Which Statistical test?

Introduction

Irrespective of the statistical package that you are using, deciding on the right statistical test to use can be a daunting exercise. In this document, I will try to provide guidance to help you select the appropriate test from among the many variety of statistical tests available. In order to select the right test, the following must be considered:

  1. The question you want to address.
  2. The level of measurement of your data.
  3. The design of your research.

Statistical Analysis (Test)

After considering the above three factors, it should also be very clear in your mind what you want to achieve.

If you are interested in the degree of relationship among variables, then the following statistical analyses or tests should be use:

Correlation

This measures the association between two variables.

Regression

Simple regression - This predicts one variable from the knowledge of another.
Multiple regression - This predicts one variable from the knowledge of several others.

Crosstabs

This procedure forms two-way and multi-way tables and provides measure of association for the two-way tables.

Loglinear Analysis

When data are in the form of counts in the cells of a multi-way contingency table, loglinear analysis provides a means of constructing the model that gives the best approximation of the values of the cell frequencies. Suitable for nominal data.

Nonparametric Tests

Use nonparametric test if your sample does not satisfy the assumptions underlying the use of most statistical tests. Most statistical tests assumed that your sample is drawn from a population with normal distribution and equal variance.

If you are interested in the significance of differences in level between / among variables, then the following statistical analyses or tests should be use:

  • T-Test
  • One-way ANOVA
  • ANOVA
  • Nonparametric Tests

If you are interested in the prediction of group membership then you should use Discriminant Analysis.

If you are interested in finding latent variables then you should use Factor Analysis. If your data contains many variables, you can use Factor Analysis to reduce the number of variables. Factor analysis group variables with similar characteristics together.

If you are interested in identifying a relatively homogeneous groups of cases based on some selected characteristics then you should use Cluster Analysis. The procedure use an algorithm that starts with each case in a separate cluster (group) and combines clusters until only one is left.

Conclusion

Although the above is not exhaustive, it covers the most common statistical problems that you are likely to encounter.