1. 150th EAAE Seminar: The spatial dimension in analysing the linkages between agriculture, rural development and the environment Edinburgh, 22 Oct 2015 Julian Sagebiel IÖW – Institute for Ecological Economy Reseach, Berlin A simple method to account for spatially-different preferences in discrete choice experiments Julian Sagebiel, Klaus Glenk, Jürgen Meyerhoff

2. Aim of this presentation –Present an easy approach to attain spatially different willingness to pay from discrete choice experiments –Exemplify it with data from a discrete choice experiment (DCE) on local land use changes –Focus is on forest as alternative to agriculture –Establish willingness to pay function to predict values on different spatial units –Function depends on spatial variables –Related to Benefit Transfer methods (Rolfe et al. 2015) –Map WTP for German counties 2

3. 3 Introduction: Need for valuation –Land use conflicts and need for land use changes –Interaction with climate change –New political perspective on nature conversation –E.g. Convention on Biological Diversity, European Water Framework Directive –Sustainable land use requires incorporation of all costs and benefits –On farm level –Climate effects –Societal effects e.g. landscape, aesthetic value

4. 4 Introduction: Discrete choice experiments –DCEs to inform policy decisions –Several studies on land use conducted in Europe (van Zanten et al. 2014) –Can be integrated into cost-benefit analysis –Method to elicit willingness to pay (WTP) for non-market goods –Survey based –Respondents choose among alternative land use scenarios –Alternatives vary in their attributes

5. 5 Introduction: Willingness to pay in space –Willingness to pay varies in space –People have different preferences –People have different reference points –Several attempts to integrate space in WTP estimates –Kriging (Campbell 2009, Czajkowski 2015) –Global and Local Hotspots (Campbell 2008, Johnston & Ramachandran 2013, Meyerhoff 2013)

6. The survey –March/April 2013 –About 1,400 randomly sampled respondents all over Germany –Online questionnaire of about 30 minutes –Socio-demographics –Land use and climate change: attitudes, perceptions, knowledge –Recreational activities –Each respondent revealed his place of residence on a map (WGS84) 6

7. Spatial distribution of sample 7

8. The discrete choice experiment –Local land use changes –Within 15 kilometre radius of place of residence –Each respondent has a unique status quo situation –27 choice sets in three blocks –D-efficient design for multinomial logit model –Minimize willingness to pay standard error –Three alternatives of which one is status quo (“as today”) –Six land-use related attributes 8

9. Attributes and levels 9 AttributeLevels Share of forest (ShFor)As today, decrease by 10%, increase by 10% Field size (FiSiz)As today, half the size, twice the size Biodiversity in agrarian landscapes (Biodiv) As today, slight increase (85 points), considerable increase (105 points) Share of maize on arable land (ShMai) As today, max. 30% of fields, max. 70% of fields Share of grassland on agricultural fields (ShGra) As today, 25%, 50% Annual contribution to fund (Price) 0, 10, 25, 50, 80, 110, 160 €

10. The approach in a nutshell 10 A B C D Data preparation Calculate within the 15 km radius the status quo of all relevant attributes (here forest share) Incorporate status quo of respondent e.g. by substiting attribute level "as today" with the status quo situation Model estimation Specify indirect utility function: For forest share, a quadratic utility function may be adequate Estimate the parameters of the utility function with discrete choice model Derive willingness to pay function WTP prediction For each county, calculate a (discrete) distribution of forest share for the population Use the willingness to pay function to predict the share of forest for each level of the distribution Value Mapping Multiply the willingness to pay with the number of eligible inhabitants and create a map with these aggregate values Map further statistics, e.g. the standard deviation of willingness to pay in each county

11. Step A: Data preparation 11 A B C D Data preparation Calculate within the 15 km radius the status quo of all relevant attributes (here forest share) Incorporate status quo of respondent e.g. by substiting attribute level "as today" with the status quo situation Model estimation Specify indirect utility function: For forest share, a quadratic utility function may be adequate Estimate the parameters of the utility function with discrete choice model Derive willingness to pay function WTP prediction For each county, calculate a (discrete) distribution of forest share for the population Use the willingness to pay function to predict the share of forest for each level of the distribution Value Mapping Multiply the willingness to pay with the number of eligible inhabitants and create a map with these aggregate values Map further statistics, e.g. the standard deviation of willingness to pay in each county

12. Step A: Data preparation –Calculate within the 15 km radius the status quo of all relevant attributes (here forest share) –Any GIS software (ArcGIS, QGIS) –Requires land use data –Incorporate status quo of respondent e.g. by substituting attribute level "as today" with the status quo situation 12 Attribute levelOriginal CodingSQ Modification Status quo is25%85%25%85% As today002585 10% less111575 10% more223595

13. Step A: Data preparation 13 Attribute levelOriginal CodingSQ Modification Status quo is25%85%25%85% As today002585 10% less111575 10% more223595

14. Step B: Model estimation 14 A B C D Data preparation Calculate within the 15 km radius the status quo of all relevant attributes (here forest share) Incorporate status quo of respondent e.g. by substiting attribute level "as today" with the status quo situation Model estimation Specify indirect utility function: For forest share, a quadratic utility function may be adequate Estimate the parameters of the utility function with discrete choice model Derive willingness to pay function WTP prediction For each county, calculate a (discrete) distribution of forest share for the population Use the willingness to pay function to predict the share of forest for each level of the distribution Value Mapping Multiply the willingness to pay with the number of eligible inhabitants and create a map with these aggregate values Map further statistics, e.g. the standard deviation of willingness to pay in each county

15. Step B: Model estimation 15

16. Estimation Results: Coefficients 16 CoefficientStandard error ASCsq0.18 * 0.049 ShFor0.073 * 0.0040 ShForSquared-0.00064 * 0.000084 FiSiz: Half-0.24 * 0.044 FiSiz: Double-0.21 * 0.038 Biodiv0.18 * 0.020 ShMai0.0080 * 0.0030 ShMaiSquared-0.00020 * 0.000037 ShGra0.017 * 0.0038 ShGraSquared-0.00033 * 0.000067 Price-0.0067 * 0.00040 * p < 0.01

17. Step B: Model estimation 17

18. Quadratic Utility Function 18

19. Marginal willingness to pay 19

20. Step C: WTP prediction 20 A B C D Data preparation Calculate within the 15 km radius the status quo of all relevant attributes (here forest share) Incorporate status quo of respondent e.g. by substiting attribute level "as today" with the status quo situation Model estimation Specify indirect utility function: For forest share, a quadratic utility function may be adequate Estimate the parameters of the utility function with discrete choice model Derive willingness to pay function WTP prediction For each county, calculate a (discrete) distribution of forest share for the population Use the willingness to pay function to predict the share of forest for each level of the distribution Value Mapping Multiply the willingness to pay with the number of eligible inhabitants and create a map with these aggregate values Map further statistics, e.g. the standard deviation of willingness to pay in each county

21. Step C: WTP prediction –For each county, calculate a (discrete) distribution of forest share for the population –Requires high resolution data on population –Here 250x250m raster data –Similar to step A, calculate for each raster cell the share of forest –For each county, draw a distribution of forest share 21

22. Example Distribution of Forest Share 22

23. Step C: WTP prediction – 23

24. Step C: WTP estimation 24

25. Step C: Value mapping 25 A B C D Data preparation Calculate within the 15 km radius the status quo of all relevant attributes (here forest share) Incorporate status quo of respondent e.g. by substiting attribute level "as today" with the status quo situation Model estimation Specify indirect utility function: For forest share, a quadratic utility function may be adequate Estimate the parameters of the utility function with discrete choice model Derive willingness to pay function WTP prediction For each county, calculate a (discrete) distribution of forest share for the population Use the willingness to pay function to predict the share of forest for each level of the distribution Value Mapping Multiply the willingness to pay with the number of eligible inhabitants and create a map with these aggregate values Map further statistics, e.g. the standard deviation of willingness to pay in each county

26. Step D: Value mapping 26 –Map the average MWTP values –Multiply the willingness to pay with the number of eligible inhabitants and create a map with these aggregate values –Map further statistics, e.g. the standard deviation of willingness to pay in each county

27. Share of forest and marginal willingness to pay 27

28. Results –In general, higher MWTP where forest share is low –But: Population density is very important –MWTP hotspots are in the north of Germany –Eastern Midlands are already equipped with large forests 28

29. Discussion Advantages –Straightforward estimation –Prediction on arbitrary scales –Inclusion of several exogenous variables possible –Flexible utility functions possible Limitations –Data requirement –Unobserved heterogeneity –Accuracy? 29

30. References –Campbell D, Scarpa R, Hutchinson W. Assessing the spatial dependence of welfare estimates obtained from discrete choice experiments. Letters in Spatial and Resource Sciences 2008;1:117–26. doi:10.1007/s12076-008-0012-6. –Campbell D, Hutchinson WG, Scarpa R. Using Choice Experiments to Explore the Spatial Distribution of Willingness to Pay for Rural Landscape Improvements. Environment and Planning - Part A 2009;41:97–111. doi:10.1068/a4038. –Czajkowski M, Budziński W, Campbell D, Giergiczny M, Hanley N, others. Spatial heterogeneity of willingness to pay for forest management. 2015. –Johnston RJ, Ramachandran M. Modeling Spatial Patchiness and Hot Spots in Stated Preference Willingness to Pay. Environ Resource Econ 2013;59:363–87. doi:10.1007/s10640-013-9731-2. –Meyerhoff J. Do turbines in the vicinity of respondents’ residences influence choices among programmes for future wind power generation? Journal of Choice Modelling 2013;7:58–71. doi:10.1016/j.jocm.2013.04.010. –Rolfe J, Windle J, Bennett J. Benefit Transfer: Insights from Choice Experiments. In: Johnston RJ, Rolfe J, Rosenberger RS, Brouwer R, editors. Benefit Transfer of Environmental and Resource Values, Springer Netherlands; 2015, p. 191– 208. –van Zanten, B. T., Verburg, P. H., Koetse, M. J., and van Beukering, P. J. H. 2014. Preferences for European agrarian landscapes: A meta-analysis of case studies. Landscape and Urban Planning, 132: 89–101. 30

31. Thank you for your time and attention Julian Sagebiel IÖW – Institute for Ecological Economy Research, Berlin julian.sagebiel@ioew.de Klaus Glenk SRUC, Edinburgh klaus.glenk@sruc.ac.uk Jürgen Meyerhoff Technical University Berlin juergen.meyerhoff@tu-berlin.de

32. Estimation Results: Statistics 32 Observations33291 Pseudo R 2 0.103 AIC21890.5 BIC21983.1 Chi Squared2514.1 Lok-Lik. (Null)-12191.3 Log-Lik.-10934.3