PARRACK: FISHING EFFORT FROM AERIAL SEARCH DATA 



Socialist Republics (USSR) reported days on 

 grounds in 1969 and 1972 only, so that estimates 

 for 1970-72 were based on calculations of P{G/N) 

 and K from 1969 data. The 1973 estimate was 

 based on the average of the 1969 and 1972 values. 

 Spanish paired trawl days on grounds were 

 reported in these same years so that calculations 

 via estimator II were achieved in the same way as 

 for the USSR. Spanish stern trawl and Japanese 

 days on grounds were first reported in 1972 so that 

 the 1973 calculation of days fished by estimator II 

 was based on the 1972 data only. Poland and 

 Romania reported days on grounds in 1969 only, so 

 that all calculations by estimator II were based on 

 P{G/N) and /C values computed from 1969 data. 



Estimates of days fished were then made for 

 each country-gear partition, for each country's 

 total effort, and for all stern trawlers combined 

 and all side trawlers combined (Table 3). Estima- 

 tor II was not used to estimate effort for Bulgaria, 

 the Federal Republic of Germany (FRG), Japan 

 for 1970 and 1971, and Spanish stern trawlers in 

 1972 because of the absence of reported days on 

 grounds which is required by the estimator. 



A coeflficient of estimation error was calculated 

 to establish a measure of estimator performance: 



f = (/-/)//• 



(4.0) 



This error coefficient, then, is the difference 

 between the estimated days fished (^0 and the 

 reported days fished (/) expressed as a proportion 



of the reported value. An error coefficient was 

 computed for each estimate made and these 

 coefficients (Table 4) were then evaluated to es- 

 tablish the results of partitioning, to compare the 

 relative abilities of the three estimators, and to 

 establish estimator dependability. 



Inspection of error coefficients indicated that 

 they decreased considerably (especially those of 

 estimators II and III) after 1970, likely as a result 

 of the averaging of estimation parameters. Since 

 the error coeflftcients then tended to stabilize, only 

 values of e for the 1971-73 period were used to 

 analyze estimator performance. 



The frequency distribution of e for estimator II 

 is slightly negatively skewed, a characteristic also 

 exhibited by the distribution of t for estimator III 

 (Figure 2). This indicates a positive bias in both 

 estimators (approximately 10% in each case). Each 

 of these two distributions is also noticeably lepto- 

 kurtic indicating a marked clustering of error 

 coefficients in the interval *0.10. The distribution 

 of e for estimator I appears to be approximately 

 symmetrical and without the pronounced peaked- 

 ness exhibited by the other two. Statistics were 

 computed from the calculated error coefficients to 

 establish the probability that the f came from 

 normal distributions. (These statistics, a and 6,, 

 and tables of their probabilities are given by 

 Pearson and Hartley 1956:61-62, 183.) In the case 

 of the error coefficients of estimators II and III, 

 the probability that the error coefficients come 

 from normal distributions is extremely remote, 



Table 3.-Estimated days fished calculated by three different algorithms. 



509 



