EPA's Proposal for MOBILE6
                  Definition of High Emitter Category

                             March 7, 1997


Abstract

     The distribution of emission levels for vehicles is skewed, such
that a small number of vehicles account for a large fraction of the
total emissions as compared to an otherwise normally distributed vehicle
sample.  This was accounted for in MOBILE5 by stratifying vehicles into
four "emitter" groups for hydrocarbon and carbon monoxide emissions and
estimating the emission levels of each group separately.

     o Normal
     o High
     o Very High
     o Super

     Emissions of oxides of nitrogen were divided into only two groups,
normal and high emitters.  The rate of occurrance versus age (mileage)
of these groups was also determined.  A fleetwide emission estimate was
then calculated from the combination of the emissions from each emitter
group, weighted by their occurrance in the fleet at that age.

     Specific emission levels were chosen to define each of the emitter
categories.  The definition of these emitter groups has been different
in different versions of the MOBILE model, depending on the available
vehicle sample. For MOBILE5, the emitter categories, for the most part,
were multiples of the Federal Test Procedure certification emission
standards.  

     For MOBILE6, start emissions will be separated from running
emissions.  As a result, the existing MOBILE5 definition of emitter
levels, based on a combination of start and running emissions, cannot be
used for MOBILE6.

Summary of the Proposal

     There are two principal candidates for methodology to define the
high emitter category. The first approach is to estimate an upper
percentile for emissions above which a vehicle is high emitting by
definition. By this method, a constant percentage of all vehicles are
considered high emitters. A simple model of this approach that accounts
for vehicle deterioration can be constructed using the methods of
regression analysis. The regression line, which represents average
emissions as a function of mileage or some other usage measure, is
fitted to in-use data. Given some assumptions about the distribution of
emissions at a particular mileage level, we can compute the percentage
of vehicles expected to show emissions above any level at that mileage.
It is common to assume that this distribution is normal with constant
variance at all mileages. This basic approach can be made more realistic
in several ways. For example, average deterioration may follow a
nonlinear function of mileage; the constant variance assumption may not
be appropriate with raw emissions values, but may be appropriate for a
transformation such as the logarithm of emissions. Moreover, stratifying
the analysis by technology categories is likely to yield different
estimates of variance and, thus, of high emitter class boundaries.

     One criticism of this method is that, assuming an upward slope in
emissions with usage, using a fixed percentile to define the high
emitter implies that a vehicle might be judged a high emitter at low
mileage, while a high-mileage vehicle with the same emissions could be a
normal emitter. Some modification of the simple approach could
presumably be used to address questions like this. 

     The second basic approach is to treat the high emitter as a
fundamentally different vehicle whose occurrence and emissions are
generated by processes different from those of normal emitters. With
this model, it is necessary to estimate both the probability that a
vehicle is a high emitter and the model describing its deterioration
characteristics. As an estimation problem, this is a more complex
methodology because it requires identification of high emitting
vehicles, which is generally subjective. Given this identification, a
model for discriminating between high and normal emitters is needed. If
the sample of high emitters is small, there may be problems in achieving
sufficient confidence in the estimates. As with the previous
methodology, this approach would likely involve stratifying the analysis
by technology categories. 

     EPA's proposal is to analyze the data using both approaches and
determine at that time which of the two methods is superior.  For more
details on the sample of vehicles to be used for this analysis, refer to
the document, "EPA's Proposal for MOBILE6: Basic Exhaust Emissions and
Deterioration."