A FURTHER ANALYSIS OF BUMPUS' DATA: THE INTENSITY OF NATURAL SELECTION.

Abstract

On February 1, 1898, a very severe snowstorm hit New England. A number of English sparrows, presumably exhausted in the storm, were brought to the Anatomical Laboratory of Brown University in Providence, Rhode Island. Out of a total of 136 birds, 72 survived and the rest died. Bumpus (1898) measured nine characters on the birds. They were also classified by sex and the males into young and adults. The characters measured were the following: (i) total length in millimeters; (ii) alar extent from tip to tip of the wings, in millimeters; (iii) weight in grams; (iv) length of head in millimeters; length of (v) humerus, (vi) femur, (vii) tibio-tarsus, (viii) bones across skull and (ix) sternum, all measured in inches to three decimal places. These data have been analysed several times by different authors. Most recently, Johnston et al. (1972) have applied principal component analysis to the data to examine the differences between the surviving and dead birds. Their analysis of the nine characters confirms Bumpus' own calculations which showed that the main differences were in the means and mean deviations of the characters of total length, wingspan, lengths of humerus and femur and weight. Johnston et al. analysed the data in more detail when the characters of total length, wingspan and weight were excluded. Weight must necessarily depend largely on the condition of the birds and should obviously be excluded from the calculations. But total length and wingspan were excluded on the grounds that ornithologists consider them to be too variable and too subject to error to be useful. Since total length is not more variable than the other characters in Bumpus' data, this seems an inadequate reason for excluding it. I have calculated simple discriminant functions from Bumpus' data, using the characters of total length, and length of humerus and femur, since these were the characters that were most obviously subjected to selection in the storm. I have then used the statistics of the surviving and dead birds to calculate the proportion of selective deaths and the variance in fitness according to various models of natural selection (O'Donald, 1968, 1970). A severe snowstorm is a very extreme environment which animals are exposed to on only a few occasions each winter. But on these occasions, a population may well be subjected to a large part of the total selection imposed upon it in the whole year. If so, most of the selective deaths will then be suffered at these times. The variance in fitness will also be much increased. The actual selective values should therefore vary much more widely and it is interesting to calculate what variations in fitness affected the sparrows that Bumpus observed.