often graphed as a function of distance (x), and the resulting graph is called the correlogram [y = f(x)]. For discrete characters, the convention is to transform the autocorrelation into a standard normal deviate with mean 0 and variance 1, and this transformed function is graphed as the correlogram. Let us again take the concrete example of the common morning glory P/p locus. Three graphs result from the autocorrelation of blue with blue ( y_bb), pink with pink (y_pp), and blue with pink (y_bp), each as a function of distance. When graphed as a function of distance, the data from many local I. purpurea populations in the southeastern U.S. revealed a common pattern for the P/p locus y_pp or y_bb > 2 and y_bp < -2 initially (for the smallest values of x). This result indicates a positive autocorrelation over short spatial distances among like phenotypes (blue with blue or pink with pink) and a negative autocorrelation at short spatial scales among unlike phenotypes (blue with pink). The distance (x), where the autocorrelation first crosses the x axis (y = 0), provides an operational definition of patch size. These analyses revealed substantial patchiness for the P/p locus and yielded a minimum estimate of about 120 pink (recessive homozygous) plants per patch. The estimates of patch size are highly consistent with simulations of spatial distributions based on a pattern of pollinator flight distances that is strongly biased toward nearest neighbor moves (Turner et al., 1982; Sokal and Wartenberg, 1983).

In contrast to the P/p locus pattern, the W/w locus genotypes show little or no spatial patchiness (that is, y = 0 at the smallest values of x, in populations in which y > 2 or < -2 initially for the P/p locus). Because both loci are transmitted to successive generations within populations through the same mating process, we expect homogeneous spatial patterns. Heterogeneous spatial patterns argue that isolation-by-distance is not the sole factor governing spatial patterns within local populations. Other forces must be invoked to explain the W/w locus pattern, and the most likely appears to be selection against white homozygous plants. Epperson (1990) has carried out extensive simulation studies that tend to validate this conclusion, and, further, the simulations suggest an intensity of selection against white homozygotes (ww) of approximately 10%. Despite this conclusion, the frequency of white alleles varies from a low of 0% to a high of 43% across local populations with a mean value of about 10% (Epperson and Clegg, 1986). Because the various local populations can be thought of as quasi-independent experiments in which selection has operated for a number of generations, the persistence of white alleles argues strongly for countervailing selective forces favoring the retention of white phenotypes.

The spatial pattern of albino alleles, including unstable alleles (a*), is of interest because this locus also determines a white recessive phenotype that may experience selective pressures common to those experienced by