Latent Profile Analysis (LPA)

Latent profile analysis (LPA) is a finite mixture model that identifies unobserved subgroups of individuals who share similar patterns of means on a set of continuous indicators.

잠재프로파일분석(LPA)은 여러 연속형 지표에서 평균 패턴이 비슷한 관측되지 않은 하위집단을 찾아내는 유한혼합모형이다.

How do I identify hidden subgroups of people based on several continuous measures?

여러 연속형 척도 점수로 숨어 있는 하위집단을 어떻게 찾나요?

Latent profile analysis fits mixture models with increasing numbers of profiles and assigns each person a probability of belonging to each profile. It fits person-centered research questions where you suspect distinct response patterns across several continuous scale scores.

When to use Latent Profile Analysis (LPA)

• Several continuous scale scores measured for each respondent
• Person-centered research questions about typologies or profiles
• Choosing the number of subgroups using information criteria and likelihood ratio tests

When to use something else

• The indicators are categorical: use latent class analysis
• A quick partition without a probabilistic model is enough: use k-means or hierarchical clustering

Assumptions and requirements

• Indicators are continuous and approximately normal within profiles
• Local independence or a specified within-profile covariance structure
• Sufficient sample size so the smallest profile is not trivially small
• Multiple random starts to avoid local maxima

What the analysis reports

• Fit indices for each solution: AIC, BIC, sample-size-adjusted BIC
• Entropy and average posterior probabilities
• Likelihood ratio tests such as the bootstrap LRT or Lo-Mendell-Rubin test
• Profile sizes and profile means on each indicator
• Profile plot

How to report it

A [k]-profile solution was retained based on the lowest BIC ([value]), a significant BLRT (p = [value]), and an entropy of [value].

Related methods

• Latent Class Analysis (LCA)
• Gaussian Mixture Model
• K-Means Clustering

Run Latent Profile Analysis (LPA) on Quorant