2×2 repeated measures ANOVAs are common in my work. For that, I now created a flexible function in R. The function outputs assumption checks (outliers and normality), interaction and main effect results, pairwise comparisons, and produces a result plot with within-subject error bars (SD, SE or 95% CI) and significance stars added to the plot.

The function rm_2by2_anova.R and its accompanying notebook is available here
https://github.com/dcdace/R_functions

Below is the HTML version of the notebook. In the notebook, I test the function on three different datasets and also provide a step-by-step walk through the function elements.

Having spent many hours on this I hope the function will be useful and time-saving also to others.

Please let me know if you find any statistics or other errors!

Notebook_rm_2by2_ANOVA

1 Example 1: Using rm_2by2_anova function
2 Example 2: Step-by-step example
3 Example 3: Test on another dataset
4 Example 4: Test on another dataset
5 Control what to display

2x2 Repeated Measures ANOVA¶

In [1]:

# Dace Apšvalka, 2020, www.dcdace.net
# ----------------------------------------------------------------------
# LIBRARIES
# ----------------------------------------------------------------------
shhh <- suppressPackageStartupMessages # It's a library, so shhh!

shhh(suppressWarnings(library(psycho)))    # to get the Emotion dataset for the example

Example 1: Using rm_2by2_anova function¶

Load the function¶

In [37]:

source("https://raw.githubusercontent.com/dcdace/R_functions/main/RM-2by2-ANOVA/rm_2by2_anova.R")

Load example data¶

In [3]:

# ----------------------------------------------------------------------
# Get the data
# ----------------------------------------------------------------------
# As an example I use Emotion dataset from psycho package, https://github.com/neuropsychology/psycho.R

# Get the dataset
dataset <- psycho::emotion

# See how it looks
head(dataset)

A data.frame: 6 × 11
	Participant_ID	Participant_Age	Participant_Sex	Item_Category	Item_Name	Trial_Order	Emotion_Condition	Subjective_Arousal	Subjective_Valence	Autobiographical_Link	Recall
	<fct>	<dbl>	<fct>	<fct>	<fct>	<dbl>	<fct>	<dbl>	<dbl>	<dbl>	<lgl>
1	1S	18.38467	Female	People	People_158_h	1	Neutral	12.239583	0.5208333	63.281250	TRUE
2	1S	18.38467	Female	Faces	Faces_045_h	2	Neutral	16.406250	4.1666667	22.135417	FALSE
3	1S	18.38467	Female	People	People_138_h	3	Neutral	25.520833	25.5208333	55.989583	TRUE
4	1S	18.38467	Female	People	People_148_h	4	Neutral	0.000000	0.0000000	44.791667	FALSE
5	1S	18.38467	Female	Faces	Faces_315_h	5	Neutral	25.781250	45.8333333	30.208333	FALSE
6	1S	18.38467	Female	Faces	Faces_224_h	6	Neutral	2.604167	0.0000000	8.072917	FALSE

Input data for the function¶

In [4]:

# ----------------------------------------------------------------------
# Specify items and parameters
# ----------------------------------------------------------------------

# From the dataset, specify subject ID, Dependent Variable, and 2 within-subject Factors 
columns <- list(
  sID = "Participant_ID",     # subject ID
  DV  = "Subjective_Valence", # dependent variable
  Fc1 = "Item_Category",      # within-subj factor 1
  Fc2 = "Emotion_Condition"   # within-subj factor 2
)
# Define plot label names, tilte, colors and what type of error bars to show on the plot
param <- list(
  y.label    = "Subjective Valence",
  Fc1.label  = "Item",
  Fc2.label  = "Emotion",
  cat.color  = c('#DF4A56', '#5284a8'),
  errorbar   = "ci"  # can be either sd, se, or ci
)
# the plot title
param$title <- sprintf("%s x %s interaction", param$Fc2.label, param$Fc1.label)

Run the function¶

The function will output several result summaries. And it will also return a list of result variables including assumption plot (boxplots & QQ plot) and the interaction plot.

In [5]:

results <- rm_2by2_anova(dataset, columns, param)

================================================================

CHECK ASSUMPTIONS

================================================================
 Outliers

# A tibble: 4 x 6
  Fc1    Fc2      sID   meanDV is.outlier is.extreme
  <fct>  <fct>    <fct>  <dbl> <lgl>      <lgl>     
1 Faces  Negative 17S    -1.91 TRUE       FALSE     
2 Faces  Neutral  9S     50.5  TRUE       FALSE     
3 People Negative 17S    -1.17 TRUE       TRUE      
4 People Negative 9S    -25.8  TRUE       FALSE     

Data has extreme outliers!

-----------------------------------------------
 Normality 

# A tibble: 4 x 5
  Fc1    Fc2      variable statistic       p
  <fct>  <fct>    <chr>        <dbl>   <dbl>
1 Faces  Negative meanDV       0.928 0.159  
2 Faces  Neutral  meanDV       0.926 0.145  
3 People Negative meanDV       0.833 0.00364
4 People Neutral  meanDV       0.947 0.351  

Not all levels are normaly distributed!

================================================================

SUMMARY DESCRIPTIVES

================================================================
     Fc1      Fc2  N    meanDV       sd       se       ci
1  Faces Negative 19 -52.75116 14.77499 3.389615 7.121317
2  Faces  Neutral 19  17.62804 11.73257 2.691635 5.654916
3 People Negative 19 -59.92947 13.65515 3.132706 6.581571
4 People  Neutral 19  19.45238 15.63017 3.585806 7.533500

================================================================

2x2 INTERACTION RESULTS

================================================================

Error: sID
          Df Sum Sq Mean Sq F value Pr(>F)
Residuals 18   9623   534.6               

Error: sID:Fc1
          Df Sum Sq Mean Sq F value Pr(>F)
Fc1        1  136.2  136.16   1.499  0.237
Residuals 18 1634.6   90.81               

Error: sID:Fc2
          Df Sum Sq Mean Sq F value   Pr(>F)    
Fc2        1 106535  106535   245.9 6.11e-12 ***
Residuals 18   7798     433                     
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Error: sID:Fc1:Fc2
          Df Sum Sq Mean Sq F value Pr(>F)  
Fc1:Fc2    1    385     385   5.831 0.0266 *
Residuals 18   1188      66                 
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Significant Emotion x Item interaction: F(1,18) = 5.83, p = 0.0266076

================================================================

Main Effects

================================================================
Main effect of Item: F(1,18) = 1.50, p = 0.2365520


-----------------------------------------------
Main effect of Emotion: F(1,18) = 245.92, p = 0.0000000

================================================================

Post-hoc

================================================================

Pairwise comparisons grouped by Factor 1
# A tibble: 2 x 11
  Fc1    .y.    group1   group2     n1    n2 statistic    df        p    p.adj
* <fct>  <chr>  <chr>    <chr>   <int> <int>     <dbl> <dbl>    <dbl>    <dbl>
1 Faces  meanDV Negative Neutral    19    19     -14.6    18 2.09e-11 2.09e-11
2 People meanDV Negative Neutral    19    19     -14.7    18 1.84e-11 1.84e-11
# ... with 1 more variable: p.adj.signif <chr>


-----------------------------------------------

Pairwise comparisons grouped by Factor 2
# A tibble: 2 x 11
  Fc2   .y.   group1 group2    n1    n2 statistic    df     p p.adj p.adj.signif
* <fct> <chr> <chr>  <chr>  <int> <int>     <dbl> <dbl> <dbl> <dbl> <chr>       
1 Nega~ mean~ Faces  People    19    19     2.37     18 0.029 0.029 *           
2 Neut~ mean~ Faces  People    19    19    -0.673    18 0.51  0.51  ns

Assumption check plots¶

In [6]:

options(repr.plot.width = 10, repr.plot.height = 5, repr.plot.res = 200) # change plot size

results$plot.assumption.checks

Results plot¶

Error bars in the plot are within-subject 95% CIs.

Basic plot¶

In [7]:

options(repr.plot.width = 7, repr.plot.height = 5, repr.plot.res = 200) # change plot size
results$plot.anova

Add some extra stuff to the plot¶

In [8]:

#options(repr.plot.width = 7, repr.plot.height = 6, repr.plot.res = 200) # change plot size
# Adding post-hoc comparison stars
# Prepare the values

# post-hoc pairwise comparisons for each Fc1 level between the Fc2 levels
p.Fc1.L1 <-
  ifelse(results$pwc1$p[1] < 0.05, stars.pval(results$pwc1$p[1]), "n.s.")
p.Fc1.L2 <-
  ifelse(results$pwc1$p[2] < 0.05, stars.pval(results$pwc1$p[2]), "n.s.")


# add horizontal line BEHIND other elements
results$plot.anova$layers <- c(geom_hline(aes(yintercept = 0), size = 0.1), # the new layer
                               results$plot.anova$layers) # original layers

# add other things
results$plot.anova +
  
  # Add post-hoc pairwise comparisons
  # might need to adjust the x and y positions
  annotate(
    "text",
    x = 1.30,
    y = mean(results$dataSummary$meanDV),
    label = p.Fc1.L1,
    color = param$cat.color[1],
    size = 8
  ) +
  annotate(
    "text",
    x = 1.7,
    y = mean(results$dataSummary$meanDV),
    label = p.Fc1.L2,
    color = param$cat.color[2],
    size = 8
  ) +
  
  # add individual datapoints
  geom_point(
    colour = "black",
    alpha = .3,
    aes(fill = Fc1),
    size = 2,
    stroke = 0.5,
    shape = 21,
    position = position_jitterdodge(),
    show.legend = FALSE
  )

Example 2: Step-by-step example¶

Here the above function elements are executed ste-by-step (the function itself is not required).

Specify parameters¶

In [9]:

# ----------------------------------------------------------------------
# Specify items and parameters
# ----------------------------------------------------------------------

# From the dataset, specify subject ID, Dependent Variable, and 2 within-subject Factors 
columns <- list(
  sID = "Participant_ID",     # subject ID
  DV  = "Subjective_Valence", # dependent variable
  Fc1 = "Item_Category",      # within-subj factor 1
  Fc2 = "Emotion_Condition"   # within-subj factor 2
)

# Define plot label names, tilte, colors and what type of error bars to show on the plot
param <- list(
  y.label    = "Subjective Valence",
  Fc1.label  = "Item",
  Fc2.label  = "Emotion",
  cat.color  = c('#DF4A56', '#5284a8'),
  errorbar   = "ci"  # can be either sd, se, or ci
)
# the plot title
param$title <- sprintf("%s x %s interaction", param$Fc2.label, param$Fc1.label)

Prepare the dataset¶

In [10]:

# ----------------------------------------------------------------------
# Prepare the working dataset
# ----------------------------------------------------------------------
# Subset the dataset to only the columns of interest
# Because dplyr::select function clashes with MASS::select need to specify to use dplyr:: select
data_subset <- dataset %>% dplyr::select(all_of(unlist(columns)))

# Average same Factor/Level values for each subject if there are several. 
# Will use this data frame df for the rest of the results and plots
df <- ddply(data_subset, .(sID, Fc1, Fc2), 
             summarise, meanDV = mean(DV, na.rm = TRUE))

# sID, Fc1 and Fc2 must be factors (in case they are not already)
df$sID <- as.factor(df$sID)
df$Fc1 <- as.factor(df$Fc1)
df$Fc2 <- as.factor(df$Fc2)

# See how it looks
head(df)

A data.frame: 6 × 4
	sID	Fc1	Fc2	meanDV
	<fct>	<fct>	<fct>	<dbl>
1	10S	Faces	Negative	-53.472222
2	10S	Faces	Neutral	32.508681
3	10S	People	Negative	-62.673611
4	10S	People	Neutral	20.833333
5	11S	Faces	Negative	-76.189605
6	11S	Faces	Neutral	8.388238

Check assumptions¶

Outliers¶

In [11]:

# ----------------------------------------------------------------------
# Check for outliers
# ----------------------------------------------------------------------
# Identify if there are extreme outliers
check.outliers <- df %>%
  group_by(Fc1, Fc2) %>%
  identify_outliers(meanDV)

# Display
check.outliers

# Outlier result text
outlier.result.txt <- ifelse(any(check.outliers$is.extreme == T), 
       "Data has extreme outliers!",
       "There are no extreme outliers.")
cat("================================================\n\n")
cat(outlier.result.txt)
cat("\n\n================================================")

# Boxplot title color. It will be red if there are ouliers!
outlier.title.color <- ifelse(any(check.outliers$is.extreme == T), 
       "red",
       "black")

# Create a boxplot
plot.box <- ggboxplot(
  df, x = "Fc2", y = "meanDV",
  color = "Fc1", 
  title = outlier.result.txt, 
  ylab = param$y.label, 
  xlab = param$Fc2.label
  ) +
guides(color = guide_legend(param$Fc1.label)) +
scale_color_manual(values = param$cat.color) +
theme(plot.title = element_text(hjust = 0.5, color = outlier.title.color))

A tibble: 4 × 6
Fc1	Fc2	sID	meanDV	is.outlier	is.extreme
<fct>	<fct>	<fct>	<dbl>	<lgl>	<lgl>
Faces	Negative	17S	-1.909722	TRUE	FALSE
Faces	Neutral	9S	50.469741	TRUE	FALSE
People	Negative	17S	-1.171875	TRUE	TRUE
People	Negative	9S	-25.817472	TRUE	FALSE

================================================

Data has extreme outliers!

================================================

Normality¶

In [12]:

# ----------------------------------------------------------------------
# Check normality assumption
# ----------------------------------------------------------------------
# Shapiro-Wilk’s test (should be > .05)
check.normality <- df %>% 
  group_by(Fc1, Fc2) %>% 
  shapiro_test(meanDV)

# Display
check.normality

# Normality result text
normality.result.txt <- ifelse(any(check.normality$p < 0.05), 
       "Not all levels are normaly distributed!", 
       "All levels are normally distributed.")
cat("================================================\n\n")
cat(normality.result.txt)
cat("\n\n================================================")

# QQ plot title color. It will be red if not all levels meet normality!
normality.title.color <- ifelse(any(check.normality$p < 0.05), 
       "red",
       "black")

# Create Quantile-Quantile plots
plot.qq <- ggqqplot(df, "meanDV", ggtheme = theme_minimal(), title = normality.result.txt) +
  facet_grid(Fc1 ~ Fc2, labeller = "label_both") + 
  theme(plot.title = element_text(hjust = 0.5, color = normality.title.color))

A tibble: 4 × 5
Fc1	Fc2	variable	statistic	p
<fct>	<fct>	<chr>	<dbl>	<dbl>
Faces	Negative	meanDV	0.9279360	0.158654565
Faces	Neutral	meanDV	0.9258916	0.145440020
People	Negative	meanDV	0.8333678	0.003640378
People	Neutral	meanDV	0.9469856	0.350725108

================================================

Not all levels are normaly distributed!

================================================

Boxplot and QQ plots¶

In [13]:

# ----------------------------------------------------------------------
# PLOT Boxplot and QQ
# ----------------------------------------------------------------------
options(repr.plot.width = 10, repr.plot.height = 5, repr.plot.res = 200) # change plot size

# Add result texts as titles for each plot
plot_grid(plot.box, plot.qq, nrow = 1, rel_widths = c(1/4,3/4))

Descriptive summary¶

In [14]:

# ----------------------------------------------------------------------
# Within-Subject descriptive summary
# ----------------------------------------------------------------------
# Get the within-subject summary. Will need this to plot within-subject error bars.
# Using summarySEwithin function from the Rmisc package 

dataSummary <- summarySEwithin(
  df,
  measurevar = "meanDV",
  withinvars = c("Fc1", "Fc2"),
  idvar = "sID",
  na.rm = TRUE,
  conf.interval = 0.95
)
# See how it looks
dataSummary

A data.frame: 4 × 7
Fc1	Fc2	N	meanDV	sd	se	ci
<fct>	<fct>	<dbl>	<dbl>	<dbl>	<dbl>	<dbl>
Faces	Negative	19	-52.75116	14.77499	3.389615	7.121317
Faces	Neutral	19	17.62804	11.73257	2.691635	5.654916
People	Negative	19	-59.92947	13.65515	3.132706	6.581571
People	Neutral	19	19.45238	15.63017	3.585806	7.533500

2 x 2 ANOVA results¶

2-way interaction¶

In [15]:

# ----------------------------------------------------------------------
# Factor 1 X Factor 2 interaction, ANOVA
# ----------------------------------------------------------------------
# Within-subject repeated measures ANOVA
  results$res.anova <-
    aov(meanDV ~ Fc1 * Fc2 + Error(sID / (Fc1 * Fc2)),
        data = df)
  #Summary results
  summary.anova <- summary(results$res.anova)

  # save p-val, will need it to put on the plot
  res.anova.pval <-
    summary.anova[["Error: sID:Fc1:Fc2"]][[1]][["Pr(>F)"]][1]

  # Display result string
  sign.anova <-
    ifelse(res.anova.pval < 0.05, "Significant", "No significant")

  results$res.txt.anova <- sprintf(
    "\n%s %s: F(%d,%d) = %.2f, p = %.7f\n",
    sign.anova,
    param$title,
    summary.anova[["Error: sID:Fc1:Fc2"]][[1]][["Df"]][1],
    summary.anova[["Error: sID:Fc1:Fc2"]][[1]][["Df"]][2],
    summary.anova[["Error: sID:Fc1:Fc2"]][[1]][["F value"]][1],
    res.anova.pval
  )

cat("\n================================================\n\n")
cat(results$res.txt.anova)
cat("\n\n================================================")

================================================


Significant Emotion x Item interaction: F(1,18) = 5.83, p = 0.0266076


================================================

Main Effects of Factor 1¶

In [16]:

# Main Effect of Factor 1

  # save p-val, will need it to put on the plot
  res.Fc1.pval <-
    summary.anova[["Error: sID:Fc1"]][[1]][["Pr(>F)"]][1]

  # Display result string
  sign.Fc1 <-
    ifelse(res.Fc1.pval < 0.05, "Significant", "No significant")

  results$res.txt.Fc1 <- sprintf(
    "Main effect of %s: F(%d,%d) = %.2f, p = %.7f\n",
    param$Fc1.label,
    summary.anova[["Error: sID:Fc1"]][[1]][["Df"]][1],
    summary.anova[["Error: sID:Fc1"]][[1]][["Df"]][2],
    summary.anova[["Error: sID:Fc1"]][[1]][["F value"]][1],
    summary.anova[["Error: sID:Fc1"]][[1]][["Pr(>F)"]][1]
  )

  # display the result sentence

  cat(results$res.txt.Fc1)

Main effect of Item: F(1,18) = 1.50, p = 0.2365520

Main Effects of Factor 2¶

In [17]:

# Main Effect of Factor 2
  res.Fc2.pval <-
    summary.anova[["Error: sID:Fc2"]][[1]][["Pr(>F)"]][1]

  # Display result string
  sign.Fc2 <-
    ifelse(res.Fc2.pval < 0.05, "Significant", "No significant")

  results$res.txt.Fc2 <- sprintf(
    "Main effect of %s: F(%d,%d) = %.2f, p = %.7f\n",
    param$Fc2.label,
    summary.anova[["Error: sID:Fc2"]][[1]][["Df"]][1],
    summary.anova[["Error: sID:Fc2"]][[1]][["Df"]][2],
    summary.anova[["Error: sID:Fc2"]][[1]][["F value"]][1],
    summary.anova[["Error: sID:Fc2"]][[1]][["Pr(>F)"]][1]
  )

  # display the result sentence

  cat(results$res.txt.Fc2)

Main effect of Emotion: F(1,18) = 245.92, p = 0.0000000

Post-hoc tests¶

In [18]:

# Pairwise comparisons grouped by Factor 1
pwc1 <- df %>%
  group_by(Fc1) %>%
  pairwise_t_test(
    meanDV ~ Fc2, paired = TRUE
  )
pwc1

A rstatix_test: 2 × 11
	Fc1	.y.	group1	group2	n1	n2	statistic	df	p	p.adj	p.adj.signif
	<fct>	<chr>	<chr>	<chr>	<int>	<int>	<dbl>	<dbl>	<dbl>	<dbl>	<chr>
1	Faces	meanDV	Negative	Neutral	19	19	-14.57339	18	2.09e-11	2.09e-11	****
2	People	meanDV	Negative	Neutral	19	19	-14.68316	18	1.84e-11	1.84e-11	****

In [19]:

# Pairwise comparisons grouped by Factor 2
pwc2 <- df %>%
  group_by(Fc2) %>%
  pairwise_t_test(
    meanDV ~ Fc1, paired = TRUE
  )
pwc2

A rstatix_test: 2 × 11
	Fc2	.y.	group1	group2	n1	n2	statistic	df	p	p.adj	p.adj.signif
	<fct>	<chr>	<chr>	<chr>	<int>	<int>	<dbl>	<dbl>	<dbl>	<dbl>	<chr>
1	Negative	meanDV	Faces	People	19	19	2.3724765	18	0.029	0.029	*
2	Neutral	meanDV	Faces	People	19	19	-0.6727361	18	0.510	0.510	ns

Plot the results¶

Preparing titles and annotations¶

In [20]:

# ----------------------------------------------------------------------
# P-values for displaying significance stars
# ----------------------------------------------------------------------
# 2x2 interaction result
p.anova <-
  ifelse(res.anova.pval < 0.05, stars.pval(res.anova.pval), "n.s.")
# Main effects
p.mainFc1 <-
  ifelse(results$one.way1$p < 0.05,
         stars.pval(results$one.way1$p),
         "n.s.")
p.mainFc2 <-
  ifelse(results$one.way2$p < 0.05,
         stars.pval(results$one.way2$p),
         "n.s.")
# plot main title and factor labels (with added significance stars)
plot.title <- sprintf("%s %s", param$title, p.anova)
Fc1.title <- sprintf("%s %s:", param$Fc1.label, p.mainFc1)
Fc2.title <- sprintf("%s %s", param$Fc2.label, p.mainFc2)

The plotting function¶

In [21]:

# ----------------------------------------------------------------------
# Plot the interaction results
# ----------------------------------------------------------------------
plot.anova <-
  ggplot(data = df,
         aes(
           x = Fc2,
           y = meanDV,
           group = Fc1,
           color = Fc1
         )) +
  # connect the level means with a line
  stat_summary(fun = mean, geom = "line", size = 1) +
  
  # add mean point and error bars (taken from dataSummary created above)
  geom_pointrange(
    data = dataSummary,
    aes(
      y = meanDV,
      ymin = meanDV - .data[[param$errorbar]],
      ymax = meanDV + .data[[param$errorbar]],
      color = Fc1
    ),
    size = 1,
    stroke = 0.5,
    show.legend = FALSE
  ) +
  
  # change the default colors to the defined ones
  scale_color_manual(values = param$cat.color) +
  scale_fill_manual(values = param$cat.color) +
  
  # add labels and the title
  guides(color = guide_legend(Fc1.title)) +
  labs(x = Fc2.title, y = param$y.label) +
  ggtitle(plot.title) +
  
  # make it nicer
  theme_minimal() +
  theme(
    text = element_text(size = 16),
    plot.title = element_text(hjust = 0.5,
                              size = 16,
                              face = "bold"),
    panel.grid.major = element_blank(),
    panel.grid.minor = element_blank(),
    legend.position = "top"
  )

Add additional items to the plot and display the plot¶

In [22]:

options(repr.plot.width = 7, repr.plot.height = 6) # change plot size
eb <- "sd"

# ----------------------------------------------------------------------
# Add extra stuff to the plot
# ----------------------------------------------------------------------
plot.anova + 

# add post-hoc pairwise comparisons
# might need to adjust the x and y positions
  annotate(
    "text", x = 1.30, y = mean(dataSummary$meanDV),
    label = p.Fc1.L1, color = param$cat.color[1], size = 8
  ) +
  annotate(
    "text", x = 1.7, y = mean(dataSummary$meanDV),
    label = p.Fc1.L2, color = param$cat.color[2], size = 8
  )  +

# add individual datapoints
geom_point(
    colour = "black", alpha = .3,
    aes(fill = Fc1),
    size = 2, stroke = 0.5, shape = 21,
    position = position_jitterdodge(),
    show.legend = FALSE
  ) +

# add horizontal line at 0
    geom_hline(aes(yintercept = 0), size = 0.1)

Example 3: Test on another dataset¶

Here I am testing the rm_2by2_anova function on another dataset. The dataset is from https://ademos.people.uic.edu/Chapter21.html. The background of the dataset is as follows:

This fake data is a sample of 30 adults. All subjects were shown a set of words. Afterwards, for some of the words they were given a multiple choice test, and the rest of the words they simply restudied. Immediately afterwards their memory was tested for all words to determine if there were memory differences depending on whether the word was tested or restudied (immediate final test). Two days later, subjects came back and were given the same test on all words again (delayed final test). The dependent variable (DV) given in the file are a proportion of correctly remembered items for the immediate and delayed test for both conditions. Therefore, if we want to know if there are memory difference based on time delay and whether the word was tested or restudied, we need to conduct a within-subjects ANOVA.

Load the data¶

In [23]:

# https://ademos.people.uic.edu/Chapter21.html
# load the data
study_dataset <- read.csv("https://ademos.people.uic.edu/Chapter_21_Within_Data.csv")

# Dataset content: 
#Subject = Subject ID #
#Within_Cond = Study Method (test or restudy)
#Within_Time = Immediate or Delayed
#DV = Memory Performance

# Change the column names and levels to reflect the data content

study_dataset$Within_Cond <- factor(study_dataset$Within_Cond,
                               levels = c(1,2),
                               labels = c("Test", "Restudy"))

study_dataset$Within_Time <-factor(study_dataset$Within_Time,
                                  levels = c(1,2),
                                  labels = c("Immediate", "Delayed")) 

colnames(study_dataset)[2] <- "Study_Method"
colnames(study_dataset)[3] <- "Test_Time"
colnames(study_dataset)[4] <- "Memory_Performance"

head(study_dataset)

A data.frame: 6 × 4
	Subject	Study_Method	Test_Time	Memory_Performance
	<int>	<fct>	<fct>	<dbl>
1	1	Test	Immediate	96
2	2	Test	Immediate	81
3	3	Test	Immediate	73
4	4	Test	Immediate	90
5	5	Test	Immediate	72
6	6	Test	Immediate	83

Specify parameters¶

In [24]:

# ----------------------------------------------------------------------
# Specify items and parameters
# ----------------------------------------------------------------------

# From the dataset, specify subject ID, Dependent Variable, and 2 within-subject Factors 
columns <- list(
  sID = "Subject",
  DV  = "Memory_Performance",
  Fc1 = "Study_Method",
  Fc2 = "Test_Time"
)
# Define plot label names, tilte, colors and what type of error bars to show on the plot
param <- list(
  y.label    = "Memory Performance",
  Fc1.label  = "Study Method",
  Fc2.label  = "Test Time",
  cat.color  = c('darkblue', 'lightgreen'),
  errorbar   = "ci"  # can be either sd, se, or ci
)
# the plot title
param$title <- sprintf("%s x %s interaction", param$Fc2.label, param$Fc1.label)

Run the function¶

In [25]:

results <- rm_2by2_anova(study_dataset, columns, param)

================================================================

CHECK ASSUMPTIONS

================================================================
 Outliers

[1] Fc1        Fc2        sID        meanDV     is.outlier is.extreme
<0 rows> (or 0-length row.names)

There are no extreme outliers.

-----------------------------------------------
 Normality 

# A tibble: 4 x 5
  Fc1     Fc2       variable statistic       p
  <fct>   <fct>     <chr>        <dbl>   <dbl>
1 Test    Immediate meanDV       0.922 0.0301 
2 Test    Delayed   meanDV       0.934 0.0645 
3 Restudy Immediate meanDV       0.958 0.280  
4 Restudy Delayed   meanDV       0.871 0.00176

Not all levels are normaly distributed!

================================================================

SUMMARY DESCRIPTIVES

================================================================
      Fc1       Fc2  N   meanDV        sd       se       ci
1    Test Immediate 30 81.26667  9.475361 1.729956 3.538158
2    Test   Delayed 30 83.80000  9.232216 1.685564 3.447366
3 Restudy Immediate 30 83.40000  9.504108 1.735205 3.548892
4 Restudy   Delayed 30 68.30000 12.424643 2.268419 4.639438

================================================================

2x2 INTERACTION RESULTS

================================================================

Error: sID
          Df Sum Sq Mean Sq F value Pr(>F)
Residuals 29   2497   86.12               

Error: sID:Fc1
          Df Sum Sq Mean Sq F value  Pr(>F)   
Fc1        1   1340  1340.0   12.24 0.00153 **
Residuals 29   3176   109.5                   
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Error: sID:Fc2
          Df Sum Sq Mean Sq F value  Pr(>F)   
Fc2        1   1184  1184.4   13.21 0.00107 **
Residuals 29   2599    89.6                   
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Error: sID:Fc1:Fc2
          Df Sum Sq Mean Sq F value   Pr(>F)    
Fc1:Fc2    1   2332  2332.0   20.16 0.000104 ***
Residuals 29   3354   115.6                     
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Significant Test Time x Study Method interaction: F(1,29) = 20.17, p = 0.0001042

================================================================

Main Effects

================================================================
Main effect of Study Method: F(1,29) = 12.24, p = 0.0015321


-----------------------------------------------
Main effect of Test Time: F(1,29) = 13.21, p = 0.0010667

================================================================

Post-hoc

================================================================

Pairwise comparisons grouped by Factor 1
# A tibble: 2 x 11
  Fc1     .y.    group1    group2     n1    n2 statistic    df         p   p.adj
* <fct>   <chr>  <chr>     <chr>   <int> <int>     <dbl> <dbl>     <dbl>   <dbl>
1 Test    meanDV Immediate Delayed    30    30     -1.12    29 0.271     2.71e-1
2 Restudy meanDV Immediate Delayed    30    30      5.15    29 0.0000165 1.65e-5
# ... with 1 more variable: p.adj.signif <chr>


-----------------------------------------------

Pairwise comparisons grouped by Factor 2
# A tibble: 2 x 11
  Fc2       .y.    group1 group2     n1    n2 statistic    df         p    p.adj
* <fct>     <chr>  <chr>  <chr>   <int> <int>     <dbl> <dbl>     <dbl>    <dbl>
1 Immediate meanDV Test   Restudy    30    30    -0.870    29 0.392      3.92e-1
2 Delayed   meanDV Test   Restudy    30    30     5.17     29 0.0000159  1.59e-5
# ... with 1 more variable: p.adj.signif <chr>

Assumption check plots¶

In [26]:

options(repr.plot.width = 10, repr.plot.height = 5, repr.plot.res = 200) # change plot size

results$plot.assumption.checks

Results plot (with extras)¶

In [27]:

options(repr.plot.width = 7, repr.plot.height = 6, repr.plot.res = 200) # change plot size
# Adding post-hoc comparison stars
# Prepare the values

# post-hoc pairwise comparisons for each Fc1 level between the Fc2 levels
p.Fc1.L1 <-
  ifelse(results$pwc1$p[1] < 0.05, stars.pval(results$pwc1$p[1]), "n.s.")
p.Fc1.L2 <-
  ifelse(results$pwc1$p[2] < 0.05, stars.pval(results$pwc1$p[2]), "n.s.")

results$plot.anova +
  
  # Add post-hoc pairwise comparisons
  # might need to adjust the x and y positions
  annotate(
    "text",
    x = 1.5,
    y = 86,
    label = p.Fc1.L1,
    color = param$cat.color[1],
    size = 8
  ) +
  annotate(
    "text",
    x = 1.5,
    y = 72,
    label = p.Fc1.L2,
    color = param$cat.color[2],
    size = 8
  ) +
  
  # add individual datapoints
  geom_point(
    colour = "black",
    alpha = .3,
    aes(fill = Fc1),
    size = 2,
    stroke = 0.5,
    shape = 21,
    position = position_jitterdodge(),
    show.legend = FALSE
  )

Example 4: Test on another dataset¶

Here I test the function on kepress learning by observation dataset. This is my dataset published in https://doi.org/10.1523/JNEUROSCI.1597-18.2018

Load the data¶

In [28]:

dataOL <- 
  read.csv("https://raw.githubusercontent.com/dcdace/E2fMRI_MVPA_PPI/master/Data/BehaviouralResults_longformat.csv")

# Change the Type levels to better reflect the data content
dataOL$Type <- factor(dataOL$Type,
                               levels = c("TR", "UN"),
                               labels = c("Trained", "Untrained"))

head(dataOL)

A data.frame: 6 × 8
	sID	gender	age	righthanded	Time	Type	ET	Err
	<int>	<chr>	<int>	<chr>	<chr>	<fct>	<dbl>	<int>
1	2	m	23	yes	Pre	Trained	1195.40	20
2	3	m	26	yes	Pre	Trained	1681.00	30
3	5	f	20	yes	Pre	Trained	1427.20	0
4	6	f	21	yes	Pre	Trained	1798.30	5
5	7	m	27	yes	Pre	Trained	904.36	0
6	8	m	40	yes	Pre	Trained	504.40	20

Specify parameters¶

In [29]:

# ----------------------------------------------------------------------
# Specify items and parameters
# ----------------------------------------------------------------------

# From the dataset, specify subject ID, Dependent Variable, and 2 within-subject Factors 
columns <- list(
  sID = "sID",
  DV  = "ET",
  Fc1 = "Type",
  Fc2 = "Time"
)
# Define plot label names, tilte, colors and what type of error bars to show on the plot
param <- list(
  y.label    = "Execution Time (ms)",
  Fc1.label  = "Type",
  Fc2.label  = "Time",
  cat.color  = c('#df4a56', '#0272c6'),
  errorbar   = "ci"  # can be either sd, se, or ci
)
# the plot title
param$title <- sprintf("%s x %s interaction", param$Fc2.label, param$Fc1.label)

Run the function¶

In [30]:

results <- rm_2by2_anova(dataOL, columns, param)

================================================================

CHECK ASSUMPTIONS

================================================================
 Outliers

# A tibble: 7 x 6
  Fc1       Fc2   sID   meanDV is.outlier is.extreme
  <fct>     <fct> <fct>  <dbl> <lgl>      <lgl>     
1 Trained   Post  19     2444. TRUE       FALSE     
2 Trained   Pre   17     3676. TRUE       FALSE     
3 Untrained Post  8       463. TRUE       FALSE     
4 Untrained Post  17     2551. TRUE       FALSE     
5 Untrained Post  19     2719. TRUE       FALSE     
6 Untrained Pre   17     3992  TRUE       FALSE     
7 Untrained Pre   19     3529. TRUE       FALSE     

There are no extreme outliers.

-----------------------------------------------
 Normality 

# A tibble: 4 x 5
  Fc1       Fc2   variable statistic      p
  <fct>     <fct> <chr>        <dbl>  <dbl>
1 Trained   Post  meanDV       0.957 0.602 
2 Trained   Pre   meanDV       0.917 0.150 
3 Untrained Post  meanDV       0.919 0.164 
4 Untrained Pre   meanDV       0.902 0.0869

All levels are normally distributed.

================================================================

SUMMARY DESCRIPTIVES

================================================================
        Fc1  Fc2  N   meanDV       sd       se       ci
1   Trained Post 16 1338.271 266.3100 66.57749 141.9066
2   Trained  Pre 16 1865.416 265.2562 66.31405 141.3450
3 Untrained Post 16 1463.746 213.3426 53.33564 113.6822
4 Untrained  Pre 16 1900.056 328.3329 82.08324 174.9563

================================================================

2x2 INTERACTION RESULTS

================================================================

Error: sID
          Df   Sum Sq Mean Sq F value Pr(>F)
Residuals 15 28566155 1904410               

Error: sID:Fc1
          Df Sum Sq Mean Sq F value Pr(>F)
Fc1        1 102547  102547   1.864  0.192
Residuals 15 825140   55009               

Error: sID:Fc2
          Df  Sum Sq Mean Sq F value   Pr(>F)    
Fc2        1 3712982 3712982   25.17 0.000153 ***
Residuals 15 2212669  147511                     
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Error: sID:Fc1:Fc2
          Df Sum Sq Mean Sq F value Pr(>F)
Fc1:Fc2    1  33005   33005   1.791  0.201
Residuals 15 276434   18429               

No significant Time x Type interaction: F(1,15) = 1.79, p = 0.2007469

================================================================

Main Effects

================================================================
Main effect of Type: F(1,15) = 1.86, p = 0.1922710


-----------------------------------------------
Main effect of Time: F(1,15) = 25.17, p = 0.0001532

================================================================

Post-hoc

================================================================

Pairwise comparisons grouped by Factor 1
# A tibble: 2 x 11
  Fc1       .y.    group1 group2    n1    n2 statistic    df         p     p.adj
* <fct>     <chr>  <chr>  <chr>  <int> <int>     <dbl> <dbl>     <dbl>     <dbl>
1 Trained   meanDV Post   Pre       16    16     -5.32    15 0.0000852 0.0000852
2 Untrained meanDV Post   Pre       16    16     -4.17    15 0.000817  0.000817 
# ... with 1 more variable: p.adj.signif <chr>


-----------------------------------------------

Pairwise comparisons grouped by Factor 2
# A tibble: 2 x 11
  Fc2   .y.   group1 group2    n1    n2 statistic    df     p p.adj p.adj.signif
* <fct> <chr> <chr>  <chr>  <int> <int>     <dbl> <dbl> <dbl> <dbl> <chr>       
1 Post  mean~ Train~ Untra~    16    16    -3.05     15 0.008 0.008 **          
2 Pre   mean~ Train~ Untra~    16    16    -0.400    15 0.694 0.694 ns

Assumption check plots¶

In [31]:

options(repr.plot.width = 10, repr.plot.height = 5, repr.plot.res = 200) # change plot size

results$plot.assumption.checks

Results plot (with extras)¶

In [32]:

options(repr.plot.width = 7, repr.plot.height = 6, repr.plot.res = 200) # change plot size
# Adding post-hoc comparison stars
# Prepare the values

# post-hoc pairwise comparisons for each Fc1 level between the Fc2 levels
p.Fc1.L1 <-
  ifelse(results$pwc1$p[1] < 0.05, stars.pval(results$pwc1$p[1]), "n.s.")
p.Fc1.L2 <-
  ifelse(results$pwc1$p[2] < 0.05, stars.pval(results$pwc1$p[2]), "n.s.")

results$plot.anova +
  
  # Add post-hoc pairwise comparisons
  # might need to adjust the x and y positions
  annotate(
    "text",
    x = 1.5,
    y = 1400,
    label = p.Fc1.L1,
    color = param$cat.color[1],
    size = 8
  ) +
  annotate(
    "text",
    x = 1.5,
    y = 1800,
    label = p.Fc1.L2,
    color = param$cat.color[2],
    size = 8
  ) +
  
  # add individual datapoints
  geom_point(
    colour = "black",
    alpha = .3,
    aes(fill = Fc1),
    size = 2,
    stroke = 0.5,
    shape = 21,
    position = position_jitterdodge(),
    show.legend = FALSE
  ) + 

 # Show Pre first (instead of default alfabetical order)
 scale_x_discrete(limits = c("Pre", "Post"))

Control what to display¶

By default, the function outputs a lot of results. If you don't want do display the output, you can 'hush' (suppress) it with a 'hush' function.

The 'hush' function¶

In [33]:

# hush function
# I got it from here https://stackoverflow.com/questions/2723034/suppress-output-of-a-function 

hush <- function(code){
  sink("NUL") # use /dev/null in UNIX
  tmp = code
  sink()
  return(tmp)
}

Run the function without any output¶

In [34]:

results <- hush(rm_2by2_anova(dataOL, columns, param))

Choose what to output¶

Below is the list of possible output elements.

In [35]:

# results$outliers               # results of identify_outliers function
# results$normality              # results of shapiro_test
# results$plot.assumption.checks # boxplots and QQ plots in one plot
# results$dataSummary            # Within-Subject descriptive summary of all 4 levels
# results$res.anova              # ANOVA results of aov function
# results$one.way1               # Factor 1 Main effect results of anova_test function 
# results$one.way2               # Factor 2 Main effect results of anova_test function 
# results$pwc1                   # Pairwise comparisons grouped by Factor 1
# results$pwc2                   # Pairwise comparisons grouped by Factor 2
# results$plot.anova             # the results plot

In [36]:

# for example, output ANOVA summary results
summary(results$res.anova)

Error: sID
          Df   Sum Sq Mean Sq F value Pr(>F)
Residuals 15 28566155 1904410               

Error: sID:Fc1
          Df Sum Sq Mean Sq F value Pr(>F)
Fc1        1 102547  102547   1.864  0.192
Residuals 15 825140   55009               

Error: sID:Fc2
          Df  Sum Sq Mean Sq F value   Pr(>F)    
Fc2        1 3712982 3712982   25.17 0.000153 ***
Residuals 15 2212669  147511                     
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Error: sID:Fc1:Fc2
          Df Sum Sq Mean Sq F value Pr(>F)
Fc1:Fc2    1  33005   33005   1.791  0.201
Residuals 15 276434   18429

Facebook Tweet LinkedInShares30

Table of Contents

2x2 Repeated Measures ANOVA¶

Example 1: Using rm_2by2_anova function¶

Load the function¶

Load example data¶

Input data for the function¶

Run the function¶

Assumption check plots¶

Results plot¶

Basic plot¶

Add some extra stuff to the plot¶

Example 2: Step-by-step example¶

Specify parameters¶

Prepare the dataset¶

Check assumptions¶

Outliers¶

Normality¶

Boxplot and QQ plots¶

Descriptive summary¶

2 x 2 ANOVA results¶

2-way interaction¶

Main Effects of Factor 1¶

Main Effects of Factor 2¶

Post-hoc tests¶

Plot the results¶

Preparing titles and annotations¶

The plotting function¶

Add additional items to the plot and display the plot¶

Example 3: Test on another dataset¶

Load the data¶

Specify parameters¶

Run the function¶

Assumption check plots¶

Results plot (with extras)¶

Example 4: Test on another dataset¶

Load the data¶

Specify parameters¶

Run the function¶

Assumption check plots¶

Results plot (with extras)¶

Control what to display¶

The 'hush' function¶

Run the function without any output¶

Choose what to output¶