The role of demand-side incentives and charging infrastructure on plug-in electric vehicle adoption: analysis of US States

The transport sector is responsible for 23% of global energy-related greenhouse gas (GHG) emissions and is expected to more than double by 2050 in the business-as-usual scenario (International Energy Agency (IEA) 2013); faster than other energy end use sectors (Sims et al 2014). In addition to the global climate implications, the sector is also the leading cause of local air pollution, with near-term damage from non-GHG air pollutants more severe than those from climate change (Tessum et al 2014, Delucchi 2000, Hill et al 2009, Michalek et al 2011). In the US, transport sector is responsible for 28% of GHG emissions; with 60% coming from light duty vehicles (LDVs) (ORNL 2013). Fulton et al (2013) estimates that if strong efficiency improvements and fuel switch measures could be achieved with LDVs, it is possible to meet GHG reduction targets without major changes in travel demand (Fulton et al 2013). With urban mobility highly dependent on personal transportation modes in the US, emissions reductions in LDVs are key to achieving significant GHG reductions in the transport sector.

Transitioning from conventional gasoline vehicles to electric drivetrains is one solution to reduce GHG emissions and improve urban air quality while providing greater economic security (Graver et al 2011). The IEA recommends that plug-in electric vehicles (PEV) make up at least 40% of new vehicle sales globally by 2040 to stabilize GHG concentrations at 450 ppm (IEA 2015). PEVs can not only produce near-zero tailpipe emissions but also lower GHG emissions, depending on the electricity generation mix (Orsi et al 2016, Wood et al 2017). As the US transitions to a cleaner electricity mix, PEVs should provide ample GHG and air quality benefits (Onat et al 2015). Actualizing these benefits require significant market penetration, which depends on solving challenges related to economics, consumer behavior, and a technology that deviates significantly from existing dominant design in terms of vehicle range, maintenance, and refueling (Anderson and Tushman 1990, Hickman and Banister 2012, Bjerkan et al 2016). Throughout the rollout of PEV models in the US, more than 400 supply- and demand-side incentives and regulations related to PEVs have been implemented at the state/local level by governments and electric utilities (AFDC 2016). In addition, there have been several initiatives by the federal government to promote PEV adoption (see appendix 1 of supplementary information (SI)) available at stacks.iop.org/ERL/13/074032/mmedia. PEV sales in the US increased by 40% in 2016, reaching a total stock of over 500 000 vehicles—driven in part by advances in PEV technologies, policy support, and infrastructure development (Wood et al 2017, IHS Markit 2017). However, PEVs (plug-in hybrid electric vehicles (PHEVs) and battery electric vehicles (BEVs) combined) still comprise less than 0.9% of all LDVs sold in the US (IEA 2017). Therefore, analysis of the efficacy of various PEV-promotional policies and other contextual factors is needed to guide future action.

1.1. What influences PEV adoption?

2.1. Dependent variable—PEV, PHEV or BEV purchases

2.2. State level incentives

2.3. PEV charging infrastructure

Public charging networks expand in response to charging demand from existing PEV owners while simultaneously promoting further consumer acceptance by alleviating range anxiety (Wood et al 2017). Public charging stations (hereinafter referred to as EVSEs) in recent years have installed level 2 (L2 AC) or DC-Fast charging (DCFC) EVSEs that charge up to 48 times faster than the 120 V level-1 (L1 AC) chargers (AFDC 2018). There has been a substantial increase in L2 installations around city centers and DCFCs along highways since 2015 (see figure 1) (Francfort et al 2017).

Zoom In Zoom Out Reset image size

In the analysis, we represent PEV charging infrastructure by the number of public EVSE stations (not individual outlets) per capita drivers in each state in every quarter. We measured EVSE density as a ratio of per capita drivers and not per square mile because per capita EVSEs are a better proxy for EVSEs being in the vicinity of the users regardless of a state's population density. Figure 2 shows that the number of public EVSEs and gasoline price to a lesser degree appears to be related to vehicle purchases at the national level, inviting our in-depth analysis at the state level.

Zoom In Zoom Out Reset image size

Variables used in the analysis and the summary statistics are outlined in tables I and II of appendix 2 in SI. We first estimated the number of vehicles sold in a state s, for vehicle model v, in quarter t (2008–2016) using a panel data fixed effects model, with tax incentives, non-monetary incentives, charging infrastructure, and state demographics as key vectors influencing PEV adoption.

Model 1:

$$\begin{equation} \begin{array}{l@{}l@{}l@{}} {\rm Log}\left( {{\it Y}_{{\rm svt}} } \right) \hfill & = \hfill & {\it F}\left( {\psi ,S_{{\rm svt}} ,{\rm infra}_{{\rm st}} } \right)\hfill \\ {{\rm Log}\left( {{\it Y}_{{\rm svt}} } \right)} \hfill & = \hfill & { \propto _{{\rm sv}} + \beta *({\rm incentives}_{{\rm svt}} )} \hfill \\ {} \hfill & {} \hfill & {{\rm + }\gamma *\left( {{\rm infrastructure}_{s{\rm (}t{\rm - 1)}} } \right)} \hfill \\ {} \hfill & {} \hfill & {{\rm + }\delta {\rm *(State Characteristics}_{{\rm st}} {\rm ) + }\theta _{{\rm vt}} + \varepsilon _{{\rm svt}} } \hfill \\ \end{array} \end{equation}$$

The fixed effects model, however, does not control for the natural evolution of the technology, the influence of existing PEV owners on future owners (i.e. threshold effect) (Eppstein et al 2011, Sierzchula et al 2014), and the endogenous relationship between charging infrastructure and vehicle sales. Hence, we modified the base model with a lagged dependent variable as an explanatory variable to control for technology diffusion and the threshold effect, and lagged EVSE infrastructure variable to account for the endogeneity between charging infrastructure and PEV sales.

Model 2:

$$\begin{equation} {\rm Log} (Y_{\rm svt}) = F (\psi, {\rm Log}(Y_{{\rm svt}-1}, X_{\rm svt}, {\rm infra}_{{\rm st}-1}) \end{equation}$$

where F is the sales function with a lagged dependent variable and lagged infrastructure variable. Ψ represents the vectors of parameters and X represents the explanatory variables. Equation (1) transforms into the following:

$$\begin{equation} \begin{array}{l@{}l@{}l@{}} {{\rm Log}\left( {{\it Y}_{{\rm svt}} } \right)} \hfill & = \hfill & { \propto _{{\rm sv}} + \lambda *{\rm Log}\left( {{\it Y}_{{\rm sv}(t - 1)} } \right) + \beta *} \hfill \\ {} \hfill & {} \hfill & ({\rm incentives}_{{\rm svt}}) + \gamma *\\ {} \hfill & {} \hfill & \left( {\rm infrastructure}_{{\it s}({\it t}- 1)} \right) \hfill \\ {} \hfill & {} \hfill & {{\rm + }\delta {\rm *}\left( {{\rm State Characteristics}_{{\rm st}} } \right) + \theta _{{\rm vt}} + \varepsilon _{{\rm svt}} }. \hfill \\ \end{array} \end{equation}$$

Including the lagged dependent variable in a fixed effects model, however, violates strict exogeneity of explanatory variables with the error term. Therefore, we followed a generalized method of moments (GMM) approach (Arellano and Bond 1991) to instrument the lagged PEV sales and public EVSE infrastructure with their respective deeper lags.

Under the American Recovery and Reinvestment Act (ARRA), the federal government earmarked $37 million for installing about 4600 charging points across the nation (Peterson and Michalek 2013). Federal ARRA support to install EVSEs in early adopter states such as Illinois, New York, and Washington led to a significant buildup of EVSE infrastructure between 2011 and 2014 when PEV adoption had yet to scale (see figure 1). Instrumenting public EVSE variable with its own deeper lags, however, does not allow us to understand the influence of early adopters (i.e. ARRA recipients) and demand from existing PEV owners on public EVSE infrastructure. To account for this, we used a system of two equations; with the first stage of the model instrumenting public EVSE infrastructure with the timeline of ARRA funded EVSEs in different states and the cumulative stock of PEVs in each state. These two factors effectively account for early adopters and demand from existing PEV owners for additional public charging infrastructure. Lagged by one quarter, public EVSE estimates from the first stage are used in the main sales equation. We report results from Model 3 separately in section 4.3.

>Model 3: Infra_st = G (ϕ,Z_svt, Q_st−1)

Log (Y_svt) =  F(ψ,Log(Y_svt−1), X_svt, Infra_st−1)—Second stage is same as Model 2.

The empirical models allow for national trends that apply to vehicle models in all states and for temporal variations of a vehicle model irrespective of the state in which they are sold. To control for unobserved heterogeneities, we include (state) × (vehicle model) fixed effects, ∝_sv and (vehicle model) × (time) fixed effects, θ_vt. We accounted for potential heteroscedasticity by using (state) × (vehicle model) clustered robust standard errors on the remaining error term, ε_svt. While (state) × (vehicle model) fixed effects controls for time invariant preferences for a vehicle in a state, (time) × (vehicle model) fixed effects controls for state invariant trends in sales and countrywide production shocks for a vehicle model. We did not add federal incentives to the state incentive variables, as federal incentives apply to vehicles bought in all states. We controlled for state-level demographics that may explain a state's preference toward PEVs.

Finally, in addition to PEV specifications, we ran model specifications separately for PHEVs and BEVs as heterogeneity in purchase decisions are likely to be substantial between the two vehicle segments (Axsen et al 2015) and policies promoting PEVs overall may not affect PHEVs and BEVs equally (Vergis and Chen 2015). Additionally, households who prefer BEVs tend to have different income; education, recharge access, environmental concern, and technological awareness characteristics when compared to those who prefer PHEVs (Axsen et al 2016).

3.1. Model limitations

4.1. PEV adoption

The estimates for tax incentives and public EVSEs for the fixed and GMM models (i.e. PEV_model1 and PEV_model2 specifications) in figure 3 (table III in appendix 3 of SI for details) show that the traditional fixed effects model significantly overestimates the relationship between the key explanatory factors and PEV purchases. Controlling for the natural diffusion of a vehicle technology using lagged PEV purchases in the GMM model, a 1% increase in tax incentives relative to a vehicle's MSRP is associated with only a 1.15% increase in per capita PEV purchases versus the 2.9% increase in purchases estimated by the fixed effects model. Also accounting for the inherent endogeneity between public EVSEs and vehicle adoption using the variable's deeper lags, the GMM model shows that one additional EVSE per capita is associated with a 3% increase in per capita PEV purchases compared to the 6.2% estimated by the fixed effects model.

Zoom In Zoom Out Reset image size

4.2. Vehicle category and adoption

PEV_main estimates in the sub-plots of figure 4 (table IV in appendix 3 of the SI) shows that within tax incentives, rebates influence PEV adoption more than tax credits or sales tax waivers. Isolating the effects for PHEVs and BEVs separately, we find that both tax credits and rebates are significant for BEVs when compared to PHEVs (highlighted green in the first two sub-plots). BEV_main in sub-plots 1 and 2 show that a 1% increase relative to the vehicle's purchase price is associated with a 1.8% increase in BEV purchases with a tax-credit and 2.16% with a rebate. For instance, if the proposed increase in California's existing $2500 rebate to $10 000 (i.e. 25% increase for a $30 000 vehicle) materialized, it may have led to 50% higher adoption. Within BEVs, Nissan Leaf purchases are equally explained by both state tax credits and rebates, while Tesla purchases are equally explained by rebates and sales tax waivers. Comparing PHEV_main and BEV_main in subplot no. 4 of figure 4, we see that one additional EVSE per capita is associated with a 7.2% increase in BEV purchases yet only a 2.56% increase in PHEV purchases. Within BEVs, Nissan Leaf purchases are highly correlated to public EVSEs while Tesla Model S purchases are not influenced by the presence of Tesla-compatible public EVSEs. Within PHEVs, the purchases of Volt are strongly correlated with public EVSE stations while the Prius purchases are not.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

States with HOV lane access incentive for PEVs registered higher purchases when compared to states without the incentive. The correlations are relatively smaller but significant after we control for California (see table VII of SI). Presence of HOV lane access is associated with stronger purchases of Prius plug-in and BEVs in particular (subplot no. 5). Gasoline price levels in a state are strongly correlated with the purchases of the Prius plug-in when compared to other top selling PHEV and BEVs (subplot no. 6). BEV purchases are more correlated to median household incomes when compared to PHEVs. However, the associations become less clear when we estimate the impacts for the top-selling vehicle models in the two vehicle categories (subplot no. 7). Finally, environmental awareness, proxied by the league of conservation voters (LCV) score of house representatives from different states are significantly correlated with the purchases of PHEV vehicles such as Volt and Prius Plug-in (subplot no. 8).

4.3. Influence of early adopters

4.4. Sensitivity analysis

To render our findings more usable to policymakers (appendix 4 of SI), we discuss possible mechanisms causing the correlations that are novel, counter intuitive or conforming to existing studies. Overall, tax incentives of any type and public charging infrastructure are dominant factors in driving PEV adoption. Rebates are more effective than tax credits. This finding is in line with studies that have pointed out that incentives closer to the point of sale are more effective, likely due to hyperbolic discounting by consumers of rewards that arrive later (Hardman et al 2017, Yang et al 2016, IEA 2017). In countries such as Norway and Netherlands, where a point of sale VAT exemption for PEVs is combined with high VAT for conventional gasoline vehicles, PEV market shares are significantly higher (Hardman et al 2017, Aasness and Odeck 2015, Bjerkan et al 2016). Estimating tax incentives as a percentage of a vehicle's MSRP demonstrates a diminishing marginal return to every dollar incentive offered for higher MSRP vehicles (i.e. $300 incentive on a $30 000 vehicle has the same effect as a $700 incentive for a $70 000 vehicle). This would imply that incentives are more effective in increasing adoption when targeting less expensive PEVs.

Zoom In Zoom Out Reset image size

Tax credits and rebates may be less effective on PHEVs than BEVs because their incremental cost (over a traditional gasoline vehicle) is lower. States such as Washington and Oregon registered a higher number of PHEV purchases with nearly insignificant amount of sales tax incentive or no tax incentive. These findings do not conform to existing literature with the exception of a modeling study in California (DeShazo et al 2017), which estimates that customers are more likely to purchase PHEVs over BEVs in the absence of incentives or with the existing incentives. Among BEVs, equal correlation of tax credits and rebates on Nissan Leaf purchases may be because of high lease rates where Nissan captures the incentive and passes it on to the driver. Similarly, equal correlation of rebates and sales tax waivers for Tesla Model S purchases may indicate that Tesla buyers with higher incomes do not discount money received later. The lack of impact of tax credit on Tesla purchases may indicate the fact that high income earners may not be eligible to apply for a tax credit. The significant correlation of Tesla sales in states that offer rebates or sales tax waivers contrast findings from modeling and post purchase surveys (Tal and Nicholas 2016, Hardman and Tal 2016) that Tesla buyers would have bought their vehicle even in the absence of tax incentives.

The presence of public charging infrastructure has a strong influence on vehicle purchases decisions. The result conforms to findings from studies that analyzed adoption trends in other countries like Norway (Zhang et al 2016, Mersky et al 2016). Early investments in infrastructure, even if they would be underutilized for some time, is likely to incentivize early adopters and result in a multiplying effect on PEV adoption (figure 5). The correlation of infrastructure with vehicle purchases strengthens with the battery-only driving range of a PHEV, while weakening with increasing driving range of BEVs. Neither the sales of Tesla Model S nor the Prius (the two poles of battery range) are affected by the availability of charging infrastructure. This may be due to the following factors. First, unlike long-range BEV owners, short-range BEV owners may be more range anxious and consider public charging availability seriously when making purchase decisions. Second, PHEV drivers with a significant electric driving range may base their purchase decision more on total cost of ownership, which would include utilizing low-cost/free electric power from public EVSEs whenever possible.

Environmental awareness seems to drive PHEV adoption more than the availability of tax incentives, as embodied by the higher sales of PHEVs across environmentally friendly states with decreasing price differentials against gasoline vehicles. Higher gasoline prices and higher median household incomes significantly correlate with vehicle adoption, more consistently with PHEVs. Finally, HOV lane access correlates to higher PHEV and BEV adoption, with correlation surprisingly stronger for BEVs. If HOV lane access is the only reason to purchase an alternative vehicle, it is likely that a customer would choose a less expensive PHEV model.

The empirical results discussed in this article are important for policymakers who need to make informed decisions to maximize the adoption of emissions reducing transport technologies. In summary, our results indicate that to further increase PEV adoption, policy makers may want to incentivize public charging infrastructure along highways to cater to future vehicle models that are likely to have longer driving range, tailor tax incentives closer to the point of sale for BEVs at the lower end of the price spectrum and PHEVs with longer battery range, and better internalize the environmental cost of owning conventional gasoline vehicles. With the widespread introduction of PEVs, the data is only getting more robust. This should allow future researchers to confirm and refine the relationships described in this article as well as assesses the short and long-term emissions benefit derived from incentivizing electric vehicles.

The authors would like to thank Margo Melendez, Wendy Dafoe, Aaron Brooker, Eric Wood, and Marc Melaina of the National Renewable Energy Lab for their invaluable help throughout this project. Authors are also grateful to Kelly Sims Gallagher of Tufts University and David Greene of the University of Tennessee for their advice and the 2017 ETH Zurich SusTec Academy for their valuable inputs and suggestions. This work was authored in part by Alliance for Sustainable Energy, LLC, the manager and operator of NREL for the US. DOE under Contract No. DE-AC36-08GO28308. Funding was provided by US. DOE's Office of Energy Efficiency and Renewable Energy—Vehicle Technologies Office. The views expressed in the article do not necessarily represent the views of the DOE or the US Government. The US Government retains and the publisher, by accepting the article for publication, acknowledges that the US Government retains a nonexclusive, paid-up, irrevocable, worldwide license to publish or reproduce the published form of this work, or allow others to do so, for US Government purposes.