Dear Statalisters,

I've found that a fairly simple crossed random effects multilevel model I estimated with Stata/MP 16.0 does not work anymore after updating to Stata/MP 16.1, if the sample size is above 50 000. The model has grade point average as the outcome and the levels are school, neighbourhood and family. The school effects are crossed with other levels. There are 220 schools and 298 neighbourhoods, 43 171 families and 53 976 children in the data (children nested in families nested in neighbourhoods). The syntax below (no predictors in this example) has worked previously but after the upgrade to 16.1 it returns error r(3203) below. I have also tried to use version control in front of the command but the problem persists.

However, I noticed that when I reduce the data by random sampling so that it includes 49 999 cases (children), the model runs without issues. Do you have any idea why this occurs and is there some setting that I should tweak? It seems that others have had similar problems (probably after upgrade) but I they haven't pointed the issue to the data size (see link below).

Edit: I'm running the analysis on a server with 64GB of ram, so it's probably not related to memory issues.

Code:
mixed gpa || _all: R.school || neighbourhood: || family:

MapRowsToIntegers(): 3203 <tmp>[1,53976] found where colvector required
GenGroupSampleNumeric(): - function returned error
<istmt>: - function returned error
r(3203);

https://www.statalist.org/forums/for...l-error-r-3203