Testing heterogeneous anchoring and shift effect in double-bounded models: The case of recreational fishing in Tasmania

Size: px
Start display at page:

Download "Testing heterogeneous anchoring and shift effect in double-bounded models: The case of recreational fishing in Tasmania"

Transcription

1 Testng heterogeneous anchorng and shft effect n double-bounded models: The case of recreatonal fshng n Tasmana Sarah Jennngs School of Economcs and Fnance Unversty of Tasmana Steven Rust Australan Bureau of Statstcs Satosh Yamazak School of Economcs and Fnance Unversty of Tasmana Jeremy Lyle Tasmanan Aquaculture and Fsheres Insttute Unversty of Tasmana 25 January 2010 Australan Agrcultural and Resource Economcs Socety Natonal Conference 2010, January, 2010 n Adelade. 1

2 Abstract Ths paper explores the extent and nature of anchorng and shft effects n a doublebounded contngent valuaton of recreatonal fshng n Tasmana s nshore saltwater fshery. In partcular we model the stuaton where respondents, when answerng the second valuaton queston, evaluate the bd amount partly wth reference to the sze of the frst bd amount. The estmates of the coeffcents and mean WTP for a day of fshng are compared across dfferent contngent valuaton models, ncludng a sngle-bounded model, a conventonal double-bounded model and models that control anchorng and exogenous shft effects n both homogeneous and heterogeneous forms. Overall we fnd consstent evdence of anchorng, but mxed evdence of a shft effect. Results show that both males and females anchor n the same way, but that respondents who have a manstream vew of what recreatonal fshng represents anchor more strongly than those whose vew of fshng s not manstream. The estmated mean WTP for a day s recreatonal fshng s consstently hgher n all models whch account for bas n responses than n ether the sngle-bounded or double-bounded models. We ndcate the possblty that anchorng behavour may be more complex than s captured n our models and suggest that ths needs to be addressed f the results of contngent valuatons are to relably nform resource allocaton decsons and recreatonal fshng management. Keywords: Contngent valuaton, anchorng bas, shft effect, heterogenety, recreatonal fshng JEL codes: C35, Q26 2

3 1 Introducton Increasng pressure on stocks n wld fsheres heghtens the need for relable estmates of the value of the fsh resource n dfferent uses. Developng a better understandng of how recreatonal fshers value the resource s needed to underpn resource allocaton decsons and the desgn of management polces that govern access to, and the qualty of, recreatonal fshng. The non-market nature of recreatonal fshng mples a contnued need for research to be based on stated preference methods, ncludng contngent valuaton. Whle t s generally accepted that the dchotomous choce contngent valuaton format s preferred to the open ended queston format (Arrow et al. 1993), debate contnues about whether gans n effcency assocated wth a multple bd format (Hanneman 1991) offset the bas that can result from varous forms of startng-pont bas that may arse (Cameron and Quggn 1994). For example, respondents may use nformaton provded as part of the valuaton exercse to re-evaluate ther atttude towards the good or servce beng valued. They may also modfy ther wllngness to pay (WTP) so as to gve responses that they beleve are n some sense socally desrable. A number of studes have developed double-bounded models whch ncorporate response bas n the form of anchorng and an exogenous shft effect (e.g., Herrges and Shogren 1996, Whtehead 2002, Chen et al and Flachare and Hollard 2006). These studes have consstently found that when these effects are not accounted for, the estmates of the margnal effects and mean WTP are both nconsstent. Generally, however, ths work has been based on the assumpton that response bas occurs homogeneously across respondents. The possblty that ndvduals may dsplay dfferences n the degree to whch ther responses reflect these behavours s acknowledged, but less well explored (exceptons are, for example, Aprahaman et al and Flachare and Hollard 2008). In ths paper we begn explorng the extent and nature of anchorng and shft effect n a double-bounded contngent valuaton of recreatonal fshng n Tasmana. In partcular 3

4 we model the stuaton where respondents, when answerng the valuaton queston, evaluate the second bd amount partly wth reference to the sze of the frst bd amount. We then test two separate hypotheses regardng heterogenety n survey respondents anchorng behavour by groupng respondents accordng to gender, and to ther socal representaton of recreatonal fshng. The estmates of the coeffcents and mean WTP for a day of fshng are compared across dfferent contngent valuaton models, ncludng a sngle-bounded model, a conventonal double-bounded model and models that control anchorng and exogenous shft effects n both homogeneous and heterogeneous forms. The rest of ths paper proceeds as follows. Secton 2 revews a seres of contngent valuaton models ncludng those whch account for anchorng and a shft effect as both homogeneous and heterogeneous phenomenon among respondents. The estmaton method s dscussed n Secton 3 and the source and characterstcs of our data descrbed n Secton 4. Secton 5 explans how the sample has been parttoned n estmatons nvolvng heterogenety and n Secton 6 the results of estmatons are presented and dscussed. Concludng remarks, whch emphasse the future drecton ths research wll take, are n Secton 7. 2 The econometrc models Sngle-bounded model Assume that respondent s true wllngness to pay (WTP) s specfed as: WTP x u, * ' u N 2 (0, ) (1) where x s a k 1 vector of ndependent varables, s a k 1 vector of correspondng coeffcents and u 2 s a normally dstrbuted error term wth mean 0 and varance. In the sngle-bounded dchotomous choce model, the valuaton queston s asked only once and respondent answers yes f the bd b 1 s less than or equal to hs/her WTP, but 4

5 answers no f the bd amount s greater. WTP j s defned as the wllngness to pay used to answer the valuaton queston j, the respondent s answer s defned as y1 1 f WTP b y 0 f WTP b (2) 1 1 and where y s an ndcator varable such that y 1 f the respondent answers yes, and j j y j 0 f the respondent answers no. Conventonal double-bounded model The sngle-bounded model can be extended by ncorporatng addtonal nformaton from subsequent valuaton questons. For the double-bounded model, the frst queston s followed wth a second dchotomous choce valuaton queston. As n the sngle-bounded model, the respondent answers yes to the follow-up queston, f the bd amount less than or equal to hs/her WTP, and answers no otherwse. The respondent s answer to the second valuaton queston s defned as: b 2 s y2 1 f WTP2 b2 and y2 0 f WTP2 b2 (3) One advantage of the conventonal double-bounded model over the sngle-bounded model s that, by ncorporatng addtonal nformaton from the follow-up queston, the estmates from the double-bounded model are statstcally more effcent (Hanemann et al. 1991). Anchorng and shft effect Prevous studes have shown that estmates from the double-bounded model are unbased only f respondents answer the frst and second valuaton questons based on the same * WTP,.e., WTP WTP1 WTP2 (Cameron and Quggn 1994, Herrges and Schogren 1996, Albern et al. 1997, DeShazo 2002 and Flachare and Hollard 2006). It s well 5

6 establshed, however, that responses to double-bounded contngent valuaton questons may be subject to varous forms of response bas, suggestng that ths may not always be the case. Herrges and Shogren (1996) propose a framework that explctly models and estmates the effect of anchorng bas wthn the double-bounded model, whch arses when respondents update ther WTP when presented wth a second bd amount. The Herrges and Shogren model assumes that respondents WTP when answerng the second valuaton queston s a weghted average of ther true WTP and the frst bd amount, such that: WTP2 1 WTP1 b 1 * where WTP WTP (4) 1 where [0,1] s the weghtng parameter that measures the strength of the degree of anchorng. When 1, the respondent totally replaces the pror wllngness to pay wth the ntal bd amount, whereas there s no anchorng effect at all when 0. Albern et al. (1997) propose an alternatve model whch assumes that respondents WTP when answerng the second valuaton queston s exogenously shfted from the true WTP, such that * WTP WTP where WTP WTP (5) where s the shft parameter. The economc ntuton of the shft parameter s as follows. When the shft parameter s negatve ( 0 ), respondents systematcally devalue ther WTP after the frst valuaton queston, whch s referred to as the ncentve ncompatblty effect. By contrast, a postve shft parameter ( 0 ) represents a form of yea-sayng or acquescence behavour n whch respondents overestmate ther WTP for the second valuaton queston as a result of a tendency for respondents to agree regardless of the bd level. Legget et al. (2003) suggest that ths type of bas may be more prevalent n n-person surveys where respondents may be more nclned to respond n ways that they beleve wll please the ntervewer. 6

7 Whtehead s (2002) model allows for the possblty that response bas n the double- model mght be of a form that nvolves both anchorng and a shft effect such bounded that: 2 1 * WTP 1 WTP b where WTP WTP (6) 1 1 Heterogenety n anchorng Recent studes suggest that anchorng may be a heterogeneous process n that ndvdual respondents may dffer n ther anchorng behavour. Usng Monte Carlo smulaton, Aprahaman et al. (2008) show that f anchorng s mstakenly specfed n a homogenous form (as n equatons (4) to (6)), when true anchorng behavour occurs heterogeneously across respondents, the estmates are based and the shft effect spurously appears. Lmted emprcal evdence of heterogenety n double-bounded contngent valuaton studes confrms that falure to correctly account for ths n a double-bounded contngent valuaton model may result n a based estmate of WTP. A common approach has been to explore dfferences n anchorng behavour across dfferent groups of ndvduals, where group membershp s based on some observable characterstc (e.g. gender or ncome) or latent characterstc (e.g. atttude or belef) of the respondent. For example, allowng for only two groups, Flachare et al. (2007) and Flachare and Hollard (2008) specfy the WTP to the second valuaton queston as: 1 (1 ) (1 ) WTP I I WTP I I b * where WTP WTP (7) 1 7

8 where I s a dummy varable takng value 1 f the respondent belongs to one group, and 0 otherwse, and 1 and 2 are the correspondng parameters. Employng the theory of socal representaton to separate ther sample nto two groups, Flachare et al. (2007) and Flachare and Hollard (2008) both fnd evdence of anchorng among the group whose representaton of the good beng valued s consstent wth a manstream or conformst vew. 1 3 Estmaton The sngle-bounded model s estmated by the method developed by Cameron and James (1987). The model s estmated by maxmum lkelhood wth the log-lkelhood functon: n 1 log L y log Pr( yes) 1 y log Pr( no). (8) 1 1 The probabltes can be calculated as: ' 1 1 Pr yes Pr( y 1) 1 b x and ' 1 1 Pr no Pr( y 0) b x. (9) where s the standard normal densty functon. For the models wth a follow-up queston, there are four possble combnatons of answers to the valuaton queston,.e., (yes, yes), (yes, no), (no, yes) and (no, no). If the respondent answers yes to the frst valuaton queston, the second bd amount becomes hgher ( b H 2 b ), whle f s/he/she answers no the second bd amount becomes lower L ( b ). Thus, the probabltes that respondent answers (yes, yes), (yes, no), (no, yes) 2 b 1 1 and (no, no) to the frst and second valuaton questons are: 8

9 H yes yes y1 y2 WTP2 b2 Pr, Pr( 1, 1) Pr H yes no y1 y2 b2 WTP2 b1 Pr, Pr( 1, 0) Pr L no yes y1 y2 b1 WTP2 b Pr, Pr( 0, 1) Pr 2 (10) L no no y1 y2 b2 WTP2 Pr, Pr( 0, 0) Pr and the log-lkelhood functon s defned as: n logl y y log Pr yes, yes y (1 y ) log Pr yes, no (1 y1 ) y2 log Pr no, yes (1 y1 )(1 y2) log Pr no, no. (11) For the model ncorporatng heterogeneous anchorng and shft effects, the probabltes are computed by 2 1 ' WTP b x Pr 1 Pr Pr Pr b b b b ' ' b2 WTP2 b1 x b1 x ' b2 b1 ' b WTP b b x x 1 b b 2 1 ' b WTP x (12) where I 1 (1 I ). The probabltes for the conventonal dchotomous choce 2 double-bounded model and the model wth homogeneous anchorng and a shft effect are computed by mposng restrctons on (12). For the model wth homogenous anchorng and shft effects, the restrctons are for all and the restrctons for the conventonal double-bounded model are 0 and 0. 9

10 4 Survey desgn and data Our data came from a seres of questons asked of Tasmanan recreatonal fshers n a survey conducted as a follow-up to the 2007/08 Survey of Recreatonal Fshng n Tasmana (Lyle et al. 2009). 2 At the completon of the recreatonal survey (January 2009), respondents were asked whether they would be wllng to partcpate n a follow-up economc survey. A sample of 604 households was selected for the economc survey whch was admnstered by telephone over a seven week perod n June and August 2009 by a team of professonal ntervewers, all of whom had prevously been nvolved n the fshng survey. Intervewers were brefed on the purpose and desgn of the survey, as well as beng gven a broad overvew of the contngent valuaton method. Complete responses were receved for 480 fshers, representng an overall response rate of 79.4 per cent. Contact could not be establshed wth an actve fsher over the age of 18 for 59 households (9.7 per cent) and a further 59 households (9.7 per cent) ndcated that no members had fshed n the twelve month perod between July 2008 and June These latter households were consdered out of scope and excluded from the survey. Sx of the elgble fshers who were surveyed, or less than 1 per cent of our orgnal sample, dd not provde complete responses. Our analyss focuses only on the 314 fshers whose last reported day s fshng was n the nshore saltwater fshery (ISF). 3 Eght of these respondents were excluded due to nconsstences n the daly cost nformaton they provded. On the bass of ther answer to the valuaton queston a further 13 fshers were dentfed as protestors and ther responses were dropped from the fnal data set (Dzegelewska and Mendelsohn, 2007), whch therefore conssted of observatons on 293 fshers. The survey nstrument conssted of sx parts and generally took mnutes to complete. The survey nstrument was pre-tested on ten fshers. Pre-testng resulted n mnor changes to the survey and was used to determne the approprate range of bd 10

11 amounts used n the valuaton queston. Part four of the survey asked respondents a seres of questons related to ther most recent days fshng, ncludng detals of the locaton and nature of the trp, the number and speces of fsh caught, ther motvatons for fshng and mportantly, the level and types of avodable costs ncurred on the day only. It also ncluded a set of questons desgned to establsh fsher s economc valuaton of ther most recent days fshng. Other parts of the survey asked respondents for nformaton about ther fshng actvty over the past twelve month perod, demographc characterstcs and what fshng represented to them. Descrptve statstcs for varables derved from the 293 survey responses are reported n Table 1. About 67 percent of respondents ndcated that they were targetng a sngle speces on ther most recent fshng day, whereas about 23 percent and nearly 10 per cent were ether targetng multple speces or nomnated no specfc target speces, respectvely. In total, fshers n the ISF reported havng caught about 30 dfferent speces of fsh on the last day s fshng, wth a total catch of about fsh. Flathead (fam. Platycephaldae) comprsed almost 64 percent of the total personal catch for the sample. About 27 and 38 percent of respondents sad that ether enjoyng the outdoors or spendng tme wth famly and frends was ther man motvaton on that day. Only 19 percent of respondents sad that the man reason was to catch fsh. [Table 1 about here] We use a double-bounded valuaton queston format that s smlar to Wheeler and Damana (2001), n whch the payment vehcle s the amount respondents report havng personally spent on avodable, consumable tems for ther most recent days fshng. Ths format was chosen for ts smplcty and because t avods the need to ntroduce a lcense fee or tax, both of whch may elct protest bds from respondents. 4 The frst round valuaton queston conssted of; 11

12 Bearng n mnd that you have many calls on your ncome, f t had cost you an extra $XX on these [consumable] tems for ths day s fshng only, would you stll have gone fshng on that day? Ths was followed wth a second queston; and would you have stll gone fshng on that day f t had cost you an addtonal $YY? where $YY wll be ether double or half $XX dependng on whether the respondent answers yes or no to the frst queston. The ntal bd amounts $XX were set at $10, $20, $30, $40, $50 and $60 and were randomsed across respondents accordng to a unform probablty dstrbuton. Table 2 shows the average bd values for the frst and second questons and the jont frequences of responses (yes, yes), (yes, no), (no, yes) and (no, no) to the frst and second valuaton questons. [Table 2 about here] Fgure 1 shows the dstrbuton of frst and second queston responses for each ntal bd value. The relatvely small number of (no, yes) and (no, no) responses for respondents recevng an ntal bd of $10 suggests that the lower end of the bd range was well chosen. [Fgure 1 about here] 5 Heterogeneous anchorng In ths paper we consder two separate hypotheses regardng the source of heterogenety n respondents anchorng behavour, these beng the respondents gender and ther socal representaton of recreatonal fshng. Prevous studes have suggested that dfferences n 12

13 anchorng behavour may reflect the demographc and other soco-economc characterstcs of respondents (Aprahaman et al. 2007). The second source of heterogenety examned follows Flachare et al. (2007) and Flachare and Hollard (2008), who hypothesse that a group of people who have unque belefs (mnorty or nonconformng) about the subject of the valuaton exercse are less lkely to be nfluenced by new nformaton. Thus, fshers who have unque belefs about recreatonal fshng actvty are less prone to anchor to the frst bd amount than are people who hold common belefs. Flachare et al. (2007) and Flachare and Hollard (2008) employ the theory of socal representaton to separate ndvduals nto two groups, manstream (or conformst) and mnorty (or non-conformst). In ths paper we follow the method proposed by Flachare and Hollard (2008) to dentfy the socal representaton assocated wth Tasmanan recreatonal fshng held by respondents n our sample. Identfcaton of ths representaton and the parttonng of the sample nto manstream and mnorty groups were based on respondents answer to an open-ended queston that preceded the valuaton queston n our survey. In ths queston we asked: What s the frst word that comes to mnd when I menton recreatonal fshng?...what s the second word?...what s the thrd word? The words dentfed are taken to descrbe the representaton each respondent mantans regardng recreatonal fshng. Ths method, known as word assocaton, s commonly appled by psychologsts n ther nvestgaton of socal representaton (see for example Farr 1993). Not surprsngly, gven the open-ended nature of the queston, a large number of words were gven by respondents (.e., about 360 words were obtaned). Words were then sorted nto clusters on the bass of ther sharng smlar meanng. For example, the category Envronment ncludes words such as beach, coast and envronment. A total of eght categores were dentfed n ths way: 13

14 Envronment, Holday, Famly & Frends, Feelng, Fsh & Fshng object, Food & Drnk, Restrctons, Weather Each dentfed word was then replaced wth the correspondng category to obtan an ordered lst of categores for each respondent. In developng ths orderng, any double or trple ctatons from the same category are handled by suppressng the lower rankng ctatons. For example, f a respondent had cted words whch were from the categores {Envronment, Holday, Envronment} then they would have ther representaton recorded as {Envronment, Holday}. The core of the socal representaton assocated wth Tasmanan recreatonal fshng was dentfed by calculatng the ctaton rate of each category and rankng them accordngly. Table 3 shows that Feelng s the most cted category wth 65.2 per cent of respondents lstng a word categorsed n ths way. Any respondent who cted at least one word that belonged to one of the three hghest ranked categores (Feelng; Fsh & Fshng object; Famly & Frends) was taken to hold a manstream or conformst representaton of recreatonal fshng. [Table 3 about here] 6 Results Sngle-bounded and conventonal double-bounded models Table 4 presents the estmates of the coeffcents and the mean WTP for a day s recreatonal fshng for the sngle-bounded model (Model I) and the conventonal doublebounded model (Model II). The estmates show that men have a greater WTP for the day of fshng than women. The WTP of respondents who fshed from a boat s also hgher than that of respondents who fshed from ether the shore or a jetty. Whle the number of people n the fshng party s postvely related wth the WTP, the presence of 14

15 respondents chldren decreases the value of the fshng day. WTP s also postvely related to respondents ncome. Interestngly, Table 4 shows that all catch varables are statstcally nsgnfcant and thus addtonal fsh caught would not ncrease the WTP for a days fshng. Ths result s dfferent from a number of other valuaton studes that found a postve relatonshp between the number of fsh caught and the respondents WTP (Johnston et al. 2006). The potental reason for ths s that the number of fsh caught n the day s fshng was relatvely large (e.g., 2483 flathead were caught n total) and, as a result, the margnal WTP for addtonal catch s nsgnfcant. In addton, our data suggests that the man motvaton for gong fshng n the Tasmanan ISF s for reasons other than catchng fsh for many respondents. In fact, 66 percent of the respondents ndcated that the man motvaton for gong fshng was to ether enjoy the outdoors or to spend tme wth frends/famly. In terms of the targetng preference, the estmates from the conventonal double-bounded model show that fshers who were targetng flathead and fshers who dd not target any speces have a sgnfcantly lower WTP than fshers who were targetng multple speces. The estmates of the mean WTP are smlar across the sngle-bounded and conventonal double-bounded models. Takng nto account that the average total expendture on consumable goods on the last day of fshng (Cost) s A$42.79 (Table 1), the total mean WTP (.e. Cost WTP ) for a day of fshng n the Tasmanan ISF s estmated as A$ n the sngle-bounded model and A$108.4 n the double-bounded model. It s, however, mportant to note that snce the valuaton survey was orgnally desgned to estmate the double-bounded model, the lower and especally hgher range of the dstrbuton of bd amounts s truncated. Estmates from the sngle-bounded model are lkely, therefore, to be both neffcent and nconsstent. Further, Table 4 confrms that there are effcency gans n the estmates of the coeffcents from usng the doublebounded model. [Table 4 about here] 15

16 Homogenous anchorng and shft effects Table 5 shows the estmates of the coeffcents and mean WTP for a day s fshng from the model controllng anchorng (Model III) and the model controllng both anchorng and shft effects (Model IV). The homogeneous anchorng parameter ( ) s sgnfcant for both models and the shft parameter ( ) s also sgnfcantly postve n Model IV. The estmates, ˆ and ˆ for the two models, suggest that respondents do re-evaluate ther WTP n lght of the frst bd when answerng the second valuaton queston. The postve shft parameter ˆ 3.79 confrms yea-sayng behavour. Ths mples that respondents overvalued ther WTP when they answered the follow-up valuaton queston. Gven the statstcally sgnfcant anchorng and shft parameters, the estmated coeffcents and mean WTP from the conventonal double-bounded model are lkely to be based. The mean WTP estmated by Models III and IV s consderably hgher than that estmated by the sngle-bounded and conventonal double-bounded models. The total mean WTP s estmated as A$ n Model III and A$ n Model IV. Further, whle the sgns of the estmated coeffcents from these models are smlar to those from Models I and II, some varables become statstcally nsgnfcant after controllng the anchorng and shft effects. Ths emprcal result s consstent wth the results llustrated by Herrges and Shogren (1996). The effcency gans antcpated by ncorporatng the nformaton from a follow-up queston would be dmnshed or totally lost when anchorng effects are accounted for n the conventonal double-bounded model. [Table 5 about here] Heterogenety n anchorng Tables 6 and 7 present the estmaton results of the models ncorporatng heterogeneous anchorng behavour. To control the respondents heterogenety, the sample s grouped 16

17 nto men and women n Models V and VI (Table 6) and nto conformsts and nonconformsts n Models VII and VIII (Table 7). The results of Model V n Table 6 show that both anchorng parameters 1 and 2 are statstcally sgnfcant at the 5% level and the estmated values are smlar to each other. Ths confrms that men and women are both nfluenced by the frst bd amount but the degree to whch they anchor s the same between the two groups. By contrast to the results n Model VI, the postve shft effect s lost when the heterogeneous anchorng and exogenous shft effect are both controlled n Model VI. Where the sample s grouped nto conformsts and non-conformsts, we confrm heterogenety n anchorng effects. The estmaton result of Model VII n Table 7 shows that the anchorng effect s nsgnfcant for non-conformst, whereas the estmate of the anchorng parameter for conformsts s smlar to those obtaned n Models III, IV, V and VI (Tables 5 and 6). Whle the anchorng effect s sgnfcant for non-conformsts at the 10% level n Model VIII, agan the postve shft effect found n Model VI s no longer evdent. The estmated mean WTP from Models IV to VIII, whch all control for heterogeneous anchorng, are smlar to each other. They are also of a smlar magntude to the estmates obtaned from Models III and IV, whch treat anchorng as a homogeneous phenomenon. Ths s expected because our results do not suggest strong dfferences n anchorng behavour, between males and females or between fshers whose representaton of recreatonal fshng s consstent wth a manstream belef and those who hold a mnorty belef. [Table 6 about here] [Table 7 about here] 17

18 7 Concludng remarks In ths paper we report estmates of a seres of models of the WTP of fshers for a day s recreatonal fshng n the Tasmanan nshore saltwater fshery. These nclude models that utlse the nformaton contaned n the double-bounded format but whch adjust for anchorng and a shft effect. Overall we fnd consstent evdence of anchorng, but mxed evdence of a shft effect. The estmated mean WTP for a days recreatonal fshng s consstently hgher n all models whch account for bas n responses than n ether the sngle-bounded or double-bounded models. We tackle the possblty of heterogeneous anchorng by comparng the anchorng behavour of dstnct groups of respondents. Results show that both males and females anchor n the same way, but respondents who have a manstream vew of what recreatonal fshng represents anchor more strongly than those whose vew of fshng s not manstream. As s generally the case n the lterature, our treatment of anchorng and shft effects has been lmted to the case where respondents answer to the second queston s made by comparng the second bd amount ( b 2 ) wth a measure of WTP that reflects both the * respondents true WTP ( WTP ) and the value of the frst bd amount ( b ). Lechner et al. (2003), however, pont out that responses to the frst valuaton queston may be anchored * to the frst bd value, whch would mean that WTP 1 s a weghted average of WTP and b 1. Moreover, the partcular form of our payment vehcle, namely an ncrease n the daly cost of consumables, may ntroduce a further opportunty for anchorng to occur. We conjecture that daly cost may be a partcularly strong anchor, as ntervewers are often nstructed to ensure that respondents are remnded of ths amount pror to beng presented wth the frst bd amount. We antcpate that falng to allow for these forms of anchorng n both homogeneous and heterogeneous specfcatons of the double-bounded model wll result n based coeffcents and mean WTP estmates. 1 18

19 Endnote 1. Whle Flachare and Hollard (2008) dd not explctly test whether anchorng by the mnorty group s statstcally sgnfcant, Flachare et al. (2007) fnd that the mnorty group does not anchor ther WTP to the ntal bd amount. 2. The 2007/08 Survey of Recreatonal Fshng n Tasmana (Lyle et al. 2009) was a phone-dary survey whch recorded the fshng actvtes of partcpant fshers between December 2007 and November 2008, The selecton of the sample for the phone-dary survey was based on a randomly chosen set of Tasmanan lsted telephone numbers, and employed some spatal stratfcaton usng the Australan Bureau of Statstcs (ABS) Statstcal Dvsons. 3. Respondents were allocated to one of 8 fsheres on the bass of nformaton provded n relaton to target speces (f defned), catch speces composton (f not nl), fshng locaton (regon and water body type) and fshng method used on the most recent days fshng. About 65 percent of most recent days reported occurred n the ISF. Major fsh speces n the ISF are flathead, Australan salmon, squd, black bream and cod. The freshwater fshery was the next largest fshery n our sample wth nearly 20 percent of respondents ndcatng that ther most recent days fshng was n ths fshery. 4. Remanng protest bds were dentfed by ncludng a not wllng to answer opton for the valuaton questons, and askng respondents who answered (no,no) to explan ths response. These measures are consstent wth those recommended by the NOAA Panel on Contngent Valuaton (Arrow et al. 1993). 19

20 Acknowledgement The authors would lke to acknowledge the Tasmanan Department of Prmary Industres, Parks, Water and Envronment and Water for ts support of ths study and to the team of survey ntervewers. 20

21 References Albern, A., Kannnen, B. and Carson, R.T., Modelng response to ncentve effects n dchotomous choce contngent valuaton data. Land Economcs, 73: Aprahaman, F., Chanel, O. and Luchn, S., Modellng startng pont-bas as unobserved heterogenety n contngent valuaton surveys: an applcaton to ar polluton. Amercan Journal of Agrcultural Economcs. 89: Aprahaman, F., Chanel, O. and Luchn, S., Heterogenous anchorng and the shft effect n teratve valuaton questons, Resource and Energy Economcs, 30: Arrow, K., Solow, R., Portney, P.R., Leamer, E.E., Radner, R. and Schuman, H., Report of the NOAA panel on contngent valuaton. Federal Regster, 58: Cameron, T.A. and James, M.A., Effcent methods for closed-ended contngent valuaton surveys. The Revew of Economcs and Statstcs, 69: Cameron, T.A. and Quggn, J., Estmaton usng contngent valuaton data from a Dchotomous choce wth follow-up questonnare. Journal of Envronmental Economcs and Management, 27: Chen, Y.L., Huand, C.J. and Shaw, D., A general model of startng pont bas n double-bounded dchotomous contngent valuaton surveys. Journal of Envronmental Economcs and Management, DeShazo, J.R., Desgnng transactons wthout framng effects n teratve queston formats. Journal of Envronmental Economcs and Management, 43: Dzegelewska, D.A. and Mendelsohn, R., Does No mean No? A protest methodology. Envronmental and Resource Economcs, 38: Farr, R.M., Common sense, scence and socal representatons. Publc Understandng of Scence. 2: Flachare, E. and Hollard, G., Controllng startng-pont bas n double-bounded contngent valuaton surveys. Land Economcs, 82: Flachare, E. and Hollard, G., Indvdual senstvty to framng effects. Journal of Economcs Behavor and Organzaton, 67: Flachare, E., Hollard, G. and Luchn, S., Heterogeneous anchorng n dchotomous choce valuaton framework. Louvan Economc Revew, 73:

22 Hanemann, M., Looms, J. and Kannnen, B., Statstcal effcency of doublebounded choce contngent valuaton. Amercan Journal of Agrcultural Economcs, 73: Herrges, J. and Shogren, J., Startng pont bas n dchotomous choce valuaton wth follow-up questonng. Journal of Envronmental Economcs and Management, 30: Johnston, R.J., Ranson, M.H., Besedn, E.Y. and Helm, E.C., What determnes wllngness to pay per fsh? A meta-analyss of recreatonal fshng values. Marne Resource Economcs, 21: Lechner, S., Rozan, A. and Lasney, F., A model of the anchorng effect n dchotomous choce valuaton wth follow-up. Workng Papers of BETA Leggett, C.G., Kleckner, N.S., Boyle, K.J., Dufeld, J.W. and Mtchell, R.C., Socal desrablty bas n contngent valuaton surveys admnstered through n-person. Land Economcs, 79: Lyle, J.M., Tracey, S.R., Stark, K.E. and Wotherpoon, S., survey of recreatonal fshng n Tasmana, Tasmanan Aquaculture and Fsheres Insttute Report. Wheeler, S. and Damana, R., Valung New Zealand recreatonal fshng and an assessment of the valdty of the contngent valuaton estmates. The Australan Journal of Agrcultural and Resource Economcs, 45: Whtehead, J.C., Incentve ncompatblty and startng-pont bas n teratve valuaton questons. Land Economcs, 78:

23 Table 1 Descrptve statstcs Varable Descrpton Mean Std Dev As_Caught The number of Australan salmon caught Fh_Caught The number of flathead caught Oth_Caught The number of other speces caught As_Target Specfcally targetng Australan salmon (yes = 1, no = 0) Fh_Target Specfcally targetng flathead (yes = 1, no = 0) Oth_Target Specfcally targetng other speces (yes = 1, no = 0) Non_Target Not targetng any speces Male Gender (male = 1, female = 0) Age Age (<20 = 1, = 2,, = 6, >70 = 7) Fulltme Currently workng full tme (yes = 1, no = 1) Income Income (<$20k = 1, $20k-40k = 2,, $80k-100k = 5, >$100K = 6) Days The number of days spent fshng n the last 12 months PubHolday The most recent day of fshng was a publc holday (yes = 1, otherwse 0) DayTrp The most recent day of fshng was a day-trp (yes = 1, otherwse 0) Hours The amount of tme spent fshng on the most recent day of fshng Importance How mportant fshng was on that fshng day (most mportant = 3,, less mportant = 1) OtherPersons The number of other persons n the fshng party Chldren Respondent went fshng wth hs or her chldren (yes = 1, no = 0) Boat Fshed from boat (yes = 1, no = 0) Shore Fshed from a shore (yes = 1, no = 0) Jetty Fshed from a jetty (yes = 1, otherwse = 0) MotEatng The man motvaton for gong fshng was to catch fsh for eatng (yes = 1, no = 0) MotOut The man motvaton for gong fshng was to enjoy the outdoors (yes = 1, no = 0) MotFrends The man motvaton for gong fshng was to spend tme wth frends/famly (yes = 1, no = 0) MotSport The man motvaton for gong fshng was to fsh for sport (yes = 1, no = 0) MotOther The man motvaton for gong fshng was other reasons (yes = 1, no = 0) Condtons Overall fshng condton (excellent = 5,, terrble = 1) Cost Total amount spent for the last day of fshng

24 Table 2 Descrptve statstcs for valuaton questons Varable Descrpton Mean Std Dev Valuaton questons b 1 Bd value for frst queston b 2 Bd value for second queston y 1 Response to frst queston (yes = 1, no = 0) y 2 Response to second queston (yes = 1, no = 0) Jont frequences of responses Pr(y 1 = 1, y 2 = 1) 0.43 Pr(y 1 = 1, y 2 = 0) 0.32 Pr(y 1 = 0, y 2 = 1) 0.17 Pr(y 1 = 0, y 2 = 0)

25 Table 3 Ctaton rate of each category Category Ctaton rate Feelng 65.2% Fsh & Fshng object 37.6% Famly & Frends 32.3% Food & Drnk 29.5% Envronment 22.7% Holday 22.0% Restrctons 11.5% Weathers 6.2% 25

26 Table 4 Estmaton results for sngle-bounded and double-bounded models Model I Model II Independent varables (sngle-bounded) (double-bounded) Estmates t-statstcs Estmates t-statstcs Constant (1.69)* (2.49)*** As_Caught (-0.13) (-0.34) Fh_Caught 0.28 (-0.06) (-0.022) Oth_Caught 0.93 (0.78) 0.49 (0.67) As_Target (-0.21) (-0.61) Fh_Target (-1.55) (-1.72)* Oth_Target (-0.81) (-1.1) Non_Target (-1.64) (-1.87)* Non_Target As_Caught (0.16) (0.97) Non_Target Fh_Caught 0.93 (0.45) 1.01 (0.8) Non_Target Oth_Caught 4.32 (-0.85) (-0.38) Male (2.06)** (2.06)** Age (-1.79)* (-1.76)* Income 7.24 (2.06)** 4.95 (2.39)*** Days 0.07 (0.33) 0.09 (0.66) DayTrp (-0.95) (-1.47) Hours (0.16) 1.15 (0.63) Importance 5.38 (0.81) 0.89 (0.21) OtherPersons 5.28 (1.72)*** 7.93 (4.13)*** Chldren (-2.14)** (-2.15)** boat (1.64) (1.75)* MotOther (-2.14)** (-2.73)*** Condtons 0.21 (0.1) 1.25 (0.95) σ (p-value) (0.002) *** (<0.001) *** log-lkelhood Pseudo-R^ Predcton success 81.2% 55% Observatons Mean WTP *** = 1% level of sgnfcance, ** = 5 % level of sgnfcance, * = 10 % level of sgnfcance 26

27 Table 5 Estmaton results for homogeneous anchorng models Independent varables Model III Model IV (homogeneous anchorng (homogeneous anchorng) and shft effect) Estmates t-statstcs Estmates t-statstcs Constant (1.53) (1.43) As_Caught (-0.47) (-0.5) Fh_Caught 0.01 (0.03) (0.05) Oth_Caught 0.93 (0.76) 0.93 (0.75) As_Target (-0.5) (-0.51) Fh_Target (-1.42) (-1.42) Oth_Target (-0.82) (-0.81) Non_Target (-1.54) (-1.57) Non_Target As_Caught (0.95) (0.94) Non_Target Fh_Caught 1.33 (0.68) 1.38 (0.69) Non_Target Oth_Caught (-0.33) (-0.3) Male (1.81)* (1.82)* Age (-1.56) (-1.61) Income 7.77 (2.01)** 7.92 (2.02)** Days 0.08 (0.37) 0.08 (0.37) DayTrp (-1.3) (-1.32) Hours 0.85 (0.3) 0.91 (0.3) Importance 3.56 (0.52) 3.57 (0.51) OtherPersons (2.53)** (2.58)*** Chldren (-1.7)* (-1.73)* boat (1.65)* (1.68)* MotOther (-2.07)** (-2.15)** Condtons 1.65 (0.78) 1.65 (0.95) σ (p-value) (0.005) *** (0.003) *** γ (p-value) (0.023) ** (0.0098) *** δ (p-value) - (0.0247) ** log-lkelhood Pseudo-R^ Predcton success rate 55.6% 55.3% Observatons Mean WTP *** = 1% level of sgnfcance, ** = 5 % level of sgnfcance, * = 10 % level of sgnfcance 27

28 Table 6 Estmaton results for heterogeneous anchorng models (gender) Model V Model VI Independent varables (heterogeneous anchorng) (heterogeneous anchorng and shft effect) Estmates t-statstcs Estmates t-statstcs Constant (1.53) (1.46) As_Caught (-0.48) (-0.47) Fh_Caught 0.01 (0.033) (0.031) Oth_Caught 0.93 (0.76) 0.90 (0.73) As_Target (-0.5) (-0.5) Fh_Target (-1.43) (-1.43) Oth_Target (-0.82) (-0.81) Non_Target (-1.55) (-1.52) Non_Target As_Caught (0.95) (0.95) Non_Target Fh_Caught 1.33 (0.68) 1.31 (0.68) Non_Target Oth_Caught (-0.33) (-0.34) Male (1.80)* (1.88)* Age (-1.57) (-1.56) Income 7.78 (2.02)** 7.70 (1.99)** Days 0.08 (0.38) 0.08 (0.39) DayTrp (-1.3) (-1.3) Hours 0.86 (0.29) 0.88 (0.31) Importance 3.60 (0.52) 3.60 (0.53) OtherPersons (2.53)** (2.40)** Chldren (-1.70)* (-1.68)* boat (1.66)* (1.64) MotOther (-2.07)** (-2.00)** Condtons 1.66 (0.78) 1.66 (0.8) σ (p-value) (0.005) *** (0.005) *** γ (p-value) (0.0246) ** (0.0205) ** γ (p-value) (0.0458) ** (0.0825) * δ (p-value) - (0.286) log-lkelhood Pseudo-R-squared Predcton success rate Observatons Mean WTP % % *** = 1% level of sgnfcance, ** = 5 % level of sgnfcance, * = 10 % level of sgnfcance 28

29 Table 7 Estmaton results for heterogeneous anchorng models (manstream) Independent varables Model VII Model VIII (heterogeneous anchorng (heterogeneous anchorng) and shft effect) Estmates t-statstcs Estmates t-statstcs Constant (1.53) (1.43) As_Caught (-0.5) (-0.49) Fh_Caught 0.02 (0.038) (0.03) Oth_Caught 0.94 (0.76) 0.92 (0.74) As_Target (-0.53) (-0.49) Fh_Target (-1.42) (-1.43) Oth_Target (-1.1) (-0.82) Non_Target (-1.55) (-1.51) Non_Target As_Caught (0.96) (0.93) Non_Target Fh_Caught 1.36 (0.69) 1.33 (0.66) Non_Target Oth_Caught (-0.32) (-0.33) Male (1.81)* (1.79)* Age (-1.56) (-1.56) Income 7.74 (2.00)** 7.95 (2.00)** Days 0.08 (0.37) 0.08 (0.35) DayTrp (-1.3) (-1.28) Hours 0.79 (0.27) 0.85 (0.28) Importance 3.54 (0.51) 3.88 (0.55) OtherPersons (2.52)** (2.50)** Chldren (-1.70)* (-1.7)* boat (-1.64) (1.66)* MotOther (-2.06)** (-2.10)** Condtons 1.70 (0.8) 1.64 (0.76) σ (p-value) (0.005) *** (0.005) *** γ (p-value) (0.0232) ** (0.0145) ** γ (p-value) (0.3096) (0.0705) * δ (p-value) - (0.277) log-lkelhood Pseudo-R-squared Predcton success rate Observatons Mean WTP % % *** = 1% level of sgnfcance, ** = 5 % level of sgnfcance, * = 10 % level of sgnfcance 29

30 Fgure 1 Dstrbuton of responses for each ntal bd value 100% 80% 60% YY YN NY NN 40% 20% 0% Intal bd value (b 1 ) 30