"Auto" Estimator in SEM/Mediation module

Comments

• mpopa

  April 2020

  Hi, Kathi

  In the SEM the data must meet certain conditions, including the univariate normality of the observed variables. In most cases, at least in psychology and social sciences, the observed variables are often responses on Likert scales (categorical, with different levels, usually between 3 and 7). The distribution of the categorical variables does not respect normality, which could affects the results of the analysis. Here the estimation technique intervenes. For continuous variables (normally distributed), the appropriate estimation technique is Maximum Likelihood (ML). When the variables are not subject to normal distribution, different alternative techniques have been proposed, each with its advantages and disadvantages.

  In my opinion, when the indicators are evaluated on Lickert levels, it is preferable that instead of ML one of the other options to be chosen, for example. Weighted Least Square (WLS). My observations with JASP show that the Auto option favors the ML estimation, even if the observed values ​​are categorical.

  Here, some good source of information:

  https://www.youtube.com/watch?v=NmF1blf2zDI

  http://psychmodels.ucdavis.edu/uploads/8/6/3/9/86398102/rhemtullabrosseauliardsavalei_pm.pdf

  https://www.researchgate.net/publication/297604832_Comparison_of_Different_Estimation_Methods_for_Categorical_and_Ordinal_Data_in_Confirmatory_Factor_Analysis

  Cheers
• mpopa

  April 2020

  A short description af each estimation technique, here:

  http://lavaan.ugent.be/tutorial/est.html

  
  
• Kathi1227

  June 2020

  Hi,

  I saw your response only just now and wanted to say thank you! That's very helpful. :)

  
  

  Best,

  Kathi