Analyses

Canonical correlation

Relate two sets of columns, not just one pair.

Edit on GitHub

When to use it

Use canonical correlation when each side has more than one column. For example: how does the set of study hours and anxiety line up with the set of recall and confidence? Bivariate correlation only looks at one pair at a time. This finds a weighted combination on each side that correlates as strongly as possible, then a second combination, and so on.

Assumptions

The columns are numeric. The first canonical correlation is the one to read. Later functions are leftover association after the first is removed. Tensr does not test multivariate normality.

Running it in Tensr

Analyze → Correlate → Canonical. In chat: “Canonical correlation of hours and anxiety with recall and confidence.”
Set A and set B each need at least one column.

Options

Prop

Type

Reading the output

One set is exam score and confidence. The other is hours and anxiety. The sets share a moderate link and a lot of leftover variance.

FunctionWilks' ΛFdfp-value
Function 10.73922.1714< .001
Function 20.9643.521.064
VariableFunction 1Function 2
score0.996-0.089
confidence0.2770.961

Canonical correlation: score, confidence ↔ hours, anxiety, n = 96. Function 1 accounts for 23.3% of shared variance between the two variable sets. Canonical correlation: score, confidence ↔ hours, anxiety, n = 96. Function 1 accounts for 23.3% of shared variance between the two variable sets. Metrics: Canonical r1 = 0.483; Canonical r2 = 0.191.

Reporting (APA 7)

Canonical correlation: score, confidence ↔ hours, anxiety, n = 96. Function 1 accounts for 23.3% of shared variance between the two variable sets. Report the estimate in the table. This procedure is not summarised by one p-value.

Coming from SPSS

Analyze → Correlate → Canonical.

SPSS canonical correlation is a macro in many installations rather than a base dialog. Tensr’s menu path is Analyze → Correlate → Canonical.

One pair of columns is bivariate correlation. Grouping people from several scores is discriminant analysis.