Paper 3 · June 2023 · Statistics: measurement scales, Poisson, chi-square, spatial autocorrelation
Why does spatial autocorrelation matter when a geographer applies classical statistical tests to spatial data?
AClassical tests assume observations are independent, and spatially autocorrelated residuals break that assumption, giving artificially small standard errors and false significance.
BIt converts ordinal data into interval data, which is what classical tests such as the t-test are designed to accept as input.
CIt removes the need to sample at all, because a spatially autocorrelated variable can be measured once and applied to the whole study area.
DIt guarantees that the distribution of the variable is normal across the study area, so a parametric test such as the t-test can always be used in place of a non-parametric one without any further checking of the data.
Explanation
Nearby observations that resemble one another are not independent, so each carries less new information than the test assumes. The result is understated standard errors and relationships that appear significant when they are not, which is why autocorrelation must be tested for before a classical test is trusted.
Derived from ZIMSEC Advanced Level Geography 6037/3 Paper 3 Practical Test, June 2023 session, Q1