United States Air and Radiation EPA420-R-02-003
Environmental Protection January 2002
Agency M6.EXH.002
svEPA Analysis of Emissions
Deterioration Using
Ohio and Wisconsin
IM240 Data
@9 Printed on Recycled
Paper
-------
EPA420-R-02-003
January 2002
Of
Ohio
IVi6.EXH.002
Phil Enns
Ed Glover
Penny Carey
Assessment and Modeling Division
Office of Transportation and Air Quality
U.S. Environmental Protection Agency
NOTICE
This technical report does not necessarily represent final EPA decisions or positions.
It is intended to present technical analysis of issues using data that are currently available.
The purpose in the release of such reports is to facilitate the exchange of
technical information and to inform the public of technical developments which
may form the basis for a final EPA decision, position, or regulatory action.
-------
1.0 INTRODUCTION
This report documents the collection and analysis of IM240 data conducted in support of
EPA's study of in-use deterioration and the revision of the MOBILE6 emissions inventory model.
EVI240 data have certain strengths and weaknesses relative to other emissions data. After extensively
comparing results based on various data sources, EVI240 data from Ohio were used to correct
deterioration estimates computed from FTP data. These results and a description of the analysis are
reported elsewhere.1 In addition to this limited application, the EVI240 test data have added
considerably to the understanding of vehicle emissions in the overall fleet.
A substantial effort was devoted to collecting and studying data from state inspection and
maintenance (I/M) programs that employ the EVI240 test cycle. This cycle was developed to provide
a relatively short (239 second) test that captures the essential features of the well-established Federal
Test Procedure (FTP), which has long served as the standard for exhaust emissions testing. EVI240
data offer two principle advantages. First, very large samples of data are available since state I/M
programs aim to test all or most registered vehicles. Second, as a result of this comprehensive
testing, the samples approach a census of all vehicles and are thus relatively unbiased.
There are several disadvantages to using EVI240 data when compared with FTP testing. The
test is shorter than the FTP and is therefore considered less representative of real driving. In
particular, the EVI240 test contains no cold start portion. Vehicle preconditioning and ambient
temperature conditions are uncontrolled in EVI240 testing. Data recording at EVI test lanes is generally
less thorough and accurate, especially with regard to odometer readings. Finally, it is necessary to
make several transformations of the EVI240 data to obtain FTP-comparable results. Thus, from the
outset, EVI240 data were considered less satisfactory than FTP data, and regarded as supplementary
rather than primary.
In MOBILE6, vehicle exhaust emissions will be allocated between engine start (start
emissions) and travel (running emissions). This split enables the separate characterization of start
and running emissions for correction factors such as fuel effects and ambient temperature. It also
allows a more precise weighting of these two aspects of exhaust emissions for particular situations
such as morning commute, parking lots and freeways. Because the EVI240 test does not contain a
cold start, data from that test are appropriate only for the study of running emissions; start emissions
are considered in a separate document.2
Section 2 describes the data obtained from Colorado, Arizona, Ohio, and Wisconsin, and
provides a brief summary of the I/M programs in each state. Section 3 defines important terms and
JEnns, P., E. Glover, P. Carey, and M. Sklar, "Determination of Running Emissions
as a Function of Mileage for 1981-1993 Model Year Light-Duty Cars," Report No.
M6.EXH.001.
2Glover, E. and P. Carey, "Determination of Start Emissions as a Function of Mileage and
Soak Time for 1981-1993 Model Year Light-Duty Vehicles," Report No. M6.STE.003.
-------
describes the methodology and analysis of the data. Section 4 presents results and conclusions
obtained from the analysis. Data from Phoenix, Arizona and Colorado were analyzed by an EPA
contractor. The results from that work are described briefly, with details found in referenced reports.
Data from Dayton, Ohio, are discussed more thoroughly in this paper. The Ohio data received
special attention because that location had no previous I/M program; as a result, these data were
considered more representative of intrinsic emissions deterioration.
2.0 IM240 DATA
2.1 Colorado and Phoenix
EVI240 data collected by the state of Colorado between April and August, 1995, were
supplied to an EPA contractor, Systems Application International, Inc. (SAI) for evaluation of
emissions deterioration. The Colorado program uses a fast-pass procedure whereby most passing
vehicles are not required to complete the full 240-second test. However, a special study was
conducted on a random sample of vehicles that were required to complete the full test. These full test
results comprised the data used by SAI.
A significant issue in working with EVI240 test data concerns the accuracy of reported
odometer readings. State I/M programs record odometer incidentally, and under closer scrutiny,
numerous anomalies often are revealed. The most common problem is the occurrence of
unrealistically low mileage given the vehicle's age. In addition, a few cases of exorbitantly high
mileages were observed. The best explanation for many of these observations is that the recorder
miscopied the left-most digit of the odometer, e.g., a car with 121,300 miles is recorded as having
only 21,300 miles. For older vehicles, the odometer may only include five digits, precluding accurate
recording of mileages greater than 99,999.
SAI identified problems with the reported odometer readings, and developed a procedure for
modifying these values so that they would better conform with known properties of vehicle mileage
accumulation. Their approach corrects the odometer based on a probabilistic model of the relation
between vehicle age and mileage. Based on the results from its application to the Arizona and
Colorado data, this methodology appears to have shortcomings. It produces corrected data sets with
certain unnatural features of mileage distribution. The best explanation is that the method's
assumptions are simply not sophisticated enough to describe the true error patterns in the data. A
more satisfactory method was not developed, so this model was used in SAI's subsequent analyses
and was also applied, in modified form, to EPA's study of Ohio EVI240 data.
SAI also developed a statistical technique for identifying emission value outliers for the
purpose of deleting unusually large or small values in the analyses of deterioration. A small
percentage of emission values were adjusted on this basis. Extensive summary statistics were
-------
prepared for both the edited and unedited data.3
Data from the Phoenix, Arizona I/M program collected from January to June, 1996, were also
analyzed by SAL The Arizona program uses a fast-pass/fast-fail (FPFF) algorithm which can pass
or fail the vehicle prior to the end of the JJVL240 sequence. The state also conducts full JJVL240 tests
without using the FPFF algorithm on a randomly-selected two percent of the vehicle population.
These random full JJVI240 test results comprised the data used by SAL These data contained
odometer anomalies of a somewhat different nature, so the contractor modified its correction
software accordingly. It also applied the emission outlier screening procedure developed for the
Colorado data to produce a modified Phoenix data set. Summary statistics were again generated for
the raw and screened data.4
2.2 Ohio
LM240 data collected in Dayton, Ohio were of special interest in this study. This city had no
previous I/M experience, so there is reason to believe that deterioration of measured emissions would
be more "natural" than in regions with earlier I/M programs. Like Colorado, Ohio employs a fast-
pass algorithm to speed up the testing process, but random full LM240 tests are not conducted. As
a result, it is necessary to estimate full 240-second values from the fast-pass results. These estimates
were developed using a regression model developed with data from the Wisconsin I/M program.
Both data sets are discussed below.
The Ohio data includes LM240 test results on all of the 1981 and later registered cars and
light-duty trucks scheduled to be tested from April, 1996 through March, 1997. Since the program
testing frequency is biennial, the sample of vehicles represents approximately half of the overall
vehicle population. Also, because the first month is April, the data set does not contain testing
results collected during early startup in 1995 and the beginning of 1996. The original database
contained more than one million vehicles from three separate Ohio cities (Cleveland, Akron/Canton,
and Dayton/Springfield). Only the data from Dayton/Springfield (referred to as the "Dayton" data
elsewhere in this report) were used in the analysis, and these data were further restricted to the valid
initial tests (no post-repair retests were used). The sample was restricted to Dayton because it had
never implemented any I/M or anti-tampering (ATP) program in the past. The other two sites had
implemented decentralized I/M or ATP programs in the past. These programs may have had some
influence on the prior condition of the vehicles, and thus potentially bias emissions in the sample
from those cities.
3 Cohen, J.P., R.K. Iwamiya, and R.E. Looker, "In-Use Deterioration Data Analysis: Task
1, Initial Data Analysis and Quality Review of Colorado LM240 Data," SYSAPP-96/72d, Draft
Report, Systems Application International, Inc., November, 1996.
4 Cohen, J.P., R.K. Iwamiya, and R.E. Looker, "In-Use Deterioration Data Analysis: Task
1, Initial Data Analysis and Quality Review of Phoenix LM240 Data," SYSAPP-96/76d, Draft
Report, Systems Application International, Inc., November, 1996.
-------
The sample contained both cars and trucks, with significant numbers of data points for all
model years from 1981 through 1995. A vehicle identification number (VIN) decoder developed by
Radian Corp. was used to determine each vehicle's model year, vehicle type (car or truck), and fuel
metering type (fuel injected or carbureted). The VIN decoder was believed to have achieved a high
rate of proper decoding; nevertheless, because of undecodable "VTNs, voided or suspicious emission
tests (unusually low or high CO2 emissions, i.e., less than 100 g/mi or greater than 1500 g/mi), or
missing data in key fields, the sample size was reduced to approximately 211,000 vehicles from
Dayton/Springfield. Table 1 provides a breakdown of the Dayton sample by model year, vehicle
type, and technology (fuel delivery type).
The State of Ohio does not conduct full EVI240 tests on all its vehicles or on a random sample
of vehicles. A small fraction of vehicles take the full 239 seconds to pass. All failures take the full
239 seconds. The vehicles displaying low emissions during the beginning of the test have their test
terminated early under what is called a fast-pass (FP) procedure. The purpose of the FP test is to
speed up the testing process and increase the lane volume so as to more efficiently utilize fixed
testing resources. The FP test is not a standardized test in terms of length, because it can be
terminated as early as 30 seconds into the cycle and as late as 238 seconds.
The FP results are also a function of the specific FP algorithm used to determine whether an
individual vehicle passes out of the test early (fast-passes) or stays in the test the full 239 seconds.
The algorithm is expressed as a large table of cut points (pass/fail standards) that is applied
simultaneously for all three pollutants at each second of the test. The four states which conduct fast-
pass EVI240 testing use slightly different FP algorithms, which makes comparison between the states
somewhat difficult. Ohio performs its fast-pass (FP) testing in accordance with an EPA fast-pass
algorithm recommended in the "EPA High Tech I/M Guidance Document."
As with the Arizona and Colorado data, the mileage recorded in the Ohio data is of
questionable quality for many of the individual vehicles. An odometer correction was judged to be
necessary because of two systematic problems. First, in about 200 cases, the odometer value was
found to be unrealistically high, i.e., over 300,000. The most likely explanation for these outliers is
the accidental recording of the tenth's digit, with the effect of multiplying mileage by ten. For
example, under this scenario, a proper reading of 61,000 miles becomes an unreasonable 610,000
miles. This problem was addressed by eliminating all readings over 300,000 miles. However, for
vehicles with less than 300,000 miles it was difficult to determine whether improper recording of
the tenth's digit occurred.
As with the Colorado and Arizona EVI240 data, inspection of the Dayton, Ohio, data set
suggests the second type of problem, odometer rollover. This problem was considered far more
widespread and difficult to rectify. One reason for odometer rollovers is that many vehicle odometers
in earlier model years were not designed to record mileage greater than 99,999, and frequently during
the I/M process the inspector merely records the displayed mileage. Moreover, the collection of
accurate odometer data by an I/M program usually is not an important priority, giving rise to
concerns about the values even in newer vehicles. Consequently, in a substantial number of vehicles,
mileage is unrealistically low for the given vehicle's age. Initially, a correction for this problem was
attempted using a modified version of the methodology created by SAI for use with the data from
-------
Colorado and Phoenix. In the modified approach, the odometer for a vehicle with low mileage may
be incremented by 100,000 miles if it fits a particular profile. In effect, this procedure adds 100K
miles to selected vehicles; the proportion of vehicles so adjusted grows with age. The choice of
vehicles assigned a new mileage is made probabilistically according to distributions fitted to the data
under certain assumptions underlying the S AI methodology. While not an ideal solution, it was felt
that this method yielded an improvement to the uncorrected odometer values in the raw data. After
reviewing the corrected odometers, however, a decision was made to use region-specific mileage
accumulations instead for subsequent analysis.
2.3 Wisconsin Second-by-Second Data
Raw emission values from the Ohio data are not directly comparable to one another because
they correspond to varying test durations. To address this concern, a model for predicting the full
239-second emissions rate from a partial fast-pass test score was developed. This model was fitted
using second-by-second data from the Wisconsin EVI240 test program. These data were collected as
a random sample over three different months (December, 1995, April, 1996, and October, 1996).
It contained data on 3,148 cars and 1,192 light trucks with a range of model years from 1981 through
1995. The Wisconsin EVI240 data were chosen over data from the other two EVI240 states (Arizona
and Colorado) with second-by-second data because of the geographic, demographic, and
meteorological similarities between Ohio and Wisconsin. Furthermore, both states use the same
testing contractor, so analyzers and specific test procedures are likely to be similar. Results from the
Wisconsin data analysis are reported in the next section.
3.0 METHODOLOGY AND ANALYSES
3.1 Colorado and Phoenix
SAI performed an extensive analysis of emissions deterioration for the combined
Phoenix/Colorado data set. A variety of regression models was investigated, and several are
discussed in detail.5 They focus on the logarithmic transformation of emissions, which tends to give
better model fits than the untransformed raw data. Model coefficients estimated in this way provide
multiplicative adjustments to a baseline zero-mile emission rate. In general, the models do not fit the
data especially well due to the lack of good baseline values for individual vehicles. Highlights of
SAI's findings include the following:
1. As expected, emissions deteriorate with increasing mileage.
2. The rate of deterioration of emissions is less at higher mileages and in older model year
5 Cohen, J.P., R.K. Iwamiya, and R.E. Looker, "Analysis of In-Use Deterioration of
Emissions Using I/M 240 Data," SYSAPP-97/06d, Draft Final Report, Systems Application
International, Inc., February, 1997.
-------
vehicles.
3. CO and HC emissions in Colorado tend to be higher than in Arizona; with NOx, the
opposite effect is observed.
3.2 Fast-Pass to Full IM240 Conversion
3.2.1 EPA Approach
As noted earlier, the Ohio EVI240 test uses a fast pass algorithm, i.e., the test is terminated
prematurely for vehicles displaying low emissions at an elapsed time of as little as 30 seconds.
Partial and full test emissions are all measured in grams per mile. Nevertheless, examination of
second-by-second data gives ample evidence that emissions from tests of varying duration are not
directly comparable, since the speed-acceleration mix changes over the cycle. Therefore, estimation
of a "simulated" full EVI240 test score was undertaken for all passing vehicles for which only a fast
pass score was available.
From the Wisconsin second-by-second data, regression models were constructed in which
the full EVI240 emissions are predicted based on several independent variables using only the tests
from passing vehicles. The natural logarithm of emissions was used as the independent variable.
Model regressors include vehicle type (car or truck), fuel metering type, model year, and simulated
length of test (in seconds). The simulated test length was determined by applying the fast-pass
algorithm used in Ohio to the second-by-second emissions. These models give good fits, with R-
squared values ranging from 70% to 82%. Table 2 reports the coefficients for these models.
The coefficients from these models were then used to predict full EVI240 scores for each Ohio
fast pass test. Because the models fit the natural logarithm of emissions, the antilog transformation
was employed to obtain values in EVI240 space.
3.2.2 RFF Approach
An alternative approach to estimating full EVI240 scores from fast-pass scores was proposed
by Resources for the Future (RFF). This methodology involves regressing the EVI240 emissions
against emissions at a given time point in the test using all tests. The RFF model includes the model
year variable (but not vehicle type or fuel metering system). This approach produces a different
equation for every test duration between 30 and 239 seconds. This approach avoids the problem of
correlation between the regressor and error terms that may affect the EPA models. However, there
is concern that it may be inappropriate to use tests from failing cars with high emissions in fitting
an equation intended to represent a fast pass (i.e., low-emission) outcome.
RFF tested its method on second-by-second data from Arizona, with acceptable results.
When applied to the Ohio data, the EPA and RFF models produced similar mean estimates of full
EVI240 scores (see Table 3). Therefore, it was decided to use the values generated by the EPA
approach.
-------
3.3 IM240 to Running LA4 Conversion
The next step involved estimating running LA4 scores for each Ohio vehicle. (The Running
LA4 is a test cycle comprised of the 1372-second LA4 trace that underlies the FTP, with no start
component. Unlike previous versions of MOBILE, MOBILE6 will treat start and running emissions
separately.1) This estimation was achieved using two sets of regression models. The first set of
models was developed from a sample of 77 tests conducted for the purpose of estimating a Running
LA4 from a conventional three-bag FTP. This test program and its analysis are described in a
separate report.6 The results of this work were then used with a sample of FTP and EVI240 paired
tests conducted on vehicles chosen from I/M lanes in Hammond, Indiana and Phoenix, Arizona. This
sample was comprised of 997 vehicles, of which 93 8 are from model year 1981 or later. Coefficients
from the 78-vehicle study were used to convert the FTP bag scores to Running LA4 values for
subsequent correlation with matching EVI240 scores.
The EVI240 tests at the inspection lane were, of course, based on the vehicle's tank fuel.
When moved to the lab, each vehicle's fuel was replaced with Indolene in accordance with standard
test protocol. Then, in addition to the FTP, a lab EVI240 was conducted. For this analysis, however,
the IM240-to-FTP conversion was made between the lane EVI240 (tank fuel) and the lab FTP
(Indolene), since EVI240 data from Ohio is based on tank fuel tests.
Table 4 shows the results of modeling the log of running LA4 emissions as a function of log
of EVI240 plus dummy variables representing vehicle fuel metering technology and model year group
(defined in the table). The coefficients from these models were then applied to the Ohio data to
produce fitted Running LA4 scores for that much larger data set. The modeling approach is similar
to that used to simulate full EVI240 scores from fast pass scores. The fitted values of the natural log
of Running LA4 are converted to gram per mile space using the antilog transformation.
4.0 RESULTS AND CONCLUSIONS
Inspection and maintenance programs provide large samples of emission test data. These data
are not subject to the types of recruitment bias found in samples collected for other purposes such
as the EPA emissions factor program. The EVI240 test is designed to better emulate actual on-road
driving than older I/M procedures. Therefore, these data offer a number of possibilities for improved
modeling of emissions deterioration and other types of behavior. Despite these benefits, as this report
explains, currently available EVI240 test data also suffer from shortcomings that need to be addressed
before the data can be used directly in modeling emissions.
For MOBILE6, the Ohio EVI240 data are used indirectly to modify emission rates derived
from FTP data. In this application, FTP data are employed to determine running LA4 deterioration
for Tier 0 cars. The resulting emission rates are then adjusted by applying a high emitter correction
6 Brzezinski, D., E. Glover and P. Enns , "The Determination of Hot Running Emissions
From FTP Bag Emissions," Report No. M6.STE.002.
-------
factor based on the Ohio IM240 data. These correction factors were judged necessary to adjust for
possible bias in the samples of FTP test data collected by EPA and the vehicle manufacturers. This
bias is attributed to questions concerning vehicle owner willingness to participate in emissions
testing programs. The Ohio EVI240 scores were used to develop the high emitter correction factors,
but regional-specific annual average odometer readings by vehicle age were substituted for the
reported odometers. This approach takes advantage of the Ohio EVI240 data set, which represents
a large sample of vehicles not subject to a previous I/M program, while also limiting one of the
major shortcomings of the Ohio EVI240 data, i.e., the reported odometer readings. The methodology
used to develop the high emitter correction factors is described in the paper cited earlier.1
-------
Table 1
Distribution of Dayton, Ohio IM240 Data
MODEL YEAR
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
ALL
VEHICLE
CAR
FUEL METERING
GARB
924
2,767
2,791
7,105
4,329
6,771
2,777
3,092
1,399
267
7
32,229
FI
140
882
998
4,146
4,542
11,207
8,041
16,367
10,117
16,606
9,519
16,604
10,646
13,740
7,895
783
132,233
ALL
1,064
3,649
3,789
11,251
8,871
17,978
10,818
19,459
11,516
16,873
9,526
16,604
10,646
13,740
7,895
783
164,462
TRUCK
FUEL METERING
GARB
158
862
739
2,182
1,749
1,805
626
443
82
119
6
8
8,779
FI
7
19
5
87
316
2,873
2,278
5,236
3,231
4,271
3,074
5,289
3,517
5,061
2,528
230
38,022
ALL
165
881
744
2,269
2,065
4,678
2,904
5,679
3,313
4,390
3,080
5,289
3,517
5,061
2,528
238
46,801
ALL
FUEL METERING
GARB
1,082
3,629
3,530
9,287
6,078
8,576
3,403
3,535
1,481
386
13
8
41,008
FI
147
901
1,003
4,233
4,858
14,080
10,319
21,603
13,348
20,877
12,593
21,893
14,163
18,801
10,423
1,013
170,255
ALL
1,229
4,530
4,533
13,520
10,936
22,656
13,722
25,138
14,829
21,263
12,606
21,893
14,163
18,801
10,423
1,021
211,263
10
-------
Table 2
Fast-Pass to Full IM240 Regression Models
Using Wisconsin Second-by-Second IM240 Data
These models use the following independent variables:
LFxx = Natural Log(fast past gram/mile value of pollutant xx)
F=0 (carbureted fuel metering), =1 (fuel injected)
V=0 (truck), =1 (car)
D;=0 (not model year i), =1 (model year i) , i=1981 to 1994
D;*LFxx = Crossproduct to capture slope change with model year
The Di and crossproduct coefficients are not shown.
The dependent variable is Lxx = Natural Log(240-second gram/mile value of xx).
Dependent Variable: LCO - Log(IM240 CO)
Root MSB
Dep Mean
C.V.
0.68473
1.17901
58.07704
R-square
Adj R-sq
0.7034
0.7013
Parameter Estimates
Variable DF
Parameter
Estimate
Standard
Error
T for HO:
Parameter=0
Prob
INTERCEP
LFCO
F
V
FSEC
-0.042493
0.497165
-0.238492
0.070382
0.003180
0.04651906
0.01568796
0.03735494
0.02377776
0.00019780
-0.913
31.691
-6.384
2.960
16.075
0.3611
0.0001
0.0001
0.0031
0.0001
11
-------
Table 2 (Continued)
Dependent Variable: LHC - Log(IM240 HC)
Root MSE
Dep Mean
C.V.
0.50372
-1.37212
-36.71100
R-square
Adj R-sq
0.8189
0.8176
Parameter Estimates
Variable DF
Parameter
Estimate
Standard
Error
T for HO:
Parameter=0
Prob
INTERCEP
LFHC
F
V
FSEC
-1.594231
0.529476
-0.109558
0.161715
0.003870
0.04709904
0 . 01496448
0.02765615
0.01776551
0.00014103
-33.848
35.382
-3.961
9.103
27 .444
0.0001
0.0001
0.0001
0.0001
0.0001
Dependent Variable: LNOX - Log(IM240 NOX)
Root MSE
Dep Mean
C.V.
0.43297
-0.26214
-165.16871
R-square
Adj R-sq
0.7359
0.7340
Variable DF
Parameter Estimates
Parameter
Estimate
Standard
Error
T for HO:
Parameter=0
Prob
INTERCEP
LFNO
F
V
FSEC
-0.681601
0.480133
-0.023815
0.082967
0 . 001471
0.03277202
0.01264385
0.02363659
0.01511348
0.00011878
-20.798
37.974
-1.008
5.490
12.382
0.0001
0.0001
0.3137
0.0001
0.0001
12
-------
Table 3
Comparison of EPA and RFF Full 240-Second Means by Model Year
Carbureted Vehicles
RFF
FP
I EPA | RFF
IM240 FP IM240 FP IM240
EPA
FP IM240
RFF
FP IM240
EPA
FP IM240
MODEL
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1996
HC
1
1
| 1.42
1. 84
| 1.88
1.67
| 1.55
1. 57
| 1.28
| 1.13
| 0.89
| 0.75
| 0.96
| 0.81
| 0.40
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
HC
1.
1.
1.
1.
1.
1.
1.
0.
0.
0.
0.
0.
0.
1
1
1
22 |
74 |
76 |
52 |
38|
37|
07 |
90
64 |
51 1
62 |
72
H|
HC
1.
1.
1.
1.
1.
1.
1.
1.
0.
0.
0.
0.
0.
1
42 |
85|
89 |
67|
56 |
57 |
29 |
13 |
89 |
75 |
96 |
82
41|
HC
1.
1.
1.
1.
1.
1.
1.
1.
0.
0.
0.
0.
0.
1
1
1
32 |
79
95 |
55 |
47|
46
17 |
00
77 |
56 |
61|
78
20 |
CO
21.
29.
27.
23 .
23.
22 .
18.
17.
13.
12 .
11.
13 .
4.
1
1
. 69
.73|
.44
.90 |
.97|
.24|
.10|
.30|
.92 |
.93|
CO
20.
29.
27.
23 .
23.
22 .
18.
15.
11.
10.
10.
12 .
1.
1
1
1
25|
69
52 |
14 |
19 |
06
07 |
14 |
05 |
98
59 |
CO
21.
29.
27.
23 .
23.
22 .
18.
17.
13.
12 .
11.
13 .
4.
1
16 |
70
73 |
44
91 |
92
97 |
24 |
10 |
30|
93 |
72
94 |
CO
22.
31.
34.
22 .
25.
23 .
21.
16 .
13.
11.
10.
13 .
3.
1
1
1
66 |
23 |
48 |
71 |
68 |
03 |
70 |
83 |
50 |
08
51 1
15 I
NOX
1.
2 .
2.
2 .
1.
1.
1.
1.
1.
1.
1.
1.
0.
1
85|
13 |
22 |
13 |
93 |
95 |
79 |
68
35|
28
02 |
H|
45|
NOX
1.
2 .
2.
2 .
2.
2 .
1.
1.
1.
1.
0.
1.
0.
1
1
1
91 |
27
33 |
25|
02 |
00
83 |
69
36|
22
99 |
08
38|
NOX
1.
2 .
2.
2 .
1.
1.
1.
1.
1.
1.
1.
1.
0.
1
1
85|
14 |
22 |
14 |
93 |
95 |
79 |
68
35|
28
02 |
H|
46 |
NOX
1.96
2 . 17
2.24
2 . 19
1.92
2 . 04
2.14
1. 71
1.53
1.23
0.93
1. 11
0.42
M6.EXH.002
13
-------
Table 3 (continued)
Fuel Injected Vehicles
RFF
FP IM240
EPA
FP IM240
RFF
FP IM240
EPA
FP IM240
RFF
FP IM240
EPA
FP IM240
MODEL |
VFP AT? -i-
1
1980
1981 |
1982
1983 |
1984
1985 |
1986
1987 |
1988
1989 |
1990
1991 |
1992
1993 |
1994
1995 |
1996
HC
0.
1.
1.
1.
1.
1.
1.
1.
0.
0.
0.
0.
0.
0.
0.
0.
0.
. 74
.69
.42
.35
.34
.37
.25
.10
. 99
.90
. 75
.59
. 50
.43
.31
.23
. 16
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
HC
0.
1.
1.
1.
1.
1.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
1
1
48
53 |
30|
18 |
13 |
10 |
93 |
79 |
65|
56 |
43 |
38|
29
21|
H|
06 |
03 |
HC
0.
1.
1.
1.
1.
1.
1.
1.
1.
0.
0.
0.
0.
0.
0.
0.
0.
1
75 |
70 |
43 |
36|
34 |
37|
26
H|
00
91 |
76 |
59 |
51 1
43 |
31|
24 |
17 |
HC
0.
1.
1.
1.
1.
1.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
1
1
53 |
54 |
40
15 I
14 |
12 |
98
85|
75 |
59 |
40
42 |
39|
26|
20
12 |
09
CO
9.
23.
17.
16.
15.
15.
13 .
12.
10.
10.
10.
8.
7.
6.
5.
4.
2 .
1
.27
.37|
.21|
.19|
. 88
.58|
•H|
.16|
. 86
.76|
.13|
.24|
.50|
.51|
.03|
.21|
.58|
CO
7.
23.
17.
16.
15.
15.
12 .
10.
8 .
8.
7.
6.
5.
4.
2 .
2.
0.
1
1
36|
73 |
53 |
09 |
38|
65|
18 |
99 |
98
36|
29
49 |
46
32 |
86
02 |
95 |
CO
9.
23.
17.
16.
15.
15.
13 .
12.
10.
10.
10.
8.
7.
6.
5.
4.
2 .
1
28
37|
21|
19 |
89
58 |
H|
17 |
86
77 |
13 |
25|
51 1
52 |
03 |
22 |
58 |
CO
7.
23.
21.
14.
14 .
14.
12 .
10.
9.
8.
6 .
7.
7.
4.
3 .
2.
1.
1
1
80
21|
42 |
60 |
98
02 |
91 |
66 |
56 |
10 |
79
00 |
39|
65|
78
25|
53 |
NOX
1.
1.
2 .
2.
2 .
1.
1.
1.
1.
1.
1.
1.
0.
0.
0.
0.
0.
1
19 |
87|
70
42 |
33 |
96 |
56 |
64 |
42 |
37|
26
17 |
96
90 |
57 |
46 |
33 |
NOX
1.
2.
2 .
2.
2 .
2.
1.
1.
1.
1.
1.
1.
0.
0.
0.
0.
0.
1
1
15 I
09 |
83 |
55 |
43 |
05 |
66
67|
43 |
32 |
16 |
04 |
84
75 |
52 |
42 |
32 |
NOX
1.
1.
2 .
2.
2 .
1.
1.
1.
1.
1.
1.
1.
0.
0.
0.
0.
0.
1
1
20
88 |
71 |
43 |
33 |
96 |
57 |
64 |
42 |
37|
27
17 |
96
91 |
58 |
47|
34 |
NOX
1.24
1.88
2 . 67
2.42
2 .27
2.03
1. 92
1.64
1. 59
1.30
1. 12
1.03
1. 12
0.76
0. 78
0.38
0.30
M6.EXH.002
14
-------
Table 4
Regression models: Running LA4 vs. IM240
These models use the following independent variables:
LFxx = Natural Log(fast past gram/mile value of pollutant xx)
FI=0 (carbureted fuel metering), =1 (fuel injected)
Ml=l (model years 1981-82), =0 (otherwise)
M2=l (model years 1983-87), =0 (otherwise)
The dependent variable is
LxxRUN = Natural Log(Running LA4 gram/mile value of xx).
Dependent Variable: LHCRUN - Log(Running LA4 HC)
Root MSB 0.79242 R-square 0.7068
Dep Mean -1.00124 Adj R-sq 0.7055
C.V. -79.14334
Parameter Estimates
Parameter Standard T for HO:
Variable DF Estimate Error Parameter=0 Prob > |T|
INTERCEP 1 -0.750184 0.09554650 -7.852 0.0001
LHCIM 1 0.948738 0.02344691 40.463 0.0001
FI 1 -0.006354 0.08638848 -0.074 0.9414
Ml 1 0.827859 0.12575290 6.583 0.0001
M2 1 0.383417 0.05923913 6.472 0.0001
15
-------
Table 4 (continued)
Dependent Variable: LCORUN - Log(Running LA4 CO)
Root MSE
Dep Mean
C.V.
0.91602
1.65384
55.38721
R-square
Adj R-sq
0.6434
0.6419
Parameter Estimates
Variable DF
Parameter
Estimate
Standard
Error
T for HO:
Parameter=0
Prob
INTERCEP
LCOIM
FI
Ml
M2
-0 .700927
0.956897
0.173259
0.770544
0.193573
0.11976277
0.02533007
0.09985916
0.14457986
0.06678826
-5.853
37 .777
1.735
5.330
2.898
0.0001
0.0001
0.0831
0.0001
0.0038
Dependent Variable: LNOXRUN - Log(Running LA4 NOX)
Root MSE
Dep Mean
C.V.
0.57696
-0.34445
-167.50296
R-square
Adj R-sq
0.6533
0.6518
Parameter Estimates
Variable DF
Parameter
Estimate
Standard
Error
T for HO:
Parameter=0
Prob
INTERCEP
LNOXIM
FI
Ml
M2
-0.623836
0.849380
-0.089006
0.280035
0.209848
0.06824633
0.02276379
0.06267262
0.09190557
0 . 04307749
-9 .141
37.313
-1.420
3.047
4 . 871
0.0001
0.0001
0.1559
0.0024
0.0001
16
------- |