Figure 7.2How health boards compare

HANDLING ATTENTION POINTS but it will sometimes obscure blocs of data where several indices are shown together, unless they are very well designed. New users of D charts also may not realize that it is often tricky to achieve a consistent angle, orientation and appearance across a succession of D charts, making them harder for readers to interpret. This is also a key reason why journals and book publishers are less keen on them. By contrast Figure 7.3 uses a simpler two-dimensional (flat) format, which is much easier and quicker to design and implement in a consistent fashion. It is preferred by most journals and academic publishers, and is straightforward for readers to interpret.
All these differences between Figures 7.2 and 7.3 are generally applicable to every chart you have to design. The most important principles are:
◆
Always have completely informative headings and labelling,
including details of units of measurement. Appropriate labels and scales must be shown for both horizontal and vertical axes, and legend labels are needed wherever the chart includes multiple data series (shown in several lines,
bars or shadings 100 200 300 400 500 600 700
Orkney
Fife
Lanarkshire
Grampian
Shetland
Forth Valley
Western Isles
Dumfries and Galloway
Greater Glasgow
Lothian
Argyll and Clyde
Ayrshire and Arran
Highland
Tayside
Border
Cataract treatment rates per 100,000
Upper outliers
217 229 239 277 282 297 308 317 318 318 332 332 339 503 723
Upper quartile
Median
Lower quartile
Figure 7.3
How Scotland’s health boards compared in
treating cataracts, 1998–9 financial year

1 8 AUTHORING AP H D
◆
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 percent of the observations. Never let the scale beset 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) overtime 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 soon, 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 overtime 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 beheld on spreadsheet and presentation packages, separate from your word-processed main text. There are good reasons for this, notably avoiding creating very long document files which cannot then be backed upon 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.

HANDLING ATTENTION POINTS Other techniques for data reduction
The only way to grasp a mathematical concept is to see it in a multitude of different contexts, think through dozens of specific examples, and find at least two or three metaphors to power intuitive speculations.
Greg Evans
6
To present data well you have to really understand them. And to do that, you have to look hard at them fora longtime, and ask an array of well-thought-out questions about what they show. Yet modern PCs and software give all of us the ability to crunch far more numbers than perhaps we have fully analysed,
and then to inflict them on our readers in an undigested way.
Data reduction means simplifying the numbers we are working with. The field of exploratory data analysis offers many powerful techniques for doing this, and has an interesting literature which I will only briefly touch on here.
7
Properly exploring and reducing data is an essential principle for making progress in understanding any set of numbers that you have to analyse, let alone conveying that information accurately to readers. The key principles of data reduction are:
◆
Look hard at your primary data. Do not rely on analysis packages to give you an intuitive picture of what you are dealing with or to tell you what questions to ask. Analysis packages can only work well for you if you already know what shape of data you have. This is easy enough in coursework where you are replicating someone else’s prior analysis, but often very difficult for brand-new information that you have just generated by research.
◆
Always put your data in a numerical progression (easily done in any spreadsheet. Chart them wherever possible,
and then look hard at the results. Never engage in more complex forms of multivariable analysis, such as correlations or regression analysis, without understanding the visual shape of the primary data you are handling.
◆
When trying to see patterns in your data remove as much of the clutter as possible. For instance, try looking at aversion div

---

Download 2.39 Mb.

Share with your friends: