Table 3 gives evidence that the index of error varies from age to 

 age, being higher for the older tlian for the yo^unger classes; and, there- 

 fore, the error on age composition must vary according 'to the number of 

 old fish present'. The index of error varies from reader to reader; also 

 for any given reader, from time to time, being more or less proportional 

 td the degree of his fatigue. It has been our experience that rreaders some- 

 times unconsciously adopt faulty reading techniques irhich result in their 

 making some types of errors consistently over a peribd of time. Thus, 

 Reader C (table 3) tended during the test to overlook the marginal ring 

 on scales with two annuli, and -consequently he recorded an excess of ' year- 

 lings. Faults of this kind ape.; detected and eliminated during the joint 

 readings . 



To avoid such errors, frequent tests should be made. Indeed, it would 

 be desirable to make parallel readings of all scales wore it not impractical 

 in routine work involving several hundred specimens weekly. Fortunately, 

 at least 95 percent of the California .catch is composed of fish younger 

 than 5 years, for v/hich the index of error is low enough to justify sub- 

 stituting for parallel readings the procedure described below. For the. 

 remainder, the relatively high error necessitates making parallel readings 

 on all specimens. 



In the light of the foregoing results, the following procedure was 

 adopted for future routine v/ork: The slides will be dealt among the several 

 readers like playing cards, so thas-^ all have equal-sizedj randomly distrib- 

 uted portions that should prove to be satistically identical as to body 

 length composition. That being true, the age compositions of the several 

 portions should also be statistically identical; and significant differences 

 among them should be detectable by the chi-square test, as described below. 

 IVherover such differences occur, the scales v/^ill -be re-read and the cause 

 of the differences found and cori'octcd. Mcanvvliile, specimens aged as five, 

 years or older iid.ll be given parallel readings; and joint readings will then 

 be made of the disagreements. 



Follovfing thia procediire, the scale collections taken during the 19liO- 

 Ul season at San Francisco,. 3i210 in number, v^ere dealt between Vifalford and 

 Mosher. Tha1^ the t^TO portions were statistically similar in size composition 

 is indicated by a chi-square comparison, ivhich gave a probability of 0.53 

 that a second pair of samples dravm from the same population vro\ild differ 

 in size composition as much or more by chance alone. 



For the same reason, if these tv;o random portions mqto alike in size 

 composition, they should be alike in age composition. A chi-square coiiparison 

 was therefore made between tho totals of each year class in Ifalford's portion 

 •and those in Lloshcr's. This test gave a probability of 0.13 that a sucond 

 pair of samples dravm from the same population would differ in ago composi- 

 tion as much or more by chance (chi-squaro vj-as 8,U Vi'lth 5 degrees of freedom). 

 Though high enough to be acceptable as evidence th^it Walford and Mosher had, 

 in the main-, read their respective lots of scales alike, it is at the same 

 time low enough to suggest the presence of sc«nc non-random differences between 

 the tvra scries of age distributions, due, possibly to some difference in the 

 interpretations of the two readers. 



104 



