HerfstkleurenHelpdesk IBM SPSS Statistics 20

Introduction Sample size Table design Graph design Syntax Testing Links SPSS Statistics 20

Testing Choosing the right test

What do you want to do?
A "person" you could ask for help is the SPSS Statistics Coach. This is part of the SPSS help system and it helps you in choosing the proper statistical technique for your problem. It guides you via a series of questions related to the situation you are in and the things you want to do to the best technique and shows you also where to find it in SPSS.

Let us have a closer look at some of the questions the statistics coach will ask you.

1.    What is the design of your research? Are you dealing with a:

  • One sample case: You have collected sample data from a single population regarding a variable X.
  • Two independent samples case: You are dealing with two independent groups and you have measured variable X for samples from both groups.
  • Two related or matched samples case: You are dealing with matched pairs of observations. Variable X is measured for two linked objects/subjects or in two different situations for the same object/subject.
  • k independent samples case: You are dealing with several independent groups and you have measured variable X for samples from all groups.
  • k related samples case: You are dealing with matched groups of observations for variable X or repetitions of measuring the same variable X for the same case.

2.    What is the level of measurement of the variable X you are interested in? Is it:

  • Nominal: There is just a list of answer options for variable X.
  • Dichotomous: There are two answer options (like yes/no, right/wrong, +/-). Note that this is a special case of a nominal variable.
  • Ordinal: There is a list of answer options for variable X and there is a logical ordering of these answer options. This allows you to rank the answers for X.
  • Scale: The answers for X are actual numbers. Adding and subtracting answers makes sense. You can work for example with averages and differences.

3.    What do you want to test? Is it about:

  • One ore more population means.
  • One or more population proportions.
  • One or more population variances.
  • The distribution of a population or of several populations.
  • Something else.

4.    Do you have additional information about the distribution of the variable X for your population?

  • The main question here is whether or not we may assume that X has approximately a normal distribution.
  • A possible alternative question is whether or not the sample is large enough, so we can use a limit theorem to ensure that a certain distribution (like a normal or a chi-square distribution) may be used.

You already see the next remark coming.
There are a lot of combinations of answers possible with these four questions. This implies that there are a lot of tests possible. Some of them are summarized in a scheme for testing (Word, 73 KB).

If your survey design is not in this scheme and the SPSS statistics coach can't help you either, consult a book, an expert or look on the internet for help.

to the top

Last modified 30-10-2012

© Jos Seegers, 2009; English version by Gé Groenewegen.