USING MULTIVARIATE STATISTICAL PROCEDURES TO IDENTIFY IGNITABLE
LIQUID RESIDUES IN THE PRESENCE OF INTERFERENCES
By
Kaitlin Prather

A THESIS
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
MASTER OF SCIENCE
Criminal Justice
2011

Abstract
USING MULTIVARIATE STATISTICAL PROCEDURES TO IDENTIFY IGNITABLE
LIQUID RESIDUES IN THE PRESENCE OF INTERFERENCES
By
Kaitlin Prather
Gas chromatography – mass spectrometry (GC-MS) is a common technique used in the
analysis of fire debris. In this approach, the chromatogram of an ignitable liquid residue (ILR)
extracted from the debris is visually compared to chromatograms of neat liquids in a reference
collection. However, the association of the ILR to the neat liquid can be affected by factors such
as matrix interferences, evaporation of the ignitable liquid, and thermal degradation. This
research aims to develop an objective method for associating ILRs to the corresponding ignitable
liquid standard using multivariate statistical procedures. The combination of statistical
procedures removes subjectivity in visual comparison of chromatograms, while enabling a
statistical measure of the association between the ILR and the corresponding liquid standard.
In this study, liquid standards of neat gasoline and kerosene, as well as each liquid at two
different evaporation levels, were prepared. The neat and evaporated liquids were spiked onto
unburned and burned samples of two different household matrices (nylon carpet and high density
polyethylene). In addition, separate samples of each matrix were spiked with the ignitable
liquids and then burned using a propane torch to simulate fire debris.
A combination of principal components analysis (PCA) and Pearson product moment
correlation (PPMC) coefficients was used to investigate the association of ILRs to the
corresponding liquid standards. PCA was used to assess the association of the ILRs to the
corresponding liquid standard, while PPMC coefficients provided a statistical measure of that
association despite matrix interferences, evaporation effects, and thermal degradation.
ii

Acknowledgements

I would like to thank all of the people who supported me and helped me through my time
at Michigan State. Thank you, first and foremost, to Dr. Ruth Smith for being a wonderful
adviser. Without your guidance and not-so-gentle nudges from time to time, these two years
would not have been nearly as worthwhile or fun!
A huge thank you to all the other members of the Forensic Science Program for their
friendship and support over the past two years. Thanks to John McIlroy, Monica Bugeja, Beth
Shattuck, Tiffany Van De Mark, Emily Riddell, Suzanne Towner, Karlie McManaman, Christy
Hay, Ruth Udey, Melissa Bodner, Seth Hogg, and Kari Anderson for being a generally
wonderful group of people. The only reason I survived getting this degree and remained
reasonably sane is because of all of you!
I have to thank my family for their love, support, and most of all patience while I’ve been
pursuing my Master’s degree. Thank you for pushing and encouraging me, despite all the rough
spots and gablargy-blargs along the way.
Finally, I would like to thank Dr. Bill Terrill for taking the time to sit on my committee.
Thank you also to Dr. Victoria McGuffin both for being on my committee as well as for the
expertise and help you have given me during my research.

iii

Table of Contents

List of Tables

vii

List of Figures

ix

Chapter 1 – Introduction
1.1 Classification of Ignitable Liquids
1.2 Extraction of Ignitable Liquids from Fire Debris
1.3 Analysis of Ignitable Liquids
1.4 Problems in Identifying Ignitable Liquids Residues in Fire Debris
1.5 Literature Review
1.5.1 Matrix Interferences
1.5.2 Evaporation of Ignitable Liquids
1.5.3 Multivariate Statistical Analysis of Ignitable Liquids
1.5.4 Data Pretreatment Procedures
1.6 Research Objectives and Goals
References

1
2
5
7
11
12
15
18
23
25
28

1

Chapter 2 – Instrumental and Statistical Techniques
2.1 Gas Chromatography-Mass Spectrometry
2.2 Data Pretreatment Procedures
2.2.1 Background Subtraction
2.2.2 Savitzky-Golay Smoothing
2.2.3 Retention Time Alignment
2.2.4 Normalization
2.3 Principal Components Analysis
2.4 Pearson Product Moment Correlation Coefficients
References

31
45
45
47
50
53
55
58
60

31

Chapter 3 - Using Multivariate Statistical Procedures to Identify Ignitable Liquid Residues
in the Presence of Matrix Interferences from Carpet
62
3.1 Introduction
62
3.2 Materials and Methods
63
3.2.1 Ignitable Liquid Collection
63
3.2.2 Ignitable Liquid Standard Preparation
63
3.2.3 Inherent Matrix Interference Sample Preparation
64
3.2.4 Burned Matrix Interference Sample Preparation
65
3.2.5 Thermal Degradation Sample Preparation
65
3.2.6 GC-MS Analysis
66
3.2.7 Data Pretreatment
66
3.2.8 Data Analysis
68
3.3 Results and Discussion
69
3.3.1 Selection of Retention-Time Alignment Parameters
69
iv

3.3.2
3.3.3
3.3.4
3.3.5

Selection of Normalization Method
70
Visual Assessment of Ignitable Liquid Standard Chromatograms
73
Association and Discrimination of Neat and Evaporated Liquid Standards
75
Association and Discrimination of Neat and Evaporated Liquid Standards using PPMC
Coefficients
79
3.3.6 Identification of Inherent Matrix Interferences
81
3.3.7 Association and Discrimination of Samples in the Presence of Inherent Matrix
Interferences
84
3.3.8 Assessment of Association and Discrimination of Samples in the Presence of Inherent
Matrix Interferences Using PPMC Coefficients
87
3.3.9 Identification of Burned Matrix Interferences and Selection of Burn Time
90
3.3.10 Association and Discrimination of Samples in the Presence of Burned Matrix
Interferences
94
3.3.11 Assessment of Association and Discrimination of Samples in the Presence of Burned
Matrix Interferences Using PPMC Coefficients
104
3.3.12 Association and Discrimination of Samples in the Presence of Thermal
Degradation
108
3.3.13 Assessment of Association and Discrimination of Samples in the Presence of Thermal
Degradation Using PPMC Coefficients
112
3.4 Conclusions from Carpet Matrix Study
119
References
122
Chapter 4 - Using Multivariate Statistical Procedures to Identify Ignitable Liquid Residues
in the Presence of Matrix Interferences from Plastic
124
4.1 Introduction
124
4.2 Materials and Methods
125
4.2.1 Ignitable Liquid Collection and Standard Preparation
125
4.2.2 Inherent Matrix Interference Sample Preparation
125
4.2.3 Burned Matrix Interference Sample Preparation
126
4.2.4 Thermal Degradation Sample Preparation
126
4.2.5 GC-MS Analysis
127
4.2.6 Data Pretreatment and Analysis
127
4.3 Results and Discussion
127
4.3.1 Selection of Retention-Time Alignment Parameters
127
4.3.2 Selection of Normalization Method
128
4.3.3 Association and Discrimination of Neat and Evaporated Liquid Standards
128
4.3.4 Association and Discrimination of Neat and Evaporated Liquid Standards using
PPMC Coefficients
133
4.3.5 Identification of Inherent Matrix Interferences
133
4.3.6 Association and Discrimination of Samples in the Presence of Inherent Matrix
Interferences
138
4.3.7 Assessment of Association and Discrimination of Samples in the Presence of Inherent
Matrix Interferences Using PPMC Coefficients
143
4.3.8 Identification of Burned Matrix Interferences and Selection of Burn Time
149
4.3.9 Association and Discrimination of Samples in the Presence of Burned Matrix
Interferences
153
v

4.3.10 Assessment of Association and Discrimination of Samples in the Presence of Burned
Matrix Interferences Using PPMC Coefficients
159
4.3.11 Association and Discrimination of Samples in the Presence of Thermal
Degradation
165
4.3.12 Assessment of Association and Discrimination of Samples in the Presence of Thermal
Degradation Using PPMC Coefficients
170
4.4 Conclusions from Plastic Matrix Study
176
References
178
Chapter 5 – Conclusions and Future Work
5.1 Summary of Research
5.1.1 Research Objectives and Goals
5.1.2 Carpet Matrix Study Summary
5.1.3 Plastic Matrix Study Summary
5.2 Future Work
References

180
180
180
183
184
186
191

vi

List of Tables

Table 1.1: ASTM classes for ignitable liquids and chemical composition of each class

3

Table 1.2: Table 1.2: Examples of ignitable liquids in each of the ASTM classes

4

Table 1.3: Common extracted ion profiles used to identify ignitable liquids

6

Table 3.1: Mean Pearson product moment correlation coefficients (± standard deviation) for all
replicates (n=15) of gasoline and kerosene liquid standards correlated to replicates of each
standard (n=225).
80
Table 3.2: Pearson product moment correlation coefficients for replicates (n=15) of each
ignitable liquid extracted from unburned carpet and the range of coefficients for the extracts
correlated to the corresponding ignitable liquid standard.
89
Table 3.3: Mean Pearson product moment correlation coefficients (n=225) for 70% evaporated
kerosene extracts correlated to each of the ignitable liquid standards.
89
Table 3.4: Pearson product moment correlation coefficients for replicate (n=15) extractions of
each ignitable spiked onto burned carpet and the range of coefficients for the extracts correlated
to the corresponding ignitable liquid standard.
105
Table 3.5: Comparison of mean Pearson product moment correlation coefficients (± standard
deviation) calculated between extracts of burned carpet spiked with kerosene at each evaporation
level and the corresponding ignitable liquid standards.
107
Table 3.6: Pearson product moment correlation coefficients for replicates (n=15) of each
ignitable liquid thermal degradation extract and the range of coefficients for the extracts
correlated to the corresponding ignitable liquid standard. (* - indicates coefficients calculated
excluding replicates that contained no ILR)
115
Table 3.7: Comparison of Pearson product moment correlation coefficients for extracts (n=15)
containing gasoline ILRs correlated the corresponding ignitable liquid standard as well as the
90% evaporated gasoline and neat kerosene standards. (* - indicates coefficients calculated
excluding replicates that contained no ILR)
117
Table 3.8: Comparison of mean Pearson product moment correlation coefficients (± standard
deviation) for clusters 1 and 2 of 90% evaporated gasoline extracts correlated to each other and
the two standards.
120
Table 3.9: Comparison of Pearson product moment correlation coefficients for extracts (n=15)
containing kerosene ILRs correlated the corresponding ignitable liquid standard as well as the
10% evaporated and 70% evaporated kerosene standards. (* - indicates coefficients calculated
excluding replicates that contained no ILR)
120
vii

Table 4.1: Mean Pearson product moment correlation coefficients (± standard deviation) for
replicates (n=15) of gasoline standards.
134
Table 4.2: Mean Pearson product moment correlation coefficients (± standard deviation) for
replicates (n=15) of gasoline standards.
134
Table 4.3: Pearson product moment correlation coefficients for replicates (n=15) of each
ignitable liquid extracted from unburned high density polyethylene and the range of coefficients
for the extracts correlated to the corresponding ignitable liquid standard.
144
Table 4.4: Mean Pearson product moment correlation coefficients (± standard deviation) for
replicates (n=15) of residues of kerosene at each evaporation level extracted from unburned high
density polyethylene correlated to the neat gasoline and neat kerosene ignitable liquid standards
(n=225).
148
Table 4.5: Pearson product moment correlation coefficients for replicate (n=15) extractions of
each ignitable spiked onto burned high density polyethylene and the mean (± standard deviation)
and range of coefficients for the extracts correlated to the corresponding ignitable liquid
standard.
160
Table 4.6: Comparison of mean Pearson product moment correlation coefficients (± standard
deviation) calculated between extracts of burned high density polyethylene spiked with all
evaporation levels of gasoline to the gasoline standards.
162
Table 4.7: Comparison of mean Pearson product moment correlation coefficients (± standard
deviation) calculated between extracts of burned high density polyethylene spiked with all
evaporation levels of kerosene to standards for the corresponding liquid, neat kerosene, and neat
gasoline.
164
Table 4.8: Pearson product moment correlation coefficients for replicates (n=15) of each
ignitable liquid thermal degradation extract and the range of coefficients for the extracts
correlated to the corresponding ignitable liquid standard.

171

Table 4.9: Comparison of mean Pearson product moment correlation coefficients for extracts
(n=15) containing neat and 10% evaporated gasoline ignitable liquid residues correlated the
corresponding ignitable liquid standard (n=225) as well as the neat kerosene standard
(n=225).
173
Table 4.10: Comparison of mean Pearson product moment correlation coefficients for kerosene
ignitable liquid residue extracts (n=15) correlated to kerosene standards at each level of
evaporation and neat gasoline standard (n=225).
175

viii

List of Figures

Figure 1.1: Representative chromatogram of neat gasoline. Major compounds are labeled: a)
ethylbenzene, b) p-xylene, c) o-xylene, d) propylbenzene, e) m-ethyltoluene, f) pethyltoluene, g) o-ethyltoluene, and h) 1,3,5-trimethylbenzene.
8
Figure 1.2: Representative chromatograms of neat gasoline (──) and neat gasoline ILR
extracted from burned plastic (- - -) demonstrating the change in peak ratios of the C2alkylbenzenes due to the addition of matrix interferences from the plastic.

10

Figure 2.1: Schematic diagram of a gas chromatograph.

32

Figure 2.2: Schematic of the inlet of a gas chromatograph.

34

Figure 2.3: Diagram illustrating A) band broadening due to molecular diffusion while the
analyte molecules travel through the column from t1 to t3 and B) the resultant broadening of
the chromatographic peak.
36
Figure 2.4: Diagram showing mass transfer of the analyte molecules between the mobile and
stationary phases at A) equilibrium and B) when the flow rate of the carrier gas is too high
for equilibrium to be reached, resulting in band broadening, and the corresponding
chromatographic peaks.
37
Figure 2.5: Diagram of an ion source demonstrating electron ionization.

40

Figure 2.6: Diagram of a quadrupole mass analyzer showing resonant and non-resonant
ions.

42

Figure 2.7: Diagram of continuous-dynode electron multiplier.

44

Figure 2.8: A) TIC of neat kerosene and B) the mass spectrum of decane (indicated in the
TIC with an arrow) showing the molecular ion, base peak, and fragment ions.

46

Figure 2.9: Total ion chromatograms of neat gasoline A) before and B) after background
subtraction of the caprolactam peak (indicated with arrow).

48

Figure 2.10: Portion of the total ion chromatogram of neat gasoline before smoothing (─)
and after application of the Savitzky-Golay algorithm (‐ ‐ ‐), demonstrating reduction of
high-frequency noise in the peaks.

49

Figure 2.11: Total ion chromatograms of 90% evaporated gasoline standard (──) and
extract from a 90% evaporated gasoline thermal degradation sample (─ ─) A) demonstrating

ix

the shift in retention times for the C3-alkylbenzenes, and B) those same peaks after retentiontime alignment.
52
Figure 2.12: Portion of total ion chromatograms (n=15) of neat kerosene showing the C10
normal alkane peak both A) before and B) after normalization of the data.

54

Figure 3.1: A) Misaligned 1-methylnaphthalene peak in the total ion chromatograms (TICs)
of 90% evaporated gasoline replicates (n=15) using a warp size of 5 points and a segment
size of 60 points. B) Well-aligned 1-methylnaphthalene peak in the TICs of 90% evaporated
gasoline using a warp size of 2 points and a segment size of 75 points.
71
Figure 3.2: Comparison of A) aligned but not normalized p-xylene peak in the total ion
chromatograms (TICs) of neat (─), 10% evaporated (─ • ─), and 90% evaporated gasoline
(‐‐‐) and B) p-xylene peak in the TICs of neat and evaporated gasoline after internal
standard and total area normalization, showing the minimized spread among replicates of the
three evaporation levels.
72
Figure 3.3: Total ion chromatograms (TICs) of A) neat gasoline, B) 10% evaporated
gasoline, C) 90% evaporated gasoline, D) neat kerosene, E) 10% evaporated kerosene, and F)
70% evaporated kerosene. Major compounds are labeled: a) ethylbenzene, b) p-xylene, c) oxylene, d) propylbenzene, e) m-ethyltoluene, f) p-ethyltoluene, g) o-ethyltoluene, and h)
1,3,5-trimethylbenzene. IS indicates the internal standard, nitrobenzene.
74
Figure 3.4: Scores plot of the first principal component (PC1) versus the second principal
component (PC2) based on the total ion chromatograms for the two ignitable liquids at three
levels of evaporation, denoted as follows: Neat gasoline ( ), 10% evaporated gasoline ( ),
90% evaporated gasoline ( ), neat kerosene ( ), 10% evaporated kerosene ( ), and 70%
evaporated kerosene ( ).
76
Figure 3.5: Loadings plots of A) the first principle component and B) the second principle
component based on the total ion chromatograms of the six ignitable liquid standards. Major
compounds are labeled: a) ethylbenzene, b) p-xylene, c) o-xylene, d) propylbenzene, e) methyltoluene, f) p-ethyltoluene, g) o-ethyltoluene, and h) 1,3,5-trimethylbenzene. IS indicates
the internal standard, nitrobenzene.
77
Figure 3.6: Total ion chromatogram of unburned carpet.

82

Figure 3.7: Total ion chromatograms of carpet spiked with A) neat gasoline and B) neat
kerosene showing the addition of the C9-C12 branched alkanes from the matrix. Major
compounds are labeled: a) ethylbenzene, b) p-xylene, c) o-xylene, d) propylbenzene, e) methyltoluene, f) p-ethyltoluene, g) o-ethyltoluene, and h) 1,3,5-trimethylbenzene. IS indicates
the internal standard, nitrobenzene.
83
Figure 3.8: Scores plot of first principle component (PC1) versus the second principle
component (PC2) based on the total ion chromatograms for the six ignitable liquid standards
x

and unburned carpet projections. Each standard is denoted as follows: Neat gasoline ( ),
10% evaporated gasoline ( ), 90% evaporated gasoline ( ), neat kerosene ( ), 10%
evaporated kerosene ( ), and 70% evaporated kerosene ( ).

85

Figure 3.9: Total ion chromatograms of extracts of carpet burned for A) 0 seconds, B) 10
seconds, C) 20 seconds, D) 30 seconds, E) 60 seconds, and F) 120 seconds. Major
interference compounds are identified.

91

Figure 3.10 A) Portion of total ion chromatograms of burned carpet spiked with neat gasoline
(─) versus the neat gasoline standard (─ • ─), showing the contribution of the styrene peak
from the burned carpet to the C2-alkylbenzenes in gasoline. B) Portion of total ion
chromatograms of burned carpet spiked with neat kerosene (─) versus the neat kerosene
standard (─ • ─), showing the contribution of the C15 branched alkane peaks from the burned
carpet after the C13 normal alkane peak in kerosene.

93

Figure 3.11: Scores plot of first principle component (PC1) versus the second principle
component (PC2) based on the total ion chromatograms for the six ignitable liquid standards
and projected scores for burned carpet spiked with each ignitable liquid. Each standard is
denoted as follows: Neat gasoline ( ), 10% evaporated gasoline ( ), 90% evaporated
gasoline ( ), neat kerosene ( ), 10% evaporated kerosene ( ), and 70% evaporated kerosene
95
( ).
Figure 3.12: A) Total ion chromatogram (TIC) of burned carpet spiked with neat gasoline,
showing the addition of the C15 branched alkanes from the burned carpet, and B) the same
TIC after mean centering, showing no corresponding peaks.

96

Figure 3.13: A) Portion of the total ion chromatogram of the neat gasoline standard (─) and
an extract of neat gasoline spiked onto burned carpet (─ ─), demonstrating the difference in
abundances of the C3-alkylbenzenes, and the effect of this difference on the B) meancentered data, and C) loadings on PC1, which results in the spread observed in the scores
plot.
98
Figure 3.14: Demonstration of the difference in abundances of the C3-alkylbenzenes
between the 90% evaporated gasoline standard (─) and an extract of 90% evaporated
gasoline spiked onto the burned carpet (─ ─) in the A) total ion chromatogram, B) meancentered data, and C) loadings on the first principle component, which results in the close
positioning of the spiked burned carpet extracts and the standards in the scores plot.

100

Figure 3.15: Demonstration of the difference in abundances of the C12-C15 normal alkanes
between the neat kerosene standard (─) and an extract of neat kerosene spiked onto the
burned carpet (─ ─) in the A) total ion chromatogram, B) mean-centered data, and C)
loadings on the second principle component, which results in the standard being positioned
more negatively in the scores plot than the extract.
101

xi

Figure 3.16: A) Enlarged view of the scores plot (Figure 3.11), focused on the replicate
extractions of burned carpet spiked with 10% evaporated kerosene. Each standard is denoted
as follows: Neat kerosene ( ) and 10% evaporated kerosene ( ). Extracts from the spiked
burned carpet are indicated by half fill. Total ion chromatograms showing the two peaks that
contribute most to the observed spread: B) the internal standard peak, showing one extract
with a lower abundance for all replicates, and C) the styrene peak from the burned carpet
matrix, showing that only four of the five extracts contain this peak.
103
Figure 3.17: Scores plot of first principle component (PC1) versus the second principle
component (PC2) based on the total ion chromatograms for the six ignitable liquid standards
and thermal degradation projections. Each standard is denoted as follows: Neat gasoline ( ),
10% evaporated gasoline ( ), 90% evaporated gasoline ( ), neat kerosene ( ), 10%
evaporated kerosene ( ), and 70% evaporated kerosene ( ), burned carpet with no ignitable
liquid added ( ). Extracts from the thermal degradation samples are indicated by half fill. 109
Figure 3.18: Total ion chromatograms of anomalous extracts (labeled 4 in Figure 3.17) of
carpet spiked with A) 90% evaporated gasoline, B) 70% evaporated kerosene, and C) no
ignitable liquid.

114

Figure 3.19: Enlarged view of the scores plot (Figure 3.17), highlighting two clusters of
extracts containing 90% evaporated gasoline (1 and 2). Each standard is denoted as follows:
Neat gasoline ( ), 10% evaporated gasoline ( ), 90% evaporated gasoline ( ), neat kerosene
( ), 10% evaporated kerosene ( ), and 70% evaporated kerosene ( ). Extracts from the
thermal degradation samples are indicated by half fill.
118
Figure 4.1: Total ion chromatograms of the C12 matrix interferences in five extracts of
burned high density polyethylene A) after alignment to the consensus target using the COW
algorithm and B) without alignment.
129
Figure 4.2: Scores plot of the first principle component (PC1) versus the second principle
component (PC2) based on the total ion chromatograms for the two ignitable liquids at three
levels of evaporation, denoted as follows: Neat gasoline ( ), 10% evaporated gasoline ( ),
90% evaporated gasoline ( ), neat kerosene ( ), 10% evaporated kerosene ( ), and 70%
evaporated kerosene ( ).
131
Figure 4.3: Loadings plots of A) the first principle component (PC1) and B) the second
principle component (PC2) based on the total ion chromatograms of the six ignitable liquid
standards. Major compounds are labeled: a) ethylbenzene, b) p-xylene, c) o-xylene, d)
propylbenzene, e) m-ethyltoluene, f) p-ethyltoluene, g) o-ethyltoluene, and h) 1,3,5trimethylbenzene. IS indicates the internal standard, nitrobenzene.

132

Figure 4.4: Representative total ion chromatogram of unburned high density polyethylene.

135

Figure 4.5: Total ion chromatograms of high density polyethylene (HDPE) spiked with A)
neat gasoline and B) neat kerosene showing the addition of C10-, C14-, and C16 alkenes and
xii

the C12 and C14 alkanes from the unburned HDPE matrix. Major compounds are labeled: a)
ethylbenzene, b) p-xylene, c) o-xylene, d) propylbenzene, e) m-ethyltoluene, f) pethyltoluene, g) o-ethyltoluene, and h) 1,3,5-trimethylbenzene. IS indicates the internal
standard, nitrobenzene. Cx denotes the alkane of the indicated carbon chain length (e.g., C12
is dodecane).
137
Figure 4.6: Scores plot of the first principle component (PC1) versus the second principle
component (PC2) based on the total ion chromatograms for the six ignitable liquid standards
and projected scores for ignitable liquids spiked onto unburned high density polyethylene
(HDPE). Each standard is denoted as follows: Neat gasoline ( ), 10% evaporated gasoline
( ), 90% evaporated gasoline ( ), neat kerosene ( ), 10% evaporated kerosene ( ), and 70%
evaporated kerosene ( ). Extracts from the unburned HDPE are indicated by half fill.
139
Figure 4.7: The C14 normal alkane peak in the total ion chromatogram (TIC) of the 70%
evaporated kerosene standard (──) and an extract of unburned high density polyethylene
(HDPE) spiked with 70% evaporated kerosene (─ ─) demonstrating the slight change in peak
width which results in a lower correlation between the two chromatograms. Solid vertical
lines indicate where the C14 normal alkane peak begins and ends in the TIC of the 70%
evaporated kerosene standard, while the dotted lines indicate the beginning and end of the
same alkane peak in the TIC of unburned HDPE spiked with 70% evaporated kerosene.
147
Figure 4.8: Representative total ion chromatograms of extracts of high density polyethylene
burned for A) 0 seconds, B) 10 seconds, C) 20 seconds, D) 30 seconds, E) 60 seconds, and F)
120 seconds. Major interference compounds are identified by brackets according to carbon
chain length. Triplet peaks are indicated as follows: a) the alkadiene, b) the alkene, and c)
the alkane.
150
Figure 4.9: Representative total ion chromatogram of burned high density polyethylene
(HDPE) spiked with A) neat gasoline and B) neat kerosene, showing the contribution of the
matrix interferences from burned HDPE. Major compounds are labeled: a) ethylbenzene, b)
p-xylene, c) o-xylene, d) propylbenzene, e) m-ethyltoluene, f) p-ethyltoluene, g) oethyltoluene, and h) 1,3,5-trimethylbenzene. IS indicates the internal standard, nitrobenzene.
Cx denotes the triplet peaks of the indicated carbon chain length (e.g., C12 bracket indicates
dodecadiene, dodecene, and dodecane).
152
Figure 4.10: Scores plot of the first principle component (PC1) versus the second principle
component (PC2) based on the total ion chromatograms for the six ignitable liquid standards
and projected scores for samples of burned HDPE spiked with each ignitable liquid. Each
standard is denoted as follows: Neat gasoline ( ), 10% evaporated gasoline ( ), 90%
evaporated gasoline ( ), neat kerosene ( ), 10% evaporated kerosene ( ), and 70%
evaporated kerosene ( ). Extracts from the burned HDPE are indicated by half fill.
154
Figure 4.11: Scores plot of the first principle component (PC1) versus the second principle
component (PC2) based on the total ion chromatograms for the six ignitable liquid standards
and projected scores of the thermal degradation samples. Each standard is denoted as
xiii

follows: Neat gasoline ( ), 10% evaporated gasoline ( ), 90% evaporated gasoline ( ), neat
kerosene ( ), 10% evaporated kerosene ( ), and 70% evaporated kerosene ( ), burned
HDPE with no ILR ( ). Extracts from the thermal degradation samples are indicated by half
fill.
166
Figure 4.12: Total ion chromatograms of 90% evaporated gasoline standard (──) and extract
from a 90% evaporated gasoline thermal degradation sample (─ ─) demonstrating the shift in
retention times for the C3-alkylbenzenes. Compounds are labeled as follows: a)
propylbenzene, b) m-ethyltoluene, c) p-ethyltoluene, d) o-ethyltoluene, and e) 1,3,5trimethylbenzene.
169

xiv

Chapter 1 – Introduction
To identify a fire as intentional, rather than accidental, fire debris analysts rely on the
identification of an ignitable liquid at the scene. Ignitable liquids are often used as accelerants to
increase the speed and spread of the fire. The term “ignitable liquids” encompasses a wide range
of products, including gasoline, kerosene, torch fuel, and adhesive removers, all of which are
widely available at a low cost. In cases of suspected arson, correct identification of an ignitable
liquid residue (ILR) in the resulting debris is the cornerstone of the case. Therefore, the
development of an analytical method to successfully identify the ILR is essential.

1.1 Classification of Ignitable Liquids

The American Society for Testing and Materials (ASTM) categorizes ignitable liquids
according to their chemical composition into one of eight classes: gasoline, petroleum distillates,
isoparaffinic products, aromatic products, naphthenic paraffinic products, normal alkane
1

products, oxygenated solvents, and others-miscellaneous . With the exception of gasoline,
classes are further divided into three subclasses based on carbon chain length: light (C4-C9),
medium (C8-C13), and heavy (C8-C20+). Classification using these subclasses is flexible, so
products can be characterized as “light to medium” or “medium to heavy” as necessary to better
describe the chemical compounds present. For example, an ignitable liquid containing C6-C10
hydrocarbons would be classified as light to medium. The eight classes are described further in
Tables 1.1 with several examples of everyday products that would be found in each class given
in Table 1.2.
1
 

1.2 Extraction of Ignitable Liquids from Fire Debris

ASTM outlines several procedures for the extraction of ignitable liquid residues (ILRs)
from fire debris. The extraction procedure most commonly used in forensic laboratories is a
2

passive headspace extraction using an activated carbon strip . For this method, the fire debris
collected at the scene is placed in a sealed, unlined, unused paint can or nylon bag. An activated
carbon strip is suspended in the container, which is subsequently heated in an oven at 50-80°C
for two to 24 hours. The volatile compounds from the fire debris saturate the headspace in the
container as well as the carbon strip. After heating, the carbon strip is removed, and the
2

compounds are eluted from the strip with 50-1000 µL of solvent, such as dichloromethane .
2

This extract is then typically analyzed by gas chromatography-mass spectrometry (GC-MS) .
Although widely used, this extraction method inherently discriminates according to the
volatility of the compounds present in a given sample. Compounds with boiling points higher
than the extraction temperature, such as the long-chain hydrocarbons found in heavy petroleum
distillates, do not volatilize or adsorb onto the carbon strip effectively. Therefore, the
chromatogram of the extract may not be representative of the ignitable liquid in the debris,
leading to possible misidentification of the ILR.

2
 

Table 1.1: ASTM classes for ignitable liquids and chemical composition of each class.
Class

Chemical Composition

Gasoline-all brands,
including gasohol

C3- and C4-alkylbenzene and various aliphatic
compounds
Homologous series of n-alkanes; less significant
isoparaffinic, cycloparaffinic, and aromatic components

Petroleum Distillates
Isoparaffinic Products
Aromatic Products
Naphthenic Paraffinic
Products
n-Alkanes Products

Branched chain (isoparaffinic); cyclic (naphthalenic)
alkanes and n-alkanes insignificant or absent
Aromatic compounds; aliphatic compounds absent or
insignificant
Branched chain (isoparaffinic) and cyclic
(naphthalenic) alkanes; n-alkanes insignificant or
absent
Only n-alkanes, typically containing five or less
components

Oxygenated Solvents

Oxygenated products including alcohols, esters,
ketones; major components include toluene or xylene

Others-Miscellaneous

Liquids that cannot otherwise be classified

3
 

Table 1.2: Examples of ignitable liquids in each of the ASTM classes.
Class
Gasoline-all
brands,
including
gasohol

Light (C4-C9)

Medium (C8-C13)

Heavy (C8-C20+)

Fresh gasoline is typically in the range of C4-C12

Petroleum
Distillates

Petroleum ether, some
cigarette lighter fluids,
some camping fluids

Some charcoal starters,
some paint thinners, some
dry cleaning solvents

Kerosene, diesel fuel,
some jet fuels, some
charcoal starters

Isoparaffinic
Products

Aviation gas, specialty
solvents

Some charcoal starters,
some paint thinners, some
copier toners

Some commercial
specialty solvents

Aromatic
Products

Some paint and varnish
removers, some
automotive parts
cleaners, xylenes,
toluene-based products

Some automotive parts
cleaners, specialty
cleaning solvents, some
insecticide vehicles, fuel
additives

Some insecticide
vehicles, industrial
cleaning solvents

Naphthenic
Paraffinic
Products

Cyclohexane-based
solvents/products

Some charcoal starters,
some insecticide vehicles,
lamp oils

Some insecticide
vehicles, lamp oils,
industrial cleaning
solvents

n-Alkanes
Products

Solvents, pentane,
hexane, heptane

Some candle oils, copier
toners

Oxygenated
Solvents

Alcohol, ketones, some
lacquer thinners, fuel
additives, surface
preparation solvents

Some lacquer thinners,
some industrial solvents,
metal cleaners/gloss
removers

Single component
Othersproducts, some blended
Miscellaneous
products, some enamel
reducers

Turpentine products,
some blended products,
various specialty products

4
 

Some candle oils,
carbonless forms,
copier toners

Some blended
products, various
specialty products

1.3 Analysis of Ignitable Liquid Residues

ILRs extracted from fire debris are routinely analyzed using GC-MS. Both total ion
chromatograms (TICs) and extracted ion profiles (EIPs) of characteristic compound classes can
be used to identify the ILR based on pattern recognition, while MS data can be used to identify
specific compounds in the chromatograms and, more generally, the classes of compounds in an
ILR. TICs contain all of the compounds extracted from the fire debris. As a result, these
chromatograms may be rather complex and difficult to analyze, though more representative of
the compounds in the fire debris. On the other hand, EIPs are used to extract the ions of
compounds found in ignitable liquids, resulting in the simplification of chromatograms (i.e.,
reduction of the number of peaks observed) and the possible removal of interference compounds.
Commonly utilized EIPs include profiles for the alkane, aromatic, indane, olefin/cycloparaffin,
and polynuclear aromatic classes. A list of the mass-to-charge (m/z) ratios for each compound
class is given in Table 1.3. The EIP chosen for analysis depends on the compounds present in
the TIC as determined by the MS data.
Fire debris analysts attempt to visually match the TICs or EIPs from the sample against a
reference collection to identify the class of ignitable liquid present. The reference collection may
be maintained in-house, or the analysts may refer to a national database, such as the National
3

Center for Forensic Science’s Ignitable Liquids Reference Collection . The visual comparison
of the sample extract to the reference collection is based on pattern recognition. That is, the
analyst does not rely on the identification of a single compound in the ignitable liquid to identify
the ILR, but rather compares the sample and reference chromatograms based on all compounds
1

present and the relative abundance ratios of those compounds . For example, the identification
5
 

Table 1.3: Common extracted ion profiles used to identify ignitable liquids.
Characteristic Class
Alkane
Aromatic
Indane
Olefin/Cycloparaffin
Polynuclear Aromatic

Mass-to-Charge (m/z) Ratios
57 + 71 + 85 + 99
91 +105 + 119 + 133
117 + 131 + 145 + 159
55 + 69 + 83 + 97
128 + 142 + 156

6
 

of toluene in an ILR alone does not classify that ILR as a gasoline. Many aromatic products are
toluene-based, such as lacquer thinner, which is comprised almost solely of toluene. On the
other hand, gasoline is a mixture of toluene, alkylbenzenes, and naphthalenes (Figure 1.1). The
analyst looks at the abundance ratios of these compounds to identify a residue as a gasoline. For
instance, the three C2-alkylbenzenes (Figure 1.1) are found in a 1:3:1 ratio in a neat gasoline.
Therefore, the identification of the ILR as belonging to either the aromatic or gasoline class
would depend on the identification of these additional compounds in the ILR and the relative
amounts of these compounds.

1.4 Problems in Identifying Ignitable Liquid Residues in Fire Debris
Because fire debris analysis is currently based solely on a visual comparison between
chromatograms of the ILR extract and the reference collection, the analysis is oftentimes difficult
and subjective, dependent greatly on the experience of the analyst. Three of the biggest
problems encountered in fire debris analysis are the evaporation of the ignitable liquid,
interferences from the matrix, and thermal degradation of both the liquid and the matrix during
burning. Evaporation of the liquid will reduce the number of characteristic compounds from the
ignitable liquid observed, while the burned matrix will add interference compounds to the
chromatogram. Burning of the debris both adds compounds to the chromatogram, due to the
thermal degradation of the matrix, and reduces the compounds present in the TIC from the
ignitable liquid, due to loss of the more volatile compounds of the liquid.
The selection of an accelerant by an arsonist is often a matter of convenience. Therefore,
the ignitable liquid chosen may not be fresh. For instance, a can of gasoline may sit in a garage
for several months before it is used as an accelerant. On the other hand, sometimes an
7
 

1.0E6

Toluene

C2-alkylbenzenes
C2-alkylbenzenes

b

Abundance

CC-alkylbenzenes
3 3-alkylbenzenes
IS

h
e
c
a
f
d g

0
00

16
16

Retention Time (min)

Figure 1.1: Representative chromatogram of neat gasoline. Major compounds are labeled: a)
ethylbenzene, b) p-xylene, c) o-xylene, d) propylbenzene, e) m-ethyltoluene, f) p-ethyltoluene, g)
o-ethyltoluene, and h) 1,3,5-trimethylbenzene.

8
 

ignitable liquid is spread at a scene long before it is ignited, to help create an alibi for the
arsonist. In either case, because the ignitable liquid has been allowed to sit for a period of time,
the volatile compounds in the liquid are likely to have evaporated. Thus, the presence of these
compounds in the chromatogram is reduced or eliminated altogether. After the fire is ignited, the
volatile compounds are also lost during the burning process while compounds from the burned
matrix are added to the extract chromatogram. Thus, identification of the ILR by visual
comparison to the reference collection is more difficult due to differences in chemical
composition between the neat and evaporated liquids.
Fire debris matrices from which ILRs are extracted, such as carpet, wood, and plastic,
4

often contain compounds similar to those found in ignitable liquids . Hydrocarbons inherent in
these materials, as well as compounds generated during thermal degradation of the matrix, are
also extracted from the debris and observed in the ILR chromatogram. These compounds may
mask the presence of compounds from the ignitable liquid or may give the appearance of the
presence of an ignitable liquid when none is present. Thus, these interferences could change the
characteristic ratios that fire debris analysts use to identify ILRs. For example, interferences
from burned plastic change the abundance ratios of the C2-alkylbenzenes in a gasoline ILR
chromatogram (Figure 1.2). Ethylbenzene, p-xylene, and o-xylene are present in a 1:3:1 ratio in
neat gasoline. However, in this example, these compounds appear in a ratio of 1:3:2 in an ILR of
gasoline extracted from a burned plastic sample. As such, it is important for fire debris analysts
to analyze not only the chromatogram of the ILR, but also chromatograms of the matrix with no
ILR present to serve as a control. Examining chromatograms of both the burned and unburned
matrix with no ILR present gives the analyst an idea of the compounds that come from the

9
 

4.0E5

Abundance

p-xylene

o-xylene
Ethylbenzene

0
4.8
.8

Retention Time (min)

5.8
5

Retention Time (min)

Figure 1.2: Representative chromatograms of neat gasoline (──) and neat gasoline ILR
extracted from burned plastic (‐ ‐ ‐) demonstrating the change in abundance ratios of the C2alkylbenzenes due to the addition of matrix interferences from the plastic.

10
 

matrix and the retention times of these compounds, thereby minimizing false positive
identifications of these compounds as having come from an accelerant.
The loss of characteristic compounds due to evaporation and the addition of compounds
from matrices can make the visual comparison of the debris extract to an ignitable liquid in a
reference collection even more subjective. The National Academy of Sciences (NAS) report
suggests moving toward more objective methods of analysis that provide a way of statistically
5

evaluating forensic evidence

.

An objective method for the identification of an ILR could

minimize the possibility of false positive or negative identifications despite the effects of
evaporation and matrix interferences. In addition, the ability to statistically evaluate fire debris
evidence would help forensic scientists present the evidence to a jury in a more convincing
manner.

1.5 Literature Review

Many aspects of fire debris analysis have been addressed in the literature. Studies have
investigated the potential interferences from unburned and burned matrices as well as the effect
of evaporation of the ignitable liquid on the identification of ILRs in fire debris. A large amount
of research has been focused on addressing the subjective nature of fire debris analysis and the
development of a more objective method for identifying ILRs. To that end, several statistical
and multivariate statistical procedures have been investigated. Different methods of data
pretreatment have also been investigated as part of these procedures.

11
 

1.5.1 Matrix Interferences
Several authors have demonstrated that common household matrices could give rise to
interferences in fire debris, both inherent in the matrix itself and those that result from the
burning process

4-12

. These interferences could prevent the identification of an ILR by changing

the compounds observed in a chromatogram and the abundance ratio of those compounds.
Lentini et al. identified a wide range of common unburned matrices, including shoe and
4

furniture polishes, insecticides, clothing, and magazines, that contain petroleum solvents . Most
of these solvents are used in the manufacturing process, and trace amounts of the solvents can be
found in the matrix up to several years later. For instance, toluene and C2-alkylbenzenes, very
similar to those found in evaporated gasoline, were identified in a sample of 19-year-old
structural adhesive. Because of the prevalence of household matrices that give chromatograms
with patterns of compounds similar to petroleum-based accelerants, the authors caution strongly
against using single compounds to identify fire debris as containing an ILR. Instead, analysts
should be sure to analyze comparison samples of the unburned and burned matrices to determine
which interferences are inherent in the matrix to prevent false positive identifications of ILRs.
In addition to interferences found inherently in household matrices, Almirall and Furton
6

examined interferences generated during the burning process . A total of 21 different types of
matrices, including several varieties of carpet and flooring, as well as paper products and
clothing, were analyzed both unburned and burned. Matrices were extracted using a passive
headspace extraction with an activated carbon strip followed by GC-MS. Some matrices, such as
wood, showed little difference in compounds extracted before and after burning. However,
others, such as synthetic flooring, gave drastically different compounds in the chromatograms
12
 

after burning, though the interferences were neither shown nor identified by the authors.
Additionally, the authors identified several matrices that contained compounds that are
commonly used to identify ILRs, including both normal and branched alkanes and aromatic
compounds such as methylnaphthalenes, though these compounds were usually in different
abundance ratios than those typically observed for ignitable liquids.
The findings of the studies by Lentini et al. and Almirall and Furton are further
7

confirmed by Fernandes et al. . This study also investigated the effect of storage time on the
compounds extracted from the matrices. Two samples of each of 15 matrices were obtained.
One sample of each matrix was burned, extracted, and analyzed immediately, while the other
sample was stored in open air for one month before burning and analysis. For all samples, a
passive headspace extraction was used, followed by analysis using automated thermal desorption
GC-MS. Several of the analyzed matrices showed similar compounds to those found in ignitable
liquids. For example, the contact adhesive contained toluene, m-xylene, and ethylbenzene, three
compounds found in gasoline. However, smaller abundances of these compounds were observed
in the older samples and some were not observed at all. For instance, toluene and ethylbenzene
were not observed in the chromatogram of the month-old contact adhesive. Thus, the authors
suggest that fire debris analysts should attempt to analyze the pieces of evidence from the fire
scene that appear to be the oldest to minimize the chance of matrix interferences leading to a
false positive identification of an ILR.
Carpet is one of the most frequently studied matrices as it is often found at residential
arson scenes. Carpet is collected and submitted for analysis because it has the ability to absorb
8

and preserve residual amounts of ignitable liquids . Putorti et al., while investigating burn
patterns left by liquid fuel fires, discovered that as carpet burns it can melt over some of the
13
 

9

ignitable liquid . The ignitable liquid is then trapped and can be detected during analysis. A
study by Ma et al. demonstrated that the strands of carpet simultaneously act as wicks to increase
the surface area and, consequently, the rate of evaporation of the ignitable liquid as well as block
10

the heat transfer from the fire to the absorbed pool of ignitable liquid . Therefore, although the
carpet accelerates the evaporation of the liquid, if the fire is extinguished quickly, enough of the
ignitable liquid will remain in the carpet for possible identification.
Recognizing the prevalence of other matrices commonly found at arson scenes, Stauffer
investigated the mechanisms by which matrix interferences are generated during the burning of
11

polymers, such as polyethylene and polypropylene . Mechanisms of degradation were
presented for each of the polymers tested. For example, polyethylene was found to undergo
random scission during burning. This conclusion was based on the observation of series of
“triplets” of peaks in the chromatograms of the burned polyethylene corresponding to the
alkadiene, alkene, and alkane for each hydrocarbon in the plastic. Stauffer suggests that
knowledge of the materials in the matrices found at arson scenes and the various mechanisms of
degradation of those matrices would be helpful in interpreting chromatograms of fire debris and
comparing those chromatograms to reference standards. However, identification of the type of
plastic in the fire debris may be difficult due to melting and scorching of the matrix during the
burning process.
In each of the aforementioned studies, no ignitable liquids were spiked onto the matrices
investigated by the authors. Thus, the effect of the matrix interferences on the ability to identify
an ILR has not been demonstrated.

14
 

Borusiewicz et al. investigated several factors, including the type of accelerant, matrix,
and burning time, to determine which had the greatest effect on the ability of an analyst to
12

identify an ILR . Carpet, deciduous wood, and chipboard were each spiked with five different
accelerants, including gasoline and kerosene. The matrices were then burned until each selfextinguished, and then placed in separate glass jars and extracted using a passive headspace
procedure. Chromatograms were obtained for each of the samples for visual comparison to
chromatograms of the liquid standards using GC-MS. Of all the factors investigated, the matrix
was determined to be the most influential on identification of an ILR. Because carpet tended to
self-extinguish before it was completely incinerated, the ignitable liquids were often able to be
extracted in sufficiently high abundance for easy identification. However, chipboard and wood
burned until no ILR could be detected. In this study, the specific interferences generated by each
matrix were not discussed and the volume of ignitable liquid poured onto the matrix was not
optimized so that the resulting chromatograms contained compounds from both the liquids and
the matrices.

1.5.2 Evaporation of Ignitable Liquids
Of the few studies that have addressed the effect of evaporation of the ignitable liquid on
the identification of ILRs in fire debris, most have focused on gasoline because it is a commonly
8

used accelerant . For example, Mach investigated the effects of evaporation on the visual
13

analysis of gasoline chromatograms obtained from GC-MS . When gasoline is evaporated to
approximately 99% by volume, the chromatograms of the evaporated gasoline appear drastically
different than the neat gasoline, having lost the more volatile compounds such as the

15
 

alkylbenzenes. The major compounds of the evaporated samples were the methylnaphthalenes
and other polyaromatic hydrocarbons (PAHs), including anthracene and pyrene, due to the
concentration of these less volatile compounds during evaporation.
To test if these less volatile compounds could be used to identify gasoline residues from
fire scenes, Mach burned a sample of gasoline to simulate a gasoline residue that might be found
13

at the scene of a fire . The sample was burned in a glass beaker until an oily residue remained.
The residue was mixed with soot from the sides of the beaker for analysis. Acknowledging that
compounds from burned matrices, such as wood and plastic, could complicate the visual analysis
of chromatograms, Mach used limited mass scans of particular ions to simplify the
chromatographic data and focus on the PAHs for identification of the residue. The PAHs were
detected in the simulated residue, indicating that the compounds could be used for the
identification of gasoline ILRs. However, the author warns that the presence of these
compounds in a residue from an actual arson scene must be shown not to come from a matrix for
true positive identification of the residue as gasoline.
Barnes et al. investigated using target ions and abundance ratios to associate evaporated
gasoline to its neat counterpart as well as to distinguish between samples taken from different
14

service station locations . The authors identified several groups of hydrocarbons in gasoline
that are present in constant abundance ratios to each other despite evaporation. Four ratios were
calculated for gasoline at the 75% evaporation level and used to compare the evaporated gasoline
to the neat gasolines. All ten of the 75% evaporated gasoline samples were correctly associated
to the corresponding neat gasoline, to the exclusion of all other neat gasolines. In addition, six
different abundance ratios were calculated for gasoline that was evaporated to 50% by volume.
The gasoline was then spiked onto charred pine wood before GC-MS analysis. Using these six
16
 

ratios, all sixteen of the 50% evaporated gasoline samples were able to be associated to the
corresponding neat gasoline despite interferences from the charred wood, though the
interferences were neither identified nor addressed in detail by the authors.
There were some additional limitations to this study, including the small sample number.
The authors only analyzed sixteen gasoline samples and caution that, with a larger sample
number, gasolines from several locations may be more similar than the gasolines considered in
14

this study . Therefore, while it could be possible to eliminate locations as the source of a
particular sample of gasoline, it may not be possible to determine the definite origin of a sample
of gasoline using this method of analysis.
Bertsch conducted a study in which several samples of previously burned carpet were
8

spiked with evaporated gasoline . Carpet is usually made of synthetic fiber bound to a polymer
backing. During the burning process, these polymers yield alkylbenzenes and other aromatic
hydrocarbons. While these interferences are not generally confused with the normal alkanes
15

found in petroleum-based accelerants, their presence has been misidentified as gasoline . In
Bertsch’s study, the burned carpet did produce some compounds similar to those in gasoline, but
the extracted ion chromatograms for the alkylbenzenes and methylnaphthalenes could be used to
8

identify which samples had been spiked with gasoline . Bertsch concluded that while some
compounds found in ignitable liquids are produced during the burning of carpet (e.g., styrenes
and alkylbenzenes), the profiles of these compounds from the burned matrix are different enough
from gasoline to prevent false positive identification of the accelerant. However, Bertsch spiked
gasoline onto previously burned carpet. This method does not account for thermal degradation
of the ignitable liquid during the burning process.
17
 

1.5.3 Multivariate Statistical Analysis of Ignitable Liquids
Multivariate statistical procedures have been investigated by several researchers for the
16-23

identification of ILRs

. Doble and coworkers identified 44 target compounds in the

chromatograms of 88 gasoline samples and used principal components analysis (PCA) and linear
16

discriminant analysis (LDA) to differentiate between regular and premium grade gasoline .
However, further distinction of the samples as either summer or winter gasolines was not
possible using this procedure. Artificial neural networks (ANN) were then used to classify the
gasolines according to grade and then further subdivide the samples by season. Using half of the
samples as a training set, ANN were able to correctly classify 100% of the samples according to
grade. ANN were able to classify 96.6% of the samples correctly according to both grade and to
season. Two of the regular unleaded gasolines (one from each season) were not classified and
one premium summer gasoline was misclassified, though the authors do not clarify the reasoning
for the misclassification. In addition, this study analyzed gasoline samples only and did not use
the full chromatogram.
Sandercock and Du Pasquier collected 35 unleaded gasoline samples from different
17

service stations and evaporated the samples to four levels ( 25, 50, 75, and 90% by weight) .
Using PCA on the C0- to C2-naphthalene profile obtained from extracted ion chromatograms, the
evaporated samples were associated to their neat liquids. The 35 gasoline samples at five levels
of evaporation formed 18 unique groups on the PCA scores plot, eleven of which contained only
one type of gasoline sample. Thus, the authors were able to identify the service station the

18
 

gasoline came from despite evaporation. However, this study analyzed the ignitable liquids only,
so the effect of matrix interferences was not explored.
Sandercock and Du Pasquier also used selected ion monitoring (SIM) of the C0- to C2naphthalenes and PCA to distinguish three different types of gasoline (regular unleaded,
18

premium unleaded, and lead replacement) , as well as samples of similar types of gasoline
19

taken from different locations . Using the C0- to C2-naphthalenes as a chemical fingerprint, the
35 different gasoline samples were separated into 32 distinct groups using PCA. Of these
groups, 30 had only one member, one group had two, and one group had three gasoline samples.
Thus, the C0- to C2-naphthalenes could be used to distinguish between gasoline samples from
17,18

different sources. However, the neither of the Sandercock and Du Pasquier studies

nor the

16

study by Doble et al . investigated gasoline extracted from matrices.
Bodle and Hardy used PCA, soft independent modeling classification analogy (SIMCA),
and hierarchical cluster analysis (HCA) to classify a total of 130 ignitable liquids using a
previous five-category ASTM classification system: light petroleum distillates, gasoline,
20

medium petroleum distillates, kerosene, and heavy petroleum distillates . The abundances of
compounds in the chromatograms of the ignitable liquids were summed in thirty-second intervals
to reduce the size of the data set to 114 variables per chromatogram. In addition, variables that
were determined to have no significant effect were removed. For example, no response was
observed for any of the samples in the first or last three minutes of the chromatogram.
Therefore, the corresponding 12 variables were removed from the data set.

19
 

Using PCA, samples were clustered according to the ASTM class. However, there was
significant overlap between the light petroleum distillates and gasoline, as well as between
kerosene and the heavy petroleum distillates. The authors attribute this to similarity of
compounds in these classes. For instance, light petroleum distillates and gasoline overlap due to
20

similarities in the C8-C9 region of the chromatogram . However, these similarities could be
due to the reduction of variables in the data set caused by summing regions of the chromatogram
rather than actual similarity of compounds in the ignitable liquids. The representative
chromatograms provided by the authors for the light petroleum distillate and gasoline classes
show that each chromatogram has a large peak in the C8-C9 region. However, the two peaks
clearly have different retention times and shapes. Thus, the two samples might be distinguished
using the entire chromatogram instead of the reduced variable data set.
The chemical similarity of the samples was also observed in the HCA results. The HCA
dendrogram showed significant overlap of samples from the kerosene and heavy petroleum
20

distillates classes . However, the authors note that the new ASTM classification system (Table
1.1) has grouped these classes together, so the overlap in the dendrogram is not unexpected.
This overlap was again observed when SIMCA was performed on the PCA data. Despite the
overlap, the authors claim that SIMCA was able to correctly classify 98.5% of the samples into
the correct ASTM class. The only misclassified sample showed no correlation to any of the
classes and was considered an outlier. Thus, in this study, SIMCA was shown to be able to
predict the ASTM class of an ignitable liquid.
A study by Hupp et al. demonstrated that 25 diesel samples could be discriminated into
four groups based on aliphatic and aromatic content when PCA was applied to the complete

20
 

21

chromatograms of the samples . The authors also used the PCA loadings plots to identify the
chemical compounds contributing most to the variance among samples. When PCA was applied
to EICs of each sample, a larger number of groups was observed in the scores plot. The normal
alkane profile (m/z 57) yielded eight groups, and the aromatic (m/z 91) profile yielded nine.
This result indicated that the normal alkane and aromatic components provided the greatest
discrimination among samples in the data set.
In addition to PCA, Hupp et al. used Pearson product moment correlation (PPMC)
coefficients to place a statistical value on the similarity between pair-wise combinations of diesel
21

samples . PPMC coefficients showed strong correlation between samples of the same brand,
consistent with the close positioning of these samples observed in the PCA scores plot. Stronger
correlations between all samples were observed using the TICs, while a larger range of PPMC
coefficients were observed for the EICs. The larger number of groups observed using EICs as
well as the larger range of PPMC coefficients indicated an increased amount of discrimination
among samples. Analyzing both TICs and EICs, Hupp et al. were not only able to demonstrate a
visual assessment of the discrimination of the diesel samples with PCA, but also to place a
statistical measure on that discrimination using PPMC coefficients.
It should be noted that these studies, with the exception of Hupp et al., did not analyze
the full chromatograms, but rather focused on the examination of selected peaks or sections of
the chromatograms. Additionally, the aforementioned studies focused on the analysis of
ignitable liquids only, not ILRs extracted from fire debris. Thus, the effects of matrix
interferences were not considered.
Tan et al. used SIMCA and PCA to identify 51 ignitable liquids according to ASTM
22

class . In addition, a few select liquids were spiked onto a carpet matrix, which was then
21
 

burned to simulate fire debris. The extracts from the simulated debris were analyzed using a
solvent extraction procedure and GC-MS, from which extracted ion chromatograms were
generated using ions representative of the alkane, aromatic, and olefin compound classes. The
EICs were then divided into 19 equal time segments, and the signal in each part summed to give
a total of 19 variables per chromatogram. Using these variables, all ignitable liquids tested were
correctly identified using both SIMCA and PCA. While SIMCA was able to classify the
ignitable liquids correctly based on the 19 variables used rather than the whole chromatogram
(which may contain hundreds of variables or more), the author did not discuss either the identity
or extent of matrix interferences.
Baerncopf et al. used PCA and PPMC coefficients to evaluate the association of ILRs to
23

the neat reference standard despite the presence of matrix interferences . A total of six
ignitable liquids, one from each ASTM class, were spiked onto carpet, and then the carpet was
burned, extracted using a passive headspace extraction with a carbon strip, and analyzed using
GC-MS. The carpet was burned both lightly and heavily to investigate the effects of increased
matrix interferences. In both cases, ILRs were successfully associated with the corresponding
neat liquids using both PCA and PPMC coefficients. It should be noted, though, the
chromatograms of the ILRs were dominated by peaks from the ignitable liquids, with minimal
contributions from matrix interferences. Also, when additional ignitable liquid standards, again
one from each ASTM class, were added to the data set, ILRs containing liquids in the gasoline
and petroleum distillates classes could only be identified by ASTM class due to the similarity of
chemical composition of the liquid standards in these classes. For example, the two different
types of gasoline could not be distinguished from each other using the combination of PCA and
PPMC coefficients, but could be differentiated from the other ignitable liquids.
22
 

1.5.4 Data Pretreatment Procedures
Several data pretreatment procedures can be applied to chromatographic data to minimize
non-chemical sources of variance before statistical analysis. Because PCA is based on
identifying the greatest sources of variance between samples in a given data set, minimization of
non-chemical sources of variance is essential. Four types of pretreatment for chromatographic
data that have been reported in the literature are background subtraction, smoothing, retention
time alignment, and normalization.
Background subtraction is normally used to remove low-frequency fluctuations in the
24

baseline that could cause discrepancies when visually comparing chromatograms . The analyst
can manually subtract the baseline between points on a chromatogram, or computer software can
be used to subtract a background chromatogram (also known as a blank), a selected mass
spectrum, or other fitted model from the entire chromatogram at once. Background subtraction
is performed on each chromatogram separately and is generally performed before other data
pretreatment procedures.
Smoothing chromatograms helps to remove high-frequency fluctuations in signal
introduced during analysis. Smoothing serves to maximize the signal-to-noise ratio while
minimizing distortion in each peak

25

. Many different smoothing filters are available, including

26

the Savitzky-Golay algorithm . For this algorithm, the polynomial equation is fitted over a
section of the chromatogram, also called a window. The center data point in the window is then
24

replaced by the value predicted by the model . The window is then shifted and the process
repeated for each data point in the chromatogram. Both the window size and the order of the
polynomial are defined by the user. Smoothing must be applied with care, as too much
23
 

smoothing reduces the signal intensity and causes peak broadening and loss of chromatographic
26

features .
After chromatograms are smoothed, retention-time alignment is performed. Retention
time shifts are the result of several variations between chromatographic analyses, including oven
27

temperature gradient and mobile phase flow rate . Even small changes in the retention times of
the compounds in a sample chromatogram can make the visual comparison between that sample
chromatogram and a reference chromatogram more challenging. Retention-time alignment aims
to ensure that compounds in the chromatograms, which are assumed to be the same, have the
same retention time. Malmquist and Danielsson demonstrated the effects of retention-time
27

alignment using PCA on a data set of 54 chromatograms of equine cytochrome c digests . The
method was a combination of aligning the sample chromatograms to a target chromatogram and
then applying a corrected time scale to all of the chromatograms based on a few selected
chromatographic peaks. Analysis of the raw data identified misalignments between
chromatograms as the main source of variance in the data set. However, once retention time
alignment was performed, the chemical composition differences between the digests was the
greatest source of variance.
Tomasi et al. investigated the use of a correlation-optimized warping (COW) algorithm
28

for retention-time alignment of chromatograms of 84 coffee extracts . When PCA was
performed on unaligned data, the authors observed a “horseshoe” pattern in the data caused not
by chemical difference between the extracts but rather by retention time shifts of peaks in the
chromatograms. After the COW algorithm was applied, the horseshoe pattern was no longer
observed in the PCA data. In addition, a visual assessment of the chromatograms showed better
24
 

peak alignment after COW was applied. Taken together, these observations indicate that the
association and discrimination of the samples in the PCA scores plot were due to chemical
composition of the samples, not shifts in retention time after alignment procedures were used.
Normalization minimizes differences in abundance of peaks in chromatograms due to
24

variation in injection volume and instrument response . Generally, during normalization, the
chromatograms in a data set are scaled to a similar order of magnitude. Several types of
normalization can be used, including total area and internal standard normalizations. The type of
normalization chosen for analysis depends on the particular data set.
Reducing the variability in the abundance of peaks between chromatograms in a data set
ensures that random variance in abundance (i.e., variance due to differences in injection volume
or instrument response) does not outweigh variance in chemical composition during subsequent
26

analyses . However, normalization is most effective when the samples of a given data set are
similar in chemical composition. If a data set contains vastly different types of samples,
normalization may disproportionately alter one type of sample. Thus, the true source of variance
between the samples would be skewed toward differences in abundance rather than differences
29

in chemical composition .

1.6 Research Objectives and Goals

This research aimed to develop an objective method for associating an ILR to the
corresponding neat liquid using two multivariate statistical procedures: PPMC coefficients and
PCA. PPMC coefficients were used to place a numerical value on the similarity of two

25
 

chromatograms, while PCA was used to identify the greatest source of variance among samples,
allowing for both discrimination and association. The combination of statistical procedures
maximizes the potential of successful associations between ILRs and neat liquids, while enabling
a statistical measure of the associations.
The first goal of this research was to investigate the effect of matrix interferences on the
association of residue extracts to the corresponding neat liquids. Two ignitable liquids (gasoline
and kerosene) were evaporated to two different levels. The neat and evaporated liquids were
spiked onto separate unburned and burned subsamples of carpet and high-density polyethylene.
Spiking the liquids onto the unburned matrix was done to investigate the effect of interferences
found inherently in the matrix on the association of ILRs to ignitable liquid standards. The
burned matrix was spiked with each of the ignitable liquids to ensure that only the effects of
evaporation were being investigated on the same type of association, with no contribution from
thermal degradation due to burning. A combination of the multivariate statistical procedures
provided a measure of the association and discrimination of the extracted residues with matrix
interferences to the corresponding neat liquid.
The second goal of this research was to investigate the effects of thermal degradation on
the ability to associate an ILR to the corresponding neat liquid. Fire debris was simulated by
spiking separate samples of the matrices with each of the neat and evaporated ignitable liquids.
The matrices were then burned to generate significant matrix interferences. The simulated ILR
was extracted and analyzed. PCA and PPMC coefficients were used to demonstrate the
association of the ILRs to the corresponding ignitable liquid, despite thermal degradation and
evaporation of the liquid, as well as the presence of matrix interferences.

26
 

This research aimed to develop a more objective method for the analysis of fire debris, as
suggested by the recent NAS report. By making the analysis objective, the chance of a
misidentification of the presence or the absence of an ignitable liquid is minimized and no longer
dependent on the experience of the analyst. In addition, this research used the full
chromatograms, as opposed to other procedures reported in the literature that use only portions
of the chromatogram, to associate ILRs to neat liquids despite the combination of matrix
interferences, evaporation of the ignitable liquid, and thermal degradation effects. Thus, a more
complete analysis of the data was used to maximize the probability of a correct association
between and ILR and an ignitable liquid standard.

27
 

References

28
 

References

1. American Society for Testing and Materials, ASTM E 1618-06e1. Annual Book of ASTM
Standards 14.02.
2. American Society for Testing and Materials, ASTM E 1411-07. Annual Book of ASTM
Standards 14.02.
3. National Center for Forensic Science, Ignitable Liquids Reference Collection.
http://ilrc.ucf.edu/ (Accessed September 2010)
4. Lentini JJ, Dolan JA, Cherry C. The petroleum-laced background. J Forensic Sci 2000; 45:
968-989.
5. Committee on Identifying the Needs of the Forensic Sciences Community, National Research
Council. Strengthening Forensic Science in the United States: A Path Forward. Washington,
D.C.: National Academies Press, 2009.
6. Almirall JR, Furton KG. Characterization of background and pyrolysis products that may
interfere with forensic analysis of fire debris. J Anal Appl Pyrol 2004; 71: 51-67.
7. Fernandes, MS, Lau CM, Wong WC. The effect of volatile residues in burnt household items
on the detection of fire accelerants. Sci Justice 2002; 42: 7-15.
8. Bertsch W. Volatiles from carpet: a source of frequent misinterpretations in arson analysis. J
Chromatogr A 1994; 674: 329-333.
9. Putorti AD, McElroy JA, Madrzykowski D. Flammable and combustible liquid spill/burn
patterns. NIJ Report 604-00: 2001.
10. Ma T, Olenick SM, Klassen MS, Roby RJ, Torero JL. Burning rate of liquid fuel on carpet
(porous media). Fire Technol 2004; 40: 227-246.
11. Stauffer E. Concept of pyrolysis for fire debris analysts. Sci Justice 2003; 43: 29-40.
12. Borusiewicz R, Zieba-Palus J, Zadora G. The influence of the type of accelerant, type of
burned material, time of burning, and availability of air on the possibility of detection of
accelerant traces. Forensic Science International 2006; 160: 115-126.
13. Mach MH. Gas chromatography-mass spectrometry of simulated arson residue using
gasoline as an accelerant. J Forensic Sci 1977; 22: 348-357.
14. Barnes AT, Dolan JA, Kuk RJ, Seigel JA. Comparison of gasolines using gas
chromatography-mass spectrometry and target ion response. J Forensic Sci 2004; 49: 10181023.
15. Bertsch W, Sellers CS, Holzer G. Chemical Analysis for the Arson Investigator and
Attorney. Heidelberg, Germany, Hüthig Buch Verlag: 1993.
29
 

16. Doble P, Sandercock PML, Du Pasquier E, Petocz P, Roux C, Dawson M. Classification of
premium and regular gasoline by gas chromatography/mass spectrometry, principal
component analysis and artificial neural networks. Forensic Sci Int 2003; 132: 26-39.
17. Sandercock PML, Du Pasquier E. Chemical fingerprinting of gasoline 2: Comparison of
unevaporated and evaporated samples. Forensic Sci Int 2004; 140: 43-59.
18. Sandercock PML, Du Pasquier E. Chemical fingerprinting of unevaporated automotive
gasoline samples. Forensic Sci Int 2003; 134: 1-10.
19. Sandercock PML, Du Pasquier E. Chemical fingerprinting of gasoline Part 3: Comparison of
unevaporated automotive gasoline samples from Australia and New Zealand. Forensic Sci Int
2004; 140: 71-77.
20. Bodle ES, Hardy JK. Multivariate pattern recognition of petroleum-based accelerants by
solid-phase microextraction gas chromatography with flame ionization detection. Anal Chim
Acta 2007; 589: 247-254.
21. Hupp AM, Marshall LJ, Campbell DI, Waddell Smith R, McGuffin VL. Chemometric
analysis of diesel fuel of forensic and environmental applications. Anal Chim Acta 2008; 606:
159-171.
22. Tan B, Hardy JK, Snavely RE. Accelerant classification by gas chromatography/mass
spectrometry and multivariate pattern recognition. Anal Chim Acta 2000; 42: 37-46.
23. Baerncopf JM, McGuffin VL, Smith RW. Association of ignitable liquid residues to neat
ignitable liquids in the presence of matrix interferences using chemometric procedures. J
Forensic Sci 2011; 56: 70-81.
24. Morgan SL, Bartick EG. Discrimination of forensic analytical chemical data using
multivariate statistics. In: Blackledge, RD, editor. Forensic analysis on the cutting edge: new
methods for trace evidence analysis. Hoboken, NJ: John Wiley& Sons, Inc., 2007; 333-367.
25. Savitzky A, Golay MJE. Smoothing and differentiation of data by simplified least squares
procedures. Anal Chem 1964; 36: 1627-1639.
26. Brereton, RG. Applied chemometrics for scientists. Hoboken, NJ: John Wiley & Sons, Ltd,
2007.
27. Malmquist G, Danielsson R. Alignment of chromatographic profiles for principal component
analysis: a prerequisite for fingerprinting methods. J Chromatogr A 1994; 687: 71-88.
28. Tomasi G, van den Berg F, Andersson C. Correlation optimized warping and dynamic time
warping as preprocessing methods for chromatographic data. J Chemometr 2004; 18: 231241.
29. Rietjens M. Reduction of error propagation due to normalization: effect of error propagation
and closure on spurious correlations. Anal Chim Acta 1995; 316: 205-215.

30
 

Chapter 2 – Instrumental and Statistical Techniques

2.1 Gas Chromatography-Mass Spectrometry

Gas chromatography-mass spectrometry (GC-MS) is a technique very often used in
forensic laboratories for the analysis of a variety of types of evidence, from fire debris to
controlled substances. The nearly universal applicability of GC-MS analysis lies in the
combination of a separation technique, GC, with an identification technique, MS. Therefore,
GC-MS can be used both to separate and to identify compounds of a complex mixture.
The GC-MS system consists of an inlet where the sample is introduced to the system, the column
in which the compounds of a sample mixture are separated, and a detector, in this case a mass
spectrometer, used to detect each of the separated compounds (Figure 2.1).
In the broadest sense, gas chromatography separates the compounds, or analytes, of a
1

mixture based on interactions between the analytes and the mobile and stationary phases . For
gas chromatography, the mobile phase is an inert gas, such as helium or hydrogen, which is often
referred to as the carrier gas. The carrier gas carries the analytes through a capillary column.
Because the mobile phase is inert, separation of the compounds of a mixture is due to
interactions between the compounds and the stationary phase. In GC, the stationary phase is a
thin liquid film coated onto the inside of the column. As analytes are carried through the
column, they interact with the stationary phase to different extents, dependent on the chemical
composition of the coating. For instance, if the stationary phase is non-polar, the non-polar
compounds in the sample mixture will interact more strongly with the stationary phase than the
polar compounds in the mixture. Based on the strength of the affinity of the analytes for the
31
 

Regulator

Inlet
Detector

Column

Column Oven

Carrier Gas Cylinder
Figure 2.1: Schematic diagram of a gas chromatograph.

32
 

stationary phase, these interactions retard the progress of the analytes through the column. That
is, the stronger the attraction between the analyte and the stationary phase, the more the analyte
is retarded and the longer it remains in the column. Because of the differing degrees to which
the analytes will interact with the stationary phase, separation of the compounds in mixtures is
possible.
One way of introducing the sample to the GC-MS system is to dissolve the sample in a
solvent and then directly inject the solution into the heated inlet of the GC using a syringe
(Figure 2.2). The chosen solvent will vary according to the sample being analyzed, but generally
must not react with the compounds in the sample and must be volatilized upon injection. The
temperature of the inlet is chosen such that it is higher than the boiling point of the least volatile
1

compound in the sample and higher than the maximum temperature of the column oven.
Typically for capillary columns, one microliter of the sample solution is drawn into the syringe,
followed by one microliter of air. The air serves to prevent volatilization of the sample in the
needle of the syringe immediately before injection.
Once injected, the sample solution is volatilized immediately, and then carried onto the
column by the carrier gas. The amount of the volatilized sample that is introduced to the column
can be set by the analyst using a split valve. The analyst chooses a split ratio that determines
what fraction of the sample goes to the column and what fraction is purged to waste. For
example, if a 50:1 split is chosen, only one part of every fifty is introduced to the column. Split
injections are used to prevent excessive peak broadening and overloading of the column and
detector. On the other hand, for samples in which the analytes are present only in trace amounts,
splitless injections are used so the entire sample volume goes onto the column.

33
 

Syringe

Septum
Carrier Gas

Split Valve

Vaporization
Chamber

To Column
Figure 2.2: Schematic diagram of the inlet of a gas chromatograph.

34
 

In this research, a pulsed injection was used. With a pulsed injection, the pressure at the
column inlet is increased just before the sample is introduced, then reduced back to the normal
value after a specific time. The increased pressure ensures that the entire sample is introduced to
the column quickly and simultaneously, reducing band broadening and helping to prevent the
thermal degradation of the sample. Band broadening on the column results in increased
chromatographic peak width. Ideally, the sample should have the narrowest band possible while
on the column to produce a resolved, narrow peak in the chromatogram.
The two factors that most greatly affect band broadening in GC are molecular diffusion
and mass transfer. Molecular diffusion refers to the axial diffusion of the analyte molecules in
the mobile phase as the molecules pass through the column. The analyte molecules migrate from
the highly concentrated center of the band both forward and backward to regions of lower
concentration (Figure 2.3). Mass transfer describes the rate of transfer of analyte molecules
between the mobile and stationary phases. Ideally, the molecules would reach equilibrium
rapidly between the mobile and stationary phases as the analyte travels through the column.
However, this equilibrium is established so slowly that it is almost never reached. Because the
molecules in the stationary phase do not travel through the column as quickly as molecules in the
mobile phase, the band will broaden (Figure 2.4).
For GC, molecular diffusion can be reduced by increasing the flow rate of the carrier gas.
A faster flow rate reduces the amount of time it takes for the analyte to elute off of the column,
thereby reducing the amount of time during which diffusion can occur. On the other hand, the
effect of mass transfer is reduced with a slower flow rate and thinner stationary phase because
molecules are better able to reach equilibrium between the mobile and stationary phases.
Overall, a flow rate must be chosen that is fast enough to minimize molecular diffusion and slow

35
 

A

t1

t2

t3

Concentration

B
t1
t2
t3

Figure 2.3: Diagram illustrating A) band broadening due to molecular diffusion while the
analyte molecules travel through the column from t1 to t3 and B) the resultant broadening of the
chromatographic peak.

36
 

B

Total Concentration

Concentration of
Analyte in
Stationary
Mobile
Phase
Phase

A

Figure 2.4: Diagram showing mass transfer of the analyte molecules between the mobile and
stationary phases at A) equilibrium and B) when the flow rate of the carrier gas is too high for
equilibrium to be reached, resulting in band broadening, and the corresponding chromatographic
peaks.

37
 

enough to minimize mass transfer effects and yield well-resolved peaks.
The retention time of a compound is the amount of time it takes for that compound to
1

reach the detector after injection . Generally speaking, independent of the column stationary
phase, compounds in a sample mixture will elute from the column in order of increasing boiling
point. That is, compounds with lower boiling points will elute first and, therefore, have shorter
retention times than the compounds with higher boiling points that elute later. This separation is
possible because the column is housed in an oven, the temperature of which can be controlled by
the analyst using temperature programs.
1

Temperature programs can either be isothermal or ramped . For isothermal programs,
the temperature of the oven is held at a constant during the entire analysis. Like the inlet
temperature, the oven temperature must be kept above the boiling point of the least volatile
compound in the sample mixture. Isothermal temperature programs cannot be used with sample
mixtures whose compounds have a broad range of boiling points because the compounds will not
efficiently elute from the column, possibly resulting in contamination. For these mixtures, the
temperature can be ramped at different rates to increase the speed of analysis, while still allowing
for sufficient separation of the sample compounds. Typically, slower temperature ramps (e.g.,
5°C/min) give better resolution of sample peaks, especially if the compounds have similar
boiling points, but may cause band broadening. On the other hand, faster ramp rates (e.g.,
20°C/min) are used to reduce the amount of time needed for analysis. However, if the ramp rate
is too high, compounds in the mixture are eluted from the column too quickly for significant
interactions to occur with the stationary phase. Thus, the compounds are not well separated and
may begin to coelute, which can complicate their later identification.

38
 

Of the many types of detectors available for GC analysis, MS is one of the most
commonly used in forensic applications because it provides an additional analytical technique
that can be used for the identification of compounds in complex mixtures. The end of the GC
column passes through a heated transfer line into the mass spectrometer. The transfer line is
heated to prevent condensation of the analyte before it reaches spectrometer. The mass
spectrometer is comprised of three parts: the ion source, the mass analyzer, and the detector. All
three parts are kept under vacuum, necessary to prevent ion-ion and ion-molecule collisions after
2

ionization . Of the many types of ion sources, mass analyzers, and detectors available, mass
spectrometers with electron ionization, quadrupole mass analyzers, and electron multiplier
detectors are commonly found.
For electron ionization (EI), a current is passed through a thin filament housed in the ion
source. The filament is heated to incandesce, causing it to produce free electrons (Figure 2.5).
The electrons are then accelerated across an electrical potential (70 eV) towards an anode,
creating a beam of electrons. The GC column, coming through the transfer line, ends at the ion
source. Thus, each of the separated analytes eluted from the column are directed through this
electron beam. The electrons transfer enough energy to the analyte molecules to overcome the
2

first ionization energy of most organic molecules (10-20 eV) , causing the analytes to lose an
electron and to form molecular ions. Further interactions between the molecular ions and the
electron beam, or interactions with other molecules or ions, result in the formation of fragment
ions. The fragmentation pattern of each analyte is unique under a given set of conditions, thus
allowing for definitive identification of each individual compound in a mixture. All the
molecules that were not ionized are pumped away by the vacuum as waste.

39
 

Anode

e- + ee- e e
e- ee-

+
Repeller

Neutral
Molecules
from GC
eTransfer
- ee
Line
- e

Ions

To Mass
Analyzer

+
+

+

+

+

e- e- e- Electron
e
e- e- - Beam
e- e

Filament

Figure 2.5: Diagram of an ion source demonstrating electron ionization.

40
 

After EI, the ions have a positive charge. A positively charged plate, called a repeller, is
used together with an extractor and ion-focusing plates, all negatively charged, to propel the ions
into the quadrupole mass analyzer. The quadrupole consists of four parallel rods arranged in a
square and wired together to form two pairs of rods opposite each other (Figure 2.6). A
combination of a direct current (DC) and an alternating current, usually in the radio frequency
(RF) range, is applied to the rods, such that the charges on adjacent rods are in the opposite phase
and the charge on each pair of rods is constantly changing. As the ions enter the quadrupole,
only ions that have a stable trajectory at a given DC/RF ratio reach the detector. These ions are
known as resonant ions. All other ions, or non-resonant ions, are neutralized by collisions with
the rods and not detected (Figure 2.6).
The factors that determine the stability of the trajectory of an ion are the mass-to-charge
(m/z) ratio of the ion and the combination of DC voltage and RF frequency applied to the
quadrupole. Ions oscillate through the quadrupole differently due to their m/z ratio and the
frequency and magnitude of the voltage applied to the quadrupole. The two positive rods of the
quadrupole act as a high-pass mass filter. The heavier ions are too massive to be greatly affected
by the alternating voltage cycle, while the lighter ions will be more affected. Because of the
greater effect, the lighter ions are more likely to hit the rods and be neutralized while the heavier
ions pass through. At the same time, the combination of the DC and RF voltages acts as a lowpass mass filter. While the ratio between the DC and RF values is held constant, the values
themselves are increased simultaneously to effectively create a filter that separates ions based on
their m/z ratio. Because EI imparts a positive charge during ionization, all of the ions are
attracted to the negative rods of the quadrupole. However, because of the influence of the
positive rods, the lighter ions are less affected than the heavier ions (the combination of attractive

41
 

Resonant Ion

Detector

Non-resonant Ion

+
Ions

-

+

Ion
Source

Figure 2.6: Diagram of a quadrupole mass analyzer showing resonant and non-resonant ions.

42
 

and repulsive forces partially negates each other). Thus, the lighter ions stay within the
quadrupole, and only the heavier ions collide with the rods and are neutralized.
In total, to pass through the quadrupole, the ion must be heavy enough not to be affected
by the positive rods and light enough not to be affected by the negative rods. The only ions to
reach the detector, then, are within a limited range of m/z ratios for a given DC and RF value.
The specific range of ions that reach the detector can be changed by adjusting the applied
voltages. For a full mass scan, a broad range of m/z ratios (e.g., m/z 50-500) is scanned for each
analyte that elutes from the GC column. For selected ion monitoring (SIM), instead of scanning
a broad range of m/z ratios, the DC and RF values are held at a constant value or stepped through
a few values. SIM is used to increase sensitivity because the detector spends more time
detecting each ion. Additionally, only compounds with a particular molecular ion or fragment
ion are detected, so SIM can be used to simplify chromatographic data by reducing the number
of peaks observed in the chromatogram.
Once the resonant ions pass through the quadrupole, they are detected using a
continuous-dynode electron multiplier (Figure 2.7), used to amplify the ion signal and increase
2

the multiplier is held at a high negative potential while the innermost end goes to ground . The
ions are accelerated from the quadrupole into the wide end of the electron multiplier, attracted by
the applied negative charge. The ions collide with the wall of the multiplier and emit secondary
electrons, which are subsequently attracted to the relatively positive narrow end of the multiplier.
Each time the secondary electrons collide with the wall along the length of the detector, more
electrons are ejected. This cascade of electrons results in amplification of the signal. The signal
is further amplified by an operational amplifier (op-amp) that removes the background noise in
the electrical signal. Generally, the total amplification of signal using a continuous-dynode
43
 

Figure 2.7: Diagram of continuous-dynode electron multiplier.

44
 

6

electron multiplier is on the order of 10 .
The electrical current generated by the secondary electrons is digitized by an analog-todigital converter, and then the digitized signal is sent to a computerized data collection system,
which processes the data. The output of GC-MS analysis is the total ion chromatogram (TIC),
which is a sum of all of the ions detected over time (Figure 2.8A). The TIC is a plot of
abundance versus retention time, and each peak in the TIC has a corresponding mass spectrum
(Figure 2.8B). The mass spectrum is a plot of abundance versus m/z. Two peaks of note in the
mass spectrum are the base peak, or the most abundant ion in the spectrum, and the molecular
ion peak. These two peaks taken with all of the fragment ions are referred to as the
fragmentation pattern, which serves as a fingerprint so that compounds can be identified.

2.2 Data Pretreatment Procedures

Data pretreatment procedures are applied to chromatograms before data analysis to
reduce non-chemical sources of variance, such as differences in injection volume and
instrumental drift. In this research, background subtraction, smoothing, retention-time
alignment, and normalization were used.

2.2.1 Background Subtraction
Background subtraction is generally used to correct for low-frequency baseline drifts that
occur during instrumental analysis. In this research, the principle of background subtraction was
used to remove caprolactam from the ignitable liquid residue (ILR) extract chromatograms.
Caprolactam is a thermal degradation product from the nylon bags used during analysis. The
45
 

2E5
2E5

Abundance

Abundance

A

00
0
0

18
18

Retention Time (min)
Fragment Ions

45000
45000

B
57

40000

Base Peak

35000

Abundance
Abundance

30000
25000
20000

71

15000

Molecular
Ion

85

10000

99

5000

50

142
145

50
52
54
56.1
58.1
60
62
64
66
68.1
70.1
72.1
74.1
76.1
78.1
80.1
82
84.1
86.1
88.1
90.1
92.1
94.1
96
98.1
100.2
102.2
104.2
106.2
108.2
110.1
112.1
114.1
116.1
118.1
120.1
122.1
124.1
126.1
128.1
130.1
132.1
134.1
136.1
138.1
140
142.2
144
146
148
150

0
0

113

m/z

Figure 2.8: A) Total ion chromatogram (TIC) of neat kerosene and B) the mass spectrum of
decane (indicated in the TIC with an arrow) showing the molecular ion, base peak, and fragment
ions.
46
 

abundance of caprolactam observed in the chromatograms was not reproducible, and was
identified as a significant source of variance between samples during data analysis. Therefore, it
was determined that the caprolactam should be background subtracted from the data to ensure
that the greatest sources of variance in the data set were due to the ignitable liquids and matrix
interferences.
For this research, background subtraction was performed by first identifying the mass
spectrum of caprolactam. Then, the data analysis software subtracted this mass spectrum from
the entirety of the chromatogram, thereby removing any contributions of caprolactam to the
chromatogram (Figure 2.9). This process was repeated for each chromatogram in the data set
separately.

2.2.2 Savitzky-Golay Smoothing
After background subtraction, the chromatograms were smoothed to reduce the amount
of high-frequency noise introduced during data acquisition (Figure 2.10). In this research, a
Savitzky-Golay smoothing algorithm, a type of least-squares polynomial smooth, was used. For
this algorithm, the user defines both the number of chromatographic data points in a window
3

(window size) and the order of the polynomial function to be fit across the window . Window
sizes are smaller than the number of data points contained in a peak, and a second- or third-order
4

polynomial is usually selected, as these curves most closely match the peak shape .
Once the parameters are defined, the polynomial is fitted to each window, starting at the
beginning of the chromatogram. The algorithm solves the polynomial and replaces the center
value of the window with the value predicted by the model. The window is then shifted forward
by one data point, and the process is repeated along the entire chromatogram.
47
 

-6.0E4
6.0E4

Abundance

A

0
0

6.0E4
-6.0E4

Abundance

B

0

0

Retention Time (min)

17
-17

Figure 2.9: Total ion chromatograms of neat gasoline A) before and B) after background
subtraction of the caprolactam peak (indicated with arrow).

48
 

Abundance

1.0E4

0
12.00

Retention Time (min)

12.46

Figure 2.10: Portion of the total ion chromatogram of neat gasoline before smoothing (─) and
after application of the Savitzky-Golay algorithm (‐ ‐ ‐), demonstrating reduction of highfrequency noise in the peaks.

49
 

The Savitzky-Golay algorithm can be applied to nearly any kind of data and is relatively
fast and simple compared to other methods of smoothing. However, there are two limitations of
the Savitzky-Golay algorithm. First, each time the window is shifted, only the center value of
the window is recalculated. Thus, the first and last few data points in the chromatogram are not
smoothed because these points are never in the center of a window. For example, if a window
size of nine is chosen, the first four and last four data points are not smoothed. Also, oversmoothing may occur if the order of the polynomial does not accurately mirror the shape of the
peaks in the chromatogram. Over-smoothing can reduce the signal-to-noise ratio and can also
5

result in loss of resolution .

2.2.3 Retention-Time Alignment
Shifts in retention time can be caused by a number of variations during and between
chromatographic analyses, such as changes in carrier gas flow rate and column length as well as
6

degradation of the column over time . Retention-time alignment aims to correct for these shifts.
Several different methods of alignment are documented, but for this research, a correlation
optimized warping (COW) algorithm was used. Warping refers to the linear stretching and
compressing of segments along a chromatogram. Essentially, the COW algorithm warps a
sample chromatogram so that it more closely resembles a target chromatogram. The target
chromatogram can be randomly chosen from the data set, mathematically generated (e.g., an
average of several chromatograms), or prepared and analyzed as a separate sample. The chosen
target varies for the type of data, but must be representative of all of the chromatograms being
aligned.

50
 

First, the chromatogram is divided into segments. The number of data points in a
segment (i.e., the segment length) is user defined but should be more than the number of data
6

points of any single peak in the chromatogram . The other user-defined parameter is the warp
size, also known as the slack. The warp size is the number of data points that can be added or
subtracted from a segment. For example, with a warp size of 2, two points could be added, one
point could be added, no points could be added, one point could be subtracted, or two points
could be subtracted from each segment. All of these permutations are performed on each
segment beginning at the end of the chromatogram and moving toward the start.
For each permutation, the COW algorithm calculates the local correlation coefficient
between the warped segment and the corresponding segment in the target chromatogram. The
warping with the best correlation to the target chromatogram is determined and retained for use
in aligning the remaining segments. In addition, the local correlation coefficients for all other
warpings for that section are also retained. The algorithm then repeats the process for each of the
remaining segments. Once each segment has been aligned to the target chromatogram, a global
correlation coefficient is calculated by summing all combinations of the local correlation
coefficients. The combination of local correlation coefficients that yields the highest value for
the global correlation coefficient is determined to be the best alignment for the defined
parameters. This entire process is then repeated for each of the sample chromatograms being
aligned to the target. An example of chromatograms before and after retention-time alignment is
shown in Figure 2.11.
The COW algorithm can be used to align chromatograms of complex mixtures despite
6

variations in noise and baselines within a data set . However, one limitation of this algorithm is
the way in which the correlation coefficients are calculated. Because the correlations are
51
 

Abundance

1.0E6 A
1.0E6

0
0
6.5
6.5

Retention Time (min)

8.0
8.0

Abundance

1.0E6
1.0E6 B

0
0
6.5
6.5

Retention Time (min)

8.0
8.0

Figure 2.11: Total ion chromatograms of 90% evaporated gasoline standard (──) and extract
from a 90% evaporated gasoline thermal degradation sample (─ ─) A) demonstrating the shift in
retention times for the C3-alkylbenzenes, and B) those same peaks after retention-time
alignment.
52
 

insensitive to differences in relative peak abundance, the fronting or tailing edge of a peak in a
sample chromatogram may be aligned to the apex of a peak in the target chromatogram in order
to optimize the local correlation coefficient. Thus, although the local correlation coefficient
would increase, the peaks are visually misaligned in the chromatogram. Additionally, although
general guidelines can be found regarding the selection of warp and segment sizes, there is no
way to predict what combination of these parameters will yield the best alignment to the target.
Therefore, some amount of trial-and-error is required, which can be time consuming and
therefore limiting in a forensic setting.

2.2.4 Normalization
Differences in the abundance of peaks in the chromatograms of similar samples are the
3

result of variation in injection volume and detector response . Normalization is used to
minimize these differences (Figure 2.12). Many procedures for normalization are available, but
a combination of internal standard and total area normalizations was used for this research. An
internal standard normalization was used to correct for variation in peak abundances between
samples in the data set and ensure that peak maxima were on the same order of magnitude. Total
area normalization was used to reduce spread in abundance among replicates of samples.
For the internal standard normalization, each data point in the chromatogram was divided
by the abundance of the internal standard in that chromatogram. The data points were then
multiplied by the average abundance of the internal standard for all the chromatograms in the
data set. A similar process is also used for total area normalization, in which each data point in a
chromatogram is first divided by that summed abundance of the entire chromatogram. Then, the
data points are multiplied by the average total sum of all chromatograms in the data set.
53
 

Abundance

5.0E4
5.0E4 A

0
0

7.5

7.5

Retention Time (min)

7.7

7.7

5.0E4

Abundance

5.0E4 B

0
0

7.5
7.5

Retention Time (min)

7.7
7.7

Figure 2.12: Portion of total ion chromatograms (n=15) of neat kerosene showing the C10
normal alkane peak both A) before and B) after normalization of the data.

54
 

2.3 Principal Components Analysis

Principal components analysis (PCA) is a multivariate statistical technique that is used to
reduce the dimensionality of large data sets to a few variables that contribute most to the
variance in that data set. PCA is used to visually assess the association and discrimination of
samples, a task made easier because only a few variables are used rather than the entire data set.
Before PCA is applied, the pretreated data are mean centered. Mean centering redefines
the average of the data set as zero and ensures that the first principal component (PC1) describes
7

the most variance in the data set . In PCA, each data point in each sample is considered to be a
variable. The average value across each variable in the data set is calculated, and then subtracted
from each data point in the data set. To be more specific, for the chromatographic data in this
research, the average abundance of all of the chromatograms at each retention time is subtracted
from the abundance at the same retention time for each chromatogram in the data set.
After the data have been mean centered, the covariance matrix is calculated. Covariance
8

is the measure of how two dimensions vary from the mean with respect to each other .
Covariance is calculated according to the following equation:

,

∑
̅

Equation 2.1

where x and y are data points in the x and y dimensions, respectively, and n represents the total
8

number of dimensions . The covariance is calculated between all variables in the data set, and
the values form an n x n matrix. For chromatographic data, the number of dimensions is equal to
the number of data points in the chromatogram (approximately 5,000 in this research). For a

55
 

simplified example, a data set with x, y, and z dimensions would yield the following covariance
matrix:

,
,
,

,
,
,

,
,
,

It should be noted that because cov(x,y) = cov(y,x), the covariance matrix is symmetrical around
the diagonal. Also, along the diagonal, the covariance is measured between the dimension and
itself, which is equivalent to the variance of that dimension.
From the covariance matrix, eigenvectors and eigenvalues are calculated. The
eigenvector is a unit vector that, when multiplied by the data matrix, yields a resulting vector that
is a multiple of the original unit vector. The eigenvalue is the factor by which the eigenvector
changed from the original data matrix. For an n x n matrix, there are n eigenvectors, all
orthogonal to each other and each with an associated eigenvalue that corresponds to the amount
of variance described by that eigenvector. For example, the eigenvector with the largest
associated eigenvalue is considered the first principal component (PC1). The remaining n-1
eigenvectors are ranked in order according to decreasing eigenvalue. The amount of variance
described by each eigenvector is determined by the percentage of its associated eigenvalue to the
total sum of eigenvalues for all eigenvectors.
The score for each sample is calculated by first multiplying the mean-centered data by the
eigenvector for each PC. The product of each variable of the mean-centered data and the
eigenvector at the corresponding retention time is referred to as a loading. The loadings at all
retention times are then summed to yield the score of the sample on that PC. For example, if the
data are multiplied by the eigenvector for PC1, the sum of the loadings would be the sample’s
score on PC1. The score of a sample is calculated for all PCs, but generally only scores on the
56
 

first two principal components are used to generate a scores plot. The scores plot is a scatter plot
in which chemically similar samples are clustered together for simplified visual analysis.
In addition to the scores plot, loadings plots can also be generated. For chromatographic
data, the loadings plot is a plot of the eigenvector of a PC versus retention time and is used to
identify the compounds in the chromatograms that contribute most to the variance in the data set.
The compounds are weighted in the loadings plot to reflect the relative amount of variance each
contributes. Loadings plots can also be used to explain the positioning of the samples on the
scores plot.
For this research, eigenvectors and eigenvalues were calculated using a set of ignitable
liquid standards, and then used to calculate the projected scores of ILR extracts. Projecting the
scores of the extracts ensured that only the compounds in the standards contributed to the scores
of the extracts. To calculate the projected scores, the ILR data were first mean centered by
subtracting the average value of each variable from the liquid standard data used to generate the
eigenvectors. That is, the same average was subtracted from each of the extracts as was from the
standards. The mean-centered data were then multiplied separately by the eigenvectors for PC1
and PC2 calculated for the standards to determine the loading of the extracts at each retention
time. The loadings on each PC were summed separately to yield the score of the extracts on PC1
and PC2. These scores were then plotted on the same scatter plot as the standards, so that the
association of the ILR extracts to the corresponding ignitable liquid standard could be visually
assessed.
While PCA can be used to identify the variables that contribute most to the variance
observed in a data set, only the PCs with the highest eigenvalues are assessed. Therefore, while
the majority of the variance is considered, some is disregarded. In addition, association and

57
 

discrimination in the PCA scores plot is visually assessed, introducing some subjectivity into
interpretation of results.

2.4 Pearson Product Moment Correlation Coefficients

Pearson product moment correlation (PPMC) coefficients provide a statistical measure of
the correlation between two samples. For the chromatographic data generated in this research,
the correlation is calculated between two chromatograms at every data point (i.e., every retention
time). To calculate the PPMC coefficient, r, between two chromatograms, x and y, the following
equation is used:

∑
̅

Equation 2.2

∑

∑
̅

where n represents the total number of data points in the chromatogram. In this equation, the
numerator is equal to the covariance in the data, while the denominator is the variance. The
value of r can range from +1 to -1, where +1 indicates a perfect positive correlation between the
two samples and -1 is a perfect negative correlation. Coefficients between ±0.8 and ±1 indicate a
strong correlation, between ±0.5 to ±0.79 indicate a moderate correlation, and between 0 and
9

±0.49 indicate a weak correlation .
The correlation between two chromatograms is insensitive to differences in abundance.
Instead, the correlation depends only on changes in the slope of the line, such as where peaks
start and end in the chromatogram and the inflection point at the apex of each peak. Thus, data
that have been normalized will yield the same PPMC coefficients as data that have not been
58
 

normalized, but coefficients may be different for data that have been retention time aligned and
unaligned data due to shifts in the retention times of peaks (i.e., r should be higher for the
aligned data).
In this research, PPMC coefficients were first calculated between replicates of extracts to
assess precision in the extraction and analysis procedures. These replicates are expected to have
PPMC coefficients very close to one as the chromatograms should be nearly identical. In
addition, coefficients were used to measure the similarity between ILRs and the corresponding
ignitable liquid standard. These PPMC coefficients were expected to be less than one, due
differences between the standard and sample chromatograms, such as the addition of
interferences from the matrix. However, the coefficients should show stronger correlations
between the ILRs and the corresponding standard than between the ILRs and the other ignitable
liquid standards.

59
 

References

60
 

References

th

1. Skoog DA, Holler FJ, Crouch SR. Principles of instrumental analysis. 6 ed. Belmont,
CA: Thomson, 2007.
th

2. Watson JT, Sparkman OD. Introduction to mass spectrometry. 4 ed. Hoboken, NJ: John
Wiley & Sons, Ltd, 2007.
3. Morgan SL, Bartick EG. Discrimination of forensic analytical chemical data using
multivariate statistics. In: Blackledge, RD, editor. Forensic analysis on the cutting edge:
new methods for trace evidence analysis. Hoboken, NJ: John Wiley& Sons, Inc., 2007;
333-367.
4. Beebe KR, Pell RJ, Seasholtz MB. Preprocessing the samples. In: Chemometrics: a
practical guide. New York: Wiley, 1998; 26-55.
5. Brereton RG. Applied chemometrics for scientists. Hoboken, NJ: John Wiley & Sons,
Ltd, 2007.
6. Nielsen NPV, Carstensen JM, Smedsgaard J. Aligning of single and multiple wavelength
chromatographic profiles for chemometric data analysis using correlation optimized
warping. J Chromatogr A 2003; 996: 141-55.
th

7. Miller JN, Miller JC. Statistics and chemometrics for analytical chemistry. 4 ed.
Harlow, England: Pearson Education Limited, 2000.
8. Smith LI, A tutorial on principal components analysis. 2002.
<www.cs.otago.ac.nt/cosc453/ student-tutorials/principal-components>
rd

9. Devore JL. Probability and statistics for engineering and the sciences. 3 ed. Belmont,
CA: Duxbury Press, 1991.

61
 

Chapter 3 - Using Multivariate Statistical Procedures to Identify Ignitable Liquid Residues
in the Presence of Matrix Interferences from Carpet

3.1 Introduction
Carpet is a common household material oftentimes found at residential fire scenes. When
1

burned, carpet is known to contain compounds that can complicate the analysis of fire debris .
This research aims to develop an objective method for associating ignitable liquid residues
(ILRs) from fire debris to the corresponding ignitable liquid standard despite the presence of
interferences from the matrix, as well as evaporation of the ignitable liquid and thermal
degradation of both the liquid and the matrix during the burning process. A method of analysis
was developed using a combination of multivariate statistical procedures, namely principal
component analysis (PCA) and Pearson product moment correlation (PPMC) coefficients.
First, gasoline and kerosene were each evaporated to two different levels. The neat and
evaporated liquids were used as a set of standards to generate a PCA scores plot. Each of the
ignitable liquids was then spiked onto unburned carpet. The samples were extracted and
analyzed, and the scores for the extracts calculated and projected onto the scores plot generated
using the standards. The scores plot was used to investigate the effect of interferences inherent
in the matrix on the association of the extracts to the corresponding liquid standards. Next, the
liquids were spiked onto burned carpet to assess the effects of additional interferences introduced
during the burning process. The carpet had previously been burned for an amount of time
chosen to generate maximum interferences. Finally, each of the liquids was spiked onto
unburned carpet. The carpet was then burned to simulate fire debris to investigate the effect of
thermal degradation on the association of the extracts to the corresponding liquid standards.
62
 

All samples were analyzed using gas chromatography-mass spectrometry (GC-MS). The
total ion chromatograms (TICs) were compiled into two data sets and subjected to data
pretreatment procedures (smoothing, retention time alignment, and normalization). PCA and
PPMC coefficients were applied to assess the association of the sample extracts to the
corresponding ignitable liquid standard despite the presence of matrix interferences, evaporation,
and thermal degradation.

3.2 Materials and Methods

3.2.1 Ignitable Liquid Collection
Gasoline and kerosene were collected from service stations in the Lansing, Michigan,
area. Gasoline was evaporated to 10% and 90% by volume, and kerosene was evaporated to
10% and 70% by volume, using filtered air and magnetic stir bars for agitation. Once
evaporated, the liquids were stored in acid-washed, 7.4 mL amber screw-cap vials (Fisher
Scientific, Pittsburgh, PA). Vials were labeled, wrapped in Parafilm® (American National Can,
Menasha, WI), and stored at 4°C until analysis.

3.2.2 Ignitable Liquid Standard Preparation
Each of the neat and evaporated ignitable liquids was diluted 1:25 (v:v) in
dichloromethane (CH2Cl2) (spectrophotometric grade, Jade Scientific, Canton, MI) containing
nitrobenzene (0.03 M, Mallinckrodt, Inc., Parts, KY) as an internal standard.

63
 

2

A 20 µL aliquot of each diluted ignitable liquid was then spiked onto a 4 × 4 cm

Kimwipe™ (Kimberly-Clark, Irving, TX) in a nylon bag (Grand River Products, LLC, Grosse
Pointe Farms, MI). Five replicate samples were prepared for each of the neat and evaporated
ignitable liquids. Samples were placed in an 80°C oven and extracted using a passive headspace
extraction with one-fourth of an activated carbon strip (Albrayco Technologies, Inc., Cromwell,
CT), following procedures recommended by the American Society for Testing and Materials
1

(ASTM) . After extraction, the activated carbon strip was eluted with 200 µL of CH2Cl2 and
analyzed in triplicate by GC-MS.
A consensus standard (necessary for alignment) was prepared by diluting neat gasoline
and neat kerosene 1:1:25 (v:v) in the same aliquot of CH2Cl2 and spiking the mixture onto a
single Kimwipe™ in a nylon bag. The consensus standard was extracted as previously
described and analyzed by GC-MS.

3.2.3 Inherent Matrix Interference Sample Preparation
2

Nylon carpet (source unknown) was cut into 4 × 4 cm squares, and each square was
placed into a separate nylon bag. A 20 µL aliquot of each diluted ignitable liquid standard was
spiked onto separate carpet samples (n=5 for each standard). The samples were extracted as
previously described and analyzed by GC-MS.

64
 

3.2.4 Burned Matrix Interference Sample Preparation
Burn times were investigated to determine the time that generated the most abundant
2

matrix interferences. To do this, nylon carpet squares, measuring 4 × 4 cm , were burned with a
propane torch (Benzomatic, Medina, NY) for 10 to 120 seconds. The samples were extinguished
by smothering, then placed in a nylon bag and extracted according to the previously described
procedures. The extracts were then analyzed by GC-MS.
2

After determining the burn time, samples of 4 × 4 cm squares of carpet were burned for
60 sec, as described previously, and placed into separate nylon bags. The samples were then
spiked with 20 μL of the diluted neat and evaporated ignitable liquids (n=5 for each liquid). The
samples were extracted as described previously and analyzed by GC-MS.

3.2.5 Thermal Degradation Sample Preparation
2

Separate samples of carpet, measuring 4 × 4 cm , were spiked with each of the ignitable
liquid standards (n=5 for each liquid standard) then burned for 60 s to generate significant matrix
interferences. Aliquots of 750 μL of each gasoline standard were spiked onto the matrix, while
aliquots of 250 μL of each kerosene standard were spiked. The spike volumes were chosen such
that the ignitable liquid did not mask the interferences from the matrix in the chromatogram. In
addition, five samples of carpet were burned for 60 seconds with no ignitable liquid added, to
serve as a control.

65
 

3.2.6 GC-MS Analysis
All extracts were analyzed in triplicate using an Agilent 6890 gas chromatograph, with an
Agilent 7683B automated liquid sampler, coupled to an Agilent 5975 mass spectrometer (Agilent
Technologies, Santa Clara, CA). The GC was equipped with an Agilent HP-5Ms capillary
column (30 m × 0.25 mm internal diameter × 0.25 µm film thickness). A sample volume of 1 μL
was injected in the pulsed, splitless mode, with a pressure pulse of 15.0 psi for 0.25 min. The
inlet temperature was 250°C, and the carrier gas, ultra-high purity helium (Airgas, East Lansing,
MI), was held a nominal flow rate of 1 mL/min. The following GC temperature program was
used: 40°C for 3 min, 10°C/min to 280°C, hold for 4 min at 280°C. The transfer line was held
at 280°C. Electron ionization (70 eV) was used and the quadrupole mass analyzer was operated
in full scan mode (m/z 50-550) at a scan rate of 2.91 scans/s.

3.2.7 Data Pretreatment
The total ion chromatograms (TICs) generated for all samples were background
subtracted to remove the caprolactam peak. Caprolactam is a thermal degradation product from
the nylon bags and was removed from the sample TICs so that it did not contribute to the
variance in the data set. The background subtraction was performed by first identifying the mass
spectrum of caprolactam using ChemStation© Enhanced Data Analysis Software (Agilent
Technologies, version E.01.01.335). The mass spectrum for caprolactam shows a molecular ion
peak at m/z 113 and several significant mass fragments at m/z 55 and 85. The software
subtracted this mass spectrum from the entirety of the chromatogram, thereby removing any

66
 

contributions of caprolactam to the chromatogram. After background subtraction, the TICs were
smoothed with the Savitzky-Golay algorithm using ChemStation© software.
The TICs were then retention time aligned to a target chromatogram, generated by
averaging the replicate chromatograms of the consensus standard. A commercially available
correlation optimized warping (COW) algorithm (LineUp™, version 1.0.62, Infometrix, Inc.,
Bothwell, WA) was used for the alignment. A variety of warp and segment sizes were
investigated to select appropriate parameters for the alignment. To determine these parameters,
the chromatograms of the ignitable liquid standards were visually assessed for peak
misalignments and also evaluated using PPMC coefficients calculated in Excel (version
12.0.6425.1000, Microsoft Corp.). The selected parameters were then used to align the
remaining chromatograms in the data set.
All of the chromatograms were subsequently divided into two separate data sets for
normalization and data analysis. The first data set consisted of TICs of the ignitable liquid
standards and the extracts of the unburned carpet spiked with the ignitable liquids. The second
data set consisted of the ignitable liquid standards with the burned carpet extracts and the thermal
degradation extracts.
Normalization was conducted in the same manner for both data sets. Each chromatogram
was first normalized to the internal standard in that chromatogram. Normalization to the internal
standard was performed to ensure that the abundance of the maximum peak in each of the
chromatograms was generally on the same order of magnitude. For this normalization, each data
point in the chromatogram was divided by the maximum abundance of the internal standard peak
in that chromatogram and then multiplied by the average abundance of that peak in all
chromatograms in the data set. A total area normalization of the chromatograms was then
67
 

performed to minimize the spread among replicates. For the total area normalization, each data
point in the chromatogram was divided by that total area of that chromatogram and then
multiplied by the average area of all chromatograms in the data set.

3.2.8 Data Analysis
PCA was performed on chromatograms of the ignitable liquid standards only, to
minimize the effect of the addition of matrix interferences and maximize the differentiation
among the standards. Thus, only compounds from the matrix that are found in gasoline or
kerosene or compounds from the matrix that have the same retention time as compounds in those
two ignitable liquids contributed to the scores of the sample extracts. All other interference
compounds did not contribute to the variance and, therefore, had no effect on the positioning of
extracts on the PCA scores plot.
Eigenvectors and eigenvalues for the liquid standards were generated using Matlab
(version 7.7.0.471, The MathWorks, Inc., Natick, MA), while scores and loadings plots were
generated in Microsoft Excel. Scores for all sample extracts were calculated in Microsoft Excel.
To do this, the data was first mean centered by subtracting the average abundance of the
ignitable liquid standards at each retention time from each of the sample chromatograms. After
mean centering, the chromatograms were multiplied by the eigenvector for PC1 to find the
loading of the sample at each retention time. The loadings at all retention times were summed to
determine the score of the sample on that PC. This process was repeated for PC2. The
calculated scores were projected onto the scores plot generated using the standards. The scores
plot was then used to assess the association of the sample extracts to the corresponding neat
liquid in the presence of the matrix interferences found in unburned. This association was also

68
 

assessed in the presence of interferences from burned carpet as well as thermal degradation of
the ignitable liquid and matrix.
PPMC coefficients were calculated for all pair-wise combinations of the pretreated
chromatograms using Microsoft Excel. The PPMC coefficients were calculated between
replicates to assess the precision of the extraction and analysis procedures. Coefficients were
also calculated for each data set to assess the correlation of the sample extracts to the ignitable
liquid standards.

3.3 Results and Discussion

3.3.1 Selection of Retention-Time Alignment Parameters
Selection of alignment parameters was determined based on the analysis of the ignitable
liquid standards only. For the COW algorithm, several combinations of warp and segment sizes
were investigated. Warp sizes ranged from 2 to 5 points, while segment sizes ranged from 25 to
85 points. PPMC coefficients were calculated for each combination of alignment parameters and
compared to the coefficients calculated for the unaligned data. Alignments that had the highest
PPMC values compared to the unaligned data were considered to be well-aligned and were then
visually assessed for misalignments.
For most combinations of warp and segment size, misalignments of peaks in the
chromatograms of the ignitable liquid standards were still present. The greatest number of
misalignments was observed for combinations with larger warp and segment sizes. Large
segments may contain multiple peaks and, when combined with large warp sizes (e.g., a warp
size of five), can drastically shift the peaks, resulting in misalignments. For example, a warp size

69
 

of 5 points and a segment size of 60 points showed major misalignments for the 1methylnaphthalene peak in 90% evaporated gasoline (Figure 3.1A). Visual assessment of the
chromatograms indicated that the best parameters for the COW alignment were a warp size of 2
points and a segment size of 75 points (Figure 3.1B).
PPMC coefficients were calculated between all pair-wise combinations of
chromatograms. When summed, the total of the coefficients for the unaligned data was 2169.
This value was then compared to the summed PPMC coefficients of TICs after application of
each combination of alignment parameters. For example, the sum of coefficients for a warp size
of 5 points and a segment size of 60 points was 2055, 114 less than the unaligned data. This
smaller value indicates that PPMC coefficients are generally lower and is most likely due to the
misalignments in the chromatograms discussed previously. On the other hand, a warp size of 2
points and a segment size of 75 points gave a sum of 2201, 32 greater than the unaligned data
and 146 greater than the other alignment parameters. This increase indicates that the peaks in the
chromatograms are aligned better using these parameters compared to alignments using larger
warp and larger segment sizes as well as the unaligned data.

3.3.2 Selection of Normalization Method
Because gasoline and kerosene share no common compounds and the chemical
composition of each liquid changed drastically with evaporation, an internal standard was added
to the samples for normalization. In addition, several types of total area normalization were
investigated, such as normalizing by ignitable liquid type and normalizing the entire data set, to
minimize spread among replicates (Figure 3.2). For each, the normalized data were subjected to
PCA and the resultant scores plot was visually assessed for close clustering of extracts.

70
 

3.0E
3.0E5

Abundance

A

0
0

12.35
12.35

Retention Time (min)

12.55
12.55

Retention Time (min)

12.55
12.55

Abundance

3.0E5
3.0E5 B

00

12.35
12.35

Figure 3.1: A) Misaligned 1-methylnaphthalene peak in the total ion chromatograms (TICs) of
90% evaporated gasoline replicates (n=15) using a warp size of 5 points and a segment size of 60
points. B) Well-aligned 1-methylnaphthalene peak in the TICs of 90% evaporated gasoline
using a warp size of 2 points and a segment size of 75 points.

71
 

Abundance

2.0E5
2.0E5 A
A

0
0

5.63
5.63

Retention Time (min)

5.75
5.75

Retention Time (min)

5.75
5.75

2.0E5

Abundance

2.0E5 B B

00

5.63
5.63

Figure 3.2: Comparison of A) aligned but not normalized p-xylene peak in the total ion
chromatograms (TICs) of neat (─), 10% evaporated (─ • ─), and 90% evaporated gasoline (‐‐‐)
and B) p-xylene peak in the TICs of neat and evaporated gasoline after internal standard and
total area normalization, showing the minimized spread among replicates of the three
evaporation levels.

72
 

Total area normalization of each data set yielded the closest clustering of extract and was,
therefore, chosen as the normalization method. However, it should be noted not all differences
in abundance could be corrected due to the vastly different chemical compositions both between
the gasoline and kerosene and between different evaporation levels of each liquid.

3.3.3 Visual Assessment of Ignitable Standard Chromatograms
The chromatogram of neat gasoline shows the characteristic compounds of gasoline,
namely toluene and the C2- and C3-alkylbenzenes, in high abundance (Figure 3.3A).
Additionally, smaller peaks are present from the C4-alkylbenzenes and the methylnaphthalenes.
Evaporation has very little effect on the chromatogram of 10% evaporated gasoline, indicating
minimal loss of the volatile compounds: toluene and the alkylbenzenes (Figure 3.3B). However,
evaporation to 90% by volume drastically changes the appearance of the chromatogram (Figure
3.3C). Toluene is no longer observed in the chromatogram, and the abundance of the C2alkylbenzenes is greatly reduced. At the same time, the abundance of the C3- and C4alkylbenzenes increases due to concentration of these compounds after evaporation of the more
volatile toluene and C2-alkylbenzenes.
The chromatogram of neat kerosene shows small contributions from some aromatic
compounds, but is dominated by the C10-C16 normal alkane peaks in a bell-curve distribution
(Figure 3.3D). As was observed with gasoline, evaporation by 10% volume has little effect on
the chromatogram (Figure 3.3E). The aromatic compounds are lost and the C10 normal alkane
abundance is reduced, but the remaining normal alkanes are unchanged. Evaporation to 70% by
73
 

1.5E6

0
1.5E6

C3-alkylbenzenes
C2-alkylbenzenes
IS
b
h
Toluene
ef
Naphthalenes
a c dg

A

C3-alkylbenzenes
C2-alkylbenzenes
IS
Toluene
Naphthalenes

B

Abundance

0
1.5E6

C

C3-alkylbenzenes
IS Naphthalenes

C2-alkylbenzenes
0

1.5E6

D
C10

IS C11 C12 C13 C
14

C15

0

C16

1.5E6

E

C10

0

IS C11 C12 C13 C
14

C15

1.5E6

F
IS

0

C16

0

C13
C12

Retention Time (min)

C14

C15
C16

17

Figure 3.3: Total ion chromatograms of A) neat gasoline, B) 10% evaporated gasoline, C) 90%
evaporated gasoline, D) neat kerosene, E) 10% evaporated kerosene, and F) 70% evaporated
kerosene. Major compounds are labeled: a) ethylbenzene, b) p-xylene, c) o-xylene, d)
propylbenzene, e) m-ethyltoluene, f) p-ethyltoluene, g) o-ethyltoluene, and h) 1,3,5trimethylbenzene. IS indicates the internal standard, nitrobenzene.
74
 

volume results in the loss of the aromatic compounds as well as the C10 and C11 normal alkanes
(Figure 3.3F). The abundance of the C12 and C13 normal alkanes are reduced, while the
abundance of the C14-C16 normal alkanes increases due to concentration. The peak ratios of the
alkanes also change, and C14 is the most abundant normal alkane at 70% evaporation level.

3.3.4 Association and Discrimination of Neat and Evaporated Liquid Standards
The scores plot of the first principal component (PC1) and the second principal
component (PC2) accounts for 87.7% of the variance among the neat and evaporated ignitable
liquid standards (Figure 3.4). Spread among replicates is observed due to variation in
abundances not corrected by normalization and some variation in the extraction procedure.
Overall, the liquids are differentiated in PC1 and PC2, with the exception of neat and
10% evaporated gasoline. Gasoline samples are positioned negatively on PC1, while kerosene
samples are positioned positively on this PC. Neat gasoline samples, as well as samples of 10%
evaporated gasoline and all of the evaporation levels of kerosene, are positioned negatively on
PC2, though some replicate samples of neat kerosene are positioned on the x-axis. Samples of
90% evaporated gasoline are positioned positively on PC2.
The loadings plots for PC1 and PC2 are shown in Figure 3.5. PC1 (Figure 3.5A)
discriminates the ignitable liquids based on the presence of toluene, C2-alkylbenzenes, and C3alkylbenzenes, which load negatively, and the C10 to C16 normal alkanes, which load positively
on this PC. The second principal component (Figure 3.5B) discriminates the liquids based on the
positively loading C3- and C4-alkylbenzenes and the negatively loading toluene, C2-

75
 

PC2 (14.6%)

2E6

-3E6

3E6

-2E6

PC1 (73.1%)
Figure 3.4: Scores plot of the first principal component (PC1) versus the second principal
component (PC2) based on the total ion chromatograms for the two ignitable liquids at three
levels of evaporation, denoted as follows: Neat gasoline ( ), 10% evaporated gasoline ( ),
90% evaporated gasoline ( ), neat kerosene ( ), 10% evaporated kerosene ( ), and 70%
evaporated kerosene ( ).

76
 

Loadings on PC1

0.3

A

C11

C12

C13 C14
C15

C10
0

0

C16

6d
a

c

18

12
g
f

IS

e
h

Toluene

-0.3

0.3

C3-alkylbenzenes
C2-alkylbenzenes
Retention Time (min)
b

B

C3-alkylbenzenes

Loadings on PC2

h

0

C4-alkylbenzenes

ef
g
d

0

a c

6

12

C16

C13
b

IS

C2-alkylbenzenes

C14

18

C15

Toluene
-0.3

Retention Time (min)

Figure 3.5: Loadings plots of A) the first principle component (PC1) and B) the second principle
component (PC2) based on the total ion chromatograms of the six ignitable liquid standards.
Major compounds are labeled: a) ethylbenzene, b) p-xylene, c) o-xylene, d) propylbenzene, e) methyltoluene, f) p-ethyltoluene, g) o-ethyltoluene, and h) 1,3,5-trimethylbenzene. IS indicates the
internal standard, nitrobenzene.
77
 

alkylbenzenes, and the C13-C16 normal alkanes. The loadings plots of PC1 and PC2 can be used
to explain the positions of the neat and evaporated liquids on the scores plot based on chemical
composition.
Neat gasoline is positioned negatively on PC1 due to the presence of toluene and the C2and C3-alkylbenzenes, which load negatively on PC1. Compared to neat gasoline, the 10%
evaporated gasoline samples are positioned slightly more negatively on PC1 and slightly more
positively on PC2. The loss of toluene and the C2-alkylbenzenes during evaporation would be
expected to position the 10% evaporated gasoline samples more positively on PC1. However,
the reduced negative contribution to the scores of the samples from the evaporative loss of these
compounds is balanced by an increased negative contribution from the concentrated C3alkylbenzenes. Thus, the 10% evaporated gasoline samples are positioned slightly more
negatively on PC1 than the neat gasoline samples. On the other hand, the C3-alkylbenzenes load
positively on PC2, so both the evaporative loss of toluene and the C2-alkylbenzenes and the
concentration of the C3-alkylbenzenes results in a more positive positioning of the 10%
evaporated gasoline samples compared to the neat gasoline samples.
Gasoline at the 90% evaporation level is positioned less negatively on PC1 than neat and
10% evaporated gasoline due to the loss of toluene and the C2-alkylbenzenes and the resulting
concentration of the C3-alkylbenzenes during evaporation. Due to dominance of the C3alkylbenzenes and the C4-alkylbenzenes at the 90% evaporation level, these gasoline samples are
positioned positively on PC2 because these compounds load positively on this PC.
78
 

Neat kerosene is positioned positively on PC1 due to the presence of the C10-C16 normal
alkanes, which load positively on this PC. Because the C13-C16 alkanes load negatively on PC2,
neat kerosene is positioned negatively on PC2. The 10% evaporated kerosene is positioned more
positively on PC1 than neat kerosene due to the concentration of the C10-C16 normal alkanes
during evaporation and more negatively on PC2 due to the concentration of the C13-C16 normal
alkanes during evaporation. Similarly, kerosene evaporated to the 70% evaporation level is
positioned positively on PC1 and negatively on PC2 due to the dominance of the C13-C16 normal
alkanes as a result of evaporation.

3.3.5 Association and Discrimination of Neat and Evaporated Liquid Standards using PPMC
Coefficients
PPMC coefficients are insensitive to relative differences in peak abundance between
chromatograms. Therefore, the differences in relative abundance of compounds in the
chromatograms, which resulted in spread among the liquid standards on the PCA scores plot, had
no effect on the correlation between the standards. Mean PPMC coefficients of replicates of the
liquid standards are not equal to the theoretical value of 1.00, due to variation in the extraction
procedure (Table 3.1). However, all of the PPMC coefficients for replicates are greater than
0.98, indicating acceptable precision in the extraction procedure. Additionally, all of the
gasoline standards show weak correlations (<0.4) to the kerosene standards, thus the two types of
ignitable liquid can be differentiated from each other.
Replicates of neat and 10% evaporated gasoline are positioned closely on the scores plot,
indicating that the two are chemically similar. This similarity is further demonstrated by the
79
 

Table 3.1: Mean Pearson product moment correlation coefficients (± standard deviation) for all
replicates (n=15) of gasoline and kerosene liquid standards correlated to replicates of each
standard (n=225).

Neat
Gasoline
Neat
Gasoline
10%
Evaporated
Gasoline
90%
Evaporated
Gasoline
Neat
Kerosene
10%
Evaporated
Kerosene
70%
Evaporated
Kerosene

10%
90%
Evaporated Evaporated
Gasoline
Gasoline

10%
70%
Evaporated Evaporated
Kerosene
Kerosene

0.987 ± 
0.012
0.988 ±
0.011

0.993 ±
0.006

0.63 ±
0.03

0.624 ±
0.03

0.997 ±
0.0002

0.30 ±
0.05

0.290 ±
0.04

0.395 ±
0.015

0.994 ±
0.005

0.25 ±
0.04

0.236 ±
0.03

0.338 ±
0.013

0.989 ±
0.005

0.994 ±
0.004

0.11 ±
0.02

0.102 ±
0.018

0.104 ±
0.007

0.608 ±
0.012

0.641 ±
0.013

80
 

Neat
Kerosene

0.994 ±
0.005

strong correlation (0.988 ± 0.011) observed when PPMC coefficients are calculated between the
two standards. On the other hand, 90% evaporated gasoline replicates are positioned quite far
from the other gasoline standards and show only a moderate correlation to these standards
Therefore, the 90% evaporated standard is not correlated to the other standards, but the neat and
10% evaporated gasoline are correlated to each other. This result was expected because of the
similarity of the compounds observed in the TICs of neat and 10% evaporated gasoline and the
comparable difference between those two standards and the 90% evaporated gasoline standard
due to evaporative loss of toluene and the C2-alkylbenzenes. A similar trend is observed for the
kerosene standards, where only the 70% evaporated kerosene can be differentiated from the other
kerosene standards. This result is also expected due to the loss of the aromatic compounds as
well as the C10 and C11 normal alkanes going from neat kerosene to the 70% evaporation level.

3.3.6 Identification of Inherent Matrix Interferences
The TICs of the unburned carpet with no ignitable liquid present show inherent matrix
interferences dominated by the C9-C12 branched alkanes and 2-ethyl-1-hexanol (Figure 3.6).
These interferences are potentially from the adhesive, the yarn, and the backing material in the
carpet, though this has not yet been confirmed.
The addition of these matrix interferences can be observed in the chromatograms of the
carpet samples spiked with gasoline and kerosene (Figure 3.7). As expected, the spiked samples
contain the characteristic components from the ignitable liquids, such as toluene and the
alkylbenzenes for gasoline (Figure 3.3A) and the C10-C16 normal alkanes for kerosene (Figure
3.3D). However, the presence of the matrix interferences has visually altered the
81
 

Abundance

1E5
1E5

C9-C branched
C9-C1212 branched
alkanes
alkanes
2-ethyl-1-hexanol

00

00

16
16

Retention Time (min)
Figure 3.6: Total ion chromatogram of unburned carpet.

82
 

A

Toluene

C2-alkylbenzenes
b
C9-C12 branched
alkanes
C3-alkylbenzenes
h
IS
e

Abundance

4.0E5
4.0E5

a

f
g

c
d

00

0
4.0E5
4.0E5

17

Retention Time (min)

B

C9-C12 branched
alkanes
C11

Abundance

C12
IS
C10

C13

C14
C15 C16

00

0

Retention Time (min)

17

Figure 3.7: Total ion chromatograms of carpet spiked with A) neat gasoline and B) neat
kerosene showing the addition of the C9-C12 branched alkanes from the matrix. Major
compounds are labeled: a) ethylbenzene, b) p-xylene, c) o-xylene, d) propylbenzene, e) methyltoluene, f) p-ethyltoluene, g) o-ethyltoluene, and h) 1,3,5-trimethylbenzene. IS indicates the
internal standard, nitrobenzene.
83
 

chromatograms. The C9-C12 branched alkanes from the matrix coelute with the C10-C12 alkanes
in kerosene as well as the C4-alkylbenzenes in gasoline, thus changing the ratios between these
compounds and the other compounds observed in the chromatograms. Thus, the addition of
these compounds complicates visual identification of the ignitable liquid present in sample
extracts.

3.3.7 Association and Discrimination of Samples in the Presence of Inherent Matrix
Interferences
To assess the effect of inherent matrix interferences on the association of extracts to the
corresponding standard, the ignitable liquid standards were spiked onto unburned carpet,
extracted, and analyzed. Scores calculated for the extracts were then projected onto the scores
plot generated for the liquid standards. The liquids extracted from unburned carpet are generally
positioned closely to their corresponding ignitable liquid standard (Figure 3.8). However, it
should be noted that because the neat and 10% evaporated gasoline standards overlap on the
scores plot, it is not possible to visually assign the sample extracts to one evaporation level.
Neat and 10% evaporated gasoline extracted from the unburned carpet are positioned
slightly more positively on both PC1 and PC2 compared to the corresponding liquid standards.
This is due to the addition of the C9-C12 branched alkanes from the matrix that elute between
nine and ten minutes in the chromatogram. Some of these alkanes elute at the same retention
times as the C11 normal alkane, which loads positively on PC1, and the C4-alkylbenzenes, which
load positively on PC2. Thus, the presence of the C9-C12 branched alkanes from the matrix
cause the extracts to be positioned more positively on both PCs.
84
 

PC2 (14.6%)

2E6

-3E6

3E6

-2E6

PC1 (73.1%)
Figure 3.8: Scores plot of first principle component (PC1) versus the second principle
component (PC2) based on the total ion chromatograms for the six ignitable liquid standards and
unburned carpet projections. Each standard is denoted as follows: Neat gasoline ( ), 10%
evaporated gasoline ( ), 90% evaporated gasoline ( ), neat kerosene ( ), 10% evaporated
kerosene ( ), and 70% evaporated kerosene ( ).

85
 

In addition, the difference in the abundance of the internal standard affects the
positioning of the neat and 10% evaporated gasoline extracts. Despite normalization in data
pretreatment, there is still variation in the abundances of the internal standard peak in the
normalized chromatograms. When the TICs of the unburned carpet extracts containing these two
ignitable liquids are mean centered, the area of the internal standard peak in each chromatogram
is smaller than the area of the peak subtracted from the chromatogram (i.e., the average area of
the internal standard peak in the 90 liquid standard chromatograms). Thus, the corresponding
peak area in the mean-centered data is negative, and when it is multiplied by the negative
eigenvector of PC1 at this retention time, the loading for this peak is positive and contributes to
the positive positioning of the carpet extracts compared to the corresponding standards on this
PC.
Unburned carpet samples spiked with 90% evaporated gasoline are in similar positions to
the corresponding standard. Matrix interferences were present in these samples; however, the
interferences coelute with the C4-alkylbenzenes already in high concentration in 90% evaporated
gasoline. Thus, the contribution of the matrix interferences was negligible.
The extracts containing neat and 10% kerosene are positioned more positively on PC2
relative to the liquid standards, due to the addition of the C9-C12 branched alkanes from the
matrix. Because the interferences elute at the same retention times as the C4-alkylbenzenes,
which load positively on this PC, the extracts are positioned more positively on PC2.
Additionally, because the C4-alkylbenzenes load negatively on PC1, the addition of the C9-C12
branched alkanes from the matrix causes these extracts to be positioned less positively on PC1
than the corresponding kerosene standards.
86
 

Unburned carpet samples that were spiked with 70% evaporated kerosene are positioned
similarly on PC1 compared to the corresponding standard. The addition of the matrix
interferences had minimal effect because of the high concentration of the C13-C16 normal
alkanes in the 70% evaporated kerosene. The normal alkanes in the 70% evaporated kerosene
are approximately one order of magnitude greater in abundance than the interferences; hence, the
interferences have minimal effect on the positioning of the extracts in the scores plot.
The spread observed in PC2 for the replicates of the 70% evaporated kerosene extracts is
due to a misalignment of the internal standard peak in the TICs. When the TICs of the carpet
extracts containing the 70% evaporated kerosene are mean centered, essentially the area of the
internal standard peak is subtracted from the wrong retention time range in the chromatogram
due to this misalignment (9.24-9.28 minutes in the standard versus 9.28-9.32 minutes in the
sample extract). Thus, a negative peak is created in the mean-centered chromatogram, and when
it is multiplied by the negative eigenvector of PC2 at this retention time, the loading for this peak
is positive. Therefore, although the internal standard loads negatively in the loadings plot for
PC2, the peak has a positive contribution to the score on this PC. Additionally, though the
contribution is comparatively small, some of the C9-C12 branched alkanes from the matrix elute
at the same retention times as the C4-alkylbenzenes, which load positively on PC2. Thus, the
combination of these two effects causes the 70% evaporated kerosene extracts to be positioned
more positively on PC2 than the corresponding standard.

3.3.8 Assessment of Association and Discrimination of Samples in the Presence of Inherent
Matrix Interferences Using PPMC Coefficients

87
 

PPMC coefficients were calculated between replicates of the spiked unburned carpet
extracts and the corresponding liquid standards (Table 3.2). PPMC coefficients of replicates are
not equal to 1.00, due to variation in the extraction procedure. However, coefficients of
replicates are generally greater than 0.95, indicating acceptable precision in the extraction
procedure. Replicates of the carpet extracts with 70% evaporated kerosene have a mean PPMC
coefficient of 0.93, most likely due to the misalignment of the internal standard peak previously
discussed.
PCA and PPMCs are complimentary statistical procedures. That is, PCA determines the
variance in a data set while PPMC coefficients provide a measure of similarity. By using both of
these procedures, it is possible to statistically assess the positioning of extracts on the scores plot.
When the range of PPMC coefficients is calculated between each set of unburned extracts and
the corresponding standard, all of the gasoline extracts show strong correlations to the
corresponding ignitable liquid standard. However, extracts from the carpet spiked with the
various evaporation levels of kerosene demonstrate strong to moderate correlations to the
corresponding ignitable liquid standards. The weaker correlations for the kerosene extracts
compared to the gasoline extracts are reflective of the greater spread between the kerosene
extracts and the corresponding standards on the PCA scores plot (Figure 3.8).
The widest range of PPMC coefficients is observed for the 70% evaporated kerosene
extracts. This is expected due to the greater spread of replicates observed in the scores plot for
these samples compared to the other kerosene extracts caused by the misalignment of the internal
standard peak. PPMC coefficients were calculated between the 70% evaporated kerosene carpet
extracts and each ignitable liquid standard (Table 3.3). As expected, the 70% evaporated
kerosene extracts show moderate correlations to each of the kerosene standards, regardless of

88
 

Table 3.2: Pearson product moment correlation coefficients for replicates (n=15) of each
ignitable liquid extracted from unburned carpet and the range of coefficients for the extracts
correlated to the corresponding ignitable liquid standard.

Ignitable Liquid Standard

Mean PPMC Coefficient ±
Standard Deviation for
Replicates of Extracts
(n=225)

Neat Gasoline
10% Evaporated Gasoline
90% Evaporated Gasoline

0.992 ± 0.006
0.996 ± 0.004
0.997 ± 0.002

Range of PPMC
Coefficients Between
Spiked Carpet Samples
and Corresponding
Ignitable Liquid
Standard (n=225)
0.951 - 0.846
0.945 - 0.881
0.950 - 0.915

Neat Kerosene
10% Evaporated Kerosene
70% Evaporated Kerosene

0.996 ± 0.004
0.95 ± 0.07
0.93 ± 0.07

0.891 - 0.812
0.893 - 0.732
0.851 - 0.678

Table 3.3: Mean Pearson product moment correlation coefficients (n=225) for 70% evaporated
kerosene extracts correlated to each of the ignitable liquid standards.

Neat Gasoline
10% Evaporated Gasoline
90% Evaporated Gasoline

Mean PPMC Coefficient ±
Standard Deviation for Replicates
of Carpet Spiked with 70%
Evaporated Kerosene and Each
Ignitable Liquid Standard (n=225)
0.15 ± 0.07
0.13 ± 0.08
0.19 ± 0.07

Neat Kerosene
10% Evaporated Kerosene
70% Evaporated Kerosene

0.68 ± 0.08
0.72 ± 0.08
0.7 ± 0.2

Ignitable Liquid Standard

89
 

evaporation level. However, when correlated to standards for the three gasoline standards,
PPMC coefficients of less than 0.2 are observed, indicating weak correlations. Thus, while it is
not possible to associate the 70% evaporated kerosene extracts to the correct evaporation level, it
is possible to identify the residue as being kerosene rather than gasoline.

3.3.9 Identification of Burned Matrix Interferences and Selection of Burn Time
The TICs of the burned carpet show a variety of matrix interferences depending on the
burn time. The interferences observed in the burned carpet are products from the burning of the
adhesive, the yarn, and the backing material in the carpet. As previously discussed, the TIC of
the unburned carpet was dominated by the C9-C12 branched alkanes, though a non-reproducible
peak for 2-ethyl-1-hexanol was observed (Figure 3.9A). Burning for 10 seconds had no
significant effect on the interferences present in the TIC (Figure 3.9B). After burning for 20
seconds, styrene, benzaldehyde, and acetophenone were apparent in the TIC, as well as some
residual peaks from the C9-C12 branched alkanes inherent in the matrix (Figure 3.9C).
After burning for 30 seconds, peaks from the C9-C12 branched alkanes were no longer
observed in the TIC (Figure 3.9D). In addition, the abundance of the styrene peak quadrupled
between the 30-second and 60-second burn time (Figure 3.9E). Two peaks for toluene and
dimethylheptene were clearly visible in the TIC at 60 seconds, though the abundances of these
peaks were not reproducible. In addition, the abundance of benzaldehyde decreased while the
abundance of the C15 branched alkanes increased significantly, though the abundance of
acetophenone showed no appreciable change. After 120 seconds of burning, styrene was still the
most abundant interference observed (Figure 3.9F). The abundances of benzaldehyde and
90
 

1.3E5 A

C9-C12 branched
alkanes
2-ethyl-1-hexanol

0
1.3E5 B

C9-C12 branched
alkanes

0
C9-C12 branched
alkanes

Abundance

1.3E5 C

Benzaldehyde
Acetophenone
Styrene

0

Retention Time (min)

0

16

1.3E5 D
Benzaldehyde
Styrene
Acetophenone
0
1.3E5 E

Styrene
Dimethylheptene
Toluene

0
1.3E5 F

0

0

C15 branched
Acetophenone alkanes
Benzaldehyde

Styrene
C branched
Bicyclopentylone 15
alkanes
Dimethylheptene
Benzaldehyde
Toluene
Acetophenone

Retention Time (min)

16

Figure 3.9: Total ion chromatograms of extracts of carpet burned for A) 0 seconds, B) 10
seconds, C) 20 seconds, D) 30 seconds, E) 60 seconds, and F) 120 seconds. Major interference
compounds are identified.
91
 

acetophenone were greatly reduced compared to the 60-second burn time, while the abundance
of the C15 branched alkanes remained virtually unchanged (Figure 3.9F). A significant peak was
observed for bicyclopentylone at this burn time; however, this peak was not reproducible.
Compounds that were not reproducibly observed in the TICs were not considered when
determining the burn time. A burn time of 60 seconds was chosen for future studies based on the
number of different interferences observed in the TIC and the relative abundances of those
interferences. While the burn time of 120 seconds generated more abundant peaks for styrene
and the C15 branched alkanes, the loss of benzaldehyde and acetophenone meant that fewer
interferences would be added to sample chromatograms.
The contribution of the matrix interferences from the burned carpet can be observed in
the chromatograms of the extracts of burned carpet that was spiked with the ignitable liquids. As
expected, the extracts contain the characteristic compounds from the ignitable liquids, such as
toluene and the alkylbenzenes for gasoline and the C10-C16 normal alkanes for kerosene.
However, the matrix interferences have visually altered the chromatograms. For example,
styrene coelutes with o-xylene, which is found in gasoline (Figure 3.10A). The addition of
styrene from the burned carpet changes the visual pattern of the C2-alkylbenzenes, thus
complicating identification of the presence of gasoline in the extract. In contrast, the C15
branched alkanes from the burned carpet elute at similar retention times to the compounds found
in kerosene but at such a low abundance that the visual appearance of the chromatogram is only
minimally altered (Figure 3.10B).

92
 

Abundance

1.0E6 A

Styrene

o-xylene

0
4.5

Retention Time (min)

6.5

3.5E5 B

Abundance

C13

C15 branched alkanes

0
12

Retention Time (min)

13.5

Figure 3.10: A) Portion of total ion chromatograms of burned carpet spiked with neat gasoline
(─) versus the neat gasoline standard (─ • ─), showing the contribution of the styrene peak from
the burned carpet to the C2-alkylbenzenes in gasoline. B) Portion of total ion chromatograms of
burned carpet spiked with neat kerosene (─) versus the neat kerosene standard (─ • ─), showing
the contribution of the C15 branched alkane peaks from the burned carpet after the C13 normal
alkane peak in kerosene.
93
 

3.3.10 Association and Discrimination of Samples in the Presence of Burned Matrix
Interferences
To assess the effect of interferences introduced through burning of the matrix on the
association of extracts to the corresponding standard, the ignitable liquid standards were spiked
onto previously burned carpet samples, extracted, and analyzed. Scores calculated for the
extracts were then projected onto the PCA scores plot generated for the liquid standards.
Because some of the compounds from the burned carpet, specifically styrene and
benzaldehyde, elute at similar retention times to compounds in gasoline but not kerosene, the
addition of the matrix interferences was expected to affect the positioning of the kerosene
extracts more than the gasoline extracts. The veracity of this expectation is demonstrated in the
scores plot, in that the gasoline extracts generally overlap their corresponding standards, while
the kerosene extracts are further spread from the corresponding standards (Figure 3.11).
However, while the prediction was correct, the reasoning was not.
Samples of neat and 10% evaporated gasoline extracted from the burned carpet are
positioned more positively on PC1 compared to the corresponding liquid standards. This change
in positioning could be attributed to the addition of the C15 branched alkanes from the burned
carpet that elute between 13 and 14 minutes in the chromatogram. These branched alkanes elute
at a similar retention time to the C13 normal alkane present in kerosene. Because the C13 normal
alkane loads positively on PC1, the sample extracts would be positioned more positively than the
corresponding standards. However, the mean-centered data show that the C15 branched alkanes
do not make a significant contribution to the chromatogram, and therefore do not appreciably
affect the scores or the positioning of the extracts (Figure 3.12).

94
 

PC2 (14.6%)

2E6

-3E6

3E6

-2E6

PC1 (73.1%)
Figure 3.11: Scores plot of first principle component (PC1) versus the second principle
component (PC2) based on the total ion chromatograms for the six ignitable liquid standards and
projected scores for burned carpet spiked with each ignitable liquid. Each standard is denoted as
follows: Neat gasoline ( ), 10% evaporated gasoline ( ), 90% evaporated gasoline ( ), neat
kerosene ( ), 10% evaporated kerosene ( ), and 70% evaporated kerosene ( ).

95
 

Abudance

1.0E6
1.0E6 A

00

0
0

Retention Time (min)

16
16

Abundance

3.0E6
3.0E6 B

-3.0E6
-3.0E6

0

16
16

Retention Time (min)

Figure 3.12: A) Total ion chromatogram (TIC) of burned carpet spiked with neat gasoline,
showing the addition of the C15 branched alkanes from the burned carpet, and B) the same TIC
after mean centering, showing no corresponding peaks.

96
 

On the other hand, the contribution of styrene and benzaldehyde to the chromatograms of
gasoline extracted from the burned carpet should position the extracts more negatively on PC1
because these compounds elute at retention times similar to compounds present in gasoline, all of
which load negatively on this PC. However, because of the slight difference in retention times,
styrene and benzaldehyde do not contribute to the scores of the extracts. For example, styrene
from burned carpet elutes at 5.679 minutes and o-xylene from gasoline elutes at 5.707 minutes
(Figure 3.10A). Because only the ignitable liquid standards were used in generating the scores
plot and loadings plot, styrene would have to elute at the same retention time as p-xylene to
contribute to an extract’s score on PC1 and PC2. Styrene elutes at a different time, so the
addition of this peak to the chromatogram has a minimal effect on the score of the extract
projected onto the scores plot.
Because the interferences from the burned matrix have minimal effects on the positioning
of the extracts on the scores plots, the difference in the positioning between the burned carpet
extracts containing neat and 10% evaporated gasoline versus their respective standards is
explained simply by a difference in the abundances of the C2- and C3-alkylbenzenes in each
group of samples. The TICs of these burned carpet extracts have less abundant peaks of the
characteristic alkylbenzenes found in gasoline than do the TICs of the corresponding standards.
Therefore, when the data are mean centered, the peaks from the burned carpet extracts are
smaller than the peaks in the standard. When the smaller peaks are multiplied by the eigenvector
of PC1 for the alkylbenzenes, the loadings make a smaller negative contribution to the scores of
these extracts. Therefore, the extracts are positioned less negatively on PC1 than the
corresponding standards (Figure 3.13).

97
 

Abundance

1.0E6
1.0E6 A

00
6.8
6.8

Retention Time

7.8
7.8

Abundance

5.0E5 B
5.0E5

0
0
6.8
6.8

-5.0E5
-5.0E5

Abundance

5.0E4
5.0E4

0
0

7.8
7.8

Retention Time (min)

C

7.8
7.8

6.5
6.8

-5.0E4
-5.0E4
Retention Time (min)

Figure 3.13: A) Portion of the total ion chromatogram of the neat gasoline standard (─) and an
extract of neat gasoline spiked onto burned carpet (─ ─), demonstrating the difference in
abundances of the C3-alkylbenzenes, and the effect of this difference on the B) mean-centered
data, and C) loadings on PC1, which results in the spread observed in the scores plot.
98
 

A similar argument can be made to explain why the burned carpet extracts containing
90% evaporated gasoline are positioned on top of the standard 90% evaporated gasoline extracts.
It has already been discussed that the matrix interferences themselves do not affect the
positioning of the gasoline extracts; rather, it is the abundance of the compounds that affects the
positioning. In contrast, abundances of the C3- and C4-alkylbenzenes in 90% evaporated
gasoline are similar in the TICs of the extracts and standards (Figure 3.14). Because there is no
significant difference in the abundances of the compounds found in the TICs of each group of
extracts, the calculated scores for all of the extracts will be very similar, as seen by the close
positioning in the scores plot (Figure 3.11).
Extracts from the burned carpet containing neat, 10%, or 70% evaporated kerosene are
generally positioned more negatively on PC1 and more positively on PC2 than the corresponding
kerosene standards. The positioning of these extracts is due to differences in abundance of the
C13-C15 normal alkanes between the liquid standards and extracts. Despite normalization, there
is still variation in the abundances of these alkanes in the normalized chromatograms (Figure
3.15). When the TICs of the burned carpet extracts are mean centered, the area of the alkane
peaks in the chromatogram of the burned carpet extract are smaller than the average area of the
peaks from the ignitable liquid standards subtracted from the chromatogram. Thus, the peak area
in the mean-centered data are negative, and when it is multiplied by the positive value of the
eigenvector of PC1 at this retention time, the contribution of the peak to the extract’s score is
negative. Likewise, when the mean-centered data are multiplied by the negative value of the
eigenvector of PC2 at this retention time, the loading for the peak is positive. Thus, the scores of
the burned carpet extracts containing any evaporation level of kerosene are less positive on PC1
and more positive on PC2 than the corresponding standards.
99
 

Abundance

1E6 A
6

0
0

6.8
6.8

Retention Time (min)

7.8
7.8

Abundance

7E5 B

0

6.5
6.8

8.0
7.8

-7E5
Retention Time (min)

Abundance

-1E5 C

0
6.5
6.8

8.0
7.8

-1E5

5

Retention Time (min)

Figure 3.14: Demonstration of the difference in abundances of the C3-alkylbenzenes between
the 90% evaporated gasoline standard (─) and an extract of 90% evaporated gasoline spiked onto
the burned carpet (─ ─) in the A) total ion chromatogram, B) mean-centered data, and C)
loadings on the first principle component, which results in the close positioning of the spiked
burned carpet extracts and the standards in the scores plot.
100
 

Abundance

3E6 A

0
12.0
12

Retention Time (min)

15.5
15.5

5

Abundance

4E5 B

0

12
12.0

-4E5

15.5
15.5

Retention Time (min)

Abundance

3E4 C

0

12
12.0

16
15.5

-3E4
Retention Time (min)

Figure 3.15: Demonstration of the difference in abundances of the C12-C15 normal alkanes
between the neat kerosene standard (─) and an extract of neat kerosene spiked onto the burned
carpet (─ ─) in the A) total ion chromatogram, B) mean-centered data, and C) loadings on the
second principle component, which results in the standard being positioned more negatively in
the scores plot than the extract.
101
 

The spread among the five extracts of the burned carpet containing 10% evaporated
kerosene (highlighted in Figure 3.16A) is due in large part to a significant difference in the
abundance of the internal standard peak (Figure 3.16B). Replicates of one of the extracts
(labeled 1 in Figure 3.16A) show an abundance for the internal standard approximately 40% less
than that observed in replicates of the other four extracts (Figure 3.16B). Because the internal
standard peak loads negatively on both PC1 and PC2, the smaller area of the peak means that the
scores of this extract will be more positive on PC1 and PC2 than the other extracts, as observed
in Figure 3.16A.
Also contributing to the spread among the extracts is the presence of the styrene
interference from the burned carpet. Due to the lack of o-xylene in the kerosene-containing
samples, the styrene peak in these extracts was aligned to the o-xylene peak in the consensus
target during alignment procedures, despite the difference in retention time discussed previously.
Thus, while styrene did not contribute to the scores of the gasoline-containing extracts, it does
contribute to the calculated scores of the extracts of burned carpet spiked with kerosene. This is
an inherent limitation of the alignment procedure.
Since the abundance of styrene will be different for each extract due to variability in the
burning process, some spread is to be expected. Replicates of one of the extracts (labeled 2 in
Figure 3.16A) have the highest abundance for styrene among the 10% evaporated kerosenespiked samples (Figure 3.16C). Because of this high abundance, when the data are mean
centered, this peak still has a positive area. When multiplied by the eigenvector for PC1, the
styrene peak contributes negatively to the score of the extract because p-xylene loads negatively
on this PC. Thus, this extract is positioned more negatively on PC1 than the other extracts. The
same argument holds for PC2, but the p-xylene peak has less weight in this PC, thus the negative

102
 

5E6

PC2 (14.6%)

1

A

2
0

2E6

-5E5

PC1 (73.1%)

5
6E5 BB

Abundance

Abundance

6
1E6 CC

00

9.1
9.1

Retention Time (min)

Retention Time (min)

9.4
9.4

00

5.6
5.6

Retention Time (min)

Retention Time (min)

5.8
5.8

Figure 3.16: A) Enlarged view of the scores plot (Figure 3.11), focused on the replicate
extractions of burned carpet spiked with 10% evaporated kerosene. Each standard is denoted as
follows: Neat kerosene ( ) and 10% evaporated kerosene ( ). Extracts from the spiked burned
carpet are indicated by half fill. Total ion chromatograms showing the two peaks that contribute
most to the observed spread: B) the internal standard peak, showing one extract with a lower
abundance for all replicates, and C) the styrene peak from the burned carpet matrix, showing that
only four of the five extracts contain this peak.

103
 

contribution is minimal.
In addition to a smaller peak for the internal standard previously discussed, the TIC of
extract 1 shows no styrene peak (Figure 3.16C). When the replicate TICs of this extract are
mean centered, the peak area at this retention time is negative. Because the styrene from the
burned carpet is aligned to p-xylene from gasoline, which loads negatively on both PC1 and
PC2, extract 1 is positioned more positively on both PCs than the other extracts.
Spread in the replicates of the burned carpet extracts containing 70% evaporated kerosene
can be explained in a similar manner, though differences in the abundances of C13-C15 normal
alkanes also contribute to the spread, as previously discussed.

3.3.11 Assessment of Association and Discrimination of Samples in the Presence of Burned
Matrix Interferences Using PPMC Coefficients
PPMC coefficients were calculated between replicates of the burned carpet extracts and
the corresponding standards (Table 3.4). Coefficients of replicates are greater than 0.92,
indicating acceptable precision in the extraction procedure despite variability of the burning
procedure. Replicates of the carpet extracts with 10% evaporated kerosene have a mean PPMC
coefficient of 0.92, most likely due to the lack of the styrene peak from the burned carpet in the
TICs of one replicate, as previously discussed.
When the range of PPMC coefficients are calculated between each set of sample
replicates and the corresponding standard, all of the gasoline extracts show strong correlations to
the corresponding ignitable liquid standard. The highest PPMC coefficients and smallest range
of coefficients are observed for burned carpet samples spiked with 90% evaporated gasoline.
This is expected due to the high similarity between the extracts and the standard. Extracts from
104
 

Table 3.4: Pearson product moment correlation coefficients for replicate (n=15) extractions of
each ignitable spiked onto burned carpet and the range of coefficients for the extracts correlated
to the corresponding ignitable liquid standard.

Ignitable Liquid Standard

Mean PPMC Coefficient
± Standard Deviation for
Replicates of Extracts
(n=225)

Neat Gasoline
10% Evaporated Gasoline
90% Evaporated Gasoline

0.990 ± 0.012
0.993 ± 0.007
0.992 ± 0.005

Range of PPMC Coefficients
Between Spiked Burned
Carpet Samples and
Corresponding Ignitable
Liquid Standard (n=225)
0.993 - 0.927
0.978 - 0.929
0.994 - 0.968

Neat Kerosene
10% Evaporated Kerosene
70% Evaporated Kerosene

0.97 ± 0.04
0.92 ± 0.10
0.97 ± 0.04

0.923 - 0.720
0.896 - 0.668
0.781 - 0.546

105
 

the carpet spiked with the various evaporation levels of kerosene demonstrate strong to
moderate correlations to the corresponding ignitable liquid standards. The widest range of
PPMC coefficients is observed for the 70% evaporated kerosene extracts due to variability of the
burned matrices present in the TICs of the extracts as well as the varying abundances of the C13C15 normal alkanes discussed previously. The wide range of PPMC coefficients is reflective of
the spread of the extracts in the scores plot.
Close positioning on the scores plot (Figure 3.11) indicates that extracts of neat and 10%
evaporated gasoline contain similar abundances of the compounds in gasoline. This conclusion
is reinforced using PPMC coefficients. Neat gasoline extracts have a mean PPMC coefficient of
0.970 ± 0.015 when compared to the corresponding standard (n=225) and a mean coefficient of
0.950 ± 0.019 when compared to the 10% evaporated gasoline standard (n=225), indicating a
strong correlation to both standards. Correlation of the 10% evaporated gasoline extracts to each
standard yields similar results. Thus, based on both the visual analysis of the PCA scores plot
and the statistical analysis using PPMC coefficients, it is not possible to associate the extracts to
one evaporation level over the other.
PPMC coefficients were calculated between carpet extracts containing kerosene at all
three evaporation levels and each of the kerosene liquid standards to assess whether the extracts
could be assigned to a particular evaporation level (Table 3.5). The neat and 10% evaporated
kerosene extracts show moderate correlations to both their respective standard as well as the
standard of the other evaporation level, ranging from 0.82 to 0.85. However, when compared to
the 70% evaporated kerosene standard, weak correlations of 0.35 and 0.37 were observed for the
neat kerosene and 10% evaporated kerosene extracts, respectively. Thus, while the extracts
could not be identified as containing neat or 10% evaporated kerosene, they could be
106
 

Table 3.5: Comparison of mean Pearson product moment correlation coefficients (± standard
deviation) calculated between extracts of burned carpet spiked with kerosene at each evaporation
level and the corresponding ignitable liquid standards.
Ignitable Liquid Standard
Neat Kerosene
Extract of
Ignitable
Liquid
Spiked onto
Burned
Carpet

10% Evaporated
Kerosene

70% Evaporated
Kerosene

Neat Kerosene

0.85 ± 0.06

0.83 ± 0.06

0.35 ± 0.03

10% Evaporated
Kerosene

0.83 ± 0.08

0.82 ± 0.08

0.37 ± 0.05

70% Evaporated
Kerosene

0.65 ± 0.06

0.66 ± 0.06

0.67 ± 0.07

107
 

distinguished as not containing 70% evaporated kerosene. On the other hand, the extracts
containing 70% evaporated kerosene showed similar PPMC coefficients indicating moderate
correlations to standards at all three evaporation levels. Therefore, extracts containing 70%
evaporated kerosene could be identified as containing kerosene, but it would not be possible to
specifically state which evaporation level of kerosene was present in the sample.

3.3.12 Association and Discrimination of Samples in the Presence of Thermal Degradation
To assess the combined effect of evaporation, matrix interferences, and thermal
degradation on the association of sample extracts to the corresponding standards, the ignitable
liquid standards were spiked onto carpet matrix that was subsequently burned using the burn
time of 60 seconds. The burned matrices were extracted and analyzed. Scores were calculated
for the extracts, and then projected onto the scores plot generated using the liquid standards.
Overall, when the scores of the extracts were projected onto the standard scores plot, a
close visual similarity is generally observed between the kerosene extracts and the corresponding
standards (Figure 3.17). However, all of the gasoline extracts are now more closely associated to
the 90% evaporated gasoline standard than to the other gasoline standards.
During the burning process, the more volatile components of gasoline are evaporated,
including toluene and the C2- and C3-alkylbenzenes. This loss mimics the evaporative loss of
these compounds, observed in the evaporated standards (Figure 3.3). Therefore, the extracts of
samples containing either neat or 10% evaporated gasoline are more chemically similar to the
90% evaporated gasoline standard and are positioned closer to this standard than either the neat
gasoline or the 10% evaporated gasoline standard. Spread in the scores plot for these samples is
attributable to variable loss of the volatile compounds during the burning process and not the
108
 

2E6
1

PC2 (14.6%)

2

-3E6

3E6
3

4
-2E6

PC1 (73.1%)
Figure 3.17: Scores plot of first principle component (PC1) versus the second principle
component (PC2) based on the total ion chromatograms for the six ignitable liquid standards and
thermal degradation projections. Each standard is denoted as follows: Neat gasoline ( ), 10%
evaporated gasoline ( ), 90% evaporated gasoline ( ), neat kerosene ( ), 10% evaporated
kerosene ( ), and 70% evaporated kerosene ( ), burned carpet with no ignitable liquid added
( ). Extracts from the thermal degradation samples are indicated by half fill.

109
 

addition of compounds from the burned carpet itself, as discussed previously.
It should be noted that one of the extracts containing neat gasoline (labeled 1 in Figure
3.17) is positioned more positively on PC2 than the other extracts due to less abundant internal
standard peak as well as a more abundant 1,3,5-trimethylbenzene peak observed in the TIC of
this sample. When the TICs of this replicate are mean centered, the peak area for the internal
standard becomes negative. Therefore, when multiplied by the eigenvector for PC2, in which the
internal standard loads negatively, the peak area becomes positive. Thus the score on PC2 of this
replicate will be more positive than the other replicates that contain larger peaks for the internal
standard in the TICs. The 1,3,5-trimethylbenzene at a retention time of 7.63 minutes is one of
the C3-alkylbenzenes that load positively on PC2. A comparatively high abundance of this peak
in the replicate also contributes to the extract’s more positive positioning on PC2. However, on
PC1, the effects of a less abundant peak for the nitrobenzene internal standard peak and a more
abundant 1,3,5-trimethylbenzene peak cancel on PC1 because the internal standard loads
negatively while the C3-alkylbenzenes load positively on this PC. Therefore, extract 1 is
positioned comparably to other neat gasoline extracts on PC1. Similar arguments can be made
regarding the positioning and spread of the 10% evaporated gasoline sample extracts.
Extracts containing residues of 90% evaporated gasoline are positioned more positively
on PC1 than the corresponding standard due to loss of the C2-, C3-, and C4-alkylbenzenes during
the burning process. All of the compounds load negatively on PC1; therefore their loss will
result in a more positive score for the extracts on this PC. Likewise, because the C3- and C4alkylbenzenes load positively on PC2, the loss of these compounds reduces the positive
contributions of the compounds to the score on PC2. Hence, the extracts will be positioned less

110
 

positively on this PC. This logic also explains why replicates of one extract of the 90%
evaporated gasoline burned sample extracts (labeled 2 in Figure 3.17) are positioned more
positively on PC1 and more negatively on PC2 than the other extracts. In this extract, a greater
loss of the C2-, C3-, and C4-alkylbenzenes was observed, thus the extract is spread from the
cluster of the other extracts.
Extracts containing 10% evaporated kerosene are generally positioned on top of the
corresponding standard, indicating minimal loss of the normal alkanes from the ignitable liquid
during the burning process. This is expected because the heavier chain alkanes are not very
volatile and will be less affected by the burning process. However, one extract (labeled 3 in
Figure 3.17) is spread from the rest of the extracts containing 10% evaporated kerosene. This
spread is due to a comparatively high abundance of the internal standard in the TIC of this
particular extract. Because the internal standard loads negatively on both PCs, this extract is
positioned more negatively on both PCs than the other replicates.
When carpet spiked with neat kerosene is burned, the loss of the C10 normal alkane
mimics the evaporative loss observed in the ignitable liquid standards. Therefore, extracts of the
burned carpet containing neat kerosene are expected to be more closely associated with the 10%
evaporated kerosene liquid standard versus the neat kerosene standard. For some extracts, this is
the case. However, three of the five extracts are positioned slightly less positively on PC1 and
more positively on PC2 than the other replicates as well as the neat kerosene standard. This is
due to a significant reduction in the abundance of the internal standard peak as well as the C10C16 normal alkanes observed in the TICs of these replicates.

111
 

Because the C10-C16 normal alkane peaks load positively on PC1, a reduction of the
abundances of these peaks will position the neat kerosene extracts less positively on this PC.
However, the reduction in the abundance of the internal standard peak will have the opposite
effect, as previously discussed. Therefore, the two effects partially offset each other and the
result is that the extracts are only slightly less positively positioned on PC1. However, the
internal standard and the C13-C16 normal alkanes load negatively on PC2. Thus, a reduction in
the abundances of these peaks results in a score that is less negative on PC2.
The positive positioning of the neat kerosene extracts on this PC is further explained by
the addition of styrene and benzaldehyde from the burned carpet matrix. During alignment,
these matrix interferences were aligned to o-xylene and m-ethyltoluene, respectively. O-xylene
loads negatively on PC2, while m-ethyltoluene loads positively. However, o-xylene makes a
comparatively small contribution to the score versus that of m-ethyltoluene. Therefore, the
positive contribution from the addition of benzaldehyde from the burned carpet positions the
extract more positively on PC2.
For extracts containing 70% evaporated kerosene, loss of the C12-C16 normal alkane
compounds during burning is not expected due to the relatively low volatility of these
compounds. The spread in the scores plot among the extracts is due to significant differences in
the abundances in the peaks for the nitrobenzene internal standard and the normal alkanes
observed in the TICs of the extracts, as discussed above.
The one exception to this generalization regarding is the 70% evaporated kerosenecontaining extract positioned negatively on both PC1 and PC2 (labeled 4 in Figure 3.17).
Replicates of this extract are positioned closely to an extract of 90% evaporated gasoline, but

112
 

neither of the extracts is visually associated with any of the ignitable liquid standards. When the
scores of burned carpet with no ignitable liquid added are projected onto the scores plot, they are
positioned closely to these two thermal degradation extracts (Figure 3.17).
Visual examination of the TICs of the two thermal degradation extracts shows that
neither contains the expected compounds from an ignitable liquid, such as the C3-alkylbenzenes
from the 90% evaporated gasoline and the normal alkanes of the 70% evaporated kerosene
(Figure 3.18A and B, respectively). In fact, when the chromatograms of the two extracts are
compared to a chromatogram of burned carpet (Figure 3.18C), all of the peaks observed in the
TICs of the extracts are due to the compounds from the burned carpet. Thus, no ignitable liquid
residue (ILR) was identified for these samples.

3.3.13 Assessment of Association and Discrimination of Samples in the Presence of Thermal
Degradation Using PPMC Coefficients
PPMC coefficients were calculated between replicates of the thermal degradation carpet
extracts and the corresponding standards (Table 3.6). Coefficients of replicates are generally
greater than 0.90, indicating acceptable precision in the extraction procedure despite variability
of the burning procedure and misalignments in the data set, such as the alignment of the styrene
peak to o-xylene discussed previously. PPMC coefficients for extracts containing 90%
evaporated gasoline and 70% evaporated kerosene are lower than the other extracts (0.7 ± 0.3
and 0.7 ± 0.4, respectively) due to the two extracts for which no ILR was identified. When these
extracts are excluded and the PPMC coefficients recalculated, the 90% evaporated gasoline
extracts have a mean PPMC coefficient of 0.91 ± 0.09, while the 70% evaporated kerosene
extracts have a mean coefficient of 0.95 ± 0.05.
113
 

4E6

A

0

Abundance

4E6

B

0
4E6

IS

C

Acetophenone
Styrene
Benzaldehyde
0

0

C15 branched
alkanes
17

Retention Time (min)

Figure 3.18: Total ion chromatograms of anomalous extracts (labeled 4 in Figure 3.17) of carpet
spiked with A) 90% evaporated gasoline, B) 70% evaporated kerosene, and C) no ignitable
liquid.

114
 

Table 3.6: Pearson product moment correlation coefficients for replicates (n=15) of each
ignitable liquid thermal degradation extract and the range of coefficients for the extracts
correlated to the corresponding ignitable liquid standard. (* - indicates coefficients calculated
excluding replicates that contained no ILR)
Range of PPMC Coefficients
Between Thermal Degradation
Mean PPMC Coefficient ±
Samples and Corresponding
Ignitable Liquid Standard
Standard Deviation for
Ignitable Liquid Standard
Replicates of Extracts (n=225)
(n=225)
Neat Gasoline
0.94 ± 0.05
0.776 - 0.569
10% Evaporated Gasoline
0.96 ± 0.03
0.715 - 0.518
90% Evaporated Gasoline
0.7 ± 0.3
0.774 - 0.340
90% Evaporated Gasoline*
0.91 ± 0.09
0.774 - 0.518
Neat Kerosene
10% Evaporated Kerosene
70% Evaporated Kerosene
70% Evaporated Kerosene*

0.90 ± 0.11
0.92 ± 0.08
0.7 ± 0.4
0.95 ± 0.05

115
 

0.944 - 0.701
0.889 - 0.719
0.980 - 0.283
0.980 - 0.843

For the previous study in which the liquids were spiked onto already burned carpet, the
gasoline extracts all showed strong correlations to the standards, while the kerosene extracts
showed moderate correlation (Table 3.4). Here, when thermal degradation is considered, the
opposite trend is observed for the PPMC coefficients. The kerosene extracts show strong to
moderate correlation to their respective standards. On the other hand, the gasoline extracts are
moderately correlated to their respective standards. This is largely due to the loss of volatile
compounds, such as toluene and the C2-alkylbenzenes, during the burning process. However,
extracts containing neat gasoline or 10% evaporated gasoline show strong correlations when
compared to the 90% evaporated gasoline standard and only moderate correlations to the neat
kerosene standard (Table 3.7). Thus, the extracts can be correctly identified as containing
gasoline, even though they cannot be assigned to the correct evaporation level. The wider ranges
of coefficients for this study compared to the previous study (Section 3.2.11) are to be expected
and can be explained due to the variability in the burning of both the ignitable liquid and carpet
matrix.
The extracts from the 90% evaporated gasoline samples are visually positioned farthest
from their class of liquid standard (i.e., gasoline versus kerosene) in the scores plot (Figure 3.19).
In fact, these extracts appear to be positioned equidistantly between two standards: 90%
evaporated gasoline and neat kerosene. Replicates of one extract (labeled 1 in Figure 3.19)
appear to be spread from the other extracts (labeled 2 in Figure 3.19). This is due to a greater
loss C2-, C3-, and C4-alkylbenzenes in this extract, as previously discussed. When the replicates
in cluster 1 are correlated to each of the two standards, both PPMC coefficients indicate a
moderate correlation (Table 3.8). Thus, this extract cannot be identified as containing one liquid
over the other. However, when the extracts in cluster 2 are analyzed, a moderate correlation to
116
 

Table 3.7: Comparison of Pearson product moment correlation coefficients for extracts (n=15)
containing gasoline ILRs correlated the corresponding ignitable liquid standard as well as the
90% evaporated gasoline and neat kerosene standards. (* - indicates coefficients calculated
excluding replicates that contained no ILR)

ILR in Extract

Mean PPMC Coefficient ± Standard Deviation for
Replicates of Extracts Correlated to:
Corresponding
90% Evaporated
Neat Kerosene
Standard
Gasoline Standard
Standard

Neat Gasoline

0.64 ± 0.05

0.85 ± 0.05

0.55 ± 0.06

10% Evaporated
Gasoline

0.62 ± 0.05

0.91 ± 0.04

0.51 ± 0.06

90% Evaporated
Gasoline*

0.67 ± 0.09

0.67 ± 0.09

0.51 ± 0.02

117
 

2E6
2.0E6

PC2 (14.6%)

2

1

-6.0E5
-6E5

0

2.0E6
2E6

-3.2E5
-3.2E5

PC1 (73.1%)
Figure 3.19: Enlarged view of the scores plot (Figure 3.17), highlighting two clusters of extracts
containing 90% evaporated gasoline (1 and 2). Each standard is denoted as follows: Neat
gasoline ( ), 10% evaporated gasoline ( ), 90% evaporated gasoline ( ), neat kerosene ( ),
10% evaporated kerosene ( ), and 70% evaporated kerosene ( ). Extracts from the thermal
degradation samples are indicated by half fill.

118
 

the 90% gasoline standard is observed, as opposed to the weak correlation to the kerosene
standard. Therefore, these replicates can correctly be identified as containing a gasoline residue.
Additionally, when cluster 1 is correlated to cluster 2, a strong correlation is observed. Thus,
clusters 1 and 2 can be considered chemically similar.
A strong correlation is observed for both the neat kerosene extracts and the 10%
evaporated extracts and the corresponding standard (Table 3.9). However, the neat kerosene
extracts also show a strong correlation to the 10% evaporated kerosene standard and the 10%
evaporated extracts show a strong correlation to the neat kerosene standard. Thus, extracts
containing neat and 10% evaporated kerosene cannot be associated with one evaporation level
over the other. On the other hand, the extracts containing 70% evaporated kerosene show a
strong correlation to the corresponding standard, and only moderate correlation to the other two
kerosene standards. Thus, these samples could be associated with the correct evaporation level
of kerosene.

3.4

Conclusions from Carpet Matrix Study

Interferences, both those found both inherently in carpet and those generated through burning the
carpet, can alter the appearance of ILR chromatograms and can prevent the visual identification
of the ILRs. In this research, PCA was applied to the ignitable liquid standards only, so only
compounds found in these liquids contributed to the variance. Hence, the majority of interfering
compounds both from the unburned and burned carpet did not affect the positioning of the
projected ILR scores. Instead, the factors that most interfered with the visual association of
extracts and standards on the PCA scores plot were misalignments during the alignment

119
 

Table 3.8: Comparison of mean Pearson product moment correlation coefficients (± standard
deviation) for clusters 1 and 2 of 90% evaporated gasoline extracts correlated to each other and
the two standards.

Cluster 1

Cluster 2

Cluster 2

0.84 ± 0.08

-

90% Evaporated
Gasoline Standard

0.536 ± 0.017

0.73 ± 0.03

Neat Kerosene
Standard

0.539 ± 0.008

0.497 ± 0.017

Table 3.9: Comparison of Pearson product moment correlation coefficients for extracts (n=15)
containing kerosene ILRs correlated the corresponding ignitable liquid standard as well as the
10% evaporated and 70% evaporated kerosene standards. (* - indicates coefficients calculated
excluding replicates that contained no ILR)

ILR in Extract

Mean PPMC Coefficient ± Standard Deviation for
Replicates of Extracts Correlated to:
Neat Kerosene
10% Evaporated
70% Evaporated
Standard
Kerosene Standard Kerosene Standard

Neat Kerosene

0.81 ± 0.04

0.83 ± 0.04

0.69 ± 0.09

10% Evaporated
Kerosene

0.80 ± 0.08

0.82 ± 0.09

0.60 ± 0.09

70% Evaporated
Kerosene*

0.51 ± 0.03

0.54 ± 0.04

0.91 ± 0.05

120
 

procedure and differences in abundances not accounted for during normalization.
Nevertheless, using a combination of PCA and PPMC coefficients, the association of
ILRs to the corresponding standards could be demonstrated despite matrix interferences,
evaporation of the ignitable liquids, and thermal degradation. The extracted residues could be
identified as either gasoline or kerosene, but not necessarily the correct evaporation level of
those liquids. However, this is not a limitation in the forensic setting, as arson investigators look
to identify the type of ignitable liquid used, not the specific evaporation level.

121
 

References

122
 

References

1. Bertsch W. Volatiles from carpet: a source of frequent misinterpretations in arson analysis. J
Chromatogr A 1994; 674: 329-333.
2. American Society for Testing and Materials, ASTM E 1411-07. Annual Book of ASTM
Standards 14.02.

123
 

Chapter 4 - Using Multivariate Statistical Procedures to Identify Ignitable Liquid Residues
in the Presence of Matrix Interferences from Plastic

4.1 Introduction
Plastic is a very common material used in everyday items ranging from cups to children’s
toys. Thus, it is not unusual for plastics to be found at suspected arson scenes. In addition, many
ignitable liquids can be purchased or stored in plastic containers. Plastics, burned or unburned,
are known to contain compounds similar to those observed in some ignitable liquids and,
therefore, can complicate the analysis of fire debris. In this chapter, the objective method
previously developed in Chapter 3 for carpet will be applied to a plastic matrix.
As with the investigation of interferences from carpet, gasoline and kerosene were each
evaporated to two different levels of evaporation. The neat and evaporated standard liquids were
analyzed using principal components analysis (PCA) to generate a scores plot. Each of the
ignitable liquids was spiked onto unburned high density polyethylene (HDPE) to investigate the
effect of interferences inherent in the HDPE matrix on the association of the extracts to the
corresponding liquid standards. Then, the liquid standards were spiked onto burned HDPE and
analyzed in the same way to assess the effects of additional interferences introduced during the
burning process. The HDPE had been previously burned for an amount of time selected to
generate maximum interferences. Finally, each liquid standard was spiked onto unburned
HDPE. The plastic was then burned to simulate fire debris to investigate the effect of thermal
degradation on the association of the extracts to the corresponding liquid standards.
The total ion chromatograms (TICs) were compiled into three data sets and subjected to
data pretreatment procedures (background subtraction, smoothing, retention-time alignment, and

124
 

normalization). PCA and Pearson product moment correlation (PPMC) coefficients were then
applied to assess the association of the sample extracts to the corresponding ignitable liquid
standard despite the presence of matrix interferences, evaporation, and thermal degradation.

4.2 Materials and Methods

4.2.1 Ignitable Liquid Collection and Standard Preparation
Gasoline and kerosene previously collected were re-evaporated to generate a new set of
liquid standards. The collection and preparation of the neat and evaporated standards for
gasoline and kerosene were described in Chapter 3 Sections 3.2.1 and 3.2.2.

4.2.2 Inherent Matrix Interference Sample Preparation
High density polyethylene (HDPE) was obtained in the form of recycled industrial
2

solvent bottles. Plastic from the sides of the bottles was cut into 4 × 4 cm squares while the
bottom of the bottle was cut in half, and the top of the bottle (excluding the handle) was also cut
in half. A sample consisted of either eight squares cut from the side or half of the top or half of
the bottom of the bottle. This sample size was necessary to provide a sufficient abundance of
matrix peaks during GC-MS analysis. Because the bottles varied in shape, size, and thickness of
plastic, some variation between samples was unavoidable. However, this was minimized by
using the same type of HDPE sample for a set of extractions. For example, for unburned HDPE
spiked with neat gasoline, the liquid was spiked onto HDPE samples from the bottom of the
solvent bottle for all five samples.

125
 

To investigate the interferences inherent in HDPE, samples of unburned HDPE were each
placed into separate nylon bags. The samples were then spiked with 20 μL of the neat and
evaporated ignitable liquids (n=5 for each liquid) diluted 3:100 (v:v) in CH2Cl2. The samples
were extracted as described in Chapter 3 Section 3.2.3.

4.2.3 Burned Matrix Interference Sample Preparation
Burn times were firstly investigated to determine the time that generated the most
abundant matrix interferences. To do this, samples of HDPE were burned with a propane torch
(Benzomatic, Medina, NY) for 10 to 120 seconds. The samples were extinguished by
smothering, then placed in a nylon bag and extracted and analyzed according to the procedures
described in Chapter 3 Section 3.2.4.
Following investigation of the burn time, HDPE samples were burned for the selected
time and placed into separate nylon bags. A 20 µL aliquot of each diluted ignitable liquid
standard was spiked onto separate HDPE samples (n=5 for each liquid). The samples then were
extracted as previously described.

4.2.4 Thermal Degradation Sample Preparation
Separate samples of HDPE were spiked with each of the undiluted ignitable liquid
standards (n=5 for each liquid) then burned for 60 s to generate significant matrix interferences.
For samples containing gasoline, a spike volume of 350 μL was used, while a spike volume of
175 μL was used for kerosene-containing samples. The spike volumes were chosen such that the
ignitable liquid did not mask the interferences from the matrix in the chromatogram. In addition,

126
 

five samples of HDPE were burned for 60 seconds with no ignitable liquid added, to serve as a
control.

4.2.5 GC-MS Analysis
The experimental parameters for GC-MS analysis were the same as those described in
Chapter 3 Section 3.2.6.

4.2.6 Data Pretreatment and Analysis
The total ion chromatograms (TICs) generated for all standards and sample extracts were
background subtracted, smoothed, and retention-time aligned as one data set, but the data were
subsequently separated into two data sets for normalization. The first data set included TICs of
the ignitable liquid standards and the extracts of unburned HDPE samples spiked with the
ignitable liquids, while the second included the ignitable liquid standards, the spiked burned
HDPE extracts, and the thermal degradation extracts. The procedures for background
subtraction, smoothing, alignment, and normalization are described in Chapter 3 Section 3.2.7.
PCA was performed on each pretreated data set and PPMC coefficients were calculated on the
sample TICs, as described in Chapter 3 Section 3.2.8.

4.3 Results and Discussion

4.3.1 Selection of Retention-Time Alignment Parameters
Alignment parameters were selected based on analysis of the ignitable liquid standards
only, according to procedures described in Chapter 3 Section 3.3.1. However, visual assessment

127
 

of the aligned chromatograms showed that major misalignments were still present using all
combinations of alignment parameters. Misalignments were most likely due to the close spacing
of peaks in the sample chromatograms, particularly interference peaks from the burned HDPE
matrix (Figure 4.1A). Visual assessment of the unaligned chromatograms showed no significant
misalignments (Figure 4.1B). Therefore, all TICs were left unaligned for data analysis.

4.3.2 Selection of Normalization Method
Before normalization, the TICs showed differences in abundance among replicate
extractions for both the neat and evaporated ignitable liquid standards. Normalization to the
internal standard ensured that the maxima of the chromatograms were generally on the same
order of magnitude, while total area normalization reduced spread in abundance among
replicates (see Figure 3.2 in Chapter 3 for example TICs).
After internal standard normalization, several types of total area normalization were
investigated, such as normalizing by ignitable liquid type and normalizing the entire data set.
For each, the normalized data were subjected to PCA and the resultant scores plot was visually
assessed for clustering of extracts. Total area normalization of each set of replicate extractions
yielded the closest clustering of extracts and was, therefore, chosen as the normalization method.
Thus, following internal standard normalization, each set of replicate extractions was subjected
to total area normalization prior to data analysis.

4.3.3 Association and Discrimination of Neat and Evaporated Liquid Standards
A discussion of the compounds found in gasoline and kerosene can be found in Chapter 3
Section 3.3.3. Chromatograms showing the loss of characteristic compounds in the ignitable

128
 

A

Abundance

3.0E5

0
10.6

11.4

Retention Time (min)

11.4

B

Abundance

3.0E5

Retention Time (min)

0
10.6

Figure 4.1: Total ion chromatograms of the C12 matrix interferences in five extracts of burned
high density polyethylene A) after alignment to the consensus target using the COW algorithm
and B) without alignment.

129
 

liquids during evaporation can be seen in Figure 3.3 in Chapter 3.
The scores plot of the first principal component (PC1) and the second principal
component (PC2) accounts for 89.6% of the variance among the neat and evaporated ignitable
liquid standards (Figure 4.2). Standards are positioned differently than observed in the carpet
matrix study due to the fact that the standards were reanalyzed and a different normalization
procedure used for the plastic matrix study. This is an inherent limitation in PCA analysis, in
that the use of a new set of standards or different pretreatment procedures can drastically alter the
scores and loadings plots. However, the same general trends in positioning are observed, such as
the close positioning of neat and 10% evaporated gasoline samples.
Overall, the liquids are differentiated in PC1 and PC2. The positioning of the standards
can be explained using the loadings plots (Figure 4.3). Gasoline samples are positioned
negatively on PC1 because toluene and the C2-, C3-, and C4-alkylbenzenes all load negatively on
PC1. Neat and 10% evaporated gasoline samples are positioned negatively on PC2, due to the
dominance of toluene and the C2-alkylbenzenes in the TICs of these samples, which load
negatively on this PC. On the other hand, samples of 90% evaporated gasoline are positioned
positively on this PC. This is due to the evaporative loss of toluene and the C2-alkylbenzenes
that load negatively on PC2 and the concentration of the C3- and C4-alkylbenzenes that load
positively on PC2.
Kerosene samples are positioned positively on both PCs because the C10-C16 normal
alkanes load positively on both PCs. As the level of evaporation increases, the kerosene

130
 

PC2 (24.0%)

1.0E6

-2.0E6

2.0E6

-1.0E6

PC1 (65.6%)
Figure 4.2: Scores plot of the first principle component (PC1) versus the second principle
component (PC2) based on the total ion chromatograms for the two ignitable liquids at three
levels of evaporation, denoted as follows: Neat gasoline ( ), 10% evaporated gasoline ( ), 90%
evaporated gasoline ( ), neat kerosene ( ), 10% evaporated kerosene ( ), and 70% evaporated
kerosene ( ).

131
 

Loadings on PC1

0.3
0.3

A

C13 C14

0.0
0.0

C10
0

Toluene

a

C11 C12

C15
C16
21

c d

b

g
f
e

IS
C4-alkylbenzenes

C2-alkylbenzenes
-0.3
-0.3

Loadings on PC2

0.3

0.0

h
C3-alkylbenzenes

B

C3-alkylbenzenes
C4-alkylbenzenes
C13 C14
C11
h
C12
C15
IS
g
C16
ef

0

21

a

-0.3

c

b
Toluene C2-alkylbenzenes
Retention Time (min)

Figure 4.3: Loadings plots of A) the first principle component (PC1) and B) the second principle
component (PC2) based on the total ion chromatograms of the six ignitable liquid standards.
Major compounds are labeled: a) ethylbenzene, b) p-xylene, c) o-xylene, d) propylbenzene, e) methyltoluene, f) p-ethyltoluene, g) o-ethyltoluene, and h) 1,3,5-trimethylbenzene. IS indicates the
internal standard, nitrobenzene.
132
 

standards are positioned more positively on both PCs due to concentration of the C13-C16
normal alkanes.

4.3.4. Association and Discrimination of Neat and Evaporated Liquid Standards using PPMC
Coefficients
For the liquid standards, mean PPMC coefficients of replicates are not equal to the
theoretical value of 1, due to variation in the extraction procedure (Table 4.1 and 4.2). However,
all of the PPMC coefficients for replicates are greater than 0.99, indicating acceptable precision
in the extraction and analysis procedures.
As was seen with the ignitable liquid standards for the carpet matrix study, only gasoline
and kerosene at the highest evaporation levels could be differentiated from the other evaporation
levels of the respective ignitable liquid (Table 4.1 and 4.2). The neat and 10% evaporated
gasoline standards show a strong correlation to each other (0.985 ± 0.016), but only a moderate
correlation to the 90% evaporated standard (0.65 ± 0.02). A similar trend is observed for the
kerosene standards, where only the 70% evaporated kerosene can be discriminated from the
other kerosene standards.

4.3.5 Identification of Inherent Matrix Interferences
Polyethylene is manufactured from petroleum, and therefore would be expected to
contain the long chain hydrocarbons found in petroleum. Accordingly, the TICs of the unburned
HDPE, with no ignitable liquid present, show matrix interferences dominated by dodecene,
tetradecene, and hexadecene (Figure 4.4). Other interferences include decene, dodecane, and
tetradecene, though these are at a significantly lower abundance.
133
 

Table 4.1: Mean Pearson product moment correlation coefficients (± standard deviation) for
replicates (n=15) of gasoline standards.

10% Evaporated 90% Evaporated
Gasoline
Gasoline

Neat Gasoline
Neat Gasoline
10% Evaporated
Gasoline
90% Evaporated
Gasoline

0.997 ± 0.003
0.985 ± 0.016

0.997 ± 0.003

0.65 ± 0.02

0.68 ± 0.02

0.999 ± 0.001

Table 4.2: Mean Pearson product moment correlation coefficients (± standard deviation) for
replicates (n=15) of gasoline standards.

Neat Kerosene
Neat Kerosene
10% Evaporated
Kerosene
70% Evaporated
Kerosene

10% Evaporated 70% Evaporated
Kerosene
Kerosene

0.997 ± 0.003
0.971 ± 0.009

0.997 ± 0.002

0.680 ± 0.011

0.750 ± 0.010

134
 

0.999 ± 0.001

6
1.0E6

Tetradecene

Abundance

Abundance

Dodecene

IS
Tetradecane
Hexadecene
Dodecane
Decene

0
0

0

Retention Time (min)
Retention Time (min)

17
17

Figure 4.4: Representative total ion chromatogram of unburned high density polyethylene.

135
 

The addition of the matrix interferences from HDPE is observed in the chromatograms of
the unburned HDPE samples spiked with the ignitable liquids (Figure 4.5). As expected, the
spiked samples contain the characteristic components from the ignitable liquids, such as toluene
and the alkylbenzenes for gasoline and the C10-C16 normal alkanes for kerosene. However, the
matrix interferences have significantly changed the visual appearance of the chromatograms.
Dodecene, tetradecene, and hexadecene do not elute at the same retention times as any
compounds found in either gasoline or kerosene. However, these compounds do elute at
retention times very similar to the corresponding alkanes, which are found in kerosene. For
example, dodecene from the matrix elutes at 10.88 minutes and dodecane which is present in
kerosene elutes at 11.01 minutes. As a result, the peak pattern in the TICs of kerosenecontaining extracts is different from the expected peak pattern for petroleum distillates.
Therefore, the addition of these interferences has the potential to complicate visual identification
of ignitable liquids present in sample extracts, as discussed below.
The long chain alkene compounds elute after the gasoline compounds in the
chromatogram. Although gasoline could still be identified by visual assessment of the
chromatogram, the addition of the matrix interference peaks from the unburned HDPE could lead
an analyst to believe an additional petroleum distillate was present in the extract. On the other
hand, while the C12, C14, and C16 alkenes are not present in kerosene, these alkenes will have
similar retention times to the corresponding alkanes, which are present in kerosene. Therefore,
based on visual analysis of the chromatogram, an extract of unburned HDPE with no ignitable
liquid present may be misidentified as containing a petroleum distillate. Likewise, an extract
containing kerosene may be misidentified because the added alkenes alter the characteristic

136
 

4E5

A

C2-alkylbenzenes
Toluene b
C3-alkylbenzenes
IS

Tetradecene

Abundance

Dodecene

h

e
a c

Dodecane

f
g

Tetradecane
Hexadecene

d
0
5E5

B

Tetradecene
Dodecene

Abundance

IS

Hexadecene
C11
Toluene

Decene

C10

C12
C14
C13

0

0

Retention Time (min)

17

Figure 4.5: Total ion chromatograms of high density polyethylene (HDPE) spiked with A) neat
gasoline and B) neat kerosene showing the addition of C10-, C14-, and C16 alkenes and the C12
and C14 alkanes from the unburned HDPE matrix. Major compounds are labeled:
a) ethylbenzene, b) p-xylene, c) o-xylene, d) propylbenzene, e) m-ethyltoluene, f) p-ethyltoluene,
g) o-ethyltoluene, and h) 1,3,5-trimethylbenzene. IS indicates the internal standard,
nitrobenzene. Cx denotes the alkane of the indicated carbon chain length (e.g., C12 is dodecane).
137
 

distribution of normal alkanes analysts look for when identifying residues of petroleum
distillates.
It should be noted that the toluene peak observed in the TIC of unburned HDPE spiked
with kerosene (Figure 4.5B) is from the HDPE matrix. Toluene is used in the manufacturing
process, and residual toluene was observed during the analysis of some HDPE samples. Because
gasoline at the neat and 10% evaporation levels contain a significant amount of toluene, the
addition of toluene from the HDPE to extracts containing these ignitable liquids did not
significantly alter the appearance of the chromatogram of the extract. However, for the ignitable
liquids that do not contain toluene (i.e., 90% evaporated gasoline and all evaporation levels of
kerosene), the addition of toluene was much more obvious and could potentially cause
misidentification of the ignitable liquid present in the extract.

4.3.6 Association and Discrimination of Samples in the Presence of Inherent Matrix
Interferences
PCA was used to assess the association of the ignitable liquid residues (ILRs) to the
corresponding ignitable liquid standard despite the addition of matrix interferences inherent in
HDPE. For the gasoline extracts, residues extracted from unburned HDPE are generally
positioned closely to their corresponding ignitable liquid standard, though considerable spread
among replicate extractions is observed in the extracts of HDPE spiked with 10% and 90%
evaporated gasoline (Figure 4.6). On the other hand, extracts containing all evaporation levels of
kerosene are most closely visually associated with the neat kerosene standard.
Looking at Figure 4.5, the addition of dodecene, tetradecene, and hexadecene from the
HDPE matrix changes the compounds observed in the TICs of sample extracts. Therefore, it is

138
 

PC2 (24.0%)

1.0E6

4
-1.8E6

5
3

1.8E6
2
1

-1.0E6

PC1 (65.6%)
Figure 4.6: Scores plot of the first principle component (PC1) versus the second principle
component (PC2) based on the total ion chromatograms for the six ignitable liquid standards and
projected scores for ignitable liquids spiked onto unburned high density polyethylene (HDPE).
Each standard is denoted as follows: Neat gasoline ( ), 10% evaporated gasoline ( ), 90%
evaporated gasoline ( ), neat kerosene ( ), 10% evaporated kerosene ( ), and 70% evaporated
kerosene ( ). Extracts from the unburned HDPE are indicated by half fill.

139
 

expected that these compounds would also affect the scores of the extracts compared to the
standards. However, these alkenes do not elute at the same retention times as any of the
compounds in either gasoline or kerosene. Because only the ignitable liquid standards were used
in generating the scores plot and loadings plots, these interference compounds would have to
elute at the same retention time as compounds in the liquid standards to contribute to an extract’s
score on PC1 and PC2. The C12, C14, and C16 alkenes do not coelute with compounds from the
standards, so the addition of these compounds to the extract chromatograms has no effect on the
score of the extract projected onto the scores plot. On the other hand, the C12 and C14 alkanes
from unburned HDPE will elute at the same retention time as the C12 and C14 alkanes in
kerosene. Thus, these interferences do alter the position of the extracts compared to the standard,
as discussed below.
Extracts containing neat gasoline are positioned slightly more negatively on both PC1
and PC2 than the corresponding standard. The C12 and C14 alkanes, present in HDPE and
kerosene, load positively on both PCs. Therefore, the extracts would be expected to load more
positively on both PCs than the corresponding standard. However, the effect of the addition of
these alkanes is negated by a higher abundance of the toluene and alkylbenzenes from gasoline
in the extract compared to the neat gasoline standard not corrected for during normalization.
Toluene and the C2- and C3-alkylbenzenes load negatively on PC1, therefore an increase in
abundance of these compounds positions the extracts of the spiked HDPE more negatively on
this PC. Toluene and the C2-alkylbenzenes also load negatively on PC2. Although the C3alkylbenzenes load positively on this PC, these compounds are weighted much less heavily than

140
 

toluene and the C2-alkylbenzenes. Therefore, the extracts load negatively on PC2.
Spread is observed among replicate extracts containing 10% evaporated gasoline. The
TICs of these extracts show high abundance peaks for the C12, C14, and C16 alkenes but no
peaks for the corresponding alkanes, due to variation in the plastic samples used in this study.
Therefore, the difference in position between the extracts and the corresponding standard is due
to differences in abundance of compounds from gasoline not corrected for during normalization
instead of the interferences from HDPE, as described below.
Three of the extracts (labeled 1 in Figure 4.6) are positioned more negatively on PC1 and
slightly more negatively on PC2 than the 10% evaporated gasoline standard. These extracts
show a higher abundance of the toluene and the C2- and C3-alkylbenzenes compared to the
standard. Toluene and the C2- and C3-alkylbenzenes load negatively on PC1, so an increased
abundance of these compounds positions the extracts more negatively on this PC. On PC2,
toluene and the C2-alkylbenzenes load negatively, but the C3-alkylbenzenes load positively.
Therefore, the positive contributions from the loading of the C3-alkylbenzenes partially offset
the negative contributions from toluene and the C2-alkylbenzenes, positioning the extracts only
slightly more negatively than the standard on PC2.
The other two extracts of unburned HDPE spiked with 10% evaporated gasoline (labeled
2 in Figure 4.6) are positioned more positively on PC1 and slightly more positively on PC2 than
the standard. The reasoning is parallel but opposite to that described above. The reduced
abundances of the gasoline compounds, specifically toluene and the C2-alkylbenzenes in these
extracts, contribute less negatively to the loadings of the extracts on PC1 and PC2. Thus, the
141
 

extracts are positioned more positively on both PCs compared to the standard.
Spread among extracts of unburned HDPE spiked with 90% evaporated gasoline is due to
differences both in the abundances of the compounds from gasoline as well as differences in the
abundances of interferences from the matrix. For example, only two of the extracts (labeled 3 in
Figure 4.6) show significant abundances for the C12 and C14 alkanes from the unburned HDPE.
The addition of these interferences would be expected to position the extracts more positively on
both PC1 and PC2 than the standard. However, the extracts are positioned more negatively on
both PCs, due to a combination of a significant toluene peak from the HDPE and a higher
abundance of the C3- and C4-alkylbenzenes in these extracts compared to the standard. Toluene
and the alkylbenzenes load negatively on PC1. The negative contribution to the score of the
extracts from these compounds from the evaporated gasoline is greater than the positive
contribution from the C12 and C14 alkanes from the matrix; therefore, the extracts are positioned
more negatively on PC1 than the standard. While toluene loads negatively on PC2, the C3- and
C4-alkylbenzenes and the C12 and C14 alkanes load positively. However, toluene is much more
heavily weighted on PC2 than the alkylbenzenes and alkanes, so the extracts are positioned more
negatively than the standard on PC2.
Similarly, another one of the 90% evaporated gasoline extracts (labeled 4 in Figure 4.6) is
positioned more negatively on both PCs compared to the standard. The TIC of this extract
shows a significant toluene peak from the HDPE matrix, but no interferences from the C12 and
C14 alkanes. Toluene loads negatively on both PCs, so this extract is positioned more negatively
on both PCs, as previously discussed.
142
 

The TICs of two of the extracts (labeled 5 in Figure 4.6) show less abundant C3- and C4alkylbenzenes than the 90% evaporated gasoline standard and no C12 and C14 alkane or toluene
peaks from the unburned matrix. These alkylbenzenes load negatively on PC1, so a decrease in
abundance positions the extract less negatively on PC1 than the standard. Likewise, the C3- and
C4-alkylbenzenes load positively on PC2, so the extracts are positioned less positively on PC2
compared to the standard.
Extracts containing neat, 10% evaporated, and 70% evaporated kerosene are positioned
more negatively on PC1 and PC2 than the corresponding standards. The TICs of these extracts
show less abundant peaks for the C13-C16 normal alkanes compared to the standards, even
though tetradecane is added from the matrix. Because the C13-C16 normal alkanes load
positively on both PC1 and PC2, reduced abundances of the C13-C16 normal alkanes will
position the extracts less positively on both PCs. Additionally, the TICs of all extracts
containing kerosene show a toluene peak from the HDPE matrix. Toluene loads negatively on
both PC1 and PC2, so the addition of this compound to the TICs will also contribute to the less
positive positioning of the extracts on these PCs, as previously discussed.

4.3.7 Assessment of Association and Discrimination of Samples in the Presence of Inherent
Matrix Interferences Using PPMC Coefficients
PPMC coefficients were calculated between replicates of the spiked unburned HDPE
extracts and the corresponding liquid standards (Table 4.3). PPMC coefficients of replicates are
not equal to 1, mainly due to variation in the extraction procedure. However, coefficients of
143
 

Table 4.3: Pearson product moment correlation coefficients for replicates (n=15) of each
ignitable liquid extracted from unburned high density polyethylene and the range of coefficients
for the extracts correlated to the corresponding ignitable liquid standard.

Ignitable Liquid Standard

Mean PPMC Coefficient ±
Standard Deviation for
Replicates of Extracts
(n=225)

Neat Gasoline
10% Evaporated Gasoline
90% Evaporated Gasoline

0.991 ± 0.007
0.91 ± 0.10
0.8 ± 0.2

Neat Kerosene
10% Evaporated Kerosene
70% Evaporated Kerosene

0.97 ± 0.03
0.95 ± 0.05
0.97 ± 0.04

144
 

Range of PPMC Coefficients
Between Spiked Unburned HDPE
Samples and Corresponding
Ignitable Liquid Standard
(n=225)
0.931 - 0.773
0.702 - 0.342
0.991 - 0.535
0.819 - 0.514
0.674 - 0.346
0.805 - 0.538

replicates are generally greater than 0.91, indicating acceptable precision in the extraction
procedure. The lower coefficients observed, especially for extracts of unburned HDPE with 90%
evaporated gasoline, which have a mean PPMC coefficient of 0.8, is due to the variability in the
matrix interferences present in the samples. While PCA takes into account approximately 90%
of the variance in the data set, PPMCs take into account the entire chromatogram. Therefore, the
interferences, such as the C12, C14, and C16 alkenes, which did not affect the positioning of the
extracts on the scores plot, impact the PPMC coefficients calculated between the extracts and the
corresponding standards.
When the range of PPMC coefficients is calculated between each set of unburned HDPE
extracts and the corresponding standard, neat and 90% evaporated gasoline extracts show strong
to moderate correlations to the corresponding ignitable liquid standard. However, extracts
containing 10% evaporated gasoline show moderate to weak correlations to the standard.
Similarly, extracts from the HDPE spiked with the various evaporation levels of kerosene
demonstrate strong to moderate correlation for neat and 70% evaporated kerosene extracts and
moderate to weak correlations for 10% kerosene extracts.
The wide range of coefficients observed for each of the ignitable liquids is due to the
variety of different matrix interferences from the unburned HDPE observed in the sample
chromatograms, which is a consequence of the variability in sample sources. For example, one
extract of unburned HDPE spiked with 90% evaporated gasoline (labeled 4 in Figure 4.6)
showed only a toluene interference peak from the matrix, as discussed previously. Therefore,
this extract would be strongly correlated to the 90% evaporated gasoline standard. On the other
hand, the TICs of the other four extracts containing 90% evaporated gasoline show multiple
matrix interference peaks, including toluene, dodecane, dodecene, tetradecane, tetradecene, and
145
 

hexadecene. The increased number of interferences results in a lower correlation to the standard.
This variability in the matrix interferences observed in the TICs also explains the range of
correlations observed for the neat and 10% evaporated gasoline samples.
Large differences in abundance between the kerosene standards and extracts lowers the
PPMC coefficients calculated between the two because the peak widths for compounds in the
TICs are slightly different. PPMC coefficients are insensitive to differences in peak height but
are sensitive to differences in peak width. For example, the C14 normal alkane peak in the 70%
evaporated kerosene standard is four times more abundant than the C14 normal alkane peak in an
extract of unburned HDPE spiked with 70% evaporated kerosene (Figure 4.7). In addition to
having a greater abundance, the peak in the standard chromatogram is wider than the
corresponding peak in the sample extract. While the apexes of the peaks are at the same
retention time, the tails of the peaks are not. Because PPMC coefficients are calculated using a
point-by-point comparison of the chromatograms, minute differences in the retention time at
which the peaks begin and end result in reduced coefficients and poorer correlations.
Kerosene-containing extracts were positioned between the neat gasoline and neat
kerosene standards on the scores plot, albeit closer to the neat kerosene standard, so PPMC
coefficients were calculated correlating the kerosene extracts to each of these standards (Table
4.4). Within the standard deviation range, both the neat kerosene and 10% evaporated kerosene
extracts show moderate correlations to both the gasoline and the kerosene standards. Thus, none
of these extracts can be assigned to one ignitable liquid over the other. Only the 70% evaporated
kerosene extracts can be correctly identified as containing kerosene, as these extracts show a
strong correlation to the neat kerosene standard (0.84 ± 0.03) and only a moderate correlation to
the neat gasoline standard (0.57 ± 0.08).
146
 

Abundance

3.5E5

0
13.70

Retention Time (min)

13.85

Figure 4.7: The C14 normal alkane peak in the total ion chromatogram (TIC) of the 70%
evaporated kerosene standard (──) and an extract of unburned high density polyethylene
(HDPE) spiked with 70% evaporated kerosene (─ ─) demonstrating the slight change in peak
width which results in a lower correlation between the two chromatograms. Solid vertical lines
indicate where the C14 normal alkane peak begins and ends in the TIC of the 70% evaporated
kerosene standard, while the dotted lines indicate the beginning and end of the same alkane peak
in the TIC of unburned HDPE spiked with 70% evaporated kerosene.

147
 

Table 4.4: Mean Pearson product moment correlation coefficients (± standard deviation) for
replicates (n=15) of residues of kerosene at each evaporation level extracted from unburned high
density polyethylene correlated to the neat gasoline and neat kerosene ignitable liquid standards
(n=225).

ILR in Extract
Neat Kerosene
10% Evaporated
Kerosene
70% Evaporated
Kerosene

Mean PPMC Coefficient ± Standard Deviation for Replicates of
Extracts Correlated to:
Neat Kerosene Standard
Neat Gasoline Standard
0.65 ± 0.10

0.47 ± 0.07

0.58 ± 0.13

0.47 ± 0.10

0.84 ± 0.03

0.57 ± 0.08

148
 

This result, and also the moderate to weak correlations observed between the other
extracts and the respective standards, suggests that PPMC coefficients cannot be used to
statistically identify ILRs found on unburned HDPE due to the addition of the interferences from
unburned HDPE, including the C12, C14, and C16 alkenes as well as the C12 and C14 alkanes to
the TICs of the samples.

4.3.8 Identification of Burned Matrix Interferences and Selection of Burn Time
The TICs of the burned HDPE showed a variety of matrix interferences depending on the
burn time (Figure 4.8). The interferences observed in the burned HDPE are products from the
degradation of the polyethylene polymer.
As discussed previously, the TIC of the unburned HDPE with no ignitable liquid present
was dominated by the alkenes dodecene, tetradecene, and hexadecene (Figure 4.8A). After
burning for 10 s, peaks for the C11 and C15 alkenes were more significant (Figure 4.8B). These
alkenes derive from the thermal degradation of the C12 and C16 alkenes in HDPE.
1

Polyethylene and other polymers undergo random scission when heated . This scission
results in the formation of a triplet of peaks where the first peak of the series is the alkadiene, the
second is the alkene, and the third is the alkane. The peaks are usually present in the shape of a
bell curve, with the alkene as the largest peak. This bell curve distribution was first noticeable
after 20 s of burning (Figure 4.8C). In addition, at this burn time C9 and C13 peaks were also
visible. Thus, it was determined that 20 s was the minimum burn time to observe significant
thermal degradation of HDPE.

149
 

1.8E4

C14

C12

A

b

C10
0
1.8E4

c
C14

B

C12
C10 C11

Abundance

0
1.8E4

C16

C15 C16
C14

C

C12
C9

C10 C11 a

b
C
c 13

C15 C16

0

C14

1.8E4

D
C9

0
1.8E4 E

C10

C10
C8

C11

C11

C12

C13
C14

C12
C13

C9

0
1.8E4 F

C10
C8

0

0

C15 C16

C9

C15 C16
C14

C11 C12

C13

Retention Time (min)

C15 C16
17

Figure 4.8: Representative total ion chromatograms of extracts of high density polyethylene
burned for A) 0 seconds, B) 10 seconds, C) 20 seconds, D) 30 seconds, E) 60 seconds, and F)
120 seconds. Major interference compounds are identified by brackets according to carbon chain
length. Triplet peaks are indicated as follows: a) the alkadiene, b) the alkene, and c) the alkane.
150
 

Increasing the burn time to 30 s had little effect on the appearance of the TIC (Figure
4.8D). However, when the burn time was further increased to 60 s, the bell-shaped distribution
of the alkadiene, alkene, and alkane peaks was visible for compounds with carbon chains of eight
to sixteen carbons (Figure 4.8E).

No significant changes were observed in the TICs between a

burn time of 60 s and 120 s (Figure 4.8F). Therefore, 60 s was chosen as the burn time for
convenience.
The contribution of the matrix interferences from the burned HDPE can be observed in
the chromatograms of the extracts of the burned HDPE spiked with gasoline and kerosene
(Figure 4.9). As expected, the extracts contain the characteristic compounds from the ignitable
liquids, such as toluene and the alkylbenzenes for gasoline and the C10-C16 normal alkanes for
kerosene. However, the matrix interferences have visually altered the chromatograms
significantly. For example, peaks for the C8-C16 alkadienes, alkenes, and alkanes were present in
the TICs of neat gasoline extracts (Figure 4.9A). Nonene and decene coelute with o-xylene and
1,3,5-trimethylbenzene, respectively, masking the triplet peak distribution from HDPE and
possibly changing the characteristic ratios of the alkylbenzene groups in gasoline.
The most obvious additions to the TICs of the extracts of burned HDPE spiked with
kerosene are C10-C16 alkenes, which make a doublet peak with the alkanes present in kerosene,
and the C8 and C9 triplet peaks (Figure 4.9B). The C10-C16 alkadienes are masked by other
compounds in kerosene. The addition of the C10-C16 alkanes from the matrix also changes the
ratios of these alkanes compared to the kerosene standard. Thus, the identification of an ILR as
kerosene could be difficult based solely on a visual analysis of the chromatograms.

151
 

5E5

A
C3-alkylbenzenes IS
C2-alkylbenzenes

b

Abundance

Toluene

h

a
C8

C14

C12

e
C11

c
C9

d

f C10
g

C13
C15

C16

0
0
5E5

21

Retention Time (min)

B
IS

Abundance

C11

C14

C12
C10
Toluene
C8
0

0

C9

C13

C16
C15

Retention Time (min)

21

Figure 4.9: Representative total ion chromatogram of burned high density polyethylene (HDPE)
spiked with A) neat gasoline and B) neat kerosene, showing the contribution of the matrix
interferences from burned HDPE. Major compounds are labeled: a) ethylbenzene, b) p-xylene,
c) o-xylene, d) propylbenzene, e) m-ethyltoluene, f) p-ethyltoluene, g) o-ethyltoluene, and h)
1,3,5-trimethylbenzene. IS indicates the internal standard, nitrobenzene. Cx denotes the triplet
peaks of the indicated carbon chain length (e.g., C12 bracket indicates dodecadiene, dodecene,
and dodecane).
152
 

4.3.9 Association and Discrimination of Samples in the Presence of Burned Matrix Interferences
To assess the effects of interferences introduced through burning of the matrix on the
association of extracts to the corresponding standard, the ignitable liquid standards were spiked
onto previously burned HDPE samples, extracted, and analyzed. Scores calculated for the
extracts were then projected onto the PCA scores plot generated for the liquid standards (Figure
4.10).
The TICs of the gasoline and kerosene extracts contain C8-C16 triplet peaks from the
burned matrix. The C8 peaks do not elute at the same retention time as any compounds in either
gasoline or kerosene, and therefore have no effect on the positioning of the extracts on the scores
plot. Likewise, only the C11-C16 alkanes in each hydrocarbon triplet from the burned HDPE
correspond to compounds in the loadings plots. Thus, these alkanes contribute to the scores of
extracts while the corresponding alkenes and alkadienes do not. The two exceptions to this are
nonene and decene, which elute at the same retention time as o-xylene and 1,3,5trimethylbenzene, respectively. These compounds are in gasoline, therefore nonene and decene
do contribute to the positioning of the gasoline extracts, but the corresponding alkanes and
alkadienes do not. These trends are described in detail below for extracts containing each of the
ignitable liquids.
Generally speaking, some spread among extracts is expected due to variability in the
burning process and the generated interferences. Spread among extracts containing neat and 10%
evaporated gasoline causes these extracts to overlap each other as well as the two corresponding
standards. The observed spread is due, in large part, to varying amounts of matrix interferences
in the TICs of these extracts. Some extracts containing neat gasoline are positioned more

153
 

PC2 (24.0%)

1.0E6

-1.8E6

1.8E6

-1.0E6

PC1 (65.6%)
Figure 4.10: Scores plot of the first principle component (PC1) versus the second principle
component (PC2) based on the total ion chromatograms for the six ignitable liquid standards and
projected scores for samples of burned HDPE spiked with each ignitable liquid. Each standard is
denoted as follows: Neat gasoline ( ), 10% evaporated gasoline ( ), 90% evaporated gasoline
( ), neat kerosene ( ), 10% evaporated kerosene ( ), and 70% evaporated kerosene ( ).
Extracts from the burned HDPE are indicated by half fill.

154
 

negatively on PC1 and less negatively on PC2 than the corresponding standard. The C11-C16
alkanes from the burned HDPE contribute positively to the score of the extracts on PC1 because
these compounds load positively on this PC. On the other hand, because nonene and decene
coelute with o-xylene and 1,3,5-trimethylbenzene from gasoline, which load negatively on PC1,
the addition of these compounds contributes negatively to the scores on PC1 for the extracts
containing neat gasoline. Toluene from the burned matrix was also added to the TICs of the
extracts. Toluene contributes negatively to the score of the extracts on PC1 because it loads
negatively on this PC.
The net positioning of the neat gasoline extracts on PC1 is more negative than the
corresponding standard due to the heavier weight given to the compounds from gasoline in the
loadings plot for PC1. On this PC, 1,3,5-trimethylbenzene is the most heavily weighted
compound. In addition, the C11-C12 normal alkanes make smaller positive contributions to the
loadings on PC2 compared to the other normal alkanes. The C11-C12 normal alkanes also make
smaller contributions compared to the loadings of the C11-C12 normal alkanes on PC1 (Figure
4.2A). The peak resulting from the coelution of decene from the matrix and 1,3,5trimethylbenzene from gasoline has a height two to five times larger than any of the C11-C16
normal alkanes in the TICs of the extracts. Therefore, despite the positive contribution of the
alkanes from the matrix to the score, the negative loadings from toluene, nonene (coelutes with
o-xylene), and particularly decene (coelutes with 1,3,5-trimethylbenzene) from burned HDPE
positions the extracts more negatively on PC1 than the neat gasoline standard.

155
 

On PC2, toluene and the C2-alkylbenzenes load negatively, and all other compounds load
positively. While 1,3,5-trimethylbenzene loads positively on this PC, the weighting given to the
compound is significantly less than the weighting on PC1 (Figure 4.2B). Thus, although the
peak resulting from the coelution of 1,3,5-trimethylbenzene and decene is the largest in the
chromatogram, it does not contribute significantly to the loadings on PC2. Instead, toluene is the
most heavily weighted compound. Additionally, the C11-C16 normal alkanes are weighted more
heavily on PC2 than on PC1. Therefore, the positive loadings from decene and the C11-C16
normal alkanes negate the negative loadings from toluene and nonene, positioning the extracts
similarly on PC2 to the corresponding standard.
Two extracts containing neat gasoline are positioned more positively on both PCs than
the corresponding standard. These extracts show less abundant peaks for toluene and the C2alkylbenzenes than the neat gasoline. These compounds load negatively on PC1 and PC2, so a
reduction in the abundance of toluene and the C2-alkylbenzenes results in a less negative
positioning on both PCs. In addition, these extracts had C12-C16 interference peaks with the
highest abundance compared to the other extracts. The addition of the C12-C16 alkanes from the
burned HDPE contributes positively to the extracts’ score on PC1 and PC2. Therefore, due to a
combination of the decreased abundances of toluene and the C2-alkylbenzenes and the high
abundances of the C12-C16 alkanes, these two extracts are positioned more positively on PC1
and PC2 than the corresponding standard. Spread among the extracts positioned more positively
on both PCs than the standard is due to the varying amounts of the C12-C16 alkanes from the

156
 

burned HDPE present in the TICs of the extracts, which is to be expected due to variability in the
burning process and HDPE samples used.
Similar to what was observed with the neat gasoline extracts, some extracts containing
10% evaporated gasoline are positioned more positively on both PCs than the corresponding
standard while others are positioned more negatively on both PCs. The 10% evaporated gasoline
extracts positioned more positively on both PCs compared to the standard showed reduced
abundances of toluene and the C2-alkylbenzenes and higher abundances of interferences from
the burned HDPE than other extracts. On the other hand, extracts positioned more negatively on
both PCs show higher abundances of toluene and the C2-alkylbenzenes compared to the standard
and lower abundances of interferences from the burned HDPE than other extracts. The
reasoning behind the positioning of both groups of extracts is the same as described above.
Extracts containing 90% evaporated gasoline are positioned more positively on PC1 and
more negatively on PC2 than the corresponding standard. In the TICs, the extracts show less
abundant peaks of the compounds from the gasoline compared to the standard. Despite the
negative contributions to the scores of the extracts due to the addition of toluene, nonene, and
decene from the burned HDPE, the reduction in the abundance of the C2-, C3-, and C4alkylbenzenes from the evaporated gasoline positions the extracts more positively on PC1
compared to the standard. Also, the C11-C16 normal alkanes from the burned HDPE contribute
to the positive positioning on PC1, as previously discussed. Likewise, these alkanes contribute
positively to the loadings of the extracts on PC2. However, the reduction in abundance of the
C3- and C4-alkylbenzene, which load positively on this PC, as well as the addition of toluene and
nonene, which load negatively, contribute more negatively to the loadings than decene and the
157
 

C10-C16 normal alkanes contribute positively. Therefore, the net effect is a more negative
positioning of the extracts containing 90% evaporated gasoline on PC2 compared to the standard.
All extracts containing kerosene are positioned more negatively on PC1 and PC2 than the
respective standards due to a combination of the addition of toluene, nonene, and decene from
the burned HDPE as well as reduced abundances of the C13-C16 normal alkanes observed in the
TICs of the extracts compared to the corresponding standards. For example, TICs of extracts
containing neat kerosene show significant peaks for toluene, nonene, and decene. Nonene and
decene elute at similar retention times as o-xylene and 1,3,5-trimethylbenzene, two compounds
in gasoline. Toluene, o-xylene and 1,3,5-trimethylbenzene load negatively on PC1, so the
compounds will contribute negatively to the loadings of the extracts on this PC. Toluene and oxylene load negatively on PC2 while 1,3,5-trimethylbenzene loads positively. Toluene is much
more heavily weighted on this PC than the other two compounds. Therefore, the extracts will be
positioned more negatively on PC2 than the standard.
As discussed previously, neither the alkadienes nor the C11-C16 alkenes elute at the same
retention time as any compounds in gasoline or kerosene. Thus, these compounds do not affect
the positioning of the extracts on the scores plot. However, the C11-C16 normal alkanes will
contribute, as these alkanes are found in kerosene. The C11-C16 normal alkanes load positively
on both PCs. Despite the addition of these alkanes from the burned matrix, the TICs of the
extracts containing neat kerosene show less abundant peaks for these alkanes compared to the
standard. Therefore, the positive contribution of the normal alkanes to the loadings of the
extracts on both PCs will be reduced. Taken together with the negative contribution to the
loadings on both PCs caused by the addition of toluene and nonene from the burned matrix, the
158
 

reduction in the abundance of the C11-C16 normal alkanes results in the positioning of the neat
kerosene extracts more negatively on both PCs compared to the corresponding standard.
Similar arguments explain the positioning of extracts containing 10% evaporated
kerosene and 70% evaporated kerosene, though it should be noted that the reduction in the
abundance in the normal alkane peaks for the 70% evaporated kerosene extracts was much more
significant compared to the corresponding standard than was observed for the other kerosene
extracts. Thus, these extracts are positioned much farther from the corresponding standard than
the neat and 10% evaporated kerosene-containing extracts. The difference in abundances of the
normal alkane peaks was not accounted for during normalization procedures.

4.3.10 Assessment of Association and Discrimination of Samples in the Presence of Burned
Matrix Interferences Using PPMC Coefficients
PPMC coefficients were calculated between replicates of the burned HDPE extracts and
the corresponding standards (Table 4.5). Coefficients of replicates of extracts containing 90%
evaporated gasoline and 10% and 70% evaporated kerosene are greater than 0.99, while extracts
containing neat and 10% evaporated gasoline and neat kerosene have coefficients greater than
0.94. The difference in coefficients is due to differences in the matrix interferences observed
between replicates due to variability in the burning procedure and plastic samples used.
However, all coefficients indicate acceptable precision in the extraction procedure despite this
variability.
When the range of PPMC coefficients is calculated between each set of sample replicates
and the corresponding standard, extracts containing 10% and 90% evaporated gasoline show
strong correlations to the corresponding ignitable liquid standard (Table 4.5). The comparably
159
 

Table 4.5: Pearson product moment correlation coefficients for replicate (n=15) extractions of
each ignitable spiked onto burned high density polyethylene and the mean (± standard deviation)
and range of coefficients for the extracts correlated to the corresponding ignitable liquid
standard.

Ignitable Liquid Standard

Mean PPMC Coefficient ±
Standard Deviation for
Replicates of Extracts
(n=225)

Range of PPMC Coefficients
Between Spiked Burned
HDPE Samples and
Corresponding Ignitable
Liquid Standard (n=225)

Neat Gasoline
10% Evaporated Gasoline
90% Evaporated Gasoline

0.95 ± 0.05
0.97 ± 0.03
0.995 ± 0.004

0.937 - 0.661
0.985 - 0.843
0.974 - 0.945

Neat Kerosene
10% Evaporated Kerosene
70% Evaporated Kerosene

0.94 ± 0.06
0.991 ± 0.006
0.990 ± 0.007

0.948 - 0.657
0.846 - 0.739
0.725 - 0.597

160
 

small range of coefficients for extracts containing 90% evaporated gasoline is due to the fact that
the high abundance of the C3- and C4-alkylbenzenes masks the C8-C13 triplet interference peaks
from the burned HDPE. Neat gasoline extracts demonstrate strong to moderate correlations.
The moderate correlations observed for the neat gasoline extracts are due to the presence of a
wider peak resulting from the coelution 1,3,5-trimethylbenzene and decene in one of the extracts,
similar to what was seen previously when the liquids were spiked onto unburned HDPE.
Close positioning on the scores plot indicates that extracts of burned HDPE spiked with
neat and 10% evaporated gasoline contain similar abundances of similar compounds from both
gasoline and the burned matrix. This conclusion is reinforced using PPMC coefficients. The
mean PPMC coefficient between the extracts of burned HDPE spiked with neat gasoline and
those spiked with 10% evaporated gasoline is 0.93 ± 0.07, showing a strong correlation (Table
4.6). In addition, neat gasoline extracts have a mean PPMC coefficient of 0.84 ± 0.10 when
compared to the corresponding standard (n=225) and a mean coefficient of 0.84 ± 0.11 when
compared to the 10% evaporated gasoline standard (n=225), indicating a strong correlation of the
extracts to either standard. Correlation of the 10% evaporated gasoline extracts to each standard
yields similar results. Thus, based on both the visual analysis of the scores plot and analysis
using PPMC coefficients, it is not possible to associate the extracts to one evaporation level over
the other.
Extracts from the HDPE spiked with neat and 10% evaporated kerosene demonstrate
strong to moderate correlations to the corresponding ignitable liquid standard, while extracts
containing 70% evaporated kerosene show moderate correlations to the standard. The moderate
correlations observed for 70% evaporated kerosene extracts are due to a more significant altering
of the chemical compounds observed in the chromatograms compared to the other evaporation
161
 

Table 4.6: Comparison of mean Pearson product moment correlation coefficients (± standard
deviation) calculated between extracts of burned high density polyethylene spiked with all
evaporation levels of gasoline to the gasoline standards.

ILR in Extract
Neat Gasoline
10% Evaporated Gasoline
90% Evaporated Gasoline

Mean PPMC Coefficient ± Standard Deviation for Replicates of
Extracts Correlated to:
Neat Gasoline
10% Evaporated
90% Evaporated
Standard
Gasoline Standard
Gasoline Standard
0.84 ± 0.10
0.84 ± 0.11
0.61 ± 0.03
0.92 ± 0.05
0.93 ± 0.05
0.62 ± 0.03
0.68 ± 0.03
0.71 ± 0.03
0.96 ± 0.01

162
 

levels of kerosene. At the 70% evaporation level, the C10 and C11 alkanes have been evaporated
completely. The burned matrix essentially adds these alkanes back into the chromatogram along
with the other interferences from HDPE, such as the alkenes and alkadienes previously
discussed. Thus, the extracts more closely resemble neat kerosene than 70% evaporated
kerosene in terms of the normal alkanes present in the TIC.
Accordingly, PPMC coefficients were calculated between HDPE extracts containing
kerosene at all three evaporation levels and each of the kerosene liquid standards to assess
whether the extracts could be assigned to a particular evaporation level (Table 4.7). The extracts
were also compared to the neat gasoline standard because extracts of neat and 10% kerosene
were positioned between the neat kerosene and neat gasoline standards. The neat kerosene and
10% evaporated kerosene extracts show strong correlations to the neat kerosene standard (0.81 ±
0.10 and 0.83 ± 0.03, respectively) and only moderate correlations to the neat gasoline standard
within the standard deviation (0.51 ± 0.08 and 0.47 ± 0.05, respectively). Similarly, extracts of
burned HDPE spiked with 70% evaporated kerosene show moderate correlations to both the neat
kerosene standard (0.75 ± 0.03) and the 70% evaporated kerosene standard (0.66 ± 0.03).
However, the 70% evaporated kerosene extracts have a weak correlation to the gasoline standard
(0.41 ± 0.05). Thus, all of the kerosene extracts can be correctly identified as containing
kerosene and not gasoline, but not necessarily the level of evaporation of kerosene.
When the ignitable liquids were spiked onto the unburned HDPE matrix, the peaks from
the matrix interferences were at a very high abundance, resulting in different peak widths
between the standard and the extract. PPMC coefficients are calculated on a point-by-point basis
between two chromatograms, so wider interference peaks correspond to more data points that are
not well-correlated and overall lower coefficients calculated for the chromatograms. However,
163
 

Table 4.7: Comparison of mean Pearson product moment correlation coefficients (± standard
deviation) calculated between extracts of burned high density polyethylene spiked with all
evaporation levels of kerosene to standards for the corresponding liquid, neat kerosene, and neat
gasoline.

ILR in Extract
Neat Kerosene
10% Evaporated Kerosene
70% Evaporated Kerosene

Mean PPMC Coefficient ± Standard Deviation for Replicates
of Extracts Correlated to:
Corresponding
Neat Kerosene
Neat Gasoline
Standard
Standard
Standard
0.81 ± 0.10
0.81 ± 0.10
0.51 ± 0.08
0.80 ± 0.03
0.83 ± 0.03
0.47 ± 0.05
0.66 ± 0.03
0.75 ± 0.03
0.41 ± 0.05

164
 

with thermal degradation of the matrix, the abundances of the interference peaks are reduced,
resulting in interference peaks in the chromatograms that are narrower. Also, the width of the
normal alkane peaks in the sample extracts is more similar to those in the liquid standards. Thus,
although a greater number of different interference compounds are observed in the TICs of the
burned HDPE sample extracts, the correlations observed between the samples and the standards
are actually stronger than was observed for the unburned HDPE extracts.

4.3.11 Association and Discrimination of Samples in the Presence of Thermal Degradation
In the last part of this study, the ignitable liquid standards were spiked onto unburned
HDPE samples, and then the samples were burned for 60 s to simulate fire debris. All samples
were extracted and analyzed as previously described. Scores calculated for the extracts were
then projected onto the PCA scores plot generated for the liquid standards (Figure 4.11). The
scores plot, as well as PPMC coefficients, was used to assess the association of the ignitable
liquid residues extracted from the burned HDPE to the corresponding ignitable liquid standard
despite thermal degradation of both the ignitable liquid and the matrix during the burning
process.
When looking at the scores plot, it should be noted that extracts containing neat and 10%
evaporated gasoline are positioned near the origin and between the gasoline and kerosene
standards. Thus, based on a visual analysis of the score’s plot alone, it would be difficult to
associate the extracts with one type of ignitable liquid over the other. The reasoning for the
positioning of these extracts is detailed below.
During the burning process, the more volatile components of gasoline are lost, including
toluene and the C2-akylbenzenes. This loss mimics the evaporative loss of these compounds, as

165
 

1.0E6

PC2 (24.0%)

2

-2.0E6

2.0E6

1

-1.0E6

PC1 (65.6%)
Figure 4.11: Scores plot of the first principle component (PC1) versus the second principle
component (PC2) based on the total ion chromatograms for the six ignitable liquid standards and
projected scores of the thermal degradation samples. Each standard is denoted as follows: Neat
gasoline ( ), 10% evaporated gasoline ( ), 90% evaporated gasoline ( ), neat kerosene ( ), 10%
evaporated kerosene ( ), and 70% evaporated kerosene ( ), burned HDPE with no ILR ( ).
Extracts from the thermal degradation samples are indicated by half fill.

166
 

observed in the evaporated standards (Chapter 3 Figure 3.3). Therefore, the extracts of samples
containing either neat or 10% evaporated gasoline are positioned less negatively on PC1 than the
standard due to the loss of these compounds and more positively on PC2 than the standard due to
the concentration of the C3-alkylbenzenes, as previously discussed. At the same time, addition
of the C12-C16 alkanes from the burned HDPE positions the extracts more positively on both
PCs, closer to the extracts of burned HDPE with no ignitable liquid added. Spread in these
samples is attributable to variability in both the loss of the volatile compounds during the
burning process and the addition of compounds from the burned HDPE.
One of the neat gasoline extracts (labeled 1 in Figure 4.11) is positioned more negatively
on PC1 and more positively on PC2 than the corresponding standard. This extract shows
significantly higher abundances of the C2- and C3-akylbenzenes than the standard. These
compounds load negatively on PC1, so the extract is positioned more negatively on PC1 than the
standard. The C2-alkylbenzens load negatively on PC2, but the C3-akylbenzenes load positively.
Despite the heavier weighting of the C2-akylbenzenes, the positive contributions of the C3akylbenzenes combined with those from the C12-C16 alkanes are greater than the negative
contributions of the C2-akylbenzenes. Therefore, the extract is positioned more positively on
PC2 than the neat gasoline standard.
The TICs of extracts of HDPE that contain 90% evaporated gasoline are visually very
similar to the TIC of the standard 90% evaporated gasoline, indicating minimal loss of the
ignitable liquid during the burning process, which is expected due to the non-volatile nature of
the C3- and C4-alkylbenzenes. In addition, because the 90% evaporated gasoline was not lost
167
 

during the burning process, the abundances of the C3- and C4-alkylbenzenes are very high,
masking the interference compounds from the burned matrix. Accordingly, the simulated ILR
extracts are positioned very closely to the standard. Extracts are positioned slightly more
negatively on PC1 and similarly on PC2 compared to the standard due to slightly higher
abundances of the C3- and C4-akylbenzenes in the extracts compared to the standard.
One of the 90% evaporated gasoline-containing extracts (labeled 2 in Figure 4.11) is
positioned more positively on PC1 and more negatively on PC2 than the standard due to a slight
drift in retention times of the compounds in the TIC, particularly in the C3-akylbenzenes (Figure
4.12). The retention time of the apexes of the C3-akylbenzenes in the TICs of the ILR extracts
do not match the apexes of these compounds in the loadings plots. Only the portions of the
peaks that have corresponding retention times to those in the loadings plots will contribute to the
scores. Thus, although the peaks are similar in abundance for this extract as the other extracts,
the drift causes the C3-akylbenzenes to contribute only partially to the loadings on the PCs and,
consequently, the extract is not positioned as negatively on PC1 or as positively on PC2 as
expected.
Thermal degradation extracts containing neat and 10% evaporated kerosene are
positioned closely to their respective standards. This positioning is expected due to the less
volatile nature of the normal alkanes present in kerosene. Extracts containing neat kerosene are
positioned slightly more positively on PC2 but similarly on PC1 compared to the corresponding
standard due to an increased abundance in the C11 and C12 alkanes, resulting from the
combination of the alkane from the ignitable liquid and from the burned HDPE. These alkanes

168
 

1.0E6
1.0E6

Abundance

e

b

c
d
a

00
6.5
6.5

Retention Time (min)

8.0
8

Figure 4.12: Total ion chromatograms of 90% evaporated gasoline standard (──) and extract
from a 90% evaporated gasoline thermal degradation sample (─ ─) demonstrating the shift in
retention times for the C3-alkylbenzenes. Compounds are labeled as follows: a) propylbenzene,
b) m-ethyltoluene, c) p-ethyltoluene, d) o-ethyltoluene, and e) 1,3,5-trimethylbenzene.

169
 

load positively on both PCs, but are more heavily weighted on PC2. Therefore, the effect of the
increased abundance of the alkanes on the positioning of the extracts is greater on PC2 than on
PC1.
On the other hand, extracts containing 10% evaporated kerosene are positioned slightly
more negatively on PC2 but similarly on PC1 compared to the corresponding standard. All of
these extracts have a toluene interference peak from the matrix in the TICs, a peak not observed
at a high abundance in the ILR extracts containing neat kerosene. Toluene loads negatively on
both PCs, so the addition of this compound will contribute negatively to the loadings on both
PCs. However, toluene is weighted more heavily on PC2, so it’s presence in the TICs of the
extracts has a greater effect on the positioning of the extracts on PC2 than on PC1.
Extracts containing 70% evaporated kerosene are positioned more negatively on both
PCs compared to the standard. The TICs of these extracts show significant peaks for the C11 and
C12 alkanes. These alkanes load positively on both PCs, so the extracts would be expected to
load more positively on both PCs. However, these extracts also have significantly reduced
abundances of the C13- C16 alkanes compared to the standard. The C13- C16 alkanes also load
positively on both PCs, so a reduced abundance in these alkanes positions the extracts less
positively on both PCs than the standard.

4.3.12 Assessment of Association and Discrimination of Samples in the Presence of Thermal
Degradation Using PPMC Coefficients
PPMC coefficients were calculated between replicates of the thermal degradation extracts
and the corresponding standards (Table 4.8). While not equal to the theoretical value of 1, most

170
 

Table 4.8: Pearson product moment correlation coefficients for replicates (n=15) of each
ignitable liquid thermal degradation extract and the range of coefficients for the extracts
correlated to the corresponding ignitable liquid standard.

Ignitable Liquid Standard

Mean PPMC Coefficient ±
Standard Deviation for
Replicates of Extracts
(n=225)

Neat Gasoline
10% Evaporated Gasoline
90% Evaporated Gasoline

0.85 ± 0.17
0.95 ± 0.05
0.95 ± 0.06

Range of PPMC Coefficients
Between Thermal
Degradation Samples and
Corresponding Ignitable
Liquid Standard (n=225)
0.846 - 0.435
0.797 - 0.550
0.979 - 0.774

Neat Kerosene
10% Evaporated Kerosene
70% Evaporated Kerosene

0.93 ± 0.06
0.96 ± 0.03
0.90 ± 0.10

0.887 - 0.705
0.945 - 0.815
0.949 - 0.607

171
 

coefficients are greater than 0.9, indicating acceptable precision in the extraction procedure
despite variability of the burning procedure. Extracts containing neat gasoline have an average
PPMC coefficient of 0.85 due to the one extract that had very few matrix interferences present in
the TIC.
Extracts containing 10% evaporated gasoline all show moderate correlation to the
corresponding standard, while extracts containing neat gasoline show strong to weak correlations
to the corresponding standard. The mean PPMC coefficient of extracts containing neat gasoline
is 0.85 ± 0.17, and these extracts also have the widest range of coefficients when correlated to
the corresponding standard. The low mean and wide range observed for the neat gasoline
extracts are due to vast differences in the compounds from the ignitable liquid and the matrix
observed in the TICs of the extracts and is reflective of the spread among the extracts in scores
plot. As discussed previously, the TIC of one of the extracts is dominated by peaks from the
gasoline with few compounds from HDPE, while the TICs of other extracts have much more
significant interferences from the matrix. The extract that has few matrix interference
compounds in its TIC has a strong correlation to the standard. On the other hand, TICs that have
significant interferences from the burned HDPE have weak correlations to the neat gasoline
standard.
Extracts containing neat and 10% evaporated gasoline are positioned between the neat
and 10% evaporated gasoline standards and the kerosene standards on the PCA scores plot due to
reduced abundances of toluene and the C2- and C3-alkylbenzenes and the addition of the C12C16 alkanes from the burned HDPE matrix in the TIC of the extracts. The neat gasoline extracts
show moderate correlations to both the neat gasoline standard and the neat kerosene standard of
0.63 ± 0.11 and 0.63 ± 0.2, respectively (Table 4.9). A similar result is observed for extracts
172
 

Table 4.9: Comparison of mean Pearson product moment correlation coefficients for extracts
(n=15) containing neat and 10% evaporated gasoline ignitable liquid residues correlated the
corresponding ignitable liquid standard (n=225) as well as the neat kerosene standard (n=225).

ILR in Extract
Neat Gasoline
10% Evaporated Gasoline

Mean PPMC Coefficient ± Standard Deviation for Replicates
of Extracts Correlated to:
Corresponding Standard
Neat Kerosene
0.6 ± 0.2
0.63 ± 0.11
0.71 ± 0.07
0.71 ± 0.07

173
 

containing 10% evaporated gasoline. Therefore, taken as a group, the neat and 10% evaporated
gasoline extracts cannot be assigned to one standard over the other. However, if the extracts
were analyzed individually, some extracts could be correctly identified, such as the ILR of neat
gasoline positioned closer to the standard than the other extracts.
In the carpet study, thermal degradation samples containing neat and 10% evaporated
gasoline were more strongly correlated to the 90% evaporated gasoline standard than the
respective standards. This strong correlation was due to the loss of the volatile compounds in
gasoline, namely toluene and the C2-alkylbenzenes, during the burning process, which mimicked
the evaporative loss of these compounds. However, this is not the case when HDPE is the
matrix. Due to the addition of the C12-C16 normal alkanes from the burned matrix in the
chromatograms of these samples, the extracts are not correlated to the 90% evaporated gasoline
standard.
Extracts containing 90% evaporated gasoline show strong to moderate correlations with
the corresponding standard. The strong correlations observed are reflective of the extracts’ close
positioning to the standard on the PCA scores plot. The moderate correlations observed are for
the extract that had a shift in the retention times of the C3-alkylbenzenes described previously.
The neat and 70% evaporated kerosene thermal degradation samples have strong to moderate
correlations to the corresponding standards, while extracts containing 10% evaporated kerosene
show strong correlations to the standard (Table 4.10). The moderate correlations for the neat and
70% evaporated kerosene extracts are due to differences in peak widths of the normal alkanes
previously discussed. Additionally, the neat kerosene ILRs show moderate correlation to the
neat gasoline standard, while the 10% and 70% evaporated kerosene ILRs show weak

174
 

Table 4.10: Comparison of mean Pearson product moment correlation coefficients for kerosene
ignitable liquid residue extracts (n=15) correlated to kerosene standards at each level of
evaporation and neat gasoline standard (n=225).

ILR in Extract

Mean PPMC Coefficient ± Standard Deviation for Replicates of Extracts
Correlated to:
10% Evaporated 70% Evaporated
Neat Kerosene
Neat Gasoline
Kerosene
Kerosene
Standard
Standard
Standard
Standard

Neat Kerosene

0.82 ± 0.05

0.80 ± 0.05

0.50 ± 0.05

0.51 ± 0.08

10% Evaporated
Kerosene

0.85 ± 0.06

0.88 ± 0.05

0.70 ± 0.04

0.47 ± 0.05

70% Evaporated
Kerosene

0.77 ± 0.06

0.78 ± 0.02

0.76 ± 0.13

0.41 ± 0.05

175
 

correlations to the neat gasoline standard. Therefore, while the extracts containing kerosene
could not be associated to one specific evaporation level of kerosene, extracts could be identified
as not containing gasoline.

4.4

Conclusions from Plastic Matrix Study

The interferences from the unburned and burned HDPE significantly reduce the ability of
PCA and PPMC coefficients to correctly identify ILRs in sample extracts. While performing
PCA on the liquid standards ensures that the long chain alkanes are the only compounds in
HDPE that affect the positioning of the ILR extracts from the burned and unburned matrix, these
alkanes are not the only interferences from the matrix. In addition to the alkanes, the alkenes and
alkadienes from the burned matrix lower the correlation between the ILRs and the corresponding
standards, as reflected in lower PPMC coefficients. Because PCA and PPMC coefficients were
not able to successfully identify all of the ILRs analyzed, this study indicates that the
combination of these multivariate statistical procedures is not applicable to all matrices, and is
thus limited in its applicability in a forensic setting.
Generally, the addition of the matrix interferences from HDPE prevents the association of
gasoline extracts to the corresponding standard more so than the association of the kerosene
extracts. The interferences from HDPE are unlike any of the compounds found in gasoline.
Because of this, the addition of the interferences greatly affects both the positioning of the ILRs
on the scores plot as well as the correlation coefficients calculated between the ILRs and the
corresponding standards. On the other hand, the various C10-C16 normal alkanes both found
inherently in HDPE and generated during the burning of HDPE are also in kerosene. Therefore,
176
 

the addition of the matrix interferences slightly affects the visual association of the kerosene
ILRs to the corresponding standards on the PCA scores plot and the correlation between the
ILRs and standards using PPMC coefficients, and the ILRs can be identified as kerosene.
Overall, the combination of PCA and PPMC coefficients could not be used to identify all
of the ILRs extracted from the plastic matrix. However, the combination of these two statistical
procedures was useful in identifying ILRs extracted from carpet. This result indicates that
further research should focus on the application of these procedures to a variety to matrices to
investigate the potential use of these statistical procedures in a forensic setting.

177
 

References

178
 

References

1. Stauffer E. Concept of pyrolysis for fire debris analysts. Science and Justice 2003; 43(1): 2940.

179
 

Chapter 5 – Conclusions and Future Work

5.1 Summary of Research

5.1.1 Research Objectives and Goals
This research aims to develop an objective method for associating ignitable liquid
residues (ILRs) to corresponding ignitable liquid standards using a combination of two
multivariate statistical procedures: Pearson product moment correlation (PPMC) coefficients and
principal components analysis (PCA). PPMC coefficients are used to place a numerical value on
the similarity of two chromatograms, while PCA is used to discriminate between samples by
identifying the greatest source of variance among the samples, allowing for both discrimination
and association. The combination of two statistical procedures was used to maximize the
potential of successful associations between ILRs and neat liquids, while enabling a statistical
measure of those associations. Data pretreatment procedures, namely background subtraction,
smoothing, retention time alignment, and normalization, were used to minimize non-chemical
sources of variance among the samples, ensuring that the greatest sources of variance are due to
the chemical compositions of the ignitable liquids.
The first step of this research was to investigate interferences caused by common
household matrices. Unburned and burned samples of two matrices, nylon carpet and high
density polyethylene (HDPE), were analyzed to identify the interferences both inherent to the
matrix and those generated by burning. The burned matrix samples were prepared by charring
the matrix with a propane blow torch for times ranging from 10 to 120 seconds. The unburned
and burned matrix samples were placed in separate nylon bags and extracted using a passive

180
 

headspace extraction procedure with activated carbon strips, which were then eluted and
analyzed by gas chromatography-mass spectrometry (GC-MS). The burn time for future
experiments was chosen as the time that generated the maximum amount of matrix interference,
based on a visual examination of the chromatograms.
Next, the effect of matrix interferences on the association of ILRs to the corresponding
liquid was investigated. In this study, liquid standards of neat gasoline and kerosene, as well as
each liquid at two different evaporation levels, were prepared. The evaporated liquids and the
neat liquids were spiked onto separate unburned and burned subsamples of each matrix. Spiking
the unburned matrix allowed for the investigation of the effects of interferences found inherently
in the matrix, while spiking the liquids onto the previously burned matrix ensured that only the
effects of evaporation were being investigated, with no contribution from thermal degradation of
the ignitable liquid during the burning process.
A combination of the multivariate statistical procedures provided a statistical measure of
the association and discrimination of the sample extracts to the corresponding neat liquid. PCA
was used to calculate scores and loadings for the data set. The loadings plots were used for the
identification of the greatest sources of chemical variance among the liquid standards. The
scores plot was generated using the neat liquids and the scores for the extracts of the unburned
and burned samples were calculated and projected onto the scores plot. Projecting the scores
ensured that any distinctions made between the samples were based on chemical composition of
the ignitable liquids alone. Therefore, only matrix interferences that eluted at the same retention
time as compounds in the ignitable liquids affected the positioning of the extracts. The scores
plot was then used to determine if the matrix interferences prevented the association of the
extracted ILRs samples from the unburned and burned matrices to the corresponding neat liquid.

181
 

In addition to PCA, PPMC coefficients were used to assess the association of the ILRs
from the unburned and burned matrices to the liquid standards. While PCA is used to evaluate
those variables that contribute the greatest source of variance among samples, PPMC coefficients
describe the similarities between pairs of samples using all of the variables. Thus, the
combination of the two procedures was used to determine the association of the sample extracts
to the ignitable liquid standards despite the presence of matrix interferences and evaporation of
the liquid.
In addition to the effects of matrix interferences and evaporation, the effects of thermal
degradation of both the liquid and the matrix on the association of an ILR to the corresponding
ignitable liquid standard were investigated. Fire debris was simulated by spiking the neat and
evaporated ignitable liquids onto separate samples of each matrix. The matrices were then
burned to generate significant matrix interferences. The simulated ILR was extracted and
analyzed as previously described. PCA and PPMC coefficients were again used to determine if
the ILRs could be associated to the corresponding liquid standard, despite evaporation of the
liquid, as well as the presence of matrix interferences and thermal degradation products.
By making the analysis of fire debris objective, the chance of a misidentification of the
presence or the absence of an ignitable liquid is minimized and no longer dependent on the
experience of the analyst. In addition, this research used the full chromatograms, as opposed to
other procedures reported in the literature that use only portions of the chromatogram, to
associate ILRs to ignitable liquid standards despite the presence of matrix interferences,
evaporation effects, and thermal degradation. Thus, a more complete analysis of the data was
used to maximize the probability of a correct association between and ILR and an ignitable
liquid while minimizing the subjectivity in that analysis.

182
 

5.1.2 Carpet Matrix Study Summary
Matrix interferences found inherently in carpet, and also generated by during the process
of burning the carpet, change the peak patterns in the chromatograms of ILRs. Additionally,
evaporation of the ignitable liquid reduces or eliminates some of the characteristic compounds
found in the ignitable liquids, thus making the identification of the corresponding ILR more
challenging. The effect of thermal degradation during the burning process is a combination of
these two effects in that thermal degradation of the ignitable liquid can mimic evaporative loss of
the compounds in the liquid while degradation of the matrix adds compounds to the
chromatogram.
Spiking ignitable liquids onto unburned and burned carpet demonstrated that despite
interferences from the matrix, extracts could still be associated with their respective standards
both visually and statistically. Simulating fire debris demonstrated that thermal degradation of
both the ignitable liquid and the carpet matrix affect the positioning of ILR extracts on the PCA
scores plot. Kerosene extracts could still be associated with their respective standards both
visually and statistically, though the gasoline extracts were more closely associated with the most
evaporated gasoline standard. PCA was also able to provide a visual association of extracts to
standards despite the presence of evaporation, matrix interferences, and thermal degradation.
PPMC coefficients were used to qualify these associations as moderate to strong. Generally,
residues in the sample extracts from each part of the carpet matrix study were correctly identified
as being either gasoline or kerosene, but not necessarily be assigned to the correct evaporation
level of those ignitable liquids.
Analysis of the data indicated that the majority of the spread observed in the scores plot
between extracts and corresponding standards was due to misalignments and normalization

183
 

issues in the chromatograms of the samples. This suggests that with further optimization of the
data pretreatment procedures, stronger associations between the ILR extracts and the
corresponding standards could be possible.

5.1.3 Plastic Matrix Study Summary
As was seen with the carpet matrix study, the interferences found both inherently in
HDPE and those generated during thermal degradation of the matrix drastically change the visual
appearance of chromatograms of ignitable liquid residues. When the ignitable liquids were
spiked onto unburned HDPE, the hydrocarbons from the plastic significantly affected the
compounds present in the TICs of sample extracts. While a visual association of extracts to the
corresponding standard could be possible on a PCA scores plot, the addition of the matrix
interferences prevented successful statistical association of the extracts to their corresponding
standards using PPMC coefficients.
Additionally, it was observed that matrix interferences from burned HDPE significantly
affect the compounds present in the TICs of sample extracts, even more so than the interferences
found inherently in the matrix. However, despite the addition of these interferences, correct
associations of extracts to the type of ignitable liquid present (i.e., gasoline versus kerosene)
were possible using both the scores plot and PPMC coefficients, although the ILRs were not
necessarily associated to the correct level of evaporation of that ignitable liquid.
When thermal degradation was taken into account, visual and statistical association of
90% evaporated gasoline extracts and all kerosene containing extracts to the standards was
possible, though again the kerosene extracts could not be identified as containing one
evaporation level of kerosene over the others. However, extracts containing neat and 10%

184
 

evaporated gasoline could not be identified as containing gasoline or kerosene to the exclusion of
the other liquid, due to the loss of the characteristic gasoline compounds and the addition of
matrix interferences.
In total, the combination of PCA and PPMC coefficients was not able to successfully
identify all residues of ignitable liquids either spiked onto unburned HDPE or those that were
subjected to thermal degradation during burning. However, this combination of statistical
procedures was successful when carpet was used as a matrix. In that study, the interferences
from the unburned and burned carpet had very little effect on the statistical association of
extracts to the standards, especially using PCA. On the other hand, the HDPE matrix generates a
greater variety of interferences at much higher abundances than the carpet matrix. Also
complicating the analysis, HDPE contains long chain normal alkanes, which are found in
kerosene. The addition of these alkanes affects the appearance of the chromatograms of samples
containing gasoline more so than the samples containing kerosene, possibly resulting in incorrect
identifications of the ILRs present in sample extracts.
Because PCA was performed on the ignitable liquid standard only, the long chain alkanes
are the only compounds in HDPE that affect analysis of the chromatograms using PCA. Thus,
only the alkanes affected the positioning of the ILR extracts on the scores plot. The alkenes and
alkadienes from the burned matrix did not contribute the scores on the PCA scores plot, but these
compounds, as well as the alkanes, did affect the calculated PPMC coefficients between the
extracts and the corresponding standards.
The addition of matrix interferences from HDPE was expected to reduce the correlation
of ILRs to the corresponding ignitable liquid standard. However, in this study, the correlation
was also reduced due to difference in peak width because of large differences in abundance. For

185
 

example, due to the high abundance of the alkane interference peaks from the matrix, the width
of the normal alkane peaks in the TICs for samples of neat HDPE spiked with kerosene differed
greatly from the kerosene standard, resulting in lower PPMC coefficients calculated between the
kerosene sample extracts and the corresponding standards. This result, along with differences in
abundance not accounted for by normalization and retention time drift observed in the sample
TICs, suggests that further optimization of data pretreatment procedures may yield stronger
correlations between the ILRs and ignitable liquid standards.
The variety of matrix interference compounds added to the chromatograms of the extracts
from the burned and unburned HDPE samples as well as the thermal degradation samples
lowered the PPMC coefficients calculated between the extracts and the corresponding standard.
These interferences considerably reduced the ability of PCA and PPMC coefficients to correctly
identify ILRs in sample extracts. Because of this, the combination of these two statistical
procedures could have limited utility in the forensic analysis of arson debris without further
investigation and optimization.

5.2 Future Work

While the combination of PCA and PPMCs could be used to identify all the ILRs
extracted from carpet, the interferences from plastic prevented the association of all of the ILRs
to the corresponding ignitable liquid standard. The success of the carpet study indicates that
these multivariate statistical procedures could be used by arson investigators. However, the
method of analysis would have to be further optimized to be more universally applicable to
identify ILRs extracted from a variety of matrices.

186
 

One way in which the method could be optimized would be to investigate the use of
extracted ion profiles (EIPs) instead of total ion chromatograms (TICs). EIPs could be used as a
filter to remove matrix interference compounds from the chromatograms of ILR extracts. For
example, the alkadiene, alkenes, and alkanes from the burned HDPE matrix would be removed
from the TIC if an aromatic EIP is used. The removal of the interference compounds could make
the TICs of ILR extracts more visibly similar to the corresponding standards, resulting in closer
positioning of the extracts to the standards on the scores plot and strengthening the correlation
between them. However, it should be noted that the use of EIPs is dependent on the chemical
composition of the ignitable liquid. For example, use of the aromatic EIP would only be useful
for ignitable liquids that have a significant aromatic profile.
The data pretreatment procedures, especially retention time alignment and normalization,
should also be optimized further. Analysis of the data in both the carpet and plastic matrix
studies indicates that a significant amount of the spread observed between extracts and
corresponding standards is due to misalignments and normalization issues in the chromatograms
of the samples. Some spread among samples is to be expected, due to variations in sample
preparation and analysis for example. However, a reduction of spread may be possible using
different alignment algorithms and normalization procedures.
Misalignments were observed in chromatograms aligned using the COW algorithm,
especially in the plastic matrix study in which several compounds eluted at similar retention
times. Other alignment algorithms include a peak matching algorithm. Unlike the correlation
optimized warping (COW) algorithm which attempts to maximize the correlation between the
target and sample chromatograms, the peak matching algorithm is used to align the apexes of
2

peaks in the chromatograms . To do so, the algorithm calculates the first derivative of the target

187
 

and sample chromatograms, and then aligns the data point before and after each of the zero
crossings in the two chromatograms. The peak matching algorithm could be used alone or in
combination with the COW algorithm to determine the best alignment for the data set.
Although a combination of internal standard and total area normalizations was used in
this research, difference in abundance of compounds in the chromatograms, particularly between
the sample extracts and the standards, resulted in spread among the extracts on the PCA scores
plot and lower PPMC coefficients. Nitrobenzene was chosen as the internal standard because it
is not found in any of the ignitable liquids or matrices used. However, when the internal
standard normalization was performed, it was observed that compounds with very early retention
times (i.e., the very volatile compounds, particularly those found in gasoline) were not well
normalized. In addition, total area normalizations are most effective at reducing differences in
1

abundance when the samples being normalized contain similar compounds . If chromatograms
with very different compounds present are normalized together, total area normalization is less
effective. For example, normalizing replicates of the neat kerosene standard with replicates of
the neat kerosene thermal degradation ILR extracts would not correct for differences in the
abundances of compounds between the two well because the thermal degradation ILRs would
have reduced abundances of compounds from the ignitable liquid (lost during the burning
process) and additional peaks from the matrix interferences.
Alternative methods of normalization that could be investigated include the use of
multiple internal standards. One internal standard would be chosen that elutes early in the
chromatogram and another chosen that elutes later in the chromatogram. The combination of
two internal standards might be able to better normalize across the chromatogram, as the early

188
 

eluting internal standard would normalize the more volatile compounds while the later eluting
standard would normalize the less volatile compounds.
Once the method of analysis has been further optimized for the two matrices presented
here, other matrices and ignitable liquids should be investigated and the results published so that
these peer-reviewed procedures and findings would be admissible in court. Other common
household matrices, such as wood, are commonly submitted to crime laboratories for analysis
and will have vastly different interference compounds than either carpet or plastic. The
successful association of ILRs to the corresponding liquid standard using the combination of
multivariate statistical procedures presented here for a number of different matrices would
demonstrate the universal applicability of the analysis method.
Additionally, this research focused on the analysis of gasoline and kerosene, two
ignitable liquids commonly used as accelerants that have very different chemical compositions.
Ignitable liquids from the other ASTM classes should also be investigated to determine if correct
identification of ILRs from these classes would be possible with PCA and PPMC coefficients.
Liquids from the same class could also be investigated to determine if the combination of
statistical procedures can be used to determine a specific ignitable liquid or if the identification is
limited to the class of ignitable liquid.
Other statistical procedures that provide a statistical measure of the error of an
identification of an ILR should be investigated as well. For example, soft independent modeling
of class analogy (SIMCA) could be used, which allows for the classification of unknown
samples into previously-defined groups. For example, a series of known ignitable liquid samples
would be used as a training set to develop a classification model based on known classes of
ignitable liquids. A test set of known ILRs would be used to validate the model, generating a

189
 

classification error associated with the model. Then, new simulated ILR samples would be
prepared and introduced to the model for classification. With the SIMCA approach, the new
samples are classified into each of the previously defined groups and probability of group
membership determined. The sample is considered a member of the group for which the highest
probability is associated. Samples may be determined to fit into one, more than one, or none of
the groups. Two of the advantages of SIMCA are that samples can be added without having to
redefine the classes each time and samples are not forced into a group, as with other
classification methods.
This research was aimed at developing a more objective method for the analysis of fire
debris, to help minimize the chance of misidentifying the presence or the absence of an ignitable
liquid. This method also provided a statistical means of assessing the association of the ILR to
the neat liquid. Such statistical associations are consistent with the recommendations made by
the recent National Academy of Sciences report for forensic science. The combination of
statistical procedures also offers arson analysts an alternative way of presenting data to a jury.
Instead of a series of complex chromatograms, the analyst could show a visually simpler scatter
plot (the PCA scores plot) with a table of numbers (PPMC coefficients) to identify ILRs in fire
debris. Thus, the same data and results could be presented but in a way that is more easily
understood by those not familiar with arson analysis.

190
 

References

191
 

References

1.  Rietjens M. Reduction of error propagation due to normalization: effect of error propagation
and closure on spurious correlations. Anal Chim Acta 1995; 316: 205-215.
2. Johnson KJ, Wright BW, Jarman KH, Synovec RE. High-speed peak matching algorithm for
retention time alignment of gas chromatographic data for chemometric analysis. J
Chromatogr A 1998; 805: 17-35.

192