186 ◆AUTHORING A PHD
that cuts out confusing and unnecessary decimal points or
where numbers are rounded. And transform your data using
index numbers or ratios so as to put the data in number
ranges that are most easily understandable, ideally between
0 and 100. Operating with unsimplified numbers (especially
very large or very small ones) will make it more difficult for
you to find patterns in them.To get more of a fix on exploratory data-analysis techniques, I
briefly consider three useful approaches: stem-and-leaf analysis
(including measures of central level and spread); box-and-
whisker plots; and data-smoothing for over-time graphs.
Stem-and-leaf analysisis a simple technique for looking hard
at a set of data. Suppose that some data collection you have
done generates the following 27 numbers for a particular vari-
able (in the random order of their occurrence in your data set):
One way to analyse these data would be as a bar chart or
frequency count. Here we could set up some category boxes and
count the number of cases in each, yielding a result like this:
This pattern looks like a conventional single-peaked one (the
misleadingly termed ‘normal’ distribution, popularised as ‘the
bell curve’). But we have lost a lot of information here about
the precise numbers in the original data, and may be missing
a trick as a result.
Category No. of cases
50 1
40–9 1
30–9 7
20–9 9
10–19 8
0–9 125 46 52 29 15
23 22 18 12 33
19 22 34 19 22
34 18 31 17 3
19 22 21 32 20
32 33