Tag Archives: Hoen

Hoen, again

Ben Hoen has just published his latest property value study [backup link].  It was issued under the auspices of Berkeley Labs, sponsored by the Office of Energy Efficiency and Renewable Energy (Wind and Water Power Technologies Office) of the U.S. Department of Energy.  This study contains 51,276 sales, more than any previous study, spread over 9 states, 27 counties and 67 different wind projects.  The study goes for 38 pages, concluding:  “Across all model specifications, we find no statistical evidence that home prices near wind turbines were affected in either the post-construction or post-announcement/preconstruction periods.”

No doubt the industry and its supporters like AWEA will be loudly broadcasting this result.  No doubt sympathetic reporters and politicians will pick up on the buzz and accept its conclusions with no further reflection, not to mention actually taking the time to read the study.  That’s a shame, as this latest Hoen study is just one more example of how statistics can be used to prove just about anything you want proved.  And the DOE really wants to prove that wind turbines don’t affect home prices.

Take the AWEA blog entry linked to above, titled “Wind Power Has No Effect On Property Values”.  Except that isn’t what the study actually said.  What Hoen said was they didn’t find any statistical evidence of any effect.  There is quite a bit of evidence in this study that home prices near turbines take a beating, but when the statisticians finish with it all that somehow disappears.

If we look at the pre-statistical-manipulation numbers we can get a pretty fair idea of the effects of wind turbines on prices.  Below is a snapshot of the first set of figures in Hoen’s Table 7 (page 25).hoen-2013-table-7aThe PA, PAPC and PC columns show the average prices as a nearby project moves from Pre-Announcement to Post-Construction.  As always, I’m interested in comparing the close-in properties (in this case, less than 1 mile) with the far-away properties (in this case, more than 3 miles).  Even before the project is formally announced (and I can tell you first hand that a project is known about well before any formal announcement) nearby home prices are 15% lower than their lucky neighbors.  Even worse, as time marches on and the project goes into operation the price difference widens to 28%.  These numbers just happen to be in the same ballpark as those quoted by McCann, Lansink and Sunak.

Hoen admits these price differences exist at the top of page 24, but goes on to say: “Both conclusions of adverse turbine effects, however, disregard other important differences between the homes, which vary over the periods and distances.”  And what would those important differences be?  His raw data includes just house square feet, acres and age.  The close-in houses have, on average, about 100 more square feet than the far-aways (roughly 1550 vs 1650).  They average 2 acres vs. somewhat less than 1.  They are older, averaging about 60 years vs 50.  Would you be willing to trade off an extra 10 years of age for a much larger lot and some extra space?  I think most people would, so there’s nothing in the raw data to explain the difference – except, of course, the elephant in the room, the turbines.  You could try to argue that maybe there’s some other problem with the close-ins.  Did I mention these data came from 9 states, 27 counties and 67 projects?  It is quite a stretch to propose something else, as even a landfill at every one of the close-in neighborhoods wouldn’t be enough (Hoen, page 4).  It seems that wind turbines are in a class of their own when it comes to decreasing house prices.

So how does Hoen ever get to the conclusion that he does?  I have to confess, the study is written in such a way that I can’t figure it out.  I’m reasonably literate and numerate for a layperson, and if I can’t figure it out it means one of two things.  Either it was written for a specific audience (in this case, professional statisticians) or it was meant to obfuscate.

Hoen tries to make it look like he tried to do an unbiased study, and maybe he did.  But to get from these raw data to his conclusions requires a lot of manipulations and there’s no way we can ferret out all of them.  As just one example, he pools the sales data from all 9 states, as disparate as Oklahoma and New Jersey.  Are properties in Oklahoma and New Jersey from the same population?  Of course not, but pooling them increases the Standard Deviation, making statistical significance more elusive – and NOT finding statistical significance is what his sponsor really wants.  When the raw data shows one thing and the conclusions show another I get suspicious, especially when the flip is in the sponsor’s interests.

Partly as a result of the pooling, Hoen’s average house price was (from Table 3 on page 21) $122,475 with a standard deviation of $80,367.  To get to “statistical significance” you generally have to get two populations (in this case, the close-ins and the far-aways) about 2 SD’s apart.  Clearly that won’t happen here, no matter how much the wind turbines affect the prices.  Finding a group of smaller apples on size alone is pretty hard work when you’ve pooled apples and peaches together.

I have no doubt that someone could take this same data and more convincingly make the opposite case, that wind turbines do affect house prices.  Unfortunately, we all know the payer calls the tune, and in this case Mr. Hoen appears to be making a living by faithfully playing the right tune.

A Nail for Hoen’s Coffin

Ben Hoen was the lead author of the LBNL studyThe Impact of Wind Power Projects on Residential Property Values in the United States: A Multi-Site Hedonic Analysis“.  This Hoen 2009 study remains the largest ever undertaken, with nearly 7500 properties.  It found:  “Specifically, neither the view of the wind facilities nor the distance of the home to those facilities is found to have any consistent, measurable, and statistically significant effect on home sales prices“.  Hoen is the primary study quoted by wind energy proponents to rebut the common-sense notion that the nuisance created by nearby wind turbines will lower home prices.  This study is not without its problems;  I’ve previously posted on some of them.

Recently I read over another property value study, this one from Germany, “The Impact of Wind Farms on Property Values: A Geographically Weighted Hedonic Pricing Model“, by Yasin Sunak, 2012.  The main purpose behind it was to study the effect of pooling together properties from different locations and running regressions against the entire pool, compared with analyzing different neighborhoods separately.  Many of the studies routinely quoted by proponents pooled their data.  As an example, Hoen’s 7500 properties came from within 10 miles of 24 projects in 9 states and were all pooled together before he did any analyses.  I’ll cover the pooling issues first, but this study contained additional results that are equally important.

Pooling Example

To give you an example of how this might be a problem I created the following simplified example.  Below are shown close-in (<2km) and more distant prices in two separate neighborhoods.  Neighborhood #1’s price averaged $100,000 and #2 averaged $300,000.  After the project goes in the close-in neighborhoods suffer a 25% decline in prices for homes within 2km of the project.

Given that the means in both neighborhoods are over 2 standard deviations apart,  no statistician would suggest that the turbines had no effect.  A t-test shows that the chance of either of these distributions occurring by chance is 5 in a million.  But if we simply combine (“pool”) the neighborhoods and then run the analysis we get quite a different answer:

This is the same data as before, with the same declines as before, but now the standard deviation is large enough that most statisticians (especially those who are serving the interests of their sponsors) would declare there is no statistically significant effect of wind turbines on home prices.

Although my sample just considered the effect of pooling on standard deviations, the same principle applies to the effects on regression analysis: similar computational techniques are used to calculate both.  Sunak provides an analogous regression example at the bottom of page 7, figure 2.

Sunak

Sunak’s study involved several neighborhoods around just one project in Germany.  Although it was just one project he was still working with a surprisingly large number of sales, slightly over 1400.  Even more surprising, he used just vacant residential lots, so any complications from the number of bathrooms etc are avoided.  This large a sample in this small an area is unprecedented, and provides what is arguably the most accurate assessment of the effects of wind turbines on property values to date.

Sunak first pooled all the sales and ran traditional regressions against them.  For Sunak’s purposes this was just a prelude, but his results are extraordinary.  Lot prices decreased 0.209% for every 1% closer a property was to the project, a result that was statistically significant.  Put another way, prices of lots within 2km of the project decreased an average of 25.2%, again statistically significant.  Another surprising result was that being subject to shadow flicker also reduced the price enough that it was almost statistically significant as well.

Sunak then tested these results for something called spatial autocorrelation and finds some amount of it, which indicates that prices are influenced not just by the distance to the project, but also by where they are.  He then runs the regression using Geographically Weighted Regression and finds that the spatial autocorrelation has decreased markedly.  The single result that was previously applied across the entire area was not indicative of what happened within each neighborhood.  Sunak writes: “The application of the GWR revealed a more complex picture of the influencing effects through the weighting of spatial relationships and local variations in the data. Based on local estimates, the GWR revealed negative, wind-farm-related effects that are attributable to strong local influences of the wind farm site. Therefore, predominantly biased by local clustering, global estimations are not appropriate in capturing the impacts of wind farm proximity on property prices.

 Conclusion

There were two major results of this study.  First, the practice of pooling sales data from disparate areas is inappropriate and would tend to downplay the effect of the wind turbines on prices.  In the case of Hoen, one has to wonder if he was aware of this effect and went ahead with the pooling anyway, knowing that doing so served the interests of his sponsor.  Second, when enough data is collected from one area the effect of wind turbines is unmistakeable and irrefutable.

My thanks to Michael McCann, who alerted me to this study, and whose own studies indicate a loss of, guess what, 25% (or more) for properties within 2 miles of a project.