Results and analysis script from a discrete choice experiment assessing public preferences for rewilding in the Oder Delta
Data files
Jan 19, 2024 version files 455.75 KB
-
ger_data18.1.24.mat
-
MXL_d_results_pooledGER18.1.24.xls
-
MXL_d_results_pooledPOL18.1.24.xls
-
pl_data18.1.24.mat
-
README.md
Abstract
1. Rewilding is an emerging paradigm in restoration science, and is increasingly gaining popularity as a cost-effective ecosystem restoration option. A rewilding framework was recently proposed that contains three integral components: restoring trophic complexity, allowing for stochastic disturbances, and enhancing species’ potential to disperse. However, as of yet, there has been limited quantitative analysis looking at public preference for rewilding and each of its elements.
2. We used a discrete choice experiment approach to determine public preference for rewilding in the Oder Delta. The unique geographical context of the Oder Delta, spreading evenly across two countries, allowed us to analyze differences between the German (n = 1,005) and Polish (n = 1,066) samples.
3. In both countries, we found respondents were willing to pay for rewilding interventions when compared against a status quo option. Notably, preferences were strongest for restoring trophic complexity through promoting the comeback of large mammals.
4. In addition, we found respondents living locally to the study region had significantly different preferences than the nationwide samples, exhibiting negative willingness to pay for the restoration of natural flooding regimes and the presence of large predator species.
README: Results and analysis script from a discrete choice experiment assessing public preferences for rewilding in the Oder Delta
Paper title:
Public preference for the rewilding framework: a choice experiment in the Oder Delta
Authors:
Rowan Dunn-Capper a,b, Marek Giergiczny a,c, Néstor Fernández a,b, Fabian Marder a, Henrique M. Pereira a,b,d
a German Centre for Integrative Biodiversity Research (iDiv) Halle-Jena-Leipzig, Puschstrasse 4, 04103, Leipzig, Germany
b Institut für Biologie, Martin-Luther-University Halle-Wittenberg, Halle, Germany
c Faculty of Economic Science, University of Warsaw, ul Długa 44/50 00-241, Warsaw, Poland
d CIBIO (Research Centre in Biodiversity and Genetic Resources)–InBIO (Research Network in Biodiversity and Evolutionary Biology), Universidade do Porto, Vairão, Portugal
Summary:
· File count – 5 files
· File formats - .xls, .mat, .md
· File sizes – 108-115 KB
Dataset overview:
· MXL_d_results_pooledGER18.1.24 – .xls file containing the results from respondents to the choice experiment residing in Germany. Results show the estimations from a mixed logit model. Description of variables can be found below. More information on the methodology and interpretation of the results can be found in the manuscript.
· ger_data18.1.24.mat – choice experiment data from respondents residing in Germany stored as a matlab structure. Columns show the outcomes of individuals choice tasks. Respondents are assigned anonymous ids. More information on analysing choice experiment data is readily available online e.g., see Train, K 2009.
· MXL_d_results_pooledPOL18.1.24 – .xls file containing the results from respondents to the choice experiment residing in Poland. Results show the estimations from a mixed logit model. Description of variables can be found below. More information on the methodology and interpretation of the results can be found in the manuscript.
· pl_data18.1.24.mat – choice experiment data from respondents residing in Poland stored as a matlab structure. Columns show the outcomes of individuals choice tasks. Respondents are assigned anonymous ids.
Detailed file information
· ger_data18.1.24.mat & pl_data18.1.24.mat
List of variables:
· id – anonymized respondent id from survey
· cs – choice situation number, there are 12 cs per respondent
· alt – alternative number, there are 3 alt per person
· choice – dummy coding choice (takes 1 if alt was chosen)
· sq – dummy coding status quo alt (=1)
· forest - forest variable
o takes 4 levels. 1 = SQ level, 2 = lowest rewilding level (intensified land use), 3 = middle rewilding level, 4 = highest rewilding level (deadwood on floor)
· river – river variable
o takes 4 levels. 1 = SQ level, 2 = lowest rewilding level (intensified land use), 3 = middle rewilding level, 4 = highest rewilding level (natural flooding regimes)
· agri – agriculture variable
o takes 4 levels. 1 = SQ level, 2 = lowest rewilding level (intensified land use), 3 = middle rewilding level, 4 = highest rewilding level (land abandonment)
· connect – connectivity variable
o takes 4 levels. 1 = SQ level, 2 = lowest rewilding level (intensified land use), 3 = middle rewilding level, 4 = highest rewilding level (eco bridges and road removal)
· lc – large carnivore variable
o takes 4 levels. 1 = SQ level, 2 = lowest rewilding level (intensified land use), 3 = middle rewilding level, 4 = highest rewilding level (both large carnivores present)
· lh – large herbivore variable
o takes 4 levels. 1 = SQ level, 2 = lowest rewilding level (intensified land use), 3 = middle rewilding level, 4 = highest rewilding level (both large herbivores present)
· cost – cost in EUR of choice alternative
· booster – 0 = national representative sample, 1 = booster sample on local residents
More detail on the attribute levels and rationale for specification can be found in the paper
· MXL_d_results files
List of variables:
· SQ – status quo alternative
· Forest2 – second first level (compared against level 1 forest – intensification)
· Forest3 – third first level (compared against level 1 forest – intensification)
· Forest4 – fourth first level (compared against level 1 forest – intensification)
· *Same for *River2, 3, 4 (compared against level 1 river – intensification)
· *Same for *Agri2, 3,4 (compared against level 1 agri – intensification)
· *Same for *Connect2, 3, 4 (compared against level 1 river – intensification)
· Lynx – presence of lynx (compared against level 1, no lynx or wolf)
· Wolf – presence of wolf (compared against level 1, no lynx or wolf)
· LC_both – presence of lynx and wolf (compared against level 1, no lynx or wolf)
· Elk – presence of elk (compared against level 1, no elk or bison)
· Bison – presence of bison (compared against level 1, no elk or bison)
· LH_Both – presence of elk and bison (compared against level 1, no elk or bison)
· -Cost/100(EUR) – cost variable
Model diagnostics – measures model fit
LL – log-likelihood
Detailed description and analysis of results found in paper
Methods
Choice experiment set-up
To determine respondent WTP for rewilding interventions in the Oder Delta, we conducted a discrete choice experiment. The survey was conducted online in both Germany and Poland, with respondents selected by professional online survey companies to provide a representative sample of the respective national populations. The survey consisted of three sections. The first section contained standard socio-demographic questions and also studied respondents’ prior use of the Oder Delta. The second section contained the choice experiment along with follow up questions assessing respondent comprehension and motivation, and the third section included novel motivational questions assessing respondents’ relationship to the Oder Delta and nature more broadly.
We prepared final choice sets for our study using an efficient Bayesian design optimized for Mixed Logit models, using the software Ngene 1.3. Prior values were obtained from a pilot study conducted on a sample of 250 respondents; in the pilot study an optimal in difference design was used (Street and Burgess, 2007). A total of 48 choice situations were generated, which were subsequently grouped into 4 blocks, with each block comprising 12 choice situations. Respondents were presented with a total of twelve different choice situations, each including three alternatives, with one fixed as a “status quo” option (no change in the current management, leading to intensification by 2050). Each scenario was described using two stochastic disturbance attributes, the dispersal attribute, one trophic complexity attribute, and a cost attribute. Attributes levels and icons were carefully selected to be understandable and credible for respondents. As part of this process, we consulted in detail with local experts (including members of the Rewilding Oder Delta local practitioners’ association) and rewilding experts. Focus groups formed part of an iterative process, in which the attributes and icons were consistently refined in response to feedback. Before the choice tasks, respondents first carried out a simple ranking exercise using photographs of the different landscape levels. This element of the survey was designed to help familiarize respondents with attribute levels.
Cost was defined as a new annual obligatory tax paid by all citizens in the respondents’ country for the foreseeable future. It was explained to the respondent that the tax would be used to help finance the landscape management interventions alongside supporting a local rewilding fund that would offset any associated adverse economic impact on local residents of the proposed changes. Care was taken to ensure that the survey design followed the state of the art recommendations of Johnston et al., (2017), presenting respondents with “an incentive-compatible valuation exercise that involves a plausibly consequential decision”. Notably, efforts were made to mitigate against hypothetical bias (Carson et al., 2014) such that the respondent perceived the payment vehicle as binding (if the proposed change goes ahead) and that their decisions could plausibly lead to changes in the attributes being valued (Carson and Groves, 2007).
The master version of the survey was written in English and translated into German and Polish by native speakers.
Choice experiment data collection
The primary test for Germany and Poland was conducted in August and September 2022, resulting in a total of 1,657 completed surveys collected for Germany and 1,514 for Poland. Of these, the survey company collected approximately 500 respondents from each country residing within or neighboring the national state containing the Oder Delta region (e.g., Mecklenburg-Vorpommern in Germany and the West Pomerania Province in Poland). The aim of this was to try and collect respondents living locally to the Oder Delta.
Quality control questions were included in the survey to ensure respondents were reading the questions and responding accurately. The collected data was then cleaned to remove respondents identified as speeders (those below 50% of the median time), those that failed the quality control questions, or those identified to have given protest responses. After cleaning, there were 1,005 completed surveys for Germany and 1,066 for Poland.
Given the nature of this study, and in compliance with university guidelines at the time, ethical approval for this study was granted by the lead supervisor responsible for the project. Care was taken to ensure that all data collected was anonymized, that no personal or sensitive data was collected and all participants were over the age of 18. Individual respondents were only identifiable in the results through a unique numerical ID. Participants took part in the survey voluntarily and by submitting the survey gave consent for their anonymized data to be used. They could back out at any time before submitting.
Choice experiment analysis
We used the data from our DCE to estimate respondents’ WTP for rewilding interventions in the Oder Delta. In a DCE, respondents are presented with a set of available alternatives and asked to choose their preferred option. Discrete choice models are founded in the theories of economic value (Lancaster, 1966) and random utility theory (McFadden, 1974).
We employed the mixed logit (MIXL) model to analyze the data. MIXL models were estimated for the individual rewilding attribute levels of the pooled populations in Germany and Poland and for respondents living locally (rural, within 100km) to the Oder Delta in both countries. For the MIXL model, we used the panel specification proposed by Revelt and Train (1998).
We asked respondents to choose between management alternatives in the Oder Delta in 12 different choice situations. Following cleaning of the data to remove speeder surveys and individuals that failed the comprehension questions, MIXL models were estimated from a representative sample of 1,005 respondents in Germany and 1,066 respondents in Poland. Of these, 108 were classed as local in Germany, and 103 in Poland.