This function classifies the deciders based on their allocation to the components of the mixing distribution.

## Usage

classification(x, add_true = FALSE)

## Arguments

x

An object of class RprobitB_fit.

Set to TRUE to add true class memberships to output (if available).

## Value

A data frame. The row names are the decider ids. The first C columns contain the relative frequencies with which the deciders are allocated to the C classes. Next, the column est contains the estimated class of the decider based on the highest allocation frequency. If add_true, the next column true contains the true class memberships.

## Details

The function can only be used if the model has at least one random effect (i.e. P_r >= 1) and at least two latent classes (i.e. C >= 2).

In that case, let $$z_1,\dots,z_N$$ denote the class allocations of the $$N$$ deciders based on their estimated mixed coefficients $$\beta = (\beta_1,\dots,\beta_N)$$. Independently for each decider $$n$$, the conditional probability $$\Pr(z_n = c \mid s,\beta_n,b,\Omega)$$ of having $$\beta_n$$ allocated to class $$c$$ for $$c=1,\dots,C$$ depends on the class allocation vector $$s$$, the class means $$b=(b_c)_c$$ and the class covariance matrices $$Omega=(Omega_c)_c$$ and is proportional to $$s_c \phi(\beta_n \mid b_c,Omega_c).$$

This function displays the relative frequencies of which each decider was allocated to the classes during the Gibbs sampling. Only the thinned samples after the burn-in period are considered.

update_z() for the updating function of the class allocation vector.