On Tue, 26 Mar 2013, SHISHIR MATHUR wrote:
Thanks for the reply Achim. The reason I suspect autocorrelation is
because I think that within the same neighborhood, homes sold a few
months back are likely to impact the price of homes sold subsequently.
This may well be spatial (auto)correlation rather than temporal
autocorrelation.
In fact the DW test and Breusch-Pagan test come out to be significant.
So even though the data is not time series (that is, I do not have
repeated observations for the same house), however, the houses sold
close in time to each other are in the data set.
If there is a unique ordering of all observations by time, then you could
in principle apply an autocorrelation correction for the data, e.g., via
Newey-West.
But from what you describe above, it seems to be more important to capture
spatial effects in the data, e.g., by using a spatial lag model (see
lagsarlm in "spdep") or by using an additive spatial effect (see e.g. gam
in "mgcv").
Thanks,
Shish
On Tue, Mar 26, 2013 at 3:51 PM, Achim Zeileis <achim.zeil...@uibk.ac.at>
wrote:
On Tue, 26 Mar 2013, SHISHIR MATHUR wrote:
Hello:
My dataset set contains several thousand rows of
data, with each row
containing information for a house. The variables
include the sale price of
the house, the quarter and year of sale, the
attributes of the house, and
the attributes of the neighborhood and the city in
which the house is
located. The data is for a 10-year period. No house
is repeated in the
dataset. In summary, the dataset can be termed
pooled cross-section data.
My question: Can I estimate Newey-West HAC standard
errors for a model that
estimates the effect of various independent
variables on the sale price of
the house? My understanding is that Newey-West can
be used for time series
and panel data. However, I am not sure whether it
can be used for pooled
cross-section data. If yes, can you refer me to a
specific source, such as
a paper or a book?
The result of your aggregation is a cross-section data set.
Thus, there should be no correlation between the different
observations - or in other terms, the ordering of your
observations is completely arbitrary.
Consequently, there may be heteroskedasticity but not
autocorrelation. So you may use HC standard errors but HAC
should not be necessary. (Using HAC standard errors will still
be consistent but less efficient.)
--
Best,
Shish
[[alternative HTML version deleted]]
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained,
reproducible code.
--
Best,
Shishir
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
