184 ◆AUTHORING A PHD
◆ Use the ‘need to know’ criterion to pick an appropriate level
for numbers, so that the chart can be easily scaled. Make sure
that the charts are large enough to show clearly any
important features mentioned in the accompanying analysis.
◆ Scale charts so that variations are still apparent in the
middle mass of data (the middle 50 per cent of the
observations). Never let the scale be set just to accommodate
one or two extreme observations, untypical of the rest of
the data. Try not to suppress the zero.
◆ Allocate axes appropriately. Use horizontal bar charts where
long bar labels are needed. In scatterplot charts, always
allocate the horizontal (X) axis to the independent (causing)
variable, and the vertical (Y) axis to the dependent variable
(the one which is being caused or influenced).
◆ Design all line and bar charts with a numerical progression
in them – except for two special cases where you are
showing (i) over-time trends, or (ii) categorical data which
have to be kept in fixed order to be meaningful. Pie charts
should generally have a numerical progression also, with
the largest pie starting at the upper vertical and the wedge
going right and downwards, followed by the second largest
wedge, then the third largest, and so on, all going
clockwise. The only exception here is a pie chart showing
fixed-order categorical data. (Of course, you should never
use pie charts to show over-time data.)
Overall, the most important test for charts and graphs is to try
and ensure that each of them is independently intelligible to
readers who have not lived with the data for months or years, as
you will have done by the time that the thesis is printed and
bound. Again make sure that your charts are revised and updated
with your main text as it changes. Far more often than tables,
charts tend to be held on spreadsheet and presentation pack-
ages, separate from your word-processed main text. There are
good reasons for this, notably avoiding creating very long docu-
ment files which cannot then be backed up on diskettes. But it
does mean that stronger version control problems can arise
unless you are careful to keep charts and their accompanying
main text passages in close agreement. All charts should clearly
show what the main text says that they show.