Factor Analysis w/ Missing Data

[Allison Cusano]

I am using Stata 18.0 and want to run an exploratory factor analysis on a data set with missing values. N=10,118 There are 243 numeric variables that represent scores on test items that measure program quality and 0 complete cases. Variables are either n=1,131 or n=3,928. This is a result of using a different tool to collect data on the same construct (class portfolio, program portfolio, class observation, program observation). All four methods contribute to the overall score across constructs.

I ran the following:

factor x1b_3-x10f_4, ipf

and received an error:

no observations

This makes sense because of all the missing data.

Ideally, I'd like to determine construct validity for the tool overall, but am unsure how to manage the missing data. I'd appreciate any thoughts re: code for imputing missing values or insight into how else I might ensure the tool measures what it intends to! Thank you all!

[Felix Bittmann]

I have found this approach here: https://stats.oarc.ucla.edu/stata/fa...data-in-stata/

However, your case is a bit peculiar since no case has all variables observed. I would suggest checking for missingness patterns. Potentially, some variables cannot be used at all.

[Allison Cusano]

Felix,

Thank you so much for the helpful resource. I committed a Statalist faux pas and did not mention having run something similar - which essentially timed out. I think your conclusion is correct - I cannot use some variables. I am also going to try running each data collection method on its own. Thank you again for taking the time to respond.

-Allison

[Erik Ruzek]

I cannot imagine working with the output of a factor analysis on 243 items. I think your approach of breaking the items into meaningful chunks is a good one that will ease not only the EM computation suggested in the UCLA link but also interpretation on your end.