Analyses

Multidimensional scaling

Place cases on a map so that distances reflect how different they are.

Edit on GitHub

When to use it

Use multidimensional scaling when you want a picture of cases, not a picture of variables. People who gave similar answers land near each other. Two dimensions is the usual map. This is not a factor analysis of the items.

Assumptions

Columns are numeric. Distance is Euclidean after listwise deletion. The request does not standardize the columns, so a column with a much larger scale dominates the distances. You need at least two more complete rows than the number of dimensions.

Running it in Tensr

Analyze → Scale → Multidimensional Scaling. In chat: “MDS of these four items in two dimensions.”
Select at least two columns. Dimensions default to 2, from 1 to 10.

Options

Prop

Type

Reading the output

Five scale items placed in two dimensions from their distances. Items that share the factor should sit nearer each other than a pure noise pair would.

Multidimensional scaling on 5 variables, n = 120. Multidimensional scaling on 5 variables, n = 120. Metrics: Stress = 5333; Dimensions = 2.

Reporting (APA 7)

Multidimensional scaling on 5 variables, n = 120. Report the estimate in the table. This procedure is not summarised by one p-value.

Coming from SPSS

Analyze → Scale → Multidimensional Scaling (PROXSCAL / ALSCAL). SPSS stress is usually scaled to a 0-to-1 range. Tensr reports the unscaled stress from the fit.