Appendix A
Detailed Statistical Analysis Results
A-l
-------
A.l Response Time
The ANOVA analyses of response time are contained in the following sections (see Chapter 5
for more details on the ANOVA procedure used). It should be noted that, in all cases, the log
response time was modeled. The geometric mean of each result from the ANOVA model was
then used to put the findings back into the original scale (as opposed to the log scale).
A.l.l Effect of Temperature on TIC Response Time
The HAZMATCAD Plus response time for each TIC was tested at low, medium, and high
temperature. Test IDs included in this analysis are contained in Table A-l. Over the range of
temperature settings, average response time varied from a low of about 8 seconds with the AC
runs to a high of about 12 seconds with the CG runs.
Table A-l. IDs of Tests Included in the Test of Effect of Temperature on TIC Response
Time
AC-01-A
CG-01-A
Cl2-01-A
SA-01-A
AC-01-B
CG-01-B
ci2-oi-b
SA-01-B
AC-05-A
CG-05-A
Cl2-05-A
SA-05-A
AC-05-B
CG-05-B
Cl2-05-B
SA-05-B
AC-07-A
CG-07-A
Cl2-07-A
SA-07-A
AC-07-B
CG-07-B
Cl2-07-B
SA-07-B
P-values for tests of the effects of temperature are contained in Table A-2. For each test, the
p-value is greater than 0.05. Thus, there is no evidence that temperature has an effect on
HAZMATCAD Plus response time.
Table A-2. Tests for Effects of Temperature on Response Time by TIC
TIC
P-value
AC
0.18
CG
0.78
ci2
0.16
SA
0.93
A.l.2 Effect of Humidity on TIC Response Time
The HAZMATCAD Plus response time for each TIC also was tested at low, medium, and high
humidity. Test IDs for this analysis are contained in Table A-3. P-values for tests of these effects
are contained in Table A-4.
A-2
-------
Table A-3. IDs of Tests Included in the Test of Effect of Humidity on TIC Response Time
AC-01-A CG-01-A Cl2-01-A SA-01-A
AC-01-B CG-01-B Cla-Ol-B SA-01-B
AC-03-A CG-03-A Cl2-03-A SA-03-A
AC-03-B CG-03-B Cl2-03-B SA-03-B
AC-04-A CG-04-A Cl2-04-A SA-04-A
AC-04-B CG-04-B Cl2-04-B SA-04-B
Table A-4. Tests for Effects of Humidity on Response Time by TIC
TIC P-value
AC 0.56
CG 0.03
Cl2 0.27
SA 0.64
As evidenced by Table A-4, humidity only has a significant effect on the response time for CG.
For each TIC, Figure A-l provides the modeled geometric means of response time based on the
appropriate ANOVA model.
From Figure A-l, the longest response time for CG is for the highest level of humidity. It should
be noted that while there is a statistically significant difference among the response times for the
different levels of humidity for CG, the difference does not seem practically significant.
A. 1.3 Effect of Start State on TIC Response Time
The HAZMATCAD Plus response time for AC was recorded under medium temperature and
humidity with three start states:
1. Cold soak/cold start
2. Hot soak/cold start
3. Room temperature/cold start
The average response time varied from a low of about 7 seconds with cold soak/cold start runs to
a high of about 8 seconds with the room temperature/cold start runs. Start-state results were
combined with the responses from the medium humidity and medium temperature AC results
from the previous tests for the sake of comparison. The test IDs for this analysis are contained in
Table A-5. The p-value for the significance of start state is 0.63, so there is no evidence that start
state has a significant effect on HAZMATCAD Plus response time.
A-3
-------
15
14
13
12
,X—
CO
11
"O
c
o
L>
10
(L>
9
V
E
8
f—
7
w
C
6
D
Cl
10
5
V
Cd
4
3
2
1
O
& 41 6. 34
8. 71 8. SO
M
fiC-
M
CE3"
M
CL2
M
Humidity
Gas
Figure A-l. Modeled geometric mean of the response time by TIC and level of
humidity.
Table A-5. IDs of Tests Included in the Test of Effect of Start State on TIC Response Time
AC-01-A
AC-01-B
AC-20-A
AC-20-B
AC-21-A
AC-21-B
AC-22-A
AC-22-B
A. 1.4 Effect of Temperature on CW Agent Response Time
The tests used to assess the effect of temperature on response time are identified in Table A-6.
For GB, there were 5 runs at each temperature level. For HD there were 10 runs at each of two
temperature levels: medium and high. It was not possible to include HD data at low temperature
at the targeted HD concentration because the test was run at a lower concentration.
A-4
-------
Table A-6. IDs of Tests included in the Agent Testing of the Effect of Temperature on
Response Time
GB-01- HD-1A-
BA1 1
GB-5-1 HD-1B-1
GB-7-1 HD-7A-
1
HD-7B-1
With both GB and HD, there was no evidence that temperature has an effect on HAZMATCAD
Plus response time (p-values of 0.79 and 0.92, respectively).
A.1.5 Effect of Humidity on CW Agent Response Time
The tests used to assess the effect of humidity on response time are identified in Table A-7. For
HD there were 10 runs at each humidity level. For GB there were 5 runs at each humidity level.
One of the GB high humidity runs had no response and thus no associated response time. The
remaining high humidity runs were associated with a response, but the response was not stable.
For those runs, an initial response time was captured and that time is used in the analysis.
Table A-7. IDs of Tests included in the Agent Testing of the Effect of Humidity on
Response Time
GB-01- HD-1A-1
BA1
GB-03-B HD-1B-1
GB-04-B HD-3A-1
HD-3B-1
HD-4AA-1
HD-4BA-1
With both GB and HD, there was no evidence that humidity has an effect on HAZMATCAD
Plus response time (p-values of 0.18 and 0.06, respectively).
A.1.6 Summary of Response Time Analysis
Variations in temperature, humidity, and start state appear to have little effect on
HAZMATCAD Plus response time. Over all testing, the only significant finding was for CG
with variation in humidity, but the effect did not appear to be of practical significance.
A-5
-------
A.2 Recovery Time
The humidity and temperature recovery time data were analyzed with a standard ANOVA
model. None of the recovery times in those tests exceeded the maximum allowable time of 600
seconds.
For the analysis of start state, some of the TIC recovery time data were "censored," i.e.,
truncated at 600 seconds, even though HAZMATCAD Plus response had not yet returned to
baseline after that length of time. A survival model as described in Chapter 5 was used.
It should be noted that, in all cases, the log recovery time was modeled. The geometric mean of
each result from the appropriate model was then used to put the findings back into the original
scale (as opposed to the log scale).
A.2.1 Effect of Temperature on TIC Recovery Time
The HAZMATCAD Plus recovery time for each TIC was tested at low, medium, and high
temperature. P-values for tests of these effects are contained in Table A-8. (Test IDs included in
this analysis can be found in Table A-l in Section A. 1.1)
Table A-8. Tests for Effects of Temperature on Recovery Time
TIC
P-value
AC
<0.01
CG
<0.01
ci2
<0.01
SA
0.49
As evidenced by Table A-8, temperature has a significant effect on recovery time for every TIC
except SA. Figure A-2 contains the geometric mean recovery time for each TIC by level of
temperature based on the ANOVA model for each TIC.
While differences among the levels of temperature are evident for each of the first three TICs,
the greatest differences are apparent for AC. In this case, recovery time increases as temperature
decreases. The mean recovery time for AC under low temperature is more than twice as long as
that for AC at higher temperatures, as well as that for any other TIC under any temperature level.
A-6
-------
361
lmh lmh lmh lmh Temperature
I PC—I I—as—I I CL2 —I I——I Gas
Figure A-2. Modeled geometric mean of recovery time by TIC and level of
temperature.
A.2.2 Effect of Humidity on TIC Recovery Time
The HAZMATCAD Plus recovery time for each TIC also was tested at low, medium, and high
humidity. P-values for tests of these effects are contained in Table A-9. (Test IDs included in
this analysis can be found in Table A-3 in Section A. 1.2.)
As evidenced by Table A-9, humidity has a significant effect on the recovery time for AC, CG
and Cl2. Figure A-3 contains the geometric mean for recovery time based on the ANOVA
models by TIC and level of humidity.
Once again, the largest differences among recovery times for the different levels of humidity
occur for AC. The recovery time for AC is longest for medium humidity.
Table A-9. Tests for Effects of Humidity on Recovery Time
TIC P-value
AC <0.01
CG <0.01
Cl2 0.02
SA 0.84
A-7
-------
C/)
~o
140
130
120
110
100
O
a) 90
w
<0 80
E
r- 70
^ 60
CD
>
O 50
o
en 40
30
20
10
o
_sa_
25 24 24
M
fiC-
M
Ct3-
L M
I CL2
m h Humidity
s*—I Gas
Figure A-3. Modeled geometric mean of recovery time by TIC and level of
humidity.
A.2.3 Effect of Start State on TIC Recovery Time
The HAZMATCAD Plus recovery time for AC was recorded under medium temperature and
humidity with three start states:
1. Cold soak/cold start
2. Hot soak/cold start
3. Room temperature/cold start
These results were combined with the responses from the medium humidity and medium
temperature AC results from the previous tests for the sake of comparison. The p-value for the
significance of start state is <0.01, indicating that start state does have a significant effect on
HAZMATCAD Plus recovery time. (See Table A-5 in Section A. 1.3 for a list of test IDs
included in this analysis.) The p-values comparing the recovery times for each of the three start
states with the control start state are contained in Table A-10.
A-8
-------
Table A-10. Effect of Each Start State on Recovery Time Compared with the Control
Start State
Comparison
P-value
Cold Soak/Cold Start vs. Control
0.06
Hot Soak/Cold Start vs. Control
0.02
Room Temperature Cold Start vs. Control
<0.01
As evidenced by Table A-10, both the Hot Soak and Room Temperature start states have
recovery times that differ significantly from the recovery time for the control start state. Figure
A-4 contains the modeled geometric mean recovery times by start state.
400
"crT 300
XJ
c
o
N
CD
>
O
-------
the modeled geometric mean of recovery time for each agent by temperature level. Recovery
time appeal's to be greater for lower temperatures.
A.2.5 Effect of Humidity on CW Agent Recovery Time
Data available for the agent analysis of the effect of humidity on recovery time came from the
tests identified in Table A-7. For GB, the high humidity runs at room temperature had either no
400
350
V 300
"O
C
o
u
£ 250
03
¦p 200
QJ
> 150
u
CD
CE
100
50
0_u LI u u u
L M H L M H Temperature
I— GB —I I— HD —I Gas
Figure A-5. Modeled geometric mean of recovery time by agent and temperature level.
response or an unstable response. For all five high humidity runs, no recovery time could be
captured. These runs were not included in the analysis.
For both agents, there was evidence that humidity had an effect on HAZMATCAD Plus
recovery time (p-value <0.01 for both agents). Figure A-6 summarizes the modeled geometric
mean of recovery time for each agent by humidity level. The trends were not consistent.
Recovery time was longer for higher humidity in the case of GB, but longer for lower humidity
in the case of HD.
357
143
80
31 "
0
L M H L M H
I GB 1 I HD 1
A-10
-------
500
400
"ST
"O
c
o
u
CD
^ 300
CD
E
I—
b
(u 2flfl
>
o
u
CD
Q1
1Q Q
a
L M H L M H Humidity
I GB 1 I HD 1 Gas
Figure A-6. Modeled geometric mean of recovery time by agent and humidity level.
A.2.6 Summary of Recovery Time Analysis
In contrast to response time, variation in temperature, humidity, and start state appeal' to have an
effect on HAZMATCAD Plus recovery time. Over all the testing, only SA was not associated
with a statistically significant finding.
A.3 Accuracy
The following sections present the results of analyses of the accuracy of HAZMATCAD Plus
response. The HAZMATCAD Plus was considered to be "accurate" under a given set of
conditions if the HAZMATCAD Plus:
1. Alarmed in the presence of a TIC or CW agent challenge
2. Correctly identified the TIC or CW agent
A.3.1 Effects of Temperature and Humidity on TIC Accuracy
The correct identification for each TIC was as follows:
1. AC: blood or choke (BLOD or CHOK)
2. CG: choke (CHOK)
3. Cl2: halogen (HALO)
4. SA: hydride (HYDR)
43
35"
281
20
m o 1
L M H L M H
I GB 1 I HD 1
A-ll
-------
The HAZMATCAD Plus performed with perfect accuracy (by the criteria above) with the TICs
under all levels of temperature and humidity. The test IDs for TIC accuracy are contained in
Table A-ll.
Table A-ll. IDs of Tests for Determining TIC Accuracy
AC-01-A
AC-01-B
AC-03-A
AC-03-B
AC-04-A
AC-04-B
AC-05-A
AC-05-B
AC-07-A
AC-07-B
CG-01-A
CG-01-B
CG-03-A
CG-03-B
CG-04-A
CG-04-B
CG-05-A
CG-05-B
CG-07-A
CG-07-B
Cl2-01-A
ci2-oi-b
Cl2-03-A
Cl2-03-B
Cl2-04-A
Cl2-04-B
Cl2-05-A
Cl2-05-B
Cl2-07-A
Cl?-07-B
SA-01-A
SA-01-B
SA-03-A
SA-03-B
SA-04-A
SA-04-B
SA-05-A
SA-05-B
SA-07-A
SA-07-B
A3.2 Effects of Temperature and Humidity on CW Agent Accuracy
Data available for the agent analysis of the effect of temperature on accuracy came from the tests
identified in Table A-6. The HAZMATCAD Plus performed with perfect accuracy under all
levels of temperature for both agents. Given 35 out of 35 accurate responses, a lower bound on
the chance of an accurate response with medium humidity under varying temperature conditions
is 92%.
Data available for the agent analysis of the effect of humidity on accuracy came from the tests
identified in Table A-7. For HD, the HAZMATCAD Plus performed with perfect accuracy
under all humidity conditions. For GB, one of the HAZMATCAD Plus units did not respond.
The other HAZMATCAD Plus unit performed with perfect accuracy under low and medium
humidity. However, all of the 5 high humidity runs were inaccurate due to the lack of a stable
response.
It would seem that the HAZMATCAD Plus has little chance of accurate performance in the
presence of GB at high humidity. However, only 5 runs were made. Hence an estimate of the
true chance of inaccuracy at high humidity has a large error. Given 5 out of 5 inaccurate
responses, a lower bound on the chance of inaccuracy with high humidity is only 55%.
A.3.3 Summary of Accuracy Analysis
Variation in temperature and humidity do not appear to affect the accuracy of the
HAZMATCAD Plus response to TICs; and variation in temperature does not appear to affect the
accuracy of the HAZMATCAD Plus response to agents. The HAZMATCAD Plus responded
with perfect accuracy throughout this testing.
A-12
-------
The humidity testing of agents, however, did not follow the same pattern. While the
HAZMATCAD Plus responded with perfect accuracy to HD under all humidity conditions, for
all the high humidity runs, response to GB was unstable.
A.4 Repeatability
The following sections contain the statistical analyses of the repeatability of HAZMATCAD
Plus response, response time, and recovery time. For TICs, see Table A-l in section A. 1.1 for
the test IDs used to determine the effects of temperature on repeatability; and Table A-3 in
section A. 1.2 contains the test IDs used to determine the effects of humidity on repeatability.
A.4.1 Repeatability of TIC Response
For testing repeatability of response, the mode (the most common response) of all responses
observed under a given condition was computed. Then the number of observed responses that
equal that value was determined. The proportion of responses equaling the most common
response was the measure for HAZMATCAD Plus response repeatability. The effects of
temperature and humidity were tested using a logit model (see Chapter 5 for more details).
For all TICs except Cl2, repeatability was perfect under all levels of temperature and humidity (a
high response was recorded consistently for AC, CG, and SA). Chlorine was associated with
more variability in response. Typically the HAZMATCAD Plus registered a medium response
with Cl2. For one of the 10 runs at high temperature, it responded with a low; and for 8 of the 10
runs at high humidity it responded with low. However, neither of the deviations was statistically
significant. The p-values for temperature and humidity for Cl2 were 0.32 and 0.10, respectively.
Given the mix of concentrations and TICs used in the repeatability analysis, there appears to be
no evidence for a temperature or humidity effect on HAZMATCAD Plus response repeatability.
A.4.2 Repeatability of TIC Response Time
A Brown-Forsythe test of equal variances was used to test the effect of temperature and
humidity on repeatability of response and recovery time. When there is a difference between the
variability in time for the different levels of temperature or humidity, there is evidence that
temperature or humidity has an effect on the repeatability of the response or recovery time.
For the TICS, there were no significant differences in variability of response time for the
different levels of temperature or the different levels of humidity.
A.4.3 Repeatability of TIC Recovery Time
Table A-12 contains the p-values for the effects of temperature on the repeatability of recovery
time.
A-13
-------
Table A-12. Effects of Temperature on the Repeatability of TIC Recovery Time
TI
C
AC
CG
Cl2
SA
P-value
0.13
0.34
0.01
0.37
The effect of temperature on the repeatability of recovery time is significant for Cl2. Figure A-7
illustrates the spread in observed recovery times with Cl2 through box plots. The figure indicates
that recovery time is most repeatable under medium temperature.
(/)
"O
c
o
o
CD
(/)
CD
E
CD
>
O
o
CD
DC
50
40
30
M
Temperature
Figure A-7. Box plots for recovery time by temperature for Cl2.
A-14
-------
Table A-13 contains the p-values for the effects of humidity on the repeatability of recovery
time. This table indicates that humidity is only significant for AC.
Table A-13. Effects of Humidity on the Repeatability of TIC Recovery Time
TI P-value
_C
AC <0.01
CG 0.46
Cl2 0.19
SA 0.94
Figure A-8 contains box plots for the observed times at different levels of humidity for AC.
Clearly the greatest variability in recovery time occurs under medium humidity.
230"
200~
190"
"J 160
OJ 150
QJ 130~
>
120"
M
Humidity
Figure A-8. Box plots for recovery time by humidity for AC.
A-15
-------
A.4.4 Repeatability of CW Agent Response
Data available for the agent analysis of the effect of temperature on the repeatability of response
came from the tests identified in Table A-6. For each agent, there was no evidence that
temperature affects repeatability of response. At each temperature level, the response was
perfectly consistent.
Data available for the agent analysis of the effect of humidity on the repeatability of response
came from the tests identified in Table A-7. At each humidity level, the response of the
HAZMATCAD Plus to HD was perfectly consistent. For GB at high humidity there was either
no response or the response was not stable, while at both medium and high temperature, the
response was perfectly consistent. In summary, the data from the agents provides no evidence
that humidity affects repeatability of response, with the caveat that at high humidity the
HAZMATCAD Plus appears to consistently have problems responding to GB.
A.4.5 Repeatability of CW Agent Response Time
Data available for the agent analysis of the effect of temperature on the repeatability of response
time came from the tests identified in Table A-6. With both GB and HD, there was no evidence
that temperature has an effect on repeatability of response time (p-values of 0.94 and 0.19,
respectively).
Data available for the agent analysis of the effect of humidity on the repeatability of response
time came from the tests identified in Table A-7. The runs used for the analysis were those used
for the analysis of the effect of humidity on response time. With both GB and HD, there was no
evidence that humidity affects repeatability of response time (p-values of 0.27 and 0.45,
respectively).
A.4.6 Repeatability of CW Agent Recovery Time
Data available for the agent analysis of the effect of temperature on the repeatability of recovery
time came from the tests identified in Table A-6. With both GB and HD, there was no evidence
that temperature has an effect on repeatability of recovery time (p-values of 0.91 and 0.07,
respectively).
Data available for the agent analysis of the effect of humidity on the repeatability of recovery
time came from the tests identified in Table A-7. The runs used for the analysis were those used
for the analysis of the effect of humidity on recovery time. With both GB and HD, there was no
evidence that humidity has an effect on repeatability of recovery time (p-values of 0.71 and
0.79, respectively).
A.4.7 Summary of Repeatability Analysis
Variation in temperature and humidity appear to have little affect on the repeatability of
HAZMATCAD Plus performance. Over all the repeatability testing, only two TICs were
associated with statistically significant findings, both in the testing of the repeatability of
A-16
-------
recovery time. Temperature appeared to have an effect on the repeatability of recovery from Cl2
and humidity on the repeatability of recovery from AC.
A.5. Effects of Temperature, Humidity, and Start State on Response
A.5.1 Effect of Temperature on TIC Response
The HAZMATCAD Plus response to each TIC was tested under medium humidity at low,
medium, and high temperature. The test IDs used in the following statistical analysis are
included in Table A-l in Section A. 1.1. A high response was recorded for all tests under all
temperatures for every TIC except Cl2. Typically the HAZMATCAD Plus registered a medium
response with Cl2. For one of the 10 runs at high temperature, it responded with a low. A
Jonckheere-Terpstra test was used to test the effect of temperature on HAZMATCAD Plus
response to Cl2 (see Chapter 5 for more details). The p-value for this test was 0.67, indicating
that there is no evidence that temperature has a significant effect on the HAZMATCAD Plus
response.
A.5.2 Effect of Humidity on TIC Response
The HAZMATCAD Plus response to each TIC also was tested under medium temperature at
low, medium, and high humidity. The test IDs used in the following statistical analysis are
included in Table A-3 in Section A. 1.2. As in the temperature tests, a high response was
recorded for all tests under all temperatures for every TIC except Cl2. A Jonckheere-Terpstra test
was used to test the effect of humidity on HAZMATCAD Plus response to Cl2 (see Chapter 5 for
more details). The p-value for this test was <0.01, indicating that humidity has a significant
effect on HAZMATCAD Plus response. Figure A-9 contains the counts for each response by
level of humidity for Cl2.
A-17
-------
10
D
O
o
o
m
i
8
iff
1
Iff
1
IS
M
2
o
1
o o
O
O
L M H
L M H
I M 1
L M H
Response
Humidity Level
Figure A-9. Response to Cl2 as a function of humidity.
Figure A-9 indicates that the typical HAZMATCAD Plus response is lower for the highest level
of humidity.
A.5.3 Effect of Start State on TIC Response
The HAZMATCAD Plus response to AC was recorded under medium temperature and humidity
with three start states:
1. Cold soak/cold start
2. Hot soak/cold start
3. Room temperature/cold start
These results were combined with the responses from the medium humidity and medium
temperature AC results for the sake of comparison. The highest response was recorded for all
start states and for the control. No effect of start state could be detected for AC run at 1 IDLH.
(See Table A-5 in Section A. 1.3 for a list of test IDs included in this analysis.)
A.5.4 Effect of Temperature on CW Agent Response
Data available for the agent analysis of the effect of temperature on HAZMATCAD Plus
response came from the tests identified in Table A-6. For HD, response was a consistent high for
all runs within each temperature level. Thus for HD there was no evidence of an effect of
A-18
-------
temperature on HAZMATCAD Plus response. For GB, at low temperature the response was
consistently high, while at medium and high temperature the response was consistently at the
medium level. These data provided sufficient statistical evidence to conclude that temperature
has an effect on HAZMATCAD Plus response to GB (p-value <0.01).
A.5.5 Effect of Humidity on CW Agent Response
Data available for the agent analysis of the effect of humidity on HAZMATCAD Plus response
came from the tests identified in Table A-7. For HD, response was a consistent high for all runs
within each humidity level. Thus for HD there was no evidence of an effect of humidity on
HAZMATCAD Plus response. At high humidity there was either no response to GB or the
response was not stable for the HAZMATCAD Plus unit that typically responded to the presence
of GB. At low and medium humidity the response was consistently at the medium level. These
data provide sufficient statistical evidence to conclude that humidity has an effect on
HAZMATCAD Plus response to GB (p-value <0.01).
A.5.6 Summary of Response Analysis
Variation in temperature, humidity, and start state appear to have little affect on HAZMATCAD
Plus response to TICs. Over all TIC testing, the only significant finding was for Cl2 with
variation in humidity.
For the agent testing, variation in temperature and humidity appeared to have little affect on
HAZMATCAD Plus response to HD. However, both temperature and humidity appeared to
affect response to GB.
A.6 Interference Effects
The HAZMATCAD Plus response, response time, and recovery time were tested under medium
temperature and humidity with each of the following interferents, both in the absence of any TIC
or CW agent and in the presence of each TIC and CW agent.
1. Latex paint fumes
2. Floor cleaner vapors
3. Air freshener vap ors
4. Gasoline engine exhaust
5. DEAE
Results for each TIC and CW agent without interferent present served as the control data for the
interferent results. The following sections summarize the statistical analyses of the effect of
interferents on HAZMATCAD Plus response, response time, and recovery time. The test IDs
included in the analysis for TICs are contained in Table A-14 below; the corresponding test IDs
for the CW agents are shown in Table A-17.
A-19
-------
Table A-14. IDs of Tests included in the Analysis of Interferents with TICs
AC-01-A
AC-01-B
AC-09-A
AC-09-B
AC-10-A
AC-10-B
AC-11-A
AC-ll-B
AC-12-A
AC-12-B
AC-13-A
AC-13-B
CG-01-A
CG-01-B
CG-09-A
CG-09-B
CG-10-A
CG-10-B
CG-ll-A
CG-ll-B
CG-12-A
CG-12-B
CG-13-A
CG-13-B
Cl2-01-A
ci2-oi-b
Cl2-09-A
Cl2-09-B
Cl2-10-A
ci2-io-b
Cl2-ll-A
Cl2-ll-B
Cl2-12-A
Cl2-12-B
Cl2-13-A
Cl2-13-B
SA-01-A
SA-01-B
SA-09-A
SA-09-B
SA-10-A
SA-10-B
SA-ll-A
SA-ll-B
SA-12-A
SA-12-B
SA-13-A
SA-13-B
A. 6.1 Effect of Interferent on TIC Response
As described in Chapter 5, the effects of interferent on HAZMATCAD Plus response were tested
using a Kruskal-Wallis test. The highest response was recorded for all interferents for all TICs
except Cl2. Thus, no effect of interferent could be determined for AC, CG, and SA. For Cl2, a
p-value of < 0.01 indicates that interferent had a significant effect on HAZMATCAD Plus
response. Figure A-10 contains the response by interferent counts for Cl2. It is apparent from the
figure that air freshener may reduce the level of response to Cl2, while floor cleaner appears to
increase the response level. It should be noted that according to Dunn's multiple comparison
procedure (see Chapter 5), neither of the response levels for these two interferents is
significantly different from the control. Thus, while there is an overall effect of interferent, there
is insufficient power to detect differences between the interferents and the control.
A-20
-------
10
o
o
o
9
»8 8
1
i
1
1
9
i
I
9
i
1
2
I
1
1
o
o
o
o
o o
O
I
o o
o
LMH LMH LMH LMH LMH LMH Response
Gbntrol O eaner CE°E Exhaust Freshener Ffei nt I interfere fit
Figure A-10. Response to Cl2 by interferent.
A. 6.2 Effect of Interferent on TIC Response Time
An ANOVA model was used to analyze the effect of interferent on response time. Table A-15
contains the p-values for these effects.
Table A-15. Effects of Interferent on TIC Response Time
TI P-value
£
AC 0.77
CG 0.46
Cl2 <0.01
SA 0.07
As evidenced by Table A-15, interferent has a significant effect on the response time for Cl2.
Figure A-11 contains the modeled geometric means of the Cl2 response times. Floor cleaner
appeared to increase response time to Cl2, while the other interferents had little effect on
response time.
A-21
-------
A.6.3 Effect of Interferent on TIC Recovery Time
An ANOVA model was used to analyze the effect of interferent on recovery time for each TIC.
Control
Cleaner
DEAE
Exhaust Freshener
Paint
Interferent
Figure A-11. Modeled geometric means of response time by interferent for Cl2.
Table A-16 presents the p-values for these effects.
Table A-16. Tests for Effects of Interferent on Recovery Time
TIC
P-value
AC
CG
Cl2
SA
<0.01
<0.01
0.02
0.19
As evidenced by Table A-16, interferent has an effect on the recovery time for AC, CG, and Cl2
Figures A-12, A-13, and A-14 contain the geometric mean recovery times by interferent for AC,
CG, and Cl2, respectively. For all of these TICs, interferents either had little effect on or tended
to increase recovery time.
A-22
-------
259
Control Cleaner DEAE Exhaust Freshener Paint
Interferent
Figure A-12. Modeled geometric means of recovery time by interferent for AC.
Freshener Paint
time by interferent for CG.
o
Control Cleaner DEAE Exhaust
Interferent
Figure A-13. Modeled geometric mean of recovery
A-23
-------
Control Cleaner DEAE Exhaust Freshener Paint
Interferent
Figure A-14. Modeled geometric mean of recovery time by interferent for Cl2.
Upon review of the interferent recovery times, a trend was apparent for AC. Figure A-15
presents AC recovery time by test sequence number for each interferent. It is evident that
recovery times tended to increase with each successive challenge. One of the assumptions of the
ANOVA model is that the successive tests are independent. The figure suggests that this
assumption was violated for AC testing. The p-value associated with AC in Table A-16 should
be interpreted with caution.
A.6.4 Effect of Interferent on CW Agent Response
The tests used to assess the effect of interferents on HAZMATCAD Plus performance are
identified in Table A-17. For GB there were 5 runs for each of the five interferents. For HD there
were 10 runs for four of the interferents and 12 for the fifth (paint).
For both GB and HD, interferents appeal' to have a significant effect on the response of the
HAZMATCAD plus (p-value <0.01 for both agents). The HAZMATCAD plus did not respond
to GB in the presence of air freshener and paint; and in two of the exhaust runs, response was
not stable. The HAZMATCAD plus did not respond to HD in the presence of ammonia cleaner
and paint.
For those interferents that were associated with a machine response, there was no statistically
significant difference between the response of HAZMATCAD with or without the interferent.
A-24
-------
400
t—s
in
"~
g
fcj 300
in
^^
QJ
E
f—
>N
§| 200
o
u
QJ
(Y.
100
O
Figure A-15. AC recovery time by test sequence number.
Table A-17. IDs of Tests included in the Agent Testing of the Effect of Interferents on
Performance
GB-01-BA1
HD-10A-1
GB-10-1
HD-10B-1
GB-ll-B
HD-11A-1
GB-12-1
HD-11B-1
GB-13-1
HD-12A-1
GB-9-1
HD-12B-1
HD-13A-1
HD-13B-1
HD-1A-1
HD-1B-1
HD-9A-1
HD-9B-1
Sequence Number
I nt erf er ent :
1 • a eaner
* Exhaust
1 * Rai nt
CEAE
Fr eshener
A-25
-------
A.6.5 Effect of Interferent on CW Agent Response Time
Because the HAZMATCAD plus did not respond in the presence of some interferents, response
times were collected for only a subset of the interferents. For GB, analysis of the effect of
interferent on response time was limited to the runs for ammonia cleaner, DEAE, and exhaust;
for HD, analysis of the effect of interferent on response time was limited to the runs for air
freshener, DEAE, and exhaust.
The data for GB provided no evidence of interferent effect on the HAZMATCAD response time
(p-value=0.18). However the data for HD did (p-value <0.01). Figure A-16 contains the
modeled geometric means of the response time to HD in the presence of air freshener, DEAE,
and exhaust. DEAE and exhaust decreased response time, while air freshener increased response
time.
~u
c
o
u
0
w
a;
E
c
~
Q_
a:
Gbnt rol
Exfiaust
Freshener
Interferent
Figure A-16. Modeled geometric mean of response time to HD by
interferent.
A.6.6 Effect of Interferent on CW Agent Recovery Time
As with response times, recovery times were collected for only a subset of the interferents. For
GB, analysis of the effect of interferent on recovery time was limited to runs for ammonia
cleaner, DEAE, and exhaust. Two of the five exhaust runs were met with unstable response and
no associated recovery time. For HD, analysis of the effect of interferent on recovery time was
limited to the runs for air freshener, DEAE, and exhaust.
The data for HD provided no evidence of interferent effect on the HAZMATCAD recovery time
(p-value=0.95). However the data for GB did (p-value <0.01). Figure A-17 contains the
modeled geometric means of the recovery time to GB in the presence of ammonia cleaner,
A-26
-------
DEAE, and exhaust. Recovery in the presence of exhaust was shorter than that required from
GB without the interferent.
A.6.7 Summary of Interferent Analysis
Presence of interferents appeal's to have an effect on HAZMATCAD Plus performance. The
effect of interferents was most pronounced for agents, where there was no response to GB in the
presence of air freshener and paint, and no response to HD in the presence of ammonia cleaner
and paint.
<0
TJ
a
o
o
0)
>
o
Gbrrt rol <3 eaner
Interferent
Figure A-17. Modeled geometric mean of recovery time from GB by
interferent.
A.7 Analysis of False Positives
The machine response was tested under medium temperature and humidity without TICs in the
presence of each of the following interferents:
1. Latex paint fumes
2. Floor cleaner vapors
3. Air freshener vapors
4. Gasoline engine exhaust
5. DEAE
A false-positive was defined as any machine response. The number of false positive readings
was recorded, and Clopper-Pearson 95% confidence intervals were constructed for the
proportion of false positives for each interferent. Table A-18 contains the test IDs included in
this analysis.
A-27
-------
Table A-18. IDs of Tests included in the False Positive Analysis
AC-14-A
AC-14-B
AC-15-A
AC-15-B
AC-16-A
AC-16-B
AC-17-A
AC-17-B
AC-18-A
AC-18-B
The HAZMATCAD Plus performed perfectly during the false positive testing. Over the range of
interferents, there was never any machine response. Thus an estimate of the chance of false
positive is 0%. Six tests were performed for each interferent. For a given interferent, the
Clopper-Pearson upper bound on the chance of a false positive is about 46%.
A.8 Effect of Oscillation on Response
Twelve cycles were conducted in which the HAZMATCAD Plus alternated sampling from high
and low challenge plenums. For the TICs, the high challenge was IDLH, and the low challenge
was as follows: AC at 0.1 DLH, CG at 0.2 DLH, Cl2 at 0.1 DLH, and SA at 0.2 DLH. In six
of the 12 cycles, the high plenum was sampled first, then the low plenum; in the other six, the
order was reversed. Clean air was sampled before the first cycle and again after every high/low
or low/high cycle. This test with alternating concentrations was conducted only at the medium
temperature and humidity levels. Test IDs included in this analysis are contained in Table A-19.
Table A-19. IDs of Tests included in the Oscillation Analysis
AC-01-A CG-01-A Cl2-01-A SA-01-A
AC-01-B CG-01-B Cl2-01-B SA-01-B
AC-06-A CG-06-A Cl2-06-A SA-06-A
AC-06-B CG-06-B Cl2-06-B SA-06-B
A-28
-------
A. 8.1 Effect of Order of Challenge Levels on TIC Response
One of the purposes of this testing was to assess whether HAZMATCAD Plus response to a
given concentration was affected by initial exposure to an alternate concentration. The effect of
an initial alternate concentration was investigated using a Cochran-Mantel-Hansel statistic (see
Chapter 5 for more details). Table A-20 contains the p-values for these tests.
Table A-20. Effects of Order of Challenge Levels on TIC Response
TI P-value
C
AC 0.29
CG 0.95
Cl2 0.34
SA 0.95
According to Table A-20, there is no evidence that machine response to a given concentration is
affected by preceding alternate concentrations.
A.8.2 Difference in Response to the Challenge Levels
When challenged by a high concentration after being challenged by a low concentration, the
HAZMATCAD Plus response might be expected to increase. Similarly, when challenged by low
concentration after being challenged by a high concentration, the response might be expected to
decrease. The proportion of tests exhibiting this behavior for each TIC was recorded. In the case
of AC, CG, and SA, all tests performed as expected (12 per gas). However, in the case of Cl2,
only one of the 12 tests performed as expected.
For AC, CG, and SA, a Clopper-Pearson lower bound was placed on the probability that the
machine response would perform as expected. The lower bound is about 74%. For Cl2 it appears
likely that the expected response will not be observed. The chance that expectations will not be
met with Cl2 has a lower bound of about 62%.
A.8.3 Effect of Oscillation on CW agents
The tests used to assess the effect of fluctuating concentration on HAZMATCAD Plus
performance are identified in Table A-21. For GB, 6 cycles of alternating concentration were
conducted; for HD, 12 were conducted. The high concentration challenge for GB was 11 IDLH;
the low was 4 IDLH. The high concentration challenge for HD was 7 AEGL-2 ; the low was
2 AEGL-2.
A-29
-------
Table 21. IDs of Tests included in the Agent Testing of Fluctuating Concentration
GB-01- HD-1A-1
BA1
GB-6-1 HD-1B-1
HD-6A-1
HD-6B-1
There was no evidence that the HAZMATCAD Plus response to a given concentration of GB
was affected by an initial alternate concentration. However, there was evidence that the
HAZMATCAD Plus response to a given concentration of HD was affected by an initial alternate
concentration (p-value<0.01). The response to HD at high concentration was consistently high.
But the response to HD at low concentration tended to be higher when preceded by the alternate
concentration.
When challenged by a high concentration after being challenged by a low concentration or visa
versa, the HAZMATCAD Plus response to both agents was as expected: the response was
consistently higher for the former condition and consistently lower for the latter condition. A
Clopper-Pearson lower bound was placed on the probability that the machine response would
perform as expected to each of the gases. For GB with 6 observations the lower bound is 61%.
For HD with 12 observations the lower bound is 74%.
A.8.4 Summary of the Oscillation Analysis
HAZMATCAD Plus response appears to be little affected by a preceding challenge of different
concentration. An exception was found during the HD testing, where response to low
concentration was elevated when preceded with a higher concentration.
It appears that, with fluctuating concentrations, the HAZMATCAD Plus does what you might
expect: when challenged by a high concentration after being challenged by a low concentration,
the response seems to increase. Similarly, when challenged by low concentration after being
challenged by a high concentration, the response seems to decrease.
A-30
------- |