Special features

Analysing data — choosing appropriate statistical methods

By Sarah L. Vowler, MSc

There are many different statistical tests that can be used to analyse data. When reporting results it is important that the tests used are appropriate for the type of data that have been collected. This article, the first in a special feature on statistics, describes the different types of data and the tests used to analyse them

This article as a PDF (410K)

Sarah Vowler is a medical statistician at the Centre for Applied Medical Statistics, Department of Public Health and Primary Care, University of Cambridge

R Maisonneuve, Publiphoto Diffusion/SPL

Analysing data

Panel 1: Assumptions

An assumption is an assertion that is assumed to be true for the test to be valid. Each statistical test has it own set of assumptions. It is important to check what the assumptions of a particular test are before carrying out the analysis, and to check that the data meet the assumptions.

For example, the chi-squared test assumes that the data are counts, that the two variables are nominal, observations are independent of each other, 80 per cent of the expected counts exceed five and all exceed one. Standard texts will list the assumptions of statistical tests and methods.

Knowing the assumptions of the particular test should help you decide whether the test is appropriate for the data in question. If the assumptions are not met, the test is invalid and any conclusions drawn from the results of the test may not be correct.

SUMMARY

The choice of which statistical test to use to analyse a set of data is completely dependent on the type of data that have been collected. This article describes how different types of data are categorised and which tests should be used to analyse them.

The following article in this feature (p47) will describe how to assess and interpret clinical papers, depending on the way results have been analysed and presented.

There are many different statistical methods that can be used in different situations. Each test makes particular assumptions about the data, as described in Panel 1 (right). These assumptions should be taken into consideration when deciding which is the most appropriate test.

When analysing data it is useful first to get a feel for the data, what distributions the variables have and what inter-relationships exist. This can be done pictorially using scatterplots for comparing two continuous variables, matrix scatterplots for several continuous variables, histograms for individual variables and dot plots or boxplots for continuous variables between groups. Categorical data can be plotted using pie charts or bar charts.

Full text article PDF (410K)