Hello,   The question is simple but it has been driving me crazy. I have a large dataset (~150GB) with billions of observations. I would like to perform a linear regression on a list of both continuous and categorical variables (which contains thousands of categories), and keep the observations that make X'X matrix non-singular (invertible).   In Stata it is simple because Stata automatically drops observations so that X'X is always invertible. All I need to do is to use e(sample) to get the observations that are used in the regression, like the following:     reg y x1 x2 i.x3 keep e(sample)     However, in SAS it seems like most procedures keep all observations and, when X'X appears to be singular, perform a generalized inverse. That is not what I want, but I cannot find any options in proc reg or proc glm to suppress such behavior. I cannot do it in Stata as well because the dataset is too large to fit in the memory.    Does anyone have suggestions on how to do this?   Thank you in advance. ... View more