There seems to be an issue with the "effects" option when using xsmle. "Effects" provides direct, indirect, and total effects for spatial autoregressive models. However, the total effects reported are slightly off. Total effects can be calculated within Stata in another way, via the "margins" command. They can also be calculated manually using a regressor's estimated coefficient and the estimated value of rho (the coefficient on the spatially lagged dependent variable). I can get the correct total effect via margins and by hand, but when "effects" is used, numbers are off by a small amount. See below. The total effect using "effects" is 2.7038, while the total effect using "margins" is 2.6988. Other situations I've encountered show larger disparities. Not being able to trust "effects" is a problem because direct and indirect effects are not available through the margins command and are very difficult to calculate by hand.

Please let me know if you have a guess about what might be happening here.
Array