Yes, sample size can differ across paths in a path analysis. Path analysis is a form of structural equation modeling (SEM) that examines the relationships between variables in a causal model. In path analysis, each path represents a relationship between two variables, and the sample size used to estimate each path can differ.

There are a few scenarios where this might occur:

However, it's essential to note that differing sample sizes across paths can impact the accuracy and reliability of the path estimates. To address this, you may need to:

It's crucial to carefully evaluate the impact of differing sample sizes on your path analysis results and consider appropriate methods to address any issues that arise.

Hi, as you use lavaan, you can use the FIML estimator that deals with the missing data.

You specify that by using missing = "FIML" in the sem() function.

Best,

Holger

Holger Steinmetz Thank you for your answer! Can I combine missing="FIML", with estimator = "MLR"? And what does FIML do differently compared to computing the 410-outcome coefficients using only the 410 cases? Also, if I were to use FIML, would I report 558 as the sample size for the mediation analysis? Best, Lina

Hi Lina,

yes FIML and MLR (Yuan-Bentler correction) can be combined while this is not the case for MLM (Satorra-Bentler correction).

Unfortunately, I cannot say what is done statistically behind the curtain :) But yes, applying FIML uses the whole sample. Hence, your N will be the total N.

I don't know whether this will still make sense for you given your progress but I have a small book on doing SEM with lavaan:

Steinmetz, H. (2014). Lineare Strukturgleichungsmodelle: Eine Einführung mit R (Sozialwissenschaftliche Forschungsmethoden). München: Hampp.

HTH

Holger