ages are compared.* (2) Statistical tests cannot be depended upon 
to detect unrandomized types of sampling; very often common sense 
will. Among Paynter's artificial recoveries, for instance, there 
were those from short-term retrapping operations carried out at 
the nesting colony by members of the Bowdoin Scientific Station. 
These could hardly be depended upon to record nonbreeding birds in 
the first two or three years of life. Scientific collecting and 
local gull-contrc] work on this species likewise seem to me to be 
unrandomized types of sampling. Out of ll herring gulls thus re- 
ported, only 3 were in their first year of life. Removal of the 
li from the "artificial" sample would change the percentage of first- 
year birds in it from 60.1 to 66.1 per cent. 
Several statistical tests can be used to pring out differ- 
ences in survival or mortality series based upon different methods 
of recovery, such as birds reported shot, found dead, captured by 
fishermen, and the like. In many cases, of course, the samples may 
be too small to demonstrate significant differences, and in any 
case a purely mathematical test can only yield a limited amount of 
information. It can, for instance, prove that the first of two 
sources of recovery could not be draw from the same statistical 
universe as the second one, but it cannot prove which one is 
actually biased. Here one can see the overlap expected in the 
percentage of first-year birds found in two samples drawn from the 
same statistical universe. The probability that any two specific 
samples are significantly different in this respect can also be 
readily determined by a chi-square test. Computations of this type 
would show that the 46 per cent occurrence of first-year birds 
among Paynter's "natural recoveries" is significantly different 
from the 66 per cent occurrence in the "artificial recoveries" 
mentioned above. One or the other of these two sources is biased; 
a productivity analysis is still needed to indicate which one. 
Other statistical tests can similarly be used to test the differ- 
ence between the slope of two life curves, but the investigator 
must still decide which curve coincides with the known productivity 
of the species. 
Composite life tables for birds are only as reliable as the 
assumptions on which they are based. These assumptions, I am convinced, 
must be independently tested for each species or each group of species 
having similar habits. Mortality for waterfowl may, for instance, differ 
considerably from that reported for scavengers or raptores or songbirds. 

lpaynter's calculations of the mean length of life for both his 
samples were unfortunately carried out as if all his birds died 
at the conclusion of a year running from July 1 to June 30. A 
better approximation would be to assume that the recoveries were 
distributed around the midpoint of each year of life as Paynter 
did in calculating life expectancies in another part of his paper. 
30 
