I think your study is genuinely a spatial issue. So I think spatial econometric can be a suited tool. But first, you have to find out the underlying model behind your analysis. A good literature on spatial econometric would be the book of Sir Elhorst (Spatial Econometrics from Cross-Section Data to Spatial Panels). Concerning software, I think Matlab would be a better choice. Many codes are available on internet (Matlab toolbox by Lesage or that of Elhorst available on their homepage).

The books of Luc Anseling and James LeSage could be interesting for you.

http://geodacenter.asu.edu/

http://scholar.google.com/citations?user=WDHfG4gAAAAJ&hl=es

I would suggest you use a so called spatial lag model. If the spatially lagged dependent variable (weighted neighbour matrix, either by distance or other specification), is significant, there is spatial correlation. As you have panel data, I would suggest the following model:

http://facweb.knowlton.ohio-state.edu/pviton/courses/crp87105/millo-piras.pdf

It is implemented in R which is a great program if you like spatial analysis.

If I was you, I would actually use R to start a simple spatial correlation analysis using local Moran's I. Have a look at this:

http://www.bias-project.org.uk/ASDARcourse/unit6_slides.pdf

Presence of spatial correlation however does not tell you anything about the direction of diffusion of cultural values. You could test your assumptions by building in temporal lags into spatial lags. Cultural values in the origin region in t-1 should be correlated with cultural values in the neighbouring regions in year t.

Good luck

Since you mentioned that you wanted to measure attraction and/or repulsion of cultural values between countries, you might consider using a gravity model in your analysis. You can find a number of examples of using spatial interaction, gravity-based models in the journal Geographical Analysis, among others. It has been used in consumer behavior analysis to estimate the probability of customers from a given trade area patronizing different retail establishments (Huff, D.L. 1964. Defining and estimating a trade area. Journal of Marketing, 28:34-38). I am not aware of using a gravity model in estimating culturally-based attraction but it might be a good idea to try it, given that you can quantify cultural characteristic.

First of all, "spatial analysis" is a rather broad term and means different things to different people but in fact I don't think that the question you posed is really the one you want to answer since you already have specific data you want to analyze.

Other responders have given you good suggestions and the referral to R is especially good, there are many packages in R for various forms of spatial analysis. You mention STATA which is a commercial software, do you have access to any other commercial software such as ARCGIS (there are open source GIS available as well) GEODA is a free package (from Luc Anselin and there are tutorial materials available).

Do a search on Google for "Spatial Analysis" and you will find a very large number of links (although some may not be relevant for you).

The real question is what questions do you really want to answer? Note that this might lead you to ask whether you have the right data set

You mentioned that you observed spatial correlation meaning there is a gravitational effect of the cultural value. If you already have the value of Moran I, an index of global clustering, then you have the magnitude of influence. If you want to do more than just the univariate analysis, then you need to define what will be your outcome and what will be the explanatory variables. I do not think STATA can handle spatially related data, but you can use Geoda (an open source software) for analyzing spatially related data.

Of course your research can be enriched with the tools that provides spatial statistics. First, it is possible to calculate measures of spatial association to see if nearby locations are similar with respect to certain characteristics. You can then propose a model to explain this possible relationship. Some computer packages for this type of analysis are R and GeoDa (both are free).

GeoDa can be downloaded at GeoDa Center for Geospatial Analisys and Computation (http://geodacenter.asu.edu/)

Applied Spatial Data Analysis with R is an accessible text that demonstrates and explains the handling of spatial data using the R Software Platform

My expertise is more in the realm of ecology and evolution (than econometrics), but your question seems analogous (in construct) to the sort of problems we work on in landscape genetics.

Since the hypothesis here is partly formulated based on distance between countries, another interesting option (beyond gravity models and spatial regression) would be a multiple regression on distance matrices.

http://people.biology.ufl.edu/jlichstein/downloads/publications/Lichstein_2007_Plant_Ecol.pdf

I think a graph theory framework would also be appropriate here:

http://onlinelibrary.wiley.com/doi/10.1111/j.1752-4571.2008.00047.x/full.

Spatial analysts of the world....unite.

Thank you all for your comments and suggestions!

Anatole Goundan, what do you mean by “underlying model behind your analysis”? Do you mean to hypothesize what would be the origin country and the direction of influence?

Pawel Paluchowski, spatial lag model with distance matrix can be implemented in STATA and this is the only thing I know about spatial analysis so far. But as you said, it is quite limited in what answer it can give me. I am not sure if I got your suggestion with the temporal lags very well. What specifications do I need in my model? How can the values in (only one?) origin region in t-1 be correlated with values in neibouring regions in t0?

Peter Lawrence, I tried already different types of cluster analysis but it seems like there is no good cluster solution for my data. I also did Multidimensional Scaling analysis which provided me with some insides of the relationship between countries. It appears that there are 2 big and nearly equal in size clusters (though not very well defined) – Western European with center Germany and Eastern European with center Russia. Because these are the biggest countries in the clusters I assume that they have some influence on the rest of the countries which so to say “gravitate” around them. Also, your assumption is correct that I am interested in temporal evolution - my ultimate research question is to conclude whether there will be convergence or divergence of cultural values in the future. I honestly have no idea how to design the metric in order to incorporate my idea of “influences". I have no access to Matlab, unfortunately. Ideally, I would like to use STATA as I am already familiar with it.

Piotr Jankowski, thanks for your suggestion. Gravity models sounds like something relevant for my analysis and I will search more about it.

Donald Myers, it is true that spatial analysis is too broad term. I have only heard about spatial error and spatial lag models but I was sure that there is a lot more that I am missing and this is why my question here was quite general too. I didn't understand why you have doubts that I am interested in the question which I asked? My ultimate interest is to find whether European nations diverge or converge culturally over time. I have no access and experience in using other statistical software apart from STATA and SPSS.

Mohammad Ali, Moran I might give me an idea about the magnitude of influence but as far as I know, it doesn't tell me anything more specific, for example what the origin and the direction are.

Alejandro Alegría, could you be more specific about how to propose a model to explain the possible relationship?

Corey Sparks, I will start with analysis on country level even though I think I may have too few observations for a good analysis. I thought about doing the analysis on regional level but many of my variables and the theoretical framework are only on country level. I think multilevel analysis might be too complicated for my current statistical skills as well.

Another thing that worries me is the multicolinearity between all of my independent variables (GDP per capita, religious denomination, language family and experiencing communism) and geographical location as well as the too little observations to test all factors in a multiple regression model. What are your thoughts about this issue?

Here is how my analysis looks like so far.

If you can access to ArcGIS, it has a complete tool for Spatial Statistics, that ranges from univariate (different spatial autocorrelation indexes, i.e. global and Local Moran and Getis indexes), to multivariate techniques (i.e., Cluster analysis and Geographically Weighted Regression). The ArcGIS package provides a good help that well explains the different tool applications and theory. If you are familiar with R you can find a lot of codes for spatial analysis. There is another open source GIS, Quantum GIS, that has a spatial statistics tool, but I do not know if it is comparable to that of ArcGIS.

Plamen: if you want to do spatial analysis, you are going to have to be more flexible about your software platform. Most commercial packages are good for standard parametric and nonparametric stats, but are generally weak when it comes to spatial stats. ArcGIS is good for data storage, mapping, raster analysis, and data interpolation...but does not offer much in terms of pattern analysis (esp. if you want to apply causal modeling to explain your spatial pattern). R is a headache, but it might be your best bet.

I was not aware that you could do GWR in ArcGIS! Cool. Plamen, GWR is basically a local regression method that allows your predictors to vary spatially. This is particularly useful if the process you are studying is nonstationary (i.e., the statistical properties are dependent on spatial location). Most landscapes are nonstationary, especially at the scale you are working at, so this is something to consider.

I am acoording with Ciotoli, Anderson and Siska, The point is that ArcGIS software let you can to analise cultural values in Europe using data from ESS.

I think your problem can be addressed by using the approach of "spatial analysis using distance indices". This is an approach of local clustering for studying spatial patterns. I think the article on "Spatial analysis in ecology by Marie-Jos´ee Fortin, Mark R.T. Dale & Jay ver Hoef, Volume 4, pp 2051–2058 in Encyclopedia of Environmetrics (ISBN 0471 899976) Edited by Abdel H. El-Shaarawi and Walter W. Piegorsch  John Wiley & Sons, Ltd, Chichester, 2002" will help you on statistical approach to be used for your problem. All the best.

I do agree with Anderson's comment. I think MATLAB software will be a better choice than ArcGIS for such statistical analysis.

Alternatively you can opt for R and the multitude of packages it has on spatial modeling and statistics (google e.g. spdep and spgwr packages), but the learning curve may be steep if you are unfamiliar with the software. If you have the time I sincerely recommend the summer course by Spatial Econometrics Advanced Institute (http://www.spatialeconometricsadvancedinstitute.org) arranged annually in Rome.

If you want to know the zone of influence you can use the variogram or spatial correlogram. Values at surrounding locations within this zone are correlated with the value at the specific location. These tools are used to examine patterns of spatial autocorrelation.

Hi Plamen,

What I was saying is that before attempting to estimate a spatial model, you should first identify the best specification for the model that fits your research questions. But this is only possible if you know your subject well and possible relationships between cross sections. Thus, the stylized facts are a good start for your model specification.

Anatole

Other approach is the DEA. It si for measure distance but productivity análisis. http://www.uq.edu.au/economics/cepa/software.php

Before looking at space per se., you could simply add some spatial context variables to your MDS, e.g. a neighbourhood connectivity matrix, or one say on length of common border, topological distance (i.e. n countries in between) etc.

I wasn't clear if you were interested in geographical spatial analysis or spatial techniques to look at the output of your MDS? Geographical distance is a tricky term in this context (to the coutries centroid? nearest border? travel time (in which era?) etc.). Spatial stats on an MDS space would, I suspect, only be valid if the resulting space between the dimensions was a real integral and not toooo none Euclidean.

Dear Plamen,

Have a look at this paper. It talks about spatial and temporal lags. The model that I had in mind was lagged spatial dependence (there could be the case that the more complicated mixed model contemporaneous and lagged dependence is given which I would expect. It is much more difficult to establish). With panel data that could get quite difficult to specify.

The paper talks about space-time correlation so that would be the starting point which would already inform your research question

http://ideas.repec.org/p/wpa/wuwpem/0411005.html

Jim LeSage has a nice website with a ton of spatial econometrics materials, lots of links to other resources and his FREE matlab code for estimating a ton of different spatial econometric models.

http://www.spatial-econometrics.com/

If you look at ArcGIS, Spatial Statistics Tools extensions, You will find a lot of spatial statistical techniques (spatial statistical and autocorrelation indexes and spatial multivariate methods, etc.,) that you can use for a new and detailed spatial description of your data. If you need more and specific information you can have a look to the exhaustive ArcGIS online help, or to contact me, if you want.

In particular, as I have correctly understand your topic, you could use hot-spot analysis (monovariate) and spatial cluster analysis (multivariate) to reply to your question "... to measure whether there is diffusion or repulsion of cultural values from some countries to others?" However, you could also construct a multivariate spatial regression model by using Geographycally Weigthed Regression to undestand which of the variables that you are studying more/less influence your model in the study area.

I am agree with Ciotoli, but I think that you get better results if the panel data is considered.

I second the suggestion of exploring GeoDA. It was developed by Luc Anselin (one of the "forefathers" of spatial analysis in the social sciences) and his research team. I've included a link to a short article which introduces the software and the methods you can employ. GeoDA is great for exploratory analysis (e.g., the Moran's I mentioned above) and then you can move to progressively more complex methods such as spatial autoregressive models. If you do not already have a spatial dataset, it is easy enough to create one with a data table (in Excel or similar format) and a shape file of the countries of interest (which you can get through DIVA-GIS or some other free source). Good luck!

https://geodacenter.asu.edu/pdf/geodaGA.pdf

Well, this is a very general question and could have many different answers. Here are some of science questions I would propose:

1) What percentage of wildland have been converted to urban and croplands in a region during certain period? How strong is the ecological footprint?

2) How an alpine glacier was melted or accumulated during a period? How about the mass balance?

3) How many acres of woodland was damaged by the most recent hurricane? How severe?

...

I am agree with Jia. so in spatial analysis is important consider other variable that have a significan with the life resource as use land and wáter, fertilizer. The agroecology approach is my suggest in this point. The FAO have the smarth agricultural software to measure it.

This is the link where you can to check my observation http://www.fao.org/climatechange/climatesmart/en/

You've really got to think about the spatial relationships a bit more, are there structural relationships that must be captured (like distance to an amenity in hedonic modeling) or is there spatial dependency that is more of a nuisance that needs to be corrected for? In most regional economic applications there is a structural spatial relationship and that takes you one way in the empirical modeling. If its a "nuisance" (like the unit of observation does not adequately reflect the relevant region ) then you go another way with the empirical modeling.

Given that you have repeated cross-sectional data from the ESS , that there is a multilevel-structure (individuals in regions in waves in countries)and that many (all?) of you response variables are discrete you may want to consider random effects space-time discrete outcome modelling that can handle all these characteristics simultaneously.

I have uploaded this book which considers these elements in Chapters 15 and 16 to Research gate

Developing multilevel models for analysing contextuality, heterogeneity and change using MLwiN, Volume 2

Kelvyn Jones, VS Subramanian

https://www.researchgate.net/publication/260772180_Developing_multilevel_models_for_analysing_contextuality_heterogeneity_and_change_using_MLwiN_Volume_2?ev=prf_pub

If you are a Stata user you can run MLwin via runmlwin

http://www.bristol.ac.uk/cmm/software/runmlwin/

In my opinion , what you must not do is to aggregate to just countries and then use standard spatial models to look at the patterns - you will be missing a lot of what is going on by simply aggregating.

In my practice I model the implicit spatial dependency by the standard hierarchical model and include types of country ( eg based on welfare regimes such as Scandinavian or former Eastern bloc) in the fixed part of the model to help try to explain that dependence. That is I do not usually perform explicit spatial modelling as set out in Chapter 16.

This issimpler to do, and you are then trying to account for the dependency by specific theoretically-informed variables instead of rather vague forms of adjacency amd spillover effects at the country level. But of course you could do both at the same time with these random effects models.

CHapter 16 also shows how to implement random effects version of Geographically Weighted Regression

Book Developing multilevel models for analysing contextuality, he...

The questions must be including a spatial dimension as well as geographical measurement. For example, what are the contributing factors that affect migration distribution in urban areas? In this case, the 'term' distribution inherently has a spatial dmension so that it can be analysed using GIS techniques or others spatial analysis models. By using GIS, the factors/ criterion map can be the areas with acceptable cost of living, nearness to place of work, affordable housing areas, etc.

The multilevel approach that I was suggesting for the cross-wave ESS data is shown in action here

Chris Deeming and Kelvyn Jones (2013) Social Policy and Economic Development in Europe: Investigating the Macro Determinants of Individual Health and Well-being in a Multilevel Multivariate Analysis of Thirty European Nations

a preliminary version is avaialable from here

www.esrc.ac.uk/my-esrc/.../6cff352a-b28b-4234-aba1-ed4834560946‎

it is currently under review; so any comments would be particularly welcome.

Kelvyn

I just browsed your interesting paper. I am trying to investigate the macro and micro determinants of depression using the ESS. Individuals are nested across countries and a multilevel approach seems to be approprıate.

Antonio

Antonio

the reason I suggested multilevel modelling is that this approach fits the repeated structure of the ESS like a hand in a glove.

I have also put up two volumes on Research Gate to facilitate such an analysis

Book: Developing multilevel models for analysing contextuality, heterogeneity and change using MLwiN, Volume 1

Book: Developing multilevel models for analysing contextuality, heterogeneity and change using MLwiN, Volume 2

Let us know how you get on.

Kelvyn

thank you for the detailed response. The idea is to use several waves of the ESS to come up with a measure of depression. In my previous paper, I only employed the wave number 3.