broom (version 0.4.3)

poLCA_tidiers: Tidiers for poLCA objects

Description

Tidiers for poLCA latent class regression models. Summarize the probabilities of each outcome for each variable within each class with tidy, add predictions to the data with augment, or find the log-likelihood/AIC/BIC with glance.

Usage

# S3 method for poLCA
tidy(x, ...)
# S3 method for poLCA
augment(x, data, ...)
# S3 method for poLCA
glance(x, ...)

Arguments

x

A poLCA object

...

Extra arguments, not used

data

For augment, the original dataset used to fit the latent class model. If not given, uses manifest variables in x$y and, if applicable, covariates in x$x

Value

All tidying methods return a data.frame without rownames, whose structure depends on the method chosen.

tidy returns a data frame with one row per variable-class-outcome combination, with columns:

variable

Manifest variable

class

Latent class ID, an integer

outcome

Outcome of manifest variable

estimate

Estimated class-conditional response probability

std.error

Standard error of estimated probability

augment returns a data frame with one row for each original observation, augmented with the following columns:

.class

Predicted class, using modal assignment

.probability

Posterior probability of predicted class

If the data argument is given, those columns are included in the output (only rows for which predictions could be made). Otherwise, the y element of the poLCA object, which contains the manifest variables used to fit the model, are used, along with any covariates, if present, in x.

Note that while the probability of all the classes (not just the predicted modal class) can be found in the posterior element, these are not included in the augmented output, since it would result in potentially many additional columns, which augment tends to avoid.

glance returns a one-row data frame with the following columns:

logLik

the data's log-likelihood under the model

AIC

the Akaike Information Criterion

BIC

the Bayesian Information Criterion

g.squared

The likelihood ratio/deviance statistic

chi.squared

The Pearson Chi-Square goodness of fit statistic for multiway tables

df

Number of parameters estimated, and therefore degrees of freedom used

df.residual

Number of residual degrees of freedom left

Examples

# NOT RUN {
if (require("poLCA", quietly = TRUE)) {
  library(poLCA)
  library(dplyr)
  
  data(values)
  f <- cbind(A, B, C, D)~1
  M1 <- poLCA(f, values, nclass = 2, verbose = FALSE)
  
  M1
  tidy(M1)
  head(augment(M1))
  glance(M1)
  
  library(ggplot2)
  
  ggplot(tidy(M1), aes(factor(class), estimate, fill = factor(outcome))) +
    geom_bar(stat = "identity", width = 1) +
    facet_wrap(~ variable)
  
  set.seed(2016)
  # compare multiple
  mods <- data_frame(nclass = 1:3) %>%
    group_by(nclass) %>%
    do(mod = poLCA(f, values, nclass = .$nclass, verbose = FALSE))
  
  # compare log-likelihood and/or AIC, BIC
  mods %>%
    glance(mod)
  
  ## Three-class model with a single covariate.
  
  data(election)
  f2a <- cbind(MORALG,CARESG,KNOWG,LEADG,DISHONG,INTELG,
               MORALB,CARESB,KNOWB,LEADB,DISHONB,INTELB)~PARTY
  nes2a <- poLCA(f2a, election, nclass = 3, nrep = 5, verbose = FALSE)
  
  td <- tidy(nes2a)
  head(td)
  
  # show 
  
  ggplot(td, aes(outcome, estimate, color = factor(class), group = class)) +
    geom_line() +
    facet_wrap(~ variable, nrow = 2) +
    theme(axis.text.x = element_text(angle = 90, hjust = 1))
  
  au <- augment(nes2a)
  head(au)
  au %>%
    count(.class)
  
  # if the original data is provided, it leads to NAs in new columns
  # for rows that weren't predicted
  au2 <- augment(nes2a, data = election)
  head(au2)
  dim(au2)
}

# }