Evaluation of real-world fuel

consumption of light-duty

vehicles in China: A 2021 update

Ruoxi Wu, Liuhanzi Yang, Zhinan Chen, Zifei Yang

BACKGROUND

The gap between real-world carbon dioxide (CO

2

) emissions/fuel consumption of

light-duty vehicles (LDV), on one hand, and their laboratory type approval values on

the other, is increasingly apparent around the world

of consumer-reported fuel consumption data in order to evaluate real-world fuel

consumption of China’s LDV fleet from model year (MY) 2007 to MY 2021.

DATA SOURCE AND METHODOLOGY

Several unocial platforms collect real-world fuel consumption information in China,

including Sina auto, Sohu auto, and XiaoXiongYouHao (XXYH). Sina auto and Sohu

auto are two mainstream websites that collect and summarize consumer-reported fuel

1 Uwe Tietge, Sonsoles Díaz, Zifei Yang, and Peter Mock. From Laboratory to Road International: A Comparison

of Ocial and Real-World Fuel Consumption and CO

2

Values for Passenger Cars in Europe, the United States,

Acknowledgments: The authors thank all internal and external reviewers of this report for their guidance and

constructive comments, with special thanks to Zhang Xiong (XiaoXiongYouHao) and Ma Dong (Vehicle Emission

Control Center, Ministry of Ecology and Environment). Their reviews do not imply any endorsement of the content

of this report. Any errors are the authors’ own.

Funding for this work was generously provided by the Rockefeller Brothers Fund

and the Federal Ministry for the Environment, Nature Conservation and Nuclear

Safety. This publication is part of NDC Transport Initiative for Asia (NDC-TIA).

NDC-TIA is part of the International Climate Initiative (IKI). The Federal Ministry

users to track and compare their fuel consumption.

We use XiaoXiongYouHao information to summarize consumer-based fuel

consumption, in part because the proprietor of XiaoXiongYouHao graciously provided

the data set used in this study. The mobile application of XiaoXiongYouHao was

launched in 2010. By the end of 2020, it had collected data on more than 2.4 million

individual vehicles. The data set includes information on more than 34,000 vehicle

model variants with model years ranging from 1986 to 2021.

incentive to enter fake data.

Data processing and statistical analysis of this study follow the standardized methods

described in prior ICCT publications studies.

4

The data proprietor removed outliers using reasonable ranges based on statistical

analysis. For individual vehicles with more than five reported fuel consumption records,

the average fuel consumption is considered normal if it lies within 2.5 standard

deviations left of the mean and 4.5 standard deviations right of the mean of all data

from the same vehicle variant (the same vehicle type and model year). It is normal for

vehicles to have higher fuel consumption in real driving. As a result, the distribution

For each vehicle model variant, the dataset included information on the model year,

engine type (naturally aspirated or turbocharged), transmission type, vehicle segment,

average on-road fuel consumption, ocial fuel consumption rating, and number of

vehicles in the sample. After removing data for motorcycles, trucks, electric vehicles,

and missing values, and removing model years 2002 to 2006 as they included fewer

than 5000 samples per year, about 2,088,835 individual vehicles of more than 21,000

vehicle model variants remained, with model years ranging from 2007 to 2021. In this

report, MY 2021 vehicles refers to vehicles with 2021 model variants sold in or before

2020, given that data records are collected by the end of 2020. Table 1 details the

number of samples by model year. Vehicles from MY 2009 to MY 2020 constitute the

3 Some vehicles of model year 2021 were sold in or before 2020. The sample size of vehicles of model year 2021

is larger than 12,000.

4 Uwe Tietge, Sonsoles Díaz, Zifei Yang, and Peter Mock. From Laboratory to Road International: A Comparison

of Ocial and Real-World Fuel Consumption and CO

2

Values for Passenger Cars in Europe, the United States,

China, and Japan. (ICCT: Washington, DC, 2017). https://theicct.org/publications/laboratory-road-intl.

Zifei Yang, and Liuhanzi Yang. Evaluation of Real-World Fuel Consumption of Light-Duty Vehicles in China.

(ICCT: Washington, DC, 2018). https://theicct.org/publications/real_world_fuel_consumption_ldv_China.

dataset with the entire light-duty vehicle fleet in market

5

, in order to show data

representativeness. The market dataset includes sales information for the 2009–2010

year and the 2012–2019 period. It should be noted that “year” in the market dataset

means sales year, whereas “year” in the XiaoXiongYouHao dataset refers to the model

year of the vehicle.

Figure 1 presents the share of vehicles by transmission type. Continuous variable

transmissions, dual clutch transmissions, and automated manual transmissions are all

categorized as automatic transmissions. We compared the vehicle fleet composition

the same year. One possible reason for this is that the year for market data is sales year,

while the year for XiaoXiongYouHao data is model year. For a given sales year, vehicles

for sale will include those of the current MY as well as previous MYs. As a result, the

change of fleet composition by sales year lags behind the change by model year.

5 Market data was purchased from CATARC.

Because vehicles with automatic transmissions tend to have a larger type-approval

versus real-world gap than manual transmission vehicles, we further dive into

automatic transmission vehicle data and analyze the lab-to-road dierence by engine

type. Figure 2 plots the trend of automatic transmission vehicle fleet composition

by engine type. The share of vehicles with turbocharged engines continues to rise

dramatically. The engine type distribution of XiaoXiongYouHao data agrees very

We further explore the influence of engine type on divergence between

XiaoXiongYouHao and ocial data on CO

higher divergence than vehicles with naturally aspirated engines, especially in the past

ten years. Divergence for turbocharged vehicles reached 38% in MY 2020, while that

of vehicles with naturally aspirated engines reached 28% in the same model year. One

possible explanation for the large gap for turbocharged vehicles is that consumers

may use the higher torque to accelerate faster in the real world, increasing real-world

fuel consumption compared with that of the lower load type-approval driving cycle.

Therefore, the fuel benefits from turbocharged engines may not be fully realized in

real-world driving.

Divergence of turbocharged vehicles of MY 2007 has been removed from the plot,

since the sample size is merely 67, which is too small to be representative.

Despite the change in composition of the vehicle fleet, the divergence between

consumer-reported and ocial fuel consumption increased across all segments (see

Figure 7). Lower medium cars and SUVs exhibited a divergence close to the average

for all segments. The medium, upper medium, and large segments tended to show a

higher divergence than the other segments, but the market share of these segments

has been relatively consistent over time and consequently these vehicles do not

contribute to the widening divergence for the entire fleet over time.

Results for mini cars are only kept for MY 2008–2015, a period in which the sample size

exceeded 500 for each MY. Some results of small vehicles (MY 2021), upper medium/

large vehicles (MY 2007–2008, 2021), and MPV (MY 2020–2021) are also removed due

CONCLUSIONS AND POLICY RECOMMENDATIONS

The findings from the updated analysis of XiaoXiongYouHao data illustrate that the gap

between real-world and ocial fuel consumption in China has widened from 14% for

MY 2007 to 37% for MY 2021. The accelerated expansion of the gap clearly indicates

the need for regulatory changes to reverse the trend and close the gap.

Policy recommendations are summarized as follows:

emission data for research purposes. We suggest that data

2

7

. Measuring real-world fuel consumption is a key recommendation since

2