Analyses

Latent class analysis

Find groups of people who share a response pattern.

Edit on GitHub

When to use it

Use latent class analysis when you think the sample is a mixture of a few types, and you only see their answers. Each person gets a probability of belonging to each class. The class profiles say what a typical member answered.

Assumptions

Indicators are numeric. On 0/1 items Tensr fits a binary measurement model. On other numbers it treats the indicators as continuous. Rows with a missing indicator are dropped. You need at least five rows per class. The fit uses a fixed random seed, so the same data and the same class count give the same solution. If a class ends up empty, the run stops instead of printing a zero-size class.

Running it in Tensr

Multivariate → Latent Class Analysis. In chat: “Two latent classes from these three items.”
Select the indicators. Set the number of classes from 2 to 10. The default is 2.

Options

Prop

Type

Reading the output

Four yes/no items scored 1 and 0, and two latent classes. One class was built to endorse the items more often. Membership is probabilistic, not a clean split.

Classn%
19760.6%
26339.4%
IndicatorClass 1Class 2
q10.23710.8254
q20.29900.7778
q30.09281.0000
q40.43300.5873

Latent class analysis with 2 classes on 4 indicators. Latent class analysis with 2 classes on 4 indicators. Metrics: Classes = 2; Observations = 160; AIC = 842.435; BIC = 870.111; Entropy = 19.841; Log-likelihood = -412.217.

Reporting (APA 7)

Latent class analysis with 2 classes on 4 indicators. Report the estimate in the table. This procedure is not summarised by one p-value.

Coming from SPSS

The product menu is Multivariate → Latent Class Analysis. SPSS does not ship this under Analyze. The profile chart is one bar group per class.