Measuring Renewable Energy Externalities: Evidence from Subjective Well-being Data

Data

Our data come from several sources. The data on life satisfaction along with information on the respondents’ socioeconomic situation are provided by the German Socioeconomic Panel Study (SOEP) of the German Institute for Economic Research (DIW); see Schupp et al. (2014). The SOEP survey has been conducted on a yearly basis since 1984. Annual waves of the survey include more than 20,000 individuals in about 11,000 households. With respect to the spatial dimension, SOEP data are identified by respondents’ postcode district from 1993 onward (Wagner, Frick, and Schupp 2007).⁴

An important attribute of the SOEP is its panel structure (i.e., that the same individuals are reinterviewed each year), which facilitates analysis of changes on the individual level and controlling for unobserved time-invariant characteristics (Andreß, Golsch, and Schmidt 2013). As respondents may join the panel at a later stage (late entry), or drop-out in a single wave (temporary nonunit response) or permanently (panel attrition) (e.g., due to refusal, death, or relocation), the set of respondents is slightly changing over time; that is, the data are unbalanced (Andreß, Golsch, and Schmidt 2013).

The dependent variable in our life satisfaction regressions is the answer to the following question: “How satisfied are you at present with your life, all things considered? Please respond using the following scale, where 0 indicates ‘not at all satisfied’ and 10 indicates completely satisfied’.” Figure 2 shows that life satisfaction exhibits no apparent trend over time.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 2

Development of Happiness across Time

Source: Data from SOEPv29 (Schupp et al. 2014).

The dataset used in this paper refers to the waves 1994–2012 and includes 324,763 observations for 46,678 individuals. The variables of the SOEP used in this analysis are summarized in Table A2 in the Appendix.

As to the energy data, the four German transmission system operators (TSOs)—Amprion, 50Hertz, Tennet, and TransnetBW— provide data on all RE plants that come under the Renewable Energy Act.⁵ Though the four sources are different with regard to their comprehensiveness, they all give information about the postcode district (five-digit hierarchical system) where the plant was installed, its type of technology (wind, solar, biomass, hydro, geothermal energy, as well as landfill, mine, and sewage gas), the commissioning date, and the installed capacity. Unfortunately, there is no information that would help us to distinguish between rooftop solar and freestanding installations.⁶ Nor can we obtain information about the sort of biomass plant, which may have implications for odor nuisance because of different materials being combusted. For data on wind energy, a different source was used (BDB⁷), which gives concise information on construction of wind turbines in Germany.⁸ The energy variables used in this analysis are summarized in Table A3 in the Appendix. From the summary statistics we can tell that wind energy has the greatest installed capacity of the three considered RE technologies and that most respondents have solar installations in their neighborhood. The standard deviations indicate that there is a considerable variance across observations.

We matched the data on RE plants and the socioeconomic data of the SOEP on the basis of the respondents’ postcode district. Moreover, as the exact dates of the interview and the plant commissioning were available, we could identify for each respondent the number of plants and capacity installed per type of technology by the time of the interview.⁹ For each wave during 1994–2012 the final dataset gives information about the respondents’ socioeconomic situation (see Table A2) as well as the presence of RE technologies in the respondents’ postcode district (see Table A3). In Germany, there are about 8,200 postcode districts, comprising an area of 44 km² on average.¹⁰ In order to take account of possible spillover effects, we widened the spatial scale to include RE plants of neighboring postcode districts. Information from the open source platform OpenStreetMap was used to identify for each postcode district the adjacent postcode districts.

General Methodological Issues

Our approach to measuring externalities from RE involves approximating utility by data on subjective well-being, specifically, life satisfaction. Though this approach relies on subjective data, a major feature of this method is that it does not rely on people’s stated attitude toward or stated evaluation of the issues under study. Instead, life satisfaction data are being elicited independently of those issues, and it is the purely statistical association between life satisfaction and the independently measured variables of interest that is taken as a measure of preference.

Assumptions necessary for using reported life satisfaction in longitudinal analysis are a positive monotonic relationship between life satisfaction and the underlying true utility, and ordinal intrapersonal comparability (see Ferrer-i-Carbonell and Frijters 2004; Faßhauer and Rehdanz 2015). This means that if the satisfaction score at time t is greater than at time t ′, this reflects the same ranking of underlying utility. If, more restrictively, it is assumed that differences in satisfaction scores are proportional to differences in underlying utility, life satisfaction can be treated as a cardinal variable. Using SOEP data, Ferrer-i-Carbonell and Frijters (2004) found that assuming the data to be ordinal or cardinal and applying the corresponding estimation methods has little effect on qualitative results. In particular, the ratios of coefficients are similar, which is important for monetary valuation. Similar results were obtained by many others. In addition, Ferrer-i-Carbonell and Frijters (2004) stress the importance of including individual fixed effects (thereby controlling for time-invariant characteristics such as personality traits), as their omission biases the results substantially. In contrast to estimators for cardinal data (least squares), there is no consensus as to appropriate methods for implementing individual fixed effects in ordered regression models (Baetschmann, Staub, and Winkelmann 2015).

Econometric Approach

We estimated microeconometric life satisfaction functions in which the self-reported life satisfaction (LS) of individual i in postcode district s and year t depends on indicators of renewable energy (RE) in her postcode district, income, and a standard set of time-variant sociodemographic controls (age squared, number of children in household, internal migration, living in property, health status, partner status, employment status, person in household needing care). Time-invariant factors are implicitly captured through person-specific fixed effects. In addition, we use region-year fixed effects involving the states (Bundesländer) of Germany. The estimating equation can be stated as follows: [1] where person_i and region_year_rt denote person and region-year fixed effects, respectively, and ε_ist denotes the error term. Person fixed effects control for unobserved time-invariant characteristics of the individual (such as personality traits), whereas region-year fixed effects control for time-varying unobserved factors common to a particular region-year (such as, e.g., non-RE power generation or regional unemployment rates). The income variable is specified as the net monthly household income adjusted for inflation and equivalized according to the OECD-modified scale. As is common in the well-being literature, income is included in logarithmic form to account for decreasing marginal utility.

Equation [1] is estimated for RE referring to wind power, solar power, and biomass, respectively. Several alternative indicators are used for RE. One indicator is a dummy variable that takes the value 1 if at least one of the respective RE plants exists in the respondent’s postcode district and 0 otherwise. Alternative RE indicators are the number of plants and the installed capacity. It should be noted that over time the RE dummy variable changes its value only once (from 0 to 1), at the time of the first installation (unless the first installation took place before the period of observation or there is none).¹¹ In contrast to the number of units and the installed capacity, the coefficient of the dummy variable hence measures the effect of presence versus nonpresence of RE installations.

In addition to RE in a respondent’s own postcode district, we include the corresponding RE indicator in the neighboring (adjacent) postcode districts. This specification serves to measure the existence and strength of spatial well-being spillovers. Given that the presence of RE installations in the own district and in adjacent districts are correlated (at r = 0.87, r= 0.54, and r=0.66 in the case of solar, wind, and biomass, respectively), omission of the adjacent districts may imply that effects from adjacent districts are falsely attributed to the own district.

In order to study the dynamics of RE externalities, we extended Equation [1] to include dummy variables that represent leads and lags of the first installation of RE units. Specifically, letting T denote the year of the first installation, D(T + i) is a dummy variable that takes the value 1 in T + i and 0 otherwise, where i= ≤ -3, -2, -1, 0, 1, 2, and i ≥ 3. The analysis is restricted to those individuals who actually experienced an RE installation in their postcode district. This reduces the number of observations to 94,324 for solar, 29,830 for wind, and 71,252 for biomass.

We minimize the risk of omitted variable bias by controlling for the observed timevarying life satisfaction factors known to be relevant (see Dolan, Peasgood, and White 2008 for a review) as well as for unobserved person-specific factors (through the fixed effects modeling framework). Though life satisfaction is likely to be measured with error, there is no reason to expect that measurement error is correlated with our independent variables of interest. Finally, including person fixed effects is an effective way of dealing with reverse causation in life satisfaction regressions (Ferrer-i-Carbonnel and Frijters 2004).

Following Ferrer-i-Carbonell and Frijters (2004), we treat the dependent variable, 11-point life satisfaction, as a cardinal variable and estimate Equation [1] and variants thereof using least squares. We report robust standard errors corrected for clustering at the individual level.

Main Results

Table 1 presents the estimation results for the main variables of interest, whereas more detailed results, including those for the control variables, are reported in the Appendix.¹² Three alternative specifications are reported that differ according to whether the RE installations are captured by a dummy variable, by their number, or by installed capacity. We note that the coefficient estimate for the dummy variable measures the difference between those district-years in which there was no RE and those in which there was. The coefficient thus picks up both the effect of the initial installation and also the effect of subsequent expansions.

With respect to solar power, we find non-significant coefficients in the own district for the dummy specification as well as for the number and capacity, whereas the number and capacity in the adjacent districts have weakly significant and significant negative coefficients, respectively. The presence of solar installations (dummy specification) in adjacent districts has a weakly significant positive coefficient.

With respect to wind power, the coefficient on the dummy variable in the own district is significantly negative, whereas the number and capacity have weakly significant negative coefficients. The wind power variables in the adjacent districts are nonsignificant.

In the case of biomass, the presence of at least one biomass plant (dummy specification) in both the own district and in the adjacent districts has significant negative coefficients. The numbers of plants in the own and adjacent districts are insignificant, but the installed capacity in the own district as well as that in the adjacent districts have weakly significant negative coefficients. The coefficients on the dummy and the capacity in the own district are greater than their counterparts in the adjacent districts, and the coefficient on the capacity in the own district is almost four times as large as its counterpart for wind power.

We conducted a falsification test regressing capacity expansion in T+4 on life satisfaction in t using the same empirical model. The coefficients of the RE variables turn out insignificant. As this procedure reduces the number of observations, we tested whether the results of our original model hold up, which is the case.

We thus find that the well-being effects of the technologies considered—solar, wind, and biomass—differ qualitatively and quantitatively. In particular, there are differences (1) with regard to whether effects originating from the own or the adjacent districts matter, (2) with regard to whether the mere presence of installations or their number and capacity matter, and (3) with regard to effect sizes. These differences can be explained in terms of the following features of the technologies: (1) the specific impairments through which the various technologies affect well-being (spatial scope of impairments and the possibility to alleviate impairments by averting behavior), (2) the ownership structure, and (3) the timing of initial installations and expansion (bulk vs. continuous expansion). With respect to the latter point we note that on average, the number and capacity of both wind and biomass plants in a district increased by a factor of 1.6 within the first three years. The number of solar installations increased by a factor of 4.6, while capacity increased by a factor of 6.6.

For solar power, well-being effects arise only from solar installations in adjacent districts. The absence of effects from solar installations in one’s own district suggests that at the level of the postcode district, benefits to private owners (residents and farmers) may offset the negative externalities. Benefits to the owners may be of a financial (reduced electricity expenditures and/or feed-in revenues), moral (warm glow), and psychological nature (status effects associated with the presence of solar panels on ones’ rooftop, see Sonnberger 2015). With respect to adjacent districts, an unexpected finding is that the dummy specification yields a weakly significant positive coefficient. A possible explanation for this finding may relate to the fact that a considerable number of solar parks are owned by citizen energy cooperatives. Assuming that citizen-owned solar parks are located not only in people’s own district but also in adjacent districts, people may enjoy the financial, moral, and psychological benefits from those, while being less affected by the associated impairments than if the installations are in closer proximity. In contrast to the dummy, the number and capacity have (weakly) significant negative coefficients. This suggests that the dominance of benefits over impairments may be a matter of size, becoming reversed as the scale of installations gets bigger.

For wind power, well-being effects arise only from installations in people’s own district. There do not seem to be spatial externalities from neighboring postcode districts. These findings are explicable because wind power installations differ in important ways from solar installations: They are more often owned by external investors and utilities than solar installations, such that there are fewer local benefits than in the case of solar power. In addition, citizens’ participation in the decision process is often limited. Moreover, visual impairments can be avoided by averting behavior, in particular when installations are not in close proximity. This may explain the absence of spillovers from adjacent districts. Acoustic impairments depend on proximity and are thus also less liable to spatial spillovers. In addition, while the dummy specification yielded a significant coefficient, the coefficients on the number and capacity are only weakly significant. This is consistent with the fact that initial installations are bigger and the expansion rate after the initial installation is much slower in the case of wind power than in the case of solar power.

In the case of biomass, effects arise from plants both in the own district and in adjacent districts, and they are highly significant in the dummy specification but weakly significant with respect to the capacity and insignificant with respect to the number. The generally negative effect of biomass plants reflects the circumstance that benefits accrue only to a small number of farmers, whereas the odor nuisance affects everybody and is difficult to avoid by averting behavior. Moreover, depending on the wind conditions, odor nuisance may be less related to proximity than are the visual impairments from wind turbines. The result that the dummy variable is significant whereas the numbers are insignificant and the capacity is only weakly significant reflects the circumstance that expansion rates after the initial installation are low.

With respect to effect sizes, we found that 1 MW of installed capacity in one’s own district has an effect that is almost four times as large in the case of biomass than in the case of wind power. This finding may be related to odor nuisance being more difficult to avoid than the mainly visual impairments from wind turbines. This finding is consistent with the result obtained in a cross-national setup that electricity from biomass is the most disliked form of electricity generation not only in comparison to other renewable energies but also in comparison to fossil-fuel-based and nuclear electricity generation (Welsch and Biermann 2014).

Dynamics of Renewable Energy Externalities

Table 2 reports the estimation results for the dynamics of RE externalities. This analysis focuses on wind power and biomass plants because the presence of solar power in one’s district was found to have no significant well-being effect. The coefficients of the dummy variables D(T + i) indicate whether and how life satisfaction in the respective year differs from life satisfaction before T − 2, where T is the year of the first installation. The lead coefficients indicate well-being effects from people’s anticipation of the installation of RE facilities. Such effects may arise when the installation is publicly debated in advance. The lag coefficients represent a blend of two effects. One is hedonic adaptation, which implies that coefficients become smaller in magnitude and less significant over time. The other is the effect of expansions after the first installation, which implies that coefficients become bigger.

For wind power, the coefficients on D(T +i) before and after T + 1 are nonsignificant. The coefficient on D(T + 1) is weakly significantly negative. We thus find an absence of any lead effects and a dominance of hedonic adaptation over any effect from expansion after the first installation. In the case of biomass, there is a weakly significant lead effect in T − 1, followed by significant effects thereafter, which persist even after three years. An F-test of whether the six time dummies are equal to each other and to 0 can be rejected for biomass, but not for wind. Considering that the expansion rate of power from biomass is the same as that of wind power, this suggests that hedonic adaptation to the nuisance from biomass plants is weaker than adaptation to the nuisance from wind turbines. A possible explanation for this finding may be that the nuisance from biomass plants (odor) is more difficult to avoid by averting behavior than is the nuisance from wind turbines (mainly visual impairment).

Quantifying Renewable Energy Externalities

Table 3 presents an overview of the estimated well-being effects of RE both in units of life satisfaction and in monetary terms. The monetary values are based on the results presented in Table 1 and obtained by dividing the coefficients on the respective RE variables by the coefficient on log income, which yields the marginal rate of substitution between RE and log income or the percentage change of monthly equivalized income required to compensate for a one-unit increase in the RE variable.

As reported in Table 3, the presence of wind energy plants in one’s own postcode district is associated with a reduction in the 11-point life satisfaction score of about 0.0346. This corresponds to about 7% of the effect of being unemployed, which is one of the most important adverse factors for life satisfaction (the coefficient being –0.52, see Appendix). In monetary terms it corresponds to 11% of equivalized income. Evaluated at the mean of equivalized monthly income (€1,648.266, see Table A2), the presence of at least one wind turbine in one’s postcode district is equivalent to a decrease in equivalized monthly income of about €181.9.

The presence of biomass plants in one’s own postcode district is associated with a reduction in 11-point life satisfaction score of 0.0347, corresponding to €182.4 of equivalized monthly income. The presence of biomass plants in adjacent districts is associated with a reduction in life satisfaction score of about 0.029, corresponding to €150.4.

Turning to RE capacities, an increase in installed wind power capacity of 1 MW is equivalent to 0.35% in equivalized monthly income, which is about €5.8. With respect to biomass, an increase in installed capacity of 1 MW is associated with a drop in life satisfaction of 0.039, which relates to 1.25% of income, or €20.5.

Since a major motivation of expanding renewable energies in Germany is the abatement of greenhouse gases, it is useful to compare the local externalities from RE to the benefit of CO₂ reduction. Such a comparison depends on several parameters such as the number of full-load hours of the renewable power plants on the one hand, and the type of fossil-fuel technology being replaced on the other hand. Focusing on wind power, we note that the average installed capacity in a district-year in which at least one wind turbine exists is 10.9 MW. Assuming 1,600 full-load hours per year (BMWi 2014), we obtain an average yearly power production per district of 17,440 MWh. If the same amount of electricity were to be produced from lignite-fired plants (at a heat rate of 34.7%, implying 1.190 tons of CO2 per MWh, see Wagner et al. 2007), it would imply 20,753.6 tons of CO₂. If it were to be produced from hard coal–fired plants (at a heat rate of 36.9%, implying 0.931 tons of CO2 per MWh, see Wagner et al. 2007) it would imply 16,236.64 tons of CO₂. Applying a medium estimate of worldwide damage from CO₂ of €50 per ton (Foley, Rezai, and Taylor 2013), the climate damage avoided in a district-year with mean wind power capacity comes down to €1,037,680 (lignite) and €811,832 (hard coal), respectively. Applying the same assumptions to the district year with the maximum installed wind capacity (378.671 MW) yields an estimated avoided climate damage worth €36,049,479 (lignite) and €28,200,623 (hard coal).

Comparing these estimates of avoided global climate damage with the externalities from having wind turbines in one’s postcode district (€2,182 of equivalized yearly income per affected person) depends on the population size of the respective postcode district, which is unknown to us. For a postcode district with average installed capacity (10.9 MW), the affected population should not be greater than 476 individuals (if wind power replaces lignite) and 372 individuals (if wind power replaces hard coal) for (local) externalities to be less than (global) avoided damage. For the postcode district with maximum installed wind power capacity (378.671 MW), the critical numbers are 16,521 individuals (lignite) and 12,924 individuals (hard coal). Since the average population in a postcode district is about 10,000 (81 million divided by 8,200 districts) and since wind turbines can be found more often in rural (sparsely populated) districts than in urban (densely populated) districts, it is likely that the local externalities created by wind parks are less than the climate damage that they avoid. It may be added that nonmarket values obtained using subjective well-being data are frequently deemed to be biased upward due to the estimated marginal utility of income being biased downward (see Welsch and Ferreira 2013 for a discussion). The estimated values of climate damage, though typically obtained using other techniques, are of course also subject to considerable caveats.

In addition to climate damage, renewable energies may also avoid traditional forms of pollution. We deem this to be less relevant in the case of Germany since all fossil-fueled power stations in Germany are equipped with scrubbers for particulates, sulfur dioxide, and nitrogen oxides. Using a multicountry dataset, however, Welsch and Biermann (2014) found that a higher share of renewable energies in the national electricity mix is associated with higher well-being. This finding may be the net effect of local externalities created and air pollution avoided by renewable energies. In addition, it may reflect avoided subjective nuclear risk when the share of wind power is greater and the share of nuclear power is smaller (Welsch and Biermann 2014).