Part 1: Setting the scene

Across Europe, there has been significant growth in the uptake of electric vehicles (EVs) in recent years. For example, 24% of all new cars purchased in the UK in 2023 were hybrids or fully electric.

Source: UK Department for Transport & Driver and Vehicle Licensing Agency

However, this growth in EVs has not been equally spread. At present, battery electric vehicle (BEV) ownership is unevenly distributed both within and between countries in Europe. Scandinavian countries lead the way with very high rates of BEV take-up, with BEVs making up 20% of Norway’s entire car stock in 2021. In France, Germany and the UK, this figure was around 1.5 - 2% - comparatively low compared to other high-income European countries. One likely explanation of this difference is national variations in BEV policies. For example, Norway introduced incentives for BEV uptake as far back as the 1990s.

Source: Eurostat

However, policy incentives alone cannot account for all differences in BEV uptake. Policies are typically applied nationwide, yet within-country variations similarly show a very unequal distribution. For instance, in Britain the top local authority for privately owned BEVs (Westminster, at 5.9% of all private cars) has over 15 times the rate of EVs than the lowest (less than 0.4% in Kingston upon Hull). Attempted explanations of these intra-country differences often  focus on regional variations in income and population density. But is this a sufficient explanation? Or could more mysterious behavioural forces be at play?

Source: UK Department for Transport & Driver and Vehicle Licensing Agency

Part 2: Diagnosing regional EV variation

So how would one harness data science to dissect the factors behind these regional uptake differences? Using supervised machine learning, we started by investigating how much of the regional differences in privately owned BEVs – ignoring plug-in hybrids and company EVs, since different factors may drive the uptake of these vehicles - can be explained by the traditional factors of income and population density. We trained a Random Forest model to predict BEV uptake across about 300 British local authorities based on income and population density as well as a suite of extra regional variables, including average age, taxi numbers, housing types and multiple-car ownership. Ultimately, the variables of this model were able to account for 54% of the variation in uptake (known as the ‘R^2’ statistic), leading us to question what could explain the remaining 46%...

One explanation might be differing consumer attitudes to the environment, so we then investigated this possibility using a linear regression model. Unsurprisingly, we found that a higher proportion of people with ‘green’ habits (based on Natural England’s ‘MENE’ data) is associated with a higher share of BEVs. A 10 percentage point increase in the number of people reporting pro-environmental behaviour is associated with an increase of 1 BEV per 1,000 cars. Though this effect seems small, it means that if a local authority with the median proportion of BEVs (falling exactly in the middle of the distribution) saw a 10 percentage point increase in environment-friendly behaviour, it would move up to the 61^st percentile (meaning it would have more BEVs than 61% of all local authorities).

However, even after accounting for these green habits, a lot of the variation in uptake in our modelling remains unexplained. The graph below highlights the local authorities that are particularly poorly explained. Those above the dashed red line are under-predicted by the model (the actual number of BEVs is higher than predicted by the model) while those below the line are over-predicted.

Source: Frontier Economics modelling 

Two local authorities (Camden and Westminster) with much greater BEV ownership than elsewhere are substantially under-predicted by the model. Such outliers are to be expected: there are so few examples of areas with uptake this high that a statistical model is unlikely to be able to adequately capture what makes them special. However, this is also a tantalising sign that in such areas high levels of BEV ownership might themselves be spurring more people to purchase a battery electric car. Indeed, feedback loops and network effects may be playing a role in creating ownership clusters.

Part 3: The missing piece of the puzzle? 

The figure below illustrates some of the channels through which feedback loops might occur:

These types of feedback loops are commonly observed during the spread of new technologies. As they may not become significant until uptake reaches a certain threshold, statistical models trained on data during the early stages of a rollout may underestimate the speed with which the market can change. And some variables, such as seeing an EV on a neighbour’s driveway or hearing friends discuss their new Tesla, can be very difficult to capture in standard modelling.

To help assess the impact of these sorts of effects we can use agent-based modelling, a technique that explicitly models how individuals and organisations interact. An agent-based model (ABM) would therefore be well equipped to simulate how such micro interactions between agents result in the observed quirks and variations in EV uptake at a macro level.

Frontier’s recent application of ABMs has yielded new insights into key climate mitigation policies. We have used the technique to investigate vehicle-to-grid charging, where owners of electric cars return the power in their battery to the grid. We have also examined the take-up of heat pumps for the National Infrastructure Commission. One strand of the work, enabled by ABMs, was an exploration of the distributional impacts on homeowners compared to renters: a landlord deciding whether or not to install a heat pump in homes they rent out will be more concerned by the upfront bill, and less attentive to the long-term cost saving, than an owner-occupier.

This analysis of the electric vehicle market in the UK suggests that ABMs could help generate a deeper understanding of the driving forces behind EV uptake and thus help shape policy to encourage more people to make the switch.