The error of determination of toxicity / by J.W. Trevan.

  • Trevan, J. W. (John William)
Date:
1927?]
Catalogue details

Licence: In copyright

Credit: The error of determination of toxicity / by J.W. Trevan. Source: Wellcome Collection.

    23/36 (page 503)
    (= D) is 454 out of 455. The calculated frequency is 453-8. 436 times the error did not exceed ± 25 per cent. The calculated value is 435. 274 did not exceed d: 10 per cent. The calculated number is 263. Table V.—Assay of insulin. Comparison of observed frequency of deviations with frequencies expected if the distribution were normal with a standard deviation of 12-45 per cent. Percentage Devia¬ tion from Means of Frequency of Negative Deviations. Frequency of Positive Deviations. each Batch. Observed. Calculated. Observed. Calculated. 0 to 5 90 71-5 63 71-5 5 to 10 64 60 57 60 10 to 15 46 44-25 34 44-25 15 to 20 33 27-25 ' 20 27-25 20 to 25 11 19-5 16 19-5 25 to 30 4 6-3 8 6-3 30 to 37-4 1-0 3-1 7 3-1 >37-4 {Actual value 42) 0 0-6 1 0-6 The figures are plotted in fig. 8. The dotted lines represent the normal distribution for a standard deviation of i 12-45 per cent, and the full lines the observed frequencies. Inspection of this figure suggests that the distribution actually obtained is not quite normal. There is an excess of values between 0 and — 5 per cent., a deficit for extreme values in the negative direction and an excess of extreme positive values. These facts suggest that the true form of the distribution is slightly skew, with the mode on the negative side of the mean between 0 and — 5 per cent. I have tested the significance of the variation from the normal distribution by calculating y2 (Fisher, Cap. 4) for the figures in Table V. The value of f after combining the frequencies of deviations above + 30 per cent, and also those below — 30 percent, works out at 24-224 ; P is therefore 0-03, taking n as 13. A larger value of y2 would only occur about once in 35 such sets of 455 tests if the distribution were normal. The difference between the normal distribution and that obtained is probably therefore just significant. That the distribution is not quite normal depends in the end on the asymmetry of the insulin characteristic. The dose errors for a convulsion rate of 50 per cent, obtained from the horizontal distances between the characteristic and the dotted lines (fig. 7) are + 37-6 per cent. — 34-4 per cent., an asymmetry in the