Which statistical test?

Categorical could be confusing, only because people have made it so. Statistics books describe different sub-types, which are not of much practical use (in analysis). But you may need to know these for examinations. – Nominal : named groups, depending on various characteristics. One group is not 'better' than another. E.g. race, blood group, etc. – Binary: similar to above, but only two groups. E.g. dead/alive, cured/not cured, etc. – Ordinal/ordered/ranked: groups are ordered and each group has a rank. E.g. social classes, age groups, etc. – ('qualitative', 'discrete')1


Asoka Weerakkodi
As there is a bewildering variety of statistical tests, it could be very difficult for the clinician to know which particular test one should use for a given study.Unfortunately, most books written by statisticians are not very helpful to clinicians.
We need a simple practical guide.To understand statistical tests, one needs to know about data and different data types.Which test you use depends very much on the type of data you have (rather than the type of study you have done).

Data types
For practical purposes, there are only two: Categorical: where you count (whole numbers only).E.g. number of people, or events.
When it comes to different data types, Continuous is easy; there are no sub-types.'Continuous' is continuous -that's it.
Categorical could be confusing, only because people have made it so.Statistics books describe different sub-types, which are not of much practical use (in analysis).But you may need to know these for examinations.
-Nominal: named groups, depending on various characteristics.One group is not 'better' than another.E.g. race, blood group, etc.
-Binary: similar to above, but only two groups.E.g. dead/alive, cured/not cured, etc. -

Continuous data
For continuous data, it is a bit more complicated; which test you use depends on certain conditions:

Data normally distributed:
For paired 4  If you measure the VC of asthmatics before and after using an inhaler, the second set of readings is dependent on the first.They are called 'paired samples'.
Paired samples are less subject to variation (as each individual acts as its own control), and hence needs less numbers than unpaired samples to show significance.

Comment
There are a large number of other statistical tests for various situations, but these are beyond the scope of the average clinician.In practice, it is unlikely that you would use, or be required to know about, anything other than the t test and the chi square test.
My sincere thanks are due to Mrs T S Usha Kiran, Consultant Obstetrician and Gynaecologist at the Royal Gwent Hospital, Newport, South Wales, UK, for valuable advice and for simplifying seemingly complex statistics.

[Practical point: we could convert continuous data into categorical by 'grouping' them, but not vice versa] 2 . Which statistical test? Asoka Weerakkodi 1 Sri Lanka Journal of Obstetrics and Gynaecology 2010; 32: 106-108 Continuous Professional Development
When it comes to analysis, categorical data is easy.Essentially there is only one test: 3 2.You must insert actual numbers into the cells of the contingency table, and not percentages or proportions.3. Number in each cell must not be less than 5; in which case you should use a variant called the Fisher's Exact Test (in practice, the computer packages assign this test automatically).

Practical points: If the number of observations is >100, assume normal distribution, and non-normal when it is <30. Between 30-100, one has to prove normality by various tests or by 'eye-balling'. Sometimes it is possible to convert non-normal into normal by manipulating data, e.g. by taking the reciprocal or the logarithm.]
data, use Student's paired t-test; These are parametric tests 3 .

Summary: which test for which data?
For Categorical, use Chi Square test.For Continuous, see chart below.
P P Pa a ar r r a a am m me e et t tr r ri i ic c c t t te e es s st t ts s s