Advances in Choice Experiment for the evaluation of environmental goods and services

Over the last years, Choice Experiment (CE) methodology has increased its diffusion in several environmental contexts. Despite the increasing popularity of the method, there are still aspects that are not fully explored yet. In particular, the research areas explored in the thesis are: i) the analysis of the effect of information treatments in CEs; ii) the development of frameworks to include spatial variables in discrete choice models; iii) the analysis of the effect of individuals’ psychological traits on preferences towards environmental goods and services; iv) the comparison of existing model specifications which allow to account for preference heterogeneity. Most of the research questions were investigated by applying discrete choice modeling to data collected in two case studies: i) the analysis of social demand for landslide protection in Val del Boite (Veneto region), ii) the analysis of the demand of different heating system of households of the Veneto region. The remaining part of the analysis, instead, involved data generated by means of a simulation study. The thesis is organized in five chapters. Chapter 1 introduces the Choice Experiment method, outlines the research objectives and illustrates the case studies. Chapter 2 focuses on the exploration of the effect of information treatments on the stability of preference estimates and it is based on data analysis carried out from the first case study. Preferences were retrieved before and after providing respondents with scientific-based information, based on visual simulations of possible landslide events. This enabled to measure information effects. Choice data were used to estimate a Mixed Logit model (MXL) in WTP space to obtain robust estimates of marginal willingness-to-pay estimates and control for the effect of information. Overall, it was found that mWTP estimates are dependent on information. The geographical distribution of such effects was illustrated by means of maps of willingness to pay values. Chapters 3-4 illustrate analysis carried out on data retrieved from the second case study. Chapter 3 aims to analyze how geographical variables influence individuals’ sensitivity to key features of heating systems. A MXL model was estimated to spatially characterize preference heterogeneity. The results showed that geographical variables are in fact significant sources of variation of individual’s sensitivity to the investigated attributes of heating systems. Thematic maps were produced to illustrate the distribution of willingness to pay to avoid CO2 emissions across the region and to validate the estimates ex-post. Chapter 4 roots on previous theoretical evidence which suggests that beliefs and attitudes of individual consumers play a crucial role in the diffusion of innovative products. A Latent Class-Random Parameter (LC-RPL) model was estimated to analyze preferences of households for key features of ambient heating systems. The model specification allowed to evaluate the coherence of the underlying preference structure using as criteria psychological constructs from the Theory of Diffusion of Innovation by Rogers. The results broadly support this theory by providing evidence of segmentation of the population consistent with the individuals’ propensity to adopt innovations. It was also found that preferences for heating systems and respondents’ willingness to pay for their key features vary across segments. Chapter 5 illustrates the results of a Monte-Carlo experiment aimed at retrieving the required number of parameters and sample sizes to obtain good approximations of true distributions with Logit-mixed logit (LML) models. These models were recently introduced by Train (2016) and are a key advancement in methods to represent the random taste heterogeneity in logit-type models as they generalize many previous parametric and seminonparametric specifications. The performance of LML models are also compared with those of parametric specifications based on normal mixing distributions. The results suggest that LML models outperform parametric models only at large sample sizes. LML Mmdel specifications with large number of parameters outperformed those with small number parameters only at large sample sizes as well. Finally, chapter 6 draws the conclusions of the thesis.

Negli ultimi anni, la metodologia degli Esperimenti di Scelta (CE) si è diffusa in diversi contesti ambientali. Nonostante la crescente popolarità del metodo, vi sono ancora alcuni aspetti non del tutto esplorati. Tra questi, il presente lavoro di tesi si concentra su: i) L’analisi dell’effetto di trattamenti di informazione nei CE; ii) L’inclusione di variabili spaziali nei modelli a scelta discreta; iii) L’analisi dell’effetto dei tratti psicologici degli individui sulle preferenze verso beni e servizi ambientali; iv) Il confronto tra diversi modelli a scelta discreta. La maggior parte delle domande di cerca sono state investigate applicando il metodo a dati raccolti in due casi di studio: i) l’analisi della domanda di sistemi di protezione dalle frane in Val del Boite (regione Veneto); ii) l’analisi della domanda di diversi sistemi di riscaldamento in Veneto. La rimanente parte delle analisi è stata condotta su dati generati con una simulazione. La tesi è strutturata in sei capitoli. Il primo capitolo introduce il metodo degli Esperimenti di Scelta ed illustra gli obiettivi della tesi e i casi di studio. Il secondo capitolo si focalizza sull’esplorazione dell’effetto di trattamenti di informazione ed è basato sull’analisi dei dati raccolti nel primo caso di studio. I risultati ottenuti supportano l’evidenza di un effetto dei trattamenti di informazioni sulle stime ottenute da modelli a scelta discreta. I capitoli 3 e 4 illustrano i risultati delle analisi condotte nel secondo caso di studio. Il capitolo 3 si concentra sull’effetto di variabili geografiche sulle preferenze verso diversi sistemi di riscaldamento, analizzato tramite la stima di un modello Mixed Logit. I risultati evidenziano un effetto significativo delle variabili geografiche incluse nel modello. Nel capitolo 4, un modello Latent Class-Random parameters è stato stimato per analizzare le preferenze degli individui verso i diversi sistemi di riscaldamento. Tale modello ha permesso di valutare la coerenza delle stime con la teoria della diffusione delle innovazioni formulata da Rogers. I risultati supportano tale teoria, mostrando una segregazione della popolazione coerente con la propensione individuale ad adottare prodotti innovativi. Inoltre, i risultati suggeriscono una variazione delle preferenze tra i diversi segmenti. Il capitolo 5 illustra i risultati di una simulazione Monte Carlo condotta per confrontare risultati ottenuti con diversi modelli a scelta discreta. Infine, il capitolo 6 riporta le conclusioni della tesi.