Bugs code for Categorical response
Usage
categorical_bugs(
nodename,
nodesCatIdx,
parentnames,
nodesintercepts,
parentcoefs
)
categorical_bugsGroup(
nodename,
nodesCatIdx,
nodesintercepts,
parentnames,
parentcoefs,
sigma,
sigma_alpha
)Arguments
- nodename
character string of response variable name.
- nodesCatIdx
integer vector of length \(|K-1|\) and starting at \(k+1\) (see Examples).
- parentnames
single character string (for one parent) or vector of characters (for multiple parent nodes) with parent node (predictor variables) names.
- nodesintercepts
overall mean of response. Parameter from fixed-effects intercept.
- parentcoefs
overall slope for each predictor (parent node) variable (fixed-effects).
- sigma
within-group variance. Parameter from random-effects residual.
- sigma_alpha
between-group variance-covariance matrix. Parameters from random-effects intercept.
Details
The output of fitAbn with method = "mle" is based on
the output of logistic regression models fit with either lm,
glm, glmer, multinom,
mblogit or internal irls methods.
They all use the first factor level as reference level.
Therefore, nodesCatIdx starts with index \(2\) and not \(1\).
nodesintercepts and parentcoefs refer to the values of
(Intercept) and Estimate of the respective model output.
Predictor names build the keys in parentcoef.
Examples
# A -> B
# Where B is a categorical variable with 4 levels.
categorical_bugs(nodename = "b",
nodesCatIdx = c(2, 3, 4),
parentnames = "a",
nodesintercepts = c(2.188650, 3.133928, 3.138531),
parentcoefs = list("a"=c(a=1.686432, a=3.134161, a=5.052104)))
#> b ~ dcat(p.b) # Categorical response
#> p.b[1] <- phi.b[1]/sum(phi.b) # soft-max
#> log(phi.b[1]) <- 0 # Reference category
#> p.b[2] <- phi.b[2]/sum(phi.b) # soft-max
#> log(phi.b[2]) <- 2.18865 + 1.686432*a
#> p.b[3] <- phi.b[3]/sum(phi.b) # soft-max
#> log(phi.b[3]) <- 3.133928 + 3.134161*a
#> p.b[4] <- phi.b[4]/sum(phi.b) # soft-max
#> log(phi.b[4]) <- 3.138531 + 5.052104*a