vii ] DISCUSSION OF THE DATA OF STATURE. 131

and secondly of a tendency on the part of my correspondents to record medium statures when they were in doubt, whose effect would be to reduce the value of the Prob. Error. The R.F. F. data in Table 12 run so irregularly that I cannot interpret them with any assurance. The value they give for Fraternal Regression certainly does not exceed.. , and therefore a correction, amounting to no less than j- of its amount, is required to bring it to a parity with that derived from the Special data (because I + I x I- = 1). Hence it might be argued, that the value of Regression from Mid-Parent to Son, which the R.F.F. data gave as 4, ought to receive a similar correction. If so, it would be raised to 2+1= 4 ; but I cannot believe this high value to be correct. My first estimate made from the R.F.F. data, was g, as already mentioned. If this be adopted, the corrected value would be g, or. instead of 1, which might possibly pass. Curiously enough, this value of g for Regression from Mid-Parent to Son, coincides with the value of g for Regression from a single Parent to Son, which the direct observations showed (see page 99), but which owing to their paucity and to the irregularity of the way in which they ran, I rejected and still reject, at least for the present. While sincerely desirous of obtaining a revised value of average Filial Regression from entirely different and more accurate groups of data, the provisional value already adopted of 4 from Mid-Parent to Son may be accepted as being near enough for the present. It is impossible. to revise one datum in the

K2