The basics
You can get quite far with hist, mean, sd, summary, boxplot.
Standard error
Here's one, found somewhere online1:
st.err <- function(x)
{
nn <- function(x)
{
length(na.omit(x))
}
if (is.matrix(x))
n <- apply(x, 2, nn)
else if (is.data.frame(x))
n <- sapply(x, nn)
else if (is.list(x))
cat("Hmm, this won't work with a list...\n")
else if (is.vector(x))
n <- nn(x)
sd <- sd(x, na.rm = TRUE)
sd/sqrt(n)
}
Formatting things
The rround function forces R to round x to d decimal places:
rround = function(x,d) format(round(x, d), nsmall = d)
An example use of this is a function which takes a vector, x, and gives you back a string of its mean and sd in parentheses:
meansd = function(x) paste(rround(mean(x),2)," (", rround(sd(x),2),")", sep="")
Note also the helpful paste function there. Can come in handy for combining factors and other useful things.
Tables
Suppose you have data like this:
| ID | score1 | score2 | sex | hand |
|---|---|---|---|---|
| 1 | 5.19 | 4.09 | M | L |
| 2 | 1.66 | 14.91 | M | R |
| 3 | 8.98 | 14.23 | F | L |
| 4 | 4.09 | 5.35 | F | R |
| 5 | 2.67 | 13.38 | M | L |
| 6 | 1.62 | 5.52 | M | R |
| … | … | … | … |
And want a table of means and sds like:
| sex | F | F | M | M |
| hand | L | R | L | R |
| score1 | 5.07 (2.62) | 4.91 (3.12) | 4.68 (3.30) | 4.90 (2.82) |
| score2 | 9.99 (2.57) | 9.72 (2.70) | 10.02 (3.58) | 10.55 (2.90) |
Handy way to do it (the beef here is summarise, llist, merge, and t):
require(Hmisc)
rround = function(x,d) format(round(x, d), nsmall = d)
meansd = function(x) paste(rround(mean(x),2)," (", rround(sd(x),2),")", sep="")
d = data.frame(ID = 1:100
, score1 = rnorm(100,5,3)
, score2 = rnorm(100,10,3)
, sex = rep(c("M","M","F","F"),25)
, hand = rep(c("L","R"),50))
t1 = with(d, summarize(score1, llist(sex,hand), meansd))
t2 = with(d, summarize(score2, llist(sex,hand), meansd))
t3 = merge(t1,t2)
t(t3)
The t function is matrix tranpose. Have a nosy at t3 versus t(t3) to make sense of what it's doing.
Graphs
Means with error bars
Simple example of a barchart with error bars:
bardata <- 1:3 # create some data
error <- c(.2,.4,.7) # and error
bar <- barplot(bardata, ylim = c(0,4), col="lightblue", names.arg=c("A","B","C"))
arrows(bar, bardata + error, bar, bardata - error,
length = 0.10, # width of the arrowhead
angle = 90,
code = 3
)
(Remember to describe what kind of bars they are.)
Another graph
require(gplots)
n = 60
g1 = rnorm(n/2, 100, 40)
g2 = rnorm(n/2, 60, 40)
y = c(mean(g1),mean(g2))
ys = c(g1,g2)
group = c(rep("A",n/2),rep("B",n/2))
plotmeans(ys ~ group, xlab="", ylim=c(0,150), main = "", connect=T)
By default you get 95% CIs. Check the help for more options.