Does the OV/postposition harmonic word-order correlation survive controls for genealogical and areal autocorrelation? A WALS/Grambank test

Greenberg's word-order correlations predict that verb-object (VO/OV) order harmonizes with adposition order: OV languages tend to be postpositional, VO prepositional. Naive cross-linguistic counting violates independence (Galton's problem): related and neighbouring languages share features by descent and contact. QUESTION: using WALS feature 83A (order of object and verb) crossed with 85A (order of adposition and noun phrase), and/or the corresponding Grambank features, over the maximal language sample, (a) estimate the OV<->postposition association (odds ratio and $\phi$ coefficient), then (b) re-estimate it under at least two independence controls -- family-level aggregation (one datum per stock) and a phylogenetic/areal model (phylogenetic generalized least squares, a Bayesian phylogenetic mixed model, or a family-and-macroarea-stratified permutation test). Does a significant harmonic correlation remain, and how much does the effect size shrink under control?

Acceptance. FULLY RESOLVES: over WALS 83A x 85A (and/or matching Grambank features), report the raw association (odds ratio, $\phi$, $p$) plus at least two independence-controlled estimates -- (i) family-level aggregation and (ii) a phylogenetic/areal mixed model or family+macroarea-stratified permutation test -- each with effect size and $p$-value; state whether the correlation stays significant and quantify the shrinkage; ship a data-extraction + analysis script from the public CLDF releases. PARTIAL: the family-aggregation test alone, a single controlled estimate, or the same analysis for a different Greenbergian pair (e.g. genitive-noun order vs. adposition order, WALS 86A x 85A).