correlogram for missing values
On Thu, 26 May 2011, Advait Godbole wrote:
Dear List, I am performing a spatial analysis for gridded data with ~8000 cells. Each grid cell is 10 x 10 km and has one dependent and 4 independent variables. The dependent variable is log-transformed and has some NAs. I generated a first order queen neighbour object:
The missing values are almost certainly the source of your problem. Subset both your data and your neighbour object to retain only complete cases. You did not state your command for Moran's I, but I guess that you set na.action=na.omit - you do not seem to have passed this to correlog, which gives the message you see irrespective of the cause of the error. The problem will be resolved by removing incomplete cases. Roger
vulcgrid.queen.nb
Neighbour list object: Number of regions: 8281 Number of nonzero links: 64944 Percentage nonzero weights: 0.09470512 Average number of links: 7.842531 and then ran a Moran's I (after accounting for the missing values using na.omit):
lnemitpcap.morans.queen.1st
Moran's I test under randomisation
data: LNEMITPCAP
weights: vulcgrid.queen.nb.w
omitted: 7, 176, 259, 1304, 1424, 1473, 1532, 2101, 2273, 2436, 2451, 2884,
2921, 4260, 4271, 4454, 4577, 4957, 5111, 5126, 5589, 5837, 5871, 5994,
6084, 6315, 6527, 6528, 6696, 6697, 6865, 6946, 7022, 7034, 7115, 7558,
7638, 7658, 7780, 7898
Moran I statistic standard deviate = 58.0851, p-value < 2.2e-16
alternative hypothesis: greater
sample estimates:
Moran I statistic Expectation Variance
3.249438e-01 -1.213592e-04 3.131927e-05
I wanted to test for the suitability of higher order neighbours and hence
tried sp.correlogram which does not work with NAs. Is there a workaround?
Also, on trying correlog() following suggestions in the ASDAR book, gives
the following output:
vulcgrid.correlog <- correlog(coords, LNEMITPCAP, method = "Moran",
nbclass=NULL) Error in correlog(coords, LNEMITPCAP, method = "Moran", nbclass = NULL) : Bad selection of class breaks, try another one... How may I go about this? Do I need to manually specify distance bands? the data are not on a regular grid. The GIS file is an extract from a larger dataset for the conterminous US in North American Lambert Conformal Conic projection with NAD 1983 datum. Suggestions are greatly appreciated. Regards,
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