The light-duty vehicle fleet appears to be at the beginning of a fundamental transformation. For over a century, light-duty vehicles have been powered directly by fossil fuels-gasoline and diesel. Yet in the past decade, battery electric vehicles (EVs) have become a viable alternative, with EVs making up 15 percent of global new vehicle sales in 2022:III. EVs have been heavily promoted by policymakers around the world, with both vehicle subsidies and subsidies for charging stations. With this policy tailwind, many automakers have committed to fully phasing out fossil fuel-powered internal combustion engine vehicles by 2035 and have invested billions of dollars toward EV development and production facilities.

This study examines how EV attributes, prices, and quantities sold have changed in recent years, with a focus on comparisons between EVs and the most similar conventional (internal combustion engine) vehicles. We also explore heterogeneity across geography, demographics, vehicle classes, and price ranges to underscore that the United States cannot be seen as a monolithic vehicle market with respect to EVs. We leverage data on all new light-duty vehicles sold in the United States from 2014 to 2020, where an observation in the data is an individual vehicle in a zip code. Using these data, we perform a matching analysis and explore a set of descriptive results to glean new insights into how EV sales stack up against the sales of similar conventional vehicles.

Our matching analysis compares EVs with similar conventional vehicles and shows that EVs are becoming increasingly competitive where they are competing, but EV sales shares are still well below 50 percent in nearly all vehicle and price segments. Many market segments (e.g., vehicle classes) remain deeply untapped, with minimal market penetration of EVs so far. In contrast, EVs are overrepresented in the luxury market segments. Interestingly, while the attributes of EVs have been dramatically improving, the average price of the vehiclesand the price of each vehicle relative to the closest competitors-has remained relatively constant. These findings suggest that there is a taste penalty for EVs-the fraction of vehicle buyers that prefer an EV over a similar gasoline vehicle is small-and/or that EV model availability is limited. Our matching analysis indicates that when EVs are available, they tend to do well relative to similar conventional vehicles.

This paper relates to several recent literatures on EVs. One recent group of papers studies how EV demand responds to public charging infrastructure (Li et al. 2017;Li 2019;Ou et al. 2020;Sinyashin 2021;Springel 2021) or home charging availability (Davis 2022). Another object of study is the effect of financial incentives on EV adoption (Muehlegger and Rapson 2022;Armitage and Pinter 2022;Xing, Leard, and Li 2021;Remmy 2022;Barwick, Kwon, and Li 2022). Demand for EVs may also be determined by peer effects (Tebbe 2023). Finally, several papers have studied how tastes for EVs are heterogeneous across geography and demographics (Linn 2022;Archsmith, Muehlegger, and Rapson 2022) and how they change over time (Forsythe et al. 2022).

Our primary dataset was obtained from Experian and is sourced from automobile dealers and state agencies. The data consist of all new vehicles registered in the United States from 2014 to 2020. The data are at the level of the vehicle identification number (VIN), which is a 17-digit alphanumeric code that uniquely identifies an individual vehicle. The data also contain the zip code of each vehicle registration. The first ten digits of the VIN contain information on the make, model, model year, and trim of the vehicle. We decode each VIN in the dataset, which also provides additional vehicle attributes, such as the fuel type, wheelbase, horsepower, and manufacturer suggested retail price (MSRP) for each vehicle.

In a separate dataset from Experian, we also observe all vehicles in the entire vehicle fleet, along with make, model, model year, trim, year of registration, county, and a set of demographic variables. We use the model year to focus on new vehicles, and we leverage these data for an analysis of how demographics correlate with EV adoption. We further bring in data on fuel economy from the US Environmental Protection Agency's fueleconomy.gov website. Finally, we merge in data from the US Department of Energy on charging station availability and census data on population to develop a measure of charging station density. Online Appendix 1 presents summary statistics for both datasets.

In the primary dataset, there are over 111 million observations. For a small fraction of the sample, we are missing key vehicle attributes, such as MSRP or wheelbase. Thus, our final sample contains approximately 106 million observations (the separate sample that contains the demographics has over 58 million observations). In the final full sample, we observe 913,619 dedicated battery EVs and 531,723 plug-in hybrid electric vehicles (PHEVs).foot_0

We begin our analysis by exploring key EV trends for the period 2014-2020 (Figure 1). In the top panel, we observe the share of light-duty vehicle sales that are EVs. There is a clear trend upward in the EV market share, with the market share in 2020 reaching just over 1.5 percent. The price (i.e., MSRP less federal subsidies) of EVs has been steadily increasing if we simply take the unweighted average across vehicle offerings. This reflects greater higher-priced luxury offerings in 2019 and 2020 and the expiration of incentives for several major EV manufacturers. However, the sales-weighted average price for EVs has stayed relatively constant since 2014 at 0.5 1 1.5 2 10 20 30 40 50 60 2014 2016 2018 2020 EV market share (percent) Price (thousands) Sales-weighted price Unweighted price Share of sales Price and (inside) shares 1 2 3 4 2014 2016 2018 2020 Value relative to 2014 Fast chargers/1,000 sq. mi Offerings Sales-weighted range Unweighted range The bottom panel plots the trends in EV offerings, EV range ( sales weighted and unweighted), and Level 2 and direct current fast charger density. All four of the time series are normalized to be 1 in 2014, and the panel plots the growth relative to 2014. The bottom line (green) presents the number of EV offerings on the market, defined as unique make-model-range combinations. We observe that they have increased by less than 1.5 times between 2014 and 2020, which is not nearly as dramatic as the relative increase in range or the number of charging stations. In fact, we see automakers add and drop EV offerings even if there is a general trend upward. The average EV range ( sales weighted and unweighted) increased substantially over the time period, by 2 to 2.5 times; the average density of fast chargers increased even faster, reaching nearly 4 times 2014 levels by 2020.

There is remarkable heterogeneity in EV sales, across geographies (e.g., within and between states, and in urban versus rural areas), across demographics, and across vehicle classes or segments. Figure 2 shows the county-level share of EVs in total new vehicle sales across the United States in the last year of our data, 2020. One of the most striking findings is that in most of the United States, the share of EVs is zero or near-zero. Only a few states have many counties with EV market shares above 5 percent. California is a prominent example, with the highest market shares in the counties in the Bay Area.

One driver of this geographic dispersion is that many of the states with higher market shares of EVs have " zero-emission vehicle" (ZEV) policies that mandate that a certain percentage of the vehicles that each automaker sells are ZEVs (EVs count as ZEVs) or they must buy credits from other automakers to meet their target. Thus, automakers will provide more EVs to dealers in these states and in some cases may offer EVs for lower prices. Outside of these states, EVs are not as commonly offered to dealers.

But there could be other drivers as well. Most of the counties that have near-zero market shares for EVs are in rural areas. This is true even in many of the ZEV states, such as Washington and Oregon. Demand for EVs may be low in these rural areas, and as a result auto manufacturers may not provide EVs to the dealerships in these areas. Demand may also be low due to limited charging stations in some of these rural countries, which would make owning an EV more difficult. There can also be supply-side explanations. Shipping EVs to rural dealerships may incur higher costs.

We can further explore the urban/rural heterogeneity-along with additional aspects of demographic heterogeneity-with an ordinary least squares regression of adoption of an EV (or PHEV) using the second Experian dataset that includes demographics. This descriptive regression is presented in Table 1. A time trend is included as well to capture the possibility of demand for EVs and PHEVs improving over time due to unobserved factors. The results suggest that households are more likely to adopt an EV or PHEV if they reside in urban areas, have higher incomes, or are more highly educated. The time trend is significant and positive. The magnitudes of the coefficients may make the changes appear small, but recall that the mean sales percentage of EVs and PHEVs is only 2.3 percent (higher toward the end of the sample), so an increase of one-half of a percentage point, such as we see with the coefficient on urban, is a notable increase.

We also examine heterogeneity by the price of the vehicles sold. For this, we go back to our primary dataset. Figure 3 shows the share of total sales by price range bracket that are EVs or PHEVs. The numbers next to each bar show the number of vehicles sold in that price category. A clear insight emerges: the market share of EVs and PHEVs is quite high in several price brackets at the high end, but the number of vehicles sold in these high price brackets is relatively small. Moreover, this pattern shows only a slight shift toward higher market shares in the lower price brackets by 2020. These results immediately indicate that EVs can make up a large market share in the US new car market but that there is a great deal of untapped product space for EVs in the lower price brackets. These untapped markets make up a much larger fraction of the total vehicle market than the luxury segments in which EVs thrive. It may have been much harder for EVs to penetrate the lower-price markets due to the high cost of batteries, but these markets hold substantial promise as the cost of batteries declines. As our analysis is only descriptive, we cannot conclude definitively if the untapped EV markets in the lower-price segments are caused by limited demand (e.g., a "taste penalty" relative to similar gasoline vehicles) or limited supply (e.g., supply chain constraints); decomposing this is a promising avenue to explore in future research.

To shed further light on the potential untapped markets for EVs, we next explore heterogeneity by vehicle class. Figure 4 presents the share of total sales in each of the major vehicle classes over time. The most striking finding is that in the hatchback category, sales of EVs and PHEVs are close to 15 percent of the market in some years. Hatchbacks are a small market segment with a 2,484 3,514 3,052 1,238 1,589 1,656 531 321 269 169 73 67 59 66 37 46 2016 2,063 4,455 2,248 1,573 1,466 1,540 1,235 493 118 23 182 106 43 58 36 9 2018 1,355 3,241 1,276 1,395 Note: Dependent variable is a dummy for EV or PHEV.

relatively large number of EV offerings, including the Chevrolet Bolt and the Nissan Leaf. The market for sedans is much larger, and there we see the Tesla Model S capture 3-5 percent of the market after 2018. Yet in nearly all other vehicle classes, the market share of EVs and PHEVs is extremely small. This again points to the existence of untapped markets.

We explore the idea of untapped markets further by comparing the sales of EVs to similar conventional vehicles. For this analysis, we use a nearest-neighbor matching approach. We estimate a linear probability model where the dependent variable is an EV dummy. The included attributes are price (net of incentives), vehicle class (i.e., body type, including sport-utility vehicle, coupe, hatchback, and so forth), drive type ( all-wheel drive or not), wheelbase, number of doors, and the log of the horsepower/weight ratio. We then obtain the propensity score for each conventional vehicle, indicating its similarity to EVs. We estimate the model separately for each model year and for prices above and below the median. For each EV model, we select a set of the three most similar conventional vehicles based on the propensity score. We find very similar results using an alternative matching approach (see online Appendix 2).

Figure 5 shows the ratio of sales of EVs to those in the comparison group of conventional vehicles most similar to the EVs. Not surprisingly, for most model years, the ratio is much less than one, indicating that EVs are sold less than comparable vehicles. However, beginning EV share PHEV share Coupe Hatchback Sedan Sport-utility vehicle Minivan Ratio of EVs to comparison group in 2018, the ratio exceeds 0.5, suggesting a market share larger than one-third when zooming in on higher-similarity EV and conventional vehicles. In 2020, the ratio is greater than one, so the EV market share rises above 50 percentvastly higher than the actual fleet-wide market share of less than 2 percent in those years. The main takeaway from this figure is that when we focus on only the conventional vehicles that are comparable to EVs, EVs are becoming increasingly competitive. The low overall market share stems from the (near-)absence of EV offerings in many segments of the vehicle market, suggesting that constrained offerings play an important role in explaining the untapped markets.

The findings of this analysis show that EV sales have been on the rise from 2014 to 2020. The attributes of EVs and the availability of fast charging infrastructure have been steadily improving even as the sales-weighted price of EVs has remained relatively flat, all contributing to greater sales of EVs. However, EVs have insignificant market share in vast swaths of the United States, likely due in part to ZEV standards and automaker decisions about where to sell their EVs. Similarly, EVs have very low market share across many price brackets and vehicle segments. Yet there are regions and market segments where EVs are extremely competitive, and increasingly so over time. Perhaps surprisingly, when compared to similar (matched) gasoline vehicles, EVs have seen relative sales shares exceeding 30 percent in recent years.

These results suggest that constrained supply of EVs is an important determinant of the near-absence of EV sales in large parts of the vehicle market, and more generally plays a key role in explaining the heterogeneity in EV sales, likely along with heterogeneity in the taste for EVs. This exploratory analysis sets the stage for formal demand estimation and structural modeling to disentangle the relative contribution of demand-and supply-side factors in explaining the increase in EV sales over the past decade, and to disentangle trends in demand caused by enhanced EV attributes versus a reduction in the "taste penalty" for EVs. A richer understanding of EV demand is important for automakers and policymakers.