medicalnerds.com

If you wish to analyse longitudinal data sometimes it is helpful to plot each experimental subject’s data at each timepoint to get a feel for the patterns or trends. You can often learn more from this raw data than a boxplot or line graph of the means. I am a habitual SPSS user but have found myself limited with it’s capabilities. In this short article, aimed at SPSS users who are new to R, I will show you how to produce a ‘Spaghetti Plot’ using R.

For this example I have measurements of 50 patient’s serum rhubarb concentrations after ingestion of a large rhubarb based meal. Data is recorded at several time-points: 0,3,6 and then every 6 months. Many of the patients do not have observations at each timepoint. The current data Format in SPSS is

However to plot this in SPSS you need turn the data into long format using

Menu>>Data>>Restructure>>Case to Variables

to give you this format:-

When you have your data in this format you can use

Menu>>Graphs>>Line Charts>>Multiple>> "Values of individual Cases"

Then place Time as the Category label and “Lines Represent” as “Rhubarb1, Rhubarb2, etc”. This will give you the following graph. (I have only plotted a few cases for simplicity)

The major flaw with this graph is that the x-scale is ordinal rather than being a continuous scale..really I would like to see the distance between month 0 and 3 half the distance between 6 and 12. The only possible solution in SPSS is to plot the data as a multiple overlay scatter plot with interpolation lines. This works for a few subjects but is tedious for more than a couple and I have found impossible for 50.

My solution is to use R (R-Project for Statistical Computing). R is an open-source programming environment that is well suited to statistics and graphical applications. You can visit the R website here. (To produce this graph you do not need to learn R, you can do it by cutting and pasting.)

Once you have downloaded R you need to save your SPSS file in a tab delimited format (.DAT). Do this using the Save as dialog in SPSS. A library is available to import SPSS files directly, but that is beyond the scope of this article.

Luckily R can cope with your data in Long or univariate format.

The code to produce the graph is below, you will need to edit the input path of the input and export files, and also change the field names from ID, Time, Rhubarb etc. Then just cut and paste into R and press return to generate the PDF


#Open Lattice Library
library(lattice)
#Load tab delimited file with headers for Field names
data1 <-read.table("D:/Documents and Settings/Nerd/Desktop/Rhubarb.dat",header = TRUE ,sep = "\t")
#Set output to PDF and set output path and filename
pdf("D:/Documents and Settings/Nerd/Desktop/Output.pdf",width=8)
#Plot data by Month
xyplot(Rhubarb ~ Time, groups = ID,
data = data1,
type = "l" ,xlab="Months from Study Start",ylab="Plasma Rhubarb mg/dl",
scales =list(x=list(at=c(0,3,6,12))))
dev.off()
#Close the PDF

You can specify the x-axis labels that are displayed by changing the list (0,3,6,12) , or let them be generated automatically by R by removing

scales =list(x=list(at=c(0,3,6,12)))

The output file in png format is shown below.

The real beauty of this technique is how easily you can increase the number of subjects in your graph by cutting and pasting more data and simply re-running the same script. A more complicated example (The example I was originally working on) is shown below:-