Dear all, My model consists of observations for all four seasons, and each of the seasons is analyzed with respect to five different time slots. All of these observations are referring to the same grid, which means that my neighborhood matrix is repeated twenty times (I took the neighborhood matrix of the original grid and modified the entries in such a way that I have a diagonal matrix where each diagonal block entry consists of the original neighborhood object, with adjusted values). I recently run my model with the errorsarlm() and lagsarlm() functions. It worked fine, and the results coincide well with the linear regression results, which is great. However, there are a few other strange things: In the lag-model, I have only estimates for some variables, but the z-value is indicated as NA (for the error model it works fine). After running the SAR models, R tells me that the two dummy variables for the seasons are aliased. However, in the linear regression I did not have any problems, and the variables are not correlated. I think the second issue might be related to the fact that the number of observations seems to correspond to the number of observations for one season, instead of all four seasons. In other words, it looks like R has realized that the neighborhood repeats itself for each season. Interestingly enough, R has not realized that the pattern repeats itself inside the seasons as well for the time slots. Or is there another possible reason for the fact that the number of observations corresponds to only one fourth of all the observations actually included? Has anybody experienced these issues as well? Any idea on how to resolve these problems would be greatly appreciated. Thank you in advance! Best regards, Raphael Mesaric
Question spdep package - aliased variables and NA z-values
2 messages · Raphael Mesaric, Roger Bivand
On Mon, 27 May 2019, Raphael Mesaric via R-sig-Geo wrote:
Dear all, My model consists of observations for all four seasons, and each of the seasons is analyzed with respect to five different time slots. All of these observations are referring to the same grid, which means that my neighborhood matrix is repeated twenty times (I took the neighborhood matrix of the original grid and modified the entries in such a way that I have a diagonal matrix where each diagonal block entry consists of the original neighborhood object, with adjusted values). I recently run my model with the errorsarlm() and lagsarlm() functions. It worked fine, and the results coincide well with the linear regression results, which is great. However, there are a few other strange things:
These are not really appropriate models, maybe look at the splm package for panel/longitudinal equivalents. To debug the aliasing in the model fitting functions in spatialreg (no longer spdep), please provide a reproducible example, preferably from a built-in data set. Roger
In the lag-model, I have only estimates for some variables, but the z-value is indicated as NA (for the error model it works fine). After running the SAR models, R tells me that the two dummy variables for the seasons are aliased. However, in the linear regression I did not have any problems, and the variables are not correlated. I think the second issue might be related to the fact that the number of observations seems to correspond to the number of observations for one season, instead of all four seasons. In other words, it looks like R has realized that the neighborhood repeats itself for each season. Interestingly enough, R has not realized that the pattern repeats itself inside the seasons as well for the time slots. Or is there another possible reason for the fact that the number of observations corresponds to only one fourth of all the observations actually included? Has anybody experienced these issues as well? Any idea on how to resolve these problems would be greatly appreciated. Thank you in advance! Best regards, Raphael Mesaric [[alternative HTML version deleted]]
_______________________________________________ R-sig-Geo mailing list R-sig-Geo at r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-geo
Roger Bivand Department of Economics, Norwegian School of Economics, Helleveien 30, N-5045 Bergen, Norway. voice: +47 55 95 93 55; e-mail: Roger.Bivand at nhh.no https://orcid.org/0000-0003-2392-6140 https://scholar.google.no/citations?user=AWeghB0AAAAJ&hl=en