On the use of aggregated vs individual data in assessment models
Working Paper
2016
Permanent link to this Item
Authors
Journal Title
Link to Journal
Journal ISSN
Volume Title
Publisher
Publisher
University of Cape Town
Faculty
Series
Abstract
The conventional two-step process in fisheries assessments, whereby data are first aggregated to provide typically annual values before those are input to the assessment model, is compared to a single-step process where the individual data are input directly to the assessment model. The key point at issue is whether or not the latter process would provide estimates of key parameters that are (and are reliably estimated to be) more precise in circumstances where there is non-independence in the individual data. Arguments are offered that this non-independence does not introduce bias into estimates of precision for the aggregated case when observation error variance in the data is much less than process error variance in the assessment model. The utility of the random effects approach for addressing non-independence through working with individual data in a single-step process is queried; this is because of uncertainty about the bias in estimates of precision that may arise because of a lack of knowledge in most situations whether the structure assumed for the random effects will adequately account for the actual (and usually unknown) sources of non-independence in the data. Some aspects of the issue are illustrated by quantitative examples.
Description
Reference:
Butterworth, D.S. 2016. On the use of aggregated vs individual data in assessment models. DAFF Branch Fisheries document: FISHERIES/2016/AUG/SWG-PEL/65: 6pp.