Would developing country commitments affect US households' support for a modified Kyoto Protocol?

15
ANALYSIS Would developing country commitments affect US households’ support for a modified Kyoto Protocol? Hui Li a , Robert P. Berrens a, * , Alok K. Bohara a , Hank C. Jenkins-Smith b , Carol L. Silva b , David L. Weimer c a Department of Economics, University of New Mexico, Albuquerque, NM 87131, USA b George Bush School of Government and Public Service, Texas A&M University, College Station, TX 87443, USA c Robert M. LaFollette School of Public Affairs, University of Wisconsin, Madison, WI 53706, USA Received 3 October 2002; received in revised form 29 September 2003; accepted 6 October 2003 Abstract Would US households be willing to pay more to support a modified Kyoto Protocol (MKP) if developing countries had binding future limits on greenhouse gas production? We explore this question using data from a unique set of national Internet samples and web-based surveys. Using an advisory referendum format, the contingent valuation method is applied to estimate annual household willingness-to-pay (WTP) for US Senate ratification of the Kyoto Protocol for a split-sample treatment: the basic Kyoto Protocol (BKP) (control group) versus a MKP that includes limits on future greenhouse gas production for major developing countries (treatment group). The results indicate that the treatment significantly increases the probability of a Yes vote on the advisory referendum; econometric modeling results provide evidence that the MKP significantly increases US households’ median WTP to support the treaty. D 2004 Elsevier B.V. All rights reserved. Keywords: Contingent valuation; Developing countries; Kyoto Protocol; Willingness-to-pay 1. Introduction The Kyoto Protocol is perhaps the most far-reach- ing international environmental treaty ever consid- ered. A core feature of the treaty is its binding emissions reduction targets for developed countries, but the absence of binding commitments for develop- ing countries. To date, the US has declined to ratify the treaty. As the largest economy in the world, US withdrawal will greatly impair the effectiveness of the Kyoto Protocol, which still appears poised to take effect. Consequently, an important point of concern is what might bring the US back to the treaty, or back to serious international negotiations. Within this broader debate, one question is whether binding commitments by the developing countries would increase popular support by US households for ratifying a modified Kyoto Protocol (MKP). Policymakers realize that developing countries will play a significant role in determining the success of international efforts to respond to global climate 0921-8009/$ - see front matter D 2004 Elsevier B.V. All rights reserved. doi:10.1016/j.ecolecon.2003.10.010 * Corresponding author. Tel.: +1-505-277-9004; fax: +1-505- 277-9445. E-mail address: [email protected] (R.P. Berrens). www.elsevier.com/locate/ecolecon Ecological Economics 48 (2004) 329 – 343

Transcript of Would developing country commitments affect US households' support for a modified Kyoto Protocol?

www.elsevier.com/locate/ecolecon

Ecological Economics 48 (2004) 329–343

ANALYSIS

Would developing country commitments affect US households’

support for a modified Kyoto Protocol?

Hui Lia, Robert P. Berrensa,*, Alok K. Boharaa, Hank C. Jenkins-Smithb,Carol L. Silvab, David L. Weimerc

aDepartment of Economics, University of New Mexico, Albuquerque, NM 87131, USAbGeorge Bush School of Government and Public Service, Texas A&M University, College Station, TX 87443, USA

cRobert M. LaFollette School of Public Affairs, University of Wisconsin, Madison, WI 53706, USA

Received 3 October 2002; received in revised form 29 September 2003; accepted 6 October 2003

Abstract

Would US households be willing to pay more to support a modified Kyoto Protocol (MKP) if developing countries had

binding future limits on greenhouse gas production? We explore this question using data from a unique set of national Internet

samples and web-based surveys. Using an advisory referendum format, the contingent valuation method is applied to estimate

annual household willingness-to-pay (WTP) for US Senate ratification of the Kyoto Protocol for a split-sample treatment: the

basic Kyoto Protocol (BKP) (control group) versus a MKP that includes limits on future greenhouse gas production for major

developing countries (treatment group). The results indicate that the treatment significantly increases the probability of a Yes

vote on the advisory referendum; econometric modeling results provide evidence that the MKP significantly increases US

households’ median WTP to support the treaty.

D 2004 Elsevier B.V. All rights reserved.

Keywords: Contingent valuation; Developing countries; Kyoto Protocol; Willingness-to-pay

1. Introduction

The Kyoto Protocol is perhaps the most far-reach-

ing international environmental treaty ever consid-

ered. A core feature of the treaty is its binding

emissions reduction targets for developed countries,

but the absence of binding commitments for develop-

ing countries. To date, the US has declined to ratify

0921-8009/$ - see front matter D 2004 Elsevier B.V. All rights reserved.

doi:10.1016/j.ecolecon.2003.10.010

* Corresponding author. Tel.: +1-505-277-9004; fax: +1-505-

277-9445.

E-mail address: [email protected] (R.P. Berrens).

the treaty. As the largest economy in the world, US

withdrawal will greatly impair the effectiveness of the

Kyoto Protocol, which still appears poised to take

effect. Consequently, an important point of concern is

what might bring the US back to the treaty, or back to

serious international negotiations. Within this broader

debate, one question is whether binding commitments

by the developing countries would increase popular

support by US households for ratifying a modified

Kyoto Protocol (MKP).

Policymakers realize that developing countries will

play a significant role in determining the success of

international efforts to respond to global climate

1 The Kyoto Protocol will not take into effect until it is ratified

by at least 55 of the Annex I countries, representing at least 55% of

the total 1990 carbon dioxide emissions. Annex I countries refer to

industrialized countries and those in transition to a market economy.

The Convention of the Kyoto Protocol includes 41 Annex I countries,

including Australia, Canada, Japan, Russia and the US. As of

H. Li et al. / Ecological Economics 48 (2004) 329–343330

change. In addition to concerns about implementation

costs to the US economy, a major concern of the Bush

administration and others regarding the Kyoto Proto-

col is that it exempts developing countries from

binding commitments for future production of green-

house gases. The objective of this study is to assess

whether US household support for the treaty would be

greater if the Kyoto Protocol were modified to impose

binding greenhouse gas production commitments on

developing countries. Our assessment uses data from

a web-based contingent valuation (CV) survey, and a

unique set of large, national (US) Internet samples.

In the split-sample survey, a control group of

respondents is presented with a national referendum

to advise the US Senate on ratification of the Kyoto

Protocol. A treatment group is presented with a

modified version of the Kyoto Protocol, which

includes binding commitments by the developing

countries. Using Internet sample data, we test a set

of hypotheses about the effects of selected explanato-

ry variables, particularly the MKP treatment, on

voting responses and annual US household willing-

ness-to-pay (WTP) to support ratification of the treaty.

The results indicate that the treatment significantly

increases the probability of a Yes vote on the advisory

referendum. Econometric modeling results provide

evidence that the MKP significantly increases US

households’ median WTP to support the treaty.

In the next section we briefly discuss the status of

the Kyoto Protocol. Section 3 discusses data collec-

tion and survey design. Section 4 discusses modeling

considerations and presents a set of hypotheses related

to the impact of the MKP on US household WTP for

Senate ratification. The final two sections assess the

empirical results and present conclusions.

February 2003, 111 instruments have been ratified, approved or

acceded to the treaty including Japan and Canada. However, the total

emissions only represent 44.4% of the 1990 level. Because the United

States, which has the largest global share of emissions (about 36.1%

of greenhouse gases), withdrew from the Kyoto Protocol, the treaty

effectively cannot enter into force without ratification by Russia.

Russia has a global share of 17.4% of emissions, and has recently

made statements at the World Summit on Sustainable Development

in September 2002, and the G-8 Summit in Evian in June 2003

confirming its intention to ratify. Further, in July 2003, Russia

finished their analysis of the costs and benefits analysis of complying

with the Kyoto Protocol, and the results favored ratification. For

status updates, visit the UN web page on climate change at: http://

unfccc.int/resource/kpstats.pdf (accessed on July 22, 2003).2 COP8 was held in New Delhi; India’s role as host directs

attention to the role of developing countries.

2. Background on the Kyoto Protocol

The first world conference concerning global cli-

mate change was held in 1979. In the late 1980s the

United Nations (UN) set up the Intergovernmental

Panel on Climate Change (IPCC), which released its

first assessment report in 1990. The first landmark on

international negotiations related to global climate

change occurred in June 1992, at Rio de Janeiro, where

154 countries signed a UN Framework on Climate

Change. Subsequently, a series of ongoing meetings

were held to achieve an agreement to reduce green-

house gas emissions. On December 11, 1997, delegates

from 160 nations reached agreement on the Kyoto

Protocol to the Framework on Climate Change.1

The Protocol establishes binding commitments for

industrialized countries to reduce their greenhouses

gas emissions by at least 5% from 1990 levels by the

period 2008 – 2012, while leaving developing

countries with no obligatory commitments. For the

United States, the target is 7% below 1990 emissions.

As many rules and operational details were left

unspecified, negotiations continued through the 6th

Conference of Parties (COP6) in Hague during No-

vember of 2000. Unfortunately the highly anticipated

COP6 did not bring an agreement, and the US

eventually announced in 2001 that it did not intend

to ratify the treaty, citing the costs to the US economy

and the lack of binding limits on the future emissions

of the main developing countries (e.g. China and

India). The remaining parties reached a compromise

on the climate treaty at the resumed COP6 in Bonn

during July of 2001, which is regarded as establishing

the basic outlines for implementing the Kyoto Proto-

col. In the follow-up meetings (COP7 in 2001, and

COP8 in 2002), the parties finalized many of the

operational procedures for the Kyoto Protocol and

paved the way for countries to consider ratification of

the Protocol.2

H. Li et al. / Ecological Economics 48 (2004) 329–343 331

Considerable discussion concerning the Kyoto Pro-

tocol focuses on its feasibility. One of the key issues is

the representation of developing countries, many of

which are potentially large and growing greenhouse

gas producers, and are already given a 10-year grace

period for complying with schedules for phasing out

most ozone-depleting substances such as chlorofluo-

rocarbon and halons.3 Several ethical and political

reasons are offered as justifications for differentiated

treatment. These include transitional economies, small

historical contributions to global warming, and smaller

emissions per capita compared to industrialized

countries. However, given their potential for future

industrial growth, the lack of specific restrictions on

developing countries has been a point of considerable

concern in the US.

In recent US congressional testimony, Thorning

(1999) argues that a key drawback of the Kyoto

Protocol is its ‘‘failure to engage developing nations

in meaningful action,’’ while to meet the Kyoto

target, the US would experience slower wage

growth and a reduction in GDP of 1–4%. Suther-

land (2000) also examines expected changes in US

emissions to meet Kyoto Protocol targets. He

assesses several US market adjustments and draws

a similar conclusion that the growth in economy

would decline up to 5% per year and the energy

prices would increase by approximately 14% per

year. One of President Bush’s (2001) (p. 2) princi-

ple objections to the Kyoto Protocol is that it

exempts ‘‘80% of the world, including major pop-

ulation centers such as China and India, from

compliance’’.4

Wesley and Peterson (1999) explore the ethical

dimensions of US Senate opposition to the Kyoto

Protocol. Specifically, the treaty is perceived as unfair

because industrialized countries would bear a heavier

burden for the abatement targets than developing

countries. In their view a weakness of Kyoto Protocol

is the absence of any timetable for low-income

countries to introduce abatement requirements. Jonas

3 For the detailed statement, see Article 3, Paragraph 6 (p. 4) of

the Framework of the Kyoto Protocol (URL: http://unfccc.int/

resources/docs/convkp/kpeng.pdf).4 For details, see paragraph 2, p. 1 of ‘‘Text of a letter from the

President to Senators Hagel, Helms, Craig, and Roberts’’ (White-

house permanent document, March 14, 2001).

et al. (2000) draw a similar conclusion that the Kyoto

Protocol is inadequate and requires major modifica-

tions including consideration of commitments by

developing countries.

Recent discussions have focused on implementa-

tion issues, the likely effectiveness after US with-

drawal, and possible ways to bring the US back into

serious negotiations. For example, Jaeger (2002) sug-

gests that for the countries that have ratified or are

considering ratifying the Kyoto Protocol, imposing a

greenhouse gas tax might be a possible alternative to

meeting emission reduction commitments. Loschel

and Zhang (2002) note that with the absence of the

US, there might be no real emission reduction in all

remaining Annex I regions. In short, US withdrawal

greatly hampers the potential effectiveness of the

Kyoto Protocol. Arcas (2001) proposes an amended

Kyoto Protocol to have all the nations, including the

US, ‘‘on board’’. This protocol includes a dynamic

procedure on the emission reduction targets, and an

alternative environmental technology transfer from

developed countries to developing countries. Similar-

ly, Carraro et al. (2002) propose a new regime that

enhances cooperation on technological innovation and

diffusion without targets on emissions.

In short, US withdrawal will greatly hamper the

effectiveness of the Kyoto Protocol. However, as of

late 2003, the US Senate has never actually voted on

the Kyoto Protocol. Rather the only vote that has

actually taken place was the Byrd–Hagel Resolution

(1997) (S. RES. 98), which took place before the

Kyoto Protocol was finalized and before the flexibility

mechanism was added. The resolution states explicitly

(p. 2) that the US Senate will not ratify the Kyoto

Protocol without mandatory ‘‘specific scheduled com-

mitments to limit or reduce greenhouse gas emis-

sions’’ for developing countries such as China and

India. The development of our survey treatment in

1998–1999 was roughly based on this resolution.

Thus, in terms of policy relevance, our survey data

provides evidence on US public support for the Kyoto

Protocol, with and without the only condition on the

treaty the US Senate has ever taken a public stance on.

As continuing evidence of importance, several prom-

inent US Senators (John McCain and Joe Lieberman)

have proposed a bill, called the Climate Stewardship

Act of 2003 in January 2003, which calls for a

domestically mandatory, and economy-wide emission

H. Li et al. / Ecological Economics 48 (2004) 329–343332

reductions with flexible trading program.5 Moreover,

in July 2003, they announced their intentions to push

for the first US Senate vote on the Kyoto climate

treaty.6

To be clear, a wide variety of issues will determine

the future of the Kyoto Protocol and possible alterna-

tive future configurations of the treaty. However, in

this analysis we explore US households’ support for

the Kyoto Protocol, and the effect of a single key

change—the addition of binding emission commit-

ments for developing countries.

3. Data

3.1. Design of the survey

This targeted investigation is part of a larger re-

search program. The full database consists of four

national survey samples. In January 2000, the first

sample was collected by a national random digital

dialing telephone survey; and soon after the first

Internet sample was drawn through web-based ques-

tionnaires from a Harris Interactive (HI) panel of

willing respondents. The surveys focused on knowl-

edge and attitudes related to global climate change and

US ratification of the Kyoto Protocol. A second Inter-

net sample using the same instrument was collected in

July 2000, also from the HI panel. In November 2000,

the last Internet sample of data was collected by

Knowledge Networks (KN), which uses Web TV

technology given to the respondents recruited though

random digital dialing. For the study question investi-

gated here, wemake use of the two samples from the HI

panel.

HI’s online interviewing relies on 4.4 million

Internet users, most of whom were recruited when

they signed up for Internet services. Sampling includ-

ed sending email invitations to randomly selected

5 For details, see Summary of Lieberman/McCain Draft

Proposal on Climate Change from Joe Lieberman press office, on

January 8, 2003, available at URL: senate.gov/f lieberman/press/

03/01/2003108655.html. Site accessed September 17, 2003.6 For details, see MSNBC environmental news on ‘‘Senate

Rivals Tackle Bush on Climate’’, available at URL:

www.msnbc.com/newa/946475.asp?0cv =CB10. Site accessed Sep-

tember 17, 2003.

panel members. Each email contains a unique indi-

vidual password to the Harris Poll Online web site

where the respondent completes a questionnaire re-

garding his or her opinions about the US and the

Kyoto Protocol. Reminder emails were sent out 2–4

days later to non-respondents to encourage their

participation. The procedure ended in 13,034 com-

pleted responses of the first HI survey (HI1) in

January 2000 and 11,160 completed responses of

the second HI survey (HI2) in July 2000. The two

HI samples comprise the largest dataset among the

four multi-mode survey samples.

As discussed in more detail in Berrens et al.

(2003), a direct concern with any use of the HI

samples is their response rates: 4.0% for HI1 and

5.5% for HI2.7 This is because of the sampling

properties of the Harris online poll. HI has developed

several weighting methods to make the samples more

nationally representative (the detailed discussion on

the weights will be presented in the following sec-

tion). Several recent studies discuss the validity of HI

survey samples.

Using a set of multi-mode surveys quite similar to

our full study design, Chang and Krosnick (2003)

provide statistical tests comparing national Internet

(HI and KN) survey samples and a national telephone

survey sample (random digit dialing) to the Current

Population Survey (CPS) of March 2000. Based on a

comparison of demographic information, they find

that ‘‘none of the (samples’) average deviation was

huge, and sample representativeness was never dra-

matically poor’’ (p. 23), and weighting considerably

‘‘shrunk the demographic deviations from the popu-

lation’’ (p. 25). Although not all types of web-based

surveys should be considered the same (Couper,

2000), Chang and Krosnick (2003) conclude that

Internet-based data collection can be treated as a

feasible way to conduct representative sample sur-

veys.8 In previous analysis of the same multi-modes

dataset used in our current study, Berrens et al. (2003)

draw a similar conclusion about HI and KN samples

after comparing a variety of variables on attitudes and

8 Moreover, the HI panel results showed exceptional prediction

performance overall when they correctly predicted 99% of the 2000

election races. For details, see RFL Communications (2000).

7 The response rate is calculated as the ratio of completed

surveys to email invitations sent.

H. Li et al. / Ecological Economics 48 (2004) 329–343 333

voting intentions. It is argued that the various Internet

samples ‘‘produced relational inferences’’ close to the

national probability-based telephone sample (especial-

ly after weighting for the two HI samples). Thus, we

focus here on the use of the weighted HI samples, with

24,194 total observations.9

The basic survey template included three treat-

ments: mental accounts versus standard reminder,

enhanced information versus basic information, and

MKP against the basic Kyoto Protocol (BKP).10 The

particular focus on this study is on the split-sample

Kyoto Protocol treatment, which was implemented

only in the two HI samples.11 MKP refers to a

hypothetical single modification to the Kyoto Proto-

col, which adds mandatory emission limits for devel-

oping countries, while BKP refers to the current

Kyoto Protocol. In the HI samples, about half of the

randomly selected respondents were given a referen-

dum question on the BKP, while the rest were asked a

referendum question on MKP.

The survey instrument consists of three major

sections. The first part contains demographic, attitu-

dinal, and knowledge questions. The second part, or

valuation section, implements the experimental design

treatments and presents the CV scenario, the actual

advisory referendum valuation question for Senate

9 We also compared our HI sample data for a variety of

demographic variables with 1990 US census data. For example, we

find that after weighting, the median annual household income

($37,500) is close to census data ($39,284), the percentage of male

population (48%) is close to census data (48.4%), and the

percentage of some college population (48.7%) exactly matches

the census data (48.7%). An exception is that the median age (44),

even after weighting, is higher than the census data (32.8). The HI

survey respondents are adult Internet users, so their age is

unsurprisingly higher. Also, we construct the flow income variable

using a corresponding categorical index; $37,500 is the mean value

between $35,000 and 39,999, and this range includes the census

value. All of median or mean values of the selected demographic

variables for the raw HI data (unweighted) are considerably off the

census data. For example, the median annual income is $47,500;

and the mean percentage of some college population is 84.5%.10 For an examination of the use of mental accounts in a CV

study see Batemen and Langford (1997). Samples et al. (1986)

conducted an early investigation into CV information effects.11 Full descriptive statistics for all four national survey samples

are available upon request; and see the comparisons and discussions

in Berrens et al. (2003). For detailed discussions on the information

treatment effect, see Berrens et al. (2004). The beta version of the

survey is available on-line at: http://www.unm.edu/instpp/gcc/.

ratification of the Kyoto Protocol (basic or modified),

and a number of follow-up questions. The final

section contains questions about the credibility/fair-

ness of making public policy decisions on the basis of

WTP, political attitudes and participation, and addi-

tional demographic questions.

The implementation of the Kyoto Protocol is the

policy being valued. The payment vehicle is higher

prices for energy and gasoline. The valuation question

for the BKP uses an advisory referendum with a

dichotomous choice elicitation format as follows:

Suppose that a national vote or referendum were

held today in which US residents could vote to

advise their senators whether to support or oppose

ratifying the Kyoto Protocol. If US compliance

with the treaty would cost your household [t]

dollars per year in increased energy and gasoline

prices, would you vote for or against having your

Senators support ratification of the Kyoto Proto-

col? Keep in mind that [t] dollars spent on

increased energy and gasoline prices could not be

spent on other things, such as other household

expenses, charities, groceries or car payments.

For—Against—Do not know/No answer—.

The payment amounts [$t] were randomly assigned

across the sample from the set {6, 125, 300, 500, 700,

900, 1200, 1800, 2400}. The structure of payment

amounts was based on pre-test results, and followed

the general design suggestions of Kanninen (1995).12

The valuation question for the MKP treatment,

offered to a randomly selected sub-sample, reads as

follows:

An alternative to the Kyoto Protocol, which we

will refer to as the MKP, would make only one

change in the agreement. It would require that

developing countries, such as China, India, Mex-

12 Selection of the set of payment amounts was loosely based

on the 1997 national telephone pre-test, and an attempt to follow the

general design suggestions of Kanninen (1995) to avoid putting too

much weight in the tails of the probability of acceptance

distribution. Acceptance rates at different payment amounts were

tracked for the 1998 national telephone survey through the first

several hundred observations, and the two high payments were

added. The same payment amounts were then used for all three

national Internet surveys.

H. Li et al. / Ecological Economics 48 (2004) 329–343334

ico, Brazil, and Argentina, promise to restrict their

future production of ‘‘greenhouse gases’’ to no

more than 5% above current levels.13

Then, these selected respondents are asked an

advisory referendum question parallel to previous

one, using the same set of random payment amounts

[$t]:

Suppose that a national vote or referendum were

held today in which US residents could vote to

advise their senators whether to support or oppose

ratifying the MKP. If US compliance with the

treaty would cost your household [t] dollars per

year in increased energy and gasoline prices, would

you vote for or against having your Senators

support ratification of the MKP?

For—Against—Do not know/No answer—.

3.2. Weighting

As previously noted, the two HI samples are not

probability-based; they originate from the recruited

panel that is not constructed through random sam-

pling.14 Thus, HI applied two sophisticated weighting

approaches, raking ratio adjustment for the first HI

sample (HI1) and propensity weighting for the second

HI sample (HI2), to extrapolate from the panel sam-

ples to the US population and mitigate self-selection

bias and panel effects.

The raking ratio adjustment method is a post-

stratified weighting algorithm, and its estimate is

performed iteratively and proportionally to control

two or more known marginal population totals.15

The method has the advantage that the estimates do

13 While there has been considerable concern that future

emissions increases in the developing countries might swamp any

achievements of the Kyoto Protocol, there has never been any

international consensus concerning possible restrictions on green-

house gas emissions for developing countries. When developing our

survey we needed a specific scenario for the policy change; we

chose the 5% increase to represent a constraint on developing

country greenhouse gas emissions that allowed some limited growth

in emissions.14 See Couper (2000) for a review of web-based Internet

surveys and a specific discussion of the HI panel.15 The raking procedure is first ascribed to Deming and

Stephan (1940). Oh and Scheuren (1987) provide a detailed

description.

not depend on the order that the variables are imputed.

The samples that use raking ratio adjustment are

expected to be more nationally representative, but

the method only corrects for sample selection bias to

the extent that the selection is due to the observed

characteristics being used to construct the weights.

The propensity weighting method is an algorithm

to adjust Internet samples based on propensity scores,

or probabilities.16 To employ the method, two sets of

data, incorporating attitudinal, behavioral, and/or de-

mographic questions, need to be collected: an Internet

sample and a random digital dialing telephone data.

Then the data are combined and the propensity score

is calculated by finding the conditional probability of

being in one sample rather than the other.17 Since this

method includes some attitude and belief variables,

the weights are more likely to be correlated with

unobservable factors that affect sample selection bias.

So propensity weighting is expected to more effec-

tively correct sample selection bias, and is considered

the ‘‘state of the art’’ alternative. HI uses propensity

weighting to ensure that sample characteristics from

its online panel surveys reflect the general population.

They employ this method by using off-line informa-

tion from Census estimates and other surveys to adjust

the composition of the on-line panel by demographic,

attitudinal and behavioral factors.18

For the HI1 sample, HI employed a raking weight

to match 32 demographic marginal totals including

age, region and gender. For the HI2 sample, HI

calculated propensity score series of composite fac-

tors including three attitudinal and three behavioral

questions.19

In our study, the WTP modeling employs weighted

maximum likelihood estimation, using raking weights

for the HI1 sample and propensity weights for the HI2

sample.

18 For details, visit the HI panel website at: http://

www.harrisinterative.com.19 The attitudinal questions include: whether Washington was

in touch with the rest of the country, personal efficacy, and

information overload. The behavioral questions include: whether the

respondent had read a book, traveled, or participated in a sport over

the last month.

16 The propensity score method was first developed by

Rosenbaum and Rubin (1983).17 For details, see Berrens et al. (2003).

H. Li et al. / Ecological Economics 48 (2004) 329–343 335

4. Modeling considerations and hypotheses

4.1. Recoding voting response for uncertainty

Respondents’ certainty levels in the referendum

voting decision may differ greatly because of differ-

ences in the strength of their preferences, understand-

ing of the Kyoto Protocol, etc. We follow the general

approaches of Champ and Bishop (1997); Loomis and

Ekstrand (1998), and develop recoded voting responses

based on individuals’ responses to a follow-up question

concerning the certainty level of their answer to the

referendum question. That is, after respondents an-

swered whether they would vote for or against having

their senators support ratification of the Kyoto Proto-

col, a follow-up question was offered as to how certain

their voting decision was on a scale from 0 to 100, with

0 meaning absolutely certain of voting against it and

100 meaning absolutely certain of voting for it. For

analysis with recoded responses, the Yes votes with a

50 score and above were treated as Yes votes in this

study, while the Yes answers with certainty levels lower

than 50 were converted to No responses.20

4.2. WTP modeling

The full set of WTP models we estimate include

both weighted and unweighted data, and raw and

recoded voting responses. Our modeling approach

closely follows the conventional referendum CV

model of Cameron and James (1987) to directly

estimate a household WTP function. We begin with

the underlying WTP function:

WTPi ¼ f ðxi; b; r; eiÞ ¼ ebVxiþrei ð1Þ

20 There is no standardized approach for the design of follow-

up uncertainty questions in CV studies. Approaches include

separate continuous scales (0–10 or 0–100) for Yes or No answers,

and simple uncertainty categories (Highly Certain, Highly Uncer-

tain, etc). We have found that the follow-up question design we use

here (single 0–100 scale where uncertainty is centered around 50)

produces results quite similar to the separate continuous scales

approach for the Yes and No responses. Specifically, as found in

Berrens et al. (2002), there is a pattern that No tends to mean No

and Yes sometimes means Maybe; i.e. respondents are more likely

to give uncertain Yes responses. In our case, for the Yes votes, there

are 12% of respondents with certainty level of 10 or below, while

for the No votes, there are only 2% of respondents with a certainty

level above 90.

where, xi is a vector of the selected explanatory

variables of respondent i, b is the estimated coefficient

of corresponding explanatory variables, r is a vari-

ance parameter, and ei is a random error component

with mean zero.

Since we cannot detect WTP responses directly in

the referendum format, the latent function of individ-

ual’s true WTP can be observed by a discrete indicator

variable Wi, where

Wi ¼ 1 if WTPizti; Wi ¼ 0 otherwise ð2Þ

and ti is the payment amount that respondent i was

randomly assigned. Thus, the probability of a Yes

response is:

PrðWi ¼ 1Þ ¼ PrðWTPi > tiÞ ¼ 1� Uððti � bVxiÞ=rÞð3Þ

Based on the assumption of a log-normal distribu-

tion of the error term ei, a probit model is employed,21

and the log-likelihood function is:

logL ¼X

fWilogð½1� UððlogðtiÞ � bVxiÞ=rÞ�

þ ð1�WiÞlog½UððlogðtiÞ � bVxiÞ=rÞ�g ð4Þ

For the estimated models, goodness-of-fit is mea-

sured by McFadden’s likelihood ratio index (LRI)

(Green, 2000).

For the models using recoded voting responses, we

construct another index variable WiV. The respondent

must answer Yes to the referendum question (Wi= 1)

and provide a follow-up certainty level (Ci) which is

greater than the threshold certainty value of 50 in

order for WiV= 1. Then, parallel to the DC model

using the raw voting responses, the individual’s WTP

must be inferred through the recoded indicator WiV:

WiV ¼ 1 if WTPz1 and Ciz50;

WiV ¼ 0 otherwise: ð5Þ

21 Based upon concerns about the exclusion of negative WTP

values, we test normal distribution models as well as log-normal

ones. We find no significant negative median WTP values across a

full set of models, and the log-normal models generally have better

fits. Moreover, the signs and significance of estimated coefficients

are basically consistent across the distributions. Thus, we apply Eq.

(4) in our maximum likelihood estimation. Results from the full set

of normal models are available upon request.

H. Li et al. / Ecological Economics 48 (2004) 329–343336

Given the same distribution of the error term, the

log-likelihood function of the recoded model is the

same as in (4).

Finally, because mean WTP estimates can be very

sensitive to the outliers and the distribution of as-

sumption, we focus on the more robust median WTP,

where median WTPi = expbVxi, and xi is replaced with

its mean, x . The standard error of median WTP is

calculated using the delta method (Green, 2000).

4.3. Explanatory variables

A set of attitudinal and socioeconomic variables

are available to explain WTP responses. Detailed

definitions and the associated sample statistics are

provided in Table 1. The WTP models include a

Table 1

Variable definition and descriptive statistics

Variable Definition

EDUC Education level indicator variable, 1–7 scale: 1, less

7, completed some or all graduate school

AGE Respondents’ age, scaled by 100

MALE Respondents’ gender: 1, male; 0, female

BRINK Respondent’s attitude towards the relationship betwee

civilization, scaled by 0–10: 0, no real threat to civili

brink of collapse due to environmental threats

KNOW Familiarity with Kyoto Protocol? 0–10 scale: 0, not a

INTRTY Respondent’s view on international treaties as a way

0–10 scale: with 0, vary bad idea; 10, very good ide

CONFID Respondent assessment of the effectiveness of the Ky

no effect; 10, certain to reduce global warming

GRHOUSE Whether the respondent believes the greenhouse gase

to rise: 0, no; 1, yes

SCICERT Respondents’ perception of scientists’ certainty about

temperature to rise 0–10 scale, with 0, not at all certa

FAIR Perceived fairness of the Kyoto Protocol? 0–1 scale,

completely fair

IDEO Political ideology index, 1–7 scale, with 1, strongly l

MEMBER Indicator variable, if respondent is a member of envir

LNLINC The logarithm of annual household income (in $1000

PAY The random assigned payment amount, from $6 to 24

MA Indicator variable of mental account treatment, where

allocating monthly household budget to several menta

H2 Indicator variable of the second HI sample, with 1, H

D-MKP Indicator variable of MKP treatment: 1, MKP; 0, BK

D-MKP*CON Interaction term between modified Kyoto and CONFI

D-MKP and CONFID

Vote Dummy variable indicating respondent’s voting for Se

1, Yes; 0, No

Descriptive statistics are based on the 24,194 observations, with the exc

19,563 observations, respectively.

number of standard socioeconomic variables, such

as respondent’s education level (EDUC), gender

(MALE), environmental group membership (MEM-

BER), and household income. Additional explanatory

variables can be grouped by three aspects. We include

two attitudinal variables in our model: the respon-

dent’s attitude toward international treaties (INTRTY)

and environmental problems (BRINK). We include

three knowledge variables: respondents’ knowledge

about greenhouse gases (GRHOUSE and SCICERT)

and the Kyoto Protocol (KNOW). Finally, we include

two assessment variables as perceived by respondents:

the effectiveness (CONFID) and the fairness (FAIR)

of Kyoto Protocol.

We are particularly interested in two explanatory

variables: INTRTY and CONFID. First, since the

Mean Standard

deviation

than high school; 5, college graduates; 4.58 1.25

0.42 0.13

0.50 0.50

n environmental threats and human

zation; 10, human civilization is on the

5.79 2.25

t all familiar; 10, completely familiar 2.03 2.58

to handle environmental problems?

a

6.89 2.99

oto Protocol, 0–1 scale with 0, certain of 5.60 2.73

s cause average global temperatures 0.76 0.42

greenhouse gases causing global

in; 10, completely certain

7.03 2.51

with 0, completely unfair; 10, 4.96 2.98

iberal; 7, strongly conservative 4.13 1.63

onmental group: 1, yes; 0, no 0.16 0.36

) 3.74 0.76

00, scaled by $100 6.41 7.31

respondent is asked two questions

l accounts: 1, yes; 0, no

0.49 0.50

I2; 0, HI1 0.49 0.50

P 0.46 0.50

D, derived from the multiplication of 2.79 3.41

nate ratification of Kyoto Protocol: 0.56 0.50

eption of the variables IDEO and LNINC, which have 21,972 and

H. Li et al. / Ecological Economics 48 (2004) 329–343 337

Kyoto Protocol requires international cooperation,

the more the respondent believes in the role of

international treaties in addressing global environ-

mental problems (INTRTY), the more likely he or

she would support the Kyoto Protocol generally.

Second, it is our expectation that the more confi-

dence a respondent has in its eventual effectiveness

(CONFID), the greater the likelihood he or she

would favor the treaty.

4.4. Hypotheses tests

Prior to the WTP modeling, we investigate whether

the BKP versus MKP treatment (i.e. absence or

presence of the MKP) is significantly related with

respondent’s voting decision. We test hypothesis H1

against the null hypothesis of no effect:

H1: BKP versus MKP treatment is significantly

related with respondent’s voting decision.

We apply the Mantel–Hanzael v2-test, and we

expect H1 to be accepted.

Then based on our WTP modeling strategies, we

test the following hypotheses across the various

models incorporating both raw and recoded res-

ponses, and using the weighted data. We hypothesize

that the variable INTRTY, i.e. the belief in the role of

international treaties, will be a significant positive

determinant of voting responses. Letting bINTRTY be

the estimated coefficient on INTRTY, we are inter-

ested in testing the null hypothesis (H0: bINTRTY= 0)

against,

H2: hINTRTY>0.

We expect to accept the alternative hypothesis

(H2), and to find that the variable INTRTY will have

a positive effect on the probability of voting Yes on

the advisory referendum.

Third, we hypothesize that the variable CONFID,

i.e. belief in the Kyoto Protocol’s effectiveness, will

be a significantly positive determinant of voting

responses. Letting bCONFID be the estimated coeffi-

cient on CONFID, we are interested in testing the null

hypothesis (H0: bCONFID = 0) against,

H3: hCONFID>0.

We expect the null to be rejected and that the

variable CONFID will have a positive effect on the

probability of voting Yes on the advisory referendum.

Additionally, since our principle goal is to explore

the relationship between US household support for the

Kyoto Protocol (basic vs. modified), we want to test

this treatment effect directly. Thus, we expect

respondents to treat the BKP significantly different

than the MKP, i.e. we are expecting a latent structural

break between respondents’ voting responses between

the BKP and the MKP treatment. We test the null

hypothesis that the coefficients from the BKP and the

MKP regressions are the same, i.e. biBKP = bi

MKP,

against:

H4: BKP and MKP coefficients are not equal,

biBKP p bi

MKP.

To examine whether the MKP treatment influences

a respondent’s voting decision, a likelihood ratio test

is applied. We expect the null to be rejected.

Finally, of interest is whether these hypothesized

effects will translate into significant differences in

median WTP estimates. Let WTPBKP and WTPMKP

be the median WTP under the BKP and MKP treat-

ments, respectively. Then the last hypothesis (against

the null hypothesis of no difference) we test is,

H5: WTPMKP>WTPBKP.

A one-tailed, independent samples t-test is

employed to check if median WTPMKP is signifi-

cantly greater than median WTPBKP. We expect to

accept H5.

5. Statistical results

Before moving on to the WTP modeling results,

we investigate whether there exists overall differences

in proportions voting Yes between the BKP and the

MKP treatments. To begin, we note that the payment

amount was randomly assigned varying from $6 to

2400, and is expected to be inversely related to the

proportion of Yes votes. This expectation was verified

by the full sample probability distribution of Yes

responses across different payment amounts: as the

payment amount increases, the percentage of Yes

responses decreases. For instance, the percentage of

Table 2

Summary of the probability of Yes votes

Payment Raw data Recoded data

amount ($)Probability of Yes

votes for BKP

Probability Of Yes

votes for MKP

Probability of Yes

votes (total)

Probability of Yes

votes for BKP

Probability of Yes

votes for MKP

Probability of Yes

votes (total)

6 76.62 77.13 76.87 66.63 68.60 81.32

12 70.61 74.92 72.50 62.80 67.63 76.08

25 68.79 72.99 70.98 61.76 62.64 72.51

75 62.47 68.45 65.36 55.97 58.70 60.69

150 62.07 59.24 60.71 53.32 49.35 51.14

225 56.41 59.21 57.86 47.83 50.59 49.78

300 55.68 55.54 55.61 47.05 45.34 44.35

500 51.00 51.11 51.06 41.69 42.84 39.65

700 46.17 47.33 46.73 39.33 37.44 33.80

900 44.62 46.33 45.45 37.24 36.64 32.29

1200 46.07 45.39 45.73 38.74 36.69 33.49

1800 40.14 40.76 40.46 34.17 32.39 28.82

2400 39.03 37.38 38.23 32.70 29.82 26.06

Summary of

total votes

55.31 56.61 55.95 47.60 47.62 47.61

In the recoded data, a Yes vote is treated as a real Yes vote only when respondent’s certainty level is not less than 50%.

H. Li et al. / Ecological Economics 48 (2004) 329–343338

Yes votes to $6 (76.87%) was much higher than that

to $300 (55.61%) and was about twice that to $2400

(38.23%). Further, and not surprisingly, as a conser-

vative measurement taking account of certainty, the

recoded percentage of Yes responses is lower than raw

percentage for each level of payment amount. For

example, the percentage of recoded Yes responses to

$150 (51.14%) was approximately 10% lower than

that of raw data Yes responses (60.71%).

The detailed analysis of the proportion of Yes votes,

split by treatment and with and without recoding, is

given in Table 2. In order to examine whether the

treatment affected voting responses across payment

amounts, we post-stratified the 29,194 observations

into the BKP and MKP samples. The Mantel–Haens-

zel v2 test has a value of 4.159, which is significant at

the 5% level, and the confidence interval of the odds

ratio is mildly less than 1. Therefore, there is a

significant association between the BKP versus MKP

treatment and voting responses.22 Specifically, on

22 Although the total mean proportions of Yes votes are quite

close (see Table 2), this is not consistent across the 13 different

payment amounts, and thus can be deceiving. At lower payment

amounts the MKP sample tended to show a higher proportion of Yes

votes. The Mantel–Haenszel statistic is used for evaluating the

overall association between group and responses, adjusting for the

stratification factor (see Landis et al., 1998).

average the respondents facing the MKP treatment

were more likely to vote Yes. Results support

hypothesis H1. As such, it seems reasonable to

explore the effect of the Kyoto Protocol treatment

in the WTP modeling, while controlling for other

factors.

Further, as clearly seen in Table 2, we have a

‘‘fat-tail’’ problem in our data (i.e. the presence of

high levels of Yes responses even to the highest

payment amounts). There are two possible reasons

for this: (1) people simply have a tendency to

support the referendum regardless of the payment

amount, or yea-saying (Blamey et al., 1999;

Michelle and Carson, 1989); and (2) the payment

amounts were not high enough to pull the tail of

the acceptance rate down. There are several sug-

gested statistical approaches to handling this prob-

lem (e.g. Li and Mattson, 1995; Ready and Hu,

1995). However, for our purposes (testing the

treatment effect), we use the convenient approach

of focusing on the more robust median WTP

measure. In contrast to the mean WTP measure,

which is highly sensitive, the median WTP measure

is largely unaffected by large acceptance rates in

the upper tail of the distribution. Also, a number of

sources have argued that median WTP is an appro-

priate measure for welfare changes in various study

H. Li et al. / Ecological Economics 48 (2004) 329–343 339

contexts; e.g. given our referendum context it is

closer to the value of the majority’s WTP.23

We estimate six weighted-probit models using

various combinations of raw and recoded data.24 For

each combination, we run the BKP and MKP treat-

ment samples separately and then follow with a

pooled model, with a dummy variable (D-MKP) to

indicate the treatment. The results of these estimations

are presented in Table 3. Models 1 and 2 estimate the

BKP and MKP treatments separately using the raw

data. Model 3 uses raw data to estimate pooled BKP

and MKP treatments. Models 4 through 6 are equiv-

alent to models 1 through 3, but use recoded data.

Summary statistics for each model are presented in

the bottom section of Table 3. The v2 (bslopes = 0)statistics are all significant at 0.01 level, which indi-

cate that the null hypothesis that all the coefficients

are jointly equal to zero is not accepted. LRI values

range from 0.24 to 0.30.

We first examine the pooled models (models 3 and

6). In terms of explanatory variables, the estimated

coefficient on the role of international treaty

(INTRTY) is always significant (0.01 level) across

all models. The positive sign indicates that households

more favorably disposed towards international treaties

for dealing with environmental problems were likely

to have a higher annual WTP. Our results thus support

hypothesis H2; INTRTY has a significant positive

effect on WTP.

For the pooled models, the estimated coefficient on

the CONFID variable is positive and significant (0.01

level). Recognizing that the interaction (D-

MKP*CONFID) term between CONFID and D-

MKP has a significant negative sign, we calculated

the marginal effect of CONFID, and found a value

greater than zero for all models (for example, 0.17 for

23 Both mean and median WTP are commonly used to measure

welfare changes in CV studies, and there is considerable historical

discussion on which measure should be used. Imber et al. (1993)

suggest that in practice the median is often the preferred measure.

For example, Harrison and Kristrom (1996) argue that following the

simple majority rule, the median WTP can represent the minimal

aggregate WTP for the voting population in a referendum context.24 We also estimate a set of matched models without using the

weights, and the qualitative conclusions are consistent with the

results from weighted-probit models presented here. Results are

available upon request.

model 3).25 The positive sign of the marginal effect of

CONFID indicates that the more the respondents think

the treaty would reduce global warming, the more

they would pay for ratification of the Kyoto Protocol.

This evidence supports hypothesis H3.

Based on the results from the pooled models, we

find that CONFID has a greater effect on WTP under

BKP than under MKP; the rate of change in WTP

with the increase in confidence under BKP is higher

than MKP. This implies that MKP treatment shrinks

the effect that CONFID had on WTP. We then

calculated the median WTP at each CONFID level

while controlling for the BKP and MKP treatments.

We find that an increase in CONFID affects WTP

under BKP (WTPBKP) more than under MKP

(WTPMKP). When the respondents are not certain

about the effect of the Kyoto Protocol, the modifica-

tion of the treaty will encourage respondents to pay

more. On the other hand, when the CONFID goes up

particularly past its mean value (5.59), i.e. the

respondents are increasingly confident in the BKP,

the modification may seem redundant, and has a

moderate effect.

A variety of other explanatory variables also influ-

ence respondents’ WTP in the pooled models. The

estimated coefficients on the attitudinal variables

BRINK and GRHOUSE are positive and significant

(0.01 level). The more serious respondents think the

environmental problem is, and if they believed that

greenhouse gases cause global temperature to rise, the

higher the WTP. The estimated coefficient on FAIR is

also positive and significant (0.01 level) for both

models. The fairer the respondents consider the Kyoto

Protocol, the more they would be willing to pay to

support it. As far as the socioeconomic characteristics

are concerned, the coefficients on EDUC, MALE,

IDEO, LNINC and MEMBER are all significant

(0.01 level). Thus, being a male, a member of an

environmental group, having a higher education level,

and having a higher household income, all contribute

to greater WTP.

The pooled models (models 3 and 6) offer us the

general outlines of the key factors. The marginal

effect of MKP is positive for model 3, but negative

25 We also checked a variety of interaction terms with D-MKP

and INTRTY, FAIR, BRINK. We only found the interaction term

with CONFID (D-MKP*CONFID) to be significant.

Table 3

Estimations of log-normal WTP models (weighted)

Variables Model 1 BKP Model 2 MKP Model 3 POOL Model 4 RC-BKPa Model 5 RC-MKP Model 6 RC-POOL

INTERCEPT � 10.72*** (� 15.95) � 9.73*** (� 18.44) � 10.85*** (� 24.36) � 10.52*** (� 20.06) � 10.52*** (� 20.06) � 11.33*** (� 24.48)

EDUC 0.08 (1.39) 0.20*** (4.56) 0.17*** (4.84) 0.03 (0.52) 0.27*** (6.01) 0.20*** (5.41)

AGE 0.95** (2.09) � 2.74*** (� 7.30) � 1.20*** (� 4.27) � 1.25*** (� 2.61) � 2.26*** (� 6.24) � 2.00*** (� 6.90)

MALE 0.93*** (6.55) 0.73*** (6.14) 0.79*** (8.73) 1.05*** (6.88) 0.29*** (2.58) 0.61*** (6.72)

BRINK 0.32*** (8.5) 0.19*** (5.84) 0.24*** (9.92) 0.3*** (7.2) 0.20*** (6.26) 0.24*** (9.51)

KNOW 0.06* (1.9) � 0.04 (� 1.50) � 0.01 (� 0.59) � 0.01 (� 0.35) � 0.07*** (� 2.96) � 0.06*** (� 3.18)

INTRTY 0.34*** (10.59) 0.26*** (9.37) 0.29*** (14.11) 0.47*** (12.79) 0.29*** (10.78) 0.38*** (17.11)

CONFID 0.74*** (17.04) 0.39*** (12.91) 0.68*** (20.84) 0.74*** (15.77) 0.49*** (15.98) 0.66*** (19.73)

GRHOUSE 2.13*** (10.44) 2.25*** (13.74) 2.26*** (17.79) 1.51*** (6.79) 1.31*** (8.12) 1.40*** (10.61)

SCICERT � 0.02 (� 0.74) � 0.03 (� 1.33) � 0.04** (� 2.00) 0.08** (2.54) 0.05** (2.07) 0.06*** (3.09)

FAIR 0.21*** (7.63) 0.19*** (8.73) 0.19*** (11.22) 0.16*** (5.56) 0.17*** (7.88) 0.16*** (9.31)

IDEO � 0.40*** (� 8.12) � 0.17*** (� 4.25) � 0.28*** (� 9.06) � 0.37*** (� 7.00) � 0.06* (� 1.66) � 0.20*** (� 6.40)

MEMBER 0.79*** (3.47) 1.56*** (7.53) 1.20*** (7.94) 1.02*** (4.29) 1.19*** (6.40) 1.11*** (7.56)

LNINC 0.37*** (4.4) 1.14*** (14.76) 0.81*** (14.57) 0.49*** (5.33) 0.68*** (9.66) 0.61*** (10.94)

MA � 0.34** (� 2.55) � 0.13 (� 1.14) � 0.24*** (� 2.81) � 0.23 (� 1.56) � 0.28*** (� 2.63) � 0.32*** (� 3.65)

H2 0.02 (0.17) � 0.26** (� 2.30) � 0.09 (� 1.08) � 0.34** (� 2.33) 0.13 (1.15) � 0.05 (� 0.56)

D-MKP – – 1.85*** (7.79) – – 0.63** (2.45)

D-MKP*CON – – � 0.33*** (� 8.72) – – � 0.16*** (� 4.13)

r 4.31*** (26.29) 3.56*** (30.50) 3.89*** (40.58) 4.74*** (25.12) 3.46*** (32.50) 4.02*** (40.80)

N 9152 9032 18,184 9152 9032 18,184

LnL � 4301.39 � 4460.20 � 8867.93 � 4550.90 � 4470.79 � 9091.30

v2 (bslopes = 0) 3720.38 3138.16 6614.74 3275.8 3001.02 6135.34

McFadden’s LRI 0.30 0.25 0.27 0.26 0.24 0.25

Median WTP

(SE-WTP)

300.73*** (25.55) 443.11*** (31.91) 374.79*** (30.44) 92.56** (8.96) 129.10*** (8.85) 113.70*** (6.41)

v2 structure(bBKP vs. bMKP)

390.09*** – 175.03*** –

t comparison

(WTPBKP vs.

WTPMKP)

331.86*** – 276.66*** –

*,**,*** denote the estimate is significantly different from zero at the 0.01, 0.05 and 0.10 levels, respectively. The asymptotic standard error is listed in the parentheses. The v2

statistic =�2½ln Lr � ln Lu�, where ln Lrand ln Lu are the log-likelihood functions evaluated at the restricted and unrestricted estimates, respectively. It is distributed with k degrees of

freedom, where k is the number of restricted parameters. McFadden’s (LRI) = 1� ln Lu=nL0; 6, r is the estimated variance parameter for normalization.a RC denotes recoded models. An observed Yes vote is treated as a real Yes vote when respondent’s certainty level is not less than 50%, otherwise treated as a No vote.

H.Liet

al./Ecologica

lEconomics

48(2004)329–343

340

26 These results are also consistent across all the models with

either recoding for uncertainty or without weighting.

H. Li et al. / Ecological Economics 48 (2004) 329–343 341

for model 6, using the weighted data and recoded

responses. This one result might seem to undermine

the effect of the MKP. However, we note that the

dummy variable, D-MKP, alone in the pooled models

may be inadequate to capture the effects of the MKP

treatment. It forces respondents’ preferences under the

BKP and MKP to be the same (we already found that

CONFID played different roles under the treatments).

Therefore, we separate the sample into the BKP

treatment (models 1 and 4) and MKP treatment

(models 2 and 5) for further analysis.

For all four split treatment models, the estimates of

key explanatory variables such as INTRTY, CONFID,

BRINK, FAIR, GRHOUSE, MEMBER, and LNINC

are consistent in sign, and significant at the 0.01 level,

as in the pooled models. And as discussed previously,

it is not surprising that the coefficients on CONFID

from the BKP models are generally higher than for the

MKP models.

Turning to the predicted WTP results, the values

of median WTP are significantly different from zero

across all the models. Generally speaking, the BKP

models (models 1 and 4) have lower median WTP

than the related MKP models (models 2 and 5), and

the recoded models (models 4–6) always have a

lower median WTP than the related raw data models

(models 1–3). For example, in models 1 and 2 with

raw data, the median WTPBKP is $300, while the

median WTPMKP is $443, respectively. For the

recoded ones (models 4 and 5), the median WTPMKP

drops to $129, compared to a value of $93 for

recoded WTPBKP. As expected, the recoded models

offer lower median WTP estimates, generally 30%

less than the median WTP estimates from the raw

data models.

After reviewing the model results, we use a log-

likelihood ratio test to determine if a structural break

exists between the BKP and MKP treatments. As

indicated in the summary section of Table 3, the v2

statistics are significant (0.01 level). Thus, the MKP

treatment influences respondent’s voting decision

substantially, suggesting that US households treat

BKP and MKP differently when they are considering

supporting Senate ratification of Kyoto Protocol. The

null hypothesis is rejected, and as expected we accept

H3.

Moreover, the median WTP varies from a low of

$93 (model 4) to a high of $443 (model 2). As shown

in Table 3, the t-statistics imply that annual house-

hold median WTPMKP is statistically higher than

median WTPBKP. On average, median WTPMKP is

43% more than median WTPBKP, and ranges from

40% to 47% higher. Thus, the MKP treatment,

incorporating developing country commitments to

reduce greenhouse gases production, would increase

US households’ median WTP.26 Therefore, the evi-

dence supports H5.

6. Discussions and conclusions

The Kyoto Protocol is a complex environmental

agreement requiring extensive global cooperation, and

the question of US participation or withdrawal

remains a pivotal issue. This paper analyzes the

effects of a hypothetical modification of the Kyoto

Protocol on US households’ voting responses on an

advisory referendum to the US Senate. The modified

version of the Kyoto Protocol considers restricting

developing countries greenhouse gases production to

no more than 5% above 1990 levels. The Protocol

remains the centerpiece of international efforts to

reduce global warming, and we believe that our

specific treatment remains a critical question, which

is different than arguing that exactly our version of the

Protocol will ever get considered. But, in terms of

policy relevance, our survey data provides evidence

on US public support for the Kyoto Protocol, with and

without the only condition on the treaty the US Senate

has ever taken a public stance on.

The results from our study suggest that modifica-

tion of the Kyoto Protocol to include restrictions on

greenhouse gas production in developing countries

significantly influences respondents’ support for US

Senate ratification. The results indicate that the treat-

ment significantly increases the probability of a Yes

vote on the advisory referendum. Further, econometric

modeling results provide evidence that the MKP

significantly increases US households’ median WTP

to support the treaty. For weighted likelihood models

using the recoded data, the median annual household

WTP for the US Senate ratification of the MKP is

H. Li et al. / Ecological Economics 48 (2004) 329–343342

40% higher than for ratification of the BKP ($129.10

vs. 92.56). For the weighted models using the raw

referendum data, the percentage increase in annual

median WTP is even higher (47%) under the MKP

versus the BKP ($443.11 vs. 300.73).

We close our discussion by emphasizing two

caveats. First, our two data samples come from HI

large panel Internet surveys, which are non-proba-

bility based. Although our study uses weighted

models, the weighted samples still cannot perfectly

match the US census data. As such our results

should be treated with considerable caution in terms

of national representativeness.

Second, concerning the fat-tail problem with our

referendum data, we employ the convenient approach

of focusing our estimation results on median WTP;

this measure is resistant to the outliers of the distri-

bution (and thus the distributional assumption), and

tends to provide a conservative measure of central

tendency. Further, using the recoded data (for uncer-

tainty) helps reduce but does not eliminate the fat-tail

problem. We would note that there are suggested

alternative approaches to handling the fat-tail problem

and providing mean WTP estimates; e.g. one alterna-

tive is using a pinched-logit model (Ready and Hu,

1995). However, because our WTP models use a

fairly rich set of explanatory variables, we encoun-

tered convergence problems when attempting to esti-

mate pinched-logit and truncated probit models using

the same data samples. We leave further explorations

for future research.

Acknowledgements

This research was funded by a grant from the

National Science Foundation’s Behavior, Risk and

Decision Making program (Grant #9818108). We

thank Harris Interactive for providing the survey

samples. The authors are solely responsible for all

errors and opinions.

References

Arcas, R.L., 2001. Kyoto Protocol: an adequate agreement? Euro-

pean Environmental Law Review 10 (10), 282–294.

Batemen, I., Langford, I., 1997. Budget constraint, temporal, and

question-ordering effects in contingent valuation studies. Envi-

ronment and Planning A 29, 1215–1228.

Berrens, R., Bohara, A., Jenkins-Smith, H., Silva, C., 2002. Further

investigation of voluntary contribution contingent valuation:

fair share, time of contribution and respondent uncertainty.

Journal of Environmental Economics and Management 44 (1),

144–168.

Berrens, R., Bohara, A., Jenkins-Smith, H., Silva, C., Weimer, D.,

2003. The advent of Internet surveys for political research: a

comparison of telephone and Internet samples. Political Analy-

sis 11, 1–22.

Berrens, R., Bohara, A., Jenkins-Smith, H., Silva, C., Weimer, D.,

2004. Information and effort in contingent valuation surveys:

application to global climate change using national Internet

samples. Journal of Environmental Economics and Manage-

ment (in press).

Blamey, R.K., Bennett, J.W., Morrison, M.D., 1999. Yea-saying

in contingent valuation surveys. Land Economics 75 (1),

126–141.

Bush, G.W., 2001. Text of a Letter from the President to Senators

Hagel, Helms, Craig, and Roberts. Available at URL: http://

www.whitehouse.gov, March 13, 2001.

Byrd–Hagel Resolution, 1997. One hundred fifth Congress of the

United States, First Session. S. RES. 98, July 25, 1997.

Cameron, T., James, M., 1987. Efficient estimation methods for

‘‘closed-ended’’ contingent valuation surveys. The Review of

Economics and Statistics 69 (2), 269–276.

Carraro, C., Buchner, B., Cersosimo, I., Marchiori, C., 2002. Back

to Kyoto? US Participation and the Linkage between R&D and

Climate Cooperation Center for Economic Policy Research

(Working paper, April).

Champ, P.A., Bishop, R.C., 1997. Using donation mechanisms to

value nonuse benefits from public goods. Journal of Environ-

mental Economics and Management 33, 151–162.

Chang, L., Krosnick, J., 2003. National Survey via RDD Telephone

Interviewing vs. the Internet: Comparing Sample Representa-

tiveness and Response Quality Department of Psychology, Ohio

State University (Manuscript).

Couper, M., 2000. Web surveys: a review of issues and approaches.

Public Opinion Quarterly 64 (4), 464–494.

Deming, W.E., Stephan, F.F., 1940. On a least square adjustment of

a sampled frequency table when the expected marginal totals are

known. Annals of Mathematical Statistics 11, 427–444.

Green, W.H., 2000. Econometric analysis, 4th ed. Prentice-Hall Inc.

Harrison, G.W., Kristrom, B., 1996. On the Interpretation of

Responses to Contingent Valuation Surveys. Current Issues in

Environmental Economics. Manchester University Press.

Imber, D., Stevenson, G., Wilks, L., 1993. A Contingent Valuation

Survey of the Kakadu Conservation Zone, RAC Research Paper,

3. Resource Assessment Commission, Canberra.

Jaeger, W., 2002. Carbon taxation when climate affects productivity.

Land Economics 78 (3), 354–367.

Jonas, M., Obersterner, M., Nilsson, S., 2000. How to go from

today’s Kyoto Protocol to a past-Kyoto future that adheres to

the principles of full carbon accounting and global-scale verifi-

cation? Interim report, November.

Kanninen, B., 1995. Bias in discrete response contingent valuation.

H. Li et al. / Ecological Economics 48 (2004) 329–343 343

Journal of Environmental Economics and Management 28 (1),

114–125.

Landis, J., Sharp, T., Kuritz, S., Koch, G., 1998. Mantel–Haenszel

methods. Encyclopedia of Biostatistics 3, 2378–2391.

Li, C., Mattson, L., 1995. Discrete choice under preference uncer-

tainty: an improved structural model for contingent valuation.

Journal of Environmental Economics and Management 28 (2),

256–269.

Loomis, J., Ekstrand, E., 1998. Alternative approaches for incorpo-

rating respondent uncertainty when estimating willingness to

pay: the case of the Mexican spotted owl. Ecological Economics

27, 29–41.

Loschel, A., Zhang, Z., 2002. The economic and environmental

implications of the US repudiation of the Kyoto Protocol and

the subsequent deals in Bonn and Marrakech. Paper presented at

the internal workshop on climate policy in Asia, Tokyo.

Michelle, R., Carson, R.T., 1989. Using Surveys to Value Public

Goods: The Contingent Valuation Method Resources for the

Future, Washington, DC.

Oh, H.L., Scheuren, F.J., 1987. Modified raking ratio estimation.

Survey Methodology 13, 209–219.

Ready, R.C., Hu, D., 1995. Statistical approaches to the fat tail

problem for dichotomous choice contingent valuation. Land

Economics 71 (4), 491–499.

RFL Communications, 2000. Harris Interactive uses election 2000

to prove its online MR efficacy and accuracy, Research Business

Report, November: 1–2.

Rosenbaum, P., Rubin, D., 1983. The central role of propensity

score in observational studies for causal effects. Biometrika 70

(1), 41–45.

Samples, K.C., Dixon, J.A., Gowen, M.M., 1986. Information dis-

closure and endangered species valuation. Land Economics 62

(3), 306–312.

Sutherland, R.J., 2000. Achieving the Kyoto Protocol in the US:

how great are the needed changes? Mitigation and Adaptation

Strategies for Global Change 5, 123–142.

Thorning, M., 1999. Oversight hearing to receive on the economic

impact of the Kyoto Protocol to the Framework convention on

climate change. Paper presented to the US Congress, March.

Wesley, E., Peterson, F., 1999. The ethics of burden-sharing in the

global greenhouse. Journal of Agricultural and Environmental

Ethics 11, 167–196.