case study

← Back

Analysis of Wind Farms Impact on Property Prices Using Difference-in-Differences

A Reykjavík University × ETH Zürich research thesis measuring how wind farms move nearby house prices. Homes within 1 km sell for about 4 percent less, and actually seeing the turbines carries a penalty of its own, reaching 6 percent when many fill the view.

  • Python
  • Causal Inference
  • Geospatial
  • Econometrics
  • Research

My final BSc thesis, completed in collaboration with ETH Zürich as part of the EU-funded WIMBY project (Wind In My Backyard). The brief was simple to state and hard to answer: put a credible number on how onshore wind farms affect the price of the homes around them.

In collaboration with

The question

Wind power is central to the energy transition, yet new turbines are often fought at the local level. People support renewables in principle and still object when a wind farm is proposed next to them, and one of the loudest concerns is what it does to the value of their home. That concern shapes planning decisions, so it is worth measuring properly rather than guessing.

The catch is that homes near turbines already differ from homes far away. They tend to be more rural, more open, sometimes more scenic. A raw price comparison would blame the turbine for differences that were there all along. The real task is separating the effect of the wind farm from everything else that makes a location what it is.

The data

I brought together four spatial datasets covering southern Great Britain, all aligned to a common coordinate system so distances and areas stay consistent:

  • 5.7M residential sales from 2011 to 2019, roughly 2.9M of them clean enough to model.
  • 5,441 onshore turbines, grouped into 835 wind farms by their commissioning dates, which set the before-and-after clock for every nearby sale.
  • 28 terrain-aware viewshed rasters at 100 m resolution, giving each property a real count of how many turbines are actually visible from it, not just how close they are.
  • A digital surface model used to simulate shadow flicker, the moving shadow a rotating blade throws across the ground.

That last pair matters. Distance is a crude proxy for exposure, because two homes the same distance from a turbine can have completely different views depending on the terrain between them.

How I measured it

I treated a new wind farm as a natural experiment and used a difference-in-differences design with two-way fixed effects. In plain terms, I compared how prices moved near a new turbine against how prices moved in similar areas over the same period, so shared market swings cancel out and what remains is the turbine’s own footprint. I studied it through three separate channels: raw proximity, actual visibility, and shadow flicker. Event-study versions traced how the effect evolves year by year around construction.

The interactive below is the idea in miniature. Drag a home closer to a wind farm and watch the naive correlation pull away from the honest causal estimate. The numbers here are illustrative, but the mechanism is exactly the one behind the results further down.

The Estimatorτ: a wind farm’s effect on nearby home prices

Drag the house nearer to or further from the turbine.

Raw correlation−8.6%95% CI ±1.3%what a naive comparison suggests (misleading)
Causal estimate (DiD)−4.3%95% CI ±1.3%after differencing out area & time trends (the honest effect)

How the price effect changes with distance

Estimated price effect by distance from the wind farmTwo curves plotted against distance. The raw correlation starts far below zero close to the turbines and the causal difference-in-differences estimate sits above it inside its 95 percent confidence band. Both effects shrink toward zero as distance grows. A marker tracks the chosen distance.0−5%−10%1.0 km0246
Causal estimate + 95% band Raw correlation

Why the causal estimate is credible

Parallel-trends chart of a Difference-in-Differences designThree home-price index lines. Control and Treatment move in parallel before the turbine is built. After construction the Treatment line drops below its Counterfactual, the dashed line showing where it would have gone parallel to Control. The shaded gap between Treatment and Counterfactual, bracketed and labelled at the end, is the difference-in-differences estimate.turbine built−4y+3y−4.3%
Control Treatment Counterfactual (no turbine)The shaded gap between Treatment and its Counterfactual is the DiD estimate.
Interactive reconstruction. The real difference-in-differences analysis ran offline on 5.7M property transactions across 5,441 onshore turbines. The curve here is tuned to resemble the shape of the observed gradient, but the exact figures are illustrative teaching values, not the reported results.

What I found

The closer the home, the bigger the discount

Splitting the area around each turbine into one-kilometre rings shows a clear gradient. Homes within 1 km sell for about −4.29% relative to comparable homes 10 to 11 km away. The discount stays negative and significant out to 5 to 6 km, then fades into statistical noise by 6 to 7 km. Living near a turbine costs something, even before asking whether you can see it.

Price effect by distance from the nearest wind turbineBar chart of the estimated price effect in each one-kilometre distance band, relative to homes 10 to 11 km away. The discount is largest within 1 km at about 4.3 percent, stays negative through 5 to 6 km, and is statistically indistinguishable from zero by 6 to 7 km.−2%−4%0−4.3%0–1−2.6%1–2−3.3%2–3−1.5%3–4−2.2%4–5−1.7%5–6n.s.6–7n.s.7–8km from nearest turbine
Distance-band effect on sale price, relative to homes 10 to 11 km away. Whiskers show 95% confidence intervals. Bands marked n.s. are not statistically significant.

Seeing turbines is its own penalty

Swapping distance for a real visibility count tells a sharper story. Seeing a handful of turbines barely registers, but the effect is strongly non-linear and escalates fast. Homes that see 16 or more turbines carry a discount of about −6.08%. The jump between the 6 to 10 and 11 to 15 bands is the interesting part: a horizon dotted with a few turbines is tolerated, a horizon dominated by them is not.

Price effect by the number of visible turbinesBar chart of the estimated price effect grouped by how many turbines are visible from a property, relative to seeing none. Seeing one to five turbines lowers prices by about half a percent, while seeing sixteen or more lowers them by about six percent. The relationship is strongly non-linear, jumping sharply between the six-to-ten and eleven-to-fifteen bands.−2%−4%−6%0−0.5%1–5−1.2%6–10−3.8%11–15−6.1%16+turbines visible from the property
Effect on sale price by number of visible turbines, relative to seeing none. Whiskers show 95% confidence intervals.

Visibility is separate from proximity, and that is the point

The obvious worry is that visibility is just distance wearing a disguise, since closer homes tend to have clearer views. So I put both channels in the same model. If visibility were only a proxy for proximity, its effect should collapse once distance is controlled for.

It barely moves. The visibility penalty goes from −0.83% to −0.78% after adding distance controls, and a distance gradient survives after controlling for visibility. Proximity and visibility are related but genuinely separate forces on price. That separation is the thesis’s main contribution.

Shadow flicker: a hint, not a verdict

Within 3 km, homes exposed to shadow flicker after construction show a discount of around −1.97% under one geographic specification. But it thins to −0.23% and loses significance under a finer one, so I report it as suggestive evidence of a physical nuisance channel rather than a firm result.

Distance is not the whole story

The interesting result is not that wind farms lower nearby prices. It is that distance alone cannot tell you by how much. Two homes an equal distance from a turbine can face very different discounts depending on what they can actually see, and the penalty climbs sharply once turbines fill the view rather than dot the horizon. Visibility also holds up once distance is controlled for, so it acts as a channel of its own rather than a stand-in for being close. Most earlier work leaned on distance rings alone, and bringing in terrain-aware visibility is where this thesis earns its keep: proximity and visibility are related but genuinely distinct ways a turbine reaches a home’s price.

The estimates I trust most are the static distance and visibility ones. The year-by-year event studies are the honest soft spot: treated and control homes were already drifting apart before any turbine went up, so those paths describe the timing rather than prove it, and I would not claim to say whether prices recover in the long run.

Where it stands

A complete national-scale causal analysis, submitted as my BSc thesis and built end to end in Python (pandas, GeoPandas, Rasterio, Shapely, and Statsmodels). A manuscript extending it is in preparation with the ETH Zürich team, and the full write-up is below.

Read the full reportPDF · 57 pages