1 ## @knitr Options, include=F, cache=F, message=F
   2 opts_chunk$set(size='footnotesize', eval=T, cache=T, prompt=F, message=F, tidy=F)
   3 
   4 
   5 ## @knitr LinearityExampleVisualization, fig.show='hold'
   6 library(languageR)
   7 data(lexdec)
   8 
   9 g = glm(RT ~ I(Frequency^2) + I(Frequency^3), data=lexdec)
  10 summary(g)$coefficients
  11 
  12 library(ggplot2)
  13 ggplot(lexdec, aes(x=I(Frequency^2), y=RT)) +
  14   geom_point() +
  15   geom_abline(
  16     intercept = coef(g)['(Intercept)'], 
  17     slope = coef(g)['I(Frequency^2)'],
  18     color = "blue")
  19 
  20 ggplot(lexdec, aes(x=I(Frequency^3), y=RT)) +
  21   geom_point() +
  22   geom_abline(
  23     intercept = coef(g)['(Intercept)'], 
  24     slope = coef(g)['I(Frequency^3)'],
  25     color = "red")
  26 
  27 
  28 ## @knitr LinearityExampleVisualization3, echo=FALSE, out.width='8cm', out.height='8cm'
  29 p = function(x) {
  30   y = coef(g)['(Intercept)'] + 
  31     coef(g)['I(Frequency^2)'] * x^2 +
  32     coef(g)['I(Frequency^3)'] * x^3 
  33   return(y)
  34 }
  35 ggplot(lexdec, aes(x=Frequency, y=RT)) +  
  36   geom_point() +
  37   geom_smooth(method='lm', fill=NA, color="black", linetype="dashed") +
  38   stat_function(fun= p, geom="line", color = "green", size=2) +
  39   stat_function(fun= function(x) {coef(g)['(Intercept)'] + 
  40     coef(g)['I(Frequency^2)'] * x^2}, geom="line", color = "blue") +
  41   stat_function(fun= function(x) {coef(g)['(Intercept)'] + 
  42     coef(g)['I(Frequency^3)'] * x^3}, geom="line", color = "red") +
  43   annotate("text", label = "x^2", x=7, y=5.9, color = "blue", size=10) +
  44   annotate("text", label = "x^3", x=7, y=7, color = "red", size=10) 
  45 
  46 
  47 ## @knitr FrequencyTrialModel, echo=T, results='hide', size='scriptsize'
  48 summary(glm(RT ~ Frequency + Trial, data= lexdec))
  49 
  50 
  51 ## @knitr Centering, echo=T, size='scriptsize'
  52 lexdec$cFrequency = lexdec$Frequency - mean(lexdec$Frequency)
  53 summary(lexdec[,c('Frequency','cFrequency')])
  54 
  55 
  56 ## @knitr FrequencyTrialModel2, echo=T, size='scriptsize'
  57 summary(glm(RT ~ Frequency * Trial, data= lexdec))$coefficients
  58 
  59 
  60 ## @knitr Predict, echo=T, eval=F, results='hide', size='scriptsize'
  61 # Beyond the coefficients, another interesting thing to compare is 
  62 # the models predictions (the predicted RTs)
  63 g1 =  glm(RT ~ Frequency * Trial, data= lexdec)
  64 g2 =  glm(RT ~ cFrequency * cTrial, data= lexdec)
  65 
  66 # if the models make the same prediction the difference in their 
  67 # predictions should be zero
  68 predict(g1) - predict(g2)
  69 
  70 
  71 ## @knitr NativeSexModel, echo=T, results='hide', size='scriptsize'
  72 summary(glm(RT ~ NativeLanguage * Sex, data= lexdec))$coefficients
  73 
  74 
  75 ## @knitr ReorderingFactorLevels, echo=T, size='scriptsize'
  76 # by default R assumes alphanumeric ordering of factor levels
  77 # but we can always change that by explicitly stating what
  78 # order we would like:
  79 lexdec$NativeLanguage = 
  80   factor(lexdec$NativeLanguage, levels=c('Other','English'))
  81 
  82 # Btw, c() is simply referring to a vector in R. They can 
  83 # contain numbers, strings, or any (combination) of other 
  84 # objects, e.g. c(4, "seven", "Hans the Wurst", c(log(2), "hai"))
  85 
  86 
  87 ## @knitr AnovaCoding, echo=T, size='scriptsize'
  88 # confirm that the default contrast was treatment-coding, e.g.
  89 contrasts(lexdec$Sex)
  90 
  91 # change contrasts
  92 contrasts(lexdec$Sex) = contr.sum(2)
  93 contrasts(lexdec$Sex)
  94 # notice that default sum-coding in R assigns 1 to the 
  95 # alphanumerically first level and -1 to the second level
  96 
  97 # let's label the contrast so that we don't get confused 
  98 # (cbind() makes a matrix out of multiple vectors; in this 
  99 # case 1 vector, but it allows us to label the columns/vectors)
 100 contrasts(lexdec$Sex) = cbind("Male" = c(-1, 1))
 101 
 102 
 103 ## @knitr AnovaCoding2, echo=T, size='scriptsize'
 104 contrasts(lexdec$Sex) = contr.sum(2) / 2
 105 contrasts(lexdec$Sex)
 106 
 107 contrasts(lexdec$NativeLanguage) = contr.sum(2) / 2
 108 contrasts(lexdec$NativeLanguage)