how reliable are meta-analyses for international … · 2015. 2. 19. · meta analysis (ma) is...
TRANSCRIPT
NON-MARKET VALUATION
HOW RELIABLE ARE META-ANALYSES FOR INTERNATIONAL
BENEFIT TRANSFERS ?
Ståle NavrudDepartment of Economics and Resource Management
Norwegian University of Life Sciences
E-mail: [email protected]
Environmental Economics Research Hub Workshop
Australian Agricultural and Resource Economics Society (AARES) 53rd Annual Conference,
Cairns, February 10th 2009
MOTIVATION & KEY RESULT
Increased demand for environmental valuation by national EPAs and the EC for use in:i) Cost-Benefit Analyses of projects and programsii) External costing; to determine the correct level of environmental taxes and subsidiesiii) Green accountingiv) Natural Resource Damage Assessment (NRDA)
Often lack time and resources to conduct new primary valuation studies:- Simple transfers of unit values (WTP per household per year) from existing valuation studies often used, but high transfer errors remain a challenge - Meta-analysis (MA) claims to have the potential to increase precision in benefit transfer (BT); but few have tested this
Main research question:Using data from three countries, how does meta-analytic BT compare with simple BT techniques?
Main finding: Simple unit value transfer from domestic studies performs no worse on average than meta-analytic Benefit Transfer
Lindhjem, H. and S. Navrud 2008: How Reliable are Meta-Analyses for InternationalBenefit Transfer ? Ecological Economics, 66; 425-435.
OUTLINE
BT and MA in non-market valuation
Background and research questions
Meta-model, data sources and estimation
Benefit transfer tests and results
Main challenge to increased policy use of environmental valuation – both original research and transferred values:
“Will people actually pay the amounts they say they are willing to pay in Stated Preference Surveys?”
BT and MA IN NON-MARKET VALUATION (1)
Transfer economic value of public good from study site (where the primary valuation study) to policy site
Both benefits and costs can be transferred; rather name it “value transfer”
Meta Analysis (MA) is “an analysis of analyses” (here: previous environmetalvaluation studies)
Often mix between Revealed (travel cost, hedonic pricing) and Stated Preference methods (contingent valuation and choice experiments) in MA
Seminal work: Smith and Kaoru (AJAE 1990) MA on travel cost models of recreation benefits in the US 1970-86
BT Techniques:
1. Simple Unit Value Transfer (WTP /hh/year)
2. Benefit Function Transfer (WTP function from one study)
3. Meta Analysis (WTP function from many studies; incl. study and site var.)
MA IN NON-MARKET VALUATION (2)
Three main uses of MA:
- Research synthesis
- Hypotheses testing
- Benefit transfer
Hypotheses testing: Classic questions in non-market valuation – WTP-WTA disparity
– Scope insensitivity
– Convergent validity of estimates from different methods
– Use vs non-use values
– Actual vs hypothetical (stated) WTP (“hypothetical bias”)
MA for benefit transfer: More recent.– Using meta regression function(s) to transfer value to the “policy site”
– Promise of increased precision in value estimates for cost-benefit analysis
META ANALYSES OF ECOSYSTEMS AND BIODIVERSITY
Recreational use values of ecosystems (TC and CV)- Rosenberger and Loomis (2003), US studies- Shrestha and Loomis (2003), US studies- Zandersen and Tol (2005) (9 European countries)
Non-use values (mainly CV)- Loomis & White (1996) Rare and endangered species- Brander et al (2006) Wetlands- Brander et al (2007) Coral reefs- Nijkamp et al (2007) Biodiversity and Habitat Services- Jacobsen and Hanley (2007): Biodiversity; 46 CV studies worldwide- Tuan and Lindhjem 2008: Biodiversity in Asia and Oceania
Main problem for using these MA for BT: Too wide in scope
BACKGROUND
Economists have valued non-timber benefits (NTBs) from forests in Finland, Sweden & Norway for 20 years– Mostly contingent valuation (CV) surveys; recently also Choice Experiments
– People’s WTP for forest protection and/or multiple use forestry (MUF)
Lindhjem (2007) in Journal of Forest Economics takes stock of this literature using MA
Based on meta-data in Lindhjem (2007) we set out to :
Evaluate the performance of meta-analysis for benefit transfer (MA-BT)
Bergstrom and Taylor (2006, pp.359):
”….before widespread application of MA-BT models, there is a need for additional MA-BT validity tests across different types of natural resources and environmental commodities.”
Navrud and Ready (2007, pp. 288):
”Simple approaches [to benefit transfer] should not be cast aside until we are confident that more complex approaches do perform better.”
Lindhjem, H. and S. Navrud 2008: How Reliable are Meta-Analyses for International Benefit Transfer ? Ecological Economics, 66; 425-435.
RESEARCH QUESTIONS
Given a priori “favourable” conditions for benefit transfer:– Homogenous countries in terms of culture, GDP, institutions, forest use
– Studies using the same valuation methods for the “same” good
– MA with high explanatory power
How large are the transfer errors using MA (i.e. difference between policy site value and transferred value)?
Is meta-analytic BT more reliable (has lower transfer error) than simple BT techniques used in practice?
Many BT tests between single sites, while rare for MA:– Rosenberger & Loomis (2000), Shresta & Loomis (2001; 2003), Santos (2007)
– Mostly use values ( value per activity day)
META-MODEL
Standard meta-model we use:
– WTPms = Willingness-To-Pay obs. m from study s (Annual WTP per household, long-term, 2005-NOK values, converted using PPP)
– X, P = Site/programme characteristics of the forest good valued
– M = Study characteristics (e.g. CV elicitation method, survey modes etc)
– S = Socio-economic characteristics
– ems = Observation-level error, us = Study-level error
Meta-analytic benefit transfer (MA-BT): 1) Set X, P, S in estimated equation equal to values at policy site
2) Set M equal to ”average” or best-practice values
3) Calculate WTP for the policy site using the meta-regression equation
smsqmsS
lmsP
kmsM
jmsX0ms ueSPMXWTP ++β+β+β+β+β=
META-DATA SOURCES & CODING
More than 50 studies reporting from 30 CV surveys in Finland, Norway and Sweden since 1985 (incl. theses/reports)
All study results coded in a spreadsheet for variables X, P, M, S that are hypothesised to explain variation in WTP– Site/programme variables (X, P): I)Local, regional, national; country;
urban vs non-urban; protection, Multiple Use Forestry or mix; season; hectare/percentage protection
– Study variables (M): WTP question formats, payment vehicles, 3 levels of response rates for mail survey, individual vs household
– Income/education/gender (S): Not significant
– Other variables: Year of survey, MSc Thesis, unpublished studies
Table 1 Stated preference valuation surveys of non-timber benefits from forests in Finland, Norway and Sweden, 1985-2005
Main references Year Good1 Gain/ loss
Mode Scope2 Method
# WTP (USD)
Finland Kniivilä (2004) 2000 P L Mail R, L CV: DC 2 61-107 Lehtonen et al (2003) 2002 P G Mail R CV: DC 5 190-342 Pouta et al (2000; 2002) 1997 P G/L Mail N CV: DC 4 154-227 Pouta (2003; 2004; 2005) 1998 M G Mail N CV: DC 2 287-299 Rekola and Pouta (2005) 1995 M G Mail L CV: DC 1 20 Siikamäki and Layton (2005) 1999 P G Mail N CV: DC, CE 3 79-134 Mäntymaa et al (2002) 1999 P, B G Mail N CV: OE 4 224-380 Horne et al (2005) 1998 P, M G/L Interv. L CE 1 -16 Tyrväinen & Väänänen (1998) 1995 P, O L Mail L CV: OEPC 5 31-124 Tyrväinen (2001) 1996 P, O L/G Mail L CV: OEPC 6 22-248 Norway Simensen and Wind (1990) 1989 P, M G Interv. L CV: OE 3 21-159 Hoen and Winther (1993) 1990 P, M G Interv. N CV: OEPC 6 14-65 Veisten et al (2004a; b) 1992 B L Interv. N CV: OE/OEPC 3 138-210 Sandsbråten (1997) 1997 M L/G Interv. L CV: DC 2 43-45 Leidal (1996) 1996 P L Interv. L CV: DC/OE 3 455-504 Skagestad (1996) 1996 P, M G Interv. L CV: OEPC 1 15 Veisten and Navrud (2006) 1995 P L Mail R CV: DC/OE 4 3-104 Hoen and Veisten (1994) 1992 M G Interv. L CV: OE 1 50 Hoen and Veisten (1994) 1993 M G Interv. L CV: DC 1 48 Strand and Wahl (1997) 1997 P L Interv. L CV: OE/DC 2 172-243 Sweden Bojö (1985) 1985 P G Interv. L CV: DC 1 58 Bostedt and Mattson (1991) 1991 M, O L Mail L CV: OE 1 385 Mattson and Li (1993) 1991 M, O L Mail R CV: OE/DC 2 469-907 Mattson and Li (1994) 1992 M, O L/G Mail R CV: DC, CE 2 440-1280Kriström (1990a; b) 1987 P G Mail N CV: DC/OE 4 275-725 Johansson (1989) 1987 B L Mail N CV: OE 1 254 Bostedt and Mattson (1995) 1992 M, O G Mail L CV: OE 2 78-84 Fredman & Emmelin (2001) 1998 M, O G Mail R CV: OE 1 92 Total number of estimates 72 1, Good: P = Protection, M = Multiple Use Forestry, B = Primarily forest biodiversity, O = Other; # = Number of estimates in
the meta-regression. 2 = Local (L), Regional (R), National (N)
Table Meta-analysis variables and descriptive statistics Variable Description Sign Mean (SD) Dependent variable WTP2005 WTP in 2005 NOK …. 1192 (1374)Methodological variables: CE Binary: 1 if choice experiment, 0 if CV +/- 0.08 (0.25) OE Binary: 1 if OE without payment card, 0 if dichotomous choice - 0.36 (0.48) OEPC Binary: 1 if OE with payment card, 0 if dichotomous choice - 0.26 (0.44) Volunpv Binary: 1 if payment vehicle is described as a voluntary (unrelated to use) (e.g.
donation to a fund), 0 if otherwise (e.g. tax) + 0.18 (0.39)
Userpv Binary: 1 if payment vehicle is related to recreational use or access (e.g. Entrance fee etc), 0 if otherwise (e.g. Tax)
- 0.19 (0.4)
Otherpay Binary: 1 if payments were to occur on something other than an annual long-term basis, for example as a lump-sum, annual for a limited period, monthly or per season
+ 0.5 (0.5)
Actualpay Binary: 1 if payments were actually made, 0 if hypothetical WTP - 0.03 (0.17) Individual Binary: 1 if individual WTP, 0 if household +/- 0.32 (0.47) Mailhigh Binary: 1 if mail survey with high (more than 65% useable questionnaires), 0 if in-
person interview +/- 0.13 (0.33)
Mailmed Binary: 1 if mail survey with medium (between 50% and 65% useable questionnaires), 0 if in-person interview
+/- 0.25 (0.44)
Maillow Binary: 1 if mail survey with low (below 50% useable questionnaires), 0 if in-person interview
+/- 0.31 (0.46)
Quality of study: UnPub Binary: 1 if WTP estimate unpublished, 0 if published +/- 0.38 (0.47) Mscthesis Binary: 1 if primarily a Master thesis, 0 if otherwise +/- 0.15 (0.36) Good description: Forestpract Binary: 1 if more cautious forestry practices; 0 if full protection +/- 0.32 (0.47) Protmix Binary: 1 if mix of protection and forestry practices; 0 if full protection +/- 0.07 (0.26) Forestarea Continuous: Total forest area of proposed change (ha). + See text Impl Binary: 1 if neither percentage of total land area nor forest area (ha) are mentioned in
the survey, 0 if otherwise +/- 0.78 (0.42)
Hafrerc Continuous: Area percentage of total productive forest area in the country (estimated in year 2005, or based on info provided in study)
+ See text
Haperc Continuous: Area percentage of total land area + See text Localgood Binary: 1 if local good, 0 if nationwide + 0.42 (0.5) Reggood Binary: 1 if regional good, 0 if nationwide + 0.21 (0.41) Sweden Binary: 1 if study conducted in Sweden, 0 if Norway +/- 0.19 (0.4) Finland Binary: 1 if study conducted in Finland, 0 if Norway +/- 0.44 (0.5) Urban Binary: 1 if primarily urban forest (major town), 0 if otherwise +/- 0.33 (0.47) Season Binary: 1 if surveyed in autumn/winter (i.e. Sept-March), 0 if spring/summer (i.e.
April-August) - 0.6 (0.49)
Avoidloss Binary: 1 if it is WTP for avoiding a loss, 0 if it is for an improvement + 0.4 (0.49) Use Binary: 1 if primarily use/users, 0 otherwise (i.e. users and non-users are incl.) + 0.36 (0.48) Other variables Year Continuous: Range 1 (1985, year of first survey) to 16 (2002). + 10.6 (4.2)
EXAMPLE: CV SURVEY; DESCRIPTION OF ENV. GOOD
Veisten; K. and S. Navrud
(2006) Contingent valuation and actual payment for voluntarily provided
passive-use values: Assessing the effect of an induced truth-telling mechanism and elicitation formats. Applied Economics, 38, 735–756
EXAMPLE: : CV SURVEY; DESCRIPTION OF ENV. GOOD (2)
Veisten and Navrud (2006)
META-REGRESSION ESTIMATION & RESULTS
Four models estimated on meta-data, all Huber-White robust estimation (adjusting for heteroskedasticity & intrastudy correlation)
Models 1 & 2: Full linear and double log models
Model 3: Double log, one extreme observation excluded
Model 4: Restricted version of Model 3; Only retaining variableswhich parameters are p<0.20; commonly used in MA BT
Table 4 Meta-regression model results with different model specifications Full models Restricted models (double log) Variable I. Linear II. Double-log III. Trimmed (one obs.
excluded) IV. Restricted (p<0.2 excluded)
Intercept 1549.256* (875.5331)
4.140617** (1.170449)
1.943833** (.9250999)
1.72109*** (.5947163)
CE 192.6951 (378.0004)
.3297439 (.2406569)
.0964237 (.1543238)
OE -1334.071** (594.0914)
-.495455 (.3395935)
-.3484957 (.3404972)
OEPC 227.536 (382.0898)
-.3608809 (.2204971)
-.2795691 (.1720878)
Volunpv 3799.7*** (988.7608)
2.803627*** (.7711909)
1.716961*** (.5918412)
1.845446*** (.3816678)
Userpv -2564.024*** (424.8793)
-.3300177 (.4289763)
.2968603 (.2882608)
Otherpay 183.4371 (620.5135)
-.066285 (.4875653)
.419128 (.4353331)
Actualpay -571.5364* (320.3029)
-2.099854*** (.1061977)
-1.974489*** (.1579718)
-1.715755*** (.3672467)
Individual 1834.944*** (471.8069)
1.295294*** (.2941284)
1.58119*** (.1762362)
1.410485*** (.1699934)
Mailhigh -6477.973*** (1032.545)
-4.986712*** (.7683036)
-5.232499*** (.7537281)
-4.10506*** (.6955875)
Mailmed -4864.702*** (1043.229)
-4.270923*** (.9019158)
-4.919064*** (.8583492)
-3.735134*** (.7212115)
Maillow -2476.168** (970.375)
-3.009995*** (.9114381)
-4.18444*** (.7654645)
-3.328119*** (.6948911)
Unpub -791.1643** (320.2655)
.0190386 (.3603327)
.0845459 (.3065782)
Mscthesis -1916.265** (754.8593)
-1.730453*** (.5586125)
-1.299899*** (.3998584)
-1.121121** (.414038)
Forestpract 765.1689** (320.39)
.2771635 (.3163496)
-.1541521 (.2228194)
Protmix -1261.768 (808.1531)
-.6688487 (.5322865)
-.4740047 (.4339458)
Impl 1276.517 (934.0211)
1.279632** (.525085)
1.246919*** (.4045801)
1.168564*** (.1942462)
Localgood 649.1225 (536.0937)
-.4468539 (.4902242)
-1.327387*** (.3227563)
-1.088904*** (.1613885)
Reggood 2350.52*** (746.4256)
.821114* (.471253)
.5384419 (.6038565)
Sweden 1111.561 (822.4675)
2.147048** (.9714438)
3.889263*** (.7683388)
3.370004*** (.5675196)
Finland 644.2306 (1046.65)
2.131236* (.6016583)
2.254886*** (.707515)
1.932351*** (.6072392)
Urban -1551.158*** (552.4695)
-.5718084 (.4513243)
.1599182 (.3069824)
Season -1879.212*** (496.1174)
-.784065** (.313954)
-.6893471** (.2758698)
-.447132** (.1828208)
Avoidloss 627.9457 (415.2456)
.5853566* (.3072963)
.1907735 (.1567345)
Use 451.9457 (721.9776)
.0224779 (.3540051)
-.3166314 (.2617096)
Year/LnYear 130.3553 (82.63281)
1.242805** (.5555091)
2.380679*** (.2862772)
2.246495*** (.3421329)
Log likelihood χ2 101.47*** 121.56*** R2 0.756 0.815 0.886 0.814 N 72 72 71 71
Note: *p < 0.10, **p < 0.05, ***p < 0.01, Number of survey clusters for models = 27.
Regression results cont’d
META-REGRESSION RESULTS (CONT’D)
Models explain a large part of the variation in WTP– R2 between 0.75 and 0.88
– Higher R2 than most other MA studies
Models confirm some key expectations – Elicitation method:
Open-ended WTP questions lower than for dichotomous choice
– Survey mode:
High response mail surveys lower WTP than low response surveys
– Voluntary payment vehicle yields higher WTP; actual payment lower
– WTP increases over time (beyond inflation for market CPI goods)
META-REGRESSION RESULTS (CONT’D)
More method/study variables than site variables are significant. Typical result in the literature; potential problemfor MA-BT - Full discussion of results in Lindhjem (2007)
Site variables: Differences between countries, local good lower WTP, season important
No scope sensitivity of WTP to size (ha) of forest detected– Checked for protection only, for national and local forests, for those who
explicitly were given size in percentage or hectares in the survey
– Might mean that the size of the area is too crude a measure for scope of the good
TRANSFER ERROR TESTING METHODOLOGY
Transfer error (TE); WTPE = estimated value; WTPT = true value
Two main BT tests:
– Test 1: Analyse within-sample and out-of-sample overall mean TE of MA model
– Test 2: Compare MA-BT with simple BT techniques
– Also: Checking sensitivity across MA model specifications/restrictions
T
TE
WTP|WTPWTP|TE −=
TRANSFER ERROR TEST 1
We use all four meta-regression models for TE test 1
(a) Within-sample TE:– Predict WTP for each of the 71-72 observations
– Calculate TE for each obs. & overall mean TE (Mean Absolute Percentage Error)
(b) Out-of-sample TE:– Use N-1 of the data to predict the out-of-sample WTP observation
– Run N models and calculate mean TE each time, and then overall mean TE
RESULTS OF TRANSFER ERROR TEST 1Table 5 Transfer errors for within-sample and out-of-sample runs MA-BT models
TE for different model specifications (%) Model I:
Linear Model II: Dbl log
Model III: 1 obs. excl.
Model IV: p>0.2 excl.
Within-sample Overall mean TE Overall median TE
135 37
52 26
39 25
52 30
0 -25th percentile (obs 1-18)* 390 71 77 76 25 - 50th percentile (obs 19-36) 105 92 57 72 50 - 75th percentile (obs 37-54) 24 17 25 24 75 – 100th percentile (obs 55-71/2) 24 26 26 37 Out-of-sample Overall mean TE Overall median TE
266 51
222 40
62 34
63 31
0 -25th percentile (obs 1-18) 770 202 110 109 25 - 50th percentile (obs 19-36) 213 592 70 53 50 - 75th percentile (obs 37-54) 38 27 20 35 75 – 100th percentile (obs 55-71/2) 42 67 50 54 Notes: *Percentiles calculates the transfer errors in four different segments of the data, when WTP is sorted in ascending
order.
RESULTS (CONT’D): PREDICTED WTP VS OBSERVED WTP
24
68
10
0 20 40 60 80Observations
lnwtp05 wtp_p
24
68
10
0 20 40 60 80Observations
lnwtp05 wtp_p
Observed log WTP (lnwtp05) and predicted values (wtp_p) for Model IV (out-of-sample)
Mean TE = 63 %
Tendency to err on low side
Observed log WTP (lnwtp05) vs predicted values (wtp_p) for Model II (out-of-sample)
Mean TE = 222 %
Tendency to overshoot
RESULTS OF TRANSFER ERROR TEST 1 (CONT’D)
Relatively low median TE for all models (25-51 %)
Linear models perform much worse (mean TE of 135-266 %)
Models 3 & 4 perform quite well (mean TE of 39-63 %)– TE comparable to other studies, e.g. Brander et al (2006, 07), Shresta and
Loomis (2001; 2003)
– Predicted values err on the conservative (low) side
Higher TE for lower WTP values (also result of def. of TE)– But for practical CBA, TE for aggregated WTP over the total affected
population matters
– Overall TE thus depends on the size of the affected hh x WTP/hh
TRANSFER ERROR TEST 2: MA BT SIMULATION
One random WTP estimate drawn from each of the 26 surveys to represent a benchmark ”policy site” value
All other estimates from the benchmark survey taken out
26 predictions for the ”policy site” using the best two MA models from test 1
”Site” variables set equal to policy site
Method/study variables: Set the same as for policy site estimate– Instead of mean for sample, ”best practice”, 0.5 or other values
– Not possible in practice: lower bound TE, ”the best” the MA models can do
TRANSFER ERROR TEST 2: SIMPLE BT TECHNIQUES
Compare TE from two best MA models from Test 1 with simple BT techniques:– WTP from most similar domestic study site (”the best you could do”)
– WTP from most similar domestic and international study
– Mean WTP of domestic studies w similar site characteristics
– Mean WTP of domestic & international studies w similar site characteristics
– Use raw mean for all WTP observations (”the worst you could do”)
”Site” variables MUF, protection or mix, local, national forest set equal to policy site
Mean TE 217% 62% 71% 86% 166% 126% 47% Median 120% 7% 12% 41% 85% 70% 37% Mean TE* (same obs.)
196% 62% 62% 86% 136% 111% 33%
Median* (same obs.)
120% 7% 7% 41% 85% 70% 29%
------------------------------Table shortened-------------------------------------------
Best similar domestic (D) or
internat. (I) study
Mean of similar domestic (D) or international (I)
studies
MA-BT models
D D+I
Main reference
Site bench-mark value
Raw mean WTP
all studies
(-1)
D D+I** n Mean n Mean
III IV
Simensen and Wind (1990)
286 1225 (328%)
289 (1%)
289 (1%)
4 300 (5%)
14
756 (164%)
113 (60%)
272 (4%)
Hoen and Winther (1993)
340 1277 (275%)
na 1847 (443%)
0 na 6 3954 (1063%)
2367 (596%)
641 (88%)
Veisten et al (2004a; b)
1355 1193 (11%)
na 1638 (20%)
0 na 1 1638 (20%)
572 (57%)
1256 (7%)
Sandsbråten (1997)
277 1218 (339%)
286 (3%)
286 (3%)
4 351 (27%)
14 771 (178%)
1175 (323%)
416 (49%)
Leidal (1996)
3248 1109 (65%)
1567 (51%)
1567 (51%)
3 1047 (67%)
10 519 (84%)
1985 (38%)
2258 (30%)
RESULTS OF TRANSFER ERROR TEST 2 (CONT’D)Table: Comparison of transfer errors of simple BT techniques with meta-analytic BT
Simple benefit transfer techniques Mean TE (%)
Single, best domestic study 62
Single, best domestic & int. study Site variables = “policy site” 71
Mean WTP, domestic studies Methodological variables ignored 86
Mean WTP, int. & domest.studies 166
Mean WTP from all regardless of site/study characteristics – “Max TE” 217
Meta analytic benefit transfer
Model IV (restricted model, p<0.2) Site variables = “policy site” 47
Model III (all variables incl.) Methodolog. variables set “optimal” 126
Figure 3
Transfer errors for MA-BT model 4 and mean of similar domestic studies arranged in ascending order of TE for each BT technique,
respectively
0
100
200
300
400
500
600
700
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Number of transferred estimates
Tran
sfer
Err
or (%
)
Mean of similar domestic studies MA-BT:Model IV
Lower TE for MA-BT model IV over domestic mean due to two very high TE values for the latter
COMPARISON OF MA-BT MODEL IV AND MEAN FROM SIMILAR DOMESTIC STUDIES
70% of transfers with MA-BT IV
50% of domestic mean transfers
40% of transfers for both methods have TE<20%
Removing two extreme TE values for both methods: Mean TE = 35% for both
TE<40%
CONCLUSION
Best models yield median and mean TE of 25-34% and 39-62% in initital check of reliability (Test 1)
Use of mean WTP from domestic study sites performs no worse on average than the two MA-BT models tested (Test 2)
Including international studies in simple BT technique increases TE (Test 2)
For both tests large variation in performance between model specifications
More research needed to understand under what conditions MA-BT or other techniques may be more appropriate
What is the acceptable transfer error for different policy uses ? Highest for CBA, Lower for External costing/Environmental taxes and GA, and lowest for NRDA (EU Environmental Liability Directive). Also context-dependent
More tests of validity and reliability of Stated Preference methods
Main challenge to increased policy use of environmental valuation – both original research and transferred values:
“Will people actually pay the amounts they say they are willing to pay in Stated Preference surveys?”
Preservation of biodiversity in coniferous forests in the Oslomarka Forests
Comparing Hypothetical Donations in Contingent Valuation Surveys
with Actual Donations
Veisten, K. & S. Navrud 2006. Contingent valuation and actual payment for voluntarily provided passive-use values: Assessing the effect of an induced truth-telling mechanism and elicitation formats.
Applied Economics, 38, 735–756