Predictions from a Fitted SIR Model

Generates predictions from a fitted SIR model for the training data or for new data. Predictions can be on the link scale (linear predictor) or the response scale (expected counts, probabilities, or means).

Usage

# S3 method for class 'sir'
predict(object, newdata = NULL, type = c("response", "link"), ...)

Arguments

object: A fitted sir object from sir.
newdata: Optional named list with components W (3D or 4D array), X (3D array), and/or Z (3D or 4D array) for counterfactual prediction. Dimensions must match the original fit. Any component not supplied is taken from the original fit. If NULL (default), returns predictions for the training data. Note: unlike many R predict methods, newdata is a list of arrays, not a data frame.
type: Character string: "link" for linear predictor (eta) or "response" for expected values on the original scale. Default is "response".
...: Additional arguments (unused).

Value

An array (n1 x n2 x T) of predicted values on the requested scale.

Details

For scenario (counterfactual) analysis, supply modified arrays in newdata. For example, to see how the network would change if a covariate increased by one unit, pass the modified Z array while keeping W and X from the original fit.

Examples

if (FALSE) { # \dontrun{
model <- sir(Y, W, X, Z = Z, family = "poisson")

# In-sample fitted values
pred <- predict(model)

# Scenario: what if Z increases by 1 unit?
Z_shift <- Z + 1
pred_scenario <- predict(model, newdata = list(Z = Z_shift))

# Compare mean predictions
mean(pred, na.rm = TRUE)
mean(pred_scenario, na.rm = TRUE)
} # }