Appendix B. Description of experimental methods and statistics used to determine the effect of host density and dispersion, and presence of a habitat boundary, on *Anagrus* performance.

For this experiment, patches were created using plastic 35 cm × 47 cm × 15 cm tall bus pans that were filled with excavated clumps of emergent sprangletop (*n* = 160 bus pans). Bus pans, representing small sprangletop patches, were transported to an outdoor garden facility at the University of North Dakota (Grand Forks, ND) and sprayed with insecticidal soap to kill planthoppers (all in the nymphal stage at this time).

To achieve different densities and dispersion patterns, I manipulated the number of infested stems per cell by caging adult planthoppers on individual stems. The stem cage was constructed of a 10 cm long by 2.5 cm diameter acetate tube (see Cronin and Wilson 2007). High-density foam sealed the ends of the tube around the stem. Into each cage, I added three female planthoppers and allowed them to oviposit for 48 h. If all planthoppers within a cage died within 24 h, they were replaced. From the survey study, the mean number of infested stems per dm^{2} ranged from 0.6 ± 0.16 in small patches (*n* = 10 patches) to 1.1 ± 0.27 in large patches (*n* = 6 patches). I used two density treatments in this experiment. The low density treatment was set at a mean of 0.6 infested stems/cell (a
total of 9 infested stems per bus pan) and the high density treatment was triple that value, a mean of 2 infested stems/cell (30 infested stems per bus pan).

There were three dispersion treatments that were fully crossed with the density treatment: uniform, random, and clumped distributions. For the uniform treatment, the infested stems were evenly distributed among cells. In the low density treatment, cells 1,3,5,6,8,10,11,13,15 (counting left to right in rows 1–3) each had one
infested stem. In the high density treatment, 2 stems per cell were infested. For the random treatment, I used MATLAB 4.0 (The Mathworks, Natick, MA) to generate 1000 Poisson distributions of infested stems (*n* = 15 cells) with a pre-defined cell mean of 0.6 or 2.0 and a variance-to-mean ratio that deviated from one by no more than 10%. I randomly selected one distribution and assigned it to a bus pan for that treatment. Using a similar procedure, I generated 1000 clumped distributions of infested stems based on a negative-binomial model with fixed cell mean and a clumping parameter *k* that was within 10% of the desired value, 0.3.

A third treatment in this study was the location of the bus pans. One half of the bus pans were placed at ≥5 m from any other sprangletop and were meant to represent small island patches with discrete boundaries. The other half of the bus pans were embedded within a mainland habitat (>250 m^{2}), and spaced ≥5 m apart. These mainland bus pans lacked distinct borders. The three treatments were fully crossed and each treatment combination was replicated eight times.

The effects of planthopper density, dispersion, and bus pan location on *Anagrus* density, the proportion parasitized and per capita parasitized were determined with a full factorial ANCOVA. Total stem density within each bus pan was treated as a covariate. Differences among dispersion levels were assessed with Tukey's tests. Normality and homogeneity of variances were achieved by arcsine square-root transforming the proportion parasitized and ln-transforming *Anagrus* density (no./m^{2}), the per-capita parasitized, and no. stems/m^{2}.

LITERATURE CITED

Cronin, J. T., and S. W. Wilson. 2007. Description, life history and parasitism of a new species of delphacid planthopper (Hemiptera: Fulgoroidea). Annals of the Entomological Society of America 100:640–648.