Choosing a statistical test

Why do researchers use statistical tests?

to determine whether a difference or an association found in a particular investigation is statistically significant - that is more than could have occurred by chance. The outcome of this leads the psychologist to accept or reject the null hypothesis

What are the 3 factors to consider when choosing a statistical test?

1. whether you’re testing for a difference or a correlation

2. the design of the investigation: independent groups (unrelated) OR repeated measures/ matched pairs (related)

3. what levels of data you have

What are the 3 levels of data?

nominal data, ordinal data and interval data

What is nominal data?

Data placed into categories with no numerical order or equal intervals, in which ppts will only appear in one category

Give two examples of nominal data.

Gender; attachment type; yes/no responses.

What is ordinal data?

Data that can be ranked or ordered but with unequal intervals between values. It lacks precision as it is subjective

Give two examples of ordinal data.

Questionnaire ratings on a 1–5 scale; finishing positions in a race.

What is interval data?

Data measured on a scale with equal intervals between values. It is more of an objective measure and usually consist of a measure that is pre-existing

Give examples of interval data.

Temperature (°C), height in cm

How do researchers decide which test to use?

What tests are collectively know as parametric tests?

the related t-test, unrelated t-test and Pearson’s r

Why are parametric tests good?

they are more powerful and robust than other tests. These tests may be able to detect significance within some data sets that non-parametric tests cannot

What are the 3 criteria that must be met in order to use a parametric test?

1. data must be interval level - they use the actual scores rather than ranked data

2. the data should be drawn from a population which would be expected to show a normal distribution for the variable being measured. Variables that would produce a skewed distribution are not appropriate for parametric tests

3. there should be homogeneity of variance - the set of scores in each condition should have similar dispersion or spread. One way of determining variance is by comparing the standard deviations in each condition; if they are similar, a parametric test may be used. In a related design it is generally assumed that the two groups of scores have a similar spread