Use of subjective prediction in optimal stratified sampling with application to shrimp surveys in the Barents Sea
Publication details
Journal : Journal of Northwest Atlantic Fishery Science , vol. 27 , p. 139–150 , 2000
International Standard Numbers
:
Printed
:
0250-6408
Electronic
:
1813-1859
Publication type : Academic article
Links
:
FULLTEKST
:
http://journal.nafo.int/J27/vo...
If you have questions about the publication, you may contact Nofima’s Chief Librarian.
Kjetil Aune
Chief Librarian
kjetil.aune@nofima.no
Summary
The task of applying subjective knowledge in predicting the number of trawl samples per stratum that minimises the coefficient of variation (CV) of the abundance estimator was considered. The constraint was the vessel time available. It was assumed that the strata biomass means, arbitrarily scaled, are the only unknown parameters needed to find the optimal solution. The concept of a subjective prediction distribution of the unknown stratum means was introduced. The distribution was described as person-dependent and determined based on intervals [L,U] for the minimum and maximum subjectively predicted biomass values compared with the true measured values found after the predictions. The approach assumed a constant subjective confidence level defined as the probability of covering the true value in a random interval. A pilot subjective prediction experiment was conducted during the 1998 shrimp survey in the Barents Sea. Based on 62 [L,U] predictions of shrimp biomass in the next trawl haul combined with the true biomass, the subjective prediction distribution for the cruise leader was estimated. The distribution was applied to the stratum predictions for the next survey. 10 000 random predictions of true strata means were simulated from the distribution. For each simulation, CV values of the abundance estimator were estimated based on relative strata means predicted from historical data as well as the subjective predictions. A significant CV reduction was obtained based on a combination of subjective prediction and historical data, compared to the use of historical data alone.