There seems to be an issue with the calculation of marginal effects for spatial models containing a spatially lagged dependent variable (e.g. SAR, SAC, and SDM models). As far as I can tell, there are two ways to calculate marginal effects: (1) including the "effects" option when estimating the model and (2) using the post-estimation command "margins." These should result in the same marginal (total) effect, right? Specifically, the total marginal effect for a given regressor should = (that regressor's estimated coefficient)*[1/(1-rho)] because spatial spillovers produce a geometric series. The use of "margins" gives this result exactly, but the use of "effects" gives a result that is close, but slightly larger. See below for a simple illustration of an SAR model with only one regressor. What is going on here? Why is "effects" not giving the anticipated result? I need to figure this out because, as far as I know, direct and indirect effects can only be obtained using "effects" (not "margins"). Please help. Array
Related Posts with Spatial Model (xsmle) "Effects" vs. "Margins" - Why are they different?
Panel Data AnalysisHello How should the final estimate be in a model with the following conditions? Panel Data Random …
Obtaining a forest plot showing the standardised mean differences after teffects ipw and aipwDear Statalisters, I have been trying to produce a forest plot-style of graph to show the mean in t…
Using FEI am trying to carry out panel FE analysis using .xtreg. For cluster SE do I need to create a differ…
How to test whether sum of coefficients is significantDear Stata users, I have a question regarding the interpretation of my coefficients. I use an IV (T…
Problem analyzing outcome by dateHello STATAListers, I'm a new user and having trouble working on a simple dataset with just 5 varia…
Subscribe to:
Post Comments (Atom)
0 Response to Spatial Model (xsmle) "Effects" vs. "Margins" - Why are they different?
Post a Comment