Even some of President Donald Trump’s fiercest critics admit that his election win was largely due to his ability to connect on a visceral level with voters who had long felt ignored and disrespected.
Whether it was his unscripted, often rambling speeches or his zest for politically incorrect barbs, Trump made many of his supporters believe that he genuinely understood their priorities and values.
But there is growing evidence that — at least when it comes to clean energy — there is a wide chasm between the policy positions the president supports and what many of his Republican supporters actually believe.
Throughout the campaign, Trump ignored clean energy and focused mostly on promoting coal, oil and gas. But as reputable political surveys have shown, Republican voters actually want more support of renewable energy over fossil fuels. And they are proving that with their wallets, too.
That is one of the most intriguing findings in a new study by PowerScout, an Oakland-based company that uses data to assist consumers who are considering going solar.
PowerScout examined about 1.5 million rooftops in the top 20 solar markets nationwide and found that Democratic and Republican party donors installed residential solar systems at roughly equal rates — 3.06 percent of Democratic donors and 2.24 percent of Republicans, respectively.
In California, by far the nation’s largest residential solar market, Republicans and Democrats install solar at about the same clip, while in Hawaii Republicans install more solar than Democrats.
The timing of the study’s release to coincide with Trump’s inauguration was not an accident. “We want to put a stake in the ground and send a message to the incoming administration, to President Trump and to Secretary Perry, to show in a data-driven way that this is not a partisan issue,” said Attila Toth, the CEO and co-founder of PowerScout, who formerly ran SunEdison’s distributed generation business. “With this household-level data, you can show there is a real economic-driven, non-partisan demand for this type of clean energy asset.”
PowerScout’s study is not the only research to expose the wide clean energy divide between some Republican politicians and the party’s most ardent supporters. Last September, the energy information company SolarPulse released a study that examined data from 25,000 California homes that had installed solar between 1997 and 2015, finding that Republican-leaning communities were five times more likely to purchase rooftop solar than were communities that voted for Democrats. A 2016 poll by the Yale Program on Climate Change Communication also found that 84 percent of all registered voters, including 75 percent of Republicans, supported additional funding for clean energy sources like wind and solar.
PowerScout’s study was different from past research in two significant ways. Rather than studying the voting patterns and partisanship of geographic areas with relatively large amounts of installed solar, the company zoomed in on the household level. In other words, it wanted to find out how willing individual Democrats and Republicans were to shell out money for a solar installation. In addition, the study examined the homes of 1.5 million rooftops — a robust dataset.
To reach its findings, PowerScout pulled the Federal Election Commission’s lists of donors to the Democratic and Republican parties and candidates in the top 20 solar markets. Toth felt that focusing on donors was the best gauge available for an individual household’s party affiliation. “We have seen studies before that lump people together by the voting preference of a county where they reside. To us, that is not granular enough to explain someone’s political tendencies,” he said. “If someone forks out money, that shows their political leaning at the household level.”
From those publicly available lists of donors, PowerScout then filtered out work addresses, post office boxes and apartments so it could focus on single-family homes where a residential solar system could be installed. After winnowing the initial donor lists down, PowerScout was left with 1.5 million homes, 1 million of which were Democratic donors and the rest of which were Republican supporters.
Machine learning to identify solar
In order to get the household-level data it desired, PowerScout next gathered satellite images of all 1.5 million homes of Democratic and Republican donors. Naturally, it would be too time-consuming and cumbersome to have people search each rooftop for solar panels, so the company instead employed a machine-learning model called a convolutional neural network, or CNN.
“We went through and labeled images of houses that had solar and those that didn’t,” said Mike Ulin, head of data science at PowerScout. “You feed those into the model, and the model, based on those examples, can learn to differentiate between homes with solar and without solar.”
The model isn’t actually looking at all the images, at least as we would imagine it. Instead, it converts each image into numerical values.
“So every pixel in the image has three values, a different number for every red, green and blue channel. You end up with a model that has millions of variables as opposed to simple regression models that only have a handful,” said Ulin.
In addition, Ulin said he optimized the model through a technique known as gradient descent. “The model goes through hundreds of iterations to find the optimal settings for millions of variables. Every time it does, we score the output and how well the model identifies the labeled images,” he said, noting that the model was about 90 percent accurate at identifying images.
Toth said the technology can be used to identify the people who are most likely to purchase clean energy technologies like solar and substantially cut customer-acquisition costs. “You can build a propensity model that helps you to focus marketing dollars on high-propensity customers,” he said.
With more data and smarter marketing, Toth believes that, over time, customer acquisition costs can be sliced sixfold. “Today, solar customer acquisition costs are in excess of 15 percent to 17 percent of the total purchase price,” he said, comparing those costs with the 3 percent to 4 percent average in the car industry. “There’s no reason you should pay $6,000 to acquire a solar system [customer] when the hardware is costing less than that.”
If those cost reductions come to fruition, it’s safe to say that solar will become even less of a partisan issue than it is now.