README.Rmd

---
output: github_document
bibliography: "inst/ma.bib"
csl: "inst/apa.csl"
---

```{r setup, include = FALSE, warning = FALSE, comment = FALSE}
#library(poLCAExtra)
devtools::load_all()
set.seed(42)
```

# poLCAExtra : New and convenient functions to improve workflow of `poLCA`

The package `poLCAExtra` offers convenient functions to improve the workflow of [`poLCA`](https://github.com/dlinzer/poLCA)[@poLCA]

```{r github, echo = TRUE, eval = FALSE}
# Current development version
remotes::install_github(repo = "quantmeth/poLCAExtra")
```

# Comparing many latent class models

Either by using many latent class models like

```{r LCA, warning = FALSE, comment = FALSE, message = FALSE, results = FALSE}
# poLCAExtra has two data sets examples
jd <- ex1.poLCA

# Formula
f1 <- as.formula(cbind(V1, V2, V3, V4, V5, V6) ~ 1)

# Four latent classes analysis of increasing classes
LCA1 <- poLCA(f1, data = jd, nclass = 1, verbose = FALSE) 
LCA2 <- poLCA(f1, data = jd, nclass = 2, verbose = FALSE)
LCA3 <- poLCA(f1, data = jd, nclass = 3, verbose = FALSE)
LCA4 <- poLCA(f1, data = jd, nclass = 4, maxiter = 1000, nrep = 10, verbose = FALSE)
```

which can then be compared with

```{r CompareLCA}
anova(LCA1, LCA2, LCA3, LCA4)
```

The function readily gathers some relevant statitics to choose the number of classes.

This new version of `poLCA` can also handle many number of classes and yield the same output.

```{r, warning = FALSE, comment = FALSE, message = FALSE, results = FALSE}
LCAE <- poLCA(f1, data = jd, nclass = 1:4)
LCAE
```

Both previous tables can be easily exported.

The original `plot()` (from `poLCA`) function has been updated to account for the multiple LCA.

```{r plotLCA}
plot(LCAE, nclass = 3)
```

There is also a new plot that works with a single LCA output.

```{r}
poLCA.plot(LCA3)
```

Or all the LCA output, which may help to compare diffent number of classes.

```{r}
poLCA.plot(LCAE)
# If only a subset is required
# poLCA.plot(LCAE, nclass = 2:3)
# poLCA.plot(LCA2, LCA3)
```

# New features

## New indices

The `SABIC`, `CAIC`, and `AIC3` statistics have been added to `poLCA`.

```{r sabic}
LCA3$sabic
LCA3$caic
LCA3$aic3
```

The relative entropy, which is more often requested than the entropy, has been added.

```{r re}
poLCA.relentropy(LCA3)
```
The Lo-Mendell-Rubin adjusted likelihood ratio test and the Vuong-Lo-Mendell-Rubin likelihood ratio test have been added.

```{r lrt}
poLCA.lrt(LCA3)
```

The bootstrapped Vuong-Lo-Mendell-Rubin likelihood ratio test which is more generally recommended is also added. Note that the bootstrapped tests are quite time-consuming.

```{r tech14, cache = TRUE}
poLCA.blrt(LCAE)
# poLCA.blrt(LCA3)
```

# New functions

## A `predict()` function

Predicted probabilities and classes

```{r predict}
head(round(predict(LCA3), 3))
# head(round(predict(LCAE, nclass = 3),3))
```

## Two functions to verify LCA assumptions

Analysis of residual (also known as Tech10 in Mplus) to investigate local independence.

```{r tech10}
poLCA.residual.pattern(LCAE, nclass = 3)
# poLCA.tech10(LCA3)
```

Another one to inspect the covariance matrice.

```{r rescov}
poLCA.cov(LCAE, nclass = 3)
# poLCA.cov(LCA3)
```

<!-- ## A summary -->


## Bootstrap 3-step approach

The 3-step approaches will be improved.

Here are two examples that can be readily be implemented.

```{r 3step, cache = TRUE}
d3 <- d3step("categorical", LCAE, nclass = 3)
# d3step("categorical", LCA3)
d3

r3 <- r3step("continuous", LCAE, nclass = 3)
# r3step("continuous", LCA3)
r3
```

d3step plot

```{r d3stepplot}
plot(d3, ci = c(.05,.95))
```
r3step plot

```{r r3stepplot}
plot(r3, ci = c(.05, .95))
```

# How to cite

Caron, P.-O. (2024). *poLCAExtra : New and Convenient Functions for the Package 'poLCA'*. https://github.com/quantmeth/poLCAExtra

# References