Discriminant analysis
Predict which group a person is in from several numeric scores.
When to use it
Use discriminant analysis when the outcome is a group and the predictors are numeric. For example: can recall and confidence tell lecture students from workshop students? Logistic regression answers a similar question for two groups. Discriminant analysis reports a function, group centroids, and a classification table.
Assumptions
Predictors are numeric. The groups have similar covariance. Tensr does not test that. With two groups there is one discriminant function.
Running it in Tensr
Options
Prop
Type
Reading the output
Teaching method predicted from exam score and study hours. The groups overlap, so classification should be better than chance and well short of perfect.
| Group | LD1 | LD2 |
|---|---|---|
| Lecture | -0.438 | 0.082 |
| Online | 0.008 | -0.168 |
| Workshop | 0.429 | 0.087 |
| Functions | Wilks' λ | χ² | df | p-value |
|---|---|---|---|---|
| 1 through 2 | 0.876 | 12.44 | 4 | .014 |
| 2 | 0.986 | 1.323 | 1 | .250 |
Linear discriminant analysis of method using 2 predictor(s). Linear discriminant analysis of method using 2 predictor(s). Metrics: Classification accuracy = 44.8%; Groups = 3; Cases = 96; Chance accuracy = 33.3%.
Reporting (APA 7)
Linear discriminant analysis of method using 2 predictor(s). Report the estimate in the table. This procedure is not summarised by one p-value.
Coming from SPSS
Analyze → Classify → Discriminant.
SPSS prints eigenvalues, Wilks’ lambda, standardized coefficients, and a classification table. Tensr prints centroids, Wilks’ lambda, the classification counts, and the coefficients.
Related
A yes-or-no outcome with odds ratios is logistic regression. Several outcomes at once, rather than a group, is MANOVA.