commit 44c4a7a7abc782e3e29462f9b6e4ad1faf18a00a
parent da836be5f37fbc90baee74fd260b650f77932a5a
Author: eamoncaddigan <eamon.caddigan@gmail.com>
Date: Fri, 21 Aug 2015 17:02:07 -0400
Comments and cleanup.
Diffstat:
2 files changed, 7 insertions(+), 13 deletions(-)
diff --git a/Jags-Yord-Xnom1grp-Mnormal.R b/Jags-Yord-Xnom1grp-Mnormal.R
@@ -72,17 +72,7 @@ genMCMC = function( datFrm, yName , qName,
# Write out modelString to a text file
writeLines( modelString , con="TEMPmodel.txt" )
#-----------------------------------------------------------------------------
- # INTIALIZE THE CHAINS.
- # # Initial values of MCMC chains based on data:
- # muInit = c( mean(y[x==1]) , mean(y[x==2]) )
- # sigmaInit = c( sd(y[x==1]) , sd(y[x==2]) )
- # threshInit = 1:(nYlevels-1)+0.5
- # threshInit[1] = NA
- # threshInit[nYlevels-1] = NA
- # # Regarding initial values in next line: (1) sigma will tend to be too big if
- # # the data have outliers, and (2) nu starts at 5 as a moderate value. These
- # # initial values keep the burn-in period moderate.
- # initsList = list( mu=muInit, sigma=sigmaInit, nuMinusOne=4, thresh=threshInit )
+ # This is where the chains would be initialized, but we'll just let JAGS do it
initsList = NULL
#-----------------------------------------------------------------------------
# RUN THE CHAINS
@@ -181,6 +171,7 @@ plotMCMC = function( codaSamples , datFrm , yName , qName, compVal , #RopeEff=NU
xlab=bquote(sigma) ,
main=paste("Std. Dev.") ,
col="skyblue" )
+
#-----------------------------------------------------------------------------
# 4. effect size.
effectSize = ( mu - compVal ) / sigma
diff --git a/antivax-attitudes.R b/antivax-attitudes.R
@@ -43,16 +43,19 @@ questionnaireData <- expData.clean %>%
# "tidy" the data
gather("question", "response", -subject_number, -intervention) %>%
separate(question, c("interval", "question"), sep = "\\.") %>%
- mutate(interval = factor(interval, c("pretest", "posttest"), ordered = TRUE))
+ mutate(interval = factor(interval, c("pretest", "posttest"), ordered = TRUE),
+ question = factor(question))
# Some plots --------------------------------------------------------------
+# Check out the distribution of responses before and after the intervention
p1 <- ggplot(questionnaireData, aes(x = question, y = response, fill = interval)) +
geom_violin() +
facet_grid(intervention ~ .)
print(p1)
+# Look at each subject's change for each question
p2 <- ggplot(questionnaireData, aes(x = interval, y = response, group = subject_number)) +
geom_line(alpha = 0.2, position = position_jitter(w = 0.15, h = 0.15)) +
facet_grid(intervention ~ question)
@@ -61,7 +64,7 @@ print(p2)
# Bayesian analysis of survey data ----------------------------------------
-# For now, we'll just fit one model to all the questions pre-test.
+# Fit a model to each question using pre-intervention data.
modelData <- filter(questionnaireData, interval == "pretest")
source("Jags-Yord-Xnom1grp-Mnormal.R")
