[Openspace] Strange points on Moran scatterplot
Roger.Bivand at nhh.no
Thu Apr 21 09:41:37 CDT 2005
On Thu, 21 Apr 2005, Breego Lery wrote:
> OK, Moran's I with EB rate does improve the plots for my rate DV that has
> a lot of zeros.
> When I save this adjusted rate and use it in regressions, the models do
> not change substantively (but I do get stronger spatial autocorrelation,
> better fit statistics and coefficients that are not directly
> interpretable.) However, I have persisent non-normality and
> heteroskedasticity. Is it appropriate to run transformations on a model
> that already has this transformed EB adjusted rate?
Well, this is a classic case of the initial question assuming that the
respondents can see over the questioner's shoulder, and see the broader
context of the research situation!
Modelling is, I believe, different, and your goal is to model the cases -
what is causing the cases - given the population at risk, is that right?
Revisiting the original papers by Noel Cressie on the North Carolina SIDS
data is worthwhile, it shows how much care may be needed to handle the
cases where their means and variances can drift or trend in relation also
to explanatory variable(s). Alternatively look at Poisson or Negative
Binomial regression (possibly zero-inflated), but then we don't have
proper ways of checking the residuals for possible autocorrelation (again,
I think). I think that your cases will be over-dispersed, this is quite
common. What do the rates represent? Your email and signature don't give
much of a handle (rates of cylce punctures??).
> c/ /'_
> (*) (*)
> --- Roger Bivand <Roger.Bivand at nhh.no> wrote:
> From: Roger Bivand <Roger.Bivand at nhh.no>
> Date: Wed, 20 Apr 2005 10:09:47 +0200 (CEST)
> To: Breego Lery <breego at muchomail.com>
> Cc: Luc Anselin <anselin at mac.com>,
> <openspace at agec144.agecon.uiuc.edu>
> Subject: Re: [Openspace] Strange points on Moran scatterplot
> On Tue, 19 Apr 2005, Breego Lery wrote:
> > Well, the mean of x can't be 0 because it is a rate. The mean is 18,
> > although 8 percent of the x's are 0's.
> Yes, that's what Luc is saying (the plot centres the data on 0 by
> subtracting the mean). If the x values are min(x) - here 0, they'll stack
> up on the left, towards the y-axis, with different y values because the
> average values of their neighbours differ. If you create a raw rate with
> the NC SIDS sample data, and do a regular Moran scatterplot, you see the
> same effect (at about x = -1.5); using the better suited Moran's I with
> rate standardisation, it goes way.
> > O__
> > c/ /'_
> > (*) (*)
> > ~~~~~~~~~~
> > --- Luc Anselin <anselin at mac.com> wrote:
> > From: Luc Anselin <anselin at mac.com>
> > Date: Tue, 19 Apr 2005 16:53:32 -0500
> > To: breego at muchomail.com
> > Cc: openspace at agec144.agecon.uiuc.edu
> > Subject: Re: [Openspace] Strange points on Moran scatterplot
> > only thing I can think of right away is that you have a lot of zero
> > values
> > such that the mean of x is zero. the lining up means that the x value
> > constant (but the spatial lag may not be). the points on the x-axis of
> > the
> > moran scatter plot are rescaled with 0 as the mean, so they should
> > not all be on one side of the y-axis, unless 0 is the mean originally.
> > does this make sense.
> > L.
> > On Apr 19, 2005, at 4:47 PM, Breego Lery wrote:
> > > Ever seen a situation where the points on a univariate Moran
> > > scatterplot are all in the upper right quadrant and a bunch of points
> > > are lined up along the Y axis? When I say lined up I mean really
> > > up - exactly on the Y and not at all over into the top left quadrant.
> > > Any idea how to interpret this or does it indicate an error in
> my .gal
> > > file?
> > >
> > >
> > >
> > > Bridgette
> > >
> > >
> > >
> > >
> > >
> > > O__
> > > c/ /'_
> > > (*) (*)
> > > ~~~~~~~~~~
> > >
> > >
> > > The Free Email with so much more!
> > > =====> http://www.MuchoMail.com <=====
> > > _______________________________________________
> > > Openspace mailing list
> > > Openspace at sal.agecon.uiuc.edu
> > > http://sal.agecon.uiuc.edu/mailman/listinfo/openspace
> > The Free Email with so much more!
> > =====> http://www.MuchoMail.com <=====
> Roger Bivand
> Economic Geography Section, Department of Economics, Norwegian School of
> Economics and Business Administration, Helleveien 30, N-5045 Bergen,
> Norway. voice: +47 55 95 93 55; fax +47 55 95 95 43
> e-mail: Roger.Bivand at nhh.no
> The Free Email with so much more!
> =====> http://www.MuchoMail.com <=====
Economic Geography Section, Department of Economics, Norwegian School of
Economics and Business Administration, Helleveien 30, N-5045 Bergen,
Norway. voice: +47 55 95 93 55; fax +47 55 95 95 43
e-mail: Roger.Bivand at nhh.no
More information about the Openspace