predict.ace: Prediction of a fitted Additive Causal Expansion model
In mazphilip/AdditiveCausalExpansion: Additive Causal Expansion Model

Predict the the response surface and the marginal response with respect to the treatment using a fitted "ace" model. In case of binary treatments, the marginal response surface is the (heterogeneous) treatment effect. Optionally, the average treatment effect that takes into account all covariances of the posterior can be returned.

1
2
3

## S3 method for class 'ace'
predict(object, newX, newZ, marginal = FALSE,
  return_average_treatments = FALSE, normalize = TRUE, ...)

`object`	An "ace" object returned from the ace.train function.
`newX`	(Optional) A matrix with new data. Using the training data if missing.
`newZ`	(Optional) A vector, matrix or scalar with new treatment data. If it is a scalar, it predicts using the same value for all observations. If missing, it uses the training data.
`marginal`	Logical flag that determines whether to predict the response surface (FALSE) or the marginal response surface/heterogeneous treatment effect (TRUE) (default: FALSE).
`return_average_treatments`	Logical flag that determines whether treatment effect averages (ATE, ATT, ATU) are returned (default: False). Ignored if `marginal + FALSE`
`normalize`	For internal use (default: True). A logical scalar that determines whether new data is normalized using training moments.
`...`	Ignored.

For the estimate of the response surface (marginal = FALSE) or marginal response (marginal = TRUE), the method returns a list with

Maximum a priori (map) estimate (here: mean)
95 percent credible interval (ci) for each prediction point
Posterior variance (var) for each prediction point
If marginal = TRUE and return_average_treatments = TRUE:
- (Sample) Average Treatment Effect: ate with map, ci, and var
- (Sample) Average Treatment effect of the Treated: att with map, ci, and var
- (Sample) Average Treatment effect of the Untreated: atu with map, ci, and var