H.Onslow 217 



(though in pnictico it apjMMirs to b** Ickk), In'cnuHf the Htinmlu.s nM|uin*(l 

 U) cuujk' a chan^f of Honsation is always a (Idiinti' fraction of tin* original 

 stimuluH. Thr jmliT coIohfh win thrn^fon* he nuMiHnriMl with ^rrat^T 

 *iccuracv than thi* darker. The chief error, however, is eauHe<| mther by 

 the ditfen'nees in tht* ei>Kmr t>f the wiii^ an-a selectee I for meiuiureiiH-nt, 

 than by the inability of the eye t.o <iiscrinniiale limr shades. As a 

 rule the yi'llow cohmr bec«)nies slightly paiei- towards the jHiiphery «»f 

 the wing. Therefore since exactly the same area cannot always again 

 be found, the experimental error is incn^Jised by rei)eating the mejisure- 

 nients. The following figures give some i<lea of \,\\v, ditterences in the 

 c<^li>ur of an average specimen : 



Area rouml (liscoidal spot 



., ht'tween discoiilal s\mA and central fascia 



„ near hinder angle ... 



,, near costa ... 



,, near apex ... 

 Central orange fascia 

 C)ran>je Khoulder knot ... 

 Black diHcoidal spot 



On account of this variability a i)ractice was always niade of selecting 

 the large yellow area surrounding the discoidal spot, the more perfect 

 wing of the two always being chosen. When this area was torn, or too 

 badly rubbed, the nearest portion, and therefore the one most approaching 

 the same colour, was selected. To avoid prejudice, all the readings were 

 taken before the curves were constructed, and in most cases the readings 

 were repeated twice. It was found from experience that at about the 

 range in question there was a maximum experimental error of 01 colour- 

 units. With regard to the (juestion whether the data prove that the 

 families and groups of families dealt wdth are really heterogeneous, and 

 not single samples, I have shown the evidence to Mr Udny Yule. He 

 has most kindly considered the matter and reports as follows : "As regards 

 figures obtained in the DR x RR cross (Figs. 4 to 9) I do not think there 

 can be any doubt that the data can be held to prove segregation. The 

 gap is wide and well marked and occurs in the same position in several 

 families or groups of families. In the case of the DR x DR cross 

 (Figs. 10 to 13) so well marked a separation cannot be expected, as the 

 second maxinnim does n(it rise to a sufficient height, and the distribu- 

 tion of the 72's only causes a comparatively slight hump on the tail of 

 the distribution. A check on the result can however be obtained by 

 building up the distribution that should be expi!cted on the theoretical 

 basis, showing that fair agreement is obtained." This check has been 

 applied with fairl}' satisfactory results. In the ca.se of the DR x DR 



