Computer simulation techniques to assess angler survey methodology

(1)

COMPUTER SIMULATION TECHNIQUES

TO ASSESS

ANGLER SURVEY METHODOLOGY

David

L

Wade

Applied Marine Research Laboratory

Old Dominion University

Norfolk, Virginia 23529

Cynthia M. Jones

Old Dominion University

Norfolk, Virginia 23529

Douglas S. Robson

150 Maclaren St PHS

Ottawa, Ontario

Canada K2P OL2

Kenneth H. Pollock

Department of Statistics

North Carolina State University

Raleigh, N.C. 27695-8203

(2)

ACKNOWLEDGEMENTS

The authors extend thanks to The Maryland Department of Natural Resources for the

use of survey data, and special thanks to Dr. John Hoenig, Mr. Bill Check and Mr. Scott

Cone for their insight and help. This study was funded by Virginia Sea Grant through

(3)

INTRODUCTION

Marine recreational angling

Is

a componentoffishing which

Is

difficult to estimate. In 1985, an estimated

13.1 mUllon anglers

fished

for155.2 mUlion days In the nation's oceans, gulfs,bays,andestuaries, and total

marine recreational flshIng expenditures for 1985 were 1.2 billion dollars (USFWS, 1988). In order to effectively manage these recreational fisheries It Is Important to have unbiased and precise estimates of

catchandeffort. Angler Intercept surveys that contact angling parties duringthe prOcess offishing are an effectivemethod ofobtaining estimatesofmarine recreational catchandeffort, and In some fisheries they may be the only practical meansof acquiring these estimates. The absence of marine angling licensing

programs

In

manystatesand concentrationsofcoastalresidents makes marine~eatlonal anglers difficult toidentifyand

contact

whentheyare

not

fishing.

There are two primary methodsofconducting Intercept surveys: the access site

method

and the roving

method. The access method employs clerks who interview angling parties which have completed their trips

and are leaving the fishery at access sites, such

as:

boat ramps, piers, and marina parking lots. Oerks inspect the catchandconductInterviews In order to obtain catch and effort Information. Some fisheries have

diffuse access sites, and angling partiesInthis case may complete fishing trips

at

homes, private docks,

remote parking sites, and otherwiseInaccessible

areas.

As efficient samplingofsuchsites

Is not

possible, the accesssitemethod

Is not

applicable to these

fisheries.

Theroving clerk method isapplicable to these diffuse access fisheries, and

Is

conducted by using clerks who move through the fishery and Intercept

anglingpartiesIntheprocessoffishing. Catch and effort Information for Incomplete trips Is obtained and expanded for complete trip estimates. However, the estimates obtainedfrom aroving clerk survey have significant problema with bias and the derivation of variance components (Robson, 1961), and these problems are relatively mathematlcaJly Intractable.

In order to examine thepropertiesofbias and variance,we must know the aetuaJ total catch and effort

(4)

ofsimulation modeling

was

chosen, and Implemented In a general manner so that data from a variety of fisheries canbemodeled. In this

initial

studythemodel

was

applied to the recreational blue crab(Cailinectes

sapidus) fisheryof the Mid-AtlantIc Chesapeake Bay area.

The diffuse angler access, high valueandlackofrecreational catch and effort statisticsforthis recreational crabfisherymakeItan IdealsubjectInwhichto study the applicationof theroving derk design. This diffuse accessfishery

Is

consideredto comprise a singlestock(Cronin

at

ai,1987)that

Is

distributed In bothrural

and urban areas among a complex system of bays, tributaries and salt marshes. Crabs are captured recreatlonally with a variety of gears. Including: traps, pots, trodlnes, handllnes, and dlpnets. In 1986 comrnerclallandlngs were approximately valuedat 31 mUllan dollars (Cronin atai, 1987), but few studies havebeenperformedto determinethemagnitudeofthe recreational blue crabfishery(Williamsatai,1982).

Fisheriesmanagers have estimatedtherecreational component tobeontheorderof50%, buttheNational

MarIne FIsherIes

ServIce

(NMFS) marine recreational fishing survey has notassessed thefishery except through special requestand funding byMaryland In 1979and 1983. In addition,Virginia has undertaken no surveysofthisfisheryand hasno licensingformarine recreational anglers.

The objectives ofthisstudy were to develop a computer simulation model ofthe roving derk survey design, andapplythemodel to thebluecrabfisheryofChesapeake Bay. The feaslbUIty0: conducting a survey In thefield

was

Investigated by examiningthe bias and sampling distribution of the roving derk design, andthemodel

was

generalized toallowmodificationsofthesurvey designandacceptance of data

from otherfisheries.

MATERIALS AND METHODS

The model

was

developed as a high level programming language, and

was

coded In Pascal. This language Implementation provides a modular and flexible tool similar to commercial statistical analysis systems (e.g. SAS, SPSS). andthePascalprogramming language provides modUlarIty, peer acceptance,

(5)

Theapplication ofthemodel tothebluecrabfisheryoftheChesapeake Bayusedcatch and effortdata from a 1983Maryland recreational angler survey. The data

was

compUed and provided by the Maryland DepartmentofNaturalResources. For this applicationofthe model the Maryland recreational dipnet fishery

was

simulated.

Various commercial grapNcs packages were Interfaced to the model to allow detaDed monitoringof

Its

function,andstatisticaloutput

was

analyzed throughtheuse

of

StatJstIcaJAnalysisSystem (SAS)programs.

DESCRIPTION OF THE MODEL

The model

was

constructed In 2 main modules. The first component is a population module which

stochasticallygenerates an angler populationwiththe same characterIstlcs that are exhibited Inthefield (thus establishing realistic -ground

truth,.

and the second component is an estimation module which simulatestheexecutionofa rovingclerksurvey ontheangler population (thus providing estimatesoftotal catch and total effort). The actual valuesoftotal catch and total effortfrom the population module are compared totheestimates produced bythe estimation module, andthrough Iteration ofthis process the biasand sampling distributionofthe estimates are revealed.

POPULATION MODULE

Eachpopulationproducedbythepopulation moduleconsistsofan array

of

stochastically generated data records. eachrecord representinga singleanglingparty. These parties have the attributesofgeographic location. startingtime.

effort.

andcatch.

(6)

Thestartingtimeforeachparty

Is

generated stochastically, and

Is

based

on

a distribution

eX

starting times observed Inthefield. Aprobabilitydensity function constructedfromfield data

Is

expressed in the form a cumulative density function, andwhen a distribution

eX

uniformly distributed randomvariates Is applied to

the functfon thentheresulting distributioncloselyresemblesthe probabHIty density function observed In the

field.

Effortforananglingparty

Is

theduration (hours)

eX Its

fishingtrip,andfortheMarylandcrab data, effort

couldbefitas a linear model

eX

startingtime.Stochastlclty Is added tothesimulationby fittinglinear models

eX

mean

andvarlance

eX

efforttothenumber

eX

hours Inthefishingday minus starting time:

(where{Jand{J'arethe OLSregression coefficients,Ts

Is

thestartingtime [hrsJ foranangling party, and

TFIs the end

eX

thefishing day). Daylength minusstartingtime Isused Inthesefunctions. so that effort valueswiIdecrease

to

zero

atthe end

eX

each day. Foreachanglingparty,thesevalues

eX

mean effort and variance

eX

effort describenormal distributions, andfromthesedistributions a value

eX

effortfor theparty Ischosen randomly.

Forthe Marytand data, catchcould befit as a linear model

eX

effort , and stochastlclty Is added to the modelbyfitting models

eX

mean

and varianceofcatch toeffort.

JI(catch) ,. (effort){J

The value ofcatchfor each angling partyIsdetermined bytheapplicationofthe mean and variance for catch to anormallydistributed randomnumber generator. In theMaryland crabfishery, catch could befit

(7)

successfulpartiesmayfishforlonger periodsthanunsuccessful parties. To model relationshipsofcatch rate

to effort we add a curMlnear component tothelinear model for meanofcatch.

p(eatch) .. (effort)~ + (effoo2)~'

The relationshipofcatch rateto effort

Is

specified by assigningthe parameter B' to a negative or positive

value.

EstImationModule

When the simulation model hasgenerated a populationofanglerswithknown values oftotal catch and total effort, a roving creel survey Is conducted bytheestimation moduleto estimate total catchand effort for that population. Total effort

Is

estimated asthe productoftheduration ofa clerk's transect (hrs) and total parties counted:

n

ETPH, .. E

j-1 <x.j) •

T

_{s ' m}

whereETPH,is the estimate of totalpartyhoursfordayI,Xlj

Is

thevalueofan angling party j on day i_(Xlj=1 if the partywas encountered bytheclerk, else

XII ..0), Ts

Is

the duration of a clerk'stransect (hrs), and m

Is

the numberof

transects

In a day.

Total catch

Is

expanded

from

Interview obtained Information

on

partycatch andtrip length.

Individual survey designs are specified,andthey are Implemented as aseriesof linear equations In the cartesian coordinatesofgeographic location

on

a linear survey

route

andtimeofday.

For this seriesofsimulations

two

primary

survey

designswerechosen, a simple roving creel designand

a more elaborate designwithclerk scheduling.Thescheduling designwaschosen In an attempt to reduce the bias inherent Inthemore simple surveymethod.

Thesimplerovingclerk design,typeone, (figure 1) assigns a survey clerk a circular, linear route through

(8)

foreach survey day,and theclerkIsassigneda starting timeandcompletion time for the day. Theclerk's scheduled movement through thefisherycan be modeled by a line segment:

location • Travel Speect( Time of Day)

+

Starting location

Slopeof this line

Is

equaltothe clerk's scheduledspeed, and

start

and end coordinates are equal to the

start

and end ofthe survey day. Each stationary angling partycan also be simulated by a line segment

Slope of this linebeingequal to zero,andstartingandending coordinates beingequal to startandend of

thefishingtrip. An Intersection of a clerk linesegmentandan anglingpartylinesegmentsimulates a field Interceptofa

party

bytheclerk. During a

survey,

theclerk proceeds

at

his or her scheduledspeedthrough

the fishery untI an angling party

Is

Intercepted. and

at

this pointthe clerk must stop movement through space,

conduct

an Int8IView of somespecifiedlength,andcontinue moving at the originalspeed.Note that this causes the clerk to departfromtheoriginalschedule(theInteMew caused a delay). This entire process

can be simulatedbythe useoflinesegmentsfor each component of a clerk's travelandeach anglingparty. The secondtypeofsurvey design

Is

amodifiedmethod that uses aseriesofcheck points to ensure that

the clerk remains on schedule (figure 2).The checkpoints are distributed along a route, are chosen time

intervals (e.g. fNery 2 hours), and can be positioned as -ume-Iocated· geographic points (e.g. piers, lighthouses, boatramps).Theclerkconducts

Intervtews

as Inthesimple example,andwith each Interview.

heor shewIIbecome Increasinglyoff schedule. HowfNer,

at

somepointtheclerk's pathwAllntercept a linear schedule segment (with a slope representing some maximumspeed of travel) that if followed will

retumtheclerk tothe

Is

moving attheIncreasedspeed, he orshecountsanglingpartieswithout stopping to Int8fVlewthem.

RESULTS AND DISCUSSION

(9)

relationship between Individualcatch rate and effort then this will result In a bias In total catch estimates.

When Interview time

Is

greaterthan zero, the probabUIty of Intercepting a party becomes more complex.

There

Is

a shadowing effect on Interception probabilities between angling parties alongtheroute,andthis

Is

Dlustrated In figure 3. Dueto thetime a clerk Is stationary whOe Interviewing thefirstpartyon the route,

the probability of Interceptingthe secondparty

Is

reduced by some amount up to the lengthofthelnteMew,

andthis shadowing

Is

additive astheclerkconductsInterviews throughouttheday. This resultsIn bias In estimates of total effort.

In the simple roving clerk survey design wherethe clerk does

not

attempt to remainon schedule, this

shadowing effect can cause considerable bias In effort estimation.Asshown bythesimulationresults(figure 4) In this exampleof10,000 Independently simulated survey days,thebias rangesfromapproximately 25 percent at Interview lengths of 5 minutes to 36 percentatInterview la.-.gths of i 5 minutes,and the estimator

always under estimatesthe actual value oftotal effort forthe population. The bias In effort estimation

Is

variable, and it Is related to the length of Interviews conducted.

In an attempt to avoid this bias, survey clerks can be Instructed to retum to scheduled routes when

Interviews cause them to be off schedule, and this Is simulated by the modified survey design with check

points. Inthemodified designthe bias Ineffort estimation

was

reduced to approxknately 1 to 2 percent (figure 5),andunder normal management conditions this might be an acceptable levelofbias. However the bias Is stDl variable,and always under estimatestheactualvalueoftotaleffort.

It

Is

important to understand that biasIneffort resultswere under thebestoffield conditions. Angling

partieswere distributed In uniform distributions ontheroute,andInthe

reaJ

wortd it

Is

possible that anglers

are distributed In contagious distributions. The effect of clumping may Increasetheshadowing affect on total effort estimationand

Is

thesubjectofcurrentstudy.

The sampling distributions ofthe estimators are Important forthe construction ofconfidence Intervals

about estimates,andthey were determined (figure 6)bya simulation runof10,000 independently simulated

survey days.The populationvalues of total effort for each survey day were normallydistributed, butthe

estimated values for each survey design were non-normaIIy distributed (P<.01).Theconsequencesofthese results are that standard T tables can

not

be used with this data, and that T distributions should be

(10)

SUMMARY

ThesinUattonmodelwasdetermined tobea usefulapproach to examiningthe rovingclerk designand the

statistical

propertiesd itsestimators. Wefound theroving clerk estimatord totalfishing effortto be

both biasedandvariableIn its bias.Thisbiaswas underthe bestd circumstances. whereanglingparties were uniformly distributed along a route, and could bereducedby Instructing clerk's to

return

toscheduled

surveyroutesatspecifiedcheckpoints. These resultswerespecific totheMarylandrecreational blue crab

(caJllnect8S sapidus)dlpnetfishery,butthemodel

Is

generalandcanbeapplied todatafromotherfisheries. However, thebiasdemonstrated Is InherentIn.the roving clerk design, and

Is

caused bythe Interception

(11)

Uterature Cited

Cronin, LE. (ed.) 1988. Report oftheChesapeake Bay blue crab management workshop.

Chesapeake Bay

Commission.

MONR, Potomac River FIsheries Commission, Virginia Marine Resources Commission, 68 p.

Robson, O.S., 1961. On the statisticaltheoryofa roving creel censusoffisherman. Biometrics17, pp

415-437.

United States FIsh and WUdllfe Service, 1988. 1985

National

surveyoffishing, hunting, and wildlife associated recreation. 167 p.

(12)

figure

1

figure 2

figure3

figure 4

FIgure5

Figure 6

Ustof Figure.

A diagram representing a simple roving creel survey design Une segment A represents a

stationary angling party,andthe lengthofsegment A is the effort (trip length) of theparty. Una segment B represents the original scheduled path of movement of a survey clerk

throughthefishery(slopeisequaltospeedoftravel). Une segmentC represents the actual

path of the clerk as he/she stops moving to conduct an Interview (C

1) and continues

movingwhentheinterview is completed (C_2).

Adiagramslmlarto figure

2.

bt4in thiscasetheclerkIncreases speedd travel (slope) at Intervals.

so

that he/she will remain

on

schedule (wAI

pass

through regularty scheduled checkpoint, represented hereas open circles).

A diagram representingthereduction of the probabUity (5) of the clerk intercepting a party (P1)'The reduction is caused bythe Interviewing of a second party (P2) ~nterview is

coilduetedfrom 1₀to1_1),

A chart ofthe biasin total effortestimatfonfor the simple roving creel survey design

as

a

function ofinterviewlength. Results are from10,000Independent survey days.

A chart of the biasin total effort estimation forthemodified roving creel survey design(with

checkpoint scheduling)

as

a function of Interview length. Results are from 10,000

Independent survey days.

The sampling distributions for actualandestimated valuesoftotal effort. Results are from