Multi-source immunisation coverage model with Stan

ic_fit(
  X,
  prior_lambda = NULL,
  prior_sigma = NULL,
  upper_sigma = NULL,
  lower_sigma = NULL,
  region = TRUE,
  verbose = TRUE,
  ...
)

# S3 method for ic.df
ic_fit(
  X,
  prior_lambda = NULL,
  prior_sigma = NULL,
  upper_sigma = NULL,
  lower_sigma = NULL,
  region = TRUE,
  verbose = TRUE,
  ...
)

# S3 method for ic.df
ic_fit_single(X, region = TRUE, verbose = TRUE, ...)

Arguments

X	Object of `ic.df` for analysis
prior_lambda	Scale parameter for normal prior on source-specific intercepts. See details.
prior_sigma	Scale parameter for the truncated Cauchy prior on source-specific standard deviations. See details.
upper_sigma	Numeric value (or vector) setting the upper bounds on the source-specific scale parameter. See details.
lower_sigma	Numeric value (or vector) setting the lower bounds on the source-specific scale parameter. See details.
region	Logical. Should region-specific models be generated? Default is `TRUE`.
verbose	Logical. Should messages be displayed? Default is `TRUE`.
...	Arguments passed to `rstan::sampling` (e.g. iter, chains).

Value

An object of class icfit for single region models, or iclist for multiple regions. The icfit object contains a stanfit object returned by rstan::sampling with the fitted object along with the posterior samples, data used in the model, and other attributes related to the fitting. An iclist object is an extension of a list containing multiple icfit objects.

Details

The default priors for the source-specific intercepts (lambda) and standard deviations (sigma) are given by normal and truncated Cauchy distributions, respectively. Future version may allow users to specify different distributions. Currently, users may only specify the scale parameter of these distribution. By default, lambda for all sources is given a value of 0.5, i.e. a prior distribution of normal(0, 0.5).

For sigma, the default prior varies by source. While administrative and official data are set to a value of 2, survey data is restricted to 0.2, reflecting a prior belief in more accurate measurements. These parameters are used in a half-Cauchy distribution (e.g. cauchy(0, 0.2)).

In addition, the upper and lower bounds of the sigma parameter may be set to define the range of the possible values. In general, the lower bounds should always be zero and cannot be negative. Similar to the differences in the prior distributions, an upper bound of 0.4 is placed on 'survey' estimates. In the absence of a user-defined upper-bound, these are set to 100.

Users setting prior_lambda, prior_sigma, or upper_sigma should note that the length of values specified must be 1 or match the number of unique sources in X. When only one value is present, it will be applied to all sources. When a vector is provided, the order of values must match the order of sources as given by list_sources(X).