I have read the article by Pisati but no additional help sine I already
have been his modules. I still have same questions in my mind. That
article only explains the mechanics of the commands I am using. I still
need help for questions 1, 3 and 4 as Nick has already answered number
2.
Thanks again.
-----Original Message-----
From: [email protected]
[mailto:[email protected]] On Behalf Of Nick Cox
Sent: Wednesday, February 13, 2008 11:57 AM
To: [email protected]
Subject: st: RE: R: Moran I and spatial correleation help needed
Let me join these two postings:
1. This is precisely the set of programs that Recai Aydin is asking
about. (Recai should have indicated where the programs come from.)
2. Carlo's helpful reference points up the general fact, just made
public, that old issues of the STB are now accessible to all with
internet access.
3. Thus Recai should read Maurizio's article within STB-61 if (s)he has
not already done so.
Just to pick up one of Recai's several questions: You are at perfect
liberty to calculate Moran's I for some transformation of one of your
variables, and for any nonlinear transformation the results will differ.
The issues arising seem exactly like those with any transformed scale or
link function: what is a suitable scale for analysis will depend on your
problem, your data, and your intended analysis.
Nick
[email protected]
Carlo Lazzaro
Maurizio Pisati has suite of commands for spatial data analysis that may
be
useful.
See http://www.stata-press.com/journals/stbcontents/stb60.pdf
RECAI AYDIN
I am new in "spatial correlation" and need your help. I appreciate if
you
can help me. here are the commands I run and I have no problem getting
the
output. However, I cannot say it for the output:)
spatcorr value, bands(0 1.5 3 4.5) xcoord(x) ycoord(y)cumulative
spatwmat, name(W) standardize xcoord(x) ycoord(y) band(0 1.5)eigenval(E)
spatreg value sf sf2 age lot br fb pool d1 d2 d3 d4 d5, weights(W)
eigenval(E) model(lag)
spatreg lnvalue sf sf2 age lot br fb pool d1 d2 d3 d4 d5, weights(W)
eigenval(E) model(error)
My questions are:
1) I noticed that I can use only one variable after "spatcorr" and I
assume
it must be the dependent variable but that means the program is only
checking for the spatial correlation in that variable alone without
knowing
what my full model is. That does not make sense to me. Then how does it
determine if the spatial correlation is "lag" type or "error" type
unless it
defines the error as the deviation in Y?
2) I tried to run "spatcorr" with "lnvalue" as I normally use semi-log
form
and I got different Moran I. Can I use this command with transformed
variable. How does that affect the interpretation of the I?
3) Is there any easy way of understanding if the spatial autocorrelation
is
due to "lag" or "error" from the output?
4) When I have the output for the last two lines (spatreg) and compare
them
with OLS, I do not see much difference even though Moran I, rho and
lambda
all are significant. How can I knoe that actually one is superior to
the
others. This last one is especially important for me. Because I really
doubt
about the importance of spatial autocorrelation and I need to be
convinced
by one of the "believers":).
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
