Van Fuel Efficiency Measurement - A Successful Application of Data Envelopment Analysis

Purpose
Road fleets’ profitability has been long correlated to fuel efficiency (McKinnon, 1993). In effect, fuel
expenditures are one of road transport operations’ biggest budgets (FBP, 2005) as well as an area
where improvements are generally possible (Wilson, 1987). In order to improve vehicles’ fuel
efficiency, fleet managers need methods which can accurately measure vehicles’ fuel performance.
Regardless whether fuel information is obtained from fuel cards or vehicles’ electronic solutions such
as CANbus, fuel efficiency is generally measured in miles per gallons (mpg). Yet, mpg does not
include all the factors necessary to its interpretation such as vehicle weight or age. Furthermore, other
aspects of fuel efficiency, such as fuel costs, are not directly reflected by mpg but instead by other
measures such as pence per mile (ppm). These limitations can potentially lead to situations where a
vehicle can be mpg efficient but ppm inefficient (and vice versa) making it hard for fleet manager to
understand how efficient their vehicles are. Thus there is a need for a method which can address
these limitations. Data Envelopment Analysis (DEA) – an advanced benchmarking technique – can
potentially address these limitations. Thus, this paper will discuss an application of DEA to van fuel
efficiency measurement.
Research Approach
The fuel efficiency DEA model originally included fuel volume, fuel cost, vehicle’s weight and vehicle’s
age as inputs while mileage was the only output. The fuel information, obtained from fuel cards
records, was cleansed using a cleansing algorithm partly relying on telematics information. Another
algorithm was also used to appraise the volume used during the measurement period (the smoothing
algorithm). Data from three different companies’ fleets was used in this study.
Findings and Originality
The results indicate that DEA can address mpg’s limitations while effectively measuring van fuel
efficiency. No vehicle was found to be simultaneously mpg efficient and ppm inefficient (and vice
versa); thus using either volume or cost, provided similar efficiency levels. Vehicle weight was kept in
the model as it proved to have a significant impact on fuel efficiency while age seemed to further
segment the results in a way which fleet operators defined as incoherent with the notion of fuel
efficiency. Vehicle age was thus excluded from the models. Results from the smoothing algorithm
suggest smoothing the volume used is indispensable when using fuel cards.
Although DEA has been widely used in transport operations, the literature mainly concentrates on
ports or airports (Cullinane et al., 2006, SangHyun, 2009, Yoshida and Fujimoto, 2004, Pestana
Barros and Dieke, 2007) rather than directly on road transport (Yang and Pollitt, 2009). Only a limited
number of papers can be found dealing with the use of DEA to measure road operations efficiency
(Hjalmarsson and Odeck, 1996, Odeck and Hjalmarsson, 1996, Kerstens, 1996, Cowie and Asenova,
1999) and, except for this study, none could be found on van operations or fuel efficiency
measurement. This lack of published research brings originality to this study.
Research Impact
This case study demonstrates that it is possible to use DEA to incorporate ‘vehicle weight’ in the fuel
efficiency model in order to provide a better and more comparable vans’ fuel efficiency measure than
with simple mpg measurement.
Practical Impact
Fleet operators understood the model results and appreciated the fact the measure incorporated
vehicle weight. However, the debriefing discussions seemed to indicate that fleet managers were
more concerned about spotting very bad drivers and fuel theft rather than accurate fuel efficiency
measurement per se. These concerns were partially addressed by the fuel card data cleansing and
smoothing algorithm. Finally, recent success of driver competitions (Masternaut Three X, 2010) seem
to indicate there is a latent need in the industry for accurate driver performance measurement which
suggests that methods such as the one developed in this study could be of greater use in a near
future.