---
title: "Why use correlation-adjusted confidence intervals?"
bibliography: "../inst/REFERENCES.bib"
csl: "../inst/apa-6th.csl"
output:
rmarkdown::html_vignette
description: >
This vignette explains the difference between a regular confidence interval
and a correlation-adjusted confidence interval.
vignette: >
%\VignetteIndexEntry{Why use correlation-adjusted confidence intervals?}
%\VignetteEngine{knitr::rmarkdown}
\usepackage[utf8]{inputenc}
---
```{r, echo = FALSE, warning=FALSE, message = FALSE, results = 'hide'}
cat("this will be hidden; use for general initializations.")
library(superb)
library(ggplot2)
```
In paired-sample designs ---also called within-subject designs--- the same participants
are measured more than once. In that case, asking whether a factor influenced the
scores is the same as asking if the factor influenced all the participants. If all
the participants turn out to be influenced in the same manner, it can be safely
concluded that the factor influenced the group of participants.
## An example
Consider a study trying to establish the benefit of using exercises to improve
visuo-spatial abilities onto scores in statistics reasoning, as measured by a standardized
test with scores ranging from 50 to 150. The design is a within-subject design,
specifically a pre-exercises measure and a post-exercises measure of statistics reasoning.
The data are available in ``dataFigure2``; here is a snapshot of it
```{r}
head(dataFigure2)
```
There is a large variation in the scores obtained and as such a *t* test where
the scores are treated as independent will fail to detect a difference:
```{r}
t.test(dataFigure2$pre, dataFigure2$post, var.equal=TRUE)
```
However, when the proper, paired-sample t-test is used, we find that
the difference is quite important:
```{r}
t.test(dataFigure2$pre, dataFigure2$post, var.equal=TRUE, paired = TRUE)
```
What is going on? how can we make a plot that properly display this
difference?
## A few words on the underlying theory (optional)
Let's examine the data using a plot in which each participant's scores are
shown with a line. We get the following:
```{r, message=FALSE, echo=TRUE, fig.width = 4, fig.cap="**Figure 1**. Representation of the individual participants"}
library(reshape2)
# first transform the data in long format; the pre-post scores will go into column "variable"
dl <- melt(dataFigure2, id="id")
# add transparency when pre is smaller or equal to post
dl$trans = ifelse(dataFigure2$pre <= dataFigure2$post,0.9,1.0)
# make a plot, with transparent lines when the score increased
ggplot(data=dl, aes(x=variable, y=value, group=id, alpha = trans)) +
geom_line( ) +
coord_cartesian( ylim = c(70,150) ) +
geom_abline(intercept = 102.5, slope = 0, colour = "red", linetype=2)
```
As seen, except for 5 participants, a vast majority of the participants have an
upward trend in their results. Thus, this upward trend is probably a reality in
this dataset.
### Centering the participants to better see the trend
One solution used in @c05 is to center the participants' data on the
participants' mean. It consists in computing for each participant their mean
score and replace that participant's mean score with the overall mean. With
this manipulation, all the participants will now hover around the overall
mean (here 102.5, shown with a red dashed line).
The following realizes this *subject-centered* plot for each participant.
```{r, message=FALSE, echo=TRUE, fig.width = 4, fig.cap="**Figure 2**. Representation of the *subject-centered* individual participants"}
# use subjectCenteringTransform function
library(superb)
df2 <- subjectCenteringTransform(dataFigure2, c("pre","post"))
# tranform into long format
library(reshape2)
dl2 <- melt(df2, id="id")
# make the plot
ggplot(data=dl2, aes(x=variable, y=value, colour=id, group=id)) + geom_line()+
coord_cartesian( ylim = c(70,150) ) +
geom_abline(intercept = 102.5, slope = 0, colour = "red", size = 0.5, linetype=2)
```
Here again, we see that for 5 participants, their scores went down. For the 20 remaining
ones, the trend is upward. Thus, there is clear tendency for the exercices to be
beneficial.
Running the adequate *paired* *t* test, we find indeed that the difference
is strongly significant (t(24) = 2.9, p = .008):
```{r}
t.test(dataFigure2$pre, dataFigure2$post, paired=TRUE)
```
### What is the impact on confidence intervals?
The above suggests that within-subject designs can be much more powerful than between-
subject design. As long as there is a general trend visible in most participants, the
paired design will afford more statistical power. How to we know that there is a
general trend? An easy solution is to compute the correlation across the pairs of
scores.
In R, you can run the following:
```{r}
cor(dataFigure2$pre, dataFigure2$post)
```
and you find out that in the present dataset, correlation is actually quite high, with
$r \approx .8$. Whenever correlation is positive, statistical power benefits from
correlation.
Because increased power means higher level of precision, the error bars should be
shortened by positive correlation. Estimating the adjusted length of the error
bars from correlation is a process called **decorrelation** [@c19].
To this date, three techniques have been proposed to decorrelate the measures.
* ``CM``: This method uses *subject-centering* followed by a bias-correction step (otherwise
the error bars are slightly overestimated) [@c05; @m08].
* ``LM``: This method also uses *subject-centering* and bias-correction
but when there are more than two measurements, it also equalizes the length of
the error bars using a technique akin to pooled standard deviation measure [@lm94].
* ``CA``: this is the newest proposal. It directly uses the correlation (or the mean
pairwise correlation when there are more than two measurement) to adjust the error bar
length. In a nutshell, the error bar length are adjusted using a multiplicative term
$\sqrt{1-r}$. As an example, when $ r = .8$, the adjustment is $\sqrt{1-.8} = 0.44$.
That means that the error bars are 44% the length of the unadjusted error bars
(that is, less than half).
Whichever method you choose have very little bearing on the actual result. As
shown in @c19, all three methods are based on the same general concepts and
they generate very little difference in the amount of adjustments.
In the present dataset, the error bar are more than shorten by half! which clearly
shows the benefit of the within-subject design on precision.
## Getting a plot
### Making it simple
With ``suberb``, all the decorrelation techniques are available using the adjustment
``decorrelation`` [@cgh21].
The simplest code is the following:
```{r, message=FALSE, warning=FALSE, echo=TRUE, fig.height = 3, fig.width = 4, fig.cap="**Figure 3a**. Means and difference and correlation-adjusted 95% confidence intervals"}
superbPlot(dataFigure2,
WSFactors = "Moment(2)",
adjustments = list(
purpose = "difference",
decorrelation = "CA" ## NEW! use a decorrelation technique
),
variables = c("pre","post"),
plotStyle = "line" )
```
This will result in a basic plot, but what matters is that the confidence intervals are adjusted so that you can compare bars even
if they are repeated measures. Note that the confidence interval of one point does
not contain the other point, suggesting, as it should be, a significant differencee.
### Refining the plots
With the following code, we can decorate the plot a little bit, and show
side-by-side both the unadjusted plot (which give an erroneous result) and
the adjusted plot.
```{r, message=FALSE, warning=FALSE, echo=TRUE, fig.width = 4, fig.cap="**Figure 3b**. Means and 95% confidence intervals on raw data (left) and on decorrelated data (right)"}
options(superb.feedback = 'none') # shut down 'warnings' and 'design' interpretation messages
library(gridExtra)
## realize the plot with unadjusted (left) and ajusted (right) 95\% confidence intervals
plt2a <- superbPlot(dataFigure2,
WSFactors = "Moment(2)",
adjustments = list(purpose = "difference"),
variables = c("pre","post"),
plotStyle = "line" ) +
xlab("Group") + ylab("Score") +
labs(title="Difference-adjusted\n95% confidence intervals") +
coord_cartesian( ylim = c(85,115) ) +
theme_gray(base_size=10) +
scale_x_discrete(labels=c("1" = "Collaborative games", "2" = "Unstructured activity"))
plt2b <- superbPlot(dataFigure2,
WSFactors = "Moment(2)",
adjustments = list(purpose = "difference", decorrelation = "CA"), #only difference
variables = c("pre","post"),
plotStyle = "line" ) +
xlab("Group") + ylab("Score") +
labs(title="Correlation and difference-adjusted\n95% confidence intervals") +
coord_cartesian( ylim = c(85,115) ) +
theme_gray(base_size=10) +
scale_x_discrete(labels=c("1" = "Collaborative games", "2" = "Unstructured activity"))
plt2 <- grid.arrange(plt2a,plt2b,ncol=2)
```
In the above plot, I used ``decorrelation = "CA"``. Alternatively, you could
use ``decorrelation = "UA"`` or ``decorrelation = "CM").
When there is only two conditions, all the techniques return identical
error bars.
### Illustrating the various decorrelatin techniques.
As another example illustrating the differences between the techniques, I
generated random data for 5 measures with an
amount of correlation of 0.8 in the population. In Figure 4 below, all error
bars are superimposed
on the same plot. As seen, there is only minor differences between the
three techniques. The green lines all have the same length; this is the
main characteristic of the Loftus and Masson approach, in contrast with the
other two techniques.
```{r, message=FALSE, warning=FALSE, echo=FALSE, fig.width = 4, fig.cap="**Figure 4**. All three decorelation techniques on the same plot along with un-decorrelated error bars"}
# using GRD to generate data with correlation of .8 and a moderate effect
options(superb.feedback = 'none') # shut down 'warnings' and 'design' interpretation messages
test <- GRD(WSFactors = "Moment(5)",
Effects = list("Moment" = extent(10) ),
Population = list(mean = 100, stddev = 25, rho = 0.8) )
# the common label to all 4 plots
tlbl <- paste( "(red) Difference-adjusted only\n",
"(blue) Difference adjusted and decorrelated with CM\n",
"(green) Difference-adjusted and decorrelated with LM\n",
"(orange) Difference-adjusted and decorrelated with CA\n",
"(bisque) Difference-adjusted and decorrelated with UA", sep="")
# to make the plots all identical except for the decorrelation method
makeplot <- function(dataset, decorrelationmethod, color, nudge, dir) {
superbPlot(dataset,
WSFactors = "Moment(5)",
variables = c("DV.1","DV.2","DV.3","DV.4","DV.5"),
adjustments=list(purpose = "difference", decorrelation = decorrelationmethod),
errorbarParams = list(color=color, width= 0.1, position = position_nudge(nudge), direction = dir ),
plotStyle="line" ) +
xlab("Moment") + ylab("Score") +
labs(subtitle=tlbl) +
coord_cartesian( ylim = c(85,115) ) +
theme_gray(base_size=10)
}
theme_transparent <- theme(
panel.grid.major = element_blank(),
panel.grid.minor = element_blank(),
panel.background = element_rect(fill = "transparent",colour = NA),
plot.background = element_rect(fill = "transparent",colour = NA)
)
# generate the plots, nudging the error bars and using distinct colors
pltrw <- makeplot(test, "none", "red", 0.0, "both")
pltCM <- makeplot(test, "CM", "blue", -0.2, "left")
pltLM <- makeplot(test, "LM", "chartreuse3", -0.1, "left")
pltCA <- makeplot(test, "CA", "orange", +0.1, "right")
pltUA <- makeplot(test, "UA", "bisque4", +0.2, "right")
# transform the ggplots into "grob" so that they can be manipulated
pltrwg <- ggplotGrob(pltrw)
pltCMg <- ggplotGrob(pltCM + theme_transparent)
pltLMg <- ggplotGrob(pltLM + theme_transparent)
pltCAg <- ggplotGrob(pltCA + theme_transparent)
pltUAg <- ggplotGrob(pltUA + theme_transparent)
# put the grobs onto an empty ggplot
ggplot() +
annotation_custom(grob=pltrwg) +
annotation_custom(grob=pltCMg) +
annotation_custom(grob=pltLMg) +
annotation_custom(grob=pltCAg) +
annotation_custom(grob=pltUAg)
```
### Illustrating individual differences
In ``superb``, it is possible to ask a certain type of plot. The ``plotStyle`` used so
far is ``"line"`` (the default is ``"bar"``). Another basic style is ``"point"`` (no line
connecting the means).
Other types of plot exists that are apt at showing the summary statistics but also
the individual scores (the [6th Vignette]
(https://dcousin3.github.io/superb/articles/Vignette6.html) shows how to
develop custom-made layouts).
For illustrating individual differences, a style proposed is ``pointindividualline``
which --- as per Figure 1--- will show the individual scores along with the summary
statistics and the error bars. For example:
```{r, message=FALSE, warning=FALSE, echo=TRUE, fig.width = 4, fig.cap="**Figure 5**. Means and 95% confidence intervals along with individual scores depicted as lines"}
superbPlot(dataFigure2,
WSFactors = "Moment(2)",
adjustments = list(purpose = "difference", decorrelation = "CM"),
variables = c("pre","post"),
plotStyle = "pointindividualline" ) +
xlab("Group") + ylab("Score") +
labs(subtitle="Correlation- and Difference-adjusted\n95% confidence intervals") +
coord_cartesian( ylim = c(70,150) ) +
theme_gray(base_size=10) +
scale_x_discrete(labels=c("1" = "Collaborative games", "2" = "Unstructured activity"))
```
## In conclusion
The major obstacle to the use of adjusted error bars was the difficulty to obtain them.
None of the statistical software (e.g., SPSS, SAS) provide these adjustments. A way
around is to compute these manually. Although not that complicated, it requires manipulations,
whether they are done in EXCEL, or through macros [e.g., WSPlot, @oc14].
The present function renders all the adjustments a mere option in a function.
# References