NEWS

spldv 0.2.2

Fixed a bug in the computation of the spatial multiplier matrix used by marginal effects.
For continuous variables, impacts(..., result = "from.region", dydx = "exact") and impacts(..., result = "cumulative", dydx = "exact") now use analytical derivatives rather than finite differences. This improves numerical stability and makes the "exact" option consistent across average, from-region, and cumulative impacts.
impacts() now stops with an informative error when the fitted formula contains transformations that are not currently supported for spatial marginal effects, including poly(), scale(), bs(), and ns(). For quadratic effects, users should write explicit terms such as x + I(x^2). For scaled variables, users should create the transformed variable in the data before fitting the model.
Discrete variables, including factors, ordered factors, and logical variables, continue to be evaluated as predictive contrasts rather than continuous derivatives. This includes formulas that use inline factors such as factor(x) and interactions involving such factors.

Expanded the help page for impacts() and its summary() method. The documentation now explains average direct, indirect, and total impacts; from-region impacts; cumulative impacts; the role of from.unit; the interpretation of Q; the difference between dydx = "exact" and dydx = "numeric"; and the distinction between continuous derivatives and discrete contrasts.
Added examples illustrating average impacts, from-region impacts, cumulative impacts, Monte Carlo and Delta-method standard errors, finite changes through change = c(lower, upper), and impacts for models containing inline factor terms.

Added regression tests for inline factor terms in average, from-region, and cumulative impacts.
Ignored local diagnostic and check artifacts so they are not included in package builds.

Includes prediction function for GMM, LGMM, and RIS estimators.
Includes the argument fastmom argument in sbinaryGMM function which speed up the computation time.
New functionalities for computing marginal effects.
Fixed a bug in the gradient of GMM models.
atmeans argument for marginal effects is deprecated.
Fixed overflow in the S matrix.
Improved documentation.