Is the split incentive problem worse for college student renters: an analysis of landlord self-reported and hypothetical choices?

Abstract

In the residential housing sector, energy conservation issues may arise in the relationship between landlords and renters (a.k.a. tenants) due to principal-agent and information problems. An example is the split incentive, where one party makes the energy efficiency decisions while the other pays the energy bill. Herein, we investigate whether the landlord and renter split incentive problem may be more likely and more challenging for college student renters than those who are not college students. This may occur from landlords perceiving that college renters lack sufficient demand for energy efficient improvements.

There is a lack of studies regarding the possibility that college renters may face greater exposure to the split incentive problem. We surveyed landlords to better understand their prior energy efficiency investment decisions and used a contingent valuation question to further investigate their choices for a hypothetical return on investment scenario. The landlords had various mixes of college students and non-students in their properties. Landlords renting one single-family property exclusively to college students had, on average, completed fewer major upgrades to their rental properties and were less likely to invest in a hypothetical insulation upgrade.

1 Introduction

Parts of the introduction, background, and methods sections have been summarized from the authors’ work in a grant completion report (Haynes et al. 2020).

The split incentive is a common market failure that has been studied extensively by economists, energy experts, and environmental researchers. A type of principal-agent problem, the literature frequently refers to the split incentive problem in the context of the consumption of energy. In this context, the problem occurs most frequently between landlords and renters (a.k.a. tenants) but can exist whenever two parties have different incentives regarding the consumption of energy. Gillingham et al. (2012) outline possible scenarios in which the split incentive problem can occur. The most relevant for this study is that, if the renter pays the energy bills, fewer energy investments to the rental property might occur (see, for example: Levinson and Niemann 2004; Melvin 2018).

Herein, we investigate whether the landlord and renter split incentive problem may be more likely and more challenging for college student renters (a.k.a. tenants) than those who are not college students. This may occur from landlords perceiving that college renters lack sufficient demand for energy efficient improvements. That is, landlords who rent primarily to college students may be particularly hesitant to invest in upgrades if they would need to increase rent to partially or fully recover upgrade costs.

In 2020, the residential sector accounted for 17% of total primary energy consumption in the United States (U.S. Energy Information Administration (USEIA) 2021). While much research has focused on the concept of the split incentive and how this concept impacts energy consumption in rental housing, less is known about how landlords who rent to students make energy upgrade decisions compared to landlords who rent to non-students and whether student and non-student renters have different preferences for energy efficiency in their rental units. Specifically, this investigation will help to address the following research question: Is renting to students compared to non-students a significant factor in whether landlords have completed or are willing to pay for various energy efficient upgrades?

Employing a contingent valuation framework, this study gathers empirical insights to address this question.

This paper contributes to the split incentive literature by exploring in more detail the decision-making process of landlords regarding energy-efficient upgrades. We analyze both the completed upgrade choices and a hypothetical choice. The hypothetical choice allows us to control for several potential factors: type of product, added expense for an upgraded product, amount of utility bill savings from the upgraded product being installed, and the number of renters in the rental unit. We also allow landlords to increase rent to cover part or all of the expenses or to make a positive return on the investment. In addition, due to the mix of the landlords’ renter types, we are able to compare decisions made by landlords who rent exclusively to students to those who do not.

The paper is organized by sections. We discuss the theoretical framework in Sect. 2. In Sect. 3, methodological approaches, we outline methodological approaches used in the literature, findings from the literature, and the study context. Section 4 covers methods, including survey development, data collection, and modeling. We present results in Sect. 5 and discuss the results in Sect. 6. Section 7 concludes by summarizing, discussing limitations, and provides directions for future research.

2 Theoretical framework

Principal-agent studies in the industrial organization and regulation literature have long considered asymmetric information and its potential to reduce welfare (see for example: Leibenstein 1966; Spence and Zeckhauser 1971; Jensen and Meckling 1976; Loeb and Magat 1979; Baron and Myerson 1982; Laffont and Martimort 2001). Spence and Zeckhauser (1971) outline situations in an insurance market where, under imperfect monitoring, a benevolent insurer is unlikely to be able to provide a first-best contract. Loeb and Magat (1979) and Baron and Myerson (1982) address a cost information asymmetry between a regulator and utility.

Laffont and Martimort (2001) emphasize that often times a benevolent policymaker would face the same information problem as the principal; however, if the planner wanted to weigh the parties’ surplus differently, public policy could be utilized to assist in redistribution.

A significant literature has also developed around principal-agent and asymmetric information issues in energy conservation. When considering residential housing or commercial building rental, issues arise in the relationship between builders and purchasers or landlords and renters. An example is the split incentive problem, where one party makes the energy efficiency decisions while the other pays the energy bill. Blumstein et al. (1980) outline potential energy conservation barriers, including those that may stem from misplaced incentives or a lack of information or misinformation. Jaffe and Stavins (1994) focus on trying to separate failures leading to the energy-efficiency gap into market and non-market categories. Gillingham et al. (2009) outline energy efficiency market failures and potential policy options to alleviate the effects.

3 Methodological approaches

In this section, we outline methodological approaches used in the literature, findings from the literature, and the study context.

Environmental valuation techniques are typically divided between revealed and stated preference techniques (see, for example, Carson and Louviere 2011). Revealed preference techniques often estimate the value of environmental goods or services based on actual or reported behavior or purchases (e.g. travel cost and hedonic methodologies). Within the revealed preference realm, the hedonic method typically estimates the value of environmental goods or services as attributes of the price of a house. We do not have detailed data on energy efficient purchases made by landlords for each of their properties such as product, price, attributes, etc. Asking landlords to accurately recall or retrieve this information is an impractical request. Therefore, revealed preference techniques, such as the hedonic method, are not ideal for this application.

Stated preference techniques utilize hypothetical decisions in a survey to estimate value. Most relevant to this study are the stated preference techniques referred to as discrete choice experiments and the contingent valuation method. Discrete choice experiments vary the level of several attributes in multiple plans and ask respondents to choose which plan they prefer based on the attributes and price. A discrete choice experiment could have been designed to try to value the attributes of energy efficient products by listing those attributes, changing their levels and the payment amount in various plans, then asking respondents to choose between the plans. However, we think this indirect energy upgrade valuation method is unnecessarily complicated given our study questions and our ability to select energy efficient products with few non-energy related attributes and a high level of familiarity to the landlords (see, for example, Johnston et al. 2017, p. 18).

Contingent valuation typically creates one scenario with a change in the level of the environmental good or service and asks a respondent a single question about whether they would be willing to pay a specified amount for that change. The answer options are usually limited to two options (e.g. yes or no), also known as binary (a.k.a. dichotomous) choice. Herein we use the single bound dichotomous choice contingent valuation method to answer our research questions (see, for example, Carson and Louviere 2011). Dichotomous choice responses can be modelled using either logistic regression or binomial probit; in this case, we opted for binomial probit due to its alignment with the theoretical underpinnings of our study and its ability to accommodate the normal distribution assumption of the error terms in our data.

Given Phillips (2012) showed that landlords and renters may have different preferences and willingness to pay for different types of energy-efficiency investments, two different products were chosen to be labeled in the survey (we will refer to these as treatments).We used the U.S. Department of Energy’s “Home Energy Saver” calculator to find products that were significant financial investments and would result in relatively large utility bill savings for our area (http://homeenergysaver.lbl.gov/consumer).

3.1 Findings from the literature

There is a paucity of split incentive studies that focus on students or the potential that students have different preferences or are treated differently by landlords than other rental groups.

Carrol et al. (2016) used a discrete choice experiment to test whether a discrete change in energy rating was valuable. The design includes mentioning that the rating is tied to energy bills but no specifics on savings amounts were given. Their sample of 865 respondents was a mix of students (68%) and staff (32%) at Trinity College in Hartford, Connecticut. The results suggest renters will pay more for improvements. The results were not separated by the two groups; however, given the high percentage of students, the results may provide information about student renters from other areas.

Harvey et al. (2016) surveyed 555 off-campus student renters near Cornell University in Ithaca, New York. A single survey question was used to ask students how much more they would pay for energy efficient and comfortable housing. Increments of $25 up to $100 were used for answers. Results suggest 75% of students would pay at least $25 more per month, and about 30% of students responded with at least $50 more. It is not known whether potential utility bill savings, from the possible energy reducing investments, were incorporated.

Haynes et al. (2016) surveyed landlords and students in Duluth, Minnesota. Results suggest that students may be more prone to the effects of the split incentive for a number of reasons. Based on difference in means tests, students tended to choose properties that were larger and had more bedrooms than other types of renters did. This housing type would typically consume more energy. Also, utilizing frequencies, landlords who rent primarily to students were less likely to report furnace replacement as an energy efficient improvement to the majority of their properties than if they rented to other types of renters.

This paper contributes to the split incentive literature by exploring in more detail the decision-making process of landlords regarding energy-efficient upgrades. We analyze both the completed upgrade choices and a hypothetical choice. The hypothetical choice allows us to control for several potential factors: type of product, added expense for an upgraded product, amount of utility bill savings from the upgraded product being installed, and the number of renters in the rental unit. We also allow landlords to increase rent to cover part or all of the expenses or to make a positive return on the investment. In addition, due to the mix of the landlords’ renter types, we are able to compare decisions made by landlords who rent exclusively to students to those who do not.

3.2 Study context

Data for this study were collected from landlords of properties in Duluth, Minnesota in 2019. Although Duluth is considered a northern U.S. city, it is not unique in having a cold climate, aging housing stock, low vacancy rates, and a significant amount of both student and non-student renters. Climate is important in considering the benefits and costs of specific energy related upgrades, such as a furnace or air conditioner. Age relates to the likelihood that the properties can benefit from new technology or an increased amount of an energy related product (e.g. insulation). The vacancy rate may influence how responsive landlords may need to be, regarding property amenities, in order to compete in the market for renters.

Several major U.S. cities are in the same climate zone as Duluth and have similar efficiency recommendations for windows and insulation.² Duluth’s percentage of housing built prior to 1940 was 43.2% in 2020 (City of Duluth 2020); this is similar to Milwaukee, Chicago, and St. Paul/Minneapolis. Finally, Duluth’s vacancy rate (3.8%) is similar to Minneapolis/St. Paul (4.1%), Boston (3.8%), and Worchester, Massachusetts (2.6%) (U.S. Census Bureau 2019b). Internationally, nearly all major Canadian cities, Berlin, Prague, Vienna, Warsaw, Minsk, Kyiv, and Moscow are in the same climate zone (Beck et al. 2018). Duluth vacancy rates are similar to Montreal (3.5%), Quebec (2.7%), and Toronto (4.6%) (Statistics Canada 2022). The implication of these similarities is that our results may be relevant in these other areas.

We chose to use a furnace or insulation upgrade for the hypothetical purchase decisions for five reasons. First, they meet the large investment and large potential savings criterion. Second, these products are typically not viewed regularly by renters and therefore reduce the complication of providing potentially important types and amounts of benefits other than utility savings (such as the size, appearance, and features of windows and refrigerators). The reduction of alternate attributes allowed us to keep our purchase decision scenario descriptions relatively simple and we do not need to be concerned about strong preferences for those attributes unintentionally influencing the results. Third, although the climate is relatively cold, furnace and insulation products have changed over time and it is likely that many properties would incur some utility bill savings from upgrading these products. Furnace efficiency has varied considerably both over time and based on current standards. The U.S. Department of Energy reports that older low-efficiency systems can have be as low as 56%–70% efficient based on annual fuel utilization efficiency (https://www.energy.gov/energysaver/furnaces-and-boilers). In addition, the furnace efficiencies currently available for purchase range from 80% to 98.5%. Therefore, there is a wide range of potential upgrades available and the furnace could even be upgraded multiple times. There are several types of insulation and recommendations for the levels of insulation have changed over time. For example, the 2021 International Energy Conservation Code (IECC) increased the recommended insulation thermal resistance level (a.k.a. R-value) for ceilings from 49 to 60 in zone 7 (https://codes.iccsafe.org/content/IECC2021P1/chapter-4-re-residential-energy-efficiency). There are also insulation options for above ground walls, basement walls, sealing, etc. Again, there are a wide range of potential upgrades available and the insulation could be upgraded multiple times. Fourth, the scenario and decision is hypothetical so we can force that the furnace must be replaced or that the an audit recommended the insulation be upgraded. The change in efficiency level is implicit in the utility bill savings. Fifth, we think it may be advantageous to incorporate one product that might be considered more of a necessity (furnace) and a second product that may be considered more of a luxury (insulation) in the design.

4 Methods

In this section, we outline the methods used in the study including survey development and data collection and the models utilized.

4.1 Survey development and data collection

To explore our research questions, a contingent valuation survey was designed and implemented using Qualtrics ™ software.³ The purpose of the survey was to gather evidence on how landlords make decisions about energy efficient investments.

4.2 Pretest survey

Prior to full scale distribution, a pretest survey was conducted on January 15, 2019. Participants were recruited through a postcard mailing sent to a random sample of rental property owners in Duluth and were compensated $50. Seven of the eight landlords who signed up to participate actually attended. Five respondents took the survey on a computer, while two completed it on their phones. After providing a significant amount of context, the contingent valuation question asked landlords: “Would you upgrade to the more energy efficient replacement?”.

After the pretest survey, several adjustments were made to improve the survey design. For example, the original survey design attempted to mimic a hypothetical decision where the landlords had the option of upgrading to a generic, more energy efficient product while increasing rent by a specified amount. It was suggested to make the scenario more consistent with a return-on-investment (ROI) choice for a specific investment (e.g. furnace or insulation). Also, several noted that the cost of the investment would need to be considerably larger before they would consider increasing rent to partially or fully recuperate those costs. Our initial values were $240 and $1,000; however, landlords suggested the minimum could be closer to $2,500.

Those who participated in the landlord focus group were asked to complete and provide comments on the new ROI question sequence in mid-February 2019. Their comments were generally positive. The new survey version was also sent to Duluth Landlord Association members for comment. Thirteen responses were received from both groups. After the final adjustments were made, the survey requests were sent to landlords in April 2019.

The final survey was comprised of the following sections: introduction and screening questions, questions about the number and types of rental properties and renters, energy payments and improvements, the hypothetical improvement choice and follow-ups, demographics, and closing.⁴

4.3 Hypothetical improvement choice: scenario design

We also designed a hypothetical purchase decision scenario and a return on investment (ROI) question sequence.

It is important to note that each landlord received only one treatment and one scenario. That is, they either received the hypothetical investment choice for the furnace treatment or the insulation treatment, then the corresponding randomly selected upgrade costs, renter savings, and number of renters to complete the scenario. Although this choice creates less data, it avoids the potential problem of anchoring bias based on the price of the first upgrade decision (see, for example, Johnston et al. 2017).

4.4 Furnace treatment design

The scenario asked a single binary choice question regarding the decision to upgrade the furnace. For example:

“Imagine that you are responsible for the management of a small, single-family home rental property with one tenant.

Imagine the furnace must be replaced. A replacement furnace with similar features costs $1,200.

Alternately, you could upgrade to a more energy efficient replacement for a cost of $2,400, paying $1,200 in upgrade costs (the extra cost compared to the replacement furnace). The upgrade would save the tenant a total of $20 per month in utility payments.

Please select the option that is most similar to what you might choose. You will be given the option to increase rent in upcoming questions.

Would you upgrade to the more energy efficient replacement?”.

Next, landlords were asked whether or not they would increase rent based on their choice. (Note, if the landlord does not choose to upgrade they hypothetically purchase the replacement furnace and still may want to increase rent to recover some of those costs). Those selecting “No” were asked follow-up questions. Those choosing “Yes” continued through the ROI sequence by selecting the percentage (50% or 100%) of which type of cost they would recover (replacement, upgrade, or total).

After making this choice, options for increasing rent to recover costs in two, three, four, or five years were presented along with the specific rent increases per month that would be necessary to meet the landlords’ selected goal.⁵ Figure 1 summarizes the simplified question sequence.

Fig. 1

Summary of Landlord Survey Question Sequence

The U.S. Department of Energy’s “Home Energy Saver” calculator was used to help inform the choices for the upgrade costs and savings amounts (http://homeenergysaver.lbl.gov/consumer). When using Duluth zip codes, and defaults entered based on that zip code, the estimated furnace upgrade costs varied widely for each efficiency grade. AFUE is Annualized Fuel Utilization Efficiency, which measures a gas furnace’s efficiency. The midpoint expense of the range for the 90 AFUE furnace upgrade is about $2,600, while the midpoint of the range for the 96 AFUE furnace is $3,200. To lower the risk of overshooting on the low value (and receiving all “No” answers to the hypothetical choice question), the upgrade values were chosen to start at $1,200 and go up in $200 increments to the maximum of $3,600. The yearly utility payment savings estimates from the upgrade furnaces were about $190 to $260; therefore, $20 per month was chosen as a baseline. Since the amount of savings were uncertain but could be a factor in the landlord’s decision to purchase an upgraded furnace, an alternative savings amount of $40 per month was created.

Additionally, landlords may be more likely to raise rent due to an upgrade if there are more renters to cover the rent increase (e.g. landlord could increase a single renter’s rent $40 per month or increase rent $10 per month with four renters). The mean number of bedrooms for Duluth non-student rental properties was 3.6, and for student rental properties, it was 4.1. Therefore, the number of renters in the scenario was varied as either one or four.

4.5 Insulation treatment design

In the insulation treatment, the baseline was a sub-optimal level of insulation. A variety of insulation options exist with various expenses and potential savings.

“Imagine that you are responsible for the management of a small, single-family home rental property with one tenant.

A recent home energy audit recommended that the insulation be upgraded. The cost to upgrade would be $1,200, saving the tenant $20 per month in utility payments.

Please select the option that is most similar to what you might choose. You will be given the option to increase rent in upcoming questions.

Would you upgrade?”

Using the “Home Energy Saver,” calculator inputs to aid in choices again, one option identified would be to install wall insulation and additional attic insulation that would have approximate costs of $1,500 and estimated savings of $250 per year. If basement wall insulation was also added, the costs are about $3,400 with savings of $500 per year. These matched reasonably well to the furnace upgrade cost range of $1,200 to $3,600 and the two savings values of $20 or $40 per month.

Similar to the furnace treatment, landlords were asked whether or not they would increase rent based on their choice to upgrade and, if they chose to upgrade, they were to complete the ROI sequence.

4.6 Summary of treatment design and implementation

Table 1 summarizes the treatments and variable values that were used in the survey. We randomly assigned participants to treatment groups while ensuring similar sample size for the survey versions.

Table 1 Summary of values for hypothetical investment choices

5 Data Collection

Survey participants were recruited using an email address list of 2,289 supplied by the City of Duluth’s Life Safety Department. A random drawing for one $500 Visa gift card was utilized as an incentive for all who fully completed the survey. The online Qualtrics survey was administered from April 17 through May 3, 2019. The survey produced 369 respondents; however, some responses were incomplete and could not be used in all the analyses. The dataset contained 338 observations. Table 2 presents a description of the variables in one or both of the presented regression models.

Table 2 Variable description

5.1 Models

To model the variables associated with the count of major improvements completed to a majority of the landlord’s properties, we utilized the Poisson model.

Let, y_i = the count of major improvements completed on a majority of properties. The Poisson probabilities are⁶:

$$ {\text{Prob}}[y_{i} = j] = \frac{{e^{{ - \lambda_{i} }} \lambda_{i}^{j} }}{j!} $$

(1)

with j = 0, 1, 2, 3, or 4 major improvements completed on a majority of the landlord’s properties.

$$ {\text{Log}}(\lambda_{i} ) = \beta_{0} + \beta_{1} x_{1} + \ldots + \beta_{n} x_{n} $$

(2)

with mean = variance = $\lambda$.

To model the upgrade decision choice, the Binomial Probit model was utilized.

Let, Y = the binary upgrade decision choice where Y = 0 if no and Y = 1 if yes.

$$ {\text{Prob}}\left[ {Y = {1}} \right] = \Phi (\beta_{0} + \beta_{1} x_{1} + \cdots + \beta_{n} x_{n} ) $$

(3)

where $\Phi$ = cumulative normal distribution.

$$ {\text{Prob}}[Y = 0] = 1 - {\text{Prob}}[Y = 1]. $$

(4)

6 Results

In this section, we describe the data and results of the regression analyses. Table 3 presents the means of the variables used in regression analyses.

Table 3 Means of the variables used in regression analyses

6.1 Self-reporting of improvements

To consider our research question regarding completed improvements, we analyzed survey data regarding which resource efficient improvements had already been made to the majority of the landlords’ properties. The improvements were categorized into what are likely to be relatively major and minor investments based on costs. Specifically, programmable thermostats, plastic wrap insulation kits, and low flow shower heads are expected to cost less than $100 each and might even be purchased by renters. Furnace tune-ups seem to range between $100-$200 each. LED lighting costs would depend on whether just the bulbs were being replaced as compared to both the bulbs and housings. Similarly, Energy Star appliance costs are highly dependent on the product; for example, a dehumidifier may be quite inexpensive while a refrigerator or clothes washer could be a large expense. The remaining four improvement types (add or replace insulation, window replacement, furnace replacement, and energy efficient water heater) seem most likely to be relatively costly.

The percentage of landlords reporting completed improvements to a majority of their properties are presented in Table 4 and are separated based on whether the landlord rents exclusively to students or not. Interestingly, there are some seemingly large numerical differences for four major improvements between the landlord types (the first four listed in Table 4). While the unconditional difference is only statistically significant for adding or replacing insulation, all of the major improvement types have smaller proportions for all-student renters. Next, we further explore these differences.

Table 4 Proportion of landlords who self-reported improvements to a majority of their properties

To further analyze this difference, we created a variable that was the count of whether landlords had completed those improvements to a majority of their properties. The value is either 0 (no) or 1 (yes) for each of the four major improvements; add or replace insulation, window replacement, furnace replacement, and energy efficient water heater. Therefore, the count variable was the sum of those individual values, which created one of five possible values for each landlord: 0, 1, 2, 3, or 4. The mean for landlords renting exclusively to college students is 1.47 (n = 60), while the mean for the others is 1.82 (n = 278). The relative frequencies of the counts of the 338 observations were: 22% (0 major improvements), 22% (1), 25% (2), 21% (3), 10% (4). Therefore, based on the count nature of the dependent variable, we began the analysis with a Poisson regression with the mean specified as log-linear.⁷

The model property that the mean equal the variance may be overly restrictive. In our sample data, the discrete mean = 1.76 with variance = 1.64. This indicated that the model may have underdispersion (variance less than the mean). One extension of the Poisson model that can account for either underdispersion or overdispersion (variance greater than the mean) is the gamma model. Greene notes that allowing “… the variance of the process to differ from the mean” could be interpreted as introducing unobserved individual heterogeneity into the Poisson model (Greene 2012, p. E931). This is advantageous since allowing “…for both observed and unobserved preference heterogeneity…” is recommended in the analysis of stated preference data by Johnston et al. (2017) (p. 56). We add a robust covariance matrix to the model as well to include potential latent heterogeneity in the mean (Greene 2012).

The results of the two regression models are presented in Table 5.⁸ AIC was used as the criteria for choosing the models (while maintaining the important control variable binary sets for length of time as a landlord and income and the count and type of properties).

Table 5 Major improvement count regressions with robust covariance

The models lead to similar coefficients, significance levels, and similar overall performance. However, the statistically significant dispersion parameter, with alpha greater than one, indicates underdispersion (consistent with the descriptive statistics). Therefore, we proceed by focusing on the results from the gamma model. Referring back to our research question, is renting to students compared to non-students a significant factor in whether landlords have completed various energy efficient upgrades?

Landlords who only rented to college students made up approximately 17.5% (n = 58) of the observations. After controlling for the count of minor energy improvements made, whether energy audits had been completed on over half of the properties, the duration of time as a landlord, income, which party pays the fuel bill, age, and the number of the property types, those landlords that rent exclusively to college students have completed statistically fewer major energy improvements. While the coefficient being statistically significant at the 7% level may be considered borderline significant, we can further examine the statistical and economic significance of the partial effects by conditioning on the number of single-family properties.

About 91% of student exclusive landlords in the sample have only single-family properties (53/58) with about 74% of those having only one property (39/53) and another 15% of those having two properties (8/53). The average partial is statistically significant at the 5% level for landlords renting exclusively to students with one or two properties.⁹ The point estimate of the partial effect, about -0.37 for both one or two properties, suggests about a one-third major improvement decrease on the majority of the landlord’s properties (similar to the difference in the unconditional means). Given the sample mean of those not renting exclusively to college students equals 1.82 major improvements, this change represents a nontrivial 20% decrease in the count of major improvements.

The self-reported improvement analysis results suggest, on average, landlords renting exclusively to college students with single-family properties have invested less in the count of major upgrades than landlords with a mix of renters.

6.2 Hypothetical improvement choice results

Next, we explore whether similar results will hold for a hypothetical purchase decision of an energy efficient product. A summary of the landlord’s responses is provided in Fig. 2.

Fig. 2

Summary of Upgrade Question Responses

The upgrade decision generated qualitative no or yes answers. The answers are converted to zeros and ones to create the binary dependent variable.¹⁰

The results of the regression model are presented in Table 6.¹¹ AIC was used as the criteria for choosing the models (while maintaining the important control variable binary sets). The likelihood ratio test of pooled versus separate models, for the furnace and insulation treatments, does not reject the pooled model. We also ran a pooled model with many matching interaction terms (e.g. insulation*upgrade cost, furnace*upgrade cost). Wald tests indicate only three statistical differences in the coefficients (all at the 10% significance level): the treatment binary for the $40 utility savings from upgrades, tenant pays the fuel, and the count of single-family properties. Similarly, likelihood ratio tests do not reject the restricted model presented as compared to the model with the large number of interaction terms.

Table 6 Hypothetical improvement choice probit regression with robust covariance

We ran robustness checks with and without the robust covariance matrix for both the probit and logit model specifications. The results are very similar to those in Table 5. Adding heteroscedasticity associated with the insulation treatment led to an insignificant disturbance variance term. We present the results with the sandwich estimator for the robust covariance matrix to adjust for potential unspecified latent heterogeneity (Greene 2012).

Referring back to our research question, is renting to students compared to non-students a significant factor in whether landlords are willing to pay for various energy efficient upgrades?

Landlords who rented exclusively to college students made up approximately 16% of the observations in the insulation treatment (n = 25). Of these 25, only two have property types other than just single-family (one has a single-family property in addition to a multifamily property, one has only one multifamily property). Therefore, we focus on the effect from landlords renting single-family properties to all college student renters. The modal number of single-family properties for landlords renting to all college students was one (67.0% of the all college student landlords in the insulation treatment and 65% of all college student landlords in both treatments). The effect was statistically significant for landlords renting exclusively to college students in the insulation treatment with one single-family property. The variance inflation factor indicates little concern of multicolinearity (VIF = 1.19). The average partial effect for landlords renting exclusively to college students with one single-family property is statistically significant at the 5% level, with the point estimate suggesting a decreased likelihood of choosing to upgrade of about 33%. The negative effect was not present for more than one single-family property; however, the number of observations of these types of landlords quickly diminishes (four landlords with two properties, one with three, and two with five).

There was no statistical differences when considering a furnace upgrade. Therefore, the type of upgrade might matter; for example, whether the improvement type is considered a necessity (furnace) versus more of a luxury (insulation).

The negative and significant coefficient for landlords renting solely to college students in the insulation treatment with one single-family property is evidence that those landlords are less likely to choose to support an insulation upgrade in our hypothetical scenario. It was estimated that they were about 33% less likely to choose to upgrade, on average.

7 Discussion

The survey was designed to help understand which energy upgrades had already been made in a majority of landlord properties and to understand how landlords would respond to a hypothetical upgrade decision.

The sample included landlords who rented exclusively to students. Some landlords renting exclusively to college students clearly indicated a preference for high quality rentals: “My niche in rental properties is to have the safest and nicest student housing in Duluth… Students need to be treated as customers vs tolerated renters.”

Yet, this contrasts with the comments made by another landlord:

“It depends on the rental market. If tenants are interested in lower-cost monthly utilities, then yes, I would replace it. Typically, single families are interested in this. Students are not typically interested in lower-cost utilities. So, my answer is ‘yes’ to one of my rentals that is a single-family and ‘no’ to the student rentals.”

On average, landlords with one or two single-family rental properties that rent solely to college students completed 20% fewer major upgrades than other landlords (5% significance level). The point estimate of the partial effect, about -0.37 for both one or two properties, suggests about a one-third major improvement decrease on the majority of the landlord’s properties (similar to the difference in the unconditional means). Given the sample mean of those not renting exclusively to college students equals 1.82 major improvements, this change represents a nontrivial 20% decrease in the count of major improvements.

It is possible that the major improvement differences are related to property characteristics and energy consumption differences between student and non-student rentals. It was impractical to ask the landlords to provide detailed property characteristics for their properties as part of the survey, given the trade-off between the length and intensity of the survey and response rates (43% have more than one property). Asking for addresses would also have created an anonymity problem, as the city collects information on rentals.

However, we were able to make some generalizations utilizing a separate data set of approximately 2,000 rental properties (about 600 student properties) in Duluth from Haynes et al. (2016). Our analysis indicated that for nearly 80% of the properties (multi-family and single) there is no statistical difference in the building age of college student and non-student rentals (all means are over 90 years old), while college student rental buildings are larger on average. Additionally, natural gas and electrical consumption in single-family student exclusive rental properties were found to likely be at least that of non-student consumption on average. This information makes it considerably less likely that our result of fewer major improvements in the majority of single-family student exclusive properties are due to those properties being more efficient via age and size. It is also unlikely that energy consumption in student exclusive rental properties is lower and, therefore, that the energy improvements would yield a lower investment return.¹²

Similarly, those renting one single-family property exclusively to college students were about 33% less likely to invest in a hypothetical insulation upgrade (they were not less likely in the furnace treatment). It is unclear whether the results hold for landlords with more single-family properties as our sample sizes quickly diminish after two single-family properties for those landlords renting exclusively to college students. The negative effect was not present for more than one single-family property; however, the number of observations of these types of landlords quickly diminishes (four landlords with two properties, one with three, and two with five).

Jaffe and Stavins (1994) indicate that quality information to the adopter is not necessarily sufficient for improvements since the adopter may need to recover their investment from the those that benefit; their example is that “…landlords may not be able to recover all of the value of such investments (in the form of higher rents) where renters pay fuel bills” (p. 809). However, Harvey et al. (2016) indicated that 75% of students would pay at least $25 more per month. If we utilize that value, a back-of-the-envelope calculation suggest a single student renter would pay $1,500 more over five years or that four renters would pay $6,000 more over five years (recall that the mean number of bedrooms for non-student rental properties was 3.6 and 4.1 for student rental properties). Our highest upgrade value was $3,600. An additional $6,000 rent would likely cover that expense, even after allowing for a reasonable discount factor for the landlord. Landlords may even consider keeping rent permanently higher; this idea was mentioned in a comment, “Rent can be raised to offset the utility savings and ultimately increase returns.”

Therefore, some landlords may be mistakenly assuming that college renters lack sufficient demand for energy efficient improvements. If so, surplus opportunities, for one or both groups, are likely being missed. In addition, societal losses from missed opportunities to reduce energy may be occurring and hampering city sustainability goals (for example, Duluth’s carbon neutrality by 2050 goal; https://duluthmn.gov/sustain/). If the benefits outweighed the costs, a city policy intervention may be justified to provide an information program as suggested in Gillingham et al. (2009). An important next step would be to better understand the demand for energy efficient improvements of both student and non-student renters in the same geographic area as the surveyed landlords.

8 Conclusion

We investigated whether the landlord and renter split incentive problem may be more likely and more challenging for college student renters (a.k.a. tenants) than those who are not college students. This may occur from landlords perceiving that college renters lack sufficient demand for energy efficient improvements. We surveyed landlords about their completed energy efficiency improvements and asked them a single bound dichotomous choice contingent valuation question regarding a hypothetical upgrade decision. On average, landlords with one or two single-family rental properties that rent solely to college students completed 20% fewer major upgrades than other landlords (5% significance level). Regarding the hypothetical upgrade decision, landlords renting solely to college students in the insulation treatment with one single- family property were about 33% less likely to choose to upgrade the insulation (5% significance level). This is a problem because surplus opportunities for one or both groups are likely being missed and societal losses from missed opportunities to reduce energy consumption may be occurring. Future research in this area might focus on why some landlords may be mistakenly assuming that college renters lack sufficient demand for energy efficient improvements.