In rosiezou/440proj: Analyze Continuous Latent Variables From Ordinal Survey Responses

knitr::opts_chunk$set(echo = TRUE)

Visual Tests

Here, we'll perform some visual tests allowing the user to ensure that stat440pkg is returning sane results.

gen.imp.resp

Let's make sure gen.imp.resp is returning somewhat normally-distributed data.

library(stat440pkg)
library(tidyr)
library(ggplot2)

imp.resp <- gen.imp.resp(data = multiis, num.iter = 5)
gathered.data <- gather(imp.resp)
p <- ggplot(gathered.data) + 
  geom_histogram(aes(x = value), binwidth = 0.25) + 
  facet_wrap(~ key)

plot(p)

gen.latent.vars

Let's make sure gen.latent.vars is returning somewhat normally-distributed data. First we'll do so for Bartlett scores, then Thompson regression scores.

Bartlett factor scores

grp.indicator <- sapply(names(multiis), FUN =
                            function(x){strsplit(x, split = "_")[[1]][2]})

lv <- gen.latent.vars(data = multiis, grp.indicator = grp.indicator, num.iter = 5, scores = "Bartlett")

gathered.data <- gather(lv)
p <- ggplot(gathered.data) + 
  geom_histogram(aes(x = value), binwidth = 0.25) + 
  facet_wrap(~ key)

plot(p)

Thompson factor scores

grp.indicator <- sapply(names(multiis), FUN =
                            function(x){strsplit(x, split = "_")[[1]][2]})

lv <- gen.latent.vars(data = multiis, grp.indicator = grp.indicator, num.iter = 5, scores = "regression")

gathered.data <- gather(lv)
p <- ggplot(gathered.data) + 
  geom_histogram(aes(x = value), binwidth = 0.2) + 
  facet_wrap(~ key)

plot(p)

Results using Thompson scores for latent variables

Here, we'll create the similar plots to those that appear in the Results section of the report, but using Thompson scores.

M <- 50
latent.datasets <- gen.latent.datasets(M, multiis, grp.indicator = grp.indicator, num.iter = 5, scores = "regression")

pooled.add1 <- pool.analyses(latent.datasets, cat~comp + int, lm)
pooled.add2 <- pool.analyses(latent.datasets, comp~cat + int, lm)
pooled.add3 <- pool.analyses(latent.datasets, int~comp + cat, lm)

signif(pooled.add1$hypothesis.test, digits = 3)
signif(pooled.add2$hypothesis.test, digits = 3)
signif(pooled.add3$hypothesis.test, digits = 3)

library(scatterplot3d)
add <- function(x) Reduce("+", x)
averaged <- add(latent.datasets)/M
fit <- lm(int~comp + cat, data = averaged)
scplot <- scatterplot3d(averaged$comp, averaged$cat, averaged$int,
              main="3D Scatterplot of Latent Variables\n with Regression Plane for Int ~  Comp + Cat", angle = 120,
              xlab = "compartmentalization", ylab = "categorization", zlab = "integration",
              col.grid = "lightgrey", pch = 19, color = "lightblue")

scplot$plane3d(fit, lty = "dotted")
orig <- scplot$xyz.convert(averaged$comp, averaged$cat, averaged$int)
plane <- scplot$xyz.convert(averaged$comp, averaged$cat, fitted(fit))
i.negpos <- 1 + (resid(fit) > 0)
segments(orig$x, orig$y, plane$x, plane$y,
         col = c("blue", "red")[i.negpos], lty = (2:1)[i.negpos])


# ggplot2 pairs plot
library(ggplot2)
library(GGally)

ggpairs(averaged)

rosiezou/440proj documentation built on May 12, 2019, 6:25 p.m.