Coefficient of discrimination (CoD) and relative change in CoD (ΔCoD) as predictors are removed from logistic regression models predicting the probability of reaching 100 infected premises.

The predictors are the county-level clustering at a spatial scale relevant for the kernel, (Clustering), the type of configuration (random or FLAPS, Landsc. conf.), the logarithm of the average premises size in the seeded county, (Avg. prem. size), the logarithm of the number of premises in the seeded county (N. prem.), ln Nw and the logarithm of the premises size of the seeded premises, lnni (Seeded size). The model coefficients for these predictors as well as nine additional binary predictors indicating which out of ten different realizations of the landscape configuration are found in the supplementary material (S2 Table and S3 Table).