- model.matrix(object)
      mmDone <- TRUE
    }
    aa <- attr(mm, "assign")
    ll <- attr(tt, "term.labels")
    hasintercept <- attr(tt, "intercept") > 0L
    if (hasintercept) 
      ll <- c("(Intercept)", ll)
    aaa <- factor(aa, labels = ll)
    asgn <- split(order(aa), aaa)
    if (hasintercept) {
      asgn$"(Intercept)" <- NULL
      avx <- colMeans(mm)
      termsconst <- sum(avx[piv] * beta[piv])
    }
    nterms <- length(asgn)
    if (nterms > 0) {
      predictor <- matrix(ncol = nterms, nrow = NROW(X))
      dimnames(predictor) <- list(rownames(X), names(asgn))
      if (se.fit || interval != "none") {
        ip <- predictor
        Rinv <- qr.solve(qr.R(qr.lm(object))[p1, p1])
      }
      if (hasintercept) 
        X <- sweep(X, 2L, avx, check.margin = FALSE)
      unpiv <- rep.int(0L, NCOL(X))
      unpiv[piv] <- p1
      for (i in seq.int(1L, nterms, length.out = nterms)) {
        iipiv <- asgn[[i]]
        ii <- unpiv[iipiv]
        iipiv[ii == 0L] <- 0L
        predictor[, i] <- if (any(iipiv > 0L)) 
          X[, iipiv, drop = FALSE] %*% beta[iipiv]
        else 0
        if (se.fit || interval != "none") 
          ip[, i] <- if (any(iipiv > 0L)) 
            as.matrix(X[, iipiv, drop = FALSE] %*% Rinv[ii, 
              , drop = FALSE])^2 %*% rep.int(res.var, 
              p)
          else 0
      }
      if (!is.null(terms)) {
        predictor <- predictor[, terms, drop = FALSE]
        if (se.fit) 
          ip <- ip[, terms, drop = FALSE]
      }
    }
    else {
      predictor <- ip <- matrix(0, n, 0L)
    }
    attr(predictor, "constant") <- if (hasintercept) 
      termsconst
    else 0
  }

In general, your code suggestions look a lot different but they could fundamentally be the same thing. What do you think?

[Copilot]

@copilot looking at the source code for predict.lm regarding the use of terms, this is what the code looks like:

...

Yes — they’re fundamentally the same idea (decompose predictions by fixed-effect terms and carry a constant so terms can be summed back), but implemented differently for merMod internals. predict.lm uses QR/pivot/intercept-centering machinery specific to lm; our merMod version uses the fixed-effect model matrix (X) plus assign and then computes constant from the current prediction context. That makes it work with mixed-model features (re.form, offsets, etc.) while preserving reconstruction semantics. If you want strict predict.lm parity (including centered-term behavior), I can do a follow-up patch. Screenshot: N/A (no UI changes).