Data weighting and data conflicts in fishery stock assessments Chris Francis Wellington, New Zealand CAPAM workshop, “ Data conflict and weighting, likelihood.

Slides:



Advertisements
Similar presentations
Data Mining Methodology 1. Why have a Methodology  Don’t want to learn things that aren’t true May not represent any underlying reality ○ Spurious correlation.
Advertisements

458 Quantifying Uncertainty using Classical Methods (Likelihood Profile, Bootstrapping) Fish 458, Lecture 12.
Gavin Using size increment data in age-structured stock assessment models CAPAM growth workshop: Nov 2014.
TRIM Workshop Arco van Strien Wildlife statistics Statistics Netherlands (CBS)
An evaluation of alternative binning approaches for composition data in integrated stock assessments Cole Monnahan, Sean Anderson, Felipe Hurtado, Kotaro.
Growth in Age-Structured Stock Assessment Models R.I.C. Chris Francis CAPAM Growth Workshop, La Jolla, November 3-7, 2014.
Model Selection for Selectivity in Fisheries Stock Assessments André Punt, Felipe Hurtado-Ferro, Athol Whitten 13 March 2013; CAPAM Selectivity workshop.
AIGKC Model North Pacific Fishery Management Council Crab Modeling Workshop Report.
CMPUT 466/551 Principal Source: CMU
The current status of fisheries stock assessment Mark Maunder Inter-American Tropical Tuna Commission (IATTC) Center for the Advancement of Population.
Resampling techniques Why resampling? Jacknife Cross-validation Bootstrap Examples of application of bootstrap.
Statistics for Managers Using Microsoft® Excel 5th Edition
458 More on Model Building and Selection (Observation and process error; simulation testing and diagnostics) Fish 458, Lecture 15.
Statistics for Managers Using Microsoft® Excel 5th Edition
458 Model Uncertainty and Model Selection Fish 458, Lecture 13.
Specific to General Modelling The traditional approach to econometrics modelling was as follows: 1.Start with an equation based on economic theory. 2.Estimate.
458 Fitting models to data – II (The Basics of Maximum Likelihood Estimation) Fish 458, Lecture 9.
Handling Categorical Data. Learning Outcomes At the end of this session and with additional reading you will be able to: – Understand when and how to.
An Introduction to Logistic Regression
How many conditional age-at-length data are needed to estimate growth in stock assessment models? Xi He, John C. Field, Donald E. Pearson, and Lyndsey.
Confidence Intervals and Hypothesis Tests
Time-Varying vs. Non-Time- Varying Growth in the Gulf of Mexico King Mackerel Stock Assessment: a Case Study Southeast Fisheries Science Center Jeff Isely,
Ministry of Food, Agriculture and Fisheries Danish Institute for Fisheries Research The overlap between Science and Advice; the example of North Sea cod.
LEARNING PROGRAMME Hypothesis testing Intermediate Training in Quantitative Analysis Bangkok November 2007.
Lecture 12 Model Building BMTRY 701 Biostatistical Methods II.
WP4: Models to predict & test recovery strategies Cefas: Laurence Kell & John Pinnegar Univ. Aberdeen: Tara Marshall & Bruce McAdam.
Standardizing catch per unit effort data
Slide 1 Estimating Performance Below the National Level Applying Simulation Methods to TIMSS Fourth Annual IES Research Conference Dan Sherman, Ph.D. American.
Name: Angelica F. White WEMBA10. Teach students how to make sound decisions and recommendations that are based on reliable quantitative information During.
Surplus-Production Models
ASSESSMENT OF BIGEYE TUNA (THUNNUS OBESUS) IN THE EASTERN PACIFIC OCEAN January 1975 – December 2006.
Statistics 1: tests and linear models. How to get started? Exploring data graphically: Scatterplot HistogramBoxplot.
Evaluation of a practical method to estimate the variance parameter of random effects for time varying selectivity Hui-Hua Lee, Mark Maunder, Alexandre.
Integrating archival tag data into stock assessment models.
The Stock Synthesis Approach Based on many of the ideas proposed in Fournier and Archibald (1982), Methot developed a stock assessment approach and computer.
Discussion of time series and panel models
Workshop on Stock Assessment Methods 7-11 November IATTC, La Jolla, CA, USA.
Chapter 10 Verification and Validation of Simulation Models
Fisheries 101: Modeling and assessments to achieve sustainability Training Module July 2013.
Flexible estimation of growth transition matrices: pdf parameters as non-linear functions of body length Richard McGarvey and John Feenstra CAPAM Workshop,
M.S.M. Siddeeka*, J. Zhenga, A.E. Puntb, and D. Pengillya
U.S. Department of Commerce | National Oceanic and Atmospheric Administration | NOAA Fisheries | Page 1 Model Misspecification and Diagnostics and the.
Management Procedures (Prof Ray Hilborn). Current Management Cycle Fishery: Actual Catches Data Collection Assessment Management Decision.
The effect of variable sampling efficiency on reliability of the observation error as a measure of uncertainty in abundance indices from scientific surveys.
Estimation of growth within stock assessment models: implications when using length composition data Jiangfeng Zhu a, Mark N. Maunder b, Alexandre M. Aires-da-Silva.
Simulation analysis using stock synthesis Maunder, Piner, and Lee.
Modeling biological-composition time series in integrated stock assessments: data weighting considerations and impact on estimates of stock status P. R.
1 Federal Research Centre for Fisheries Institute for Sea Fisheries, Hamburg Hans-Joachim Rätz Josep Lloret Institut de Ciències del Mar, Barcelona Long-term.
Using distributions of likelihoods to diagnose parameter misspecification of integrated stock assessment models Jiangfeng Zhu * Shanghai Ocean University,
Estimation of selectivity in Stock Synthesis: lessons learned from the tuna stock assessment Shigehide Iwata* 1 Toshihde Kitakado* 2 Yukio Takeuchi* 1.
CAN DIAGNOSTIC TESTS HELP IDENTIFY WHAT MODEL STRUCTURE IS MISSPECIFIED? Felipe Carvalho 1, Mark N. Maunder 2,3, Yi-Jay Chang 1, Kevin R. Piner 4, Andre.
Selectivity and two biomass measures in an age-based assessment of Antarctic krill Doug Kinzey, George Watters NOAA/NMFS/SWFSC/AERD CAPAM Workshop, March.
Some Insights into Data Weighting in Integrated Stock Assessments André E. Punt 21 October 2015 Index-1 length-4.
MSE Performance Metrics, Tentative Results and Summary Joint Technical Committee Northwest Fisheries Science Center, NOAA Pacific Biological Station, DFO.
Lecture 10 review Spatial sampling design –Systematic sampling is generally better than random sampling if the sampling universe has large-scale structure.
A bit of history Fry 1940s: ”virtual population”, “catch curve”
Hypothesis Testing. Statistical Inference – dealing with parameter and model uncertainty  Confidence Intervals (credible intervals)  Hypothesis Tests.
Regression Analysis: A statistical procedure used to find relations among a set of variables B. Klinkenberg G
Testing for equal variance Scale family: Y = sX G(x) = P(sX ≤ x) = F(x/s) To compute inverse, let y = G(x) = F(x/s) so x/s = F -1 (y) x = G -1 (y) = sF.
Influence of selectivity and size composition misfit on the scaling of population estimates and possible solutions: an example with north Pacific albacore.
NWFSC A short course on data weighting and process error in Stock Synthesis Allan Hicks CAPAM workshop October 19, 2015.
Data requirement of stock assessment. Data used in stock assessments can be classified as fishery-dependent data or fishery-independent data. Fishery-dependent.
Day 4, Session 1 Abundance indices, CPUE, and CPUE standardisation
Recommended modeling approach Version 2.0. The law of conflicting data Axiom Data is true Implication Conflicting data implies model misspecification.
Is down weighting composition data adequate to deal with model misspecification or do we need to fix the model? Sheng-Ping Wang, Mark N. Maunder National.
Chapter 10 Verification and Validation of Simulation Models
Effects of Catch-at-Age Sample Size on Gulf of Mexico Gray Triggerfish Spawning Stock Biomass Estimates Jeff Isely SEFSC Miami.
Homoscedasticity/ Heteroscedasticity In Brief
Homoscedasticity/ Heteroscedasticity In Brief
Regression Analysis.
Presentation transcript:

Data weighting and data conflicts in fishery stock assessments Chris Francis Wellington, New Zealand CAPAM workshop, “ Data conflict and weighting, likelihood functions, and process error” La Jolla, October 2015 Francis (2011). Data weighting in statistical fisheries stock assessment models. CJFAS 68: Francis (2014). Replacing the multinomial in stock assessment models: A first step. Fish. Res. 151:70– 84

Why is data weighting important? 1. Can strongly affect stock status 2. All statistical inference assumes correct weighting Increasing weightings tend to decrease estimated uncertainty (and vice versa)

Example of effect of reweighting on s.e.s (Chilean ling) Eff. sample sizesStandard errors ModelAgeLengthB 0 (t)Depletion(%)F final Initial Eff. sample sizesStandard errors ModelAgeLengthB 0 (t)Depletion(%)F final Initial Reweight108 Eff. sample sizesStandard errors ModelAgeLengthB 0 (t)Depletion(%)F final Initial Reweight Change in s.e. +49% +31%+21%

Observation Model prediction Error Real world Observation error Process error The weight given to each observation should describe the likely size of its error (and thus its information content) What is “correct” data weighting?

How to set data weights? Abundance data: outside the model (see Francis 2011); heteroscedastic if possible; no iterative reweighting Composition data*: requires iterative reweighting because it’s not possible to quantify process error outside the model Some causes of process error for composition data: - misspecification of selectivity, M, and growth (for lengths) *Including tagging: propns by age or length of recaptures

Iterative reweighting of compositions 1.Set initial data weights* 2. Run assessment model 3. Adjust weights using model residuals [4. Repeat steps 2 & 3 as necessary] * Usually sample sizes; could be c.v.s

Initial weighting of compositions Two approaches: 1. To represent observation error - from bootstrapping of data (e.g., Stewart & Hamel 2014) 2. Ad hoc: e.g., number of sets or landings sampled Important: aim to capture year-to-year changes in reliability

Adjusting composition weightings From analysis of composition residuals in model output Most common method: McAllister & Ianelli (1998) - included in Stock Synthesis outputs - uses individual residuals My recommendation: Method TA1.8 etc (Francis 2011) - use residuals of mean age or mean length - adapts approach of Pennington & Vølstad (1994) - available in r4ss (SSMethod.TA1.8, SSMethod.Cond.TA1.8)

Sample sizes for compositions Adjusted Data setInitialMcA&ITA1.8 Trawl fishery Commercial longline Artisanal longline Survey Chilean hake example (Francis 2011)

Correlations amongst individual composition residuals McAllister & Ianelli approach overestimates sample sizes because it assumes residuals are uncorrelated Commercial trawl

Main sources of composition correlations Observation error: - intra-haul correlation (Pennington & Vølstad 1994) Process error: - misspecification of selectivities, M, growth

Comparison of observation & total error for hoki age compositions Actual sample sizes 1 Effective sample sizes 1 Data setTypeOtolithsLengthsObs. err.Total err. HOKwcFishery HOKcsFishery HOKcrSurvey Median values over all years Source: Table 1 of Francis (2014)

Research question: should adjustment be multiplicative or additive?

How many iterations? Pragmatic approach: - until there’s no visually significant change in fits & outputs (often, once is enough) My experience: - only large changes in N cause significant changes in the assessment

Significant & insignificant changes

An alternative to iterative reweighting (Francis 2014) Replace the multinomial with a likelihood that - allows for realistic correlations - is self-weighting Possible replacement: logistic-normal

How to deal with abundance data that may be unrepresentative? ‘Unrepresentative’ = wrong trend ≠ imprecise or noisy Common response: down-weight the data Better response: either fit well or discard

Why prioritise abundance data? Key stock assessment questions concern abundance Composition data can tell us - lots about recruitment variation - something about selectivities and growth - little about abundance (usually) Inference from composition data can easily be compromised by misspecification

What is meant by “prioritise”? Aim to achieve a good fit to all abundance data (don’t let other data stop the model fitting abundance data well) Weight abundance data outside the model - no iterative reweighting - no concentrated likelihoods

Right-weighting and down-weighting of composition data Compositions should, ideally, be right-weighted (i.e., consistent with residual sizes, as with TA1.8) Right-weighting is important for valid statistical inference Right-weighting of compositions depends on model structure Apply further down-weighting only if necessary to achieve a satisfactory fit to abundance data (often not needed)

Conflict between two abundance series Canadian 2J3KL cod stock fig. 1 from Schnute & Hilborn 1993 Survey CPUE Bad assumption: both series are correct. Better assumption: one series is correct, but we don’t know which one.

Conflict amongst multiple abundance series 1. Create two or more alternative models using different abundance series 2. Ensure that all abundance series in each model are well fitted Differences between model outputs represent our uncertainty about which abundance series are representative

Conflicts involving compositions Abundance vs compositions Amongst compositions Most likely cause: misspecification

Is restructuring an alternative to data weighting? 1. Profiles good for detecting misspecification causing data conflicts 2. Better to remove the misspecification/conflict, than to down-weight compositions 3. Without data conflict, data weighting is unimportant

Some comments 1. Restructuring to reduce misspecification is good, and profiling is a useful tool 2. Why not restructure and reweight compositions? 3. If not, how to know if we have the right weights for our compositions? 4. The aim of reweighting is right-weighting, not down-weighting

Other comments on misspecification 1. George Box: “All models are wrong [= misspecified], but some are useful” 2. Don’t add model structure that’s not estimable 3. Possibility of over-fitting (e.g., time-varying selectivities) 4. The multinomial likelihood is a misspecification commonly overlooked (especially in simulation studies) p. 424 in Empirical Model-Building and Response Surfaces (1987) by G.E.P. Box and Norman R. Draper

Summary 1: Three data-weighting principles 1. Prioritise abundance data (get good fits) 2. Don’t down-weight abundance data because they may be unrepresentative 3. For compositions: - reweight iteratively, allowing for correlations (the aim is right-weighting, not down-weighting)

Summary 2 Data conflict 1. Conflict amongst abundance data sets: - consider alternative models 2. Conflict involving compositions: - if possible, reduce misspecification (then reweight) Simulation studies - please simulate more realistic compositions