r-project package has got some handling for mapped data, as in the list https://cran.r-project.org/web/views/Spatial.html and withn it https://cran.r-project.org/web/packages/gstat/gstat.pdf For using R any stats dept can support an easy start

If you want to compare means between groups, I think the most suitable test is student-t test to know if differences between measurements are statistically significant.

But I agree with Alessandro. A centroid is a weak measurement and you would be supporting all results in a weak departing point.

I certainly agree with Alessandro's approach, although, wouldn't it be better to use populated places within counties or municipalities?

I mean, instead if using 1000 points inside each area, use every populated place within that same area (of course, I'm asuming here that you want a measure of water accesibility for human consumption).

A second point. If you use the same amount of points for municipalities and counties you'll end up with a lot more points per area in counties (in fact the density of your measurements will vary with the size of the area). I think it would be better to assume a density measure and then calculate the amount of points needed for each area.

The why question is essential for an answer. Why are you comparing distance as-the-crow-flies to open water bodies. For access to recreation by residents, (surface) water supply to households or industry, water fowl habitat? 

Hi, Karolina and everybody.

I think that the short answer to Karolina question would be the same of Josep: apply student-t significance test to these datasets.

And I agree with Alessandro and Josep that a centroid point is a weak measure to your analysis.

If you want to compare county-based and municipality-based datasets, Alessandro's and Josep's approach would be the best choices.

But if you want the best method to estimate water-distance at all levels, I would use a regular mesh of points to the whole country. Because I believe that using random points at any level could generate a bias once that Sweden (I am assuming that Sweden is your study area) has many lakes in its whole territory, but it looks like there are some "dry" areas with no lakes in some counties ( I don't know, is it by geological reasons?). A regular mesh would make your results more reliable once that your data results would not depend on randomness of how many points lie in a "dry area" or in a "lake area".

Regards from Brazil, Daniel.

You can compare them use the software of ARCGIS. It can derectly extract the parameter of distance-to-water when you have the layers of water distribution and the two compared layers no mater they are at two different geographic levels.

Thank you all for your answers!

The study was a mortality study, and the reason we studied distance to water at different geographic levels had to do with the fact that historical demographic data wasn't always available for the whole study area at the same level. Therefore, we tested if and how much results might have varied, depending on at which geographic level the calculation were performed at. The layers were different divisions of property units and something that today might have been regarded as municipalities (parishes). This fact together with the notion that we studied rural communities of the past were the reason why the centroid or the border of property units were used to estimate distance from various "high risk for getting an infection" areas. Since property units and their derivatives varied in size and shape and parishes also, I agree that Alessandro's approach would be the best one to implement. Regarding the statistical analysis of the multi-level results Josep's suggestion is very good!

I thank you all for your valuable insights!

best regards,

Karolina