As shown in the paper, we propose a simulated general method of
moments (SGMM) for the SAR model (see function smmSAR and Section 2 of
our vignette).
We can now estimate the maximal bias of the instrumental variable
estimator (see Section 1.1 and 1.2 of our vignette).
Changes in Version 1.0.2
We provide a smoother simulator of adjacency matrices in the SGMM
approach.
We add weights to the probit/logit network formation model.
We allows the use of an initial probit/logit estimate of \(\rho\), where the observed part of the
network is assumed non-stochastic in the MCMC. This is a quite different
from using an initial probit/logit estimate as prior distribution of
\(\rho\). In this latter case, \(\rho\) is updated using, among others,
information from the observed part of the network. In the first case,
\(\rho\) and the unobserved part of the
network are updated using information in \(y\), where the initial estimate acts as
prior distribution of \(\rho\).
Information from the observed part of the network is not used to update
\(\rho\). This information is included
in the initial estimate.
Changes in Version 1.0.3
Adjustments with Eigen 3.4
Changes in Version 1.0.4
Adjustments with CDatanet 2.2.0
Changes in Version 1.1.0
remove.ids function is added.
The moment function in smmSAR when GX is
observed is corrected following the revision of the paper in March
2025.
Changes in Version 1.1.1
Nothing. We accommodate changes in Armadillo 13.
Changes in Version 1.1.2
We added a new function simSAR, which simulates the
linear peer effect model. In the previous versions, those simulations
were performed using the CDatanet package. This package is no longer
required from this version.