1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
|
\name{stsls}
\alias{stsls}
\alias{print.Stsls}
\alias{print.summary.Stsls}
\alias{summary.Stsls}
\alias{residuals.Stsls}
\alias{coef.Stsls}
\alias{deviance.Stsls}
\alias{impacts.Stsls}
%- Also NEED an '\alias' for EACH other topic documented here.
\title{Generalized spatial two stage least squares}
\description{
The function fits a spatial lag model by two stage least squares, with the option of adjusting the results for heteroskedasticity.
}
\usage{
stsls(formula, data = list(), listw, zero.policy = NULL,
na.action = na.fail, robust = FALSE, HC=NULL, legacy=FALSE, W2X = TRUE,
sig2n_k=TRUE, adjust.n=FALSE)
\method{impacts}{Stsls}(obj, \dots, tr, R = NULL, listw = NULL, evalues=NULL,
tol = 1e-06, empirical = FALSE, Q=NULL)
}
%- maybe also 'usage' for other objects documented here.
\arguments{
\item{formula}{a symbolic description of the model to be fit. The details
of model specification are given for \code{lm()}}
\item{data}{an optional data frame containing the variables in the model.
By default the variables are taken from the environment which the function
is called.}
\item{listw}{a \code{listw} object created for example by \code{nb2listw}}
\item{zero.policy}{default NULL, use global option value; if TRUE assign zero to the lagged value of zones without
neighbours, if FALSE (default) assign NA - causing \code{lagsarlm()} to terminate with an error}
\item{na.action}{a function (default \code{na.fail}), can also be \code{na.omit} or \code{na.exclude} with consequences for residuals and fitted values - in these cases the weights list will be subsetted to remove NAs in the data. It may be necessary to set zero.policy to TRUE because this subsetting may create no-neighbour observations. Note that only weights lists created without using the glist argument to \code{nb2listw} may be subsetted.}
\item{robust}{default FALSE, if TRUE, apply a heteroskedasticity correction to the coefficients covariances}
\item{HC}{default NULL, if \code{robust} is TRUE, assigned \dQuote{HC0}, may take values \dQuote{HC0} or \dQuote{HC1} for White estimates or MacKinnon-White estimates respectively}
\item{legacy}{the argument chooses between two implementations of the robustness correction: default FALSE - use the estimate of Omega only in the White consistent estimator of the variance-covariance matrix, if TRUE, use the original implementation which runs a GLS using the estimate of Omega, overrides sig2n_k, and yields different coefficient estimates as well - see example below}
\item{W2X}{default TRUE, if FALSE only WX are used as instruments in the spatial two stage least squares; until release 0.4-60, only WX were used - see example below; Python \code{spreg::GM_Lag} is default FALSE}
\item{sig2n_k}{default TRUE - use n-k to calculate sigma^2, if FALSE use n; Python \code{spreg::GM_Lag} is default FALSE}
\item{adjust.n}{default FALSE, used in creating spatial weights constants for the Anselin-Kelejian (1997) test}
\item{obj}{A spatial regression object created by \code{lagsarlm}, \code{lagmess} or by \code{lmSLX}; in \code{HPDinterval.LagImpact}, a LagImpact object}
\item{\dots}{Arguments passed through to methods in the \pkg{coda} package}
\item{tr}{A vector of traces of powers of the spatial weights matrix created using \code{trW}, for approximate impact measures; if not given, \code{listw} must be given for exact measures (for small to moderate spatial weights matrices); the traces must be for the same spatial weights as were used in fitting the spatial regression, and must be row-standardised}
\item{evalues}{vector of eigenvalues of spatial weights matrix for impacts calculations}
\item{R}{If given, simulations are used to compute distributions for the impact measures, returned as \code{mcmc} objects; the objects are used for convenience but are not output by an MCMC process}
\item{tol}{Argument passed to \code{mvrnorm}: tolerance (relative to largest variance) for numerical lack of positive-definiteness in the coefficient covariance matrix}
\item{empirical}{Argument passed to \code{mvrnorm} (default FALSE): if true, the coefficients and their covariance matrix specify the empirical not population mean and covariance matrix}
\item{Q}{default NULL, else an integer number of cumulative power series impacts to calculate if \code{tr} is given}
}
\details{
The fitting implementation fits a spatial lag model:
\deqn{y = \rho W y + X \beta + \varepsilon}{y = rho W y + X beta + e}
by using spatially lagged X variables as instruments for the spatially lagged dependent variable.
From version 1.3-6, the general Anselin-Kelejian (1997) test for residual spatial autocorrelation is added.
}
\value{
an object of class "Stsls" containing:
\item{coefficients}{coefficient estimates}
\item{var}{coefficient covariance matrix}
\item{sse}{sum of squared errors}
\item{residuals}{model residuals}
\item{df}{degrees of freedom}
}
\references{Kelejian, H.H. and I.R. Prucha (1998). A generalized spatial two
stage least squares procedure for estimating a spatial autoregressive
model with autoregressive disturbances. \emph{Journal of Real Estate
Finance and Economics} 17, 99-121. \doi{10.1023/A:1007707430416}.
Anselin, L., & Kelejian, H. H. (1997). Testing for Spatial Error Autocorrelation in the Presence of Endogenous Regressors. International Regional Science Review, 20(1-2), 153-182. \doi{10.1177/016001769702000109}.
Roger Bivand, Gianfranco Piras (2015). Comparing Implementations of Estimation Methods for Spatial Econometrics. \emph{Journal of Statistical Software}, 63(18), 1-36. \doi{10.18637/jss.v063.i18}.
}
\author{Luc Anselin, Gianfranco Piras and Roger Bivand}
\seealso{\code{\link{lagsarlm}}}
\examples{
data(oldcol, package="spdep")
#require(spdep, quietly=TRUE)
lw <- spdep::nb2listw(COL.nb)
COL.lag.eig <- lagsarlm(CRIME ~ INC + HOVAL, data=COL.OLD, lw)
summary(COL.lag.eig, correlation=TRUE)
COL.lag.stsls <- stsls(CRIME ~ INC + HOVAL, data=COL.OLD, lw)
(x <- summary(COL.lag.stsls, correlation=TRUE))
coef(x)
W <- as(lw, "CsparseMatrix")
trMatc <- trW(W, type="mult")
loobj1 <- impacts(COL.lag.stsls, R=200, tr=trMatc)
summary(loobj1, zstats=TRUE, short=TRUE)
ev <- eigenw(lw)
loobj2 <- impacts(COL.lag.stsls, R=200, evalues=ev)
summary(loobj2, zstats=TRUE, short=TRUE)
require(coda)
HPDinterval(loobj1)
COL.lag.stslsW <- stsls(CRIME ~ INC + HOVAL, data=COL.OLD, lw, W2X=FALSE)
summary(COL.lag.stslsW, correlation=TRUE)
COL.lag.stslsWn <- stsls(CRIME ~ INC + HOVAL, data=COL.OLD, lw, W2X=FALSE, sig2n_k=FALSE)
summary(COL.lag.stslsWn, correlation=TRUE)
COL.lag.stslsR <- stsls(CRIME ~ INC + HOVAL, data=COL.OLD, lw,
robust=TRUE, W2X=FALSE)
summary(COL.lag.stslsR, correlation=TRUE)
COL.lag.stslsRl <- stsls(CRIME ~ INC + HOVAL, data=COL.OLD, lw,
robust=TRUE, legacy=TRUE, W2X=FALSE)
summary(COL.lag.stslsRl, correlation=TRUE)
data(boston, package="spData")
gp2a <- stsls(log(CMEDV) ~ CRIM + ZN + INDUS + CHAS + I(NOX^2) + I(RM^2) +
AGE + log(DIS) + log(RAD) + TAX + PTRATIO + B + log(LSTAT),
data=boston.c, spdep::nb2listw(boston.soi))
summary(gp2a)
}
\keyword{spatial}
|
