From lab-to-road: real-world fuel consumption and CO₂ emissions of plug-in hybrid electric vehicles

We collected data on real-world usage and FC of PHEVs from existing literature, research institutions, companies and online data bases. We collected data providing information on real-world FC, annual vehicle kilometres travelled, and UF. Our data covers China (CN), Germany (DE), Norway (NO), the United States (US), Canada (CA), and the Netherlands (NL) and includes data from private and company cars, i.e. vehicles owned by an organization that are assigned to an individual user and can also be used for private purposes. Note that our data does not cover cars owned by a company that are not allowed to be used for private purposes. Table 1 gives an overview of the sources, total sample sizes and PHEV models covered. The data is available as supplementary file (available online at stacks.iop.org/ERL/16/054078/mmedia).

In total we collected data from primary and secondary sources of over 100 000 PHEVs. Data from Spritmonitor.de and XiaoXiongYouHao.com was provided by the website operators, data from Voltstats.net and MyMPG was collected manually by the authors. As users record their PHEV FC on a voluntary basis on these websites, there is a risk of self-selection bias. Tietge et al (2017, 2019) have shown that Spritmonitor, XiaoXiongYouHao, and MyMPG closely follow the distribution of several key vehicle variables in the newly sold fleets in the respective markets. Furthermore, these four new empirical sources and combined with 13 existing studies in a joint meta-analysis ensuring the robustness of our findings and low bias in our sample. The data of company cars in a large German company was provided by the company (which wishes to remain anonymous). We obtained a sample of 95 PHEVs directly from the UC Davis. A large part of our sample is aggregate PHEV information taken from California's Advanced Clean Car Midterm Review 2017 Appendix G on electric vehicles. The sample size information is taken from chapter II of the report whereas the UF and annual driving distances are taken from table 14 as well as table 20 of the same report (CARB 2017).

Our sample is dominated by North American vehicles, but the sample sizes for individual countries are still sufficiently large for conclusions and general patterns. For Germany, for example, our sample accounts for 1% of the total stock of PHEVs at the time of data collection, while for Europe (EU, UK, EFTA, and Turkey) 1.5% of the total stock are covered (EAFO 2020). The major part of vehicles in our sample is private, yet, a substantial number of more than 10 000 PHEVs are company cars, allowing profound analyses for this user group as well. In total, our sample covers 66 models and 202 model variants (Plötz et al 2020).

To compare actual FC to test-cycle values and to analyse the effect of vehicle properties such as system power and test-cycle all-electric range, a detailed list of vehicle specifications were collected. We retrieved information on vehicle system power (i.e. the maximum combined power of electric and combustion engine), fuel type, FC (in charge-depleting mode, charge-sustaining (CS) mode, and combined), as well as all-electric range according to NEDC, WLTP, and Environmental Protection Agency (EPA) (the five cycle) test cycles from the ADAC Autokatalog (an online vehicle database, see ADAC 2020) for PHEV models in Europe and Fueleconomy.gov for PHEV models for the U.S. market (containing EPA values for combined FC and charge-depleting mode consumption). The XiaXiongYouHao.com database provided NEDC combined FC as well as NEDC all-electric range for PHEVs available in China. For those vehicles in the US that are identical to vehicles on the European market such as BMW i3 REX and the Chevrolet Volt (which has been sold as Opel Ampera in Europe), the EPA data was enriched by NEDC values from the European counterparts. For those vehicles were this is not the case, only the deviation from EPA test values could be calculated (i.e. they are missing in table 2 below).

In cases where the actual UF is missing and only the FC is given (for 10% of the sample and 203 of 244 aggregated observations), we estimate the real-world UF from the real-world FC and CS mode FC ${\text{F}}{{\text{C}}_{{\text{CS}}}}$ as ${\text{U}}{{\text{F}}^{{\text{real}}}} = 1 - {\text{F}}{{\text{C}}^{{\text{real}}}}/{\text{FC}}_{{\text{cs}}}^{{\text{real}}}$. Here, ${\text{FC}}_{{\text{cs}}}^{{\text{real}}}$ is approximated by taking NEDC values with 50% addition for real-world driving, i.e. ${\text{FC}}_{{\text{cs}}}^{{\text{real}}} = 1.5 {\text{FC}}_{{\text{cs}}}^{{\text{NEDC}}} = 1.5 {\text{F}}{{\text{C}}^{{\text{NEDC}}}}/\left( {1 - {\text{U}}{{\text{F}}^{{\text{NEDC}}}}} \right)$ with ${\text{U}}{{\text{F}}^{{\text{NEDC}}}} = {\text{AE}}{{\text{R}}^{{\text{NEDC}}}}/\left( {{\text{AE}}{{\text{R}}^{{\text{NEDC}}}} + 25\,{\text{km}}} \right)$ (AER = all-electric range). This method is slightly optimistic as a 50% deviation from NEDC is slightly above the fleet mean deviation for conventional combustion engine vehicles and hybrid vehicles (see Tietge et al 2019). It is optimistic as we increase the denominator in the second term of ${\text{U}}{{\text{F}}^{{\text{real}}}} = 1 - {\text{F}}{{\text{C}}^{{\text{real}}}}/{\text{FC}}_{{\text{cs}}}^{{\text{real}}}$, and thereby increase the calculated UF. In cases with high actual FC, this approach can lead to negative UF. We set the estimated UF to zero in these cases (11 of 244 aggregated observations covering 0.1% of the total vehicle sample). If values from the US EPA are available, we use EPA values for CS mode FC, i.e. ${\text{FC}}_{{\text{cs}}}^{{\text{real}}} = {\text{FC}}_{{\text{cs}}}^{{\text{EPA}}}$. Likewise, when the real UF is known, the actual FC is estimated by inversion ${\text{F}}{{\text{C}}^{{\text{real}}}} = {\text{FC}}_{{\text{cs}}}^{{\text{real}}}\left( {1 - {\text{U}}{{\text{F}}^{{\text{real}}}}} \right) = 1.5{\text{FC}}_{{\text{cs}}}^{{\text{NEDC}}}\left( {1 - {\text{U}}{{\text{F}}^{{\text{real}}}}} \right)$ for 18 of 244 aggregated observations covering 66% of the sample vehicles.

The distributions of real-world FCs are calculated as kernel density estimates (Wickham 2016). In the regression results, we use weighted least squares to adjust for differences in sub sample sizes. More specifically, the nonconstant variance of the individual sample means is corrected for by weighted each mean with the inverse of its standard error. In our case of mean values, the weights are thus proportional to the reciprocal of 1/√N; we thus use the square root of sample sizes as weights.

Figure 1 shows the distribution of real-world FC values in litres/100 km across the countries and user groups. We observe a broad distribution of real-world FC values, much broader than for conventional combustion engine vehicles (Tietge et al 2019). The mean FC varies between 2.1 and 7.5 l/100 km, depending on the country, user group and vehicle models. This corresponds to a range of mean tailpipe CO₂ emissions, i.e. without emissions from electricity and fuel production, from 49 to 174 g CO₂ km⁻¹ at 2.32 kg CO₂ l⁻¹ of gasoline. Table 2 below summarizes the mean real-world and test cycle FC as well as CO₂ emissions per km. As the samples in the countries cover PHEV models with different sample sizes, table 2 states unweighted and weighted means. The individual mean values by country and user group are affected by several factors such as the all-electric ranges of the models in the sample, charging and driving behaviour as well as the resulting UF. Their impact will be discussed in section 3.3.

Zoom In Zoom Out Reset image size

For the country-specific analysis, the most recent data is for Germany and China with data from 2019 to 2020. The large mean FC in China is noteworthy and mainly due to low UF (cf section 3.2). In Norway and the US, users achieve higher UF and thus lower total FC than in the other countries under consideration (cf section 3.2). Company car data is only available for Germany and the Netherlands with greater sample sizes for the Netherlands. Note, however, that the distributions in Germany and the Netherlands are similar for company cars with a peak around between 6 and 8 l/100 km, i.e. four times higher than test cycle FC, and a broad distribution.

The mean deviation from test cycle values differs between countries, but on average, real FC is two to four times higher than the test cycle for private cars and three to four times higher for company cars. As the majority of vehicles in the sample have been certified according to the NEDC, we focus mainly on the deviation from NEDC. For private vehicles, the mean relation is 300%–340% (the range indicates the mean with two standard errors) and 135%–235% for the sample size weighted mean. This relation corresponds to a deviation between real-world and test cycle FC of 200%–240% or 35%–135%. The sample-size weighted deviation is smaller as North American vehicles, mainly Chevrolet Volt, Toyota Prius and BMW i3 REX with small test cycle deviation, dominate the overall sample. Furthermore, these vehicle models have high all-electric ranges (see section 3.3). For company cars, with data from Germany and the Netherlands, the deviation is even higher. The mean relation for company cars in both countries is 305%–395% or 205%–295% deviation and the sample size weighted mean relation is 340%–410% or 240%–310% deviation.

For the US, the mean real-world FC can also be compared to EPA testing values. Figure 2 shows the distribution of the relation between real-world and EPA test cycle FC. A value of 100% corresponds to perfect match between real-world and test cycle FC. Most PHEV models and sub samples have clearly above EPA FC. The unweighted mean relation between actual and EPA FC for the US is 213% (corresponding to a deviation of 113%) and the sample size weighted relation is 179% (or 79% deviation).

Zoom In Zoom Out Reset image size

A key indicator of PHEV usage and its potential environmental benefit is the share of kilometres driven on electricity, the so-called UF. The UF is the ratio between the distance driven on electricity and the total distance driven by a vehicle. If direct UF values were not available, we calculated the UF from the real-world FC (cf section 2.2). Figure 3 shows the mean UF as a function of all-electric range for all PHEVs sub-samples. Also shown is the UF assumed in the NEDC values (dashed line). Almost all mean real-world UF are below the NEDC value. Long-ranged PHEVs in the sample tend to be closer to test cycle values. Comparing privately owned vehicles and company cars with a given range, we observe lower mean UF for company cars throughout the sample. Accordingly, the deviation to test cycle UF is even higher for company cars. The deviation between mean UF in the sub samples and test cycle values shows noteworthy differences also between the countries. It is largest for privately owned vehicles in China and for company cars in the Netherlands, but is closest to test cycle values for privately owned vehicles in Norway and the US. Furthermore, the UF in most countries shows a tendency to increase with all-electric range, as expected.

Zoom In Zoom Out Reset image size

Some country specific effects can partially be explained by country specific factors. In Norway, battery electric vehicles receive higher incentives than PHEVs and thus PHEVs are less likely to be bought to mainly benefit from the lower purchase price or taxation (especially for company cars), but to actually take advantage of the possibility to drive on electricity. The high fuel and low electricity prices in Norway are an additional incentive to maximise electric driving. Furthermore, a small additional effect could come from the fact that public charging is comparatively easy in Norway, as there is only one charging card that allows to charge at almost all public charging points across the country (Figenbaum and Kolbenstvedt 2016).

The mean UF in China show a large variation even for a given range and only a slight tendency to increase with range. Chinese authorities monitor the real-world performance of PHEVs, but there is no enforcement or regulation that effectively encourages car owners to increase electric driving or charging. Also, there are no requirements on how frequently PHEV users should charge their vehicles in China. Furthermore, the lower availability of garages and private parking spots in China makes PHEV users without regular night charging option more likely (Li et al 2021). Additionally, the restrictions on driving and purchase of conventional fuel vehicles in major cities such as Beijing, Shanghai and Hangzhou make PHEVs highly attractive irrespective of the actual usage.

The Netherlands had high incentives of PHEVs as company cars from 2012 to 2016 leading to strong sales increase in company car PHEVs. However, no additional incentives for charging these cars were enacted. Instead, many PHEV company car users in the Netherlands have fuel cards that allow refuelling conventional fuel for free but they have to pay privately for charging the PHEV at home (van Gijlswijk et al 2018), which would require the installation of a private charging spot. These financial disincentives are likely the main reason why many company car PHEV users do not frequently charge their PHEVs resulting in particularly low UF.

For the US, only private vehicles are in the sample and many of the observations are from early adopters who are more likely to purchase a PHEV only if they have an option to recharge the vehicle regularly (Chakraborty et al 2020). In addition, 21 of the 23 PHEV model variants in the US sample are Chevrolet Volt, BMW i3 and Toyota Prius PHEV which are probably preferred options for buyers with above mean environmental concern and are frequently charged (Tal et al 2018, 2020, Chakraborty et al 2020). Beyond that, Chevrolet Volt and BMW i3 are range-extended PHEVs having an NEDC all-electric range of at least 80 km to up to 330 km. Accordingly, UF in the US are higher.

We also analyse the deviation between real-world UFs and EPA UFs for the US sample in more detail. Figure 4 compares the real-world UFs to EPA values as well as the real-world UFs to NEDC UFs. Short-ranged PHEVs have lower EPA values and lower real-world UF than long-ranged PHEVs. Figure 4 demonstrates that EPA values generally match the real-world UF quite well for large span of all-electric ranges. The sample size weighted linear regression in figure 4 (dashed line with 95% confidence interval) indicates a possibly overestimation for low EPA UFs and possible underestimation for large EPA UFs. In contrast, the NEDC UFs in the right panel of figure 4 overestimate the real-world UF for almost all all-electric ranges. Figure 4 demonstrates that test-cycle UFs can be close to real-world UFs if the test-cycle values are carefully designed. There are several potential reasons why the EPA UFs are more realistic. First, the more careful electricity consumption estimate in five cycles including the use of auxiliaries and more demanding driving cycles. Second, different assumptions about the frequency of charging. Third, more realistic calculation of the UF from the combination of driving data, the all-electric test-cycle range, and the assumptions about charging (all three are combined via ${\text{U}}{{\text{F}}^{{\text{NEDC}}}} = {\text{AE}}{{\text{R}}^{{\text{NEDC}}}}/\left( {{\text{AE}}{{\text{R}}^{{\text{NEDC}}}} + 25\,{\text{km}}} \right)$ for the NEDC case). The first potential reason is very likely to play a role (see the explanation of the low NEDC UFs in the section 4), but a detailed discussion of the other factors is beyond this meta-analysis of real-world FC in the present paper.

Zoom In Zoom Out Reset image size

Lastly, for the case of Germany, the data is quite recent mainly with usage data from 2019 and early 2020. Home charging should not be a problem in Germany for the majority of PHEV users as about three quarters of passenger cars in Germany are parked in private garages or car ports overnight (MiD 2018) and the share can be expected even higher for PHEV owners, because of the higher purchase price of PHEVs and higher household income of PHEV owners (Plötz et al 2014, Frenzel et al 2015). Company car users in Germany, on the other hand, have similar financial disincentives as company car users in the Netherlands. They receive a tax benefit it they use a PHEV but this is not conditional on electric driving and many can be expected to own fuel cards and not to pay for conventional fuel, as in the Netherlands.

In summary, the real-world UF is typically only half of the NEDC test cycle values for private vehicles and even lower for company cars. Yet, the mean UF for some privately used PHEV models is close to 80% of the test cycle UF and the mean UF for users in Norway and the US are closer to test cycle UF compared to other countries.

The all-electric range and other vehicle properties such as system power can be expected to affect FC and thus tailpipe CO₂ emissions. We use regression analysis of these factors on FC to quantify these effects. We use the NEDC value for the all-electric range as it is readily available for most PHEV models. System power, i.e. combustion engine power plus electric motor power, is included to account for different vehicle size or type and engine capacity. Strictly speaking, the system power is the maximal power available for propulsion. For most PHEV models, this is the sum of combustion engine and electric motor power. In range extended electric vehicles, however, the system power is smaller than the sum of engine and electric motor power because the combustion engine is not directly used for propulsion but for battery charging.

Since FC is strictly non-negative, we use an exponential function for the effect of vehicle models' all-electric range and power and control for user group and country specific effects with the following model

$$\begin{align*}{\text{F}}{{\text{C}}^{{\text{real}}}} = & \exp \left( {\beta _0} + {\beta _1}{\text{Power}} + {\beta _2}{\text{Range}} + {\beta _3}{\text{Usergroup}} \right. \nonumber\\ &\left. \quad\quad + {\beta _4}{\text{Country}}\right) + \varepsilon .\end{align*}$$

Here, system power (${\text{Power}}$) is in units of 100 kW and all-electric range (${\text{Range}}$) in 10 km. The chosen dependence is physically meaningful: for ${\text{Range}} \to 0$, the FC approaches a finite value (i.e. the FC in CS mode) and goes to zero for ${\text{Range}} \to \infty$. Likewise, the FC approaches zero for ${\text{Power}} \to 0$ and grows with increasing power (for positive β₁). The linear regression is performed after taking logarithms by weighted least squares.

Table 3 summarizes the results of a sample-size weighted and not sample-size weighted regression and shows the impact of all-electric range and system power on FC. The model itself and all coefficients (except for Norway as country dummy) are significant (p < 0.05) and the coefficients have the expected signs (β₁ > 0 and β₂ < 0). The regression results show relatively high goodness of fit (adjusted R² > 0.8). Please note that the non-sample size weighted regression is mainly shown to demonstrate the robustness of the results as the sample size weighted model is more appropriate here. Furthermore, the regression models (2)–(4) in table 3 indicate the high explanatory power of all-electric range and system power for actual FC.

Controlling for user group and country specific effects, we find that a 100 kW increase in system power increases FC and thus CO₂ emissions by 25%–36% (including 95% confidence intervals in the full sample-size weighted regression model (1) and 23%–33% in the non-sample-size weighted regression for comparison). The system power in the PHEV models in our sample cover a range of 90–674 kW with a mean of 225 kW, a 20% increase of system power is by 45 kW and would lead to an increase of FC and CO₂ emissions by 11%–16%, keeping all other factors constant. Likewise, a 10 km increase of NEDC all-electric range decreases the FC by 11% and 14%, respectively (range from sample size weighted regression including 95% confidence intervals) with all other parameters fixed. This implies that FC and thus direct CO₂ emissions are halved with every 55 km of all-electric range (50–63 km range from sample size weighted regression including 95% confidence intervals). While a significant effect of the all-electric range on FC is expected, the effect of the models' system power is comparatively strong.

For categorical variables, the reference categories are 'private' for user group and 'Germany' for country. The regression confirms the general observations of actual FC and UF in figures 1 and 2. Controlling for other factors, the mean FC of company cars is 10%–50% higher than for private vehicles (95% confidence intervals) and Chinese PHEVs show 35%–55% higher FC than PHEVs in Germany. The change in FC is not significantly different from zero for Norway. PHEVs in the US have 1%–23% lower FC compared to Germany. The Netherlands can only be included in unweighted regression model as the sample size is not available on the individual model level.