What determines passenger car lifetimes? Insights from individual vehicle records

Determining the expected lifetime of passenger cars is vital, both when assessing the impacts of a single vehicle (for example through life cycle assessment) and when considering the dynamics of a stock of vehicles and the... [ view full abstract ]

Determining the expected lifetime of passenger cars is vital, both when assessing the impacts of a single vehicle (for example through life cycle assessment) and when considering the dynamics of a stock of vehicles and the impacts of the fleet. In the second case the turnover of stock and the uptake of new technologies is an important factor when considering long term environmental impacts, and this is partly driven by the lifetime of vehicles.

Traditionally car lifetime is measured in terms of years of use. In Great Britain published statistics on total number of cars registered can be used to estimate the distribution and median lifetime of passenger cars. The median lifetime has grown from 11.6 years in 2005 to 13.7 years in 2014. This top-down view of car lifetime cannot capture all important factors and the variation within the fleet however.

This study uses a detailed dataset of anonymised MOT (Ministry of Transport) tests and results to explore those features that influence vehicle lifetime. The MOT test is an annual test of safety, roadworthiness and emissions for most vehicles over three years of age in Great Britain. Data from these tests has been collected and made publicly available from 2007 to 2013 at the individual vehicle level, with information on mileage, age, cylinder capacity, fuel type, car make / model and other related parameters supplied alongside the results of the test procedure. By examining whether a vehicle survives between subsequent years the parameters that determine vehicle survival functions can be explored. Given the number of samples in the dataset (approximately thirty-five million per year) techniques from the fields of big data analysis and machine learning, specifically survival analysis, are utilised in analysing the data and in exploring those features that determine the likely survival characteristics of a vehicle.

The results from this approach can improve the modelling of vehicle lifetimes and illuminate the differences that are likely to exist between different vehicles (at both the individual and fleet level). Additionally, when combined with a broader understanding of the topic of product lifetimes, the findings can offer lessons for the lifetime extension of cars.