The sampling characteristics of electivity indices.

Electivity indices measure the utilization of food types (r) in relation to their abundance or availability in the environment (p). Foods that constitute a larger proportion of the diet than of the available foods are considered preferred; conversely those proportionately underrepresented in the diet are avoided. A food is eaten at random if its proportion in the diet equals its proportion in the environment. A family of electivity indices stemming from Ivlev's (1961) classic monograph exist and differ only in the particular algorithm used to calculate electivity from r and p.For each available index I graphed the values of electivity as contours for all combinations of r and p. These graphs are compared to illustrate the strengths and weaknesses of each index on the basis of the following criteria: 1) the value of the index when r=p for a food, 2) the symmetry of the electivity value as feeding deviates from random, 3) the possible range of index values, 4) the linearity of changes in electivity over the full range of r and p, 5) the sensitivity of the index to sampling errors, 6) the statistical testability of the electivity, and 7) the stability of the electivity value for a food type that changes relative abundance or occurs in combination with different food types. No one index ideally satisfies all the criteria.The host preferences of gypsy moth, Lymantria dispar, feeding on tree foliage in an undisturbed deciduous forest in southwestern Quebec, Canada were used to compare the available indices: Ivlev's electivity, E; Ivlev's forage ratio, E'; Jacob's modified electivity, D; Jacob's modified forage ratio, log Q; Chesson's alpha; Strauss' linear index, L; and Vanderploeg and Scavia's relativized electivity, E *. The electivity values calculated by each index differ one from another; host trees shown as preferred by one index will frequently appear avoided according to an alternative index. The rank order electivities for the 19 available host trees, however, are remarkably similar for all but Strauss' linear index, L. Populus grandidentata, Quercus rubra, Ostrya virginiana, and Amelanchier were the most preferred host trees in the sampled forest; Prunus serotina, Acer pensylvanicum, A. rubrum, Betula lutea, and Fraxinus americana were most avoided. The use of Vanderploeg and Scavia's E * index is recommended.