Introduction

Households consume 21% of energy in the United States1,2 through activities like heating, cooling, and cooking, among many others. Although estimates vary, energy efficiency measures like sealing doors and windows and insulating the building envelope could significantly reduce energy consumption, saving households money and reducing emissions associated with energy generation (e.g., carbon dioxide) that cause climate change (Energy.gov 2022a). Household energy efficiency, therefore, can provide multiple benefits, although its potential is largely untapped3.

Programs that encourage households to improve the energy efficiency of their homes are typically implemented by municipal governments and utilities, sometimes acting with external partners and funding from non-governmental organizations (NGOs) or the federal government. A typical energy efficiency program will provide an energy audit to a household and identify specific retrofits that could be made to improve the efficiency of the home. Generally speaking, the small body of academic and academic-adjacent literature on these programs suggests that such recommendations, if implemented well, can yield significant gains in energy efficiency and cost savings to households; however, participation in home energy efficiency programs is typically low and sporadic given that households can potentially reduce their utility bills and increase the value of their homes4,5. That is, households do not appear to be consistently engaging with programs that may be financially advantageous.

Why aren’t consumers taking advantage of programs that, at least in the long run, could reduce their utility bills? These programs are often poorly publicized, and many households may assume that they simply cannot afford to improve their homes5. Yet, at the same time that energy efficiency technologies and measures available have improved and innovations like rooftop solar have become more affordable, partisan polarization has also intensified, significantly impeding effective climate policy in the USA6,7,8. Thus, partisan polarization is a potential threat to energy efficiency and the clean energy transition yet this relationship has not been thoroughly studied9. Here, we use two sources of data to evaluate the effect of partisanship on energy efficiency program participation in a setting that is representative of mid-sized USA cities with proactive municipal and utility programming intended to reduce residential energy consumption.

What drives household decision-making about home energy efficiency?

Households make multiple decisions about their homes, such as what appliances to purchase, what colors to paint walls, and others. While some of these are conspicuous (e.g. exterior upgrades, new flooring), many that improve the efficiency of the home or safeguard the health of its residents are not readily obvious to a casual observer. Here, we discuss the relevant literature on the latter type of home upgrade and known drivers of decision-making about the home.

A small body of literature evaluates why homeowners decide to mitigate against radon, which is a known exposure risk that may be present in the home environment and for which changes made to the home to mitigate this risk are inconspicuous. In some sense, radon mitigation is like energy efficiency improvements because it is not a conspicuous upgrade like rooftop solar yet provides health benefits. Thus, this literature can provide some insights into energy efficiency upgrades. Generally, this body of work implies that a combination of financial incentives and a household’s financial situation help determine their decision-making10,11,12. Importantly, the interventions needed to mitigate radon are well defined, highly effective, and typically range in cost from several hundred to several thousand dollars. Energy efficiency upgrades are often perceived as prohibitively expensive13, and financial factors typically outweigh environmental concerns, although the latter often have some role in decision-making14,15,16,17. Multiple interventions make use of behavioral “nudges” to encourage households to reduce their energy consumption or adopt energy efficient technologies. The most well-known examples of nudges involve providing households information about the energy consumption of comparable neighbors18 or similar interventions that are meant to shift social norms19. Generally speaking, carefully designed nudges do seem to encourage households to reduce their energy consumption or improve the energy efficiency of their home by making upgrades, although the magnitude and duration of these effects continues to be studied20,21,22.

Why partisanship matters

To some extent, the state and local policy environment can shape the decision-making of households by providing rebates, incentives, or tax savings for energy efficiency upgrades. Generally speaking, city or state governments lead by Republicans are less likely to implement programs to encourage energy efficiency programs or pass policies23,24,25. Thus, partisanship has indirect effects on energy efficiency by creating policy environments where home energy efficiency retrofit decisions may be more or less likely.

There is also polarization among the public. The large volume of research on climate change polarization (e.g. refs. 8,26,27,28), and a large but smaller literature on energy attitudes and preferences and polarization (e.g. refs. 7,29,30), implies that Republicans are far less likely to believe in climate change and tend to prefer fossil fuels when compared to Democrats. However, attitudes towards renewable energy tend to be less polarized than beliefs about climate change, with studies finding minimal to small differences between Republicans and Democrats (or liberals and conservatives)31,32,33.

The research on pro-environmental behaviors and consumption / purchasing decisions is less clear. People who self-identify or are identified as conservatives are somewhat less likely to purchase electric vehicles and solar panels, but partisan differences appear to be modest34,35,36 have compared voting districts in Texas and New York and note that rooftop solar was more common in many heavily Republican districts in Texas37. have assembled political donation data with data from Google’s Project Sunroof. They find that Democratic households are only slightly more likely to have rooftop solar systems than Republican households, implying that there is very limited polarization on the decision to adopt solar power for the home. Yet these studies, while impressive, have some limitations in terms of ecological inference and relying upon political donation data to identify partisans.

Of course, adoption and implementation of rooftop solar systems and electric vehicles (EVs) at the household level are high-profile decisions that often have a set of social norms and preferences related to them38,39,40. That is, EVs and solar are costly and these products are well-known to most people in the USA—there are few people that have not formed an attitude about them41,42. However, many energy efficiency upgrades are comparatively less visible, and perhaps less likely to trigger social norms, pre-existing preferences, or partisan identities. Energy efficiency upgrades such as insulation or door and window sealing are likely of comparatively lower salience, are typically less expensive, and should not enflame partisan passions in the same way as EVs or solar. On the other hand, the differences are not always large, but several studies imply that people who identify as conservatives are less likely than others to engage in pro-environmental behavior43,44,45,46. Thus, the literature implies that it is not a given that partisanship will influence household energy efficiency decision-making.

As we noted above, there is a large body of work on “nudges” of various types to encourage households to reduce their energy consumption and/or upgrade the energy efficiency of their home. These interventions hold great promise because they are often low cost and the actions taken by households are voluntary. Yet, much of the literature on nudges has some limitations because researchers rarely have access to household or individual-level data for socio-demographics, values, or ideologies21. Other work has used voter registration and energy consumption data and found that conservatives are less likely to request information about their energy use, and less apt to respond to nudges47. Partisanship appears to have intensified United States, and it is possible that energy efficiency and conservation could become more nakedly partisan issues9. However, we are not aware of prior studies that consider the relationship between partisanship and energy efficiency program participation. In this paper, we evaluate the effect of partisanship on participation in a home energy efficiency program in a Colorado city, using voter registration and survey data described in the methods section.

Results

Voter registration and assessor data

Table 1 provides the logistic regression results for the models from the voter registration and assessor data. We start with a model that includes only the household-level variables. In this model, square footage, home value, heat pump, and the age of the home in years are all positive and statistically significant, while homes with more acreage were less likely to participate in the program. We introduce the partisanship and voter status variables in model 2, which has improved model fit as both the AIC and BIC are lower. Model 3 adds the additional person-level predictors for age and sex, yet these additional predictors reduced the AIC, but not the BIC, which suggests mixed evidence of model improvement. In this model, registered Democrats are more likely than those who identify politically as unaffiliated/other party to participate in the program (0.049, p < 0.01) and Republicans are less likely (−0.062, p < 0.001). We ran the final three models using the restricted age ranges (above 25, above 30, and above 35 only). The effects of most of the predictors are very stable, even when data is restricted to a narrower age range—we interpret this as evidence that the problem described above (i.e., young respondents may not be decision makers in the home) does not alter our results.

Table 1 Logistic Regression for program participation, voter registration, and assessor data.
Full size table

We provide average marginal effects for Democratic and Republican voter registration in Fig. 1. The results shown in the figure imply that the probability of participation for registered Democrats was 0.01 higher than “Other/Unaffiliated” group, and 0.01 lower for the registered Republicans. Many of the household-level variables were statistically significant, which imply that respondents who lived in larger or more expensive homes were more likely to participate. Active voters are also more likely to participate (b = 0.120, p < 0.001).

Fig. 1: Average marginal effects of voter registration on energy efficency program participation.
figure 1

Figure displays the change in probability of energy efficiency program participation, holding all other variables at their observed scores.

Full size image

We calculated predicted probabilities (Fig. 2) to compare active Democrats and active Republicans across the range of different home values. Recall that the mean home value in our sample was some $474,000, or 47.44 in our transformed variable. Figure 2 implies that there is a consistent and small gap between Democrats and Republicans, and respondents who live in more expensive homes are more likely to participate. However, this effect is likely small in practical terms. For instance, an increase from the mean (47.44) to two standard deviations above the mean (81, or $810,000) is only associated with a 0.018 increase in the probability of program participation.

Fig. 2: Predicted probabilities of program participation for active Democrats and active Republicans and home values.
figure 2

Note: All other predictors held at their observed values.

Full size image

Survey data results

We turn to our survey data (Table 2). The first model includes only the demographic controls, and the second model adds the partisanship predictors. In this model, the demographic predictors do not approach statistical significance. The next model adds the partisanship predictors. Political party is not statistically significant in either model, but ideology does appear to predict program participation, with individuals who identify as “somewhat” or “strongly” liberal more likely to participate than those who identify as strongly conservative. The AIC statistic has improved, but the BIC has increased, indicating that model fit has not necessarily improved with the inclusion of these new predictors. Overall, the survey data implies that partisanship, as captured by political ideology or party affiliation, is not an especially powerful predictor of program participation.

Table 2 Regression models for program participation, survey data.
Full size table

Discussion

Local and municipal governments in many parts of the USA have implemented energy efficiency programs to encourage adoption of energy efficiency upgrades. These upgrades can potentially save households money and reduce greenhouse gas emissions. Yet partisanship could be a potential barrier to household energy efficiency, in the same way that it has long been a barrier to meaningful climate legislation in the USA. We used two different sources of data to evaluate how party affiliation or political ideology predict participation in a municipal energy efficiency program.

Our results have some sanguine implications. Using voter registration and county assessor data, we found a very weak yet robust effect of voter registration, wherein Republicans were incrementally less likely to participate than Democrats. Using survey data, we found a null effect of party affiliation and a small effect of political ideology. Thus, there are partisan differences between Democrats and Republicans, but at an individual scale these differences are small.

The survey data and voter registration data also produced slightly different results, wherein voter registration was statistically significant and relatively robust with a small effect size, and the effect of political ideology was not very robust in the survey data. Some of these differences may have occurred because the predictors of interest are measured in fundamentally different ways—one is voter registration, the other is self-identified. Colorado has open primaries such that any registered voter can vote in any primary—that is, a registered Republican can vote in the Democratic primary, and vice versa. This may have impacted our results, and perhaps we would have found stronger partisan effects in a closed primary system because partisans would have a stronger incentive to register with their party.

Although there are modest to null partisan effects, we caution against dismissing partisanship as a potential challenge for residential energy insecurity and the broader goal of decarbonization. Even a small effect at the household level could have a much larger effect at the system level if scaled up to the grid level. That is, a small effect of partisanship at the household scale could have larger implications than it may initially seem—Republicans are only a few percentage points less likely to participate than Democrats, but at the grid level this could translate into thousands (or even millions) fewer participants in energy efficiency programs.

Efforts have been made to promote renewable energy technologies in a way that is non-partisan, often by emphasizing cost savings, and our work implies that the same efforts should be made for energy efficiency upgrades48,49. However, there is currently little research on how to frame energy efficiency upgrades (e.g. insulation, weatherization) in ways that would not activate conservative or Republican political identities—addressing this topic is essential to allow homeowners to harness all the benefits of energy efficiency. Partisan effects may be relatively small to null for energy efficiency program participation because program participation is unobtrusive and voluntary and may reduce utility costs and increase home equity. Partisan effects for more restrictive and less voluntary measures would likely be stronger50,51,52.

As with all research, the current study is not without limitations. As we noted earlier, Fort Collins residents have consistently elected Democrats, and registered Democrats out-number registered Republicans. Still, our survey sample likely over-represents Democrats, and this may have biased our results to an unknown degree. Further, we suggest that our results, derived from a medium-sized city in the mountain West region of the USA, may not fully generalize to all contexts. Other methodological approaches are also needed. For instance, media analysis might reveal patterns in how energy efficiency technologies are discussed in partisan news sources, which might explain some of the effects we observed here. Qualitative interviews may reveal divergent motivations between Democrats and Republicans, or liberals and conservatives, and might be useful to represent conservative voices We encourage practitioners and scholars to give careful attention to the potential challenges posed by partisanship as a barrier to energy efficiency.

Methods

Study context

Our study is situated in Fort Collins, CO, a city of 166,000 residents located in Colorado’s Front Range region approximately 60 miles north of Denver, CO. Fort Collins has a highly educated population—56% of the city’s residents have a Bachelor’s degree or higher, compared to 34.98% for the entire USA—and has an economy centered on the tech sector, a large university, several hospitals, and renewable energy (ACS 2023a) Politically, the city government has consistently been led by Democrats and the city rests in a house district that has sent a Democrat to Congress since 1975.

The city government (City of Fort Collins), often coupled with the municipal utility (Fort Collins Utility Services), has encouraged energy efficiency through a variety of programs. The current study considers a program called Efficiency Works Homes, which has evolved into what is currently called Epic Homes. It was created to assist homeowners with making energy efficiency improvements in their homes. The city provided a free energy assessment to homeowners. The comprehensive energy assessment resulted in a report for the home or property owner that identified measures capable of increasing home energy efficiency, including various insulation options, appliance replacement, and upgrades to, or installation of, heating, ventilating, and air-conditioning (HVAC) equipment, and other upgrades. Then, homeowners were linked to resources, including contractors and low interest loans (if they qualified) as well as any available rebates. Essentially, the Epic Homes program streamlines the potentially confusing and time-consuming process of identifying areas for improvement, finding a quality contractor, procuring funding, and applying for any tax rebates associated with a home retrofit. We use two different data sources to understand participation in this program53. point out that many studies of program participation have recruited participants and non-participants through different means—our data sources do not have this bias. Ethics approval was provided by the Colorado State University Institutional Review Board.

Voter registration dataset

The first portion of our analysis is in the spirt of Costa and Kahn (2013), who linked voter registration to household energy consumption data. We collected voter registration data for residents of Fort Collins and married that to data from the county assessor’s office (Larimer County) for housing characteristics. The latter dataset included information for all parcels within the county and required extensive cleaning. We first removed parcels with an address outside of Fort Collins. Next, we removed commercial properties, multifamily units, empty lots, and ranchlands to isolate single-family residential homes. Merging the voting registration data with the assessor data produced a dataset of 102,000 cases.

There was an additional complication with the final dataset. Many addresses had multiple lines per household because multiple people lived in that home. We considered a few issues to address this problem. First, we could have assumed that the deed holder as listed by the county assessor was the primary decision maker in the home and could have deleted all other residents of the same addressed from the analysis. However, we suspect that energy efficiency upgrades are often “whole house” decision wherein multiple adults in the home hold some decision-making power, not just the legal homeowner. On the other hand, younger adults in the home might be children or other former dependents of the homeowner who do not play a large role in decision-making. We opted to report a series of models, using an age variable that was provided with the voter registration data to trim the dataset of younger respondents—the logic of this approach is that, below a certain age, respondents are unlikely to be among the primary decision-makers in their home.

Our dependent variable is a binary indicator of participation in the Epic Homes program. At the individual level, our predictors are the following: a measure of voter registration (1 = unaffiliated/all others, 2 = Democrat, 3 = Republican), female sex (0 = all other genders, 1 = female) and age in years. At the household scale, we include several variables that serve as proxies for the financial status of the homeowners. These include square footage (in hundreds of square feet), logged total acreage, the value of the home (in thousands of dollars), and number of bedrooms. We also include a binary indicator for whether the home uses a heat pump (0 = other type of heating, ventilating, and air-conditioning (HVAC) system, 1 = heat pump). We use the heat pump variable because homes with a heat pump may have already been updated or improved energy efficiency. Finally, we also include a variable for home age. We enter these variables into logistic regression models in groups and evaluate AIC and BIC statistics to determine improvements in model fit54. We use average marginal effects (AMEs) to understand the effect of key predictors given the well-known challenges of directly interpreting logistic regression coefficients55,56,57.

Survey data

For our survey data, we used a mixed-mode design for data collection. First, our partners with the city of Fort Collins provided contact information for participants in the Efficiency Works Homes program (n = 1693). This list had 310 duplicate email addresses and 136 that were undeliverable. We attempted to correct the undeliverable email addresses by finding incorrect or missing characters but were only able to adjust three (3) of them. We contracted with Marketing Systems Group, a large provider of survey samples and other services, to provide a sample of Fort Collins residents. Their initial list had 3321 emails but 89 were duplicates and 269 were undeliverable. For both samples, we asked respondents if they lived in Fort Collins and were 18 years of age or older—if they did not pass this screening, they were thanked for their time and the survey was terminated. Eight-hundred thirty-five respondents began the survey over email.

For the second phase of data collection, we implemented a hybrid of the push-to-web and drop-off pick-up method. The push-to-web approach involved encouraging a potential respondent to navigate to the online survey by dropping off a door hanger. We delivered 632 door hangers, but only 33 respondents completed the survey using this method. Given this low response and the effort required to distribute door hangers, we pivoted towards completing data collection by using a push-to-web mailer to 4328 addresses, which resulted in an additional 106 completions. In total, 974 respondents began the survey, but the estimation sample for our regression models include roughly 600 cases because of missing data. Among the 974, 13 were disqualified by the screening question and another 50 did not answer a single question beyond the screening question. Consistent with our prior research efforts in this part of the USA, a non-trivial portion of respondents did not answer demographic questions, particularly questions about income—about 40 respondents who answered most of the survey questions did not answer the income question. Using 600 as the number of completions and 911 as the number of surveys that were started (e.g. the total number of respondents who navigated to the survey minus those who were not in the study area and those who did not answer a single question), our completion rate was 65.8%. Using the most conservative response rate calculation (AAPOR 1), the response rate was 9.37%. This calculation assumes that all non-responders of unknown eligibility are eligible to take the survey—a less restrictive response rate calculation could have produced a substantially higher response rate. Supplementary Tables 1 and 2 provide descriptive statistics for both datasets. Before beginning the survey, respondents were provided with an introduction that explained their rights as a participant, included contact information for the Institutional Review Board, and explained that consent to participate was given if they navigated to the survey.

We provide extensive robustness checks in Supplementary material.