Controlling Starting-Point Bias in Double-Bounded Contingent Valuation Surveys Author(s): Emmanuel Flachaire and Guillaume Hollard Source: Land Economics, Vol. 82, No. 1 (Feb., 2006), pp. 103-111 Published by: University of Wisconsin Press Stable URL: http://www.jstor.org/stable/27647693 Accessed: 11/09/2010 16:05 Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/action/showPublisher?publisherCode=uwisc. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

University of Wisconsin Press is collaborating with JSTOR to digitize, preserve and extend access to Land Economics.

http://www.jstor.org

Bias Starting-Point Contingent

Controlling

Emmanuel In

ABSTRACT. bias

Flachaire

we study starting-point sur valuation contingent in applications arises that Indeed, response questions. be influenced may questions

this paper,

in double-bounded

veys. This phenomenon use multiple valuation to follow-up valuation

in the initial valuation by the bid proposed question. Previous have in order researches been conducted to control for such an effect. However, that they find we control are lost when gains efficiency for un to a single dichot desirable relative response effects, omous choice to these results, question. Contrary we propose a way to control for starting-point bias in double-bounded

questions

with

gains

in effi

ciency. (JEL C35, Q26) I. INTRODUCTION There exist several ways to elicit individ uals' willingness to pay for a given object or policy. Contingent or CV, is a valuation, survey-based

method

to measure

nonmar

ket values, among the important literature see Mitchell and Carson (1989), Hausman and Willis (1993), Bateman (1999). To elicit the individual maximum willingness to pay, participants are given a scenario a policy to be implemented. that describes are then asked to report the amount They they are ready to pay for it. In order to elicit WTPs, the use of discrete-choice format in contingent valu ation surveys is strongly recommended by the work of the NOAA panel (Arrow et al. of asking a bid to the 1993). It consists with a question like "if it costs respondent to pay $x to obtain, would you be willing one advantage that amount?" of Indeed, the discrete-choice format is that itmimics the decision-making task that individuals face in everyday life since the respondent Land Economics February 2006 82 (1): 103-111 ISSN 0023-7639; E-ISSN 1543-8325 ? 2006 by the Board of Regents of the University

of Winconsin

System

in Double-Bounded Valuation Surveys and Guillaume

Hollard

How accepts or refuses the bid proposed. of this format is that it ever, one drawback leads to a qualitative variable dependent or answers (the respondent "no"), "yes" which reveals little about individuals' WTP. In order to gather more on information Hanemann and WTP, respondents' (1985) Carson to add a follow-up (1985) proposed discrete-choice to improve effi question of discrete-choice ciency questionnaires. as the double This mechanism is known bounded model. This basically consists of a second bid to the respondent, asking greater than the first bid if the respondent answers "yes" to the first bid, and lower otherwise. A key disadvantage of the double-bounded model is that subject's to the second bid may be in responses fluenced This is by the first bid proposed. the so-called bias. starting-point Several studies document that iterative formats produce in re anomalies question behavior. results show spondent Empirical that inconsistent results may appear, that if is, the mean WTP may differ significantly it is implied by the first question or only by the follow-up question. Different interpre tations have been proposed?the first bid can be interpreted as an anchor, a refer ence point,1 or as providing information about the cost?as well as different models to control for these anomalies (see Cameron and Quiggin and Shogren 1994; Herriges Kanninen and Carson 1996; Alberini, 1997; Whitehead 2002; DeShazo 2002). However,

The authors are assistant professors, in respectively, Universit? Paris I Panth?on-Sorbonne, and EUREQua, de Marne-la-Vall?e. The authors wish OEP, Universit? to thank Jason Shogren, Luchini, Ed Hopkins St?phane two anonymous and reviewers for their insightful comments and suggestions. 1 Kahneman clear definitions of proposes (1992) and framing effects and emphasizes the anchoring difference in the underlying mental processes.

104 Land Economics

these studies suggest that when we control for such undesirable effects, effi response ciency gains can be lost relative to a single dichotomous-choice question. At the moment, it is still difficult to con trol for such effects in an effective manner. The purpose of this paper is to address this issue. We present and compare different models in the litera proposed previously new econometric ture. We also develop models that combine the main feature of results pro existing models. Our empirical vide strong evidence that we can obtain a gain in efficiency by taking into account the follow-up question. They give a better of how subjects form their understanding to the payment responses questions. as follows. In The paper is organized mod Section 2, we review the econometric in the literature and we pro els proposed new models. In Section 3, we compare pose these different models with an application. are drawn in Section 4. Conclusions

II. ECONOMETRIC MODELS In this section, we review different mod in the literature to control for els proposed the anchoring effect, shift effect and fram new models ing effect. Then, we propose all these effects. that combine Let us first consider W0i, the true will is de /, which ingness to pay of individual fined as follows Wot

=

Xi(?)

+

ut

Ui

~ A^cj2),

[1]

where the unknown ? and a2 parameters a k x 1 vector and a scalar, are respectively, function depending where x? is a non-linear on k independent variables. explanatory The number of observations is equal to n and the error terms u? are normally distrib zero and variance a2. The uted with mean to pay (WTP) of the respondent willingness / is not observed, but his answer to a bid b? is. The subject's answers are defined as n = 1 ifWot > bi and n = 0 ifW0i < b?, where "yes"

= 1 if the r? respondent to the first question and

[2]

/ answers

the respondent

no

to the first

question.

The double-bounded model, proposed by Hanemann (1985) and Carson (1985), consists of asking a second bid (follow-up to the respondent. If the respon question) dent / answers "yes" to the first bid, bu, the second bid b2? is higher and lower other wise. The standard procedure, Hanemann assumes that (1985) and Carson (1985), are independent of the respondents' WTPs bids and deals with the second response as the first discrete in the same manner choice question: =

Wu

Woi

and W2i

=

W0i.

[3]

An individual answers "yes" to the first bid ifWu > bu and "yes" to the second bid the double-bounded if W2? > b2?. Thus, assumes that the same random util model to the both responses generates ity model first and the second bid. introduction of follow-up ques However, can between generate inconsistency tioning answers to the second and first bids. To deal with inconsistency of responses, several mod in the literature. els have been proposed Anchoring Effect con and Shogren's approach Herriges in which the follow-up ques siders a model to pay. the willingness tion can modify to them, respondents combine According their prior WTP with the value provided by effect is then the first bid, this anchoring defined as follows = Woi and W2i (1 y)Wu + ybli9

=

Wu

[4]

individual answers 0 < 7 < 1. An > bu and "yes" to first if bid the Wu "yes" to the second bid ifW2? > b2i .From [4], it follows that,

where

ru

=

1 <=> Woi

>

bu

and

r2i

=

1 ^^

W2i

>

b2i.

[5] is rather economic interpretation are to Individuals adjust supposed simple. aver their initial WTP by doing a weighted The

/ answers = 0 if r?

2006

February

82

Flachair

(1)

e and Hollard:

Star

ting-Point

amount. age of thisWTP with the proposed the importance of anchor Thus, 7 measures = 1which means that no ing. It ranges from 7 = 1 which means at is work, to 7 anchoring that subjects and ignore their prior WTP replace itwith the proposed bid. This model is thus a simple and efficient manner to test the importance of anchoring. The wider is the anchoring the less information effect, provided by the follow-up question. assumes This last model that only the follow-up question gives rise to anchoring effects and only the first bid has an influ ence on the second answer. These two last are we can restrictive and hypotheses quite show that the model is still valid if we a more consider effect, general anchoring that is, both bids can influence subject's responses.

can com Let us assume that individuals bine their prior WTP with the values pro vided by the current and by the past bids the following offer. It leads us to consider model: -

=

Wu

W2i

(1

=

(i

+ ybu

y)W0i

_

and

6)Wl/ + bb2h [6]

0 < 7 < 1 and 0 < ? < 1 .An where answers individual "yes" to the first bid offer if : m

=

1 <=* Wu

>

+ ybu > bu^

bXi <=*(1-

y)W0l

Woi > bu.

[7]

that a last condition This suggests effect of the first bid anchoring potential re the subject's offer does not influence individ An to the initial sponse question. ual answers yes to the second bid offer if: ru

=

1 ^=> W2i

>

b2i <=>

(1

8)WU

+ 6?>2i> b2i *=> Wu > b2i. [8] that a po This last condition suggests effect of the second bid tential anchoring not influence the subject's offer does to the response question. More follow-up over, we can see that the first bid offer can the second answer, because Wu influence and of is a combination of the prior WTP the value provided by the first bid.

Bias

and

Valuation

Contingent

105

Finally, these results show that the current bid offer can have an impact on theWTP but does not affect the subject's responses. Only the first bid offer can influence the answer to the follow-up It follows that the question. ? cannot be estimated. parameter This suggests the remarkable conclusion that when we use the model proposed by and Shogren in practice, Herriges (1996) we can assume a potential effect anchoring of both bids. Shift Effect and Carson Alberini, Kanninen, (1997) assume that the proposition of two bids amount to sys may cause individual WTP shift between the two responses. tematically Thus, the two answers are not based on the same WTP and this can explain inconsis is defined tency of responses. A shift inWTP as follows: Wu

=

W0i

and W2i

=

Wu

+

6,

[9]

? represents the parameter where the struc tural shift. Such a model is inspired by the follow The intuition. first bid may be inter ing as information about the providing preted cost of the object. Thus, an individual who accept the first bid offer may understand to pay an the second bid as a proposition amount It for the same object. additional that this individual may cut down follows to take that phenomenon their answers an in when into account. Symmetrically, the fol the first bid dividual offer, rejects as could be interpreted low-up question a proposition for a lower quality level of the to cut itmay lead individual object. Again, down

their

answers.

In

such

case,

the

pa

to be negative. rameter ? is expected A positive ? is however and possible as "yea saying" be could be interpreted itsWTP havior: an individual overestimate the interviewer's in order to acknowledge But, we are not aware of data proposition. that is, es this interpretation, supporting timated values of ? are negative. that a model with shift effects as Note sumes the follow-up that only question

Land 106

Economics

gives rise to shift effect and the shift is in of the bids proposed. These two dependent are quite In last hypotheses restrictive. to believe it could be difficult that deed, answers the respondent the first question re and that the behavioral truthfully, is not the same if the proposed sponses true WTP bid is close to the individual's or if it is far from it. However, these hy are required by an identification potheses condition and we cannot relax them as we have done in the anchoring model. Anchoring

and Shift Effects

=

W0i

and W2i

=

(1

+

y)Wu

ybu

+

?.

[10] The is simply a certain interpretation combination of both the anchoring and the shift effect explanations. Indeed, we can rewrite W2i = Wu + 7 (bu ~ Wu) + ? , that its prior WTP is, an individual may update term (shift) and a multi with a constant factor of the distance between the plicative and the first bid offer (anchor prior WTP and Caplan ing). See Aadland (2004) and Whitehead details. (2004) for estimation Framing Effect DeShazo (2002) proposes decomposing iterative questions into their ascending and sequences. His empirical results descending of responses occur suggest that inconsistency in It leads him to sequences. only ascending recommend in practice the double using bounded model with only decreasing follow can be written, up questions. This last model Wu

=

W0i

and W2i

=

W0i

is "yes," the first bid first subject's response as a reference is interpreted offer point: to it, the follow-up is question compared are framed as a loss and thus, individuals more likely to answer "no" to the second re if the first subject's offer. In contrast, is "no," the first bid offer is not sponse as a reference point. Thus, the interpreted versus to ascending behavioral responses are different. iterative questions descending New

Models

results based on all the pre show that in the presence of effect, shift effect or framing ef anchoring fect, the estimated mean and the estimated can be significantly of WTP bi dispersion ased. Herriges and Shogren (1996) con clude that the efficiency gains from the are lost once we con follow-up question trolled for the anchoring effect. They sug model in the gest to use the single-bounded of significant anchoring effect and presence the follow-up thus, to remove questions. DeShazo shows that most of the (2002) biases are more likely to occur in the as It leads him to recom sequence. cending to keep the follow-up questions mend from the descending and to remove sequences the follow-up questions from the ascending Empirical vious models

Whitehead the Herriges (2002) modifies and Shogren to allow model anchoring both anchoring and shift effects, Wu

2006

February

if ru

=

0.

[11]

The distinction between and ascending leads Deshazo to sequences descending attribute the parameter to inconsistency effect. framing effect rather than anchoring Indeed, using prospect theory (Kahneman and Tversky that if the 1979), he argues

sequences

only.

In order

to get information from the as to back, we propose sequences cending correct biases in the follow-up questions from the ascending sequences.2 We consider three different models, with W1? = W0?: and Anchoring

Framing

W2i

=

Wu + y(bu

-

Effects

Wu)ru-

[12]

If the subject's response is "no" to the = 0 and = first bid offer ru Wlh Wy otherwise the WTP is updated with an an 2 As

long as the model [11], proposed by DeShazo consistent results with the single (2002) provides bounded occur biases in ascending model, sequences there is no need to consider more only. Thus, compli cated models where biases occur in both ascending and descending

sequences.

82

e and Hollard:

Flachair

(1)

Star

ting-Point

in the model pro choring effect, as defined and Shogren (1996). posed by Herriges Framing and Shift Effects W2i

=

+

Wu

?ri,-.

[13]

is "no" to the If the subject's response = = 0 and first bid offer r1? W2? Wyi, is updated with a shift the WTP otherwise in the model proposed effect, as defined by and Carson Kanninen, Alberini, (1997). and Shift Effects

Framing and Anchoring Wli

=

Wu + y(bu Wu)ru + 6rlf.

[14]

is "no" to the If the subject's response = = 0 and first bid offer ru W2i Wlh is updated with both the WTP otherwise as defined in and shift effects, anchoring Whitehead the model by (2002). proposed can be of this last model Implementation with the proba based on a probit model, answers / the individual that "yes" to bility = the fh question, 1,2 equals to: j P(Wji

>

bjd

= O

XiOL

-bn

-

+

d((bu

bj?Djru)

Bias

and

Contingent

107

Valuation

in that the French Ministry search program affairs started in charge of environmental valuation 1995. It is based on a contingent involves a sample of users of survey which The pur the natural reserve of Camargue.3 valuation survey was pose of the contingent were to evaluate how much individuals reserve to to the natural preserve pay willing an

using

entrance

fee.

The

survey

was

to 218 recreational visitors administered face-to-face the 1997, using spring during se visitors were interviews. Recreational in seven sites all around the lected randomly natural reserve. The WTP question used in was a dichotomous-choice the questionnaire with follow-up.4 There was a high response of rate (92.6%). For a complete description valuation the contingent survey, see Claeys and Luchini Mekdade, G?niaux, (1999). of theWTPs were estimated using Means a linear model and Leonard (McFadden and Crooker Indeed, Herriges 1993). (2004) show that the simple linear probit the model is often more robust in estimating mean WTP and than others parametric 1 presents models. Table semi-parametric means and estimated estimated jli disper for all models, with sions a of the WTPs The standard errors given in parentheses.

[15] = where Dx = 0 and D2 = 1 ,a = ?/a, 6 7/ ? ? = 7a). Based on this 67(a 70) and X (a are interrelated the parameters equation, to according ?

=

ao,

y

=

8a/(l

+

6a)

and

?

=

\a(l

7).

[16] in [12] and [13] are The models proposed in two special cases of the model proposed with 6 = 0 and 7 = 0. It [14], respectively follows that, they can be implemented based on the probability [15], respectively with X = 0 and 0 = 0. III. APPLICATION To test our model empirically, we use the valuation of a contingent results main a re survey which was carried out within

3

is a wetland in the south of France The Camargue is a major The Camargue 75 000 hectares. covering in France and is host to many wetland fragile ecosys is the result of tems. The exceptional biological diversity area inhabited water and salt in an "amphibious" by numerous is the result of an species. The Camargue the river, the sea, and man. endless struggle between of dikes the construction the last century, while During to land for farming and embankments salvaged more it cut off the Camargue meet economic needs, region it of regular supplies of from its environment, depriving and silt previously fresh water provided by flooding. the the wildlife, and to preserve of this problem Because are are now strictly managed. resources There water and a dense and draining stations, irrigation, pumping, the river delta. How of channels network throughout are quite large. ever, the costs of such installations 4 from (5, 10, The first bid bu was drawn randomly 15, 20, 25, 30, 35, 40, 45, 50, 60, 70, 80, 90, 100}. The from the same set of second bid b2i was drawn randomly amount 3 the additional values with b2i < bu and with to the first bid was (resp. b2i > bu and 120) if the answer of answers "no" (resp. "yes"). The number fno,noj, (yes,no,), and (no,yes,), to 20, 12, 44, and 121.

(yes,yes,)

are,

respectively,

equal

108

Land

1

TABLE Mean

and

Dispersion

February 2006

Economics

of WTPs

in French

Francs /

M Single 113.52 Double 81.78 126.38 Anchoring 89.69 Shift 141.38 and Shift Anchoring Framing106.72 106.71 FramandAnch 116.98 Fram and Shift 116.39 Fram and Anch and Shift

[87.9;138.7] [78.2;85.5] [98.2;155.4] [85.6;93.8] fI07.3;176.0J [90.9;121.8] [93.6;119.3] [103.9;129.7] [101.4;131.1]

mean of WTPs is a function of parameters: its standard error and its confidence interval cannot be obtained from the es directly timation results. Confidence intervals of (l are presented in brackets, they are obtained and Robb by simulation with the Krinsky see Haab and McConnell procedure, (2003, for more details. 106-13) From Table 1 it is clear that the standard errors (in parentheses) and the confidence consider decrease intervals (in brackets) one uses usual double when the ably instead of the model bounded (Double) model single-bounded (Single). This result confirms the expected gains pro efficiency vided when the second bid is taken into ac and Kanninen count (Hanneman, Loomis, of of the mean estimates However, 1991). are very different in both models WTPs the mean of (113.52 vs. 81.78). Moreover, = of the single bounded model, WTPs ?x to not confi the 95% 113.52, does belong in the dence interval of the mean of WTPs in Such double-bounded model, [78.2;85.5]. results lead us to consider other consistent as presented in the previous section. models, a model At first, we estimate with an as defined in effect (Anchoring), choring From and Shogren (1996). [4] by Herriges Table 1, we can see that the anchoring = 7 0.52, is significant. Indeed, parameter, a Likelihood Ratio test, equals to twice the estimates difference of the log-likelihood = = 5.78, p-value rejects the (LR 0.016), = 0. This test confirms null hypothesis 7 effect in the the presence of an anchoring answers. When for correcting respondents'

45.38(23.61) 42.14 (5.23) 82.11 (40.83) 44.74 (5.77) 85.50(43.78) 40.39 (11.91) 60.19(14.77) 65.03(14.40) 64.63(16.34)

- -53.3 - -179.6 -176.7 -

0.52 (0.23) - 8.10

-7.S?

0.52(0.24) - -68.8-

-175.3 -172.8

(2.90) (2.91)

-176.9 -

0.40(0.16) -

-171.8 -171.8

-30.61(14.33) -31.60(21.77)

-0.02(0.42)

in effect, results are consistent, of WTPs of the that, the mean = ?x 113.52, belongs model, single-bounded to the 95% confidence interval of the anchor standard [98.2;115.4]. However, ing model, errors and confidence intervals increase sig so that, even if follow-up ques nificantly, of parameter increases precision tioning estimates efficiency gains are (see Double), once the lost anchoring effect is completely Accord taken into account (see Anchoring). to "the this result, ap single-bounded ing proach may be preferred when the degree of and is substantial" (Herriges anchoring a we estimate 1996, 124).Then, Shogren model with shift effect (Shift), as defined in and Carson Kanninen, [9] by Alberini, to conclusions lead similar Results (1997). model. than the double-bounded Indeed, we can see a large gain in efficiency: standard errors and confidence intervals are more pre the results are inconsistent: cise. Moreover, mean of WTPs of the single bounded model = 113.52 does not ji belong to the 95% con fidence interval of the shift model, [85.6; 93.8]. Parameter estimates of amodel with both bias and shift effects, as defined anchoring in [10] by Whitehead (2002), are given in and Shift. Based the line named Anchoring on the criterion of maximum likelihood, = is better than the others this model (? ? are consistent, in the 172.82). Results to the 95% sense that, (i = 113.52 belongs with of the model interval confidence and shift effects [107.3;176.0]. anchoring anchoring the sense

However,

compared

we

can

see

a

loss

to the single bounded

of

precision

model.

82

(1)

Flachaire

and Hollard:

Starting-Point

The only one model, presen previously ted in the literature, which give consistent results with the single bounded model and a is the model proposed gain in efficiency by as in [11]. Results DeShazo defined (2002), are presented in the line named Framing. The 95% confidence interval [90.9;121.8] = 113.52 and includes the mean of WTPs ji is narrower than the 95% confidence inter val of the single bounded model [87.9;138.7]. to In his conclusion, Deshazo recommends remove could be all the answers which that is, the influenced by framing effect, answers to the second bids if the respondents answer yes to the first bids. From the previous results, it is clear that of the problem there is no way to handle bias in an effective manner. starting-point This suggests that the best we can do in remove to the answers which is practice to starting-point bias. could be subject use of the iterative Nevertheless, questions more information about should provide of WTPs. the distribution Then, better re if all the answers sults should be expected are used and if a to iterative questions bias is used. correct model of starting-point the three new To go further, we consider in the last section, which models proposed consider all the answers to the second bids: Line Fram and Anch presents estimation in [12], that defined results of the model bias in the is, a model with an anchoring see that can We sequence only. ascending interval of the mean the 95% confidence is equal to [93.6;119.3]. of WTPs estimation Line Fram and Shift presents defined results of the model [13], that is, a model with a shift effect in the as cending

sequence

only.

We

can

see

that

interval of the mean the 95% confidence of WTPs is equal to [103.9;129.7]. and Shift presents Line Fram and Anch defined results of the model estimation in [14], that is, amodel with an anchoring in the ascending bias and a shift effect sequence only. We can see that the 95% interval of the mean of WTPs confidence is equal to [101.4;131.1]. three models, It is clear that for these are consistent with the single results

Bias

and

Contingent

Valuation

109

bounded model: the mean of WTPs to the three confidence 113.52 belongs tervals.

results

Furthermore,

are more

{i

= in

pre

cise: the standard errors (in parentheses) are smaller and the confidence intervals (in are narrower than those of the brackets) model. single-bounded In addition, we can remark that the two models Fram and Anch and Fram and Shift are special cases of the model Fram and Anch and Shift, respectively, with 6 = 0 and 7 = 0. From this last more general = 0 = we cannot model, 0.004, reject 7 (LR = 0 = p-value 0.99), but we can reject ? = = 10.31,/?-value 0.001). These results (LR lead us to select the model Fram and Shift as our contingent the one which fit better with shift valuation data, that is, a model effect in the ascending sequences only. Table 2 presents full econometric results with consistent results: of several models the the single-bounded model (Single), and our se model of Deshazo (Framing) lected model (Fram and Shift). The esti mates of the vector of coefficients ? (rather than ? / a), the standard deviation a and the shift parameter ? are directly see equations [1], [11], and [13]. presented, It is clear from this table that the standard are errors in the Fram and Shift model com reduced nearly always significantly to the standard errors in the other pared is sig models. Indeed, only one parameter nificant in the Single model when eight pa are significant in the Fram and rameters In other words, efficiency gains Shift model. are still present model in our selected account all the answers) into takes (which to the other models (which re compared answers move that could be influenced by the first bid offer). IV. CONCLUSION Follow-up are model on mation spondents.

in double-bounded questions to give more infor expected to pay of re the willingness Then,

many

economists

have

to obtain gains in this last model over the single bounded model. efficiency recent that this studies show However, can be inadequate and can give in model favored

110

Land

TABLE Parameters Variable ?0: Constant home-site ?i: Distance a car to come in ?2: Using ?3: Employee class ?4:Middle ?5: Inactive class ?6:Working collars ?: White with family ?s: Visiting alone ?9: Visiting with a group ?K):Visiting ?n: First visit facilities proposed ?i2: New ?i3: Other financing propose ?14: South-West ?15: South-East ?i6: Questionnaire type 1 ?i7: Investigator 2 ?i8: Investigator

Estimates,

February 2006

Economics

Standard

2

Errors

Single 35.43 9.30 -61.71 * 95.86 109.96 52.58 97.28 80.33 4.71 61.11 44.79 51.42 56.93 -32.03 -24.18 42.04 -28.19 23.44 -17.12 45.38

95%) Fram

Framing 39.15 * 08.93 -59.54 * 88.13 * 94.14 43.53 91.55 * 68.42 3.02 53.61 51.05 41.57 * 55.78 -30.04 -21.19 40.94 -28.19 23.43 -11.22 * 40.39

(57.27) (5.30) (41.08) (46.86) (63.60) (38.44) (68.29) (42.16) (29.61) (101.67) (47.90) (35.29) (32.12) (27.60) (33.57) (58.26) (23.34) (56.29) (57.52) (23.61)

consistent results. Many different models have been considered in the literature to correct anomalies in respondent behavior that appear in dichotomous-choice, con data. However, the cor tingent valuation rections proposed show by these models that efficiency gains given by the iterative are lost when of questions inconsistency is controlled. responses The main contribution of this paper is to propose a model to control for starting point bias in double-bounded model, and, to previous still have contrary research, gains in efficiency relative to a single dichot omous choice DeShazo question. (2002) shows that descending se and ascending quences have different behavioral responses and recommend to restrict the follow-up to the first questions only if the answers bids are "no." To benefit from more in rather than not into formation, taking account the answers which could be influ enced by the first bid offer, we propose different models of starting-point bias in iterative ascending questions only. Our results show that a model with empirical shift effects in the ascending questions

at

(*: Significant

and Shift

80.46 *5.44 ; -60.90 * 69.68 * 82.69 50.41 64.69 * 65.56 10.51 97.05 4.56 14.37 * 43.59 -22.85 -33.04 38.69 -13.21 8.43 -27.90 * 65.03 * 30.67

(47.90) (3.95) (31.37) (36.99) (46.06) (32.39) (64.34) (32.11) (25.60) (81.13) (49.65) (29.01) (22.83) (18.19) (26.83) (44.80) (18.63) (45.05) (46.35) (11.91)

(49.51) (2.54) (28.45) (30.04) (32.26) (31.37) (53.86) (26.39) (25.23) (58.27) (36.07) (17.47) (16.72) (14.59) (28.36) (37.50) (13.50) (38.95) (42.07) (14.40) (14.33)

consistent results with the single gives bounded model and provides large effi the idea that ciency gains. This supports and shift effects can be framing, anchoring combined in an efficient manner.

References

Aadland,

D.,

and A.

Incompatibility ative Valuation

Economics Alberini,

A.,

"Modeling chotomous Economics

J. Caplan.

2004.

"Incentive

and

Bias Starting-Point Comment." Questions:

in Iter Land

80 (May): 312-15. B. Kanninen, Response Choice

and R. Incentive Valuation

Carson.

1997.

Effects

in Di

Data."

Land

73

309-24. (Aug.): P. R. Portney, E. E. Learner, Arrow, K., R. Solow, R. Radner, and H. Schuman. 1993. "Report of on Contingent the NOAA Panel Valuation." Technical 58 (Jan.): 1601-14. Report 1999. Valuing Envi Bateman, I., and K. Willis. ronmental and Practice Preferences: Theory of the Contingent Valuation Method in the US, and Developing Countries. U.K.: EU, Oxford, Oxford Cameron,

University T., and

Press. J. Quiggin. Valuation

Using Contingent chotomous Choice

with

1994. Data Follow-up

"Estimation a Di from Question

82

Flachaire

(1)

and Hollard:

Starting-Point

naire."

Economics Journal of Environmental and Management 27 (Nov.): 218-34. on Contingent R. "Three 1985. Carson, Essays Ph.D. of Califor Valuation." diss., University

et Sociologie."

In La

Formats."

J. R.

Mass.:

ampton, W. Hanemann, and Discrete Studies."

Response Northeast

Economics W., Hanemann, "Statistical Dichotomous

1985.

Contingent Journal of

(Apr.):

J. Loomis,

Surveys Valuation sources

Efficiency Choice

Valuation Agricultural -.

of

Double-Bounded

Contingent

Valuation."

R.

and R. C, to Value Public Method. for

T.

1989.

Carson.

Goods: Washington

The

Using

Contingent Re D.C.:

the Future.

J. C. 2002. "Incentive Whitehead, Incompatibility Valuation Bias in Iterative and Starting-Point 78 (May): 285-97. Land Economics Questions."

in Continuous

1991.

D.,

for Data Collection Methodologies In Contingent A Crit Valuation: and Analysis." New York: Hausman. ical Assessment.

Mitchell,

and B. Kanninen.

51:

Goods:

Valuing North

14: 5-13.

Anchor

Organizational Processes

and A. Tversky. 1979. "Prospect of Decisions under Risk." Analysis 313-27. 47 (Mar.): in "Issues and G. Leonard. 1993. McFadden, D., Valuation of Environmental the Contingent

Edward

Elgar. Issues "Some

Points,

An Theory: Econometrica

Transactions "Designing in Iterative Effects Question Econom of Environmental

360-85. ics and Management 43 (May): 2003. and K. E. McConnell. T. C, Haab, Resources. and Natural Environmental

"Reference

Feelings." and Human Decision

Kahneman,

2002.

Framing Journal

1992.

296-312.

451-80. DeShazo, without

A Critical

Valuation: North-Holland.

and Mixed

Norms, Behavior

Environ 27

D.

Kahneman,

Paris:

"Para J. R., and J. A. Herriges. 2004. of Estimation and Semi-Nonparametric in Dichotomous the Choice Willingness-to-Pay Economics

Economics

of Agricultural

J. A., and J. F. Shogren. 1996. "Starting in Dichotomous Bias Choice Valuation point with Follow-up Journal of Envi Questioning." Economics 30 ronmental and Management 112-31. (Jan.):

Econ?mica.

Framework."

111

Herriges,

Crooker, metric

Valuation Contingent mental and Resource

Journal

J. 1993. Contingent Hausman, Assessment. Amsterdam:

des ?conomique et Mod?les d'?valu P. Point.

Valuation

Contingent

73 (Nov.): 1255-63.

Valeur

ed.

and

American

nia, Berkeley. and S. Luchini. C, G. G?niaux, Claeys-Mekdade, ? la Camargue? Valeur Attribuer 1999. "Quelle Une ?conomie Interdisciplinaire Perspective M?thodes Hydrosyst?mes. ation des Services D?livr?s,

Bias

2004. Starting-Point tions: Reply." 316-19.

"Incentive Bias Land

Incompatibility Valuation in Iterative Economics

80

and Ques (May):