Fig. 3. Correlates of post-fire seeding treatment application.

Marginal effects of covariates correlated with the occurrence of restoration treatments, illustrating observed sources of selection bias (n = 20,000 observations). The distribution of observed values for these covariates in treated and untreated sites is shown above each marginal effect panel. Panels illustrate predicted effects of pre-fire sagebrush cover (a), surviving, unburned sagebrush cover estimated immediately following the fire (b), November-April total precipitation (c), February-April mean temperature (d), distance from a road (e), and soil percent clay (f) on treatment probability, while holding all other covariates at their means and assuming observations occurred in a commonly treated ecoregion (Snake River Plain). Solid lines represent median posterior predictions (based on model parameters shown in Fig. 2), with shaded bands indicating 50% and 95% credible intervals around these predictions.