Heuristic decision-making

Hypothesis 3 suggests that an increase in the number of years of payback, due to an increase in either the upfront costs or a fall in annual savings, would decrease the household’s utility, and consequently decrease the probability of a household connecting to DH.

The last empirical adjustment utilises the payback period, due to its simplicity, to reflect or at least approximate the perceived risk calculated on the basis of current costs and expected annual savings by the households (Kempton and Montgomory,

1984). Arriving at the final model, the payback variable (PBi) is added to the

households’ (latent) utility function, represented thus far by Equation (5), as follows:

U_i=β₁UC_i +β₂AC_i+ δ_j IN_1i 3 j=1 + λ_j IN_2i 3 j=1 + ξ_j PB_i + 4 j=1 X_i'γ +ε_i (6)

Payback (in years) is defined as:

Payback _i=UCi Si =

UC_iDH ACiC- ACiDH

where UCi denotes the upfront cost for the installation of a district heating (DH)

interface unit. Si represents the expected annual savings – calculated by the taking

difference between annual costs (ACi) accrued by installing district heating (DH) and

remaining with the current heating system (C). It is important to note that the former refers to the values randomly presented to the households during the survey whereas

the latter represents the households reported energy bills and maintenance costs of the

current heating system27 (Table 6 and 7).

One potential issue related to the calculation of payback period arises due to the fact that around 100 households are ‘unsure’ of their energy bills. Following Palmer and Walls (2015) the present paper uses this response to control for inattention towards household energy consumption by including an indicator variable equal to 1 if the

household reported their energy bill and maintenance costs, and 0 otherwise (DK-

ANNUAL COSTS). Therefore, an estimate of the expected savings can only be used to calculate payback for the households who are attentive to annual energy bills.

An additional modelling issue arises as a result of negative savings. Obtaining individual energy consumption levels for each household prior to the survey was infeasible. To circumvent this problem a real life decision making scenario (the vignette) was simulated, randomly allocating a district heating annual heating bill estimated for an average user to each household. Due to the random allocation and varied energy consumption levels, households can potentially save money by remaining with their current heating system. Hence, one is unable to calculate the

27 _{Two additional estimates were necessarily made due to 1) households paying for their energy using a}

dual-fuel tariff, and 2) due to households using electricity as their sole energy source. For the former an Ordinary Least Squares regression (controlling for household characteristics, housing characteristics, heating needs and structural issues) is utilised to predict the proportion of each households’ dual fuel bill typically spent on gas. Using the ratio of gas expenditure to total energy expenditure for all households who provided information on their annual expenditure for both gas and electricity as the dependent variable. For brevity this regression is not reported but can be made available on request. For the latter, the same information is used to calculate the median proportion of a household’s energy bill spent on gas, for those who use both gas and electricity (conditioning on their energy usage which is based on the number of hours the household typically turns the heating on during the winter), in order to estimate the total amount a household would spend on gas, for households who solely use electricity, if they were to switch to a gas and electric system. It is assumed for all such households that if they were to connect to DH they would continue to use electricity for all other goods and services except heating.

number of payback years for the households with negative savings (431 in total). The impact of negative savings is controlled for by including an indicator variable equal to 1 if savings are less than zero and 0 otherwise (NEG-SAVINGS).

The payback variable (PBi) has been log transformed to help to control for right skew

and potential outliers. In addition, a categorical variable is created by separating the log of the payback distribution into quartiles (i.e. four equally sized categories ranging from low (PBK-L) to high (PBK-H)) to pick up any non-linearity between the households’ decision to participate in a DH scheme and the payback period. This is based on Anderson and Newell’s (2002) observation of a tendency for firms to

become increasingly willing to adopt energy efficient technologies after the payback

period exceeds 9 years. The authors suggest that this behaviour is driven by the dearth of information signalled by relatively long payback periods.

The probability of a household choosing category j within this paper’s fully nested specification, which embodies both inattention proxies and a heuristic measure of profitability in the classic framework, is defined as follows:

P(D_i= j) = Φ(α_j-β₁UC_i-β₂AC_i- δ_j IN_1i 3 j=1 - λ_j IN_2i 3 j=1 - ξ_j PB_i - 4 j=1 Xi'γ) – Φ(α_j-1-β₁UC_i-β₂AC_iβ₁UC_i-β₂AC_i- δ_j IN_1i 3 j=1 - λ_j IN_2i 3 j=1 - ξ_j PB_i - 4 j=1 X_i'_γ)