Heritability Estimates of Transformations of Normally Distributed Records

¹ Department of Animal Science, Cornell University, Ithaca, NY 14853

Milk yields were simulated for identical twin and parent-progeny parts to determine effect of transformations on genetic variance. Use of identical twin pairs excludes variability due to Mendelian sampling and contribution by the other parent present in parent-progeny records. Phenotypic variances of 616,800; 1,439,200, and 2,261,600 and means from 2267 to 13,603 by increments of 2267 were used for linear scale records. Phenotypic variances of .01000, .01875, and .02750 and means equal to the natural logarithm of the means for linear scale were used for log scale records. Heritability varied from .05 to .95 by increments of .10. For each combination of parameters, 10 replicates of 10,000 pairs of relatives were created. Data normally distributed on a linear scale were log transformed. Log normally distributed data were exponentially transformed. Heritability estimates from correlations among relatives indicated heritability is larger for normally distributed data than for transformed data. For parent-progeny pairs, the difference increased as heritability increased, and for identical twin pairs, the differences were least at extremely large and small heritabilities. For both types of relative pairs, the difference increased as phenotypic variance increased Absolute differences in heritability ranged from .0000 to .0089, a difference of little practical importance.

Key Words: heritability • logarithmic • transformations • log normal distribution

Submitted on November 20, 1989
Accepted on October 5, 1990

This article has been cited by other articles:

				D. B. Allison, B. S. Gorman, and E. M. Kucera Unicorn: A Program for Transforming Data to Approximate Normality Educational and Psychological Measurement, August 1, 1995; 55(4): 625 - 629. [Abstract]