A comparison of CVM survey response rates, protests and willingness-to-pay of Native Americans and...
-
Upload
independent -
Category
Documents
-
view
0 -
download
0
Transcript of A comparison of CVM survey response rates, protests and willingness-to-pay of Native Americans and...
ARTICLE IN PRESS
Journal of Forest Economics 13 (2007) 49–71
1104-6899/$ -
doi:10.1016/j
�CorrespoE-mail ad
(J.B. Loomis1Present ad
Street, San B2Present a
Drive, Saska
www.elsevier.de/jfe
A comparison of CVM survey response rates, protests and
willingness-to-pay of Native Americans and general
population for fuels reduction policies
Armando Gonzalez-Cabana,�, John B. Loomisb,Andrea Rodrigueza,1, Hayley Hesselnc,2
aForest Fire Laboratory, USDA Forest Service, Pacific Southwest Research Station, 4955
Canyon Crest Drive, Riverside, CA 92507, USAbDepartment of Agricultural and Resource Economics, Colorado State University, Fort Collins,
CO 80523, USAcUniversity of Montana, School of Forestry, USA
Received 28 October 2005; accepted 4 October 2006
Abstract
A contingent valuation method (CVM) study was used to compare survey response rates,
protest refusals to pay, and median willingness-to-pay (WTP) of Native American
communities in Montana compared to Montana’s general population for two wildland fire
mitigation strategies. Understanding differences in response rates, protest refusals to pay, and
median WTP between Native Americans in the United States and the general population may
shed some light on how well the method may work for indigenous people in other developed
countries such as Canada, Australia and New Zealand. Native Americans survey response
rates were not significantly different from Montana residents for the initial contact (first
see front matter r 2006 Elsevier GmbH. All rights reserved.
.jfe.2006.10.001
nding author. Tel.: +951 680 1525; fax: +951 680 1501.
dresses: [email protected] (A. Gonzalez-Caban), [email protected]
), [email protected] (A. Rodriguez), [email protected] (H. Hesseln).
dress: Research, Analysis, and Vital Statistics, Department of Public Health, 172 W. 3rd
ernardino, CA 92415, USA.
ddress: Department of Agricultural Economics, University of Saskatchewan, 51 Campus
toon, Canada SK S7N 5A8.
ARTICLE IN PRESS
A. Gonzalez-Caban et al. / Journal of Forest Economics 13 (2007) 49–7150
wave), but were significantly different for the follow-up in-depth phone interviews (second
wave). Native Americans protest rate for the prescribed burning program was not statistically
different from Montana residents. Conventionally calculated, protest rates for the mechanical
fuels reduction program are higher for Montana’s residents than Native Americans. Results
from bivariate probit with sample selection models indicate that there is no significant
difference between the Native American and Montana general populations’ median WTP for
either program. This suggests that in Montana, Native Americans and members of the general
population generally yield similar results to CVM questionnaires of forest fire management.
r 2006 Elsevier GmbH. All rights reserved.
JEL classification: Q230
Keywords: Mechanical fuels reduction; Prescribed burning; Probit model; Selection bias
Introduction
Indigenous people throughout the developed world have not always been able tofully participate in their country’s market economies. In particular, Native Americancommunities in the United States have not benefited from the economic boomsustained by the country over the last 100 years. Like many other minority groups,the impact of government programs on indigenous people such as Native Americancommunities has not been studied in depth. In 1994, pressured in part by the socialjustice movement in the United States, President Clinton enacted Executive Order12898 requiring federal agencies to evaluate the social justice of federal actions onminority populations. The Order requires policy makers to understand the impactsof their projects and policies on different cultural groups and to make sure that thesegroups do not shoulder the brunt of possible negative impacts of proposed programsor actions. Agencies frequently use surveys to ascertain the potential impact of theiractions on households. However, Loomis et al. (2002) found that past surveys,including nearly all contingent valuation method (CVM) studies in the US, excludednon-English speakers. They suggested that this could potentially lead tounrepresentative samples that would lead to an erroneous estimate of aggregatebenefits. The same argument might be made for Native American communities aslittle effort has been made to survey these populations.
To the best of the authors’ knowledge, in the US, there have been no publishedcomparisons between Native Americans’ and other US residents’ CVM willingness-to-pay (WTP) responses. However, this is not the case in Canada where mostrecently Haener et al. (2001) and Adamowicz et al. (2004) have done relevant work inassessing aboriginal people values for subsistence hunting in the province ofSaskatchewan. Haener et al. (2001) and Adamowicz et al. (2004) use choiceexperiments techniques that ‘‘ygenerate discrete choice data that can be analyzedusing methods based on random utility theory (Haener et al., 2001: 1336).’’ Haeneret al. (2001) tried to explain differences in preferences between Aboriginal groups inCanada for hunting-site characteristics, among other things. His findings suggest
ARTICLE IN PRESS
A. Gonzalez-Caban et al. / Journal of Forest Economics 13 (2007) 49–71 51
that in addition to hunter’s age that aboriginal status may affect responses tochanges in hunting attributes. They further suggest that these differences are‘‘culturally based’’ though they lack enough information on their culture to supportthis hypothesis. They concluded ‘‘y that the technique [state preference method] canbe used to provide information on values and behavioral responses to landscapechanges (Haener et al., 2001: 1339).’’
Differences in responses across cultures might be expected at any one of severalstages of a CVM design: (1) a scenario description of the problem, (2) one or moreproposed solutions, and (3) an associated payment vehicle for the solution. Eventhough focus groups and pre-tests are used to reduce scenario rejection and increasethe believability of the payment mechanism, different cultural backgrounds can leadto different interpretations of the CVM scenarios and therefore affect WTP. This isparticularly true if there are significant differences in the environmental ethic of thegroups being compared.
Minority cultures may also have a distrust of government.3 The history of brokenpromises and failed treaties between Native Americans and the US Governmentwould certainly contribute to this distrust. In addition, there may be culturaldifferences with respect to one of the two fire mitigation strategies discussed in thisstudy, prescribed burning. Specifically, Native Americans have a historic tradition ofusing prescribed burning to maintain favorable conditions for game animals (Pyne,1982), perhaps making them more likely to support prescribed burning programs. Inlight of Executive Order 12898, the relevancy of these issues to policy makers andnatural resource managers is heightened because of the need to know how differentsegments of society benefit from different environmental policies and publicprograms.
The objective of this study was to determine the level of support among NativeAmerican communities in Montana compared to Montana’s general population fortwo wildland fire mitigation strategies. Differences in WTP between NativeAmericans in Montana and members of Montana’s general population may suggestthe need to add culture-specific material to future CVM surveys. In addition, wewanted to test how well the CVM works in Native American communities relative tothe Montana general population. How well the method works for Native Americansin the United States may also shed some light on how well the method may work forother indigenous people in other developed countries such as Canada, Australia, andNew Zealand. This research may also help policy makers and natural resourcemanagers understand how respondents’ culture may influence their participation andresponse in CVM surveys.
3This is not only true of governments, as Haener et al. (2001) and Adamowicz et al. (2004) found that
trust was an important factor in the data collection process during their work. To reduce this distrust
barrier they hired a person who lived in the communities for a full year and still had to arrange the
interviews through a community resident ‘‘y who socially and culturally had access to the hunters
(Haener et al., 2001: 1336).’’
ARTICLE IN PRESS
A. Gonzalez-Caban et al. / Journal of Forest Economics 13 (2007) 49–7152
Methods
Contingent valuation method
The contingent valuation method uses survey techniques to elicit values fornonmarket goods or services. CVM uses stated preferences by respondents based ona contingent market for the good or service evaluated. This elicitation process isnecessary because there is a lack of observable market forces for services such aswildland fire risk reduction.
Although there are still some concerns about the reliability (whether WTPvariance is due to random sources or ‘‘noise’’) and validity (construct, content andcriterion) of its results (see Carson et al., 1996; Diamond and Hausman, 1994),empirical test–retest studies have demonstrated CVM results to be reliable (Loomis,1989, 1990; Reiling et al., 1990). A ‘‘blue ribbon panel’’ co-chaired by two Nobellaureates have also recommended CVM for measuring passive use values (Arrowet al., 1993). CVM has seen increasing use in valuing wildfire prevention, includingFried et al. (1999) and Loomis et al. (2002).
The blue ribbon panel also recommended that CVM surveys be conducted using adichotomous choice format, such as a voter referendum. In this format theindividual is asked whether they would vote in favor or against a preset bid amountthat varies across the sample. The voter referendum format is thought to simulaterealistic market conditions with which respondents might be most familiar, and isconsidered to be better than other elicitation methods (Arrow et al., 1993).
Cameron and James (1987) view the respondent as comparing the amount they areasked to pay to their personal maximum WTP. They answer Yes if the bid amount isless than their personal maximum WTP and No if the amount is greater than theirpersonal maximum WTP. Cameron’s valuation approach, which is now referred toas a variation function due to its link to WTP measures such as compensating orequivalent variation (McConnell, 1990), attempts to estimate this unobserved WTPusing a probit model (Cameron and James, 1987). The choice of a logit or probitmodel depends on the assumed distribution of the utility difference (Hanemann,1984).
Survey design
Program valued
The principal source of information on fire statistics (annual acres and housesburned) for development of the survey instrument used in Montana was USDAForest Service wildland fire management personnel at the regional (USDA ForestService Northern Region, R1) and local levels. Personnel provided technicalinformation on how prescribed burning and mechanical fuels reduction projects areconducted and the possible environmental effects of each project. They also reviewed
ARTICLE IN PRESS
A. Gonzalez-Caban et al. / Journal of Forest Economics 13 (2007) 49–71 53
the survey instrument to assure the correct representation of the wildland urbaninterface problem in Montana. This last concern was particularly important becausethe survey instrument used in this study was originally designed for a project inFlorida, which evaluated the response of different ethnic groups to three wildlandfire reduction strategies (Loomis et al., 2001, 2002).
The survey instrument received by all participants contained the followinginformation about the prescribed burning program.
‘‘Montana is considering using some state revenue as matching funds to helpfederal, state, tribal and private individuals finance fire prevention programs. If amajority of residents vote to pay the county share of this program, the ExpandedMontana Prescribed Burning Program would be implemented in your county andother counties in Montana on federal, state, tribal and private forest andrangelands.
Funding of the Expanded Montana Prescribed Burning Program would requirethat all users of Montana’s forest and rangelands, such as timber companies,recreation visitors, and Montana households pay the additional cost of thisprogram. If this Expanded Prescribed Burning Program were to be implemented,by law, the money would be deposited in a separate Montana Prescribed BurningFund, which could only be used to carry out the Expanded Prescribed BurningProgram described above. A citizen advisory board would review the expendituresfrom the fund annually.
If the Expanded Prescribed Burning Program was undertaken it is expected toreduce the number of acres of wildfires from the current average of approximately140,000 acres each year to about 105,000 acres, for a 25% reduction. The numberof houses destroyed by wildfires is expected to be reduced from an average of 20 ayear to about 8.
Your share of the Expanded Montana Prescribed Burning Program would costyour household $ a year. If the Expanded Montana Prescribed Burning Programwas on the next ballot would you vote?
1. ___In favor __Against’’
The ‘‘$___ a year’’ was filled in with one of 10 different dollar amounts rangingfrom $10 to $470. These monetary amounts were selected based on past wildfiresurveys elsewhere in the United States (Loomis et al., 2002) and performed well inpre-testing.
When asked to value a mechanical fuels reduction program both samplepopulations were presented with similar information. In particular, householdswere told that the following mechanical fuel reduction program was an alternativeprogram that would result in the same 25% reduction in acres burned and reductionin houses burned from 20 to 8. The key difference from the prescribed burningprogram was the description which was as follows:
‘‘Another approach to reducing the build up of fuels in the forest is to ‘‘mow’’ ormechanically chip the low and medium height trees and bushes into mulch. This isespecially effective at lowering the height of the vegetation, which reduces the
ARTICLE IN PRESS
A. Gonzalez-Caban et al. / Journal of Forest Economics 13 (2007) 49–7154
ability of fire to climb from the ground to the top or crown of the trees. Inaddition, mechanical ‘‘mowing’’ slows the new vegetation growth with the layer ofmulch acting as a barrier.
Mowing or mulching 50,000 acres of forest and rangelands is more expensive thanprescribed burning, due to increased labor and equipment needs. It would alsodecrease the number of ground cover plant species reducing food for wildlife.However, unlike prescribed burning, mulching does not produce any fire smoke.’’Then households were told that the Mechanical Fuel Reduction Program was analternative to the Prescribed Burning Program:
If the Mechanical Fire Fuel Reduction Program was undertaken instead of theExpanded Prescribed Burning Program y
Your share of this Mechanical Fire Fuel Reduction Program would cost yourhousehold $________ a year. If the Mechanical Fuels Reduction program werethe ONLY program on the next ballot would you vote? 1. —In favor 2. —Against
In this case, the ‘‘$___ a year’’ was filled in with one of 10 different dollar amountsranging from $15 to $480, because the mechanical program is more expensive toimplement than the prescribed burning program.
Focus groups
This survey design followed the basic format of one developed through four focusgroups in Florida, and successfully used there (Loomis et al., 2001). In order to refinethe survey for administration to Native Americans in Montana, the survey detailswere tailored to the Montana fire situation and the presence of extensive NativeAmerican lands in Montana. The revised and adapted survey instrument was thentested in two focus groups conducted in February of 2001 to ensure comprehen-sibility among the Native American population. One focus group was conducted atthe Confederated Salish-Kootenai reservation in Pablo, MT, and another at theBlackfeet Reservation in Browning, MT. Both Tribal Councils approved theparticipation of tribal members in this research effort. Each focus group consisted ofabout 18–20 people. There were no problems with interpretation or comprehensionof the information provided on the survey instrument. Following the focus groups acomplete mail booklet and survey script was developed. Because the same surveyinstrument was being used for Montana’s general population, and the original waspre-tested in Florida, we felt there was no need to conduct another pre-test withMontana’s general population. The final survey was an 8-page color booklet.
To assess whether refusals to pay were actually protest responses, any respondentwho declined the initial bid amount was asked if instead they would pay $1. Thosewho declined to pay $1 were asked an open-ended question as to why they voted no.These responses were recorded verbatim by the interviewer and then post-coded bythe author who conducted the focus groups into protest and non-protest refusalsto pay.
ARTICLE IN PRESS
A. Gonzalez-Caban et al. / Journal of Forest Economics 13 (2007) 49–71 55
Sample design and survey mode
The final target sample size was 500 completed interviews. The data collection wasdivided into two groups: (i) Native Americans and (ii) Montana’s generalpopulation. The target sample size for each population was 250 completedinterviews. The final sample size was 499 completed interviews, short by 23interviews with Native Americans and high by 22 interviews with general Montanaresidents (Table 1).
Random digit dialing was used to obtain a representative sample.4 Together withthe interviewer screening protocol, in which the interviewer asked to speak with theperson in the household who was 18 years or older and had the next birthday,random digit dialing should result in a sample balanced by gender. Concerned aboutthe possibility of the lack of phone availability in households on tribal lands, weconferred with tribal officials for both tribes and were assured that nearly all tribalmembers had phones.
The initial contact was used to secure participation from the household and toagree to an appointment for a detailed follow-up interview using a typeset, colorbooklet that would be mailed to them. The booklet contained questions andscenarios about the two different fire management policies (prescribed burning andmechanical fuels treatment), in addition to two illustrations contrasting wildlandfires and prescribed fires. Participants were asked to read the booklet prior to theagreed-upon phone interview. The interviews were conducted in English to obtaindemographic information and answers regarding support for each fire managementpolicy described in the scenarios.
Results and discussion
Montana general population residents and Native American residents weresurveyed starting in spring of 2001 and continuing through spring 2002. The initialresponse rates to the screening protocol were 67.4% for Montana’s generalpopulation, and 73.9% for Native Americans (first wave; Table 1). A chi-squaretest showed no significant differences at the 0.05 level between the two populations’response rates for the screening protocol.
Individuals who were not interviewed because of an incorrect phone number ornot having appropriate respondent qualifications, such as being under 18, were notincluded in the final calculated response rate. Those participants who refused tocomplete the interview or rescheduled without future contact were included in theresponse rate as unit non-responses. Likewise, any individual contacted but notinterviewed was included in the response rate as non-response. The non-responsecategory also included respondents to the initial screener who did not follow throughwith completion of the entire survey process. Accounting for all these factors
4The use of random digit dialing guarantees that nearly all households are eligible for interview. This is
especially important because of unlisted phone numbers. This problem causes difficulties in contacting
these households by mail because they will not appear on commercially available mailing lists.
ARTICLE IN PRESS
Table 1. Survey response rate comparisons
Montana general Native Americans
Total Pct. Total Pct.
First wave
Total initial
sample
602 654
Completed
initial survey
406 67.4 483 73.9
w2 ¼ 1:0117 (pp0:01) Not significant, critical value is 6.635
Second wave
Phone
disconnect,
moved,
deceased, not
available
25 33
Not called by
end of study
0 18
Net second
wave sample
381 432
Completed
second survey
272 71.4 227 52.6
w2 ¼ 7:2301 (pp0:01) Significant, critical value is 6.635
A. Gonzalez-Caban et al. / Journal of Forest Economics 13 (2007) 49–7156
produced the net sample size for the second (full) interview; 381 individuals forMontana’s general population and 432 for Native Americans (second wave,Table 1). Of the 381 individuals for Montana’s general population 272 completedthe full interview for a response rate of 71.4 (272/381)%. The Native Americansresponse rate was 52.6% (227/432) (second wave; Table 1). A chi-square test showedthat the response rates of the net samples for these populations are significantlydifferent at the 0.01 level (second wave; Table 1). One possible explanation for thelower response rate among Native Americans could be that they avoided answeringthe survey if they opposed the programs. That is, rather than saying no to ordisagreeing with the programs they simply abstained from participation.5
Analysis of protest responses
During each phone contact, the interviewer recorded open-ended statements of thereason why any respondent voted no to paying the stepped-down minimum bid of
5This hypothesis was brought to our attention by Phil Meyer, Resource Consultant, Meyer Resources,
Mtechosin, British Columbia, Canada during the 2nd World Resource Economics Congress in Monterey,
CA, June 24, 2002. Mr. Meyer does a great deal of consulting for Native American Tribes in Canada and
the United States.
ARTICLE IN PRESS
A. Gonzalez-Caban et al. / Journal of Forest Economics 13 (2007) 49–71 57
$1. A ‘‘no’’ response for reasons other than lack of value for the program or inabilityto afford it was considered a protest vote. The most often cited protest responses forthe prescribed burning program were ‘‘someone else should pay’’ (31.6% of protestresponses), ‘‘I don’t trust the state and/or federal government’’ (10.5%), and‘‘opposed to new taxes’’ (7.9%). Protests to the mechanical fuels reduction programincluded ‘‘ecologically damaging’’ (14.2%), ‘‘prescribed burning is better’’ (13.4%),and ‘‘the program will not work’’ and ‘‘log it first’’ (9.0% each). ‘‘No’’ responseswere tabulated and identified as protests or non-protests for each fuels treatmentprogram for both groups of interest (see Appendix A for a full detail of all reasonsgiven for NO responses). Both Native Americans and members of Montana’sgeneral population had higher rates of protest for the mechanical fuels reductionprogram.
To test whether there is a difference in protest rates for general Montanahouseholds and Native Americans, protest responses were aggregated to acontingency table (Table 2). Protest rates for the prescribed burning program were7.4% for Native Americans, and 11.3% for Montana residents. A chi-square testindicates there is no statistical difference between these protest rates at a 0.01 level.Protest rates for the mechanical fuels reduction program were much higher for bothgroups, at 21.9% for Native Americans, and 33.1% for Montana residents, asignificant difference at the 0.01 level. Because many more Native Americanparticipants than MT general population dropped out of the full survey we addedthem back as protest responses in the chi-square test for the mechanical program.For a valid comparison we also added the MT general population to their non-response population. This chi-square test showed no statistical difference between
Table 2. Protest response rate comparisons
Fuels treatment program
Prescribed burning Mechanical fuels reduction
Montana
generalaNative
AmericanaMontana
generalaNative
Americana
Protest
responses
24 14 86 48
Non-protest
responses
189 176 174 171
Percent protest 11.3 7.4 33.1 21.9
w2 ¼ 1:8648 (pp0:01) Not
significant, critical value: 6.635
w2 ¼ 7:5114 (pp0:01) Significant,critical value: 6.635
aThe differences between the sum of the first two rows in Table 2 and the final row in Table 1 is due to
some cases missing values for independent variables used in the models. For example, 272 residents from
Montana’s general population completed the survey, but 59 of them had missing values for one or more of
the independent variables used in the prescribed burning model. Twelve of them had missing data for the
independent variables used in the mechanical fuels reduction model. The same explanation holds true for
the Native American sample.
ARTICLE IN PRESS
A. Gonzalez-Caban et al. / Journal of Forest Economics 13 (2007) 49–7158
the two population protest response rates at a 0.01 level using this comparisonprocedure.
Statistical analysis of WTP responses
Bivariate probit regression
Typically, dichotomous choice CVM data are analyzed with the logit model,partly because the underlying logistic distribution allows for more convenientestimation compared to other binary choice models (Greene, 2000). However, theprobit, a binary choice model based on the standard normal distribution, can also beapplied to CVM data. Hanemann’s (1984, 1989) formula for calculating medianWTP is applicable to the probit model as well.
One extension of the probit model is the bivariate probit with sample selection.6
This model incorporates Heckman’s (1979) thoughts on sample selection bias intothe standard bivariate probit, a model with two simultaneously estimated equationsthat allows for correlation between the error terms in each equation. The premise ofHeckman’s sample selection model is that ‘‘using non-randomly selected samples toestimate behavioral relationships’’ results in biased and inconsistent parameterestimates (1979: 153). Because self-selected respondents may differ in somesignificant way from non-respondents, it is important to correct for this bias.Ignoring this issue could lead to inconsistent parameter and WTP estimates, makingthem unfit for generalization to the population.
Following Boyes et al. (1989) and Greene (2000), the bivariate probit with sampleselection model consists of two equations simultaneously estimated with maximumlikelihood: one equation for the binary choice of whether to pay the correspondingbid amount for the fire mitigation program (yi1), and the second equation for thebinary choice to participate in the follow-up interview (yi2). Because binary choiceoutcomes are considered reflections of underlying regressions (Greene, 2000), let *
denote a latent variable and assume that y�i1 and y�i2 follow:
y�i1 ¼ X i1b1 þ �i1,
y�i2 ¼ X i2b2 þ �i2 ð1Þ
for i ¼ 1; 2; . . . ;N, where X ij ; j ¼ 1; 2, are N � kj matrices of independent variables,bj are kj � 1 vectors of coefficients to be estimated, and eij are disturbancesdistributed bivariate normal with zero mean, unit variance and correlationcoefficient r. The variable yi1 takes on a value of 1 if the respondent is willing topay and 0 if not
y1i ¼1 if willing to pay y�i140
� �;
0 if not y�i1 � 0� �
;
((2)
6It was first introduced by van de Ven and van Praag (1981), and has since been used extensively in the
loan default and credit scoring literature (see Boyes et al., 1989, Greene, 1992, Jacobson and Roszbach,
2003).
ARTICLE IN PRESS
A. Gonzalez-Caban et al. / Journal of Forest Economics 13 (2007) 49–71 59
and the variable yi2 takes on a value of 1 if the respondent participated in the follow-up interview and 0 if not
y2i ¼1 if participated in follow� up interview y�i240
� �;
0 if not y�i2 � 0� �
:
((3)
Keeping in mind that yi1 is observed only if yi2 ¼ 1, parameter estimates areobtained by maximizing the following log-likelihood function:
‘ ¼XN
i¼1
yi1yi2 ln f2 X i1b1;X i2b2;r� �
þXN
i¼1
1� yi1
� �yi2 ln f2 �X i1b1;X i2b2;r
� �
þXN
i¼1
1� yi2
� �ln f �X i2b2
� �, ð4Þ
where F2[ � ] and F[ � ] represent the bivariate and univariate standard normalcumulative distribution functions. Simultaneous estimation allows for correlationbetween the two equations, resulting in parameter estimates that are more efficientthan those obtained from estimating the two equations separately. In addition,possible sample selection bias is corrected for, meaning that parameter estimatesderived from the bivariate probit with sample selection are generalizable to thepopulation of interest (Boyes et al., 1989), assuming that the sample selection stageof the probit model reflects a random sample of the population. This is a reasonableassumption because random digit dialing was used for the initial phone calls andfairly high initial call response rates (67–74%) were obtained.
Testing for equality of WTP function of Native Americans and general population
To determine if the general population of Montana has a different WTP functionthan Native Americans living in Montana we tested whether these two groups differby an intercept shifter and bid slope interaction term. A statistically significant bidslope interaction term would indicate a potentially significant difference in WTP, asthe bid slope has a direct influence on WTP. Regression results (Tables B1 and B2 inAppendix B) showed that the Montana general populations’ intercept (MtGenPop)and bid slope interaction variables (BIDRX_MTGEN01 and BIDMEC_MTGEN01)were not significantly different from zero for either of the two fuels reductionprograms: prescribed burning or mechanical. This suggests there are no criticaldifferences between these two groups with respect to the key bid variable and there isno independent effect on the WTP function of being Native American that is notcaptured in the other explanatory variables. Thus, the two populations are pooledfor the prescribed burning and mechanical fuels reduction program.
ARTICLE IN PRESS
A. Gonzalez-Caban et al. / Journal of Forest Economics 13 (2007) 49–7160
Prescribed burning
Preliminary probit regressions were used to determine the significant variablesaffecting the WTP for each fuels treatment program. Reduced models for each fuelsreduction program, including only the significant variables (Table 3) from the fullmodel were used so as to improve estimation efficiency and reduce the standarderrors of those variables (Greene 2000: 338). These same reduced models were usedto calculate median WTP figures and confidence intervals around median WTP(Tables 4 and 5). For the prescribed burning program, r is significant at the 0.01level, indicating that sample selection bias is present (Table 4). This means thatrespondents to the second wave (in-depth interview) are in some way systematicallydifferent from non-respondents, and this difference plays a role in how peoplerespond to the prescribed burning WTP question. In particular, second waverespondents were more likely to be older, be Montana general population residents,and less likely to believe that prescribed burning (RX burning) is too dangerous to beused.
The coefficient on the dollar amount asked of each respondent to fund theprescribed burning program (BidRx) is negative and significant at 0.05 (Table 4).RxDanger is negative and significant at the 0.01 level, indicating that those whobelieve prescribed burning is too dangerous are less likely to vote in favor of theprescribed burning program. EvacHome is positive and significant at the 0.05 level,in both the prescribed burning and mechanical fuels reduction programs indicatingthat people who have evacuated their home due to fire are more likely to vote infavor of both fuels reduction programs. ‘‘Retired’’ is positive and significant at the
Table 3. Variable names and definitions
Variable Definition
Variables used in WTP equations
BidMec Bid amount for the mechanical fuels reduction program
BidRx Bid amount for the prescribed burning program
RxRedHi Dummy where 1 ¼ prescribed burning reduces the chance
of high intensity wildfire, 0 otherwise
Variables used in sample selection equations
Age Age of respondent in years
HeardRx Dummy where 1 ¼ read or heard about the use of
prescribed burns or prescribed fires, 0 otherwise
RxDamage Dummy where 1 ¼ if wildfire occurred in an area that had
previously been prescribed burned, damage to houses and
mature trees would be reduced, 0 otherwise
SRxDangr Dummy where 1 ¼ prescribed burning is too dangerous to
be used, 0 otherwise
SRxRedFl Dummy where 1 ¼ prescribed burning effectively reduces
the amount of excess fuels in the forest, 0 otherwise
ARTICLE IN PRESS
Table 4. Results from the prescribed fire program reduced equation probit models
Variable Bivariate with
sample selection
coefficient (Std.
error)
Single equation
coefficient (Std.
error)
Variable mean
Constant 0.0691 0.7490*** 1.0000
(0.2161) (0.1239)
BidRx �0.0011** �0.0014*** 158.1230
(0.0005) (�.0005)
RxDanger �1.0444*** �1.0565*** 0.0421
(0.3081) (0.3678)
EvacHome 0.7948** 0.9406** 0.0680
(0.3787) (0.4213)
Retired 0.5182*** 0.4822** 0.1651
(0.1958) (0.2309)
Sample selection equation
Constant �1.5326*** �1.5931*** 1.0000
(0.3917) (0.4234)
HeardRx 0.2637 0.3995** 0.8543
(0.1612) (0.1723)
SRxRedFl 0.3322* 0.3343 0.9123
(0.1998) (0.2090)
RxDamage 0.2807 0.3345* 0.8891
(0.1825) (0.1892)
SRxDangr �0.3392* �0.2368 0.1573
(0.1898) (0.1872)
Age 0.0114*** 0.0099** 43.4520
(0.0038) (0.0039)
MTGenPop 0.3962*** 0.3611*** 0.4272
(0.1154) (0.1179)
r(1,2) 0.7437*** 0.0000
(0.1676)
Log likelihood �555.3457 �557.9621
N 604 604
Note: *, **, and *** denote statistical significance at the 0.10, 0.05 and 0.01 levels.
A. Gonzalez-Caban et al. / Journal of Forest Economics 13 (2007) 49–71 61
0.01 level (Table 4), indicating that people who are retired are more likely thanpeople who are not retired to vote in favor of the prescribed burning program.
WTP for the prescribed burning was calculated from the probit bivariate model(Table 4). Median WTP for the prescribed burning program, estimated at the samplemeans of the variables included in the reduced bivariate model, is $154.89 (Table 6).This compares to $605 using the single equation probit model without the sampleselection correction. Thus, when there is significant sample selection, correcting for itusing the two equations bivariate probit is quite important.
ARTICLE IN PRESS
Table 5. Results from the mechanical fuels reduction program reduced equation probit
models
Variable Bivariate with
sample selection
coefficient (Std.
error)
Single equation
coefficient (Std.
error)
Variable mean
Constant �0.1522 �0.3241 1.0000
(0.5512) (0.3878)
BidMec �0.0017*** �0.0017*** 167.6895
(0.0005) (0.0005)
RxRedHi 0.6460 0.6551* 0.9422
(0.3984) (0.3573)
EvacHome 0.7885** 0.7905** 0.0614
(0.3256) (0.3458)
OwnHome �0.2885 �0.2586 0.7762
(0.1988) (0.1934)
Income2 0.0054 0.0054* 41.0632
(0.0034) (0.0032)
Note: *, **, and *** denote statistical significance at the 0.10, 0.05 and 0.01 levels.
Table 6. Median WTP estimates from the reduced prescribed burning and mechanical fuels
reduction programs bivariate probit model
Model Median WTP Confidence intervals
95% 90%
Prescribed burning reduction
Meansa $154.89 ($816.90)–$511.81 ($414.92)–$430.26
Reducedb $65.18 ($1,398.40)–$381.93 ($698.43)–$333.28
RxDangerc ($920.13) ($7,413.40)–($210.09) ($4,118.40)–($299.81)
EvacHomed $814.95 $28.68–$4,209.70 $164.26–$2,633.90
Retirede $554.01 $54.55–$2,332.90 $159.45–$1,584.00
Mechanical fuels reduction
Meansf $295.19 ($55.55)–$844.12 $10.98–$689.81
EvacHomed $504.93 ($224.50)–$1,707.50 ($106.97)–$1,387.20
Combinations (i.e. EvacHome and Retired, EvacHome and OwnHome, etc.) would be redundant, as their
frequency in the sample is small.aEvaluated at sample means presented in Table 4.bAll dummy values are zero.cDummy for RxDanger ¼ 1, others zero.dDummy for EvacHome ¼ 1, others zero.eDummy for Retired ¼ 1, others zero.fEvaluated at sample means presented in Table 5.
A. Gonzalez-Caban et al. / Journal of Forest Economics 13 (2007) 49–7162
ARTICLE IN PRESS
A. Gonzalez-Caban et al. / Journal of Forest Economics 13 (2007) 49–71 63
When all dummy variables in the reduced bivariate probit for prescribed burningare set to equal zero, median WTP is $65.18. This value is pertinent because whileRxDanger, EvacHome and Retired are indeed significant, 74.4% of the respondentsincluded in the reduced model have zero values for each of these variables.
Mechanical program probit models
For the mechanical program, the correlation parameter r is not significant(Table 5); indicating that sample selection bias is not present. This indicates that anydifferences between the first and second wave do not play a role in how peoplerespond to the mechanical program WTP question, and the sample selectionequation results are not presented.
The coefficient on the dollar amount asked of each respondent to fund themechanical fuels reduction program (BidMec) is negative and significant at the 0.01level (Table 5). As in the prescribed burning program Evachome is positive andsignificant at the 0.05 level indicating that people who have evacuated their homebecause of fire are more likely to vote in favor of both fuels reduction programs.
Median WTP for the mechanical fuels reduction program, estimated at the samplemeans of the variables included in the reduced bivariate model, is $295.19 (Table 6).Calculating reduced median WTP where all dummies equal zero does not make sensefor the mechanical program, as the occurrence of all variables equaling zero is verysmall. Median WTP evaluated for each statistically significant dummy variable withits value set to one and the others set to zero help to illustrate the effects of eachvariable on WTP. For example, median WTP for the case in which people have hadto evacuate their home (EvacHome ¼ 1) is $504.93 (see lower half of Table 6).
Comparison of WTP between Native Americans and MT households
Median WTP is calculated using the coefficients from the reduced bivariate probitregression models and the means of the independent variables. Because the sample issplit into two sub-samples for comparison, the means of the independent variablesfor each sub-sample are different. Therefore, differences between Native Americans’WTP and Montana general population residents’ WTP are due only to differences inthe sample means of the independent variables included in the regression models.
Computed from the reduced model excluding protest responses, the NativeAmericans’ median WTP for the prescribed burning program is $134.94, and theMontana general population residents’ median WTP is $174.38 (Table 7). Incontrast, Native Americans’ WTP ($305.50) for the mechanical fuels reductionprogram is slightly higher than the Montana general population residents’ WTP($286) (Table 7). For both, the prescribed burning and the mechanical fuelsreduction programs, there are no statistical differences between Native Americans’WTP and general Montana residents’ median WTP. Caution should be used when
ARTICLE IN PRESS
Table 7. Median WTP estimates from the reduced prescribed burning and mechanical fuels
reduction program models, by group
Model Median WTPa Confidence intervals
95% 90%
Prescribed burning reduction
Montana general populationb $174.38 ($711.33)–$559.04 ($396.38)–$457.58
Native Americansc $134.94 ($801.93)–$498.38 ($483.53)–$408.30
Mechanical fuels reduction
Montana general populationd $286.17 ($65.16)–$773.80 $1.73–$668.24
Native Americanse $305.50 ($57.84)–$837.49 $13.26–$684.88
aEvaluated at sample means.bN ¼ 160; mean of RxDanger ¼ 0.02500; mean of EvacHome ¼ 0.05625; mean of Retired ¼ 0.18750.cN ¼ 149; mean of RxDanger ¼ 0.06040; mean of EvacHome ¼ 0.08054; mean of Retired ¼ 0.14094.dN ¼ 137; mean of RxRedHi ¼ 0.93431; mean of EvacHome ¼ 0.03650; mean of Own-
Home ¼ 0.80292; mean of Income ¼ 43.99270.eN ¼ 140; mean of RxRedHi ¼ 0.95000; mean of EvacHome ¼ 0.08571; mean of Own-
Home ¼ 0.75000; mean of Income ¼ 38.19643.
A. Gonzalez-Caban et al. / Journal of Forest Economics 13 (2007) 49–7164
interpreting these median WTP estimates because the large variation in theconfidence interval. The confidence intervals were calculated using the adaptationof the Krinsky–Robb method to dichotomous choice CVM developed by Park et al.(1991). The majority of the confidence intervals show a large variation from negativeto positive. The negative range of the estimate may suggest that some people must becompensated to accept implementation of the proposed programs; and some peoplehave a zero WTP. Had we anticipated this indifference as well as opposition fromrespondents, a willingness to accept question might have been asked of thoserefusing to pay. This would have allowed estimation of the extended spike model ofKristrom (1997) to model this. However, we believed the long history of NativeAmericans using prescribed fire to enhance big game habitat would make themconsistently supportive of this and the other fuels reduction program. Unfortunatelythe WTP bid distribution was limited to positive values, and there may be someinconsistency between such a bid distribution and the estimation method whichallows for negative WTP values. See Haab and McConnell (2002, pp. 89–91) formore details on this issue.7
Conclusions
This paper demonstrated how the contingent valuation method (CVM) could beused to estimate WTP for prescribed burning and mechanical fuels reduction
7The parameters b2 and r, are restricted equal to zero for the single equation model used to calculate the
restricted log-likelihood value.
ARTICLE IN PRESS
A. Gonzalez-Caban et al. / Journal of Forest Economics 13 (2007) 49–71 65
programs among Native American households on tribal lands and general residentsin Montana. The overall general similarity of results is very encouraging. Using aunivariate test (w2), we did not find a statistical difference between Native Americansand Montana’s general population for the initial random-digit dialed phoneinterview, but did find statistical differences in the response rate for the follow-upin-depth interview.
Rates of protest responses for the prescribed burning program were notstatistically different between Native Americans and general Montana residents. Ifnon-respondents are treated as protests, then there is also no significant difference inprotest rates for the mechanical fuel reduction program. Following the mostconventional approach of not including non-respondents as protest, the protest ratesfor the mechanical fuels reduction program were significantly different between thetwo populations. However, results from a separate WTP model including apopulation intercept variable and bid interaction term showed that there were nostatistical differences between Native Americans and Montana general populationresidents. Having comparable response rates and statistically insignificant differencesbetween groups with regard to the WTP function coefficients and median WTPprovide some indication that Native Americans responded to the CVM surveysimilarly to Montana’s general residents.
The above findings lead us to believe that in general the contingent valuationsurvey for natural resources programs worked reasonably well for Native Americansin Montana. The biggest difference was in follow-up survey response rates. NativeAmericans had a significantly lower response rate that would have reduced theability to generalize from the sample of Native Americans to that population.However, by using a bivariate probit with sample selection bias model the WTPestimates can be generalized to their population. Nonetheless, future surveys shouldattempt to determine whether monetary incentives or other means of motivatingNative Americans, like a letter from their Tribal Council, might improve responserates. In terms of support for forest fire management programs, support forprescribed burning was similar between general Montana households and NativeAmericans. For the mechanical fuels reduction program, the Native Americanssignificantly lower protest rate indicates that Native Americans supported thisprogram at a higher level than did Montana residents. In terms of WTP, we foundthat Native American households were willing to pay a slightly higher amount toimplement the mechanical fuels treatment programs than Montana generalpopulation households; but that these amounts are not statistically differentbetween the two groups. Native American households’ WTP for prescribed burningis less than Montana residents’, but again the difference is not statistically significant.
An important policy implication of these findings is that government agencies withfire protection responsibilities implementing prescribed burning programs for firehazard mitigation need not worry about preparing informational material specific todifferent cultural groups. This translates into potential large planning andadministrative cost savings, in turn reducing the overall costs of the prescribedburning fuels treatment program. These results are similar to those obtained byLoomis et al. (2002) and Loomis et al. (2004) when testing for language-related
ARTICLE IN PRESS
A. Gonzalez-Caban et al. / Journal of Forest Economics 13 (2007) 49–7166
differences between English versus Spanish speaking households in Florida andCalifornia for the same type of fuels reduction programs. The same is not the casefor the mechanical fuels reduction program. However, more replications are neededin other places and for other public programs to validate these findings. If furthertesting demonstrates the similarities found here hold true in other states and forother public programs this could be indicative that CVM surveys may be usefultechniques for evaluating the distribution of benefits for non-market publicprograms, both across and within different cultural groups. The result is alsoencouraging in terms of possibly applying CVM to indigenous people in otherdeveloped countries.
Our results suggest substantial WTP of Native Americans households in Montanaand Montana general population households for a prescribed burning or mechanicalfuels reduction program that would decrease the number of acres burned by wildfiresin Montana as presented in the study discussed here. This support is strong enough,as demonstrated by the high mean WTP for both the prescribed burning and themechanical fuels reduction programs to possibly make these kinds of treatmentseconomically feasible and efficient. With average WTP of at least $154 for theprescribed burning program and $295 for the mechanical fuels reduction program,and 366,000 households in Montana this would be close to $56 million for theprescribed burning program and $108 million for the mechanical fuels reductionprogram. Note, the survey explicitly indicated that only one of the programs wouldbe implemented, so that it would be incorrect to add the values of these two fuelreduction programs together.
Acknowledgments
We want to thank Mr. Dennis Dupuis and Ms. Andrea Gilham who coordinatedthe work with the Confederated Salish-Kootenai Tribe and the Blackfeet Tribe,respectively. We also want to thank the tribal councils of both tribes who allowed usto conduct this research on tribal land. This work was funded by a grant from theJoint Fire Sciences Program of the USDA Forest Service and DOI Bureau of LandManagement. We would also like to thank two anonymous referees who providedsuggestions for clarifying several important points in this manuscript.
Appendix A
Native Americans/Montana General Population (Table A1).
Appendix B
Pooled data set prescribed burning fuel treatment program (Tables B1 and B2).
ARTICLE IN PRESS
Table A1. Native Americans/Montana General Population reasons for voting NO to the
WTP question
Category Prescribed burninga Mechanical reducationa
Native
Americans
Montana
General
population
Native
Americans
Classb
Montana
General
population
Cannot afford to pay 1.3% 0.4% 0.4% 0 NP
Opposed to another government
program
0.4% 0 0.9% 0 P
State/Federal government should
pay
1.7% 0% 0.4% 0.4% P
Landowners’ responsibility 0.9% 0 0.9% 0 P
They should not have to pay 0.4% 0.7% 0.9% 0 P
More information about the
program and its effectiveness
0.4% 0 0 0.7% P
Other options/sources of funding
available
0.4% 1.5% 1.3% 2.2% P
Not happening 0.4% 0 0 0 P
Prescribed burning will limit the
availability of firewood
0.4% 0 0 0 P
Mechanical reduction is better 0.4% 0 0 0 P
Tribal members do not pay Federal
taxes
0.4% 0 0 0 P
Ecologically damaging 0 0 4.4% 3.3% P
Too expensive 0 0 1.3% 1.1% P
Prescribed burning is better 0 0 2.6% 4.4% P
Would not work 0 0 1.7% 2.9% P
Not believable 0 0 1.3% 1.5% P
Protest method of survey 0 0 0.4% 0 P
Mistrust of State/Federal agencies 0 1.5% 0.4% 1.8% P
Program will not be effective 0 0.4% 0 0 P
Only people in affected areas should
pay
0 0.4% 0 0 P
Forest Service has enough money
already
0 0.4% 0 0.4% P
Opposed to new taxes/No more
taxes
0 1.1% 0.4% 1.1% P
Logging is better and cheaper 0 0.4% 0 0 P
Pay Forest Service to do it 0 0.4% 0 0 P
Better ways to help the forest 0 0.4% 0 0 P
Not the way to control fire safety 0 0.4% 0 0 P
Not enough acreage to make it
worthwhile
0 0.4% 0 0 P
It will burn too much 0 0.4% 0 0 P
All control burns have escaped
control
0 0.4% 0 0 P
Decline to state 0 0.4% 0 0.4% P
Program not necessary/too
dangerous
0 0 0.4% 0.4% P
A. Gonzalez-Caban et al. / Journal of Forest Economics 13 (2007) 49–71 67
ARTICLE IN PRESS
Table A1. (continued )
Category Prescribed burninga Mechanical reducationa
Native
Americans
Montana
General
population
Native
Americans
Classb
Montana
General
population
Reduces food sources for wildlife 0 0 0 1.5% P
Takes local jobs away 0 0 0.4% 0.4% P
Log it first 0 0 0 4.4% P
Do not like method/program 0 0 0.9% 1.8% P
Concerned about loggers 0 0 0 0.4% P
Only at the urban wildland interface 0 0 0 0.4% P
People who build in the forest
should pay
0 0 0 0.4% P
Mulch will become a fire problem 0 0 0 0.4% P
Program should be self-financing 0 0 0 0.4% P
Too many control burns already 0 0 0 0.4% P
Montana usually does not have
many fires
0 0 0 0.4% P
Not necessary to intervene with
wildland fires
0 0 0 0.4% P
Prescribed fires are dangerous 0 0 0 0.4% P
Others 0 0 1.6% 0.4% P
aPercent of protest and non protest votes derived from Table 2.bP, protest vote; NP, not a protest vote.
Table B1. Pooled dataset prescribed burning fuel treatment program corrected for protest
with Montana General Population intercept and bid slope interaction term
Variable Coefficient Std. error z-Statistics Probability
C 1.4359 1.1356 1.26 0.2061
BIDRX �0.0031** 0.0013 �2.31 0.0207
BIDRX_MTGEN01 �0.0015 0.0020 �0.76 0.4468
BOTHERPHYSICAL 0.1898 0.4185 0.45 0.6503
BOTHERVISUAL 0.2132 0.4423 0.48 0.6297
BURNEDHOME 38.613 2.67E+08 1.45E-07 1.0000
CAMP �0.1723 0.4588 �0.38 0.7073
COLIVIN �0.0515 0.0726 �0.71 0.4777
CONORG �0.4760 0.4558 �1.04 0.2964
EVACHOME 1.6436** 0.7565 2.17 0.0298
FIREFIGHTER �0.1364 0.3721 �0.37 0.7139
GENDER 0.1624 0.3645 0.45 0.6559
INC 6.33E�06 5.86E�06 1.08 0.2801
MTGENPOP �0.4287 0.4974 �0.86 0.3888
NEIGHBURNED �0.8204 0.6038 �1.36 0.1742
OWNHOME �0.6653* 0.3961 �1.68 0.0930
PROBMINR 0.0394 0.4976 0.08 0.9369
A. Gonzalez-Caban et al. / Journal of Forest Economics 13 (2007) 49–7168
ARTICLE IN PRESS
Table B1. (continued )
Variable Coefficient Std. error z-Statistics Probability
PROBMOD �0.2499 0.5509 �0.45 0.6500
PROBSER �0.3936 0.7551 �0.52 0.6022
RET 1.0602** 0.4358 2.43 0.0150
RXDANGEROUS �0.6545 0.7148 �0.92 0.3598
RXHEALTHPROB �0.0326 0.4997 �0.07 0.9479
RXREDHIFIRE 0.8947 0.8596 1.04 0.2980
WITNESSEDFIRE �0.3219 0.4736 �0.68 0.4967
YRSINMT �0.0047 0.0094 �0.50 0.6164
Mean dependent var 0.6582 SD dependent var 0.4752
S.E. of regression 0.4550
Log likelihood �152.6551
Restr. log likelihood �176.6144 McFadden R-squared 0.1357
LR statistic (24 df) 47.9185 Probability (LR stat) 0.0026
N 275
Note: Single and double asterisk (*) denote statistical significance at the 0.10 and 0.05 levels, respectively.
Table B2. Pooled dataset mechanical fuel treatment program corrected for protests with
interaction term
Variable Coefficient Std. error z-Statistic Probability
C �1.0994 0.8957 �1.23 0.2197
BIDMEC �0.0027** 0.0012 �2.28 0.0233
BIDMEC_MTGEN01 �0.0022 0.0019 �1.18 0.2394
BOTHERPHYSICAL �0.0166 0.3808 �0.04 0.9653
BOTHERVISUAL �0.1190 0.3984 �0.30 0.7651
CAMP �0.2305 0.3765 �0.61 0.5405
COLIVIN 0.0293 0.0656 0.45 0.6556
CONORG �0.9224** 0.4409 �2.09 0.0364
EVACHOME 1.3535 0.5679 2.38 0.0172
FIREFIGHTER �0.6752** 0.3337 �2.02 0.0430
GENDER 0.4010 0.3241 1.24 0.2159
INC 1.21E-05** 5.29E-06 2.29 0.0223
MTGENPOP 0.0952 0.4455 0.21 0.8308
NEIGHBURNED 0.1771 0.5476 0.32 0.7465
OWNHOME �0.6350** 0.3321 �1.91 0.0558
PROBMINR 0.2532 0.4583 0.55 0.5808
PROBMOD �0.4826 0.4631 �1.04 0.2973
PROBSER 1.2833* 0.7434 1.73 0.0843
RET 0.4096 0.3580 1.14 0.2525
RXDANGEROUS 0.3056 0.5590 0.55 0.5846
RXHEALTHPROB 0.5772 0.4319 1.34 0.1814
RXREDHIFIRE 1.4189** 0.6066 2.34 0.0193
WITNESSEDFIRE 0.4154 0.4049 1.03 0.3050
A. Gonzalez-Caban et al. / Journal of Forest Economics 13 (2007) 49–71 69
ARTICLE IN PRESS
Table B2. (continued )
Variable Coefficient Std. error z-Statistic Probability
YRSINMT 0.0003 0.0082 0.03 0.9751
Mean dependent var 0.4951 SD dependent var 0.5008
S.E. of regression 0.4790
Sum squared resid 64.9271
Log likelihood �187.0482
Restr. Log likelihood �212.7815 McFadden R-squared 0.1210
LR statistic (23 df) 51.4666 Probability (LR stat) 0.0006
N 307
Note: Sinlge and double asterisk (*) denote statistical significance at the 0.10 and 0.05 levels, respectively.
A. Gonzalez-Caban et al. / Journal of Forest Economics 13 (2007) 49–7170
References
Adamowicz, W.L., Boxall, P.C., Haener, M.K., Zhang, Y., Dosman, D., Marois, J., 2004. An assessmentof the impacts of forest management on Aboriginal hunters: evidence from stated and revealedpreference data. Forest Science 50 (2), 139–152.
Arrow, K., Solow, R., Portney, P., Leamer, R., Radner, R., Schuman, H., 1993. Report of the NOAAPanel on contingent valuation. Federal Register 58 (10), 4602–4614.
Boyes, W.J., Hoffman, D.L., Low, S.A., 1989. An econometric analysis of the bank credit scoringproblem. Journal of Econometrics 40, 3–14.
Cameron, T., James, M., 1987. Efficient estimation methods for ‘‘closed-ended’’ contingent valuationsurveys. Review of Economics and Statistics 69, 269–275.
Carson, R., Flores, N., Martin, K., Wright, J., 1996. Contingent valuation and revealed preferencemethodologies: comparing estimates for quasi-public goods. Land Economics 72 (1), 80–99.
Diamond, P., Hausman, J., 1994. Contingent valuation: is some number better than no number? JournalEconomic Perspective 8 (4), 45–64.
Fried, J., Winter, G., Gilless, J., 1999. Assessing the benefits of reducing wildfire risk in the wildland urbaninterface: a contingent valuation approach. International Journal of Wildland Fire 9 (1), 9–20.
Greene, W.H., 1992. A statistical model for credit scoring. Working Paper No. EC-92–29, Department ofEconomics, Stern School of Business, New York University.
Greene, W.H., 2000. Econometric Analysis, fourth ed. Prentice-Hall, Upper Saddle River, NJ, 1004pp.Haab, T.C., McConnell, K.E., 2002. Valuing Environmental and Natural Resources: The Econometrics of
Non-Market Valuation. Edward Elgar Publishing, Cheltenham, UK, 326pp.Haener, M.K., Dosman, D., Adamowicz, W.L., Boxall, P.C., 2001. Can stated preference methods be used
to value attributes of subsistence hunting by Aboriginal Peoples? A case study in northernSaskatchewan. American Journal of Agricultural Economics 83, 1334–1340.
Hanemann, M., 1984. Welfare evaluations in contingent valuation experiments with discrete responses.American Journal of Agricultural Economics 66 (3), 332–341.
Hanemann, M., 1989. Welfare evaluations in contingent valuation experiments with discrete responsedata: reply. American Journal of Agricultural Economics 71 (4), 1057–1061.
Heckman, J., 1979. Sample selection bias as a specification error. Econometrica 47 (1), 153–161.Jacobson, T., Roszbach, K., 2003. Bank lending policy, credit scoring and value at risk. Journal of
Banking & Finance 27 (4), 615–633.Kristrom, B., 1997. Spike models in contingent valuation. American Journal of Agricultural Economics
79, 1013–1023.Loomis, J., 1989. Test-retest reliability of the contingent valuation method: a comparison of general
population and visitor responses. American Journal of Agricultural Economics 71, 76–84.Loomis, J., 1990. Comparative reliability of the dichotomous choice and open-ended contingent valuation
techniques. Journal of Environmental Economics and Management 18, 78–85.Loomis, J.B., Bair, L.S., Gonzalez-Caban, A., 2001. Prescribed fire and public support: knowledge gained,
attitudes changed in Florida. Journal of Forestry 99 (11), 18–22.
ARTICLE IN PRESS
A. Gonzalez-Caban et al. / Journal of Forest Economics 13 (2007) 49–71 71
Loomis, J.B., Bair, L.S., Gonzalez-Caban, A., 2002. Language related differences in a contingentvaluation study: English versus Spanish. American Journal of Agricultural Economics 84 (4),1091–1102.
Loomis, J., Bair, L.S., Gonzalez-Caban, A., 2004. Multicultural evaluation of the performance ofcontingent valuation for forest fire prevention. Contemporary Economic Policy 22 (3), 407–419.
McConnell, K., 1990. Models for referendum data: the structure of discrete choice models for contingentvaluation. Journal of Environmental Economics and Management 18 (1), 19–34.
Park, T., Loomis, J., Creel, M., 1991. Confidence intervals for evaluating benefit estimates fromdichotomous choice contingent valuation studies. Land Economics 67, 64–73.
Pyne, S., 1982. Fire in America: A Cultural History of Wildland and Rural Fire. Princeton UniversityPress, Princeton, NJ, 654pp.
Reiling, S., Boyle, K., Philips, M., Anderson, M., 1990. Temporal reliability of contingent values. LandEconomics 66 (2), 128–134.
van de Ven, W.P.M.M., van Praag, B.M.S., 1981. The demand for deductibles in private health insurance:a probit model with sample selection. Journal of Econometrics 17, 229–252.