Skip to content

Problem with moran.test function

2 messages · Michael Haenlein, Roger Bivand

Original
Michael Haenlein
# 
Dear all,

I'm trying to estimate Moran's I using the moran.test function Below
please find the summary statistics for my network and key variable. In
doing so, I'm facing the issue that my estimation results in the
following error message:

Warning message: In moran.test(x, Network_): Out-of-range p-value:
reconsider test arguments

The results are as follows:

        Moran's I test under randomisation

data:  x
weights: Network

Moran I statistic standard deviate = NA, p-value = NA
alternative hypothesis: greater
sample estimates:
Moran I statistic       Expectation          Variance
             -Inf     -0.0002926544                NA

Could anyone please tell me what this error message means?

Is there a problem with my network and/ or with my key variable?

Thanks very much for letting me know,

Michael






Characteristics of weights list object:
Neighbour list object:
Number of regions: 3418
Number of nonzero links: 6442
Percentage nonzero weights: 0.05514125
Average number of links: 1.884728
Non-symmetric neighbours list
Weights style: NA
Weights constants summary:
      n       nn S0 S1 S2
NA 3418 11682724  0 NA NA

summary(x)
Min. 1st Qu.  Median    Mean 3rd Qu.    Max.
1082    1543    2119    2250    2886    3965
Roger Bivand
#
On Wed, 16 Jun 2010, Michael Haenlein wrote:

            

      
      
      
    

        
Well, Moran's I is -Inf, and the analytical variance is NA, so something 
is not right. The problem could lie in x, Network, or the lag of x (when 
x and Network are OK but their combination is unhappy). Can you run 
moran.test() under debug() and check which values lead the value of I to 
go to -Inf? Is this what is going on:

data(columbus)
set.seed(1)
x <- log(rpois(n=49, 2))
x
moran.test(x, nb2listw(col.gal.nb))

where the current spdep release fails reporting:

Error in lag.listw(listw, z, zero.policy = zero.policy, NAOK = NAOK) :
   Variable contains non-finite values

which was a fix introduced four weeks ago, changed to a test on |Inf| from 
a test on NA:

https://r-forge.r-project.org/scm/viewvc.php/pkg/src/lagw.c?root=spdep&r1=244&r2=282

If you update spdep, you'll pick up the improvement (made thanks to a bug 
report by Matias Mayor Fernandez), and if this is the case, the problem is 
in the x.

Hope this helps,

Roger