Owing to this peculiar feature the rapidity of change in the proportion of the different forms is reduced to about one-half of what it would be if the males were also mimetic. Nevertheless the change from nearly equality to about one non-mimetic in 40 would have taken place {99} during the time _P. polytes_ has been known if a 2% selection advantage had operated during that period in favour of the mimetic. If there has been any appreciable selection going on during that time mimetics must have been far rarer when the species was first discovered, but the fact that both the mimetic forms made their way into collections before the non-mimetic tells distinctly against this supposition. Nor is there any reason to suppose that the non-mimetic form has been dwindling in numbers relatively to the mimetics during the last half century. Moore[51] in 1880 records an earlier observation of Wade's that "These three butterflies are very common, especially those of the first form; the second being perhaps least so." The first form alluded to is the M form, and the second is the A form, so that at the time Wade wrote the relative proportions of these three forms must have been very much what they are to-day. Even during half a century and with such a relatively weak selection rate as 2% in favour of the mimetics, the proportion of non-mimetics should drop from about 4:5 down to about 1:5. Therefore we must either infer that in respect of mimetic resemblances natural selection does not exist for _P. polytes_ in Ceylon, or else we must suppose its force to be so slight that in half a century certainly, and perhaps in a century and a half, it can produce no effect appreciable to the necessarily rough method of estimation employed. {100} It may, however, be argued that even an exceedingly low selection rate is able to bring about the elimination of one or other type provided that it acts for a sufficiently long time. This is perfectly true. A selective rate of .001% would reduce the proportion of recessives to dominants from 4:5 down to 1:40 in the course of about 1,400,000 generations where the mimetic resemblance is already established. Such a form of selection entails the death of but one additional non-mimetic in 100,000 in each generation. If, however, the mimetic resemblance is not fully established and the mimic bears only what supporters of the mimicry theory term a "rough" resemblance to the model, it is clear that it will have far less chance of being mistaken for the model. Its advantage as compared with the non-mimetic form will be very much less. Even supposing that the slight variations concerned are inherited, an intensity of selection which would produce a certain change in 1,400,000 generations where a mimetic resemblance is already established must be supposed to take an enormously greater time where an approach to a model has to take place from a "rough" resemblance.