I face the challenge of estimating a 2SLS (IV) model with a fractional (ranging between 0 and 1) outcome variable and spatially autocorrelated error terms (the data is spatially explicit, i.e. for each observation I have coordinates).
So far I achieved doing so but without accounting for the residual spatial autocorrelation by a first stage OLS regression, the fitted values of which are then regressed against the y in a fractional, probit-linked regression. I also bootstrapped the coefficients to correct for the bias stemming from the splitting of the to stages of the 2-stage-least-squares regression, is this right?
Literature-wise I found Lee (2003) paper titled "Best Spatial Two‐Stage Least Squares Estimators for a Spatial Autoregressive Model with Autoregressive Disturbances", but this goes into the theory without (at least to my awareness) then producing a statistical software script to replicate the methodology.
So far I achieved doing so but without accounting for the residual spatial autocorrelation by a first stage OLS regression, the fitted values of which are then regressed against the y in a fractional, probit-linked regression. I also bootstrapped the coefficients to correct for the bias stemming from the splitting of the to stages of the 2-stage-least-squares regression, is this right?
Literature-wise I found Lee (2003) paper titled "Best Spatial Two‐Stage Least Squares Estimators for a Spatial Autoregressive Model with Autoregressive Disturbances", but this goes into the theory without (at least to my awareness) then producing a statistical software script to replicate the methodology.
Comment