Does information encourage or discourage tenants to accept energy retrofitting of homes?

We study whether providing information about the consequences of residential energy retrofitting encourages public housing tenants to agree with retrofitting, and how this differs by type of information offered. We run a choice experiment in which tenants select between retrofitting packages that differ on the renewable technology used, the energy bill savings, the corresponding rent increase and the CO 2 reduction. Two subsamples of participants get additional information on the financial respectively comfort-related consequences of retrofitting. We find that an average tenant is willing to agree with retrofitting when energy bill savings are 30% higher than the rent increase. Information on comfort-related consequences of renovations makes people more likely to choose for retrofitting. On the contrary, information on financial consequences reduces the support for retrofitting.


Introduction
Incomplete information and inattention can be important contributors to the energy efficiency gap in the residential sector (e.g.Palmer and Walls, 2015;Sallee, 2014). 1 Uninformed residents might underestimate positive effects of energy efficiency improvements and not adopt energysaving home technologies, even when these are cost-effective.Providing residents with additional information does not however unambiguously increase adoption.Newell and Siikamaki (2014), Davis and Metcalf (2016), Houde (2018) document a positive impact of information on adoption, while for example Allcott and Taubinsky (2015), Fowlie et al. (2015Fowlie et al. ( , 2018) ) find lack of effect.Our paper contributes to this discussion, by studying how the type of information provided matters for the impact of information on adoption decisions.More specifically, we examine the effects of information that addresses two different psychological motives behind environmentally-friendly behavior: a financial and a comfortrelated. 2We focus hereby on lower-income households living in public housing. 3 Our study exploits recent Dutch policy changes that require public housing providers to perform energy retrofitting of their dwellings and to obtain for this consent from their tenants. 4Together with four public housing providers, we run a stated choice experiment in which tenants are offered to choose among two alternative retrofitting packages and non-retrofitting. 5Two randomly assigned treatment groups in the experiment receive additional information on the consequences of the retrofitting.We formulate the information treatments in a way as to appeal to two important motives environmental psychology distinguishes in environmentally-friendly behavior (Steg et al., 2014).The first is financial (monetary savings can be realized), the second is comfort-related (higher comfort level can be achieved in the dwelling).
The renovation packages offered to participants of the experiment closely mimic the real choice situations tenants will face in relation to the retrofitting of their dwellings.First, the packages are composed of the energy measures that are most frequently taken in the Dutch public housing (solar panels; insulation of the floor, walls, roof; substituting natural gas for another energy source, etc.) and second, they have been tailored to the specific dwellings respondents live in.Further, the packages include two financial attributes: a rent increase the housing provider charges following the retrofitting and expected savings on the energy bill.We interpret the former as an upfront investment in energy efficiency a tenant has to make.Then the ratio of the energy bill savings to the rent increase equals the gross monetary return to the investment.The experiment allows us to provide insights into what return tenants require for different retrofitting packages and how information treatments affect this required return and the willingness to accept retrofitting.A random utility model (nested logit) is used as a workhorse.
Our paper is related to several literature streams.The first one consists of studies that examine the effect of information provision on energy-conservation behavior in homes (see Delmas et al., 2013, for an overview and a meta-analysis).Health-related information (Asensio and Delmas, 2016), high-frequency information (Jessoe and Rapson, 2014) and publicly disclosed information (Delmas and Lessem, 2014) are found to reduce residential energy consumption.Among the driving mechanisms for this effect, learning plays a more important role than saliency (Lynham et al., 2016).Providing feedback about own energy use has heterogeneous impacts that depend e.g. on time of the day and age of households (Aydin et al., 2018), while information about the energy consumption of neighbours leads to a mean-reversing energy use behavior (Allcott, 2011).A smaller literature looks specifically at the effect of information on adoption of energy-saving residential technologies and appliances.Newell and Siikamaki (2014), Houde (2018) and Davis and Metcalf (2016) find that providing more detailed information leads to better energy choices than coarse certification.Alcott and Taubinsky (2017), Fowlie et al. (2015Fowlie et al. ( , 2018) ) find lack of effect.Our study contributes to this literature by studying information treatments that aim to stimulate adoption and by comparing treatments that target different behavioural motives behind energy conservation.
Another relevant literature uses stated choice experiments to figure out consumers' preferences and willingness-to-pay (WTP) for energysaving residential technologies.See Lee et al. (2017) for an extensive literature overview and Sundt and Rehdanz (2015) for a meta-analysis.Recent studies not only measure the WTP, but also examine its driving mechanisms, including: financial loss aversion and risk preferences (Bartczak et al., 2017;Fan et al., 2012), uncertainty (Williams and Rolfe, 2017;Daziano and Achtnicht, 2014), cultural impacts (Murakami et al., 2015), recent natural disaster experience (Rehdanz et al., 2017), external control of energy use (Broberg and Persson, 2016).Our study adds to this literature by examining the role of information provision and by focusing on social tenants. 6urther, there is a related literature on the economic returns on green buildings.Eichholtz et al. (2013), Chegut et al. (2016), Qiu and Kahn (2019) find that green office buildings in the UK generate substantial returns, both private for the investors as well as external for other properties in the neighbourhood.Deng and Wu (2014), Deng et al. (2012), Bruegge et al. (2016), Kahn andKok (2014), Walls et al. (2017), Zheng et al. (2012) document a green price premium on new residential developments in different countries.We focus on the returns on residential energy retrofitting which are required by social housing tenants in order to agree with renovations.
Finally, our paper adds to the ongoing public discussions on improving the energy efficiency of public housing and low-income residential sector.This topic is extremely relevant at the moment, for at least two reasons.First, poor households tend to be less informed about the costs and benefits of residential energy choices (e.g.Ghesla et al., 2020).Even when renovations are fully subsidized and strongly promoted, a non-negligible share of low-income households refrains from doing them (e.g.Long et al., 2015). 7We study the behavior of people in a situation when the investment is not free: social tenants have to incur an upfront cost in terms of a higher rent before they can profit from savings on the energy bill.We also examine the effect of information provision on the adoption behavior.Second, in the Netherlands, retrofitting the large stock of low-income public housing has been made a task of social housing providers.To be able to perform this retrofitting however, the providers are required by law to obtain agreement from 70% of their tenants for the renovation packages they propose.Drafting packages to which tenants will agree is thus of high importance for the social housing sector.We provide new practical insights into how to design an acceptable retrofitting package and how to use information treatments in order to increase the adoption rate in the public housing sector.
The rest of the paper has the following structure.In Section 2 we describe the energy retrofitting investments that take place in the Dutch public housing sector.Section 3 provides a theoretical framework.Section 4 describes the experiment and the data.Section 5 reports the results of econometric estimations.Section 6 discusses the policy implications and Section 7 concludes.

Energy retrofitting packages in the Dutch social housing sector
In the Netherlands, two thirds of the housing stock needs to undergo energy retrofitting in the coming decades, in order to achieve climate and renewable energy goals.In the social housing sector this amounts to some two million dwellings. 8In this Section, we describe the energy retrofitting policies currently carried out in the Dutch social housing sector.We will use the details of these policies to shape the stated choice experiment later in our study.
Fig. 1 below illustrates a representative social housing apartment complex subject to retrofitting: a four/six stories high condominium constructed in 1974 using a low energy efficiency technology.It is characterised inter alia by single glass, a low level of insulation of the cavity walls and the roof, and low energy-efficiency appliances.It contains some 100 apartments with a floor surface between 60 and 95 square meter. 9 For energy retrofitting, social housing providers usually choose from four standard approaches developed by the Dutch branch organization for social housing Aedes (2017b).These are described below and are indicated by letters S, M, L and XL.All the four approaches imply renewal of building installations and construction elements of the buildings.They thus not only result in energy savings, but also lead to better comfort and higher safety.Furthermore, in approach XL, natural gas is replaced by other energy sources for both, cooking and heating. 10 Approach S uses the strategy of maximally insulating the dwelling within the original body.This can include: insulation within the cavity walls, ceilings and floors; double HR++ glass, installation of a mechanical ventilation, etc.
Approach M allows for a higher level of insulation as compared to S. This is achieved through installation of a second façade in front of the information about one of the attributes in a stated choice experiment and found no effect.We focus on information that addresses specific motives behind environmentally-friendly behavior.
7 See also Elsharkawy and Rutherford (2018) for Great Britain, Grimes et al. (2012) for New Zeeland, Monteiro et al. (2017) for Italy and Portugal, Dentz et al. (2014) for the US and Tsenkova and Youssef (2012) for Canada. 8The public housing sector is large in the Netherlands, containing one third of the total housing stock of 7 mln dwellings (Statistics Netherlands, 2018). 9This particular building is located in the town of Gracht, south of the Netherlands.Apartments make some 50% of the social housing dwellings in the Netherlands (Aedes, 2017a). 10The Dutch government aims to phase out natural gas production from the Groningen gas fields by mid-2022 (e.g.IEA, 2020).This will increase the price of gas.Approach XL anticipates on this policy.A survey among all social housing providers (Aedes, 2018) suggests that 40% intends to use approach L and 40% approach XL.
I.V. Ossokina et al. current façade. 11Also here double glass and mechanical ventilation can be added.Approach L builds upon the insulation measures from approach S (insulating within the original body of the building) and extends them with investments in energy-efficient appliances, such as: a high efficiency heating boiler, ventilation with heat recovery, solar panels.
Approach XL combines measures used in approaches M and L. This includes: a second façade, ventilation with a heat recovery facility and solar panels.Further this approach provides replacement of natural gas as energy source by a heat pump for heating and by electricity for cooking.
To be able to translate the four approaches into retrofitting packages to be used in the stated choice experiment, a number of steps needs to be taken.Table 1 starts with reporting the technical characteristics of retrofitting packages that are in line with the Aedes approaches, and Table 2 documents the respective electricity consumption, gas consumption and CO 2 emissions.The technical characteristics and the energy performance of the packages have been calculated in collaboration with the experts from real estate advisory agency P-V-M and making use of the VABIsoftware widely used in the Dutch social housing sector (VABI, 2019).The energy performance has been computed for a reference household that lives in a dwelling as specified in Fig. 1.Before retrofitting, the household is assumed to have a yearly gas consumption of 1200 m 3 and the electricity consumption of 2500 kWh.Note that Table 2 reports engineering estimates, neglecting possible behavioural adjustments. 12 Table 2 suggests that the four retrofitting packages reduce the private energy consumption and the CO 2 emissions by 26 to 80%.The monthly energy bill is reduced by 13 to 50%.Investment costs of the packages lie   a We take an average of 1 m 3 natural gas = 10 kWh.
b Energy costs were calculated with the following indicators: gas €0.627/m 3 variable cost and €192/year fixed cost; electricity €0.2/kWh variable cost and €252/year fixed cost.
c We assume that one kWh electricity produces 0.413 kg CO 2 and one m 3 gas produces 1.89 kg CO 2 (Milieu Centraal, 2018).
between 9 and more than 40 thousand euro per dwelling.In terms of both the performance and the investment costs, packages S, M are what the European building performance institute (BPIE, 2011) calls 'moderate renovations', while L and XL are what is called 'extensive renovations'. 13 The social housing providers are allowed to charge the tenants part of the retrofitting investment cost through an increase in the monthly rent, as long as the rent increase does not exceed the energy bill savings.A survey among social housing providers reports that they increase rent with an amount between 30 and 70% of the energy bill reduction (Agentschap NL, 2012).In economic terms, housing providers ask their tenants to accept an upfront investment in energy efficiency (rent increase) and promise a 40%-300% return on this investment (energy bill saving).Furthermore, some social housing providers additionally stimulate tenants to cooperate by offering simultaneous renovations of the dwelling facilities (kitchen, bathroom, and toilet).
To be able to perform retrofitting, housing providers are required by law to get the tenants to agree with the renovation and the corresponding rent increase.In case of single family dwellings each tenant decides for herself.In case of an apartment buildings, 70% of tenants has to agree.If the 70% agreement is not reached, retrofitting cannot be performed.In that case housing providers usually reconsider their plans and draw a new proposal.This costs considerable money and time, and housing providers are keen to prevent these additional expenditures by drafting a package the attributes of which meet the preferences of the tenants.
Based on the discussion above, we can distinguish six important attributes on which the retrofitting packages used in the Dutch social housing differ among each other: (i) the energy retrofitting technology usedsolar panels or insulation; (ii) displacement of natural gas as energy sourceyes or no; (iii) the upfront investment tenants have to carry in terms of the monthly rent increase; (iv) the gross return in terms of the monthly energy bill reduction; (v) the environmental benefit measured as the CO 2 emission reduction; (vi) simultaneous improvement of other dwelling facilities (bathroom, toilet, kitchen) -yes or no.These attributes will be used in the stated choice experiment, see Section 4. Before discussing the setup of the experiment we present in the next section the theoretical model used in the analysis.

Baseline model
In modelling the tenants' decision-making about energy retrofitting, we build upon the random utility theory (e.g.Ben-Akiva and Lerman, 1985).We assume that the utility of living in energy-efficient dwellings is a function of the attributes of the energy retrofitting (the energy-efficiency technology used, the rent increase and the energy bill reduction, the CO 2 reduction, etc.) and a random component that captures the influence of unobserved factors.Further, we assume that the utility of non-retrofitting equals a constant plus a random component.The constant can take a positive value (reflecting e.g. a dislike of the nuisance connected with the renovation process) or a negative value (reflecting e.g. a belief that renovation is crucial to prevent environmental deterioration).When deciding about retrofitting, the tenants compare the utility of different renovation packages with each other and also with the utility of nonretrofitting, and choose the alternative that leads to the highest utility.
Formally we write the choice the individual makes as a vector: x = {m,j}.Here m can take two values: m = 0 indicates non-retrofitting and m = 1 indicates retrofitting, while j∊ J is the set of possible retrofitting alternatives (retrofitting packages).We assume that individual i makes her choice based on the maximization of the following utility. where A jk is the value of attribute k of retrofitting package j, β k is the parameter indicating the utility weight of the attribute k, α 0 is utility of non-retrofitting, ε i0 , ε i1 and ε i1j are individual-specific error terms.
Let us discuss the first term of Eq. ( 1) in more detail.We assume that the utilities of various retrofitting alternatives share a common random utility component (ε i1 ).If this component differs from zero, the utilities of different renovation packages j will be mutually correlated.In this case, if adjustments are made to one of the packages j, this will have a larger effect on the probabilities of choosing for other renovation packages than on the probability of choosing for non-renovation. 14We believe that such a feature of the model captures the actual way decision-making takes place.We will test this assumption econometrically.
To proceed further we need assumptions about the distribution of the error terms.For purposes of analytical tractability we choose for a logit specification of the model.The total errors of each of the J + 1 alternatives (so ε i0 as well ε i1 + ε i1j ) are assumed to be standard Gumbel distributed with variance π 2 6 .The alternative-specific error terms of the retrofitting alternatives ε i1j are assumed to be Gumbel distributed with scale 1 μ and a variance μ 2 π 2 6 .These assumptions are in line with a nested logit structure of the model. 15n line with the utility theory interpretation above, we expect μ ∈ (0, 1] (see e.g.Ben-Akiva and Lerman, 1985).If μ = 1, then the correlation between the utilities of the retrofitting alternatives is zero, the Independence of Irrelevant Alternatives assumption holds and we cannot reject a multinominal logit structure of the model.If μ < 1, the assumption of the independence of irrelevant alternatives can be rejected and alternatives j in the retrofitting nest m = 1 are indeed mutually correlated.The value of μ will be determined empirically, together with the other parameters of the model.
The coefficients in Eq. ( 1) can be estimated in a usual way by using Maximum Likelihood estimators (see Hensher et al., 2015).The predicted probabilities to choose one of the retrofitting packages or the non-retrofitting alternative equals (see Ben-Akiva and Lerman, 1985, chapter 10.3): (2)

Hypotheses on the valuation of the attributes
Using standard economic theory and assuming a rational behavior of tenants, we can form expectations concerning the parameter values in 13 The discussed packages and measures are further well comparable to what is used in other European countries, see e.g.Elsharkawy and Rutherford (2018) for Great Britain.In the US, however, smaller and cheaper energy retrofitting measures are often chosen for application in public housing (Dentz et al., 2014).
Eq. ( 1).Let us look first at the retrofitting alternatives.As discussed in Section 2, the energy retrofitting packages used in the Dutch social housing differ on six attributes: (i) the energy-saving technology usedsolar panels or insulation; (ii) displacement of natural gas as energy sourceyes or no; (iii) the size of the upfront investment (monthly rent increase); (iv) the gross return (energy bill reduction as % of the rent increase); (v) the CO 2 emission reduction; (vi) simultaneous improvement of other dwelling facilitiesyes or no.Take these attributes and levels to be elements of the matrix {A jk } in Eq. ( 1).Parameters β k indicate the relative valuation attached to these attributes.
First, we can expect people to place a higher value on insulation as compared to solar panels, ceteris paribus.While both lead to a lower energy bill, insulation also increases the inside thermal comfort and reduces draught.Second, it is likely that tenants place a positive value on the displacement of natural gas for heating and cooking.This value consists of a private benefit (households might appreciate the safety increase induced by the gas displacement) and a positive externality effect that has been widely covered in the press.The externality arises because gas usage reduction helps prevent further earthquakes in the North of the country that have been brought in connection with gas extraction.Third, the upfront investment (rent increase) should have a negative valuation.Fourth, the gross return should be valued positively.Fifth, because CO 2 emission reduction is an environment-improving positive externality, we expect the tenants to place a positive value on it.Finally, the simultaneous improvement of the dwelling facilities (bathroom, toilet, kitchen) is a utility increasing intervention and should be valued positively.
Let us look now at the parameter α 0 .As usual in discrete choice models, one of the alternatives has to be considered as a reference.Consider this to be the retrofitting package {A Jk }.Relative to this package, the no-retrofitting has a positive expected utility if α 0 > 0 and a negative expected utility otherwise.

Marginal utility effects
To make the estimated coefficients in Eq. (1) interpretable in the monetary space, we rewrite the equation as follows: where Aj1=∆Ej ∆Rj*100 is the gross return in % on the investment in retrofitting package j, that is defined as the ratio of the energy bill savings ∆E j to the rent increase ∆R j multiplied by 100 (see Section 3.2), and A jk * are all other attributes of the package.
We define the willingness-to-pay for an improvement in an attribute A jk * in terms of the percentage points gross return people are ready to forego while staying at the same utility level: (3)

Information treatments and their effects
Steg et al. ( 2014) distinguish three main motives behind the environment-saving behavior: gain (monetary savings can be realized), hedonic (higher level of comfort can be achieved) and normative (it is appropriate to contribute to better environment).In this paper we study the effects of information treatments targeting the former two of these motives.We refer to the motives interchangeably as: gain/financial; hedonic/comfort-related.The first treatment aims at educating people about the comfort-related (hedonic) consequences of the proposed energy renovations.The second treatment educates people about the financial (gain) consequences.
To incorporate the effects of the information treatments in the model, we assume that the utility parameters β k and α 0 can depend on the information about the alternatives a person possesses.Let θ βk , θ α be the additional effect of information.The utility after treatment becomes: We expect that highlighting the higher comfort due to retrofitting will affect the valuation of insulation as compared to solar panels and possibly the valuation of the gas displacement option.The gain information treatment, on the other hand, is likely to affect the valuation of the monetary attributes.It may also have an impact on the utility of the non-retrofitting alternative.

General
The experiment took place in the fall of 2018 through an online tool developed at the Eindhoven University of Technology.It was run in collaboration with four Dutch public housing providers, who sent to a sample of their tenants an invitation to participate, with a link to the tool.The targeted sample of tenants involved residents of dwellings constructed between 1946 and 1989, because these are largely the dwellings due to be retrofitted in the coming decades.The tenants were informed that the housing provider is considering various energy retrofitting options, in order to keep up to the Dutch climate and renewable energy goals, and would appreciate to know the residents' preferences concerning these options.There were two parts in the experiment, each involving another pair of the housing providers.The first subexperiment (September 2018) focused on the effects of the hedonic (comfortrelated) information treatment, the second subexperiment (October 2018) studied the gain (financial) information treatment. 16 When entering the online tool, respondents were randomly assigned to a treatment or a control group.The experiment consisted of three main sections.The first and the third section were identical for the treatment and the control groups.The second section differed by group: there the treatment took place.In the first section respondents had to provide data about their individual and household socio-economic characteristics, as well as answer questions about their environmental preferences and the trust in the housing provider.The environmental preferences and the trust variables were then aggregated into two psychosocial variables (see Appendix A for more details).In the second section, all respondents were offered explanation of the attributes of the energy retrofitting packages they would face in the choice tasks.In addition, the treatment group participants got information on the comfort respectively financial consequences of the retrofitting.Pictures and graphical illustrations supported the message (see Section 4.3 below and Appendix C).The third section presented the respondent eight times with a choice task.For each of these tasks, respondents were asked to choose their preferred option from among two energy retrofitting alternatives and non-retrofitting.Fig. 2 presents an example of a choice task.The next section describes how the choice tasks were constructed.

Choice tasks, attributes and levels
The retrofitting packages offered in the choice tasks, are combinations of the six attributes of the Dutch energy retrofitting approaches distinguished in Section 2. Table 3 reports the levels these attributes take in the experiment.For the most part, the levels follow directly from the 16 Sité (region Doetinchem, East of the Netherlands) and ZOwonen (region Sittard, South of the Netherlands) participated in the experiment with hedonic treatment.Rondom Wonen (region Pijnacker, West of the Netherlands) and Woonlinie (region Zaltbommel, centre of the country) participated in the experiment with gain treatment.definition of the attribute.For three attributes we had to make additional assumptions.These are discussed below.
Upfront investment (monthly rent increase).In the Dutch social sector, rents mostly vary between 200 and 800 euro/month.To ensure comparability between tenants who face different rent levels, we defined the levels of the monthly rent increase as a percentage of the current monthly rent: 3% or 7%.For use in the experiment, these percentages were converted into monetary amounts, computed using the current rent level participants self-reported (See Table B1 in Appendix B for the exact values used).Depending on the reported rent, the monthly rent increases in our experiment lie between 6 and 56 euro.
Gross return (monthly reduction in the energy bill).We defined the gross return to be equal to 110%, 130% or 150% of the rent increase. 17For use in the experiment, the energy bill reductions were also expressed in monetary terms, computed using the participant-specific calculated rent increase (see Table B1 in Appendix B for the exact values).The energy bill reductions lie between 7 and 84 euro per month and are thus in line with the predicted energy bill reductions from the four Aedes packages (Table 2).
CO 2 -emission reduction.In line with the figures in Table 2, we specified the CO 2 reduction as 30% or 60% of the current emission.To make this attribute more intuitive, in the experiment, the CO 2 reduction was also related to the number of trees that is needed to compensate for the damage from residential emissions. 18 The full experiment design includes all possible attribute-level combinations making 2 5 *3 unique retrofitting packages.As usual in stated choice experiments, we exploited the orthogonal fractional design technique (Hensher et al., 2015) which allowed to limit the number of unique packages to 27 and the number of unique choice tasks to 702.Each respondent was offered a random selection consisting of 8 of these choice tasks.

Information treatments
In the treatments the respondents received additional information

Table 3
Attributes and levels in the choice experiment.

Attribute
Level 0 Level 1 Level 2 (1) Energy retrofitting technology Solar panels Insulation (2) Natural gas displacement No displacement Displacement of gas for cooking and heating (3) Upfront investment (monthly rent increase ΔR) 3% of current rent 7% of current rent (4) Gross return on the investment (energy bill savings as % of ΔR) 110% 130% 150% (5) CO 2 emission reduction 30% 60% (6) Bathroom, kitchen, toilet Not renewed Renewed 17 This fits well within the range of the implied monetary returns to retrofitting that the housing associations report (40%-300%, see Section 2).Further note that in line with the Dutch regulations the rent increase in our experiment is always lower than the expected energy savings. 18Approximately 150 trees are needed to eliminate the CO 2 -emissions from the reference dwelling reported in Table 2.A 30% respectively 60% emission reduction implies that 50 to 100 fewer trees are needed for this purpose.
about the consequences of the retrofitting.Two information strategies were designed, one targeting the hedonic (comfort-related) motive behind the energy conservation behavior and the other targeting the gain (financial) motive, see Steg et al. (2014).In the online tool, the additional information -framed in a picture -was placed on the page introducing the attributes of the experiment.In the hedonic information treatment (Fig. 3 left panel), we stressed that the renovation would improve the comfort and thermal climate within the dwelling.In the gain information treatment (Fig. 3 right panel), we stressed that the renovation would lead to net financial savings.Appendix C shows screenshots from the experiment, for the control and the two treatment groups.

Data
Of the approached 4600 tenant households, 688 took part in the experiment and completed the choice tasks (a response rate of 15%).We removed as outliers the few student households as well as people who spent too little or too much time answering the questions (2.5% on both sides of the time distribution was dropped, leaving respondents who spent on the experiment between 4 and 34 min).Of the remaining sample of 608 respondents, 375 come from the subexperiment with hedonic treatment (185 control and 190 treatment group) and 233 come from the subexperiment with gain treatment (115 control and 118 treatment group).
Due to the random assignment to the control and treatment group, we expect no selection on observed attributes between the groups.To be sure, this is tested by performing a balance test; Table 4 reports the results.In both subexperiments, the control and treatment groups have well-balanced covariates.Moreover, in both groups participants' dwellings have a comparable geographical spread across the country.There are 13 municipalities and 57 unique four-digit zip codes in the data, where a four-digit zip code is a small area of one by one km.
We use Table 4 further to assess to what extent our sample is representative of the population of Dutch social tenants.In the Netherlands, more than 2 million households live in public housing.Of these, one third is elderly (65+), 25% is a family with children, almost 60% receives a housing allowance 19 and some 40% has paid work (Ministry of the Interior, 2019).The share households living in single family dwellings is 44% (Companen, 2016).The characteristics of the first sample are very similar to the country average.In the second sample, however, households with paid work, those not eligible for housing allowance and those living in apartments are somewhat overrepresented.Finally, Table 4 reports that an average respondent is ready to pay for improvements in their dwelling, has a medium level of trust in the housing association and a medium-high degree of environmental involvement.

Baseline results
Table 5 reports the estimation results from the choice experiment.Model A was estimated for the control group only; model B documents the effects of the treatments.Because the treatments are done at the provider level, we are worried about the intra-provider correlation and would like to correct for this by clustering the standard errors.The small number of providers (four) does not allow to cluster at the provider level however (Cameron et al., 2001;IEA, 2020).Following Abadie et al. (2017Abadie et al. ( , 2020)), we solve this challenge clustering by location, where a location is defined as a four-digit zip codea statistical unit in the Netherlands covering an area of approximately one by one kilometer. 20 The reported coefficients β indicate marginal changes in utility following a change in the respective attribute level from L0 to L1 or L2.The level L0 is defined for the six attributes (i) -(vi) as follows:

solar panels; (ii) no displacement of natural gas; (iii) investment equal to 7% increase in rent; (iv) gross return equal to 130% of the investment; (v) CO 2 emission reduction by 30%; (vi) no renewal of the dwelling facilities}.
The estimate of the inclusive value μ indicates whether there is correlation between the utilities of the retrofitting alternatives (see Section 3.1).The estimated coefficient α 0 describes the utility of non-retrofitting as compared to that of the renovation package consisting of {A jk = 0}.
Consider first the estimates of μ.In both specifications of the model, its value lies around 0.6.The hypothesis of μ = 1 can be rejected at a 1% level of statistical significance.This means that the Independence of Irrelevant Alternatives hypothesis can be rejected, the utilities of the retrofitting alternatives are mutually correlated and the nested structure of our model is supported by the data.
Let us now discuss the results for the control group, see model A. The signs of all the coefficients are intuitive and most coefficients are statistically significant.Ceteris paribus, a lower rent increase (lower required upfront investment) increases utility, while a lower return on this investment (smaller ratio of the energy bill savings to the rent increase) reduces utility.Further, a higher emission reduction, as well as the improvement of dwelling facilities have a positive utility effect.Finally, although the non-renovation constant α 0 is not statistically significant, its large negative value suggests that an average respondent prefers the

Hedonic treatment Gain treatment
Your dwelling is more comfortable and safe!-Less draught and noise nuisance, less oŌen cold feet.
-Clean and healthy air in the dwelling.
-Improved fire safety in your dwelling.
You are going to pay less! -Your energy bill will become lower.
-The rent will become a bit higher.
-We guarantee that you will save money.
Fig. 3. Treatments. 19Households with an income below 30.000 per year were eligible for a housing allowance. 20In our data, there are 57 unique zip codes and a one-to-one correspondence between a zip code and a housing provider, see also Section 4.4.

Table 4
Balance test for the differences between treatment and control groups.package with {A jk = 0} to no renovation.This is confirmed in model B where we can rely on a larger data set for the estimation of the coefficients.
We will now express the marginal utility effects β in monetary terms (see last column of Table 5).Remember that the energy bill savings in our model are defined as a percentage return on the upfront investment: the savings equal 110%, 130% (reference) or 150% of the rent increase.Using Eq. ( 3) we can express the estimated coefficients on other attributes in terms of the percentage points return that tenants are willing to forego (require as compensation) when the attribute level changes upwards (downwards).For example, if offered a package with insulation, tenants will require a 10 percentage point lower return than on an identical package with solar panels, to achieve the same utility (so e.g.120% instead of 130% return).To get a package with a double as large emission reduction, tenants are willing to forego 20 percentage points gross return.A renewal of the kitchen, bathroom and toilet is worth foregoing 35 percentage points monetary return on the retrofitting.
Consider now the effects of the treatments.In model B, for each attribute-level combination, three coefficients are reported.These are: the baseline coefficient β for the control group, and the two additional effects θ hedonic and θ gain for the treatment groups. 21The utility impact of changing the level of an attribute from L0 to L1/L2 equals for a treatment group thus the sum of the baseline and the additional effects (see Eq. ( 4)).Look first at the hedonic treatment that provides the tenants with additional information on the positive comfort-related consequences of the energy retrofitting.The estimation results suggest that this makes the insulation option more valuable, the utility effect is higher than for the control group and the effect is statistically significant and large.This is in line with the expectations: insulation not only yields energy savings, but also reduces draught and increases comfort.At the same time, we find no statistically significant change in the valuation of the natural gas displacement.This might indicate that the valuation of the gas displacement, if any, mainly comes through the positive externality mechanism (reduction of negative consequences of natural gas extraction in the North of the country) and not through the expected increase in individual comfort and safety.Turn now to the gain treatment, which provides the tenants with additional information on the expected positive financial consequences of the retrofitting.Our estimations suggest that this treatment makes the respondents more critical in evaluating the retrofitting options.For the treated group, the nonretrofitting constant (the sum of the baseline and the additional effect) is not statistically significant indicating that the respondents are indifferent between the package {A jk = 0} offering a 30% return on the energy investment, and non-retrofitting.In the control group however, the package {A jk = 0} is clearly preferred to non-retrofitting.
To better understand the financial gain effect it is important to note that whereas for a tenant the rent increase is certain, the savings on the energy bill are uncertain.The energy savings are an estimate offered by the housing provider who has an interest in convincing the tenant of the attractiveness of the package.In the financial information treatment, this financial aspect is made more salient, priming the participant to think more about it.It likely triggers secondary thoughts on the trustworthiness of the estimate and the uncertainty surrounding it, thus reducing the perceived value of the presented financial gain.
The positive effect of comfort-related information treatment on adoption is in line with the results of Newell and Siikamaki (2014), Davis and Metcalf (2016), Houde (2018) who find that more detailed information positively affects adoption.While the mentioned papers focus on the impact of the energy efficiency labelling, we contribute by studying another type of information, that on the comfort-related consequences of the retrofitting.The result that financial information might negatively affect adoption is new in the literature on energy efficiency, to our knowledge.However it is well in line with papers from behavioural finance that examine how financial information disclosure affects the investment decisions.For instance, Linciano et al. (2018) finds that higher perceived complexity of financial information increases the perceived riskiness of the product.This mechanism might well be in place in our case as well: additional emphasis on financial information may increase the perception of the riskiness of the retrofitting investment and of the uncertainty surrounding the monetary return.

Sensitivity and heterogeneity analysis
In this section we extend the above analysis.To start with, a number of sensitivity checks are performed.Then we allow for heterogeneity in the valuation of the attributes and in the effects of information treatments.
The results of the sensitivity checks are reported in Appendix D. First, it might be possible that the energy bill savings are valued differently conditional on the retrofitting technology used.Solar panels do not change the marginal cost of heating the dwelling to an agreeable temperature, they just offer a certain quantity of electricity 'for free'.The behavioural effect is therefore comparable to a lump-sum income allowance.Insulation, on the other hand, reduces the marginal cost of heating the dwelling.So the utility equivalent of the energy savings might be higher in the case of insulation.We test for this introducing a cross-effect between the retrofitting approach and the gross return on the investment (see model (1) in Table D1).The results suggest that the valuation of the energy bill reductions is quite similar across the two technologies.Second, in our experiment the hedonic and the gain treatment were applied to two separate groups of tenants.We want to test if the results stay robust when the effects for the two groups are estimated separately.Models (2) and (3) in Table D1 report the results of these two estimations.We find no statistically significant differences with the coefficients of the pooled regression reported in Table 5.
Let us now discuss the implications of possible heterogeneity among the tenants.Literature suggests that people may be heterogeneous in their propensity to invest in energy efficiency and the way they respond to information treatments (e.g.Aydin et al., 2018).First, we expect highly educated (people with a bachelor or a master grade) as well as people with a higher concern for the environment to be more inclined to show environmentallyfriendly behavior.Second, trust in the housing provider is likely to determine how much people believe the information provided about expected energy bill savings, thus tenants with a higher trust level can be expected to be more willing to agree with the offered energy retrofitting packages.
In Table 6 we report specifications in which for all coefficients of Eq. (3), interaction effects are added with: a standardized value of the score on trust in the housing provider (Model (A)), a standardized value of the score on environmental concern (Model (B)), a dummy for high education (Model (C)).For trust and environmental concern, the interaction coefficients can be interpreted as the effects of a one standard deviation change in the score, for educational level the coefficients reflect the effect of having high education as opposed to not having it.
It is reassuring that the main insights do not change.Furthermore, the utility of the non-renovation has a more negative value for people with higher environmental preferences and highly educated.The coefficient for higher trust is negative and large as well, although not statistically significant.This is in line with the hypothesis that these groups have a higher propensity to invest in energy efficiency.The gain treatment reduces or eliminates this effect however, just as it did for the full sample.Further, people with high environmental preferences turn out to be indifferent between insulation and solar panels, while high educated even prefer solar panels to insulation.Apparently, for these two groups additional comfort insulation provides is less of an issue.Environmentally concerned tenants attach a high positive value to the displacement of natural gas.

Implications
In this section we use the outcomes of the stated choice experiment to simulate the share of the tenants that will be willing to adopt retrofitting packages S to XL introduced in Section 2, with and without the information treatments.Remember that in the Netherlands, to be able to perform retrofitting, housing providers need to obtain consent of at least 70% of the tenants.
Table 7 describes the packages S to XL in terms of the attributes used in the stated choice experiment.The attributes from the experiment were matched to the packages in the first place based on the monetary value of the energy savings, this for the mean rent of 550 euro/month in our data. 22For different possible packages, we compute the shares of tenants choosing for retrofitting and choosing for non-retrofitting under four scenarios: (i) control (no information treatment, no renewal of dwelling facilities), (ii) hedonic treatment, (iii) gain treatment, (iv) control with renewal of facilities (bathroom, kitchen, toilet).Facilities renewal is an improvement in comfort, we compute its effect in a separate scenario to compare it with the effects of information treatments.Fig. 4 reports the average shares of tenants willing to renovate and choosing not to renovate, in these four scenarios.Table 8 reports the descriptive statistics of these shares across the simulations.
Our analysis suggests that the required threshold of 70% tenants agreeing with energy retrofitting can be reached for the renovation packages used in the Dutch practice, although not in all cases.Provision of additional information to tenants can increase the share of supporters, but also decrease it, depending on the type of information offered.When comfort-related consequences of retrofitting are stressed (hedonic treatment), people become more likely to participate.Providing more details on financial effects (gain treatment) decreases the propensity to cooperate.In practice, financial information needs to be provided to tenants in any case.Our results suggest that framing it together with an explanation of comfort-related consequences of retrofitting may be optimal in order to maximize the support among tenants.Finally, bundling energy retrofitting with (planned) renewal of the dwelling facilities can substantially increase the tenant's propensity to cooperate.Well-designed information treatments together with facilities renewal can thus lead to a sizable reduction in the number of non-supporters of retrofitting.In general, by optimally using the information treatments, the share of non-supporters can be reduced with 3-5 percentage points.The effect of the renewal of dwelling facilities (bathroom, kitchen, toilet) is twice as large.

Summary and conclusion
We have studied how providing different types of information on the consequences of energy retrofitting affects the willingness of social tenants to agree with retrofitting.We exploited Dutch policy changes that require public housing providers to improve the energy performance of dwellings and to obtain a consent of tenants for this retrofitting.In a stated choice experiment run together with four public housing providers, some 600 tenants were offered retrofitting alternatives that closely mimic the actual retrofitting packages used in the Dutch public housing.Two randomly selected groups of respondents got additional information on comfort-related and financial consequences of retrofitting.
Our results suggest that, on average, tenants are willing to cooperate with retrofitting, also when they have to pay for it through a rent increase.An average tenant in our data prefers solar panels to no retrofitting when the expected savings on the energy bill are 30% larger than the rent increase.Retrofitting measures that positively affect residential comfort (e.g.insulation) or environmental savings (e.g.displacement of natural gas as energy source) increase the willingness of tenants to cooperate.However, obtaining support of 70% of the tenants as required by the Dutch law does not go without saying and depends very much on the package chosen.
We find further that information provision affects the tenants' choice, and the impact can be both, positive or negative.Providing tenants with additional information on comfort-related consequences of the retrofitting increases the willingness to cooperate.Information on finance-related consequences, on the other hand, makes tenants more critical and reduces the support for retrofitting.The effects amount to a 3-5 percentage points change in the share of respondents that supports retrofitting.
Our methodology has allowed us to both, obtain insights into the retrofitting priorities of the low-income tenants and the effect of information provision on their adoption behavior.Concurrently, the stated choice approach we used to study preferences, has a number of known caveats.First and foremost, it is not incentivized.Agreeing to one of the packages is not binding, so there are no real financial consequences when participants agree to the retrofitting.Respondents might give poorly thought out answers because the situation is hypothetical (hypothetical bias), over-or understate their true valuation (strategic bias).In this sense, the gain treatment might be seen as reducing the biases as it primes participants to think more about the financial aspects.It makes the increase in rent more salient, despite the net financial gain in retrofitting.It is reassuring that, also in the gain treatment, a large share of the respondents is willing to cooperate.Still, it is important to keep these limitations of the methodology in mind when using the results.
The insights obtained contribute to the scientific literature about the role information provision can play in stimulating adoption of residential energysaving technologies and the mechanisms through which this happens.At the same time, our results ask for further research.Steg et al. (2014) distinguishes three drivers of sustainable behavior: financial (money), hedonic (comfort) and normative (care for the environment).Our experiment has yielded novel information on the impact of information affecting the former two drivers.It would be a useful extension to analyse the extent to which providing information about environmental aspects of energy retrofitting affects adoption.Furthermore, additional insights into the differences between the tenants might be desirable, among other things to fine-tune the communication strategy to the recipient.
Our paper also adds to the ongoing public discussions on improving the energy efficiency of public housing and low income residential sector.Existing studies show that even when renovations are fully subsidized and strongly promoted, a non-negligible share of low income households refrains from doing them (e.g.Long et al., 2015).Our study provides new practical insights in how to construct retrofitting packages and how to use information treatments in order to increase the adoption rate in the public housing sector. 23

Declaration of Competing Interest
None.   23 The Dutch public housing providers that participated in the experiment have been using the insights from this study to improve their communication strategy concerning energy retrofitting.

Fig. 1 .
Fig. 1.Example of a building to be renovated.

Fig. C1 .
Fig. C1.Screen shot for the control group.

Fig. C3 .
Fig. C3.Screen shot for the gain treatment group.

Table 1
Four template retrofitting packages for Dutch public housing.

Table 2
Energy performance: reference and four packages.
Here the demographic, socio-economic and psychosocial characteristics concern the household member who participated in the experiment.b Graduated from a university or a university of applied sciences.
a c See Appendix A for more details on the calculation of these variables.

Table 5
Estimation results control and treatment groups.
a So the energy savings equal ΔE = 1.1 * ΔR.22 Table2in Section 2 reports the estimated monetary energy savings from the four Aedes packages.Table A1 reports the monetary value of energy savings that tenants were offered in our experiment, depending on their reported rent level.

Table 8
Simulated shares of tenants refusing the retrofitting.

Table 7
Four renovation packages in terms of experiment attributes.

Table D1
Sensitivity analysis.
(continued on next page) I.V.Ossokina et al.