Errors in sizing bands of hypervariable DNA profiles on autoradiograms: Are they Gaussian?

The accuracy of procedures for sizing hypervariable restriction fragments by Southern blot analysis (SBA) has been tested under three different experimental conditions: (i) intrablot serial analyses: three heterozygous DNA profiles were tested 14 times each in the same gel electrophoresis; (ii) intralaboratory analyses: we replicated three profiles (six autoradiographic bands) in over 100 SBA experiments; (iii) interlaboratory analyses: 15 serial measurements produced in a recent collaborative study (Forensic Sci. Int. 1991, 49, 1-15) were taken into account. In these three cases, a typical U-shaped correlation curve between molecular size and coefficient of variation was found. We explain decrease of accuracy at both extremities of the gels in terms of: (i) enlarged shapes of bands and mean electrophoretic resolution at the cathode; (ii) diffusion and blurring of bands at the anodal edge. The three populations of data were subjected to a chi 2 test and to the Kolmogorov-Smirnov test in order to verify their compliance with normal distribution. Twenty-three out of 27 tests indicated no significant deviation from the assumption of Gaussian distribution. We recommend the adoption of tests for normality to validate the use of symmetric confidence intervals for calculating gene frequencies and asserting a match between adjacent bands.