predict.seminr_model: Predict method for SEMinR PLS models

In seminr: Building and Estimating Structural Equation Models

Predict method for SEMinR PLS models

Description

Generates out-of-sample predictions for a PLS model estimated by estimate_pls(). Supports models with and without interaction terms. For interaction models, the prediction method is automatically detected from the measurement model specification:

Usage

## S3 method for class 'seminr_model'
predict(
  object,
  testData,
  technique = predict_DA,
  na.print = ".",
  digits = 3,
  ...
)

Arguments

object	An estimated seminr_model from estimate_pls().
testData	A data.frame of held-out test data containing all indicator columns. Must not include interaction columns (these are recreated internally).
technique	The prediction technique: predict_DA (Direct Antecedents, default) or predict_EA (Earliest Antecedents).
na.print	Character string for printing NA values.
digits	Number of digits for printing.
...	Additional arguments (currently unused).

Details

• two_stage: Recreates interaction from construct-score products

• product_indicator: Recreates scaled item-level products from test data

• orthogonal: Recreates scaled products and applies stored orthogonalization coefficients from estimation

Higher-order construct (HOC) models are not currently supported for prediction. Models with mixed interaction methods (e.g., one two_stage and one product_indicator) will produce an error.

Value

A predicted_seminr_model object containing:

testData	The test data (non-interaction items only).
predicted_items	Predicted indicator scores.
item_residuals	Residuals (actual - predicted) for each indicator.
predicted_composite_scores	Predicted construct scores.
composite_residuals	Residuals for construct scores.
actual_star	Reference construct scores from re-estimation on combined data.

Examples

data(mobi)

mobi_mm <- constructs(
  composite("Image",        multi_items("IMAG", 1:5)),
  composite("Expectation",  multi_items("CUEX", 1:3)),
  composite("Satisfaction", multi_items("CUSA", 1:3)),
  interaction_term(iv = "Image", moderator = "Expectation",
                   method = product_indicator)
)
mobi_sm <- relationships(
  paths(to = "Satisfaction",
        from = c("Image", "Expectation", "Image*Expectation"))
)
model <- estimate_pls(mobi, mobi_mm, mobi_sm)
predictions <- predict(model, testData = mobi[1:20, ])

seminr documentation built on May 25, 2026, 9:06 a.m.