3.1. Method
Due to the lack of real market data, a survey was conducted to collect stated preference data. Consequently, actual behavior was not observed, but respondents were requested to make choices based on several choice tasks, each containing hypothetical choice alternatives [
38]. It is commonly agreed that individuals’ choices reflect their preferences over different alternatives of goods and services [
39]. A set of attributes characterized each choice alternative, and each attribute assumed one of several levels that described ranges of attribute variation across choice alternatives [
38,
40].
Based on random utility theory assumptions [
41,
42,
43], let us posit that the utility
perceived by the respondent
for a given alternative
in the choice task
is a function of
attributes, which characterize the
alternative as follows:
where
is the representative utility formulated by the researcher and
is the error term, which captures factors that affect utility
, but are not included in
[
44].
is usually linear in parameters:
where
is a vector of attributes, which defines the representative utility function and
is the marginal utility of the
attribute. Under the assumption that
follows a type I extreme value distribution, the probability
that respondent
selects choice alternative
in choice task
is expressed in the multinomial logit model (MNL) as follows:
The marginal utility coefficients
are estimated by maximizing the log-likelihood function [
45]:
where
is the total number of respondents,
is the number of choice tasks, which contain
K choice alternatives,
is a dummy variable that assumes the value of 1 if respondent
selects alternative k in the choice task
. A cost attribute, whose marginal coefficient is denoted by
is inserted in the formulation of the representative utility function
to estimate the marginal WTP related to each
-th attribute (
). The marginal WTP of the
-th attribute
is thus estimated as follows [
40,
46,
47]:
For the purpose of our study, the representative utility function
is grounded into two attributes: the former is related to the PV plant typology, and the latter is the cost attribute, coincident with the asset’s hypothetical (contingent) market price. To estimate the marginal WTP for each PV plant typology, attribute levels related to the PV plant typology are dummy coded. The attribute level “NO PV plant” is chosen as the reference level, and thus, its coefficient is set equal to zero [
40,
48]. The representative utility function is:
where
is the alternative-specific constant of the representative utility function,
are the coefficients of the dummy coded attribute levels,
are the PV-plant-typology attribute levels, dummy coded,
is the cost coefficient, and
is the cost attribute.
The experimental design procedure is then implemented to create the choice tasks by combining PV-plant-typology attribute levels and cost attribute levels [
40]. In detail, it adopted the efficient design methodology, which ensures the statistical significance of parameter estimates by simultaneously reducing the sample size to the minimum as possible [
49,
50,
51,
52,
53,
54]. Using the NGENE software, 9 choice tasks were obtained, and the minimum sample size of 80 respondents has been estimated.
3.2. Survey Design and Administration
Our survey aimed to investigate the effect of PV plants’ different technical characteristics on WTP and elicit the WTP for domestic rooftop PV plants for solar homes. In this DCE respondents played the role of the homebuyers of a hypothetical building and faced a set of nine choice tasks. Each choice task involved choosing among two choice alternatives, which refers to the hypothetical building equipped or not with different typologies of PV panels. For each PV plant typology, the WTP was estimated in terms of the additional price (i.e., the market price premium) that homebuyers are willing to pay for solar homes compared to homes not equipped with PV plants. As in [
13], the hypothetical building is a 200 m
2 detached house with a private garden, built in the 90 s. This is one of Italy’s most widespread building typologies [
55,
56], and the asset current market price is EUR 200,000.
We accounted for two attributes: “PV plant typology” and “cost.”
For attribute “PV plant typology,” we considered six attribute levels, namely the absence of PV plant and five different plant typologies (
Table 1). The five PV plants under investigation differed in terms of technical characteristics, such as solar cells typology (i.e., monocrystalline, monocrystalline total black and polycrystalline panels), PV plant surface (i.e., 13, 16, 22, and 26 m
2), installed power (i.e., 3 and 5 kWp) and installation typology (rooftop-integrated and rack-mounted PV plants). The relative rate of CO
2 emissions reduction and cost savings for each PV plant were estimated (
Table 2). It is worth noting that the PV plant selection was discussed in a focus group with a panel of experts (e.g., academics, engineers, and PV plants’ suppliers).
In the contingent scenario, the levels of attribute “cost” represent the asset hypothetical market price and relative market price premium regarding the hypothetical building if not equipped with PV panels. The levels of attribute “cost” were selected according to [
13]. They estimated the building’s market price premium in terms of respondents’ WTP for different PV plants by implementing an open-ended format in a contingent valuation study. They were seven cost attribute levels, and they were selected based on literature, [
13]’s findings, and discussion with the panel of experts [
40]. The lowest cost attribute level corresponds to the current market price of the hypothetical building equal to EUR 200,000; consequently, the lowest market price premium is equal to 0%. This cost attribute level accounts for the event that respondents are not willing to pay an additional price for a solar home, ceteris paribus. The cost attribute levels are reported in
Table 3 and the market price premiums ranges from 3% to 25% (i.e., 3%, 5%, 7%, 13%, 18%, and 25%). The highest market price premium (i.e., 25%) accounts for the growing interest manifested by Italian residents in investing in domestic PV plants [
14], which is expected to increase further in the near future [
57]. The above price premiums are in line with the literature, especially relative to the US context, where real market data are available, and the capitalization effect of cost savings due to PV self-consumption is robustly proven [
29,
30,
31].
The survey questionnaire was structured into five sections. In
Section 1, the research topic and aim are introduced to respondents to increase their knowledge on PV plants and their engagement in the survey and motivate the research. Some information about the effects of climate change and GHG emissions related to energy consumption is presented. The 2030 and 2050 European targets for the mitigation of climate change are briefly described. In addition, respondents are informed that renewables, such as PV solar energy, are a key factor in the energy transition from fossil fuels to renewables and the benefits generated by PV production (e.g., GHG emission reduction, cost savings due to self-consumption, etc.) are listed such as GHG emissions reduction and energy cost savings on energy bills due to self-consumption. In
Section 2, the valuation scenario is presented, and a rendering of the detached house and a detailed description of the asset’s characteristics are provided (
Figure 1).
In
Section 3, the contingent scenario and the PV plant typologies under investigation are illustrated. In detail, it is explained to respondents that they have to play the role of a homebuyer of the hypothetical detached house. In each choice task, they have to choose among a set of two alternatives, in which the typology of installed PV plants (including the absence of any PV plant) and the building’s market price vary.
In
Section 4, instructions to compile the survey and choice tasks are presented (
Figure 2). Finally, in
Section 5, respondents’ sociodemographic characteristics are collected (gender, age, educational attainment, and individual annual income).
The survey was self-administered by computer-assisted web interviewing (CAWI) and was conducted from October 2020–November 2020. The sample was randomly selected by a survey company, which stratified the sample based on the most relevant socio-demographics (e.g., gender, age, educational attainment, and income) related to the 2011 Italian census, and the questionnaire was administered to 400 Italian residents, aged between 18 and 70 years. Finally, a set of 240 fully filled questionnaires was collected and examined.