United States Air and Radiation EPA420-R-01-055
Environmental Protection November 2001
Agency M6.ACE.002
&EPA Air Conditioning Correction
Factors in MOBILES
yŁu Printed on Recycled
Paper
-------
EPA420-R-01-055
November 2001
Air Conditioning Correction Factors in MOBILE6
M6.ACE.002
John W. Koupal
Janet Kremer
Assessment and Standards 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 ABSTRACT
Revised air conditioning exhaust emission correction factors are included in MOBILE6. The
proposed factors are based on testing of 38 vehicles at two locations, using a test procedure
meant to simulate air conditioning emission response under extreme "real world" ambient
conditions. These factors are meant to predict emissions which would occur during full loading
of the air conditioning system, and will be scaled down in MOBILE6 according to ambient
conditions input by the user if appropriate. It was concluded that the data used in the
development of the proposed factors adequately represents real world conditions, based on the
results of a correlation vehicle tested at both test sites and a full environmental chamber. In
general, running emissions were found to increase during air conditioning operation, but under
some conditions HC and CO emissions decreased. Correction factors for start driving were also
assessed.
2.0 INTRODUCTION
Recent studies conducted primarily as part of the Supplemental Federal Test Procedure (SFTP)
rulemaking development process indicate that vehicle fuel consumption and exhaust emissions
increase substantially when the air conditioner is in operation. As the traditional method for
accounting for the effects of air conditioner load - increasing dynamometer horsepower by 10% -
is not adequate for characterizing this emission increase, new certification test procedures aimed
at reducing emissions when the air conditioner is in operation were implemented as part of the
SFTP rule. Air conditioning correction factors are included as an optional element of MOBILES;
however, these factors are based on testing performed in the early 1970's and are considered so
outdated that the user is discouraged from using them in the MOBILE User's Guide. Given the
recent findings on air conditioning emissions, revised air conditioning correction factors are
clearly needed.
This report presents the "full-usage" air conditioning exhaust correction factors proposed for
MOBILE6. Full-usage correction factors are meant to represent the emission increase when the
A/C system is inducing full system load on the vehicle, as would occur under extreme ambient
(temperature, humidity and solar load) conditions. Since it not appropriate to apply these factors
to all ambient conditions, MOBILE6 will scale these factors down based on the ambient
conditions under which the model is being run (the development of appropriate scaling factors is
discussed in Report Number M6. ACE.001, "Air Conditioning Activity Effects in MOBILE6").
Discussion in this report includes the testing used to generate A/C emission data, correlation
between the two test sites and with expected real-world results, and the development of the full-
usage correction factors. The treatment of air conditioning correction factors for vehicles
complying with the SFTP requirement are also addressed in this report.
Subsequent to publication of the draft version of this report in March 1998, the document was
put out for stakeholder review. Formal peer review comments were also solicited from two
independent sources. No comments were received through the stakeholder review process,
-------
hence peer review comments represent the only external feedback received on this report. A
summary of peer review comments are contained in Appendix K. Major revisions were made to
the methodology presented in the March 1998 version, and a revised report was published in
December 1999 as part of the Tier 2 regulatory support documentation. The primary update to
this final report from the December 1999 version is the inclusion of the discussion for SFTP
benefits in Section 8.0.
3.0 TESTING
3.1 Vehicles
The data used for this analysis was generated through testing performed at EPA's National
Vehicle and Fuel Emissions Laboratory and through an EPA contractor, Automotive Testing
Laboratories (ATL), in East Liberty, Ohio. 26 vehicles were tested at EPA and 12 were tested at
ATL, including one vehicle tested at both locations for correlation purposes (treated as two
separate vehicles for the purpose of this analysis). A list of the vehicles tested is contained in
Table 1 of Appendix A. The sample consisted of 1990 and later vehicles categorized as follows:
24 cars / 14 trucks, 32 Ported Fuel Injection (PFI) / 6 Throttle-Body Injection (TBI), and 28 Tier
0/10 Tier 1. Each vehicle was designated either as a "normal" emitter or "high" emitter using
the following emission cutpoints over the Running LA41 : 0.8 g/mi HC, 15.0 g/mi CO and 2.0
g/mi NOx (the cutpoints were applied independently for each pollutant, so that a vehicle could be
a high emitter for HC and a normal emitter for NOx). These cutpoints yielded five high emitters
for HC, three for CO and two for NOx. It should be noted that during the analysis the data was
divided into different strata; by facility cycle, vehicle class, etc.. There are some instances in the
analysis where a vehicles emissions classification changes. For example, a vehicle that is
classified as a normal emitter on the LA94, may have been reclassified as a high emitter for a
specific cycle, such as the local cycle.
3.2 Test Procedure
EPA's new air conditioning test procedure is based on use of a full environmental chamber at 95 °
F, 40% Relative Humidity and full solar load (850 Watts/Meter2). This type of facility was not
available to EPA at the time of testing, so use of a procedure which simulated these conditions
was required. A/C-on tests were conducted in a standard emission test cell at 95 ° F and 50
grains/pound of humidity with a standard cooling fan and the driver window down. The A/C
system was set according to the SFTP requirements; maximum A/C and blower setting with
recirculation mode if so equipped. Rather than attempting to represent a condition that would
actually occur in-use, this simulation is meant solely to induce the level of A/C system load on
the vehicle which would occur in the real world under extreme ambient conditions. Operating
"Running LA4" emissions were derived from the combination of emissions from Bag 2 and a 505 cycle run
warmed-up (i.e. without a soak). More detail on this calculation can be found in MOBILE6 Report No.
M6.STE.002, "The Determination of Hot Running Emissions from FTP Bag Emissions"
-------
with the driver window down and with standard cooling is meant to compensate for the lower
humidity level and lack of solar load inherent in the standard cell. This simulation method
showed adequate correlation with SFTP environmental cell conditions during the development of
the SFTP rulemaking2, and is a straightforward way to approximate real-world air conditioning
emissions using a standard cell setup. A/C-off tests were run in standard FTP ambient conditions
(75° F, 50 grains/pound humidity).
The vehicles were run in a warmed-up condition over EPA's facility-specific inventory cycles3,
ARB's Unified Cycle (the LA92), and the New York City Cycle one time each with the A/C on
and A/C off. A cold start ST014 cycle was also run in both conditions for the purpose of
assessing start A/C factors (information on all driving cycles used in this test program is shown
in Table 2). The EPA tests were run on a 48-inch electric dynamometer, while the ATL testing
used a twin 20-inch electric dynamometer; all tests were run without the 10% A/C load
adjustment factor typical to standard emission tests. Both bag and modal data were collected.
3.3 Correlation
Preliminary data presented at the October 1997 MOBILE6 workshop indicated a potential offset
between the results from vehicles tested at EPA and those tested at ATL5. A/C ratios from the
ATL sample were lower than EPA on average for fuel consumption and all three pollutants.
This raised a question about whether the simulation as conducted at ATL induced comparable
A/C system loading to the procedure as conducted at EPA. A related issue is whether loading
induced by the simulation as conducted at either site could be considered "full-usage", as defined
for the purpose of this analysis by the conditions used for the SFTP certification test (95 ° F, 40%
Relative Humidity, 850 W/m2 solar load). To investigate both issues, a correlation vehicle was
run over all test cycles using the simulation procedure at EPA and ATL, and on a subset of cycles
under the SFTP test conditions at GM's environmental chamber in Rochester, New York. This
vehicle was instrumented to monitor A/C compressor cycling and compressor pressures (high
and low side) on a real-time basis to gain a fuller sense of how the vehicle's A/C system was
loaded at each location.
Emission results for the four cycles tested at all three locations are shown in Table 3 of Appendix
Results from a correlation program between this simulation and a full environmental chamber over a sample of six
Tier 1 vehicles can be found in AAMA/AIAM's comments to EPA on the proposed SFTP rulemaking (EPA Docket
No. A-92-64 ItemIV-D-10).
For detail on the development of EPA's facility-specific inventory cycles, see MOBILE6 Report No. M6.SPD.001,
"Development of Speed Correction Cycles"
4 ST01 is a 1.4 mile cycle developed to specifically characterize driving behavior following startup. The cycle was
developed from an in-use driving survey conducted in Baltimore, Spokane and Los Angeles as part of the SFTP
rulemaking process.
5 "A/C Effects in MOBILE6", presentation at the October 1997 MOBILE6 workshop
-------
A. There is quite a bit of variability in the HC, CO and NOx results, making it difficult to
discern any clear trend. Judging from the large swings in each pollutant, it appears that the
vehicle went into enrichment sporadically between sites, resulting in a wide range of A/C ratio
results across the test matrix. Thus, it is difficult to draw conclusions from the emission data
(and particularly the A/C ratios) alone. The correlation analysis therefore focused on fuel
consumption (carbon) ratio and compressor operation to determine whether a difference in the
relative loading placed on the vehicle between the three sites can be distinguished. The carbon
ratio results in Table 3 show the ATL results to be slightly lower than EPA for each cycle.
However, the EPA and ATL carbon ratios are higher than the GM ratio for three of the four
cycles, and the three locations show relatively consistent carbon ratios over the New York City,
Unified and Arterial cycles. The exception to the latter point is the High Speed Freeway cycle,
for which the GM ratio (as well as the A/C-on carbon levels) are significantly lower than EPA or
ATL.
Table 4 of Appendix A contains compressor behavior data, expressed in terms of the compressor
fraction (the fraction of time the compressor in engaged during the test), and average high and
low side compressor pressures, on which compressor torque in based. The data indicate that a)
the compressor was engaged at all locations 97% or more of the time on each of the cycles, and
b) for the New York City, Unified and Arterial cycles a strong difference is not observed in the
compressor pressures. The exception again is the High Speed Freeway cycle, for which the GM
data shows significantly lower compressor pressures than ATL or EPA. From these data and the
fuel consumption results, it is apparent that the A/C system load on the high speed freeway cycle
in the full environmental cell was much less than that produced by the simulation at EPA or
ATL. The most plausible explanation for this is the use of a variable speed fan in the full
environmental cell, which would create a much higher airflow than produced by the standard
one-speed fan used on the simulation. Higher air flow across the vehicle's A/C system can
increase system efficiency, reducing relative load demand on the engine. This suggests that the
simulation could be over predicting A/C loading (and hence emissions) at the higher speed
levels; however, this effect does not appear in the overall LA92 results, a cycle which also
contains significant high speed operation. Unfortunately sufficient data does not exist over high
speed operation with representative air flow to make a more full assessment; further research will
be needed to address this issue.
From the fuel consumption and compressor data it was concluded for the purposes of this study
that despite observed emission differences between ATL and EPA, the vehicles were adequately
loaded at both sites to represent full-usage conditions. Therefore, no vehicles will be excluded
from the analysis and the emission results from the data set will be used directly (i.e. with no
scaling) to develop the full-usage correction factors.
4.0 DATA
As with previous versions of the model, MOBILE6 will contain correction factors which
estimate the emission impact of changes in temperature. Emissions at temperatures higher than
-------
75 ° F will be determined in the model first by applying a base temperature correction, then
applying the A/C correction factor appropriate for that temperature. A/C correction factors must
be developed separately from the baseline temperature corrections in order to avoid double-
counting temperature impacts. For this analysis, therefore, the A/C-off results were corrected
from the temperature the test was conducted (nominally 75°, although minor variability is
common) to the A/C-on temperature (nominally 95 °) for each paired test. Since MOBILE6
temperature correction factors will not change from the MOBILES corrections, MOBILES
temperature corrections were used6. The Bag 2 corrections were used for all running tests, and
Bag 1 corrections were used for the cold start ST01 test7.
Once the temperature correction was applied, the A/C impact was analyzed by taking the
difference between vehicle emissions with A/C on and the corrected vehicle emissions with the
A/C off (A/C base). This impact is referred to throughout this report as the "A/C effect." This
approach to looking at A/C impact differs from the proposed approach in the draft version of this
report. The draft proposal looked at the A/C impact as a ratio, therefore making the A/C
correction factor a multiplicative adjustment. This significant shift in approach between the draft
and final report because of concern that the multiplicative adjustment approach may overstate the
impact of air conditioning on emissions. The use of an additive A/C effect is meant to mute the
emission impact for technologies which were not represented in the A/C dataset. The shift from
a multiplicative to an additive approach was supported by peer review comments, as discussed in
Appendix K.
5.0 RUNNING CORRECTION FACTORS ANALYSIS
The development of running correction factors requires analysis of what vehicle groupings merit
separate treatment. Simple factorial Analysis of Variance (ANOVA) was used, for each
pollutant, to look at the A/C effect as a function of the following factors: base emissions (i.e.
without A/C), referred to as "A/C base"; vehicle class (i.e. cars vs. trucks), emitter category (i.e.
high emitter vs normal emitter), average cycle speed, and facility cycle. For the purposes of this
analysis, a factor is considered significant if it is below the 0.05 significance level. Also, for
each pollutant both linear space and log space fits were investigated. For each pollutant, the best
fit was chosen. A more detailed analysis to determine the appropriate stratifications given
sample size and technical merit followed this initial screening. A discussion of this investigation
follows for each pollutant.
5.1 NMHC
The temperature corrections will be modified to accommodate the start/running split new to MOBILE6, but the
base corrections will not change. The start/running split has not be developed, so for this analysis the MOBILES
Bag corrections were applied.
MOBILES temperature correction factors can be found in "Compilation of Air Pollutant Emission Factors,
Volume II - Mobile Sources" (AP-42), Page H-24
-------
Based on the following analysis, MOBILE6 will contain two equations that will model the A/C
effect for NMHC for vehicles classified as normal emitters. There will be no A/C effect for
NMHC for vehicles classified as high emitters. The initial screening ANOVA results indicate
significance to the 0.05 level for A/C base and for emitter category. (See Appendix B, Section A;
Test of Between-Subjects Effects8) Therefore, a more in depth analysis was performed looking at
A/C effect for both normal emitters and high emitters separately.
5.1.1 Normal Emitters - Freeway, Arterial, Ramp
ANOVA was performed over a sample of vehicles that are classified as normal emitters for
NMHC, with A/C effect as the dependant versus A/C base, average speed, facility cycle, and
vehicle class. The results of this analysis indicate facility cycle as a significant factor. (See
Appendix B, Section B; Test of Between-Subjects Effects) When looking at the pairwise
comparisons for facility cycle it is evident that the Local cycle was significantly different from
the others (for the purpose of this analysis and throughout the report, Local cycle refers to the
Local facility cycle and the NYCC facility cycle combined).(See Appendix B, Section B;
Pairwise Comparison) Therefore the Local cycle was removed from the sample (to be analyzed
separately) and ANOVA was performed on the remaining sample. Again, A/C effect was the
dependant as a function of the following factors: A/C base, average speed, and vehicle class. The
results of this analysis indicated that there would be one correction factor for both LDVs and
LDTs, and that average speed was the only significant factor.(See Appendix B, Section C; Test
of Between-Subjects Effects) Therefore, an equation that will model the NMHC correction
factors for all normal emitters, on all facility cycles, excluding Local, was developed by fitting a
linear function to the sample by average speed.(See Appendix B, Section E; Parameter
Estimates)
A/C Effect = 0.001162*(Speed); R2= .044
Figure 1 in Appendix C show the predicted A/C effect, based on the above equation, versus the
original data for all vehicles on each facility cycle.
5.7.2 Normal Emitters - Local Cycle
Next, ANOVA was performed on the sample that contained only tests performed on the Local
cycle. For this analysis, A/C effect was looked at as a function of vehicle class, average speed
and A/C base. The results of this analysis indicated that there was significance to the 0.05 level
for vehicle class. (See Appendix B, Section F; Test of Between-Subjects Effects) When looking at
the pairwise comparisons for vehicle class, it showed the significance was between LDT1 and
LDT2, and between LDV and LDT2. (See Appendix B, Section F; Pairwise Comparison)
The corrected vehicle emissions with the A/C off (i.e. A/C Base) is referred to as NMHC_Off, CO_Off, and
NOx_Off in the ANOVA results. Also "Veh_Class" refers to LDV, LDT1 and LDT2, while "Class" refers to LDV
and LOT
7
-------
However, because the LDT2 results are based on a sample of only 3 trucks, it was decided that
the truck classes should not be split, but instead would be combined and re-analyzed. This re-
analysis shows there is no significant difference between LDVs and LDTs. It also indicated no
significance for average speed, but that there is significance to the 0.05 level for A/C base. (See
Appendix B, Section G; Test of Between-Subjects Effects) Therefore, the following equation
was developed to model the NMHC A/C effect for all normal emitting vehicles on the local
cycle: (See Appendix B, Section H; Parameter Estimates)
A/C Effects = 0.506 * (A/C Base); R2 = .127
Figure 2 in Appendix C show the predicted A/C effect, based on the above equation, versus the
actual data.
5.1.3 High Emitters - Freeway, Arterial, Ramp, Local
ANOVA results indicate that vehicle class and A/C base are significant. (See Appendix B,
Section I; Test of Between-Subjects Effects) As with normal emitters, the pairwise comparison
shows LDT2 are significantly different from LDT1 and LDVs. (See Appendix B, Section I;
Pairwise Comparison) Again, due to the small sample size for LDT2 (only 1 truck), the truck
classes were combined and re-analyzed. The results of this analysis did not indicate any
significant difference between LDVs and LDTs, but did indicate A/C base to be significant. (See
Appendix B, Section J; Test of Between-Subjects Effects) An equation to model high emitters
for NMHC was developed from this analysis. (See Appendix B, Section K; Parameter Estimates)
When the predicted A/C effect, based on this equation, was plotted versus the original data, the
graphs show the equation to be over predicting the A/C effect. (See Figures 3-16 in Appendix C)
The original data lies very near or below the zero gram/mile mark. This strongly indicates that
there is no A/C effect for NMHC high emitters. The technical basis for this observation is that it
is more likely that NMHC high emitters are operating with enrichment and/or very low catalyst
efficiency without air conditioning. There is less opportunity for emissions to increase
significantly when efficiency with the A/C on won't have the same relative impact. Therefore,
based on this observation, there will be no NMHC A/C effect for vehicles classified as NMHC
high emitters.
5.2 CO
Based on ANOVA results, there will be five equations in MOBILE6 used to model A/C effect
for CO. The initial screening ANOVA results indicate that the sample should be separated by
emission category and that the Local cycle is significantly different from the other cycles,
therefore warrantying separate analysis. (See Appendix D, Section A; Pairwise Comparison)
Also, the results indicate that the sample should be separated by vehicle class, LDV vs LDTs.
(See Appendix D, Section B; Test of Between-Subjects Effects) The following will describe the
analysis used to determine each equation for each subcategory.
-------
5.2.1 Normal Emitters - Freeway, Arterial, Ramp
The normal emitter samples for LDVs and LDTs, not tested on the local cycle, were analyzed
using ANOVA, looking at A/C effect as a function of A/C base and average speed. Average
speed and A/C base are significant factors for LDVs (See Appendix D, Section C; Test of
Between-Subjects Effects), while average speed is the only significant factor for LDTs. (See
Appendix D, Section D; Test of Between-Subjects Effects) The following equations were
developed by fitting a linear function through each sample based on the significant factors for
each.
Light-Duty Vehicle A/C Effect = 0.815*(A/C base) + 0.05272*(Speed); R2 = .255
Light-Duty Truck A/C Effect = 0.104*(Speed); R2 = .059
Figures 1 & 2 in Appendix E show the predicted A/C effect, based on the above equations,
versus the original data.
5.2.2 High Emitters - Freeway, Arterial, Ramp
In MOBILE6 there will be a CO A/C effect for vehicles classified as high emitters for CO, but
only for average speeds below nineteen miles per hour (19 mph). For average speeds above 19
mph, there will be no CO A/C effect for high emitters. This is based on analysis of a sample of
vehicles classified as high emitters for CO and excludes the vehicles' test on the Local cycle.
This section describes this analysis.
The initial screening ANOVA resulting in the development of an equation based on CO base and
average speed, that modeled all vehicle classes. (See Appendix D, Section E; Parameter
Estimates) When this equation was plotted against the data, the graphs showed the model to be
overestimating the A/C effect for cycles with an average speed greater than 19mph. The data
showed the A/C effect, for these cycles, to be near or below zero. Based on this observation, the
sample was split into two sets; cycles with an average speed below 19 mph and cycles with an
average speed above 19 mph. It was concluded that for average speed above 19 mph, there will
be no CO A/C effect for high emitters. The sample with average speed below 19 mph was re-
analyzed.
ANOVA was performed on the sample of high emitters with cycles having an average speed
below 19 mph. CO effect was the independent and vehicle class, CO base, and average speed
were the factors. The results of this analysis indicates there is no significant difference between
LDVs and LDTs, therefore, vehicle classes were combined for continued analysis. (See
Appendix D, Section F; Test of Between-Subjects Effects) Continued analysis indicates that CO
base is the only significant factor for high emitters on cycles with an average speed below 19
mph. (See Appendix D, Section G; Test of Between-Subjects Effects) Based on these results, an
equation was developed by fitting a linear function through the sample based on CO base. (See
Appendix D, Section H; Parameter Estimates) The following equation models all vehicles
-------
classified as CO high emitters on cycles with an average speed less than 19 mph:
CO Effect = 0.154 * (CO base); R2 = .831
Figures 3 in Appendix E show the predicted A/C effect, based on the above equation, versus the
original data.
5.2.3 Local Cycle
The first look ANOVA of vehicles tested only on the local cycle determined that A/C base is
significant and that there is a significant difference between LDT1 and LDT2. (See Appendix D,
Section I; Test of Between-Subjects Effects) Again, based on the fact that there were very few
LDT2 in the sample (only 4 trucks), LDT were not split. Continued analysis developed an
equation to model A/C effect for all LDVs and LDTs on the Local cycle. (See Appendix D,
Section J; Parameter Estimates) The predicted A/C effect was calculated based on this linear
equation and was plotted against the original data.(Figure 4, Appendix E) Although ANOVA
analysis did not characterize emitter classification as significant, the graph clearly indicates a
need to split by emitter classification.
5.2.3a Normal Emitters - Local cycle
ANOVA was performed on a sample containing only vehicles classified as normal emitters for
CO. CO effect was looked at as a function of vehicle class, average speed, and CO base. From
this analysis it was determined that average speed was not significant and that there is a
significant difference between LDT1 and LDT2 (See Appendix D, Section K; Pairwise
Comparison), but, due to the small sample size of LDT2, the two classes were combined. When
analyzing the sample again with the two classes combined, initially the results indicate a
significance between LDVs and LDTs. (See Appendix D, Section L; Test of Between-Subjects
Effects) A more in depth look at the pairwise comparisons shows that there is no significant
difference between LDVs and LDTs. (See Appendix D, Section L; Pairwise Comparison) Based
on the pairwise comparisons, vehicle class was not considered as a factor for the analysis. The
following equation was developed based on ANOVA results indicating CO base as a significant
factor for CO effect, for all normal emitting vehicles on the Local cycle. (See Appendix D,
Section M; Parameter Estimates)
CO Effect = 0.678 * (CO base); R2 = .217
Figures 5 in Appendix E show the predicted A/C effect, based on the above equation, versus the
original data.
5.2.3b High Emitters - Local Cycle
When looking at CO effect as a function of vehicle class, average speed, and CO base, ANOVA
10
-------
results indicate that vehicle class and average speed are not significant for high emitting vehicles
on the Local cycle. (See Appendix D, Sections N & O; Test of Between-Subjects Effects)
However, CO base is considered significant and was used to develop the following linear
equation that will model CO effect for all high emitting vehicles on the Local cycle. (See
Appendix D, Section P; Parameter Estimates)
CO Effect = 0.119 * (CO base); R2 = .852
Figures 6 in Appendix E show the predicted A/C effect, based on the above equation, versus the
original data.
5.3 NOx
There will be three equations in MOBILE6 used to model the A/C effect on NOx emissions for
three different strata. Unlike CO and NMHC, these equations will be in log space. This section
will describe the analysis used to develop these equations.
ANOVA was performed on the NOx data set with A/C effect as the independent variable and
A/C base, average speed, vehicle class, and facility cycle as the factors. The conclusion from this
analysis was that there was a significant difference between LDVs and LDTs and that the Ramp
cycle should be analyzed separately. (See Appendix F, Section A; Test of Between-Subjects
Effects & Pairwise Comparison, Ramp = #5)
5.3.1 LD V - Freeway, Arterial, Local
ANOVA was performed three separate times on a sample of LDVs with tests on all cycles
excluding the Ramp cycle. For each ANOVA, A/C effect was the independent variable. The
factors that were used varied for each analysis. The following are the combinations of factors
used for the three ANOVA analyses.
1) NOx base, Average Speed
2) Log (NOx base), Log (Average Speed)
3) Log (NOx base + 1), Log (Average Speed)
Of the three analyses, the one using the factors listed as number three above had the best fit. (See
Appendix F, Section B & C; Test of Between-Subjects Effects) The log function fits well
because it is able to capture the drop in A/C effect at the lower end of the base emissions seen in
the NOx sample. Log functions stabilize at the higher end of the base emissions, where a linear
function would continue to rise. NOx base + 1 is used so that the log of a base emission equal to
zero will be zero. Based on this analysis, log (NOx base + 1) was the only significant factor and
was used to develop an equation to model A/C effect for LDVs. The following is the general
equation form that was developed.
11
-------
A/CEffect = x*Log (NOx base +1)
Although the analysis results did not indicate average speed as a significant factor, it was decided
to investigate whether there might be an interaction between average speed and base emissions.
In order to do so, the data sample was divided into three separate speed bins: below 15 mph,
between 15 mph and 31 mph, and above 31 mph. These speed bins were chosen to represent
where the average speeds for the different cycles fall between. The equation form, noted above,
was modeled for these three speed bins. This analysis showed a trend where the "x" term
decreased as average speed increased. (See Appendix F, Section D; Test of Between-Subjects
Effects) In order to capture this effect in the equation form above, the following steps were
followed:
1) If A/C Effect = x*Log (NOx base + 1) and
2) If x = a + b (Log (Average Speed)), then
3) A/C Effect = [a + b (Log (Average Speed))] * Log (NOx base + 1)
= a Log (NOx base + 1) + b (Log (Average Speed) * Log (NOx base + 1))
When this equation form (3) was modeled for the three speed bins, both the "a" and "b" terms
were deemed significant. (See Appendix F, Section E; Test of Between-Subjects Effects) The
following equation was developed from this analysis to model NOx A/C effect for LDVs on all
freeway, arterial, and local cycles.
A/C Effect = (4.867 Log (NOx base + 1) - 2.296 (Log (Average Speed)) *
Log (NOx base + 1)); R2 = 0.612
Figure 1 in Appendix G show the predicted A/C effect, based on the above equation, versus the
original data.
5.3.2 LDT- Freeway, Arterial, Local
The analysis use for LDV, described above, was also performed on the sample of LDT with tests
on all cycles, excluding the Ramp cycle. The analysis lead to similar conclusions for LDTs as
LDVs; a interactive effect between average speed and base emissions. (See Appendix F, Sections
F-H; Test of Between-Subjects Effects & Parameter Estimates) The following equation was
developed from this analysis to model NOx A/C effect for LDTs on all freeway, arterial, and
local cycles.
A/C Effect = (1.93 Log (NOx base + 1) - 0.769 (Log (Average Speed)) *
Log (NOx base + 1)); R2 = 0.371
Figures 2 in Appendix G show the predicted A/C effect, based on the above equation, versus the
original data.
12
-------
5.3.3 Ramp Cycle
ANOVA was performed on a sample of LDV and LDT with test on the Ramp cycle. For this
analysis, A/C effect was looked at as a function of vehicle class and the log of NOx base + 1.
Results indicate that there is no significant difference between LDVs and LDTs tested on the
ramp cycle. (See Appendix F, Section I; Test of Between-Subjects Effects) Also, this analysis
concluded that the log of NOx base + 1 is significant. Therefore the following equation was
developed to model A/C effect for both LDVs and LDTs on the Ramp cycle. (See Appendix F,
Section J; Test of Between-Subjects Effects)
A/C Effect = 0.655 * Log(NOx base + 1); R2 = 0.342
Figures 3 in Appendix G show the predicted A/C effect, based on the above equation, versus the
original data.
6.0 VALIDATION
Results of this analysis were compared with initial results from the Coordinating Research
Council (CRC) project E-37 investigating air conditioning emissions.9 The tests that were
performed for this project were held in an environmental chamber, under several different
ambient conditions. The models that are described in this report, EPA Report Number
M6.ACE.002, "Air Conditioning Correction Factors," were used in conjunction with CRC's
data. The purpose of doing this is to see how the models that were developed with data not from
an environmental chamber compare to the data from environmental chamber testing. Though
CRC performed tests under many different conditions, only the tests performed under conditions
similar to what EPA and ATL were trying to simulate were used for comparison. Therefore, data
from tests performed on the SCO3 driving cycle with the following conditions were used; 95 °F,
full solar load (850 watts/meter2), 100 grains water/ Ib. dry-air, with the A/C on and off. Figures
1 - 4 in Appendix H show the comparison for several of the models. Considering the variation in
CRC's data, the MOBILE6 model approach performs well on average.
7.0 START CORRECTION FACTORS
A primary change between MOBILE6 and MOBILES is the separation of FTP-based emissions
into start and running components. This change draws a distinction between start emissions and
emissions over start driving. Running emissions will represent not only emissions over warmed-
up operation, but the baseline emissions inherent in start driving; start emissions will be defined
as the incremental emission increase above this baseline which occurs during start driving. Total
emissions over start driving, therefore, will be comprised of the baseline running emissions plus
9 Draft Report, CRC Project E-37, "Effects of Air Conditioning on Regulated Emissions for In-Use Vehicles,"
Coordinating Research Council, Inc.
13
-------
incremental start emissions. In terms of air conditioning correction factors, the running
correction factors developed in Section 5 will carry over to start driving to the extent that start
driving emissions are comprised of the baseline running component. The pertinent issue for start
air conditioning correction factors is therefore whether an A/C impact exists on the incremental
start component as well.
Data required to make this assessment based on the methodology used in the development of
base start and running emission factors10 were not gathered as part of the air conditioning test
program. An assessment was therefore made by analyzing the ratio for each pollutant over a cold
start ST01 run with the A/C on and off, shown for relevant stratifications in Table 1 of Appendix
I. The NOx and fuel consumption results indicate there is an increase over start driving due to air
conditioning, but smaller (by 13% for LDV's, 7% for LDT's) than the impact over running
operation at the average speed of the ST02 cycle (20.2 mph). It is presumed from this result that
the NOx ratio observed over ST01 is attributable solely to the baseline running component, with
no A/C-related increase occurring on the start increment. Based on this presumption, a NOx
correction factor for the incremental start component is not proposed for MOBILE6.
HC and CO results vary somewhat, particularly across emitter class. Cold start HC and CO
emissions are dominated by emissions incurred by startup enrichment. Under cold start
enrichment the air-fuel ratio will likely not change due to air conditioner operation and/or
increased engine load, so increased HC or CO emissions are not expected over the start
component. It is therefore proposed that no A/C correction factor be applied to the HC or CO
start components.
It is important to note that although air conditioning correction factors are not proposed for the
start components of any pollutant, air conditioning emissions over start driving will be estimated
by MOBILE6. Because the running correction factors are carried over to start driving, they will
be applied to the extent running emissions contribute to overall start emissions. This will be true
for all starts, including those following "intermediate" soak durations in which the engine and/or
catalyst are partially warmed up. For the most part, the contribution of running emissions (and
hence the influence of the running air conditioning correction factors) will become greater as the
soak duration shortens.
8.0 BENEFITS OF THE SFTP REQUIREMENT
Increasing attention to the importance of off-cycle emissions led to the development of a new
compliance procedure, known as the Supplemental Federal Test Procedure (SFTP). In addition
to "off-cycle" emissions, the SFTP addresses emissions which are generated with the air
conditioning on, which were also inadequately represented by the FTP. The SFTP requirements
grew out of the 1990 Clean Air Act Amendments, which instructed EPA to review the existing
10 This methodology referred to is the separation of FTP emissions into Start and Running components as described
in MOBILE6 Report No. M6.STE.002, "The Determination of Hot Running Emissions from FTP Bag Emissions"
14
-------
procedures and revise them in whatever ways were necessary to make them more representative
of actual in-use conditions. Developed in conjunction with the California Air Resources Board
(ARB) and auto manufacturers, the SFTP requirement adds two additional certification cycles,
and tailpipe standards associated with those cycles, to impose control of off-cycle (US06 cycle)
and air conditioning emissions (SC03 cycle). The US06 is run with the vehicle in the hot
stabilized condition; that is, with the vehicle fully warmed up to insure that the engine and
catalytic converter have reached typical operating temperatures. The SC03 follows a 10-minute
soak and is run with vehicle air conditioning (A/C) in operation or with an appropriate simulation
of air-conditioning operation.
The assigned benefits of the SFTP rule will depend on whether a vehicle is a Tier 1 vehicle or a
LEV. EPA and ARB promulgated separate requirements applying to these standard levels, and
hence the benefits resulting from the rule must take into account the relative stringency of the
EPA and ARB rules. Under NLEV, the Tier 1 rule will only apply to LDTs above 6000 pounds
(LDT3s and LDT4s), which phase in to the SFTP requirement at 40 percent in 2002, 80 percent
in 2003, and 100 percent in 2004.l These trucks will be allowed to certify to the Tier 1 SFTP
standards until they begin phasing into the Tier 2 final standards in 2008, at which point they will
be required to comply with the SFTP provisions under the Tier 2 rule discussed below.
For Tier 1 and interim Tier 2 LDT3s and LDT4s, the benefits derived in EPA's SFTP final
rulemaking shown in Appendix J, Table 1 will be used directly in MOBILE6 (Post-SFTP CO air
conditioning emissions are a special case, as discussed below). The percent reductions shown for
the SFTP rule will be applied directly to the off-cycle adjustment to generate final off-cycle
adjustments for SFTP-compliant vehicles.
A detailed derivation of these benefits are contained in the SFTP final rulemaking.2 Because
vehicles complying with the SFTP are just starting to enter the market, an assessment of SFTP
benefit on the in-use fleet is not yet possible. We therefore consider the approach used in the
EPA SFTP rule to be the best available.
Under NLEV, the ARB rule will apply to LEV LDVs and LDTs under 6,000 pounds (LOT 1/2).
The ARB rule contains NOx and HC certification standards which differ from EPA's both in
terms of the relative stringency over the US06 and SC03 cycles, and the mileage at which a
vehicle is required to show compliance. The percent reductions derived for EPA's Tier 1
ruletherefore cannot be applied directly to vehicles complying with the ARB standards.
LEV SFTP benefits for HC are estimated to be 100 percent. This is because ARB has required
the elimination of "commanded enrichment" when the air conditioner is used, which we expect
will eradicate excess HC emissions due to air conditioning usage. Although this same provision
will reduce CO as well, we are setting the post-SFTP emission level so that CO emissions with
the air conditioner on are higher than without the air conditioner off by the amount of additional
fuel consumed. This reflects the fact that although we expect excess CO emission resulting from
commanded enrichment to be eliminated, the SFTP does not address the unavoidable load (and
15
-------
hence fuel consumption) increase that results from air conditioner usage. An analysis presented
in the draft version of this report (published in March 1998) estimated the percentage increase in
fuel consumption with the air conditioning on as a function of speed; these equations were
adopted directly for calculating a multiplicative adjustment which, when applied to running CO
emissions without air conditioning, will result in post-SFTP CO emissions accounting for the
"full-usage" air conditioning effect. These equations are as follows:
Post-SFTP CO Correction Factor (LDV/LDT1) = 1.34 -0.006134(speed)+0.000053(speed)2
Post-SFTP CO Correction Factor (LDT2/3/4) = 1.27 -0.004939(speed)+0.000048(speed)2
For estimating post-SFTP NOx air conditioning emissions, we developed a methodology which
estimated the percent reductions in NOx for the ARB standards on LEVs based on the EPA Tier
1 benefits presented in Table 1. This methodology required an assessment of the relative
stringency of the EPA and ARB SFTP standards compared to their respective FTP standard.
Several factors added complication to this analysis: first, the ARB standards are applicable at
4,000 miles whereas the EPA standards are applicable at 50,000 miles and full useful life
(100/120K miles); second, the SFTP standards are expressed at NMHC+NOx, while MOBILE
treats these pollutants separately. Third, the SFTP standards are based on operation when the
vehicle is warmed-up, necessitating that the warmed-up component of the FTP be extracted in
order to performing comparisons with the SFTP standards. An analytical step was required to
address each of these factors.
Reductions in air conditioning emissions due to ARB's LEV SFTP standards for NOx were
estimated through a determination of the stringency of the ARB and EPA SC03 standards. The
stringency of the ARB and EPA standards is characterized by how well they control air
conditioning emissions for LEVs and Tier 1 vehicles, respectively. This stringency was
determined through a direct comparison between these standards and emissions over the FTP.
The basis for this determination was a comparison between the SC03 standards and an estimation
of "running certification levels" (i.e. the running component of FTP certification levels)
calculated for Tier 1 vehicles and LEVs, according to the following steps, shown in Table J-2:
1) Average certification emissions for model year 1999 LDVs and LDTs were generated
from EPA's CFEIS database at 4,000 miles for LEVs and 50,000 miles for Tier 1 (Row
1). The certification database used to generate these averages are provided with this
report.
2) "Running certification levels" were estimated for Tier 1 and LEV by multiplying the
certification levels from Step 2 by the appropriate running BER fractions discussed in
Draft Final MOBILE6 Report M6.EXH.007 (December 1999); 0.90 for NOx and 0.23 for
HC. The FTP certification levels and the derived "running certification levels" are shown
in Row 2.
3) NMHC+NOx US06 and SC03 standards were split into separate NMHC and NOx
16
-------
standards by applying a split of 0.14/0.86 for NMHC/NOx, derived from the development
of EPA's Tier 1 standards, and discussed in EPA's final SFTP rule (Rows 3 and 4).
4) A ratio of the resulting 50,000 mile SFTP NMHC and NOx "standards" from Step 3
and the running certification levels from Step 2 were calculated for both the Tier 1 (EPA)
and LEV(ARB) requirements for US06 (Row 5). The ratio (R) represents the magnitude
of increase allowed between the FTP and US06 cycles, and hence represents the
stringency of the SFTP standard relative to the FTP standards.
5) The stringency of the ARE standards relative to the EPA standards were estimated by
comparing the value of R calculated in Step 4, according to the following equation (Row
6):
Additional Stringency of ARE Standards (%) = [(REPA -1) - (RARE-!)] / (REPA -1)
The additional stringency represents the additional off-cycle emissions which would be
eliminated above and beyond the reductions under the Tier 1 standards.
6) Benefits under the ARB rule were then derived by adjusting the Tier 1 benefits (Row 7)
from Table 7-1 according to the additional stringency contained in Step 5, according to the
following equation (Row 8):
ARB Benefit (%) = EPA Benefit + (Step 5) * (1 - EPA Benefit)
The resulting NOx SFTP benefits for LEVs are presented in Tables J-3. Full-usage air
conditioning correction factors which reflect the SFTP rule are calculated in MOBILE6 by first
estimating the additive NOx air conditioning increment without the SFTP using the methodology
presented in Section 5, and reducing this increment by the appropriate percent reduction shown
in Table J-3.
ACKNOWLEDGMENTS
Several individuals contributed considerable time and resources to gathering and analyzing the
data presented here. Carl Fulper, Carl Scarbro, Dave Boshenek and Manish Patel of OTAQ
designed and implemented the test program and developed the attendant data set. Steve Baldus
and Kevin Cullen of GM made GM's environmental chamber available and coordinated testing
at that facility.
17
-------
Appendix A:
Testing Vehicles, Cycles, and Correlation Results
-------
Table 1 - Vehicle Sample
Site
ATL
ATL
ATL
ATL
ATL
ATL
ATL
ATL
ATL
ATL
ATL
EPA
EPA
EPA
EPA
EPA
EPA
EPA
EPA
EPA
EPA
EPA
EPA
EPA
EPA
EPA
EPA
EPA
EPA
EPA
EPA
EPA
EPA
EPA
EPA
EPA
BOTH
Year
91
91
91
91
91
93
93
93
93
90
93
92
96
92
92
94
96
92
96
92
96
90
90
94
96
94
94
94
92
94
90
92
96
94
96
90
96
Vehicle
CHEVROLET CAVALIER
FORD ECONOLINE 150
FORD ESCORT
PLYMOUTH VOYAGER
CHEVROLET ASTRO VAN
CHEVROLET CORSICA
CHEVROLET S 10
TOYOTA CAMRY
HONDA ACCORD
NISSAN MAXIMA
EAGLE SUMMIT
TOYOTA COROLLA
HONDA ACCORD
SATURN SL
CHEVROLET BERETTA
FORD F150
FORD F150
MAZDA PROTEGE
CHEVROLET LUMINA
CHEVROLET CAVALIER
FORD RANGER
JEEP CHEROKEE
CHEVROLET SUBURBAN
CHRYSLER LHS
HONDA CIVIC
CHEVROLET ASTRO VAN
SATURN SL
HYUNDAI ELAN
CHEVROLET LUMINA VAN
FORD ESCORT
PLYMOUTH VOYAGER
CHEVROLET LUMINA
FORD EXPLORER
PONTIAC TRANSPORT
TOYOTA CAMRY
DODGE DYNASTY
PONTIAC GRAND PRIX
Class
LDV
LDT2
LDV
LDT1
LDT1
LDV
LDT1
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDV
LDT2
LDT2
LDV
LDV
LDV
LDT1
LDT1
LDT2
LDV
LDV
LDT1
LDV
LDV
LDT1
LDV
LDT1
LDV
LDT1
LDT1
LDV
LDV
LDV
Fuel
TBI
PFI
PFI
TBI
TBI
PFI
TBI
PFI
PFI
PFI
PFI
PFI
PFI
TBI
PFI
PFI
PFI
PFI
PFI
PFI
PFI
PFI
TBI
PFI
PFI
PFI
PFI
PFI
PFI
PFI
PFI
PFI
PFI
PFI
PFI
PFI
PFI
Std
TierO
TierO
TierO
TierO
TierO
TierO
TierO
TierO
TierO
TierO
TierO
TierO
Tierl
TierO
TierO
TierO
Tierl
TierO
Tierl
TierO
Tierl
TierO
TierO
TierO
Tierl
TierO
TierO
TierO
TierO
Tierl
TierO
TierO
Tierl
Tierl
Tierl
TierO
Tierl
Emit*
N/N/N
H/H/N
H/N/H
N/N/N
H/N/N
N/N/N
N/N/N
N/N/N
N/N/N
N/N/N
N/N/N
N/N/N
N/N/N
N/N/N
H/H/N
N/N/N
N/N/N
N/N/N
N/N/N
N/N/N
N/N/N
H/H/H
N/N/N
N/N/N
N/N/N
N/N/N
N/N/N
N/N/N
N/N/N
N/N/N
N/N/N
N/N/N
N/N/N
N/N/N
N/N/N
N/N/N
N/N/N
*HC/CO/NOx
-------
Table 2 - Test Cycles
Cycle
NYCC
LOCL
ARTE
ARTC
ARTA
FWYG
FWYF
FWYE
FWYD
FWAC
FWHS
RAMP
AREA
LA92
ST01
Description
New York City Cycle
Local Roadways
Arterial Level Of Service E-F
Arterial LOS C-D
Arterial LOS A-B
Freeway LOS G
Freeway LOS F
Freeway LOS E
Freeway LOS D
Freeway LOS A-C
Freeway High Speed
Freeway Ramp
Non-Freeway Area-Wide
California "Unified" Cycle
Start Cycle
Distance
(miles)
1.18
7.24
1.62
3.35
5.06
1.42
2.28
3.85
5.95
8.54
10.70
2.56
7.25
9.81
1.39
Average
Speed
(mph)
7.1
12.9
11.6
19.2
24.7
13.1
18.6
30.5
52.9
59.7
63.2
34.7
19.4
24.6
20.2
Max
Speed
(mph)
27.7
38.3
39.9
49.5
58.9
35.7
49.9
63.0
70.6
73.1
74.7
60.2
52.3
67.2
41.0
Max
Accel
(mph/sec)
6.0
3.7
5.8
5.7
5.0
3.8
6.9
5.3
2.3
3.4
2.7
5.7
6.4
6.9
5.1
-------
Table 3 - Correlation Vehicle Emission Results (g/mi)
NYCC
LA92
FWHS
ARTC
ATL
EPA
GM
ATL
EPA
GM
ATL
EPA
GM
ATL
EPA
GM
NMHC
Off
0.07
0.07
0.06
0.04
0.02
0.04
0.08
0.05
0.02
0.04
0.03
0.01
On
0.67
0.07
0.07
0.10
0.02
0.03
1.32
1.33
0.03
0.04
0.05
0.03
Ratio
9.39
1.03
1.25
2.76
0.94
0.61
15.72
24.23
1.75
1.11
1.48
2.83
CO
Off
1.36
0.98
0.60
0.39
0.15
0.54
2.82
4.63
0.82
1.70
1.41
0.33
On
4.34
3.32
7.71
7.92
0.53
1.83
100.04
112.24
2.41
1.41
3.00
2.99
Ratio
3.19
3.39
12.94
20.35
3.52
3.37
35.47
24.26
2.94
0.83
2.13
9.15
NOx
Off
0.03
0.11
0.11
0.38
0.38
0.67
0.25
0.22
0.30
0.11
0.13
0.19
On
0.33
0.25
0.28
0.21
0.51
1.04
0.03
0.00
0.62
0.16
0.28
0.37
Ratio
10.20
2.15
2.52
0.55
1.34
1.55
0.12
0.01
2.05
1.48
2.14
1.94
Carbon
Off
214.1
208.6
217.8
115.3
110.1
216.8
88.2
82.7
82.0
120.0
116.7
120.2
On
281.6
275.4
283.5
138.7
133.0
267.2
120.7
120.0
85.9
145.3
144.4
144.5
Ratio
1.31
1.32
1.30
1.20
1.21
1.23
1.37
1.45
1.05
1.21
1.24
1.20
Table 4 - Correlation Vehicle Compressor Behavior
NYCC
LA92
FWHS
ARTC
ATL
EPA
GM
ATL
EPA
GM
ATL
EPA
GM
ATL
EPA
GM
Compressor
Fraction
1.00
0.99
0.97
0.99
0.97
0.99
1.00
0.99
1.02
1.00
0.98
0.99
Average High
Pressure (lb/in2 )
311.5
306.4
320.9
334.2
339.4
312.1
361.1
367.3
264.8
310.7
315.3
310.8
Average Low
Pressure (lb/in2 )
49.7
58.1
44.5
48.2
57.9
40.3
43.7
50.3
34.3
46.4
54.7
39.0
-------
Appendix B:
NMHC ANOVA Results
-------
SECTION A
NMHC Univariate Analysis of Variance
LDV & LOT
All Emitter Categories
All Cycles
Between-Subjects Factors
CYCLE ID
CLASS
ART
FWY
LA92
LOCAL
RAMP
LOT
LDV
N
148
221
37
74
37
182
335
Tests of Between-Subjects Effects
Dependent Variable: NMHC_DIFF
Source
Model
CYCLEJD
NMHC_EMT
AVG_SPD
NMHCJDFF
CLASS
Error
Total
Type III
Sum of
Squares
156.5763
1.486
51.228
.213
138.427
.145
126.196
282.771
df
9
4
1
1
1
1
508
517
Mean
Square
17.397
.372
51.228
.213
138.427
.145
.248
F
70.033
1.496
206.216
.859
557.236
.585
Sig.
.000
.202
.000
.355
.000
.445
a. R Squared = .554 (Adjusted R Squared = .546)
Estimated Marginal Means
1. CYCLEJD
Estimates
Dependent Variable: NMHC_DIFF
CYCLE ID
ART
FWY
LA92
LOCAL
RAMP
Mean
9.934E-03a
7.662E-023
1.276E-023
.161a
-1.808E-023
Std. Error
.044
.039
.082
.065
.083
95% Confidence Interval
Lower
Bound
-7.706E-02
-6.782E-04
-.149
3.256E-02
-.181
Upper
Bound
9.692E-02
.154
.175
.290
.145
a- Evaluated at covariates appeared in the model: nmhc emit cat
= 7.930E-02, AVG_SPD = 27.95300, NMHCJDFF = .58475.
Page 1
-------
Pairwise Comparisons
Dependent Variable: NMHC_DIFF
(1) CYCLE ID (J) CYCLE ID
ART FWY
LA92
LOCAL
RAMP
FWY ART
LA92
LOCAL
RAMP
LA92 ART
FWY
LOCAL
RAMP
LOCAL ART
FWY
LA92
RAMP
RAMP ART
FWY
LA92
LOCAL
Mean
Difference
d-J)
-6.668E-02
-2.823E-03
-.151*
2.801 E-02
6.668E-02
6.386E-02
-8.463E-02
9.470E-02
2.823E-03
-6.386E-02
-.148
3.084E-02
.151*
8.463E-02
.148
.179
-2.801 E-02
-9.470E-02
-3.084E-02
-.179
Std. Error
.063
.092
.073
.095
.063
.092
.083
.089
.092
.092
.103
.117
.073
.083
.103
.108
.095
.089
.117
.108
Sig.a
.292
.976
.037
.769
.292
.488
.309
.287
.976
.488
.151
.792
.037
.309
.151
.098
.769
.287
.792
.098
95% Confidence Interval
for Difference3
Lower
Bound
-.191
-.184
-.294
-.159
-5.739E-02
-.117
-.248
-8.002E-02
-.178
-.244
-.352
-.199
8.819E-03
-7.851 E-02
-5.453E-02
-3.350E-02
-.215
-.269
-.261
-.392
Upper
Bound
5.739E-02
.178
-8.819E-03
.215
.191
.244
7.851 E-02
.269
.184
.117
5.453E-02
.261
.294
.248
.352
.392
.159
8.002E-02
.199
3.350E-02
Based on estimated marginal means
* The mean difference is significant at the .05 level.
a. Adjustment for multiple comparisons: Least Significant Difference (equivalent to no adjustments).
Univariate Tests
Dependent Variable: NMHC_DIFF
Contrast
Error
Sum of
Squares
1.486
126.196
df
4
508
Mean
Square
.372
.248
F
1.496
Sig.
.202
The F tests the effect of CYCLEJD. This test is based on the linearly independent
pairwise comparisons among the estimated marginal means.
2. CLASS
Page 2
-------
Estimates
Dependent Variable: NMHC_DIFF
CLASS
LOT
LDV
Mean
3.016E-023
6.683E-023
Std. Error
.042
.033
95% Confidence Interval
Lower
Bound
-5.227E-02
2.050E-03
Upper
Bound
.113
.132
a- Evaluated at covariates appeared in the model: nmhc emit cat
= 7.930E-02, AVG_SPD = 27.95300, NMHCJDFF = .58475.
Pairwise Comparisons
Dependent Variable: NMHC_DIFF
(I) CLASS (J) CLASS
LOT LDV
LDV LOT
Mean
Difference
d-J)
-3.667E-02
3.667E-02
Std. Error
.048
.048
Sig.a
.445
.445
95% Confidence Interval
for Difference3
Lower
Bound
-.131
-5.751 E-02
Upper
Bound
5.751 E-02
.131
Based on estimated marginal means
a. Adjustment for multiple comparisons: Least Significant Difference (equivalent to no adjustments).
Univariate Tests
Dependent Variable: NMHC_DIFF
Contrast
Error
Sum of
Squares
.145
126.196
df
1
508
Mean
Square
.145
.248
F
.585
Sig.
.445
The F tests the effect of CLASS. This test is based on the linearly independent
pairwise comparisons among the estimated marginal means.
SECTION B
NMHC Univariate Analysis of Variance
LDV & LOT
All cycles
Normal emitter only
PageS
-------
Between-Subjects Factors
CYCLE ID
CLASS
ART
FWY
LA92
LOCAL
RAMP
LOT
LDV
N
128
192
32
64
32
140
308
Tests of Between-Subjects Effects
Dependent Variable: NMHC_DIFF
Source
Model
NMHC OFF
AVG SPD
CYCLE ID
CLASS
Error
Total
Type III
Sum of
Squares
1.943a
7.789E-05
7.479E-02
.732
4.724E-02
22.638
24.582
df
8
1
1
4
1
440
448
Mean
Square
.243
7.789E-05
7.479E-02
.183
4.724E-02
5.145E-02
F
4.722
.002
1.454
3.555
.918
Sig.
.000
.969
.229
.007
.338
a. R Squared = .079 (Adjusted R Squared = .062)
Estimated Marginal Means
1.CYCLEJD
Estimates
Dependent Variable: NMHC_DIFF
CYCLE ID
ART
FWY
LA92
LOCAL
RAMP
Mean
2.136E-023
3.242E-023
1.977E-023
.152a
4.676E-023
Std. Error
.022
.019
.040
.032
.041
95% Confidence Interval
Lower
Bound
-2.156E-02
-5.649E-03
-5.973E-02
8.809E-02
-3.360E-02
Upper
Bound
6.429E-02
7.050E-02
9.927E-02
.216
.127
a. Evaluated at covariates appeared in the model: NMHCJDFF = .12027, AVG_SPD = 28.01429.
Page 4
-------
Pairwise Comparisons
Dependent Variable: NMHC_DIFF
(1) CYCLE ID (J) CYCLE ID
ART FWY
LA92
LOCAL
RAMP
FWY ART
LA92
LOCAL
RAMP
LA92 ART
FWY
LOCAL
RAMP
LOCAL ART
FWY
LA92
RAMP
RAMP ART
FWY
LA92
LOCAL
Mean
Difference
d-J)
-1.106E-02
1.595E-03
-.130*
-2.540E-02
1.106E-02
1.266E-02
-.119*
-1.433E-02
-1.595E-03
-1.266E-02
-.132*
-2.699E-02
.130*
.119*
.132*
.105*
2.540E-02
1.433E-02
2.699E-02
-.105*
Std. Error
.031
.045
.036
.047
.031
.045
.041
.044
.045
.045
.051
.058
.036
.041
.051
.053
.047
.044
.058
.053
Sig.a
.721
.972
.000
.588
.721
.779
.004
.743
.972
.779
.010
.639
.000
.004
.010
.048
.588
.743
.639
.048
95% Confidence Interval
for Difference3
Lower
Bound
-7.180E-02
-8.701 E-02
-.201
-.117
-4.967E-02
-7.578E-02
-.200
-.100
-9.020E-02
-.101
-.232
-.140
6.017E-02
3.919E-02
3.219E-02
9.523E-04
-6.670E-02
-7. 161 E-02
-8.610E-02
-.209
Upper
Bound
4.967E-02
9.020E-02
-6.017E-02
6.670E-02
7.180E-02
.101
-3.919E-02
7. 161 E-02
8.701 E-02
7.578E-02
-3.219E-02
8.610E-02
.201
.200
.232
.209
.117
.100
.140
-9.523E-04
Based on estimated marginal means
* The mean difference is significant at the .05 level.
a. Adjustment for multiple comparisons: Least Significant Difference (equivalent to no adjustments).
Univariate Tests
Dependent Variable: NMHC_DIFF
Contrast
Error
Sum of
Squares
.732
22.638
df
4
440
Mean
Square
.183
5.145E-02
F
3.555
Sig.
.007
The F tests the effect of CYCLEJD. This test is based on the linearly independent
pairwise comparisons among the estimated marginal means.
2. CLASS
Page 5
-------
Estimates
Dependent Variable: NMHC_DIFF
CLASS
LOT
LDV
Mean
4.328E-023
6.557E-023
Std. Error
.021
.016
95% Confidence Interval
Lower
Bound
1.576E-03
3.488E-02
Upper
Bound
8.499E-02
9.626E-02
a. Evaluated at covariates appeared in the model: NMHCJDFF = .12027, AVG_SPD = 28.01429.
Pairwise Comparisons
Dependent Variable: NMHC_DIFF
(I) CLASS (J) CLASS
LOT LDV
LDV LOT
Mean
Difference
d-J)
-2.229E-02
2.229E-02
Std. Error
.023
.023
Sig.a
.338
.338
95% Confidence Interval
for Difference3
Lower
Bound
-6.800E-02
-2.343E-02
Upper
Bound
2.343E-02
6.800E-02
Based on estimated marginal means
a. Adjustment for multiple comparisons: Least Significant Difference (equivalent to no adjustments).
Univariate Tests
Dependent Variable: NMHC_DIFF
Contrast
Error
Sum of
Squares
4.724E-02
22.638
df
1
440
Mean
Square
4.724E-02
5.145E-02
F
.918
Sig.
.338
The F tests the effect of CLASS. This test is based on the linearly independent
pairwise comparisons among the estimated marginal means.
SECTION C
NMHC Univariate Analysis of Variance
No Local Cycle
LDV and LOT
Normal Emitters
Between-Subjects Factors
CLASS LOT
LDV
N
120
264
Page 6
-------
Tests of Between-Subjects Effects
Dependent Variable: NMHC_DIFF
Source
Model
NMHC OFF
AVG SPD
CLASS
Error
Total
Type III
Sum of
Squares
.870a
.125
.113
.109
13.902
14.772
df
4
1
1
2
380
384
Mean
Square
.218
.125
.113
5.468E-02
3.658E-02
F
5.947
3.412
3.079
1.495
Sig.
.000
.066
.080
.226
a. R Squared = .059 (Adjusted R Squared = .049)
Estimated Marginal Means
CLASS
Estimates
Dependent Variable: NMHC_DIFF
CLASS
LOT
LDV
Mean
1.546E-02a
4.539E-023
Std. Error
.018
.012
95% Confidence Interval
Lower
Bound
-1.897E-02
2.221 E-02
Upper
Bound
4.989E-02
6.856E-02
a. Evaluated at covariates appeared in the model: NMHCJDFF = .10894, AVG_SPD = 31.01667.
Pairwise Comparisons
Dependent Variable: NMHC_DIFF
(I) CLASS (J) CLASS
LOT LDV
LDV LOT
Mean
Difference
d-J)
-2.992E-02
2.992E-02
Std. Error
.021
.021
Sig.a
.158
.158
95% Confidence Interval
for Difference3
Lower
Bound
-7.150E-02
-1.166E-02
Upper
Bound
1.166E-02
7.150E-02
Based on estimated marginal means
a. Adjustment for multiple comparisons: Least Significant Difference (equivalent to no adjustments).
Univariate Tests
Dependent Variable: NMHC_DIFF
Contrast
Error
Sum of
Squares
7.325E-02
13.902
df
1
380
Mean
Square
7.325E-02
3.658E-02
F
2.002
Sig.
.158
The F tests the effect of CLASS. This test is based on the linearly independent
pairwise comparisons among the estimated marginal means.
Page 7
-------
SECTION D
NMHC Univariate Analysis of Variance
LOT and LDV together
Normal emitters
No Local Cycle
Tests of Between-Subjects Effects
Dependent Variable: NMHC_DIFF
Source
Model
NMHCJDFF
AVG_SPD
Error
Total
Type III
Sum of
Squares
.761a
.108
.736
14.011
14.772
df
2
1
1
382
384
Mean
Square
.380
.108
.736
3.668E-02
F
10.373
2.945
20.076
Sig.
.000
.087
.000
a. R Squared = .052 (Adjusted R Squared = .047)
SECTION E
NMHC Univariate Analysis of Variance
Normal emitters only
All Vehicle Classes
No Local Cycle
Tests of Between-Subjects Effects
Dependent Variable: NMHC_DIFF
Source
Model
AVG_SPD
Error
Total
Type III
Sum of
Squares
.653a
.653
14.119
14.772
df
1
1
383
384
Mean
Square
.653
.653
3.686E-02
F
17.710
17.710
Sig.
.000
.000
a. R Squared = .044 (Adjusted R Squared = .042)
Parameter Estimates
Dependent Variable: NMHC_DIFF
Parameter
AVG SPD
B
1.162E-03
Std. Error
.000
t
4.208
Sig.
.000
95% Confidence Interval
Lower
Bound
6.193E-04
Upper
Bound
1.705E-03
SECTION F
PageS
-------
NMHC Univariate Analysis of Variance
All Vehicle Classes
Normal Emitters only
Local Cycle only
Between-Subjects Factors
VEHCLASS LDT1
LDT2
LDV
N
14
6
44
Tests of Between-Subjects Effects
Dependent Variable: NMHC_DIFF
Source
Model
VEHCLASS
NMHC OFF
AVG SPD
Error
Total
Type III
Sum of
Squares
2.725a
1.201
4.981 E-02
2.129E-02
7.085
9.810
df
5
3
1
1
59
64
Mean
Square
.545
.400
4.981 E-02
2.129E-02
.120
F
4.538
3.333
.415
.177
Sig.
.001
.025
.522
.675
a. R Squared = .278 (Adjusted R Squared = .217)
Estimated Marginal Means
VEHCLASS
Estimates
Dependent Variable: NMHC_DIFF
VEHCLASS
LDT1
LDT2
LDV
Mean
-5.315E-03a
.547a
.128a
Std. Error
.093
.149
.053
95% Confidence Interval
Lower
Bound
-.191
.249
2.285E-02
Upper
Bound
.180
.844
.233
a. Evaluated at covariates appeared in the model: NMHCJDFF = .18830, AVG_SPD = 10.00000.
Page 9
-------
Pairwise Comparisons
Dependent Variable: NMHC_DIFF
(I)VEHCLASS (J)VEHCLASS
LDT1 LDT2
LDV
LDT2 LDT1
LDV
LDV LDT1
LDT2
Mean
Difference
d-J)
-.552*
-.133
.552*
.419*
.133
-.419*
Std. Error
.175
.106
.175
.159
.106
.159
Sig.a
.003
.215
.003
.011
.215
.011
95% Confidence Interval
for Difference3
Lower
Bound
-.903
-.346
.201
9.992E-02
-7.963E-02
-.737
Upper
Bound
-.201
7.963E-02
.903
.737
.346
-9.992E-02
Based on estimated marginal means
* The mean difference is significant at the .05 level.
a. Adjustment for multiple comparisons: Least Significant Difference (equivalent to no adjustments).
Univariate Tests
Dependent Variable: NMHC_DIFF
Contrast
Error
Sum of
Squares
1.194
7.085
df
2
59
Mean
Square
.597
.120
F
4.970
Sig.
.010
The F tests the effect of VEHCLASS. This test is based on the linearly
independent pairwise comparisons among the estimated marginal means.
SECTION G
NMHC Univariate Analysis of Variance
LDV & LOT
Local Cycle only
Normal Emitters only
Between-Subjects Factors
CLASS
LOT
LDV
N
20
44
Page 10
-------
Tests of Between-Subjects Effects
Dependent Variable: NMHC_DIFF
Source
Model
NMHC OFF
AVG SPD
CLASS
Error
Total
Type III
Sum of
Squares
1.535a
.282
6.065E-02
1.080E-02
8.275
9.810
df
4
1
1
2
60
64
Mean
Square
.384
.282
6.065E-02
5.398E-03
.138
F
2.782
2.043
.440
.039
Sig.
.035
.158
.510
.962
a. R Squared = .156 (Adjusted R Squared = .100)
SECTION G cont.
NMHC Univariate Analysis of Variance
Tests of Between-Subjects Effects
Dependent Variable: NMHC_DIFF
Source
Model
NMHC OFF
AVG SPD
Error
Total
Type III
Sum of
Squares
1.524a
.354
.276
8.286
9.810
df
2
1
1
62
64
Mean
Square
.762
.354
.276
.134
F
5.702
2.650
2.063
Sig.
.005
.109
.156
a. R Squared = .155 (Adjusted R Squared = .128)
SECTION H
NMHC Univariate Analysis of Variance
LDV and LOT
Normal Emitters Only
Local Cycle Only
Tests of Between-Subjects Effects
Dependent Variable: NMHC_DIFF
Source
Model
NMHC OFF
Error
Total
Type III
Sum of
Squares
1.248a
1.248
8.561
9.810
df
1
1
63
64
Mean
Square
1.248
1.248
.136
F
9.187
9.187
Sig.
.004
.004
a. R Squared = .127 (Adjusted R Squared = .113)
Page 11
-------
Parameter Estimates
Dependent Variable: NMHC_DIFF
Parameter
NMHC OFF
B
.506
Std. Error
.167
t
3.031
Sig.
.004
95% Confidence Interval
Lower
Bound
.172
Upper
Bound
.839
SECTION I
NMHC Univariate Analysis of Variance
High Emitter only
All Cycles
All Vehicle Classes
Between-Subjects Factors
VEHCLASS
CYCLE ID
LDT1
LDT2
LDV
ART
FWY
LA92
LOCAL
RAMP
N
28
14
27
20
29
5
10
5
Tests of Between-Subjects Effects
Dependent Variable: NMHC_DIFF
Source
Model
NMHC OFF
AVG SPD
VEHCLASS
CYCLEJD
Error
Total
Type III
Sum of
Squares
168.4893
100.729
1.098E-02
17.928
4.098
89.701
258.190
df
9
1
1
2
4
60
69
Mean
Square
18.721
100.729
1.098E-02
8.964
1.025
1.495
F
12.522
67.377
.007
5.996
.685
Sig.
.000
.000
.932
.004
.605
a. R Squared = .653 (Adjusted R Squared = .600)
Estimated Marginal Means
1. VEHCLASS
Page 12
-------
Estimates
Dependent Variable: NMHC_DIFF
VEHCLASS
LDT1
LDT2
LDV
Mean
.418a
-.936a
.189a
Std. Error
.272
.348
.284
95% Confidence Interval
Lower
Bound
-.125
-1.633
-.379
Upper
Bound
.962
-.240
.757
a. Evaluated at covariates appeared in the model: NMHCJDFF = 3.60049, AVG_SPD = 27.55507.
Pairwise Comparisons
Dependent Variable: NMHC_DIFF
(I) VEHCLASS (J) VEHCLASS
LDT1 LDT2
LDV
LDT2 LDT1
LDV
LDV LDT1
LDT2
Mean
Difference
d-J)
1.355*
.229
-1.355*
-1.125*
-.229
1.125*
Std. Error
.405
.379
.405
.423
.379
.423
Sig.a
.001
.548
.001
.010
.548
.010
95% Confidence Interval
for Difference3
Lower
Bound
.545
-.530
-2.164
-1.971
-.988
.280
Upper
Bound
2.164
.988
-.545
-.280
.530
1.971
Based on estimated marginal means
* The mean difference is significant at the .05 level.
a. Adjustment for multiple comparisons: Least Significant Difference (equivalent to no adjustments).
Univariate Tests
Dependent Variable: NMHC_DIFF
Contrast
Error
Sum of
Squares
17.928
89.701
df
2
60
Mean
Square
8.964
1.495
F
5.996
Sig.
.004
The F tests the effect of VEHCLASS. This test is based on the linearly
independent pairwise comparisons among the estimated marginal means.
2. CYCLE ID
Page 13
-------
Estimates
Dependent Variable: NMHC_DIFF
CYCLE ID
ART
FWY
LA92
LOCAL
RAMP
Mean
-.244a
.170a
-.259a
.263a
-.477a
Std. Error
.295
.266
.550
.440
.555
95% Confidence Interval
Lower
Bound
-.834
-.361
-1.360
-.617
-1.587
Upper
Bound
.345
.702
.841
1.142
.633
a. Evaluated at covariates appeared in the model: NMHCJDFF = 3.60049, AVG_SPD = 27.55507.
Pairwise Comparisons
Dependent Variable: NMHC_DIFF
(I) CYCLE ID (J) CYCLE ID
ART FWY
LA92
LOCAL
RAMP
FWY ART
LA92
LOCAL
RAMP
LA92 ART
FWY
LOCAL
RAMP
LOCAL ART
FWY
LA92
RAMP
RAMP ART
FWY
LA92
LOCAL
Mean
Difference
d-J)
-.415
1.498E-02
-.507
.232
.415
.430
-9.244E-02
.647
-1.498E-02
-.430
-.522
.218
.507
9.244E-02
.522
.740
-.232
-.647
-.218
-.740
Std. Error
.422
.615
.487
.637
.422
.614
.559
.594
.615
.614
.693
.782
.487
.559
.693
.727
.637
.594
.782
.727
Siga
.330
.981
.302
.716
.330
.486
.869
.281
.981
.486
.454
.782
.302
.869
.454
.313
.716
.281
.782
.313
95% Confidence Interval
for Difference3
Lower
Bound
-1.259
-1.215
-1.481
-1.041
-.429
-.798
-1.210
-.542
-1.245
-1.657
-1.909
-1.346
-.467
-1.025
-.865
-.714
-1.506
-1.836
-1.781
-2.193
Upper
Bound
.429
1.245
.467
1.506
1.259
1.657
1.025
1.836
1.215
.798
.865
1.781
1.481
1.210
1.909
2.193
1.041
.542
1.346
.714
Based on estimated marginal means
a. Adjustment for multiple comparisons: Least Significant Difference (equivalent to no adjustments).
Page 14
-------
Univariate Tests
Dependent Variable: NMHC_DIFF
Contrast
Error
Sum of
Squares
4.098
89.701
df
4
60
Mean
Square
1.025
1.495
F
.685
Sig.
.605
The F tests the effect of CYCLEJD. This test is based on the linearly independent
pairwise comparisons among the estimated marginal means.
SECTION J
NMHC Univariate Analysis of Variance
High Emitters Only
LDV & LOT
All Cycles
Between-Subjects Factors
CYCLEJD
CLASS
ART
FWY
LA92
LOCAL
RAMP
LOT
LDV
N
20
29
5
10
5
42
27
Tests of Between-Subjects Effects
Dependent Variable: NMHC_DIFF
Source
Model
NMHCJDFF
CYCLEJD
AVG_SPD
CLASS
Error
Total
Type III
Sum of
Squares
151.736a
90.932
4.211
5.203E-06
1.175
106.454
258.190
df
8
1
4
1
1
61
69
Mean
Square
18.967
90.932
1.053
5.203E-06
1.175
1.745
F
10.868
52.106
.603
.000
.673
Sig.
.000
.000
.662
.999
.415
a. R Squared = .588 (Adjusted R Squared = .534)
Estimated Marginal Means
1. CYCLE ID
Page 15
-------
Estimates
Dependent Variable: NMHC_DIFF
CYCLE ID
ART
FWY
LA92
LOCAL
RAMP
Mean
-4.725E-023
.358a
-7.017E-023
.488a
-.280a
Std. Error
.317
.284
.593
.477
.598
95% Confidence Interval
Lower
Bound
-.681
-.210
-1.256
-.466
-1.476
Upper
Bound
.586
.925
1.115
1.443
.915
a. Evaluated at covariates appeared in the model: NMHCJDFF = 3.60049, AVG_SPD = 27.55507.
Pairwise Comparisons
Dependent Variable: NMHC_DIFF
(I) CYCLE ID (J) CYCLE ID
ART FWY
LA92
LOCAL
RAMP
FWY ART
LA92
LOCAL
RAMP
LA92 ART
FWY
LOCAL
RAMP
LOCAL ART
FWY
LA92
RAMP
RAMP ART
FWY
LA92
LOCAL
Mean
Difference
d-J)
-.405
2.293E-02
-.536
.233
.405
.428
-.131
.638
-2.293E-02
-.428
-.559
.210
.536
.131
.559
.769
-.233
-.638
-.210
-.769
Std. Error
.456
.665
.526
.688
.456
.663
.604
.642
.665
.663
.749
.844
.526
.604
.749
.785
.688
.642
.844
.785
Siga
.378
.973
.313
.736
.378
.521
.829
.324
.973
.521
.459
.804
.313
.829
.459
.331
.736
.324
.804
.331
95% Confidence Interval
for Difference3
Lower
Bound
-1.316
-1.306
-1.588
-1.142
-.507
-.898
-1.338
-.646
-1.352
-1.753
-2.056
-1.478
-.516
-1.076
-.939
-.801
-1.609
-1.922
-1.899
-2.338
Upper
Bound
.507
1.352
.516
1.609
1.316
1.753
1.076
1.922
1.306
.898
.939
1.899
1.588
1.338
2.056
2.338
1.142
.646
1.478
.801
Based on estimated marginal means
a. Adjustment for multiple comparisons: Least Significant Difference (equivalent to no adjustments).
Page 16
-------
Univariate Tests
Dependent Variable: NMHC_DIFF
Contrast
Error
Sum of
Squares
4.211
106.454
df
4
61
Mean
Square
1.053
1.745
F
.603
Sig.
.662
The F tests the effect of CYCLEJD. This test is based on the linearly independent
pairwise comparisons among the estimated marginal means.
2. CLASS
Estimates
Dependent Variable: NMHC_DIFF
CLASS
LOT
LDV
Mean
-6.297E-023
.242a
Std. Error
.249
.306
95% Confidence Interval
Lower
Bound
-.561
-.370
Upper
Bound
.435
.854
a. Evaluated at covariates appeared in the model: NMHCJDFF = 3.60049, AVG_SPD = 27.55507.
Pairwise Comparisons
Dependent Variable: NMHC_DIFF
(I) CLASS (J) CLASS
LOT LDV
LDV LOT
Mean
Difference
d-J)
-.305
.305
Std. Error
.372
.372
Sig.a
.415
.415
95% Confidence Interval
for Difference3
Lower
Bound
-1.049
-.439
Upper
Bound
.439
1.049
Based on estimated marginal means
a. Adjustment for multiple comparisons: Least Significant Difference (equivalent to no adjustments).
Univariate Tests
Dependent Variable: NMHC_DIFF
Contrast
Error
Sum of
Squares
1.175
106.454
df
1
61
Mean
Square
1.175
1.745
F
.673
Sig.
.415
The F tests the effect of CLASS. This test is based on the linearly independent
pairwise comparisons among the estimated marginal means.
SECTION K
NMHC Univariate Analysis of Variance
Page 17
-------
High Emitters Only
All Vehicle classes
All Cycles
Tests of Between-Subjects Effects
Dependent Variable: NMHC_DIFF
Source
Model
NMHCJDFF
Error
Total
Type III
Sum of
Squares
93.686a
93.686
164.504
258.190
df
1
1
68
69
Mean
Square
93.686
93.686
2.419
F
38.727
38.727
Sig.
.000
.000
a. R Squared = .363 (Adjusted R Squared = .353)
Parameter Estimates
Dependent Variable: NMHC_DIFF
Parameter
NMHC OFF
B
.220
Std. Error
.035
t
6.223
Sig.
.000
95% Confidence Interval
Lower
Bound
.149
Upper
Bound
.290
Page 18
-------
Appendix C:
NMHC Graphs
-------
NMHC High Emitter Graphs
Figure 1
Normal Emitters- Freeway, Arterial, Ramp
U
O)
0
2.0-
1.0-
o.oi
-1.0
1 NMHC DATA
4 NMHC MODEL
-.2 0.0 .2 .4 .6
NMHC A/C Base
.8 1.0
-------
Figure 3
SCHED ID: ART-,
0)
1 2.0-
1 1.5-
ra
J=K 1.0-
"" .5-
0)
ti 0.0-
CJ
CJ
246
NMHC A/C Base
4 NMHC_MODEL
NMHC DATA
O)
CO
E
ra
s=
CT
U
O)
O
CJ
Figure 4
SCHED ID: ART-CD
2-
0-
-2
4 6 8
NMHC A/C Base
NMHC_MODEL
NMHC DATA
-------
Figure 5
SCHED ID: ART-EF
0)
en
E
ra
en
6-
4-
2-
0-
y ^
z -4
o
0)
LJ
CJ
4 NMHC_MODEL
NMHC DATA
] 2 4 6 8 1012
NMHC A/C Base
Figure 6
SCHED ID: FVY-AC
.4
ju
'I
en
E
ra
s=
CT -
u
O)
CJ
CJ
-.4-
NMHC_MODEL
NMHC DATA
.6
1.0 1.2 1.4
NMHC A/C Base
1.6
-------
Figure 7
SCHED ID: FVY-D
1.5
0)
1 1.0
ra
en
<-> .5
0)
CJ
CJ
0.0-
-.5
2345
NMHC A/C Base
4 NMHC_MODEL
NMHC DATA
Figure 8
SCHED ID: EVY-E
'.0
ju
'I
en
E
ra
en
u .5-
O)
CJ
0.0-
o -.5-
234567
NMHC A/C Base
NMHC_MODEL
NMHC DATA
-------
Figure 9
SCHED ID: FVY-F
0)
en
E
ra
s=
CT
o
0)
LJ
CJ
<
3-
2-
0-
-2
0246
NMHC A/C Base
Figure 1 0
SCHED ID: EVY-G
O)
CO
CT
O)
O
O
3-
2
-2
4 6 8 10
NMHC A/C Base
4 NMHC_MODEL
NMHC DATA
NMHC_MODEL
NMHC DATA
-------
Figure 1 1
SCHED ID: FVY-
1.5
0)
nj
o
0)
CJ
CJ
0.0-
-1.0
23456
NMHC A/C Base
Figure 1 2
SCHED ID: LA92
2 4 6 8 10
NMHC A/C Base
4 NMHC_MODEL
NMHC DATA
,
0
0)
Cl
LJ
CJ
CJ
z
2.0-
1.5-
1.0-
.5-
0.0-
-.5-
-1.0-
-1.5
A
A
A
"
u
A NMHC_MODEL
NMHC DATA
-------
Figure 1 3
SCHED ID: LOCAL
0)
tn
CT
0)
2-
4 NMHC_MODEL
NMHC DATA
2 4 6 8 10
NMHC A/C Base
Figure 1 4
SCHED ID: NONEVY
O)
'I
CO
E
ra
i
CT
U
O)
Q
C|
LJ
CJ
^
1
Z
o.u -
2.5-
2.0-
1.5-
1.0-
.5-
0.0-
-.5-
-1.0
A
^
^A
f f
A NMHC_MODEL
NMHC DATA
4 6 8
NMHC A/C Base
-------
Figure 1 5
SCHED ID: NYCC
1 Ł-
s.
0)
E 10-
^^
en
E
ra °
CT
~" 6-
=1=
o
0)
r 4"
jj
=> ? -
-------
Appendix D:
CO ANOVA Results
-------
SECTION A
CO Univariate Analysis of Variance
All Veh. Classes
All Cycles
All Emitter Categories
Between-Subjects Factors
CYCLE ID
VEHCLASS
ART
FWY
LA92
LOCAL
RAMP
LDT1
LDT2
LDV
N
148
221
37
74
37
126
56
335
Tests of Between-Subjects Effects
Dependent Variable: CO_DIFF
Source
Corrected Model
Intercept
CYCLEJD
COJDFF
AVG_SPD
COEM_CAT
VEHCLASS
Error
Total
Corrected Total
Type III
Sum of
Squares
7114.2303
797.635
1250.449
4653.597
93.187
2547.696
353.086
93896.679
111971.495
101010.908
df
9
1
4
1
1
1
2
507
517
516
Mean
Square
790.470
797.635
312.612
4653.597
93.187
2547.696
176.543
185.201
F
4.268
4.307
1.688
25.127
.503
13.756
.953
Sig.
.000
.038
.151
.000
.478
.000
.386
a. R Squared = .070 (Adjusted R Squared = .054)
Estimated Marginal Means
1. CYCLE ID
Page 1
-------
Estimates
Dependent Variable: CO_DIFF
CYCLE ID
ART
FWY
LA92
LOCAL
RAMP
Mean
3.135a
4.01 8a
2.532a
7.962a
4.820a
Std. Error
1.298
1.174
2.300
1.846
2.313
95% Confidence Interval
Lower
Bound
.586
1.712
-1.986
4.335
.277
Upper
Bound
5.684
6.325
7.049
11.589
9.364
a- Evaluated at covariates appeared in the model: CO_OFF =
16.77908, AVG_SPD = 27.95300, COEM_CAT = 7.930E-02.
Pairwise Comparisons
Dependent Variable: CO_DIFF
(I) CYCLE ID (J) CYCLE ID
ART FWY
LA92
LOCAL
RAMP
FWY ART
LA92
LOCAL
RAMP
LA92 ART
FWY
LOCAL
RAMP
LOCAL ART
FWY
LA92
RAMP
RAMP ART
FWY
LA92
LOCAL
Mean
Difference
d-J)
-.883
.604
-4.827*
-1.685
.883
1.487
-3.944
-.802
-.604
-1.487
-5.430
-2.289
4.827*
3.944
5.430
3.142
1.685
.802
2.289
-3.142
Std. Error
1.724
2.515
1.979
2.603
1.724
2.510
2.265
2.430
2.515
2.510
2.819
3.197
1.979
2.265
2.819
2.956
2.603
2.430
3.197
2.956
Sig.a
.609
.810
.015
.518
.609
.554
.082
.741
.810
.554
.055
.474
.015
.082
.055
.288
.518
.741
.474
.288
95% Confidence Interval
for Difference3
Lower
Bound
-4.271
-4.338
-8.714
-6.799
-2.504
-3.445
-8.394
-5.576
-5.545
-6.418
-10.969
-8.569
.939
-.506
-.108
-2.667
-3.428
-3.972
-3.992
-8.950
Upper
Bound
2.504
5.545
-.939
3.428
4.271
6.418
.506
3.972
4.338
3.445
.108
3.992
8.714
8.394
10.969
8.950
6.799
5.576
8.569
2.667
Based on estimated marginal means
* The mean difference is significant at the .05 level.
a. Adjustment for multiple comparisons: Least Significant Difference (equivalent to no adjustments).
Page 2
-------
Univariate Tests
Dependent Variable: CO_DIFF
Contrast
Error
Sum of
Squares
1250.449
93896.679
df
4
507
Mean
Square
312.612
185.201
F
1.688
Sig.
.151
The F tests the effect of CYCLEJD. This test is based on the linearly independent
pairwise comparisons among the estimated marginal means.
2. VEHCLASS
Estimates
Dependent Variable: CO_DIFF
VEHCLASS
LDT1
LDT2
LDV
Mean
3.372a
4.734a
5.374a
Std. Error
1.314
1.937
.901
95% Confidence Interval
Lower
Bound
.790
.929
3.604
Upper
Bound
5.954
8.539
7.144
a- Evaluated at covariates appeared in the model: CO_OFF =
16.77908, AVG_SPD = 27.95300, COEM_CAT = 7.930E-02.
Pairwise Comparisons
Dependent Variable: CO_DIFF
(I) VEHCLASS (J) VEHCLASS
LDT1 LDT2
LDV
LDT2 LDT1
LDV
LDV LDT1
LDT2
Mean
Difference
d-J)
-1.362
-2.002
1.362
-.640
2.002
.640
Std. Error
2.211
1.450
2.211
2.052
1.450
2.052
Sig.a
.538
.168
.538
.755
.168
.755
95% Confidence Interval
for Difference3
Lower
Bound
-5.707
-4.851
-2.983
-4.671
-.847
-3.391
Upper
Bound
2.983
.847
5.707
3.391
4.851
4.671
Based on estimated marginal means
a. Adjustment for multiple comparisons: Least Significant Difference (equivalent to no adjustments).
PageS
-------
Univariate Tests
Dependent Variable: CO_DIFF
Contrast
Error
Sum of
Squares
353.086
93896.679
df
2
507
Mean
Square
176.543
185.201
F
.953
Sig.
.386
The F tests the effect of VEHCLASS. This test is based on the linearly
independent pairwise comparisons among the estimated marginal means.
SECTION B
CO Univariate Analysis of Variance
No Local Cycle
LDV & LOT
All Emitter categories
Between-Subjects Factors
CYCLEJD
CLASS
ART
FWY
LA92
RAMP
LOT
LDV
N
148
221
37
37
156
287
Tests of Between-Subjects Effects
Dependent Variable: CO_DIFF
Source
Corrected Model
Intercept
CYCLEJD
COJDFF
CLASS
Error
Total
Corrected Total
Type III
Sum of
Squares
1316.1923
2454.833
326.043
167.178
851.774
86609.227
95074.518
87925.419
df
5
1
3
1
1
437
443
442
Mean
Square
263.238
2454.833
108.681
167.178
851.774
198.190
F
1.328
12.386
.548
.844
4.298
Sig.
.251
.000
.649
.359
.039
a. R Squared = .015 (Adjusted R Squared = .004)
Estimated Marginal Means
1. CYCLE ID
Page 4
-------
Estimates
Dependent Variable: CO_DIFF
CYCLE ID
ART
FWY
LA92
RAMP
Mean
2.723a
4.111a
2.316a
5.191a
Std. Error
1.176
.969
2.324
2.325
95% Confidence Interval
Lower
Bound
.412
2.206
-2.251
.622
Upper
Bound
5.035
6.016
6.884
9.761
a. Evaluated at covariates appeared in the model: CO_OFF = 15.98205.
Pairwise Comparisons
Dependent Variable: CO_DIFF
(I) CYCLE ID (J) CYCLE ID
ART FWY
LA92
RAMP
FWY ART
LA92
RAMP
LA92 ART
FWY
RAMP
RAMP ART
FWY
LA92
Mean
Difference
d-J)
-1.388
.407
-2.468
1.388
1.795
-1.081
-.407
-1.795
-2.875
2.468
1.081
2.875
Std. Error
1.496
2.588
2.588
1.496
2.501
2.502
2.588
2.501
3.273
2.588
2.502
3.273
Sig.a
.354
.875
.341
.354
.473
.666
.875
.473
.380
.341
.666
.380
95% Confidence Interval
for Difference3
Lower
Bound
-4.328
-4.679
-7.555
-1.552
-3.121
-5.999
-5.493
-6.710
-9.309
-2.619
-3.838
-3.559
Upper
Bound
1.552
5.493
2.619
4.328
6.710
3.838
4.679
3.121
3.559
7.555
5.999
9.309
Based on estimated marginal means
a. Adjustment for multiple comparisons: Least Significant Difference (equivalent to no adjustments).
Univariate Tests
Dependent Variable: CO_DIFF
Contrast
Error
Sum of
Squares
326.043
86609.227
df
3
437
Mean
Square
108.681
198.190
F
.548
Sig.
.649
The F tests the effect of CYCLEJD. This test is based on the linearly independent
pairwise comparisons among the estimated marginal means.
2. CLASS
Page 5
-------
Estimates
Dependent Variable: CO_DIFF
CLASS
LOT
LDV
Mean
2.133a
5.038a
Std. Error
1.279
1.026
95% Confidence Interval
Lower
Bound
-.381
3.022
Upper
Bound
4.647
7.054
a. Evaluated at covariates appeared in the model: CO_OFF = 15.98205.
Pairwise Comparisons
Dependent Variable: CO_DIFF
(I) CLASS (J) CLASS
LOT LDV
LDV LOT
Mean
Difference
d-J)
-2.905*
2.905*
Std. Error
1.401
1.401
Sig.a
.039
.039
95% Confidence Interval
for Difference3
Lower
Bound
-5.659
.151
Upper
Bound
-.151
5.659
Based on estimated marginal means
* The mean difference is significant at the .05 level.
a. Adjustment for multiple comparisons: Least Significant Difference (equivalent to no adjustments).
Univariate Tests
Dependent Variable: CO_DIFF
Contrast
Error
Sum of
Squares
851.774
86609.227
df
1
437
Mean
Square
851.774
198.190
F
4.298
Sig.
.039
The F tests the effect of CLASS. This test is based on the linearly independent
pairwise comparisons among the estimated marginal means.
SECTION C
CO Univariate Analysis of Variance
Normal Emit
No Local Cycle
CLASS = LDV
Page 6
-------
Tests of Between-Subjects Effects'3
Dependent Variable: CO_DIFF
Source
Model
CO OFF
AVG SPD
Error
Total
Type III
Sum of
Squares
11535.6323
4320.606
577.457
33787.462
45323.093
df
2
1
1
274
276
Mean
Square
5767.816
4320.606
577.457
123.312
F
46.774
35.038
4.683
Sig.
.000
.000
.031
a. R Squared = .255 (Adjusted R Squared = .249)
b. CLASS = LDV
Parameter Estimates3
Dependent Variable: CO_DIFF
Parameter
COJDFF
AVG SPD
B
.815
5.272E-02
Std. Error
.138
.024
t
5.919
2.164
Sig.
.000
.031
95% Confidence Interval
Lower
Bound
.544
4.759E-03
Upper
Bound
1.085
.101
a. CLASS = LDV
SECTION D
CO Univariate Analysis of Variance
Norm Emit LOT only
No local
Tests of Between-Subjects Effects
Dependent Variable: CO_DIFF
Source
Model
AVG SPD
Error
Total
Type III
Sum of
Squares
1798.4563
1798.456
28629.094
30427.550
df
1
1
131
132
Mean
Square
1798.456
1798.456
218.543
F
8.229
8.229
Sig.
.005
.005
a. R Squared = .059 (Adjusted R Squared = .052)
Parameter Estimates
Dependent Variable: CO_DIFF
Parameter
AVG SPD
B
.104
Std. Error
.036
t
2.869
Sig.
.005
95% Confidence Interval
Lower
Bound
3.230E-02
Upper
Bound
.176
Page 7
-------
CO Univariate Analysis of Variance
All Veh Classes
High emitters only
Tests of Between-Subjects Effects
Dependent Variable: CO_DIFF
Source
Model
AVG SPD
CO OFF
Error
Total
Type III
Sum of
Squares
10004.5673
5948.248
9361.556
16434.569
26439.136
df
2
1
1
39
41
Mean
Square
5002.283
5948.248
9361.556
421.399
F
11.871
14.115
22.215
Sig.
.000
.001
.000
a. R Squared = .378 (Adjusted R Squared = .347)
Parameter Estimates
Dependent Variable: CO_DIFF
Parameter
AVG_SPD
CO OFF
B
-.462
9.863E-02
Std. Error
.123
.021
t
-3.757
4.713
Sig.
.001
.000
95% Confidence Interval
Lower
Bound
-.710
5.630E-02
Upper
Bound
-.213
.141
SECTION F
CO Univariate Analysis of Variance
High emitters only
No Local Cycle
Avg speed < 19
Between-Subjects Factors
CLASS
LOT
LDV
N
6
3
PageS
-------
Tests of Between-Subjects Effects
Dependent Variable: CO_DIFF
Source
Model
CLASS
CO OFF
AVG SPD
Error
Total
Type III
Sum of
Squares
9718.1153
518.031
1.364
150.806
940.329
10658.444
df
4
2
1
1
5
9
Mean
Square
2429.529
259.015
1.364
150.806
188.066
F
12.919
1.377
.007
.802
Sig.
.008
.334
.935
.412
a. R Squared = .912 (Adjusted R Squared = .841)
Estimated Marginal Means
CLASS
Estimates
Dependent Variable: CO_DIFF
CLASS
LOT
LDV
Mean
2.167a
58.0913
Std. Error
12.164
23.003
95% Confidence Interval
Lower
Bound
-29.101
-1.040
Upper
Bound
33.435
117.223
a. Evaluated at covariates appeared in the model: CO_OFF = 159.63389, AVG_SPD = 14.43333.
Pairwise Comparisons
Dependent Variable: CO_DIFF
(I) CLASS (J) CLASS
LOT LDV
LDV LOT
Mean
Difference
d-J)
-55.925
55.925
Std. Error
33.816
33.816
Sig.a
.159
.159
95% Confidence Interval
for Difference3
Lower
Bound
-142.853
-31.003
Upper
Bound
31.003
142.853
Based on estimated marginal means
a. Adjustment for multiple comparisons: Least Significant Difference (equivalent to no adjustments).
Univariate Tests
Dependent Variable: CO_DIFF
Contrast
Error
Sum of
Squares
514.350
940.329
df
1
5
Mean
Square
514.350
188.066
F
2.735
Sig.
.159
The F tests the effect of CLASS. This test is based on the linearly independent
pairwise comparisons among the estimated marginal means.
Page 9
-------
SECTION G
CO Univariate Analysis of Variance
Avg. Speed < 19 mph
All Veh. Classes
Tests of Between-Subjects Effects
Dependent Variable: CO_DIFF
Source
Model
CO OFF
AVG SPD
Error
Total
Type III
Sum of
Squares
9200. 084a
5780.609
337.690
1458.360
10658.444
df
2
1
1
7
9
Mean
Square
4600.042
5780.609
337.690
208.337
F
22.080
27.746
1.621
Sig.
.001
.001
.244
a. R Squared = .863 (Adjusted R Squared = .824)
SECTION H
CO Univariate Analysis of Variance
High emitters only
No Local Cycle
Avg. Speed < 19
All veh. class
Tests of Between-Subjects Effects
Dependent Variable: CO_DIFF
Source
Model
CO OFF
Error
Total
Type III
Sum of
Squares
8862. 394a
8862.394
1796.050
10658.444
df
1
1
8
9
Mean
Square
8862.394
8862.394
224.506
F
39.475
39.475
Sig.
.000
.000
a. R Squared = .831 (Adjusted R Squared = .810)
Parameter Estimates
Dependent Variable: CO_DIFF
Parameter
CO OFF
B
.154
Std. Error
.024
t
6.283
Sig.
.000
95% Confidence Interval
Lower
Bound
9.738E-02
Upper
Bound
.210
SECTION I
CO Univariate Analysis of Variance
Local Cycle Only
Page 10
-------
All Veh. Classes
All Emitter Categories
Between-Subjects Factors
VEHCLASS LDT1
LDT2
LDV
N
18
8
48
Tests of Between-Subjects Effects
Dependent Variable: CO_DIFF
Source
Model
VEHCLASS
AVG SPD
CO OFF
COEM_CAT
Error
Total
Type III
Sum of
Squares
9881.0263
1154.473
76.562
2276.352
48.846
7015.951
16896.976
df
6
3
1
1
1
68
74
Mean
Square
1646.838
384.824
76.562
2276.352
48.846
103.176
F
15.961
3.730
.742
22.063
.473
Sig.
.000
.015
.392
.000
.494
a. R Squared = .585 (Adjusted R Squared = .548)
Estimated Marginal Means
VEHCLASS
Estimates
Dependent Variable: CO_DIFF
VEHCLASS
LDT1
LDT2
LDV
Mean
3.706a
13.9793
8.798a
Std. Error
2.415
3.741
1.494
95% Confidence Interval
Lower
Bound
-1.114
6.514
5.816
Upper
Bound
8.525
21.443
11.780
a- Evaluated at covariates appeared in the model: AVG_SPD =
10.00000, CO_OFF = 21.55046, COEM_CAT = 8.108E-02.
Page 11
-------
Pairwise Comparisons
Dependent Variable: CO_DIFF
(I)VEHCLASS (J)VEHCLASS
LDT1 LDT2
LDV
LDT2 LDT1
LDV
LDV LDT1
LDT2
Mean
Difference
d-J)
-10.273*
-5.093
10.273*
5.180
5.093
-5.180
Std. Error
4.382
2.871
4.382
4.102
2.871
4.102
Sig.a
.022
.081
.022
.211
.081
.211
95% Confidence Interval
for Difference3
Lower
Bound
-19.018
-10.823
1.528
-3.005
-.637
-13.366
Upper
Bound
-1.528
.637
19.018
13.366
10.823
3.005
Based on estimated marginal means
* The mean difference is significant at the .05 level.
a. Adjustment for multiple comparisons: Least Significant Difference (equivalent to no adjustments).
Univariate Tests
Dependent Variable: CO_DIFF
Contrast
Error
Sum of
Squares
644.113
7015.951
df
2
68
Mean
Square
322.057
103.176
F
3.121
Sig.
.050
The F tests the effect of VEHCLASS. This test is based on the linearly
independent pairwise comparisons among the estimated marginal means.
Univariate Analysis of Variance
Between-Subjects Factors
CLASS
LOT
LDV
N
26
48
Page 12
-------
Tests of Between-Subjects Effects
Dependent Variable: CO_DIFF
Source
Model
AVG SPD
CO OFF
COEM CAT
CLASS
Error
Total
Type III
Sum of
Squares
9314.0643
80.405
2074.928
9.007
587.511
7582.912
16896.976
df
5
1
1
1
2
69
74
Mean
Square
1862.813
80.405
2074.928
9.007
293.756
109.897
F
16.950
.732
18.881
.082
2.673
Sig.
.000
.395
.000
.776
.076
a. R Squared = .551 (Adjusted R Squared = .519)
SECTION J
CO Univariate Analysis of Variance
Local Cycle only
LDVs and LDTs
All emit
Tests of Between-Subjects Effects
Dependent Variable: CO_DIFF
Source
Model
CO OFF
Error
Total
Type III
Sum of
Squares
6754.5313
6754.531
10142.446
16896.976
df
1
1
73
74
Mean
Square
6754.531
6754.531
138.938
F
48.616
48.616
Sig.
.000
.000
a. R Squared = .400 (Adjusted R Squared = .392)
Parameter Estimates
Dependent Variable: CO_DIFF
Parameter
CO OFF
B
.125
Std. Error
.018
t
6.972
Sig.
.000
95% Confidence Interval
Lower
Bound
8.948E-02
Upper
Bound
.161
SECTION K
CO Univariate Analysis of Variance
Local cycle only
Normal emitters only
Page 13
-------
Between-Subjects Factors
VEHCLASS LDT1
LDT2
LDV
N
16
6
46
Tests of Between-Subjects Effects
Dependent Variable: CO_DIFF
Source
Model
VEHCLASS
AVG SPD
CO OFF
Error
Total
Type III
Sum of
Squares
3803. 884a
835.916
3.965
237.803
5977.832
9781.715
df
5
3
1
1
63
68
Mean
Square
760.777
278.639
3.965
237.803
94.886
F
8.018
2.937
.042
2.506
Sig.
.000
.040
.839
.118
a. R Squared = .389 (Adjusted R Squared = .340)
Estimated Marginal Means
VEHCLASS
Estimates
Dependent Variable: CO_DIFF
VEHCLASS
LDT1
LDT2
LDV
Mean
1.267a
13.3003
7.685a
Std. Error
2.557
3.986
1.455
95% Confidence Interval
Lower
Bound
-3.843
5.334
4.777
Upper
Bound
6.378
21.265
10.594
a. Evaluated at covariates appeared in the model: AVG_SPD = 10.00000, CO_OFF = 6.00891.
Page 14
-------
Pairwise Comparisons
Dependent Variable: CO_DIFF
(I)VEHCLASS (J)VEHCLASS
LDT1 LDT2
LDV
LDT2 LDT1
LDV
LDV LDT1
LDT2
Mean
Difference
d-J)
-12.032*
-6.418*
12.032*
5.614
6.418*
-5.614
Std. Error
4.781
3.005
4.781
4.228
3.005
4.228
Sig.a
.014
.037
.014
.189
.037
.189
95% Confidence Interval
for Difference3
Lower
Bound
-21.586
-12.422
2.478
-2.835
.414
-14.064
Upper
Bound
-2.478
-.414
21.586
14.064
12.422
2.835
Based on estimated marginal means
* The mean difference is significant at the .05 level.
a. Adjustment for multiple comparisons: Least Significant Difference (equivalent to no adjustments).
Univariate Tests
Dependent Variable: CO_DIFF
Contrast
Error
Sum of
Squares
703.835
5977.832
df
2
63
Mean
Square
351.918
94.886
F
3.709
Sig.
.030
The F tests the effect of VEHCLASS. This test is based on the linearly
independent pairwise comparisons among the estimated marginal means.
SECTION L
CO Univariate Analysis of Variance
Local Cycle only
LOT & LDV
Normal Emitters Only
Between-Subjects Factors
CLASS
LOT
LDV
N
22
46
Page 15
-------
Tests of Between-Subjects Effects
Dependent Variable: CO_DIFF
Source
Model
CO OFF
CLASS
Error
Total
Type III
Sum of
Squares
3202.8913
118.764
1084.797
6578.825
9781.715
df
3
1
2
65
68
Mean
Square
1067.630
118.764
542.398
101.213
F
10.548
1.173
5.359
Sig.
.000
.283
.007
a. R Squared = .327 (Adjusted R Squared = .296)
Estimated Marginal Means
CLASS
Estimates
Dependent Variable: CO_DIFF
CLASS
LOT
LDV
Mean
43243
7.553a
Std. Error
2.195
1.500
95% Confidence Interval
Lower
Bound
.440
4.558
Upper
Bound
9.208
10.549
a. Evaluated at covariates appeared in the model: CO_OFF = 6.00891.
Pairwise Comparisons
Dependent Variable: CO_DIFF
(I) CLASS (J) CLASS
LOT LDV
LDV LOT
Mean
Difference
d-J)
-2.729
2.729
Std. Error
2.698
2.698
Sig.a
.315
.315
95% Confidence Interval
for Difference3
Lower
Bound
-8.116
-2.659
Upper
Bound
2.659
8.116
Based on estimated marginal means
a. Adjustment for multiple comparisons: Least Significant Difference (equivalent to no adjustments).
Univariate Tests
Dependent Variable: CO_DIFF
Contrast
Error
Sum of
Squares
103.579
6578.825
df
1
65
Mean
Square
103.579
101.213
F
1.023
Sig.
.315
The F tests the effect of CLASS. This test is based on the linearly independent
pairwise comparisons among the estimated marginal means.
Page 16
-------
SECTION M
CO Univariate Analysis of Variance
Local Cycly Only
Normal Emitters only
LDV & LOT
Tests of Between-Subjects Effects
Dependent Variable: CO_DIFF
Source
Model
CO OFF
Error
Total
Type III
Sum of
Squares
2118.0943
2118.094
7663.622
9781.715
df
1
1
67
68
Mean
Square
2118.094
2118.094
114.382
F
18.518
18.518
Sig.
.000
.000
a. R Squared = .217 (Adjusted R Squared = .205)
Parameter Estimates
Dependent Variable: CO_DIFF
Parameter
CO OFF
B
.678
Std. Error
.158
t
4.303
Sig.
.000
95% Confidence Interval
Lower
Bound
.364
Upper
Bound
.993
SECTION N
CO Univariate Analysis of Variance
Local Cycle only
High emitters only
Between-Subjects Factors
VEHCLASS LDT1
LDT2
LDV
N
2
2
2
Page 17
-------
Tests of Between-Subjects Effects
Dependent Variable: CO_DIFF
Source
Model
VEHCLASS
AVG SPD
CO OFF
Error
Total
Type III
Sum of
Squares
7075. 528a
980.390
945.375
54.499
39.733
7115.261
df
5
3
1
1
1
6
Mean
Square
1415.106
326.797
945.375
54.499
39.733
F
35.615
8.225
23.793
1.372
Sig.
.127
.250
.129
.450
a. R Squared = .994 (Adjusted R Squared = .966)
Estimated Marginal Means
VEHCLASS
Estimates
Dependent Variable: CO_DIFF
VEHCLASS
LDT1
LDT2
LDV
Mean
-3.5003
8.052a
69.0743
Std. Error
9.341
8.188
15.722
95% Confidence Interval
Lower
Bound
-122.192
-95.982
-130.699
Upper
Bound
115.192
112.086
268.846
a. Evaluated at covariates appeared in the model: AVG_SPD = 10.00000, CO_OFF = 197.68800.
Pairwise Comparisons
Dependent Variable: CO_DIFF
(I) VEHCLASS (J) VEHCLASS
LDT1 LDT2
LDV
LDT2 LDT1
LDV
LDV LDT1
LDT2
Mean
Difference
d-J)
-11.552
-72.574
11.552
-61.021
72.574
61.021
Std. Error
6.445
24.125
6.445
22.833
24.125
22.833
Sig.a
.324
.204
.324
.228
.204
.228
95% Confidence Interval
for Difference3
Lower
Bound
-93.438
-379.108
-70.333
-351.141
-233.961
-229.098
Upper
Bound
70.333
233.961
93.438
229.098
379.108
351.141
Based on estimated marginal means
a. Adjustment for multiple comparisons: Least Significant Difference (equivalent to no adjustments).
Page 18
-------
Univariate Tests
Dependent Variable: CO_DIFF
Contrast
Error
Sum of
Squares
388.354
39.733
df
2
1
Mean
Square
194.177
39.733
F
4.887
Sig.
.305
The F tests the effect of VEHCLASS. This test is based on the linearly
independent pairwise comparisons among the estimated marginal means.
SECTION O
CO Univariate Analysis of Variance
High Emitters Only
Local Cycle Only
LDV & LOT
Between-Subjects Factors
CLASS LOT
LDV
N
4
2
Tests of Between-Subjects Effects
Dependent Variable: CO_DIFF
Source
Model
AVG_SPD
COJDFF
CLASS
Error
Total
Type III
Sum of
Squares
6947. 853a
870.834
26.455
852.715
167.408
7115.261
df
4
1
1
2
2
6
Mean
Square
1736.963
870.834
26.455
426.357
83.704
F
20.751
10.404
.316
5.094
Sig.
.047
.084
.631
.164
a. R Squared = .976 (Adjusted R Squared = .929)
Estimated Marginal Means
CLASS
Page 19
-------
Estimates
Dependent Variable: CO_DIFF
CLASS
LOT
LDV
Mean
5.089a
63.449a
Std. Error
11.639
22.361
95% Confidence Interval
Lower
Bound
-44.990
-32.763
Upper
Bound
55.167
159.660
a. Evaluated at covariates appeared in the model: AVG_SPD = 10.00000, CO_OFF = 197.68800.
Pairwise Comparisons
Dependent Variable: CO_DIFF
(I) CLASS (J) CLASS
LOT LDV
LDV LOT
Mean
Difference
d-J)
-58.360
58.360
Std. Error
33.070
33.070
Sig.a
.220
.220
95% Confidence Interval
for Difference3
Lower
Bound
-200.650
-83.929
Upper
Bound
83.929
200.650
Based on estimated marginal means
a. Adjustment for multiple comparisons: Least Significant Difference (equivalent to no adjustments).
Univariate Tests
Dependent Variable: CO_DIFF
Contrast
Error
Sum of
Squares
260.679
167.408
df
1
2
Mean
Square
260.679
83.704
F
3.114
Sig.
.220
The F tests the effect of CLASS. This test is based on the linearly independent
pairwise comparisons among the estimated marginal means.
Univariate Analysis of Variance
Tests of Between-Subjects Effects
Dependent Variable: CO_DIFF
Source
Model
AVG_SPD
COJDFF
Error
Total
Type III
Sum of
Squares
6095. 138a
35.792
3759.501
1020.123
7115.261
df
2
1
1
4
6
Mean
Square
3047.569
35.792
3759.501
255.031
F
11.950
.140
14.741
Sig.
.021
.727
.018
a. R Squared = .857 (Adjusted R Squared = .785)
SECTION P
Page 20
-------
CO Univariate Analysis of Variance
Local Cycle Only
High Emitters Only
LDV & LOT
Tests of Between-Subjects Effects
Dependent Variable: CO_DIFF
Source
Model
CO OFF
Error
Total
Type III
Sum of
Squares
6059. 346a
6059.346
1055.915
7115.261
df
1
1
5
6
Mean
Square
6059.346
6059.346
211.183
F
28.692
28.692
Sig.
.003
.003
a. R Squared = .852 (Adjusted R Squared = .822)
Parameter Estimates
Dependent Variable: CO_DIFF
Parameter
CO OFF
B
.119
Std. Error
.022
t
5.357
Sig.
.003
95% Confidence Interval
Lower
Bound
6.206E-02
Upper
Bound
.177
Page 21
-------
Appendix E:
CO Graphs
-------
CO Graphs
Figure 1
LDV- Normal Emitters- Freeway, Arterial,
oo
U
O)
60-
40-
LJ 20-
o
o
CJ
0-
-20
-10
1 CO_DATA
4 CO MODEL
10 20
CO A/C Base
30 40
-------
Figure 3
High Emitters- Cycles with Avg. Speed <
80
en
E
ra
s=
CT
O
0)
LJ
CJ
O
CJ
60-
40-
20-
0-
-20
0 100
200 300 400
CO A/C Base
CO_DATA
CO MODEL
500 600
Figure 4
LOCAL
O)
en
E
ra
s=
CT
CJ
O
CJ
60-
40-
20-
o-
-20
7
CO_MODEL
CO DATA
-100 0 100 200 300 400 500 600
CO A/C Base
-------
Figure 5
Normal Emitters- Local
1
en
E
ra
s=
CT
0
0)
Cl_
LJ
CJ
O
CJ
J U -
40-
30-
20-
10-
0-
-10
' '" V
'.
:". ? : '
."l" A^^m
CO_DATA
CO MODEL
-10
0 10 20
CO A/C Base
30
Figure 6
High Emitters- Local
^ s
O)
en
ra
CT
^_
u
O)
LJ
CJ
0
CJ
1 u
60-
50-
40-
30-
20-
10-
0-
-10
^
A
.
A*
A
"
CO_DATA
4 CO MODEL
00 200 300 400
CO A/C Base
500
600
-------
Appendix F:
NOx ANOVA Results
-------
SECTION A
NOx Univariate Analysis of Variance
All Vehicle Classes
All Cycles
Between-Subjects Factors
VEHCLASS
FACILITY
LDT1
LDT2
LDV
1
2
3
4
5
N
126
56
335
111
221
74
74
37
Tests of Between-Subjects Effects
Dependent Variable: NOX_DIFF
Source
Corrected Model
Intercept
NOXJDFF
AVG_SPD
VEHCLASS
FACILITY
VEHCLASS *
FACILITY
Error
Total
Corrected Total
Type III
Sum of
Squares
25.9123
2.523
15.856
2.876
3.057
2.076
.741
68.638
144.847
94.550
df
16
1
1
1
2
4
8
500
517
516
Mean
Square
1.620
2.523
15.856
2.876
1.529
.519
9.264E-02
.137
F
11.798
18.376
115.504
20.950
11.136
3.780
.675
Sig.
.000
.000
.000
.000
.000
.005
.714
a. R Squared = .274 (Adjusted R Squared = .251)
Estimated Marginal Means
1. VEHCLASS
Estimates
Dependent Variable: NOX_DIFF
VEHCLASS
LDT1
LDT2
LDV
Mean
.169a
.159a
.362a
Std. Error
.040
.059
.024
95% Confidence Interval
Lower
Bound
9.094E-02
4.233E-02
.314
Upper
Bound
.248
.276
.409
a. Evaluated at covariates appeared in the model: NOXJDFF = .76540, AVG_SPD = 27.95300.
Page 1
-------
Pairwise Comparisons
Dependent Variable: NOX_DIFF
(I)VEHCLASS (J)VEHCLASS
LDT1 LDT2
LDV
LDT2 LDT1
LDV
LDV LDT1
LDT2
Mean
Difference
d-J)
1.022E-02
-.192*
-1.022E-02
-.203*
.192*
.203*
Std. Error
.070
.047
.070
.064
.047
.064
Sig.a
.885
.000
.885
.002
.000
.002
95% Confidence Interval
for Difference3
Lower
Bound
-.128
-.284
-.149
-.329
.101
7.640E-02
Upper
Bound
.149
-.101
.128
-7.640E-02
.284
.329
Based on estimated marginal means
* The mean difference is significant at the .05 level.
a. Adjustment for multiple comparisons: Least Significant Difference (equivalent to no adjustments).
Univariate Tests
Dependent Variable: NOX_DIFF
Contrast
Error
Sum of
Squares
3.057
68.638
df
2
500
Mean
Square
1.529
.137
F
11.136
Sig.
.000
The F tests the effect of VEHCLASS. This test is based on the linearly
independent pairwise comparisons among the estimated marginal means.
2. FACILITY
Estimates
Dependent Variable: NOX_DIFF
FACILITY
1
2
3
4
5
Mean
.277a
.274a
.386a
.167a
4.579E-02a
Std. Error
.047
.035
.060
.056
.080
95% Confidence Interval
Lower
Bound
.185
.205
.269
5.719E-02
-.111
Upper
Bound
.369
.343
.504
.278
.202
a. Evaluated at covariates appeared in the model: NOXJDFF = .76540, AVG_SPD = 27.95300.
Page 2
-------
Pairwise Comparisons
Dependent Variable: NOX_DIFF
(1) FACILITY (J) FACILITY
1 2
3
4
5
2 1
3
4
5
3 1
2
4
5
4 1
2
3
5
5 1
2
3
4
Mean
Difference
d-J)
3.012E-03
-.109
.109
.231*
-3.012E-03
-.112
.106
.228*
.109
.112
.219*
.340*
-.109
-.106
-.219*
.122
-.231*
-.228*
-.340*
-.122
Std. Error
.061
.072
.072
.093
.061
.074
.068
.086
.072
.074
.080
.101
.072
.068
.080
.097
.093
.086
.101
.097
Sig.a
.961
.131
.128
.013
.961
.128
.117
.008
.131
.128
.006
.001
.128
.117
.006
.213
.013
.008
.001
.213
95% Confidence Interval
for Difference3
Lower
Bound
-.117
-.251
-3.150E-02
4.826E-02
-.123
-.257
-2.668E-02
5.954E-02
-3.267E-02
-3.244E-02
6.207E-02
.142
-.250
-.239
-.375
-6.998E-02
-.414
-.396
-.538
-.313
Upper
Bound
.123
3.267E-02
.250
.414
.117
3.244E-02
.239
.396
.251
.257
.375
.538
3.150E-02
2.668E-02
-6.207E-02
.313
-4.826E-02
-5.954E-02
-.142
6.998E-02
Based on estimated marginal means
* The mean difference is significant at the .05
a. Adjustment for multiple comparisons: Least
level.
Significant Difference (equivalent to no adjustments).
Univariate Tests
Dependent Variable: NOX_DIFF
Contrast
Error
Sum of
Squares
2.076
68.638
df
4
500
Mean
Square
.519
.137
F
3.780
Sig.
.005
The F tests the effect of FACILITY. This test is based on the linearly independent
pairwise comparisons among the estimated marginal means.
SECTION B
NOx Univariate Analysis of Variance
Log Space
No Ramp Cycle
LDV only
PageS
-------
Tests of Between-Subjects Effects
Dependent Variable: NOX_DIFF
Source
Model
LG10NOX1
LG10SPD
Error
Total
Type III
Sum of
Squares
57.523a
26.665
4.680E-02
48.681
106.203
df
2
1
1
309
311
Mean
Square
28.761
26.665
4.680E-02
.158
F
182.563
169.256
.297
Sig.
.000
.000
.586
a. R Squared = .542 (Adjusted R Squared = .539)
Parameter Estimates
Dependent Variable: NOX_DIFF
Parameter
LG10NOX1
LG10SPD
B
1.865
-1.350E-02
Std. Error
.143
.025
t
13.010
-.545
Sig.
.000
.586
95% Confidence Interval
Lower
Bound
1.583
-6.225E-02
Upper
Bound
2.148
3.525E-02
SECTION C
NOx Univariate Analysis of Variance
Linear Space
LDV only
No Ramp Cycle
Tests of Between-Subjects Effects
Dependent Variable: NOX_DIFF
Source
Model
AVG SPD
NOX OFF
Error
Total
Type III
Sum of
Squares
54.725a
8.316E-02
39.687
51.478
106.203
df
2
1
1
309
311
Mean
Square
27.363
8.316E-02
39.687
.167
F
164.245
.499
238.225
Sig.
.000
.480
.000
a. R Squared = .515 (Adjusted R Squared = .512)
Parameter Estimates
Dependent Variable: NOX_DIFF
Parameter
AVG_SPD
NOX OFF
B
-5.996E-04
.459
Std. Error
.001
.030
t
-.707
15.435
Sig.
.480
.000
95% Confidence Interval
Lower
Bound
-2.270E-03
.400
Upper
Bound
1.070E-03
.517
SECTION D
Page 4
-------
Speed Bin Approach
Speed <15
Tests of Between-Subjects Effects
Dependent Variable: NOX_DIFF
Source
Model
LG10NOX1
Error
Total
Type III
Sum of
Squares
39.366a
39.366
15.308
54.674
df
1
1
95
96
Mean
Square
39.366
39.366
.161
F
244.308
244.308
Sig.
.000
.000
a. R Squared = .720 (Adjusted R Squared = .717)
Parameter Estimates
Dependent Variable: NOX_DIFF
Parameter
LG10NOX1
B
2.637
Std. Error
.169
t
15.630
Sig.
.000
95% Confidence Interval
Lower
Bound
2.302
Upper
Bound
2.971
15 < Speed < 30
Tests of Between-Subjects Effects
Dependent Variable: NOX_DIFF
Source
Model
LG10NOX1
Error
Total
Type III
Sum of
Squares
22.324a
22.324
20.303
42.626
df
1
1
143
144
Mean
Square
22.324
22.324
.142
F
157.233
157.233
Sig.
.000
.000
a. R Squared = .524 (Adjusted R Squared = .520)
Parameter Estimates
Dependent Variable: NOX_DIFF
Parameter
LG10NOX1
B
1.668
Std. Error
.133
t
12.539
Sig.
.000
95% Confidence Interval
Lower
Bound
1.405
Upper
Bound
1.931
Speed > 30
Page 5
-------
Tests of Between-Subjects Effects
Dependent Variable: NOX_DIFF
Source
Model
LG10NOX1
Error
Total
Type III
Sum of
Squares
4.673a
4.673
7.184
11.856
df
1
1
94
95
Mean
Square
4.673
4.673
7.642E-02
F
61.140
61.140
Sig.
.000
.000
a. R Squared = .394 (Adjusted R Squared = .388)
Parameter Estimates
Dependent Variable: NOX_DIFF
Parameter
LG10NOX1
B
.900
Std. Error
.115
t
7.819
Sig.
.000
95% Confidence Interval
Lower
Bound
.672
Upper
Bound
1.129
SECTION E
LDV NOx Model
No Ramp Cycle
Tests of Between-Subjects Effects
Dependent Variable: NOX_DIFF
Source
Model
LG10NOX1
LGNXLGS
Error
Total
Type III
Sum of
Squares
65.252a
18.783
7.776
40.951
106.203
df
2
1
1
309
311
Mean
Square
32.626
18.783
7.776
.133
F
246.184
141.730
58.678
Sig.
.000
.000
.000
a. R Squared = .614 (Adjusted R Squared = .612)
Parameter Estimates
Dependent Variable: NOX_DIFF
Parameter
LG10NOX1
LGNXLGS
B
4.867
-2.296
Std. Error
.409
.300
t
11.905
-7.660
Sig.
.000
.000
95% Confidence Interval
Lower
Bound
4.063
-2.886
Upper
Bound
5.672
-1.706
SECTION F
LOG Space
LOT only
No Ramp Cycle
Page 6
-------
Tests of Between-Subjects Effects
Dependent Variable: NOX_DIFF
Source
Model
LG10NOX1
LG10SPD
Error
Total
Type III
Sum of
Squares
12.0063
2.289
1.182E-02
23.684
35.690
df
2
1
1
180
182
Mean
Square
6.003
2.289
1.182E-02
.132
F
45.623
17.398
.090
Sig.
.000
.000
.765
a. R Squared = .336 (Adjusted R Squared = .329)
Parameter Estimates
Dependent Variable: NOX_DIFF
Parameter
LG10NOX1
LG10SPD
B
.786
1.247E-02
Std. Error
.188
.042
t
4.171
.300
Sig.
.000
.765
95% Confidence Interval
Lower
Bound
.414
-6.963E-02
Upper
Bound
1.158
9.458E-02
Linear Space
LOT Only
No Ramp Cycle
Tests of Between-Subjects Effects
Dependent Variable: NOX_DIFF
Source
Model
NOXJDFF
AVG_SPD
Error
Total
Type III
Sum of
Squares
10.7303
5.232
9.585E-05
24.960
35.690
df
2
1
1
180
182
Mean
Square
5.365
5.232
9.585E-05
.139
F
38.690
37.729
.001
Sig.
.000
.000
.979
a. R Squared = .301 (Adjusted R Squared = .293)
Parameter Estimates
Dependent Variable: NOX_DIFF
Parameter
NOXJDFF
AVG SPD
B
.204
3.132E-05
Std. Error
.033
.001
t
6.142
.026
Sig.
.000
.979
95% Confidence Interval
Lower
Bound
.138
-2.320E-03
Upper
Bound
.269
2.382E-03
SECTION G
Speed Bin Approach
Page 7
-------
Speed <15
Tests of Between-Subjects Effects
Dependent Variable: NOX_DIFF
Source
Model
LG10NOX1
Error
Total
Type III
Sum of
Squares
6.604a
6.604
10.373
16.977
df
1
1
51
52
Mean
Square
6.604
6.604
.203
F
32.468
32.468
Sig.
.000
.000
a. R Squared = .389 (Adjusted R Squared = .377)
Parameter Estimates
Dependent Variable: NOX_DIFF
Parameter
LG10NOX1
B
1.201
Std. Error
.211
t
5.698
Sig.
.000
95% Confidence Interval
Lower
Bound
.778
Upper
Bound
1.624
15 < Speed < 31
Tests of Between-Subjects Effects
Dependent Variable: NOX_DIFF
Source
Model
LG10NOX1
Error
Total
Type III
Sum of
Squares
4.634a
4.634
6.782
11.416
df
1
1
77
78
Mean
Square
4.634
4.634
8.808E-02
F
52.617
52.617
Sig.
.000
.000
a. R Squared = .406 (Adjusted R Squared = .398)
Parameter Estimates
Dependent Variable: NOX_DIFF
Parameter
LG10NOX1
B
.820
Std. Error
.113
t
7.254
Sig.
.000
95% Confidence Interval
Lower
Bound
.595
Upper
Bound
1.046
Speed > 31
PageS
-------
Tests of Between-Subjects Effects
Dependent Variable: NOX_DIFF
Source
Model
LG10NOX1
Error
Total
Type III
Sum of
Squares
1.798a
1.798
5.499
7.297
df
1
1
51
52
Mean
Square
1.798
1.798
.108
F
16.676
16.676
Sig.
.000
.000
a. R Squared = .246 (Adjusted R Squared = .232)
Parameter Estimates
Dependent Variable: NOX_DIFF
Parameter
LG10NOX1
B
.561
Std. Error
.137
t
4.084
Sig.
.000
95% Confidence Interval
Lower
Bound
.285
Upper
Bound
.837
SECTION H
LOT only
No Ramp Cycle
Tests of Between-Subjects Effects
Dependent Variable: NOX_DIFF
Source
Model
LG10NOX1
LGNXLGS
Error
Total
Type III
Sum of
Squares
12.7653
2.436
.735
20.961
33.726
df
2
1
1
167
169
Mean
Square
6.382
2.436
.735
.126
F
50.850
19.406
5.859
Sig.
.000
.000
.017
a. R Squared = .378 (Adjusted R Squared = .371)
Parameter Estimates
Dependent Variable: NOX_DIFF
Parameter
LG10NOX1
LGNXLGS
B
1.930
-.769
Std. Error
.438
.318
t
4.405
-2.421
Sig.
.000
.017
95% Confidence Interval
Lower
Bound
1.065
-1.395
Upper
Bound
2.795
-.142
SECTION I
NOx Univariate Analysis of Variance
Log fit
Page 9
-------
Ramp Cycle only
Between-Subjects Factors
VEHCLASS LDT1
LDT2
LDV
N
9
4
24
Tests of Between-Subjects Effects
Dependent Variable: NOX_DIFF
Source
Model
VEHCLASS
LGNOX
Error
Total
Type III
Sum of
Squares
2.147a
.466
.663
2.771
4.918
df
4
3
1
33
37
Mean
Square
.537
.155
.663
8.397E-02
F
6.391
1.850
7.897
Sig.
.001
.157
.008
a. R Squared = .437 (Adjusted R Squared = .368)
Estimated Marginal Means
VEHCLASS
Estimates
Dependent Variable: NOX_DIFF
VEHCLASS
LDT1
LDT2
LDV
Mean
8.673E-023
-6.543E-023
.270a
Std. Error
.101
.148
.061
95% Confidence Interval
Lower
Bound
-.118
-.366
.146
Upper
Bound
.291
.236
.395
a. Evaluated at covariates appeared in the model: LGNOX = .2645.
Page 10
-------
Pairwise Comparisons
Dependent Variable: NOX_DIFF
(I)VEHCLASS (J)VEHCLASS
LDT1 LDT2
LDV
LDT2 LDT1
LDV
LDV LDT1
LDT2
Mean
Difference
d-J)
.152
-.184
-.152
-.336*
.184
.336*
Std. Error
.174
.121
.174
.163
.121
.163
Sig.a
.389
.139
.389
.047
.139
.047
95% Confidence Interval
for Difference3
Lower
Bound
-.202
-.430
-.506
-.667
-6.299E-02
4.339E-03
Upper
Bound
.506
6.299E-02
.202
-4.339E-03
.430
.667
Based on estimated marginal means
* The mean difference is significant at the .05 level.
a. Adjustment for multiple comparisons: Least Significant Difference (equivalent to no adjustments).
Univariate Tests
Dependent Variable: NOX_DIFF
Contrast
Error
Sum of
Squares
.438
2.771
df
2
33
Mean
Square
.219
8.397E-02
F
2.608
Sig.
.089
The F tests the effect of VEHCLASS. This test is based on the linearly
independent pairwise comparisons among the estimated marginal means.
SECTION J
NOx Univariate Analysis of Variance
Log format
All veh. class
Ramp Cycle only
Tests of Between-Subjects Effects
Dependent Variable: NOX_DIFF
Source
Model
LGNOX
Error
Total
Type III
Sum of
Squares
1.681a
1.681
3.237
4.918
df
1
1
36
37
Mean
Square
1.681
1.681
8.992E-02
F
18.690
18.690
Sig.
.000
.000
a. R Squared = .342 (Adjusted R Squared = .323)
Page 11
-------
Parameter Estimates
Dependent Variable: NOX_DIFF
Parameter
LGNOX
B
.655
Std. Error
.152
t
4.323
Sig.
.000
95% Confidence Interval
Lower
Bound
.348
Upper
Bound
.963
Page 12
-------
Appendix G:
NOx Graphs
-------
NOx Graphs
Figure 1
LDV- Freeway, Arterial, Local
O)
3-
2-
U
O)
X
O
L NOX_MODEL
NOX DATA
-.5 0.0 .5 1.0 1.5 2.0 2.5 3.0 3.5
NOx A/C Base
-------
Figure 3
LDV & LOT- Ramp Cycle
0)
en
E
ra
CT
_,_
0
0)
Ci
UJ
"
X
o
z
1 .U '
.8-
.6-
.4-
.2-
o.o-
-.2-
-.4
- .
Al *
.
4* A
'
* ^ť^
' .
4 NOX_MODEL
NOX DATA
-.5 0.0 .5 1.0 1.5 2.0 2.5 3.0 3.5
NOx A/C Base
-------
Appendix H:
CRC Comparison Graphs
-------
MOBILES LDV A/C Model vs. CRC LDV A/C Data
NOx A/C Effect as a function of NOx A/C base, SC03
-1 J-
NOx A/C base
-------
-2 J
MOBILES A/C Model vs. CRC A/C Data
CO A/C Effect as a function of CO A/C base, SC03
CO A/C base
-------
MOBILES A/C Model vs. CRC A/C Data
NMHC A/C Effect as a function of NMHC A/C base, SC03 (Normal Emitters)
0.6
0.5
^ 0.4
J^
J 0.3 -
t/3
I 0.2
x29
Ľ 0.1
Ł
M-H _
W 0
u
u
-0.1
0
-0.3 -
-0.4 -
-0.5
0.2
0.4
0.6
0.8
1.2
1.4
1.6
1.8
NMHC A/C base
-------
Appendix I:
Cold Start Ratios
-------
Table 1 - Cold Start ST01 A/C Ratios
(average cycle speed = 20.2 mph)
Fuel
NOx
NMHC
CO
LDV
Normal
1.17
1.24
0.96
0.95
High
n/a
n/a
1.29
1.60
LDT
Normal
1.13
1.19
1.05
1.17
High
n/a
n/a
0.97
0.99
-------
Appendix J:
SFTP Benefits
-------
TABLE 1
TIER 1 SFTP BENEFITS FROM
OFF-CYCLE OPERATION and AIR CONDITIONING*
POLLUTANT
HC
CO
NOx
SFTP BENEFIT
100%
Fuel Consumption Increase
50%
*EPA rule estimated benefits of Tier 1 SFTP standards, in terms of
percent reduction of uncontrolled "excess" emissions.
TABLE 2
WORKSHEET FOR DEVELOPING LEV NOX BENEFITS
(1) Average FTP Certification Level
(2) Estimated "Running" Certification Level
(3) Estimated NOx 4K Standard (ARB)
(4) Estimated NOx 50K Standard (EPA)
(5) US06 Standard / Running Certification Level
(6) Additional Stringency of ARB Standard
(7) EPA SFTP Benefit (%)
(8) ARB Benefit (%)
Tier 1 (50K Miles)
LD,^7 LDT2 LDT3 LDT4
0.17 0.19 0.24 0.30
0.15 0.17 0.21 0.27
0.58 0.90 0.90 1.32
50 50 50 50
LEV (4K Miles)
LD,^7 LDT2 LDT3 LDT4
0.07 0.11 0.13 0.16
0.06 0.10 0.11 0.14
0.17 0.23 0.27 0.38
2.67 2.32 2.35 2.70
40% 69% 59% 56%
70% 85% 79% 78%
TABLE 3
LEV SFTP BENEFITS FROM A/C OPERATION
HC
CO
NOx
LDV/LDT1
100%
20% Remain
70%
LD2
100%
20% Remain
85%
LD3
100%
20% Remain
79%
LD4
100%
20% Remain
78%
*EPA rule estimated benefits of LEV SFTP standards, in terms of percent reduction of uncontrolled "excess"
emissions.
-------
Appendix K:
Peer Review Comments
-------
This section summarizes comments from the two formal peer reviews the original draft report
underwent after publication in March 1998.
Commentor 1:
The comment text is verbatim from the commentor, minus editorial corrections to maintain the
confidentiality of the commentor. Because the report was significantly modified after these
comments were received, many of the detailed comments are no longer applicable. The
overarching recommendations in the "general comment" section were addressed through the
change in approach from a multiplicative air conditioning correction factor to an additive
correction factor.
General comment:
You are using the ratios (with a/c)/(without a/c). Perhaps early in the paper it would be helpful to
indicate reason for using pollutant ratios instead of increments (so it will apply to large cars as
well as small; high emitters as well as low?). Such ratios for pollutants seem more logical than
for engine power. The power increment needed by the a/c compressor is more easily modeled as
an absolute number than a ratio. This observation is the basis for many comments that follow.
Basically, a/c power requirement is almost linearly proportional to engine rpm at a given
compressor suction and discharge pressure. These pressures depend, in turn, on the indoor and
outdoor temperature and airflow condition, which apparently are identical for the vehicle testing
program discussed in the draft report. Therefore improvements might be obtained by adding the
a/c power directly to the (large or small) engine power first, and then perhaps employ the
assumption that emissions are directly proportional to the total engine power (for drivetrain plus
compressor).
Page 2
"It should be noted that the correction factors presented in this report apply to vehicles which do
not comply with the SFTP requirement." We would add some comment or hint indicating why it
does not apply. What would be needed to comply? What are the effects on non-compliance?
Answers are well-known to those familiar with the issues, but probably not to many readers.
Page 3. para 1
Define "standard cooling"; it is not clear whether it refers to an engine cooling or a/c setting.
Explain rationale for the "driver window down" criterion: to ensure that the interior of the car
does not cool, so the indoor coil sees high temperature air. The justification for the "max a/c,
recirc" setting is not clear. By recirculating air that has already been dehumidified instead of
using 100% fresh air, the open window and the recirculation are working at cross purposes.
Small differences in wind outside the driver's window will therefore affect the amount of mixing
(of recirculated and fresh air) and therefore cause the evaporator to see different inlet conditions
in different tests. This might explain some of the differences observed for the same vehicle at
different test sites.
Page 3. para 3
-------
Apparently the numerator of the correction factor is measured on the 95° test, while the
denominator is a calculated (corrected) 75° test value. Next time such tests are done, a more
accurate correction factor might be obtained by testing a/c off at 95° so both figures are measured
in the same facility at the same time.
The relative humidity in the 75° test is about 38%, and it is about 20% in the 95° test. The dew
point of the air in the 95° test is 48°F, so the evaporator is probably removing sensible heat only.
The coil will be wet and dehumidify the air only if the fan speed is low enough that the
temperature of the exiting air drops below 48°. The report provides no information on exit air
temperatures, so it is not possible to know whether closeness to the dewpoint might explain some
of the variance between results obtained at different laboratories. At any rate, dry-coil conditions
are not very common in actual operation. Under "full load" conditions (95°, relative humidity =
40%) the dewpoint is 67° so the coil will become wet and latent loads will appear when the air is
cooled below that temperature.
Page 5, para 1
Notes the difference between high and low emitters for HC and CO. This raises the possibility,
on cycles where compressor cycles on/off every few seconds, that different cars' air-fuel mixture
controls might not be well-programmed to account for time constants as short as these. In other
words, by the time the mixture adjusts to the step function input due to compressor power, the
compressor cycles off.
Page 5 para 4
Typo on line 3: "is"
Page 5, para 4
It is surprising that the compressor was on for 97% of the time "on each of the cycles". Clarify
here that only 4 of the cycles were tested, or the whole list shown in Table 2? Was compressor
on full-time during the other cycles as well? On most cars the compressor shuts off when the
suction pressure falls to the level where the refrigerant evaporating temperature reaches 32°
(about 42 psia, or 28 psig). Since Table 4 says simply "psi", we suspect that it is a gage pressure
and that explains why the compressor never shut off. The description of Table 4 is hard to
understand without more knowledge of the differences between the test cells. For example if
GM reported gage pressures and EPA reported absolute evaporating pressures, that could account
for the differences noted.
Page 6. para 1
Implies that the GM test cell used a variable speed fan. Such differences between test cells
should be explained more fully. For example would the fan affect the functioning of the open
driver's window as well as the condenser face velocity? Are there any other differences, e.g.
humidity? WEagree that more research might be required to fully explain differences, but my
point is that a better partial explanation might be drawn from available data if the differences in
test conditions were better documented. Similar comments arose when looking at Table 3; here
are some comments prompted by it:
-------
Tables.
Discrepancies in this table should be better addressed; more explanations should be attempted.
There could be four possible causes:
a)instruments used in each test laboratory are not equal or calibrated
b)the test conditions are not identical
c)the test procedure (measurement) is not identical
d)the vehicle was not identically adjusted prior to each test
The report does not describe what attempts, if any, were made to identify the reasons for
nonrepeatability of measurements. In that respect, it would be good to document:
a)types of instruments, year of production, date calibrated and where, accuracy, repeatability,
range, etc. In what part of the instruments range were measurements?
b)test conditions (for vehicle and a/c system) are not exactly described. What are possible
differences? What are tolerances?
c)Did operators have the same instructions? What are tolerances? Did technician repeat the
test at one location at the same vehicle to demonstrate repeatability?
d)What were the time periods (and maybe miles) between two tests? Was vehicle hauled or
driven from one location to another?
Figures 5 to 10 and associated discussion
"Average Cycle Speed" is a term of art they will confuse mobile a/c experts who use the term
"cycling" to describe the on/off operation of the compressor, and speak of the speed or frequency
at which this occurs. It may be useful in this report to use the term "Vehicle speed" since the
cycle is identified in the same Figure.
We would question the selection of the parabolic form of the regression line. We would expect
that the influence of a/c system will decline as vehicle speed increases because engine power
increases nonlinearly with vehicle speed (with the cube of engine rpm in a given gear?) while
compressor power increases only linearly (or slightly less due to effect of ram air reducing
condensing pressure), so the relative influence of compressor power is therefore reduced. In
other words we would expect an asymptotic regression curve with ratio just above one at higher
speeds. Such behavior seems consistent with the data shown on Figures 5, 6, 8, and 10.
Figures 7 and 9 show little or no influence of vehicle speed, or high scatter around whatever
trend exists. This is not unexpected, because "vehicle speed" is a highly aggregated parameter.
Perhaps we should not even be concerned about explaining whether speed or something else is
the determining variable; this would be the case if we were interested only in emissions per cycle.
If in the future it is necessary to try to eliminate some of this scatter in order to better understand
the factors affecting emission variations among cycles, one could break the cycles into "speed
bins" and calculate a mileage-weighted average of emissions. This might provide a more solid
foundation when one wants to tie backwards to the emission standards which are expressed in
grams/mile on a given cycle, or forwards to application of these correction factors to facility
types having different average speeds. However even such a technique would fail to capture
effects of acceleration differences between cycles. Therefore it might be wise to examine ways
to model emissions based on cycle testing alone, rather than using vehicle speed as an
intermediate variable. However if the correlation approach is needed and must be improved, it
-------
might be worth trying adding two additional variables from the cycles: average acceleration and
percent idle time.
Page 7 para 3
Typo on line 2: "vehicles were"
Page 8. para 3
We would not make too much of the dip at middle speeds; we would view it as scatter until
considering how speeds and accelerations are distributed within cycles.
Commentor 2:
Charles Kowalski, PhD
Center for Statistical Consultation and Research
University of Michigan
Verbatim comments not available electronically. Hard copies available upon request. As with
the first set of comments, because the report was significantly modified after these comments
were received, many of the detailed comments are no longer applicable.
CommentrThe general recommendation was to be more systematic in how to determine what
effects are important for generating correction factors, while at the same time acknowledging that
factors such as technical judgement may play as important a role as statistical significance; in
general, statistical significance should not be the sole judge of determining correction factors.
Response: the current approach represents a good balance between technical judgement in the
absence of data, but statistical approach for determining correction factors where merited by data
------- |