Latent profile analysis

Author

DL Oberski

Latent profile analysis (a.k.a. Gaussian mixture modeling)

Exercise

An interesting example of a latent profile analysis using tidyLPA is provided, with data and R code, by Wannheden et al. (2021), at https://snd.gu.se/en/catalogue/study/2022-101#dataset.

(The paper (Wannheden et al. 2022) is available open access at: https://www.frontiersin.org/articles/10.3389/fpubh.2022.960383/full.)

  1. Go to https://snd.gu.se/en/catalogue/study/2022-101#dataset and download (at least) the files:

    • factorscores_docca.csv;
    • R-script_Step-1_Latent-profile-analysis.R.

    You should save the two files in the same directory.

  2. Now open R-script_Step-1_Latent-profile-analysis.R and step through the code until line 86 (plotting of the fit information). Which model is preferred by the various fit measures? Which model was ultimately chosen by the authors? Why?

  3. Use plot_profiles(t1m2p4) to plot the profiles of the first model. Compare this with Figure 1 in the paper. What is the main difference? Which do you prefer?

  4. Examine Table 3 in the paper. See if you can follow the authors’ rationale in labeling the classes.

  5. Now perform the LPA by using Mclust (from the mclust library) directly on t1_scaled. Try 1-10 classes and assign the result to a varible fit_mclust. Which model is selected?

  6. Plot the fit_mclust object:

    • Confirm that BIC selects the model from the previous part
    • Look at the classification (pairs) plot. What drives the clustering?

  7. Use clustCombi(fit_mclust) and call plot on the resulting object.

    1. Examine the entropy “scree plot”. Note the presence of any “elbow”.
    2. Now choose “Classification” and examine the sequence of merged clusters.

    Which step will likely give the most substantively interpretable solution?

    1. Use ClustCombiOptim to choose a merged clusters-solution. Which number is chosen?

  8. BONUS Use mclustBootstrapLRT with model = “VEE” to select the number of classes. Which number is selected?

  9. BONUS Create a pairs plot in which the points are colored by class assignment of the Wannheden et al classes. What do you see?