The Fish and Wildlife Service recovery cards contain one 
piece of information that I came to regard as being especially useful 
in sampling analyses of bird survival. This is the fiscal year in 
which each report is received. In the later stages of my study at 
Patuxent, I discarded from life tables all cards that showed a 
recovery date differing from the fiscal year received. This practice 
of using only reports sent in promptly had the following effects: 
(1) It removed from each sample all reports in which the 
memory of the band recoverer is liable te error. 
(2) If the clerk transcribed the year of recovery incorrectly 
(especially in the early days of January), the error was promptly 
caught and the card eliminated. (Fiscal year received is usually 
stamped on cards.) 
(3) Samples for the more recent years are subject to less 
bias. Suppose, for instance, recoveries are carded and punched for 
those received through June 30, 1946. If the sample includes birds 
with the fiscal-year discrepancy, items for the year 195 might be 
changed by reports received in, say, 1949 or 1950. Life tables of 
this kind thus tend to be more incomplete for the later years than 
they are for the earlier ones. 
One tally kept on 801 waterfowl cards checked in this manner 
showed that 39 (4.9 per cent) were eliminated for this discrepancy; 
this test followed the removal of 25 duplicates. The percentage 
eliminated among non-game recoveries was probably lower. 
Errors in tabulation.—-In tabulating samples of band 
recoveries, 1 found that my own work was not entirely free of error. 
At the time I began my study at the Patuxent Research Refuge, the 
work of the banding office still suffered from the shortage of 
personnel brought on by the war. The process of coding and punching 
the IBM cards was far from complete. Some thirty or forty thousand 
cards, not yet filed, were in 12 different places in the office. No 
sample could be complete until these 12 places were examined. My 
wife took on the major task of reducing these 12 piles to 3, and of 
coding and punching cards for species which I was to analyze. A file 
of letters accompanying illegible bands was also processed and carded. 
A series of some 500 incomplete cards was likewise reduced to about 
3000. These cards bore recovery dates but lacked banding dates. They 
could not be further reduced in quantity without correspondence with 
a large number of banders who had failed to list the birds on their 
schedules. Some of the banders were now dead or no longer active. 
This material extended back to 1930; it seemed to include birds that 
were recovered before the banders sent in their reports. At least 
from the cards that my wife was able to complete, the bias from this 
source in my tabulations did not appear to be statistically significant. 
A second source of minor error developed when I sorted the 
cards by machine. Even when International Business Machine Company 
22 
