Which statistical test should I use in psychology?

1. Choose the test

What is the study looking for?
What is the design?
What level of measurement are the data?

2. Sign test calculator optional

Differences in the predicted direction

Differences the other way

No difference: left out of N

Significance level
Hypothesis

AQA A-level Psychology 7182 inferential tests

Answer the questions

Check your research-methods reasoning

The eight tests AQA names, by purpose, design and level of measurement
DataDifference, unrelatedDifference, relatedRelationship
NominalChi-squared testSign testChi-squared (association)
OrdinalMann-Whitney U testWilcoxon signed-rank testSpearman's rho
IntervalUnrelated t-test parametricRelated t-test parametricPearson's r parametric

Your result

Ratings are usually ordinal. Frequencies in categories are nominal.

A-level Psychology Pro

Get an A or your money back

£129 once

Price rises to £199 on 1 January. Save £70 by joining before then.

At least 12 months' access

By paying you agree to our Terms and get access straight away, which ends your 14-day cancellation right.

No VAT is charged. Conditions

A-level Psychology uses 8 statistical tests, and you choose one by whether you test a difference or a correlation, the design and the level of measurement.

Questions people ask

When do you use a sign test?

When you are testing for a difference, the design is related (repeated measures or matched pairs), and the data are nominal, or are reduced to the direction of each difference: a plus, a minus or a tie.

What is the difference between Wilcoxon and Mann-Whitney?

Both test a difference with ordinal data. Wilcoxon is for a related design (the same or matched participants), Mann-Whitney for an unrelated design (independent groups).

Is S significant if it equals the critical value?

Yes. For the sign test, S must be equal to or less than the critical value to be significant. For example, with N = 12 at p ≤ 0.05 one-tailed, the critical value is 2, so S = 2 is significant.

When do you use chi-squared?

When the data are nominal (frequencies in categories) and each participant appears in only one category: an unrelated design testing a difference, or a test of association between two variables.

How we work this out

What it does

Answer three questions to name the test among the eight AQA lists, with the reason for each step, and work out a sign test with its critical value.

The eight inferential tests named in the AQA A-level Psychology 7182 specification (section 3.2.3.3): sign test, Wilcoxon, related t-test, chi-squared, Mann-Whitney, unrelated t-test, Spearman's rho and Pearson's r, and the calculation of the sign test.

Method

  1. The test follows from three decisions, in the order AQA's table sets them out: whether the study looks for a difference or a relationship, whether the design is related or unrelated (for a difference), and whether the data are nominal, ordinal or interval.
  2. Interval data that do not meet the parametric conditions (a roughly normal distribution, and similar variances for unrelated groups) are tested at the ordinal level.
  3. Specific checks distinguish normal paired differences from normal independent groups, check similar variances for the educational unrelated t-test route, and distinguish straight-line from monotonic relationship patterns. Explicit non-independent cases, invalid categorical inputs or a pattern that changes direction stop the recommendation. Unanswered checks remain visible actions and are never treated as confirmations.
  4. Sign test: N is the number of pairs with a difference (ties are left out) and S is the number of the less frequent sign.
  5. The critical value is the largest S whose exact binomial probability, with a half chance of each sign, is at or below the significance level (halved first for a two-tailed test). This is how printed sign-test tables are made. The result is significant when S is equal to or less than the critical value.
  6. For a directional (one-tailed) hypothesis the pluses are the predicted direction; when the minuses outnumber them, the result cannot support the hypothesis.