Computes occasion-level choice probabilities over the retained posterior draws and predicts the alternative with the largest mean probability.
Usage
# S3 method for class 'RprobitB_fit'
predict(
object,
newdata = NULL,
type = c("population", "conditional"),
uncertainty = FALSE,
level = 0.95,
ghk_draws = 500L,
progress = interactive(),
...
)Arguments
- object
[
RprobitB_fit]
Fitted choice model.- newdata
[
data.frame|NULL]
Data for prediction.NULLuses the fitted data. A response column is optional. In wide format, missing decider and occasion identifiers make every row a choice occasion of its own decider.- type
[
character(1)]
Which coefficients to predict with:"population"integrates over the estimated population distribution of the random coefficients and latent classes and applies to any decider."conditional"uses the posterior random coefficients and class allocations of the deciders that were observed when fitting the model, whichnewdatamust then name.
- uncertainty
[
logical(1)]
Add posterior standard deviations and credible intervals?- level
[
numeric(1)]
Probability of the credible intervals.- ghk_draws
[
integer(1)]
Number of draws of the GHK simulator for multivariate normal probabilities of more than three dimensions, seeoeli::pmvnorm().- progress
[
logical(1)]
Show progress?- ...
Currently not used.
Value
A data.frame with one row per choice occasion, identifier columns,
.prediction, and one probability_* column per alternative. If
uncertainty = TRUE, sd_*, lower_*, and upper_* columns are added.
Details
Conditional prediction is based on the individual-level parameters (Train,
2009, Chapters 11 and 12): the posterior distribution of a decider's
coefficients given their observed choices, which the Gibbs sampler
provides as draws when the model is fitted with
save_individual_draws = TRUE.
References
Train KE (2009). Discrete Choice Methods with Simulation, 2 edition. Cambridge University Press, Cambridge. doi:10.1017/CBO9780511805271 .
Examples
set.seed(1)
model <- fit(
choice ~ x | 0,
random_effects = "x",
dgp_parameters = list(beta = c(x = 1), Omega = matrix(0.5)),
n_deciders = 20,
iterations = 300,
warmup = 150,
chains = 1,
save_individual_draws = TRUE
)
head(predict(model))
#> deciderID .prediction probability_A probability_B
#> 1 1 B 0.2339449 0.7660551
#> 2 2 B 0.1165867 0.8834133
#> 3 3 A 0.8158130 0.1841870
#> 4 4 B 0.3933205 0.6066795
#> 5 5 A 0.7786822 0.2213178
#> 6 6 A 0.8126050 0.1873950
head(predict(model, type = "conditional"))
#> deciderID .prediction probability_A probability_B
#> 1 1 B 0.33861304 0.6613870
#> 2 2 B 0.04162912 0.9583709
#> 3 3 A 0.87322328 0.1267767
#> 4 4 B 0.38083540 0.6191646
#> 5 5 A 0.83490952 0.1650905
#> 6 6 A 0.87691676 0.1230832
head(predict(model, uncertainty = TRUE))
#> deciderID .prediction probability_A probability_B sd_A sd_B
#> 1 1 B 0.2339449 0.7660551 0.09173230 0.09173230
#> 2 2 B 0.1165867 0.8834133 0.10253290 0.10253290
#> 3 3 A 0.8158130 0.1841870 0.09909139 0.09909139
#> 4 4 B 0.3933205 0.6066795 0.04492932 0.04492932
#> 5 5 A 0.7786822 0.2213178 0.09398348 0.09398348
#> 6 6 A 0.8126050 0.1873950 0.09874129 0.09874129
#> lower_A lower_B upper_A upper_B
#> 1 0.077389421 0.57827805 0.4217219 0.9226106
#> 2 0.003576592 0.61116874 0.3888313 0.9964234
#> 3 0.591543731 0.03369682 0.9663032 0.4084563
#> 4 0.301614977 0.52965708 0.4703429 0.6983850
#> 5 0.581647759 0.06451891 0.9354811 0.4183522
#> 6 0.590683220 0.03593066 0.9640693 0.4093168
### new choice occasions
new_data <- data.frame(deciderID = 21:22, x_A = c(1, -1), x_B = c(0, 0))
predict(model, newdata = new_data)
#> deciderID .prediction probability_A probability_B
#> 1 21 A 0.796969 0.203031
#> 2 22 B 0.203031 0.796969