CausalQueries (version 1.2.1)

update_model: Fit causal model using 'stan'

Description

Takes a model and data and returns a model object with data attached and a posterior model

Usage

update_model(
  model,
  data = NULL,
  data_type = NULL,
  keep_type_distribution = TRUE,
  keep_event_probabilities = FALSE,
  keep_fit = FALSE,
  censored_types = NULL,
  ...
)

Value

An object of class causal_model. The returned model is a list containing the elements comprising a model (e.g. 'statement', 'nodal_types' and 'DAG') with the

posterior_distribution returned by stan

attached to it.

Arguments

model

A causal_model. A model object generated by make_model.

data

A data.frame. Data of nodes that can take three values: 0, 1, and NA. In long form as generated by make_events

data_type

Either 'long' (as made by make_data) or 'compact' (as made by collapse_data). Compact data must have entries for each member of each strategy family to produce a valid simplex. When long form data is provided with missingness, missing data is assumed to be missing at random.

keep_type_distribution

Logical. Whether to keep the (transformed) distribution of the causal types. Defaults to `TRUE`

keep_event_probabilities

Logical. Whether to keep the (transformed) distribution of event probabilities. Defaults to `FALSE`

keep_fit

Logical. Whether to keep the stanfit object produced by sampling for further inspection. See ?stanfit for more details. Defaults to `FALSE`. Note the stanfit object has internal names for parameters (lambda), event probabilities (w), and the type distribution (types)

censored_types

vector of data types that are selected out of the data, e.g. c("X0Y0")

...

Options passed onto sampling call. For details see ?rstan::sampling

See Also

make_model allows to create new model, summary.causal_model provides summary method for output objects of class causal_model

Examples

 model <- make_model('X->Y')
 data_long   <- make_data(model, n = 4)
 data_short  <- collapse_data(data_long, model)
 model <-  update_model(model, data_long)
 model <-  update_model(model, data_short)

   # It is possible to implement updating without data, in which
   # case the posterior is a stan object that reflects the prior

   update_model(model)

 if (FALSE) {

   # Censored data types illustrations
   # Here we update less than we might because we are aware of filtered data

   data <- data.frame(X=rep(0:1, 10), Y=rep(0:1,10))
   uncensored <-
     make_model("X->Y") |>
     update_model(data) |>
     query_model(te("X", "Y"), using = "posteriors")

   censored <-
     make_model("X->Y") |>
     update_model(
       data,
       censored_types = c("X1Y0")) |>
     query_model(te("X", "Y"), using = "posteriors")


   # Censored data: We learn nothing because the data
   # we see is the only data we could ever see
   make_model("X->Y") |>
     update_model(
       data,
       censored_types = c("X1Y0", "X0Y0", "X0Y1")) |>
     query_model(te("X", "Y"), using = "posteriors")
 }