Skip to content

Configure Post-Stratified Ratio Variance Estimation

The numerator and denominator variances are computed with the same post-stratified estimator as ve_post_strat(). Their stratum-level covariance is

snd,h=∑iyn,hiyd,hi−(∑iyn,hi)(∑iyd,hi)/nhnh(nh−1)s_{nd,h} = \frac{\sum_i y_{n,hi}y_{d,hi} - (\sum_i y_{n,hi})(\sum_i y_{d,hi})/n_h}{n_h(n_h - 1)}

Covariances are rolled up through estimation units with the same post-stratification coefficient and across independent estimation units using squared population weights, giving vndv_{nd}. For R=Yn/YdR = Y_n/Y_d, the ratio variance uses the delta-method formula

v(R)=1Yd2[v(Yn)+R2v(Yd)−2R v(Yn,Yd)],v(R) = \frac{1}{Y_d^2}\left[ v(Y_n) + R^2v(Y_d) - 2R\,v(Y_n,Y_d)\right],

and the standard error is max⁡(v(R),0)\sqrt{\max(v(R), 0)}.

This calculation also exploits sparsity. It computes only the sufficient statistics ∑iyn\sum_i y_n, ∑iyd\sum_i y_d, and ∑iynyd\sum_i y_ny_d for observed plot rows; rows absent from one side contribute zero to the cross-product and do not need to be expanded into a dense plot-by-target table.

ve_post_strat_ratio()

A PostStratifiedRatioVarianceEstimator object.