10 AN ANALYSIS OF THE EFFECT OF SELECTION. 
according to number of offspring, and is also recorded in the offspring 
generation. In the calculation of r, the parental grades are taken as 
the average grades of the two parents. When r is not given, it is not 
capable of calculation, for the reason that all parental pairs in that 
generation were of the same average grade. The correlation coeffi- 
cients given here are of doubtful significance, though many of them 
are several times their probable errors. These probable errors, like 
Taste 8.—S864, Inbred Plus Line. 
Generation. n M o r Diff. M. 
Bigsceve sect 113 1852672 0.048 1/50.-762==OKOS45| eeieclereieterersor cere —0.828 
Wiss ciasciayeiashne 121 | 5.331+ .049 1804.2 OSB IRs -rtsyatars) «toherpsrale —1.179 
Bgiiete tetera: 73 | 5.822= .031 O90 se sO | sista ave sterate eieveiave — .178 
IV Esasn coun 260 | 4.904+ .036 SSO8 ==) OZGe)|| (<5) «10e areis or stsielels —1.016 
Bigeye caevereesle 149 | 5.228+ .043 SGML tir OSO ni i covstiisveversueuanctotctons — .772 
Rice sucusvesg ails 120 | 5.450+ .044 SOB HS) esOSle ll si cfevsisiase stay scs tenet — .550 
Repke citoraete 510 | 5.190 .025 S880 se) OLS: | Perens score eee — .810 
ig ek tey.. eoe 461 | 5.475+ .023 .738+ .016 | +0.105+0.031 | — .514 
Big sicrecusicve (8 154 | 5.6434 .034 621+ .024 | + .002+ .054 | — .458 
Bygi ete cee 159 | 4.956 .051 EQGO==) ROSEN! tase seecoe tee —1.044 
tages tsetse 232 | 5.224+ .039 .867+ .027 | — .011+ .044 | — .901 
Wigs caveat 624 | 5.272+ .025 BOSSE) OLS all cetera stare tere stshenees — .728 
Bisco ay austere 353 | 5.787+ .024 .667+ .017 | — .070+ .036 | — .762 
Brant: scien 175 | 6.080+ .026 506+ .018 | + .133+ .050 | — .300 
3,504 
REVERSED SELECTION. 
ieee ess 33 | 5.152+0.102 | 0.869+0.072 | ..:..........- +0.652 
Fy. . 49 | 5.327+ .092 SO56= SSOGON |leeveretese) ocusteetesaret- +1.329 
Bua giteaessteresexe 62 | 5.710+ .052 AGNES WEST Il madgcbacooodGce +1.710 
144 
others of their kind, are intended only to give the magnitude of the 
error likely to arise from the fact that one is dealing with a sample of 
limited size—the error of random sampling. But in the present case 
the correlation coefficient is intended to measure the similarity be- 
tween the somatic appearance and the genetic possibilities of the 
parent individuals. It is known that this similarity does not amount 
to identitv, and that it may be modified in individual cases by en- 
vironmental causes. Since in any given case we are dealing with a 
rather small number of parent individuals, but a large number of off- 
spring individuals, the selection of one or two parents whose somatic 
appearance differs widely from their genetic possibility will throw 
the resulting correlation coefficient far off; but the large number of 
offspring will keep the probable error down. If, instead of entering 
each offspring individual in the correlation table separately, we enter 
only the mean grade of the offspring of each parent pair, we get what 
is perhaps a more reasonable probable error. But this method fails 
