On Fri, 7 Feb 2003 [EMAIL PROTECTED] wrote: > Dear list members, > > I am a GIS programmer at Hungarian Central Statistical Office, and > trying to make a work about statistical connections between time > required to reach some source (e.g. the capitol, the border crossing > points, or local civil services) and local social parameters (this first > time the governmental tax/person - as an indicator of the income). > > I have computed the required time data via network analysis for each > localities, and have the respond variable. Using R I have made some > linear fittings between time as predictor and the paid tax/person as > respondent, but, I suspect, the strong linear correlation I found is an > outcome from the spatial autocorrelation in the tax data.
This is an interesting analysis, with quite a lot of features. 1) What are the localities? Can their behaviour (as local councils etc.) affect the tax per capita? Or is the tax per capita more a result of the state of the local economy? Why per capita (tax comes from working people, not total population)? My guess would be that local economic conditions are the main "driver". How does tax per capita correlate with firm formation, unemployment? 2) Have you tested the residuals of your linear model for spatial autocorrelation, or just the response variable? 3) How many distance variables are you using to measure isolation - the most isolated being a long way from a) the capital, b) a border crossing, and c) local services? 4) How many localities are you examining? How are you constructing the spatial weights matrix for calculating spatial autocorrelation? > > I have mapped the local spatial autocorrelation for these data, and > found that it shows positive, negative and insignificant spatial > autocorrelations between the neighbours in large, well separated > continuous areas. The same areal distribution is typical for the > residuals from the linear correlation. > > My question is: should I use geostatistical methods based on variogram? > The argument to support this method is: My predictor is a distance-like > value - in fact the time which is a function of the available speed on > the road segments and the distance between localities. > The argument against this method: My data are not from spatially > continual variable(s), because there is not living people between > settlements. I think that using geostatistical methods would be premature while quite a lot can still be done treating the data as spatial lattice data. I would worry about the different effects of eastern and western borders, and population density, across the country. > > Or should I include a "spatial lag" - the local average of the data > weighted by the inverse distance (time) - into the regression? > I strongly suspect, that this later method is the better solution, but > can someone direct me to a publication about similar work? > > Please, excuse me for this longish "question". > Very interesting - please contact me off the list if you prefer. Roger Bivand > Thank you in advance > Jozsef Fabian > GIS programmes > HCSO Hungary > -- Roger Bivand Economic Geography Section, Department of Economics, Norwegian School of Economics and Business Administration, Breiviksveien 40, N-5045 Bergen, Norway. voice: +47 55 95 93 55; fax +47 55 95 93 93 e-mail: [EMAIL PROTECTED] -- * To post a message to the list, send it to [EMAIL PROTECTED] * As a general service to the users, please remember to post a summary of any useful responses to your questions. * To unsubscribe, send an email to [EMAIL PROTECTED] with no subject and "unsubscribe ai-geostats" followed by "end" on the next line in the message body. DO NOT SEND Subscribe/Unsubscribe requests to the list * Support to the list is provided at http://www.ai-geostats.org