Clean Electricity, Dirty Electricity: The Effect on Local House Prices

Defining and pricing externalities of electricity generation facilities is a challenging task. Roth and Ambs (2004) provide a meta-study to quantify the externality costs of 14 different electricity generation types, focusing on air pollutants. They find a wide range of damage cost estimations of individual air pollutants, such as for carbon dioxide (CO₂), ranging from $9.90 to $41.60 per ton, with coal power plants having the highest external costs, followed by gas and combined cycle power plants. In contrast, biomass and wind appear to have only limited external costs. These results are in line with a study conducted by the European Commission (2003). However, despite a common agreement over the rank of air pollution emission among electricity generation types, cost estimates of different emissions vary widely, showing the difficulty of pricing the externalities of electricity generation.

Despite the lack of air pollution stemming from wind turbines, there are significant noise and visual effects. Reported health effects such as sleep disturbance, headache, anger fatigue and loss of motivation are acknowledged as factors that can be caused by the noise from wind turbines (Farboud et al., 2013).³ Visual effects seem to have an even stronger impact than noise (Bakker et al., 2012). Even though there is no scientific evidence about causal effects, people located near wind turbines have reported health effects, claiming those effects were due to photo-induced seizures (photosensitive epilepsy) and wind turbine blade flicker (Harding et al., 2008). In addition, households located nearby report a decrease in life-satisfaction after the installation of wind turbines (Krekel and Zerrahn, 2017).

The increased number of self-reported health effects by people living near wind turbines merely seems to be caused by the annoyance over the presence of wind turbines itself, rather than originating from one aspect in particular, highlighting the difficulty to identify individual externalities (Pedersen & Waye, 2007).⁴ People generally support wind energy, but oppose it if facilities are installed close to their homes (Breukers & Wolsink 2007; Wolsink 2000, 2007; Wüstenhagen et al.2007). Wolsink (2007) states that local residents are willing to accept wind turbines in their vicinity as long as they perceive the general distribution of wind turbines as “fair”. However, in the Netherlands local residents do not have the perception that they can influence the distribution of new wind turbine sites (Wüstenhagen et al., 2007).⁵ Similar results are found in the United Kingdom (Bell et al., 2005), the United States (Pasqualetti, 2011b), and Mexico (Pasqualetti, 2011a).

Since residential real estate is fixed in location, prices are highly sensitive to factors disrupting location quality (Hilber, 2005), making real estate a good identifier of local utility or disutility from externalities. As people choose locations according to their preferences and aversions (Tiebout, 1956a), external factors, that arise from power plants and wind turbines, are incorporated in house prices (Rosen, 1974), allowing to assign monetary values to the external effects.⁶ However, since it is practically impossible determine and measure all external effects of different electricity generation facilities, we use distance (proximity) to properties as a measure in the Rosen (1974) framework. Using distance should theoretically reflect the net-external effects (Nelson, 2008).

Research focusing on local housing market effects of power plants dates back more than 40 years. However, most studies focus on the effects of individual power plants in small regional markets (Blomquist, 1974; Clark et al., 1997; Gamble and Downing, 1982; Sunak & Madlener, 2016). Large-scale studies tend to investigate just one type of electricity production, using different measures of property prices, such as transaction-based (Dröes & Koster, 2016) or survey-based (Davis, 2011). Additionally, studies use different control variables, leading to varying empirical models. Due to the heterogeneous characteristics of housing markets, changes in electricity generation technology over time, and the focus on a single type of electricity generation, it is hard to draw a coherent conclusion about the relative effects of different electricity generation types on local house prices.

In most studies,the observed external effects for conventional power plants and wind turbines are either negative or insignificant. Blomquist (1974) finds a price decrease of 0.9 percent per 500 feet, within a 2-mile distance of a coal power plant. Davis (2011) finds a discount of 3 to 7 percent within 2 miles of plants, increasing with proximity and capacity. For wind turbines, negative external effects range between 5 percent within 0.5 miles (Lang and Opaluch, 2013), 2 to 16 percent within 3 miles (Heintzelman & Tuttle, 2012), 1.2 to 2.6 percent within 2 kilometers (Dröes & Koster, 2016) and 5 to 6 percent within 2 kilometers (Gibbons, 2015). Other studies find no evidence of significant effects (Carter, 2011; Hoen, 2014; Sims et al., 2008). Since the methodology, number of observations, research area and control variables differ widely between studies, it is not possible to directly compare these findings and to draw firm conclusions regarding the relative externality costs of different forms of electricity generation.

Overall, there is no study that simultaneously includes different types of electricity generation technology, uses a large number of observations, measures transaction-based house prices, and accounts for sufficient control variables to quantify external effects in a comparable manner. Furthermore, in any study addressing the effect of locally desirable or undesirable externalities on house prices, the main challenge is identification. In the case of power plants, both renewable and conventional, the locational choice is often driven by factors such as land values and (local) politics, rather than being fully random or based on exogenous factors such as proximity to waterways or exposure to a stiff breeze. Only a few, recent studies address this issue (Davis, 2011; Dröes & Koster, 2016).

We study the effects of different electricity generation methods on house prices within one market: The Netherlands. Our focus is on coal, gas, biomass, and wind energy, due to their significance for the local electricity market. Since the findings of previous studies differ widely, we are interested in the variation of external effects between different electricity generation facility types and the variation of findings due to different model specifications. We therefore test two specifications: a difference-in-difference (DID) approach using average area price changes (e.g. see Muehlenbachs et al., 2015; Pope & Pope 2015), and a DID repeated sales model.

For both approaches, we use a similar measure of externality exposure. Since we cannot determine and measure all potential externalities of the different electricity generation facilities, we assume that externalities spread over distance (Nelson, 2008). As we focus on local external effects of electricity generation facilities and neglect global effects, such as CO₂ emissions, we focus on areas directly surrounding electricity generation facilities. However, it might be that different externalities spread differently over distance. In contrast to physical externalities such as noise and sight disturbance, economic externalities, such as employment effects, could reach further, leading to potentially unbalanced external effects over distance.

Assuming external effects spread over distance, we determine exposure to externalities by the linear distance to the closest electricity generation facility of every energy type, using longitude and latitude information.⁷ Based on a cut-off distance, we consider observations as either affected by externalities (d = 1) or not affected (d = 0). Based on empirical findings on conventional plants (Davis, 2011) and wind turbines (Dröes & Koster, 2016), we use a cut-off distance of 2.5 km for all energy types. However, we further examine different cut-off distances and interval measures.

To avoid interference among affected and control observations, we omit observations in a ring-shaped area beyond the cut-off distance as illustrated in Fig. 1. Using the externality cut-off distance of 2.5 km, we consider observations within 2.5 km distance as affected (d = 1), comparing them with control observations beyond 2.5+z km, where z is the width of the omitted area. We begin with z = 1 km, but test different lengths for robustness. Due to the heterogeneity of local residential markets, observations at externality distances do not necessarily share the same locational characteristics with observations far away. We therefore control for location fixed-effects and limit our control group to a maximum distance of 10 km (y = 6.5 km), omitting observations beyond 10 km distance from the analysis.⁸

We account for confounding factors from other electricity generation facilities. Observations within a 2.5 km distance of a nuclear power plant are omitted, since there is only one active plant in the Netherlands and we argue that the external effects from nuclear power plants differ from those of conventional plants (Gawande & Jenkins-Smith, 2001). Furthermore, we exclude observations within a 2.5 km distance from the German and Belgium border, since we cannot fully account for power plants across the border.⁹

Since the placement of power plants and wind turbines is not random, a static model might be biased by an economic endogeneity problem of the price-effect relationship. Besides infrastructure factors, such as grid infrastructure or the closeness to gas pipelines and harbors, and political factors, such as local voter opposition, land value may also determine placement decisions. Low land values make it cheaper to build a power plant or to erect a wind turbine. Since land values strongly correlate with house prices, too, observed house price discounts nearby power plants or wind turbines might be the result of low ex-ante land prices instead (Kok et al., 2014).

Beside limiting the reach of our control group, we address this potential problem by using a difference-in-difference (DID) model, investigating the effects of facility openings and closings. Although our dataset is large, it contains just a limited number of repeated sales pairs. Due to more placements of power plants and wind turbines in remote locations and imbalances in the number of facility openings and closings (e.g. few coal and biomass plant openings and wind turbine closings), we are not able to perform a DID analysis based on repeated sales for all facility types. Instead, we use a difference-in-difference model for geographic areas similar to Muehlenbachs et al. (2015), comparing the change in house prices of areas that experience facility openings / closings, with areas nearby (control areas), as illustrated in Fig. 2.

Equation 1 shows the employed model, testing the effect of facility openings on average property prices, where FAC_ki measures if observation i is in proximity to an electricity generating facility of type k, either before or after opening, post_kit (d = 1) measures transactions after opening of the closest facility of type k, and Treat_kit = FAC_kit ∗ post_kit equals 1 for transactions in close proximity to a facility, after opening. Therefore, the coefficient of interest is γ_3k, measuring the effect of facility type k openings. We control for unobserved differences in price-determining factors between areas over time and general time-trends, using X_it, consisting of structural, environmental and neighborhood characteristics, as well as time fixed-effects (Rosen, 1974). Control variables are listed in A Table 7. We further test for the assumption of common pre-treatment trends (Kuminoff and Pope, 2014). We account for potential spatial dependence and omitted variables by including neighborhood-fixed effects in X_it, hereafter referred to as location FE (Anselin & Bera, 1998; Kuminoff et al., 2010).¹⁰ As time-fixed effects we use year dummies, measuring general house price dynamics over time (see e.g. Hoen 2010; Hoen2014).¹¹ Standard errors are clustered by municipality and year.

The analysis is limited to facilities that open during the sample period. We exclude observations around plants that opened before. We only consider observations up to 10 years before and after facility openings, as it can be assumed that long-term price effects settle over time.¹² To eliminate anticipation effects and effects from construction work, we omit transactions one to two years before openings, as well as the opening year, depending on facility types.¹³ We do not consider observations that experience a replacement, being within externality distance of another facility prior to opening of a new plant. Since the number of affected houses decreases through this filtering process, we also test a 3-km cut-off distance throughout the analysis.

To examine the spatial heterogeneity of external effects, we test the external effects of different electricity generation types at different distance bands. As shown in Eq. 2, we form J distance intervals of 1 km length up to a distance of 4 km for every energy type K. We use 1 km length to have sufficient observations in every interval. To reduce complexity, we do not use a holdout distance, meaning we distinguish between treated observations up to 4 km and control observations between 4 - 10 km distance. As property prices differ between urban and rural areas (DiPasquale & Wheaton, 1996), the perception of external effects might differ due to the presence of confounding negative or positive externalities, such as from road noise in cities. We therefore investigate the urban-rural heterogeneity effect, by explicitly controlling for facility openings in urban areas. As shown in Eq. 3, we add an urban location control dummy in the control matrix X_it and add an urban-treatment interaction term represented by (Treat_kit ∗ urban_it), where β_1k represents the additional effect of facility type k opening in urban areas.¹⁴

Due to technical improvements and government regulation, conventional electricity generation facilities typically get cleaner, less noisy, and more efficient over time.¹⁵ We therefore examine the heterogeneity of opening effects over time, testing subsets of 10-year periods, moving in 5-year steps, as shown in Eq. 1. We use 10-year periods instead of individual years to have sufficient observations for every period, leading to the following, overlapping periods: 1985 - 1995, 1990 - 2000, 1995 - 2005, 2000 - 2010, and 2005 - 2015. Another potential source of heterogeneity is power plant size (Davis, 2011) or the number of wind turbines in a wind park (Dröes & Koster, 2016). Davis (2011) argues to focus on power plants above 100 MW, since disamenities are likely to be stronger for these plants. We therefore differentiate for size, using capacity for conventional plants and the number of contiguous wind turbines within a wind park, as capacity differences for individual wind turbines are relatively small compared to power plants. Due to constraints in size variation, we are only able to examine size effects for gas plants and wind turbines. Based on the available variation, we distinguish three size categories for gas plants (< 100MW, 100 - 400 MW, > 400MW), and four categories for wind turbines (single turbine, 2 to 9 turbines, 10 to 29 turbines, 30+ turbines).¹⁶

Similar to openings, we investigate the effect of facility closings, examining the effect of externality removals. We focus on areas initially nearby electricity generation facilities, experiencing a closing and being outside externality distance thereafter. We compare the average price change with areas remaining in externality distance of facilities. Equation 4 shows the model, where Close_kit (= 1) indicates observation i being close to a closing facility of type k, post_kit (= 1) indicates observation i is transacted after the closing of facility type k, and Treat_kit = Close_kit ∗ post_kit, identifies observations nearby closing facilities after closing. Similar as in Eq. 1, we control for structural and environmental characteristics by X_it. We further control for remaining facility types.¹⁷ Finally, we investigate heterogeneity in the effects, testing for facility size in a similar manner as before.

One problem of the regional difference-in-difference approach presented Section 3 is that it relies on the assumption of similar housing characteristics over time. However, even though we control for housing characteristics, we cannot rule out that transactions in affected or control areas systematically differ over time. To overcome this issue, we use repeated sales of the same property. Within our sample, there are 457,547 observations with at least one repeated sale throughout the sample period, of which 109,692 observations are sold three times, 23,959 sold four times, and 4,749 sold five times. Using the same setup, cut-off distances and time restrictions as before, we use repeated sales to measure the change in price on the same house after facility openings and closings.

We follow the approach of Aydin et al. (2016) as shown in Eq. 5. As a dependent variable, we use the percentage change in price Δp_i(t+n) of property i between period t and n. Based on the previously defined cut-off distances, we measure whether a facility of type k opened (closed) between the two sales within externality distance, indicated by ΔFAC_ki(t+n) = 1. We control for changes in house quality and amenities around, using vectors of control variables. We distinguish for positive changes \({\Delta } Q_{i(t+n)}^{\prime +}\), such as added amenities or improvements in quality, and negative changes \({\Delta } Q_{i(t+n)}^{\prime -}\). We control for sales year \(Y_{it}^{\prime }\) and time between two sales (n-t) interacted with sales year, indicated by control vector \({\Theta }_{it}^{\prime }\). It can be assumed that the housing market adjusts to openings / closings over time. In order to investigate the adjustment of the market to the opening / closing of facilities nearby, we control for the time difference between property sales and facility opening / closing years. We measure the year difference of observation i at t+n and the opening (closing) year of the nearest facility of all types K, indicated by z(k). Since we are interested in time difference effects of treated observations (ΔFAC_ki(t+n) = 1), we interact the time difference with ΔFAC_ki(t+n), indicated by Λ_(t+n)−z(k). The final model is shown in Eq. 6.

From various sources we collect detailed information on all major Dutch power plants, present between 1985 and 2015 and using coal, gas, biomass or a combination thereof.¹⁸ We subsequently verified and completed our data with the help of all major electricity suppliers in the Netherlands. Our data contains the number of electricity generation units, fuel types per unit, capacity per unit, year of operational start, year of closing (if applicable), and geocoded location.¹⁹ We exclude all cogeneration plants on industrial sites, cogeneration plants focusing primarily on heat generation and plants that do not produce electricity for the public grid (e.g. industrial plants).²⁰

The final sample includes 119 power generation units located on 45 different plant sites. Figure 3 shows the geographical distribution of the power plants in our sample. Power plants are not systematically located in low population density regions (so as to reduce possible negative externalities), but are placed rather close to urban areas, to keep supply distances in the grid short and to ensure supply stability to urban centers. Another important factor for coal and gas plants is the closeness to fuel transportation infrastructure, such as harbors or pipelines. It is therefore not surprising that a large number of plants are located in the Rotterdam harbor area, assuring direct access to overseas supply of gas and coal.

Panel A of Table 1 shows an overview of the number of power plants sorted by primary fuel type. Gas and coal plants are the most prevalent by number of plants and capacity, with coal providing more than ten times as much power as biomass, and gas almost three times as much as coal. However, the range of minimum and maximum capacity shows that capacity differences across power plants, for given fuel types, are quite large. For example, the smallest gas plant in our sample has a capacity of 13 MW, while the largest is almost 100 times larger. However, capacity distributions do not differ very much between fuel types.²¹

Panel A of Fig. 4 shows the installed capacity and the respective number of electricity generation units per fuel type per year. The majority of installed units during the sample period are gas units. The first plant primarily running on biomass was installed in 2000. The popularity of coal energy decreased over the same period, with no new facilities added for nearly 20 years between 1994 and 2014. Only in 2015 did the utility company Essent start operating the newly built Eemshavencentrale, the Netherlands’ biggest and most modern coal plant.

Information on wind turbines in the Netherlands is well-documented, including location, capacity, height, rotor diameter, setup year, dismantling year, and park affiliation. Using data from Warren and McFadyen (2010), we consider all wind turbines that were operational at some point between 1985 and 2015. Our sample consist of 2,117 individual wind turbines, clustered in 217 wind parks. Figure 3 shows the geographical distribution of all wind turbines in the sample, compared to the distribution of conventional power plants and related to population density. Most wind turbines are located in the north-western coastal area, where wind speed is highest. Furthermore, wind turbines tend to be placed in relatively low-density areas, sometimes located close to densely populated areas.

As shown in Panel B of Table 1, the average wind turbine in the Netherlands has a capacity of 1.4 MW and the largest Dutch wind turbine has a capacity of 7.5 MW, which is less than the smallest power plant in our sample. The capacity distribution of wind turbines is widely spread compared to power plants, with the largest turbine having 500 times more capacity than the smallest. Panel B of Fig. 4 shows the development of wind turbines installed in the Netherlands and the technological development over time. The average capacity per wind turbine increases over time, from 15 kW in 1982 to an average capacity of 3 MW. Modern wind turbines have not only more efficient generators but are also higher with larger rotor diameters. The average sample height is 61.5 meters and the average rotor diameter is 59 meter. During the sample period, 61 wind turbines were dismantled after an average life span of 13 years.

We employ a detailed dataset of housing transactions, consisting initially of nearly 3 million observations between 1985 and the first quarter of 2015. The dataset is provided by the Dutch realtors’ association (NVM), which covers around 70 percent of Dutch housing transactions. The dataset contains address, transaction price, structural and environmental information, as well as sales information, such as initial asking price and time on the market. We use Bing Maps through an Application Programming Interface (API) to determine longitude and latitude information per address. After excluding double entries, outliers, and observations with incomplete information, we end up with approximately 2.3 million transactions.

We match data on power plants, wind turbines and housing transactions based on longitude and latitude, using GIS. To control for systematic differences between locations, we add additional information on the municipal level to the dataset. The Dutch Statistics Office (CBS) provides information about population density and land use per municipality, which we use to identify urban centers and rural areas.²²

At a cut-off distance of 2.5 km, there are 339,931 houses within externality distance of electricity generation facilities, distributed by facility type as follows: 1,772 biomass, 10,779 coal, 152,093 gas, and 185,598 wind.²³ We notice that the number of observations within “external effect” distance is limited for biomass and coal plants. A Fig. 5 shows the percentage of affected observations over time. The number of observations close to coal, gas and biomass plants does not change significantly over time, whereas the number of observations close to wind turbines changes markedly. Since the spatial distribution of houses is relatively stable over time, this implies that more wind turbines get positioned close to housing over time, confirming the convergence of renewable electricity generation and urban space.

Table 2 shows the average characteristics of houses and apartments in close proximity to power plants, as well as the characteristics of homes in the control sample (4 km <d < 10 km). There are relatively more apartments in the affected group. While apartments in both groups have roughly the same size, houses in the affected group are smaller than houses in the control group, on average. Apartments and houses tend to be lower priced in the affected group, both in absolute terms and on a per square meter basis. However, this is not necessarily due to quality characteristics or urban location, as both characteristics show that properties in affected areas are on average of better quality and in higher urbanized areas.²⁴

A Table 8 shows the groupings of observations based on the described method in Section 3. We note a small number of observations for some facility types and times, forcing us to exclude coal plants from the opening analysis.²⁵ Due to long lifespans of coal plants and the novelty of biomass and wind, we document less observations for facility closings than for openings. A Fig. 6 plots the average price per square meter per year over time, examining the simultaneous trend assumption, and we notice that pre-opening and pre-closing trends follow a quite similar trend.²⁶

As shown in Table 3, we document a significant positive price effect for homes around gas plants, ranging from 4.0 to 4.6 percent and a negative price effect of 6.0 percent for homes around biomass plants, at a 3 km cut-off distance only. Homes in wind turbine areas show no significant price effect per se, meaning there is no ex-ante price difference. For openings, we document a negative price effect of 4.9 percent to 5.9 percent for gas plants and a negative price effect of 1.8 percent to 2.1 percent for wind turbines. In contrast, biomass plant openings lead to a positive price effect of 4.2 percent to 9.8 percent.

Examining the heterogeneity over distance, Table 4 shows the external effect results for different distance intervals. For gas plant openings, we document a positive opening effect at a distance of up to 1 km distance and significant negative opening effects thereafter. We do not observe enough observations within 1 km distance of biomass plants. Beyond this distance we document a positive opening effect thereafter. Wind turbine openings show a constant negative effect up to a distance of 4 km. Overall, we document constant effects for wind turbines, decreasing effects for biomass plants and mixed effects for gas plants.

Differentiating urban and rural areas, we document significant location discounts for homes in biomass plant areas and wind turbine areas.²⁷ In rural areas, we document a significant positive opening effect between 7.8 to 6.0 percent for gas plants, and 9.3 percent for biomass plant openings, while documenting a negative opening effect between 0.7 to 1.4 percent for wind turbine openings. In contrast, we document significant negative opening effects for all plant types in urban areas, ranging from an additive -14.1 to -10.7 percent for gas plants, -9.7 percent for biomass plants, and -1.2 percent for wind turbines. Our results show that electricity generation facilities are generally perceived negatively in urban areas.

Examining the interaction between plant openings and size, columns five and six differentiating for gas plant size (in capacity) and wind turbine park size (in number of turbines). We document a positive significant opening effect of 4.5 percent for small gas plants (< 100 MW) and an opening discount of 5.1 percent for large gas plants (> 400MW). We do not document a significant opening effect for plans in the range of 100 - 400 MW. For wind turbine areas, we document negative opening effect of 2.4 percent for single turbines, 1.9 percent for parks up to 10 turbines, and 2.8 percent for parks with more than 30 turbines. Our results show that the negative external effects of electricity generation facility openings do not necessarily increase with facility size.

Examining the heterogeneity of openings over time, we estimate our model for different sub-periods of 10 years length as shown in A Table 9. We miss sufficient number of observations for some facility types and time periods. The results are mostly in line with previous findings or insignificant. We document significant negative opening effects for gas plants in the periods 1990 - 2000 (-9.5 percent) and 2005 - 2015 (-5.1 percent). Biomass openings result in significant positive effects between 2.4 to 4.2 percent throughout all available periods. Wind turbine openings result in significant negative effects of -2.4 percent for the period 1995 to 2005, the first boom periods of wind turbines.

Table 5 presents the results for facility closings, comparing to areas with remaining facilities.²⁸ We document no significant closing effect for coal plants. Gas plant closings, show significant negative effects of -3.9 percent at a 3 km cut-off distance. For biomass plants, we document a significant negative closing effects between -7.7 and -6.0 percent. Wind turbine removals (closings) result in positive effect of 6.7 percent at a 2.5 km cut-off distance. Differentiating for gas plant size, we document that negative closing effects of gas plants are higher for plants above 100 MW, ranging from -6.6 to -5.4 percent. For gas plants below 100 MW, we document a negative closing effect at 3 km cut-off distance, significant at 10% level.²⁹ Overall, we document positive closing effects for wind turbines and negative closing effects for biomass and gas plants, compared to areas remaining in proximity to these facilities.

Since the repeated sale model analysis differs slightly from the previous specifications, we examine the fit of our model in A Table 10. We document that the explanatory power is slightly lower than of our previous model specifications. The coefficients make intuitive sense and are in line with former results. In general, adding amenities or increasing quality increases property value, whereas the removal of amenities, such as a terrace, diminishes property value.

Table 6 shows the estimation results for openings and closings. We document significant negative facility opening effects for gas plants and wind turbines. Gas plant openings in the area lead to a 9.5 percentage points lower price between sales, whereas wind turbine openings result in a 2.7 percentage points lower price. The effect is robust at 3 km cut-off distance, as shown in A Table 11. Controlling for the time between facility openings and property sales, the negative opening effect of gas plants and wind turbines remains. However, the effect diminishes over time. For wind turbines, one year difference results in an increase of 0.4 percentage points. We also document a positive effect for biomass plants (1.1 percentage points) even though we do not document a general opening effect. We do not document a significant time difference-effect for gas plants.

Since we do not observe enough transactions to investigate other closings, we focus on gas plants only. In contrast to the DID closing analysis in Section 3, we document positive closing effects of gas plants, ranging from 5.3 to 6 percentage points, not diminishing over time. However, the underlying sample of the closing analysis is small and the effects become nearly insignificant at 3 km cut-off distance (see A Table 11) and should therefore be taken with caution.