|
Doing research-an overview |
|
|
|
About the Author: Professor David CA Candy, Consultant Paediatric Gastro-Enterology and Nutrition, Royal West Sussex NHS Trust, Chichester, Visiting Senior Lecturer, Department of Child Health, University of Southampton and Visiting Professor, University of Chichester
The information in this article is correct at date of publication: September 2009
Opinions expressed by the author are not necessarily those of the publisher or editorial staff.
In the first part of this 3-part series Professor Candy introduces the subject of research and the value of a grounding in statistics. The second and third parts in the series cover ‘Why do research’ and a detailed ‘How to do your own research’ (Click on the links to access). Q. How do I make sense of statistics? It is tempting to skip over the statistical sections of papers – they don’t always look easy to read. Remember science is rarely black and white, and statistics help make sense of it. Statistics are all about confidence and probability. The aim is to determine the likelihood of a real difference between outcomes rather than a difference due to chance. A rudimentary knowledge of statistics is required to read the scientific literature. Those in the happy position of contemplating their own project can get away with leaving statistics up to the experts.
Perhaps the most important piece of advice in this article is to seek statistical advice at a very early stage. You will not please your friendly, local statistician if you collect your data and dump it onto their desk, saying ‘analyse this!’ Your first question should be ‘how many research subjects are required to get a clear-cut result (power calculation)? The result of this will demonstrate the feasibility of the study and whether you should consider a multi-centre study. Welcome the statistician into your team and offer a co-authorship if your work is published.
You will find statistical help at your Research Design Service at Universities in each Strategic Health Authority, as well as support in all aspects of designing your project. A good place to start is (www.nihr.ac.uk/infrastructure/Pages/infrastructure_research_design_services.aspx).
However, before you throw yourself at the mercy of the RDS statistician, it pays to learn the lingo, to ensure a worthwhile meeting. You’ll see from the following that it is much better to consult the experts than trying to do statistics yourself.
Q. How can I talk the statistical talk?
You will come across several terms when reading research papers, the most common ones being listed in the table overleaf. The easiest observations to analyse are called ‘normally distributed’. This means that if you plot the frequency of your observations you obtain a ‘top hat curve’. Many measurements like height and IQ are distributed in this way.
The Figure below gives an example of two sets of measurements with the average or mean values shown by the vertical lines at A and B. How can we tell whether the measurements are statistically different? We compare these sorts of measurements with a ‘t’ test. If the measurements are taken from two different groups of people we use an ‘unpaired t test’. If they are from the same group of people measured on two different occasions we use a ‘paired t test’. Stay with me, we’re nearly done!
If the information does not consist of numerical values such as kg or cm (for example the proportion of patients with a disease in two populations), the values are called non-parametric and we use different tests such as the chi square (or X2) test or Mann-Whitney U test.
Either way, the magic number in statistics is 20. If the chance of something happening is less than 1:20 then when it occurs it is statistically significant. This is expressed as a probability of less than 0.05 (P<0.05).
Of course, there’s more to research than getting a P<0.05. When evaluating a study that shows a statistically significant effect, subject the result to a reality check. Think – will a change of the magnitude described make it worthwhile changing my patients to this treatment? What is the clinical significance of the result? Are the results generalisable to my patients?
I think we’ll stop there – you have enough technical terms to drop in to convince your statistician to treat you with respect! The next sections of this article can be found here and discuss the background behind research and offer an introduction to carrying out research.
To download a PDF of the `Common terms used when discussing research`1 click here References
1. Journal of Family Health Care, Volume 15, No 3, 2005 Special Supplement 1 (Click here to download a PDF of the Bulletin)
