Conclusions

  • Measures of Central Tendency aren't enough! We need to characterize variation.
  • Variance provides error bars, characterizes distributions (e.g. the Normal), and distinguish means (ANOVA).
  • Complete Spatial Randomness is the default assumption in many spatial problems.
  • We have methods for detecting departures from complete spatial randomness; that is, spatial autocorrelation (e.g. Moran's I).
  • In estimation problems, the variogram is a sensible method for determining the effects of space, providing reasonable weights in an estimation scheme as a function of distance and direction.
  • Kriging is a good estimation scheme based on this spatial decomposition of the variance, and allows us to estimate the values of a variable away from data locations.

Links: