RED SALMON COUNTS FROM OBSERVATION TOWERS 



367 



Table 7. — Stratiflcation of the Kvichak escapement in 

 1959 



' Determined as 10-mlnute counts lor uniformity of calculation. 



The final calculations of the confidence limits, 

 with the level of significance set at 95 percent 

 (#=1.96), are made from — 



= 11,376 [60.62 ±1.96 Vs=] 



= 689,613±(ll,376)(1.96)(Vl..5213) 

 = 689,613± (22,297) (1.233) 

 =689,613±27,492 



As calculated, the 95-percent confidence limits 

 for the 1959 Kvichak River escapement are equiv- 

 alent to ±3.99 percent (±27,492) of the 689,613 

 estimated total fish passage (P). Such narrow 

 limits point to the effectiveness of the sample- 

 count estimation program in current use. 



The sum of the daily estimates given in table 1 

 (680,000) differs slightly from the estimate cal- 

 culated from the stratified data. The tabulated 

 value is considered the best estimate, and is well 

 within the calculated confidence limits. 



Effect of Changing the Sampling Design 



The length of the counting period and the fre- 

 quency of the samples are the two main variables 

 to be considered in formulating a sampling pro- 

 gram. Increasing or decreasing either one in- 

 fluences the reliability of the calculated estimates. 



To analyze these variables, systematic test sam- 

 ples were drawn from a continuous 48-hour count 



595880 O— 61 3 



obtained in 195."). Tlie lengths of the counting 

 period of the samples tested were 10, 20, 30, 40, 

 and 60 minutes, while the sampling frequencies 

 were 1, 2, 3, and 4 hours. Generally, four different 

 tests were made with each relationship: sample 

 period length to sample frequency. Totals of the 

 calculated estimates were then compared with the 

 actual total for the 48-hour period, and the per- 

 centage deviation plotted in figure 11. 



The percentage of error tends to drop with an 

 increase in the length of the samples and with an 

 increase in the frequency of the samples. Short 

 counts, under 40 minutes in length, provide re- 

 liable calculations when taken every 1 or 2 hours, 

 usually ranging within a plus or minus 6 percent. 

 These counts show a much wider range of error 

 when taken every 3 or 4 hours. This indicates 

 that, for a specific reduction in counting time, the 

 most consistent results would be obtained with 

 short samples taken frequently. The 1955 and 

 1956 estimates, based on infrequent larger samples, 

 may have been less accurate than those in recent 

 years. However, in 1955, 73 percent of the escape- 



4 hour somple frequency 



20 30 40 50 60 



Length of Samples in Minutes 



Ficii HE 11.— Distribution of sampling errors with various 

 sample period lengths and frequencies. 



