You are on page 1of 15

Psychology, Crime & Law, March 2005, Vol. 11(1), pp.

1 /15

TEST OF THE ARSON ACTION SYSTEM MODEL IN AN


INCARCERATED POPULATION
LOUISE ALMONDa*, LORNA DUGGANb, JOHN SHINEc and DAVID CANTERd
a
Centre for Investigative Psychology, University of Liverpool, Eleanor Rathbone Building,
Bedford Street South, Liverpool, L69 7ZA, UK; bSt Andrew’s Hospital, Billing Road,
Northampton, NN1 5DG, UK; cHMP Grendon and Springhill, Aylesbury, Bucks, HP18 0TL,
UK; dCentre for Investigative Psychology, Eleanor Rathbone Building, Bedford Street South,
Liverpool, L69 7ZA, UK

(Received 17 March 2003; in final form 5 September 2003)

Canter and Fritzon (Legal and Criminological Psychology, 3, 73 /96, 1998) proposed a new model of
differentiating arsonists based on the action system model. This is a replication of their study utilising
interviews with 65 inmates from one prison, HMP Grendon. A smallest space analysis (SSA) supported an
action system model of four distinct modes, adaptive, expressive, integrative and conservative, reflecting
Canter and Fritzon’s four distinct themes of arson (Canter and Fritzon, 1998). Two themes relate to expressive
acts / (a) those arsons targeted at a person and (b) those at an object. The remaining two relate to
instrumental acts / (c) those targeted at a person and (d) those at an object. Using four scales, which related to
the four themes, 55 (85%) of the acts could be assigned to either one or two adjoining themes. The study gives
further support to the utility of the action system model and the model of firesetting behaviour proposed by
Canter and Fritzon, when using a different method of data collection with a sample of serious offenders.

Keywords: Arson; Action System Model; Prisoners; Smallest Space Analysis

INTRODUCTION

In 1998, Canter and Fritzon proposed an action system model that attempts to differentiate
arson behaviours based purely on behavioural indicators rather than inferred motives.
Canter and Fritzon (1998) argued that variations in types of arson might be considered
similar to the variations that can be found in other behavioural action systems. These models
are of value as they allow data to be categorised, enabling reduction, while distilling its
essence. Models also allow the salient aspects and distinct processes that are involved in
crimes to be readily identified.
Firesetting has a long-standing link with mental disorder. Lewis and Yarnell (1951)
undertook a comprehensive survey of arson, from which they developed a classification of
firesetters. Further classificatory systems were proposed (Scott, 1978; Prins et al ., 1985).
Classification has been problematic, with mental state and motivation being utilised (Prins et
al ., 1985). Research carried out by the UK Home Office estimates that only 17% of
individuals arrested for arson suffer from some form of mental disorder. Classification

*Corresponding author. E-mail: lalmond@liv.ac.uk

ISSN 1068-316X print/ISSN 1477-2744 online # 2005 Taylor & Francis Ltd
DOI: 10.1080/1068316031009634287

1
2 L. ALMOND et al .

schemes that solely utilise the mental state of the individual are therefore not applicable to
arsonists as a whole. Motivation and human intentions are usually mixed and multiple, and
there is a danger in classifying an individual purely based on their motivation, as this is not a
static process but may change with time, place or situation. Reviews of existing arson
classificatory systems (Harris and Rice, 1996) revealed that no single framework adequately
encompassed the full range of behaviour associated with the act of firesetting until Canter
and Fritzon’s 1998 model.
The action system model, as described by Shye (1985a), is defined by a collection of
members that maintain interrelationships among themselves. It is active, open, organised and
stable. Shye focused on effective system functioning rather than deviant (criminal)
behaviours. He went on to describe that for the ‘‘open’’ part of the system, it is necessary
to consider (a) the source of the action and (b) the effect or desired target of the action. Both
these aspects of the action system can be located internally or externally to the individual.
These two facets give rise to four basic modes of functioning as shown in Table 1.

1. Adaptive mode: The action system agent (person) responds to an external event by
exploiting their environment for some instrumental gain. Cases of arson committed in this
mode are expected to be opportunistic with the specific desire being to destroy the target, for
example to conceal a crime.
2. Expressive mode: The person responds to an internal psychological situation by
influencing an aspect of their external environment. This type of arson would be displayed
in the individual’s selection of a specific target, carrying either symbolic or emotional
significance.
3. Integrative mode: The person responds to an internal psychological situation by making
alterations to the system itself. This individual sets a fire as an attention-seeking device,
through which they may endanger both their own lives and the lives of others.
4. Conservative mode: The person responds to an external event by making alterations to the
system itself. This type of arson sees the individual as setting a fire in response to some
external source of frustration, in order to restore the individuals’ own state of well-being, for
example personal revenge.

Any system under study should be expected to operate in a way that indicates one of these
hypothesised modes as a dominant theme to their activities. These four modes, Shye argued,
have logical similarities and differences and can be represented geometrically as shown in
Figure 1. Thus the conservation and expressive modes are at opposite ends of the
diagrammatic representation as they are the two polar ends in the functioning of an action
system. As the adaptive mode is derivable from these two extremes, as is the integrative
mode, they maintain an affinity with both the conservation and expressive modes. Although
they are distinct from each other (adaptive concerning external adjustments and integrative

TABLE 1 Summary of action system modes of functioning (Shye, 1985a).

Source of action in Locus of effect in Mode


relation to agent relation to agent

external external adaptive


internal external expressive
internal internal integrative
external internal conservative
ARSON ACTION SYSTEM MODEL 3

FIGURE 1 The conceptual interrelationships among the functioning modes of an action system, represented
by means of geometric /spatial proximities (Shye, 1985a, p. 113).

internal adjustments) they are not as distant from each other as are the poles of conservation
and expression.
Canter and Fritzon (1998) developed a model of firesetting behaviour using this action
system model by focusing on the actions of the arsonist rather than the internal processes
that formed the basis of some previous classifications. This model has important
implications; psychologically, they were seeking to uncover the underlying processes that
lead individuals to commit arson, making it possible to classify offenders based on their acts
of arson. Practically, this model has real benefits for arson investigators as the framework
draws on behavioural indicators, which are readily available to those conducting arson
investigations.
Canter and Fritzon (1998) carried out a content analysis of the crime scene reports and
witness statements from the case files of 175 solved arson cases from across England to
produce 42 behavioural variables. The data were then subjected to a non-metric multi-
dimensional scaling procedure known as smallest space analysis (SSA-I), which examines the
relationship each variable has to every other variable. This method was employed in order to
test the hypothesis that there were four distinct themes of arson behaviour. The results of this
study highlighted three main facets underlying firesetting behaviour. The first is a
modulating focus facet that relates to the association of the target to the offender, ranging
from variables that indicate a close association, to those that describe the target themselves.
The second facet relates to the targets as either a person or object, i.e. internal or external to
the arsonist. The third facet describes the source of the action, as either expressive or
instrumental, i.e. internal or external to the arsonist.
The term ‘‘expressive’’ used by Canter and Fritzon must not be confused with the
expressive mode of Shye’s action system model. Canter and Fritzon use the term
‘‘expressive’’ to indicate external manifestations of internal processes and drives. This term
is often used in criminology and profiling literature. Shye’s use of the term relates to a slightly
different process. The expressive mode indicates that the behaviours relate to external
4 L. ALMOND et al .

manifestations of internal processes and drives, but importantly in Shye’s use of the term,
these are specifically coupled with an external target.
As the later two facets were orthogonal this gives rise to four different themes that they
argued correspond to the four modes of action system functioning. Each of these themes
should form a scale in the sense that the variables’ combined existence is a reasonable
indication of some underlying dimension. In order to test this Crohnbach’s alpha scores were
calculated for each of the set of actions that defined a region within the SSA plot.
This is an important development in the classification of arson as it presents a framework
with a unified underlying structure. In that it models sources and targets of firesetting within
the same framework, it also brings together a number of different perspectives and helps
resolve some of the apparent contradictions in previous attempts to classify arson (Harris and
Rice, 1996). The framework is also based on the notion that the various firesetting behaviours
may stem from multiple psychological origins (Santtila et al ., 2003).
Canter and Fritzon’s work differed from earlier studies in that it relied on official data
rather than data obtained directly from the arsonists. The reliability of data from offenders is
open to question, for example they may not tell the truth or have a poor recollection of events
(Green, 1990). Canter and Fritzon demonstrated reliable structures using carefully
constructed content analysis dictionaries.
Fritzon et al. (2001) went on to collect data relating to a larger sample of 230 solved arson
cases, using the same methodology and analysis utilised in Canter and Fritzon’s (1998) earlier
study. The researchers found evidence to support the action system framework, which gave
rise to the four themes to arson similar to those identified by Canter and Fritzon (1998).
Using Fritzon et al. ’s (2001) original data, Santtila et al. (2003) identified 66 juvenile
firesetters (aged 6 /17 years). The researchers found four different themes of juvenile
firesetting similar to those identified by Canter and Fritzon (1998). In addition, the high
frequency variables were similar in both studies indicating that the basic nature of firesetting
emerges early.
Whenever a model is constructed it is important to test its utility in other settings. The
studies carried out by Canter and Fritzon (1998), Fritzon et al . (2001) and Santtila et al.
(2003) all utilised the content analysis of solved police cases and therefore certain types of
arson may be over-represented, i.e. those that are easily detected and thus solved. This study
contains a more serious sample of offenders and information will be obtained from interviews
with incarcerated arsonists as opposed to police records, in order to examine the effect this has
on the model. If, however, the model is applicable, this alternative method of data collection
and new dataset should not have a significant effect on the themes of behaviour found.

Her Majesty’s Prison Grendon


HMP Grendon in Aylesbury, UK is a category B male prison, housing serious offenders
serving long sentences. It consists of five separate residential therapeutic communities and
contains approximately 200 men. Descriptions of the treatment regime can be found in
Cullen (1995) and Genders and Player (1995). Prison medical officers based in closed prisons
throughout England and Wales refer inmates. The inmates, however, have to agree to
transfer. Most are diagnosed as having a personality disorder. Those with impaired
intellectual functioning or psychosis are excluded. The inmates of HMP Grendon are
serious offenders as evidenced by approximately 40% serving life sentences. Most have been
convicted of offences against the person. Upon arrival each inmate participates in a
ARSON ACTION SYSTEM MODEL 5

comprehensive psychological assessment relating to his personality, background and


criminal history.
The study’s first aim is to replicate Canter and Fritzon’s work using data from inmates, to
examine whether the four modes of action system are present in this sample, and if so do
their relative similarities and differences take the form proposed by Shye as shown in Figure
1. The second aim is to compare the model created by this study’s sample of serious arsonists
with the model proposed by Canter and Fritzon (1998).

METHOD

Procedure
The HMP Grendon and Springhill Research Committee granted permission for the study.
Details of 65 cases of arson were identified from HMP Grendon’s database. This database
includes information regarding current and previous offences. The sample therefore
contained inmates with either a current or previous conviction for arson. Some had both
a previous and current conviction. Using the 42 offence variables identified by Canter and
Fritzon (1998), the first author completed interviews with identified inmates after gaining
their consent to take part in the project. For those inmates who had set more than one fire,
the questioning focused on the most recent fire. Data were available for 10 inmates who
fulfilled the criteria but were no longer resident in the prison.
Every inmate approached agreed to take part in the study and many expressed an interest
in the results. The nature and purpose of the study was explained, in particular that the
questions would relate to a fire for which they had been convicted and consent was obtained.
The inmate was asked about the fire in question (for those with more than one conviction the
most recent fire was chosen for the purpose of clarity) and after a short description, the
interviewer proceeded to ask direct questions relevant to the 42 offence variables. The
interview lasted approximately 5 minutes. Data from 10 inmates were coded from
information held at the prison, as they were no longer resident in the prison. Verification
of information given at interview was impossible as the researchers tried, without success, to
obtain further information, in the form of witness statements, from the courts where the
individuals were convicted.
Each variable was coded as present or absent. This dichotomous approach to the content
analysis was used to ensure maximum clarity and reliability and replicates the approach of
Canter and Fritzon. The content dictionary used describes the 42 offence variables and is
identical to the content dictionary used by Canter and Fritzon (1998). Coding the variables
as either 1 / variable present or 0 / variable absent, across the 65 cases, produced a data
matrix.

Sample
All of the inmates were male. On reception to HMP Grendon their ages ranged from 22 to 46
years, with a mean of 30.8 years. Thirty were serving a current sentence for an offence that
involved arson; the remaining 35 had a previous conviction of arson. Life sentences were
being served by 25 of the men. The mean determinate length of sentence for the remaining 40
men was 7 years, ranging from 4 to 20.5 years. The inmates had on average 13 previous
convictions, ranging from none to 39.
6 L. ALMOND et al .

Statistical Analysis
In order to test the relationship each of the variables had to every other variable an association
matrix was produced, using Jaccard’s coefficient of association, which only takes account of
positive co-occurrence. If two variables are both absent from the records this does not increase
the association, as it is never certain that absent information was just not recorded.
A geometric representation was generated of the relationship in the association matrix
using smallest space analysis (SSA I), as used by Canter and Fritzon (1998). The algorithm
attempts to find the best representation such that the higher the correlation between any two
variables, the closer together the points representing them will be. SSA operates on the ranks
of the distance between the points and the ranks of the association coefficients. It thus
captures the relative size of associations and is therefore the most appropriate for examining
dominant themes in the present form of data.

RESULTS

None of the 65 cases contained the variable ‘‘suicide note’’ and thus we eliminated it from the
analysis. A three-dimensional SSA solution was found to have the reasonable Guttman-
Lingoes coefficient of alienation of 0.15 in 15 iterations. This coefficient is a measure of how
well the analysis has managed to ‘‘fit’’ the spatial representation of all the inter-relationships
among variables into the three-dimensional space. Figure 2 shows the first two vectors of the
three-dimensional space. In this figure each point represents an aspect of the arson derived
from the content dictionary as published in Canter and Fritzon (1998). The closer any two
variables are in Figure 2 the more likely it is that when one occurs in an offence, the other will
also occur. The SSA plot as shown in Figure 2 can be partitioned into four regions,
containing variables that reflect only one mode of functioning.
If Shye’s hypothesis is to be supported then the relative mutual orientation of the variables
in the SSA should reflect the functioning modes outlined in Figure 1.
Adaptive mode : This mode of functioning is the way in which the system interacts with
external events by modifying their external environment. Variables contained within this
mode lie in the bottom right side of the plot. These can be seen as opportunistic in that the
crime may only have occurred during other illegal activities. The act of arson being
instrumental in achieving other criminal goals, the mode therefore includes variables such as
‘‘more than one offender’’ involved, ‘‘business’’, ‘‘vehicle’’ and ‘‘theft’’.
Expressive mode : This mode reflects the arsonist responding to some internal process by
modifying their environment. Variables contained in this mode lie in the top right side of the
plot. Although the target is property, such as ‘‘school’’ and ‘‘public buildings’’, the forces
motivating the fire are internal; this fascination with fire is reflected in variables such as
‘‘remaining at scene’’, ‘‘multiple items’’ and ‘‘seats of fire’’, ‘‘previous convictions for arson’’,
‘‘serial’’ and ‘‘spree’’.
Integrative mode : This mode reflects the arsonists being internally distressed and
directing the arson internally at themselves. Variables contained in this mode lie in the top
left side of the plot. The arsonist is attempting to alleviate stress or sets fires as a cry for help;
this is reflected in variables such as ‘‘self’’, ‘‘own home’’, ‘‘hospital’’ and during the
‘‘daytime’’.
Conservative mode : This mode reflects the arson being a form of retaliation against
another person. Variables contained in this mode lie in the bottom left side of the plot. The
ARSON ACTION SYSTEM MODEL 7

FIGURE 2 A two-dimensional SSA showing action system themes in arson actions. Numbers indicate
content analysis variables as listed in the key.

arson occurs following an event involving another person, which results in the arsonist
wishing to take revenge. The variables in this mode therefore include ‘‘victim known’’, ‘‘prior
threats of violence’’ or ‘‘arson’’, ‘‘arguments’’, ‘‘planning’’ and ‘‘specific trigger’’.

Relationship among Regions


Further support for the link between the themes of arson and action system modes of
functioning comes from the positioning of the regions of the SSA. The positioning of the
four regions is indicated with the core variables that help to define that region most precisely.
8 L. ALMOND et al .

The adaptive mode (‘‘vehicle’’ variable number 6) is located opposite the integrative mode
(‘‘self’’ variable number 8), and the expressive mode (‘‘spree’’ variable number 29) is located
opposite the conservative mode (‘‘specific trigger’’ variable number 40). This is the pattern
hypothesised by Shye (1985a).

Focal Aspect of Arson


The focus facet hypothesised by Canter and Fritzon’s model reasoned that there will be
aspects of arson that will indicate some associations between the arsonist and the target, and
these will be distinct from the actions that relate to the nature of the target. In terms of the
SSA plot, this would be revealed by the associations between the arsonist and the target
being found in the centre of the configuration; these will also be the most frequent variables.
Aspects that relate to the target, Canter and Fritzon argued, should be low frequency
variables and therefore be positioned at the periphery of the plot.
The contours in Figure 3 distinguish the variables that occur in 2 /19%, 20 /39%, 40 /59%,
and more than 60% of cases. Identical frequency contours were reported in Canter and
Fritzon’s 1998 study. Slight differences exist between the variables present in each of the
contours.
As would be expected from a sample of serious offenders, this sample has a higher number
of ‘‘previous arsonists’’, the variable occurring in 63% of cases, compared with 26% in Canter
and Fritzon study. The variable ‘‘set fire’’ occurs in this sample in 37% of cases, whereas in
Canter and Fritzon’s study the frequency for this variable was much higher, 84%. This may
be explained by the difference in data collection procedures or differences between the
samples.
Although the actual frequencies of the variables differ in the two studies, the underlying
structure is the same. Variables involving the offender and the target graduate from those
focused on achieving destructive objectives to those that are specific to the interactions with
the targets in question. The data, therefore, have the same underlying meaning behind the
frequency structure as Canter and Fritzon’s model.

Themes of Arson
Source of Arson
Figure 4 shows that the SSA plot can be partitioned to discriminate between expressive and
instrumental actions, i.e. variables that are internal or external to the arsonist. The expressive
facet is evident in the top left half of the plot whilst the instrumental facet is evident in the
bottom right half of the plot. The brackets refer to the positioning of that variable in the
Canter and Fritzon model and variables denoted by stars are those that are in a different
region than that found in Canter and Fritzon’s model.

Expressive. The top half of the plot contains a number of variables that indicate the
expressiveness of the arson such as ‘‘non-specific triggers’’ and attacks on the ‘‘arsonists
themselves’’. Interestingly, ‘‘public buildings’’ and ‘‘hospitals’’ are in this region. Canter and
Fritzon argued that this suggests that these targets are selected because of their emotional
significance. Four variables from this study / ‘‘school’’, ‘‘spree’’, ‘‘set’’ and ‘‘alcohol’’ / were
ARSON ACTION SYSTEM MODEL 9

FIGURE 3 A two-dimensional SSA showing frequency contours. Numbers indicate content analysis
variables as listed in the key. Parentheses contain the percentage frequency of each variable.

positioned within the expressive region whilst in Canter and Fritzon’s model they were
positioned within the instrumental region. It can be argued that the variable ‘‘school’’ may be
more expressive as it may hold some emotional significance to the arsonist rather than as a
form of arson for material gain. The variable ‘‘spree’’ can also be argued as more expressive
as the arsonist is setting more than one fire and this may possibly reflect an expressive act.
The frequency of the variable ‘‘set fire’’ may reflect the different data collection procedures.
Conversely, to actually set fire to something as opposed to throwing a missile may be
hypothesised as more expressive, as it involves more interactions between the arsonist and
the fire. If the offender is intoxicated with alcohol this may have a significant effect on their
emotions and therefore may be seen as a more expressive variable.
10 L. ALMOND et al .

FIGURE 4 A two-dimensional SSA showing themes in arson offence behaviours. Numbers indicate content
analysis variables as listed in the key. Stars represent variables that are in a different region to that contained in
Canter and Fritzon’s (1998) model.

Instrumental. The bottom half of the plot contains a number of actions that represent the
instrumentality of the arson: ‘‘vehicles’’ set alight, the arson being part of ‘‘threatening
activity’’, as well as arson being part of a ‘‘theft’’. These all accord with Canter and Fritzon’s
proposal that the arsonists are trying, essentially, to create a change in their environment.
Three variables / ‘‘business’’, ‘‘lives deliberately endangered’’ and ‘‘drugs’’ / were positioned
within the instrumental region, whilst in Canter and Fritzon’s model they were positioned
within the expressive region. The variable ‘‘business’’, it can be argued, is more instrumental
as it may represent arson for profit. This sample of serious arsonists is willing to ‘‘endanger
ARSON ACTION SYSTEM MODEL 11

lives deliberately’’ for an instrumental outcome and this may be imperative if they are ‘‘drug’’
users as is shown in the position of these variables.

Locus of Effect
Figure 4 also shows how the SSA plot can be partitioned to discriminate between whether
the target is a property, external to the arsonist, or some form of person that may be part of
the arsonist’s self-identity, therefore internal. The object facet is evident in the top right half
of the plot whilst the person facet is evident in the bottom left half of the plot. The brackets
refer to the positioning of the variable in Canter and Fritzon’s model. Variables denoted by
stars are variables that are in a different region than that found in Canter and Fritzon’s
model.

Object. To the right of the plot are the variables ‘‘business’’, ‘‘public building’’ and
‘‘miscellaneous/uninhabited property’’ / these indicate that the target of the offence was an
object. Other variables are associated with the nature and benefits that the arson provides for
the arsonist. These include ‘‘illegal entry’’ and ‘‘theft’’ from the premises. These variables all
accord with what would normally be considered property crime. Two variables /‘‘vehicle’’
and ‘‘multiple seats’’ fired / were positioned in the object region whilst in Canter and
Fritzon’s model they were positioned within the person region. The variable ‘‘vehicle’’ is
obviously an object and is therefore positioned in the appropriate region. The variable
‘‘multiple seats’’ fired can be seen to reflect a more object target as it involves the arsonist
making sure an object is alight.

Person. The left hand of the plot contains the variables: ‘‘victim known’’, ‘‘self’’ and ‘‘victim
partner or ex partner’’. This indicates that a person was the likely target of these arsons and
typically they were people who had some significance to the arsonist. Five variables were
positioned in the person region whilst in Canter and Fritzon’s model they were positioned
within the object region: ‘‘hospital’’, ‘‘daytime’’, ‘‘set’’, ‘‘non-specific trigger’’ and ‘‘drugs’’.
The variable ‘‘hospital/institution’’ variable may be positioned within the person region, as it
may be part of the arsonists’ personal or social identity if they have spent time within a
hospital/institution. The variables ‘‘daytime’’, ‘‘set’’ and ‘‘non-specific trigger’’ show that
these arsonists are intent on setting the fire. Due to the nature of this sample it may well
contain a higher number of ‘‘drug’’ users than Canter and Fritzon’s sample / this explains
the differing positions.

Modes of Arson
Because the two primary facets of source and target were orthogonal, they create four
subsets of the form proposed by Canter and Fritzon (see Figure 4). Variables denoted by
stars are those that are in a different theme to that reported by Canter and Fritzon.

Scales of Arson
The set of action within each region should form a scale. In order to test this we calculated
Crohnbach’s alpha scores for each of the set of actions that defined a region (see Table 2).
Actions that occurred in 60% or more of cases were not used as they are core variables and
therefore occur in the majority of cases, so it would not be possible to differentiate between
12 L. ALMOND et al .

TABLE 2 Scales of actions.

Expressive object Expressive person Instrumental object Instrumental person

SERial Own HOME THEFT THReat Violence


CIVIC SELF FORCE THReat Arson
Multi ITEMS RESidential MORE 1 offender MATERials
REMained Lives end. LOCATion MISC property ARGUment
PRior HOSPital VEHicle Trigger SPECific
Multi SEATS Trigger NON-SPecific BUSIness Victim Known
Alpha/0.5590 Alpha/0.5706 Alpha/0.2489 Alpha/0.81
SCODIS/0.9986 SCODIS/0.9892 SCODIS/0.9983 SCODIS/0.9986

arsons using these variables. Only pure variables were included, as these were deemed to be
more reliable, i.e. pure variables were defined as variables found in the same region in both
the current study and Canter and Fritzon’s study. The same number of actions was used in
each scale, allowing comparisons to be more readily produced.
Three of the alpha values in Table 2 are reasonable and are very similar to those found by
Canter and Fritzon. The Instrumental Object scale has a very low alpha value / this may be a
product of the low frequencies involved.
In order to confirm that the variables chosen form a scale, an alternative measure of
scalability was also used, the SCODIS. In order to determine how well a two-dimensional,
partially ordered configuration fits the observed pattern of profiles of scores for each of the
members of the population under study, Shye (1985b) proposed the criterion of ‘‘the
proportion of profile pairs, out of all profile pairs, whose empirically observed order relation
is correctly represented by their two co-ordinates’’ (Shye et al ., 1994, p. 155). This measure,
the score distance weighted coefficient known as SCODIS, is therefore an indication of how
well the measured variables form a two-dimensional scale. As a consequence, this is an
appropriate indication of the homogeneity of a set items taken from a region of an SSA
configuration. All four scales had high SCODIS values.

Assigning Cases to Themes


Although the SSA indicates that the behaviour of arsonists can be classified in terms of four
psychologically meaningful themes, it does not classify the arsonist. The crime-scene actions
of any one individual may contain variables from more than just one SSA region, although
because these regions represent psychologically distinct modes of action, it would be
expected that the majority of behaviours would fall into one particular region. We examined
whether it was possible to classify individual cases as belonging to one of the four categories
of arson.
Each case was given a score consisting of the number of variables present in each theme.
Individuals were classified as belonging to a particular theme if the score for that theme
contained a greater number of variables than any other. Additionally, a case was considered
to be a hybrid between two themes if it contained approximately the same number of
variables for each of those themes. A case was not classified if it contained an equal number
of variables from more than two themes.
Using this method, almost all the cases (85%) could be classified as either pure types or a
hybrid of two types. The most frequent pure type was the Instrumental Person theme,
representing 13 cases. This was followed by Instrumental Object, (n/11), Expressive Person,
ARSON ACTION SYSTEM MODEL 13

(n/5) and Expressive Object, (n/3). Five hybrids were identified, the most common being
Expressive Object with Expressive Person (n/8) followed by Instrumental Person and
Expressive Person, (n/5), Expressive Object with Instrumental Object, (n/4) and
Instrumental Object with Instrumental Person, (n/3). These are all combinations which
are supported by the action systems framework in that they are all in adjoining regions.
Three cases, however, appeared to be a hybrid between Instrumental Person and Expressive
Object. Inspection of the variables contained in those regions suggests that this arson was
committed for insurance fraud. Business owners may hire individuals because of their known
proclivity to set fires.
The fact that the majority or arson cases belong to one of the four main behavioural
themes is empirical support for the hypothesis that the SSA structure represents the
dominant processes underlying arson behaviour and this structure is similar to that found by
Canter and Fritzon (1998).

CONCLUSION

This study has shown that the action system model is productive when examining firesetting
behaviour. In addition to the fact that the four different modes of functioning were reflected
in the different types of arson, the relative similarities and differences between these types
also took the form proposed by Shye. The arson action system model has therefore been
shown to be applicable to this sample of serious offenders. More studies, however, are
required utilising different sources to further strengthen the applicability of the model.
The model proposed by Canter and Fritzon (1998) has been almost entirely replicated with
a entirely different sample using only the content dictionary published in a journal. Except
for some variables, which are in different regions, the majority comply with the arson action
systems framework identified by Shye (1985a) and therefore Canter and Fritzon’s model.
Variables only differed in either one of the facets. The results of this study add weight to the
validity of the model as we obtained information from arsonists themselves as opposed to
inferring the presence/absence of variables from official records. The only variable, however,
that differed notably in its frequency was ‘‘set fire’’; therefore it could be argued that all the
other variables were correctly inferred. This may reflect a problem with data collection,
recollection of the arsonist or an unstable factor. Several variables including ‘‘vehicle’’ and
‘‘business’’ seem to be positioned in more appropriate regions in this study than in Canter
and Fritzon’s model, and in fact these two variables in a more recent article by Fritzon et al .
(2001) are contained within the regions depicted by this study. Other variables such as
‘‘drugs’’ and ‘‘daytime’’ differ in position due to the severity of the sample. The variable
‘‘suicide note’’ was not present in this sample; however, this could have been predicted as
these individuals would either be dead or would not be transferred to HMP Grendon due to
psychiatric disturbance.
The fact that some of the variables reside in different regions to that found by Canter and
Fritzon’s 1998 model may depend on either the data-gathering method, differences between
the samples or a mixture of the two. There is also the real possibility that some behaviours
may have different meanings for different samples. Any of these explanations may cause the
discrepancies found between the two studies.
Crohnbach’s alpha scores were lower in this study than in Canter and Fritzon’s study;
however, this statistic is very sensitive to the number of items and their frequencies and in this
14 L. ALMOND et al .

study each scale was represented by only six variables. When an alternative measure of
scalability was used, the score distance weighted coefficient (SCODIS), the reliability of the
scale increased dramatically.
The fact that the majority of arson cases could be assigned to either one or a hybrid of two
adjoining themes is empirical support for the hypothesis that the SSA structure represents
the dominant underlying arson behaviour. The pattern of hybrids found supports the notion
of describing offenders in terms of their predominant themes of arson behaviour, i.e. cases
could be described as expressive (without specifying the target as either a person or an
object).
To test further the validity of these four themes the typical characteristics of people within
this sample who have committed the different types of arson needs to be examined, in order
to test the hypothesis that individuals who set fires according to a particular mode of action
will also differ in their characteristics, as was found to be the case in Canter and Fritzon’s
study.
Future directions for the study of arson may include interviewing offenders about their
feelings and emotions about both the fire itself and the targets involved, for example enabling
psychological enquiry such as why fire was chosen as a means of expression to be
investigated, further strengthening the arson model. The action system model has provided a
framework that is applicable to arsonists: it is therefore plausible that the model will also be
relevant to other forms of criminal activity. Future research is needed to determine whether
categories of various crimes that exist may be applicable to this framework.
The results of this study and Canter and Fritzon’s study have both theoretical implications
in terms of enriching our understanding, and direct implications for the treatment of
arsonists. Different treatment programmes would be appropriate for the different forms of
arson, for example aimed at changing the arsonist’s view either of themselves or their targets.
These studies also extend to the investigation of fires. For example, it is likely that offenders
responsible for arson to public properties will be known to police for previous firesetting
offences.

References
Canter, D. and Fritzon, K. (1998). Differentiating arsonists: a model of firesetting actions and characteristics.
Legal and Criminological Psychology, 3, 73 /96.
Cullen, E. C. (1995). ‘G’rendon: a therapeutic prison that works’. Therapeutic Communities, 15, 301 /311.
Fritzon, K., Canter, D. and Wilton, Z. (2001). The application of an action system model to destructive
behaviour: the examples of arson and terrorism. Behavioural Science and Law, 19, 657 /690.
Genders, E. and Player, E. (1995). Grendon : a Study of a Therapeutic Prison . Oxford: Clarendon Press.
Green, G. (1990). Resurrecting polygraph validation of self-reported crime data. A note on research method
and ethics using the Deer Poacher. Deviant Behaviour, 11, 131 /137.
Harris, G. T. and Rice, M. E. (1996). A typology of mentally disordered firesetters. Journal of Interpersonal
Violence, 11, 351 /363.
Home Office (1999). Safer Communities: Towards Effective Arson Control. The Report of the Arson Scoping
Study. Home Office, London.
Prins, H., Tennant, G. and Trick, K. (1985). Motives for arson (fire raising). Medicine, Science and the Law, 25,
275 /278.
Lewis, N. D. C. and Yarnell, H. (1951). Pathological fire-setting (pyromania). Nervous and mental disease
monograph, no. 82. Coolidge Foundation, New York.
Santtila, P., Hakkanen, H., Alison, L. and Whyte, C. (2003). Juvenile firesetters: crime scene actions and
offender characteristics. Legal and Criminological Psychology, 8, 1 /20.
Scott, D. (1978). The problem of malicious fireraising. British Journal of Hospital Medicine, 19, 259 /263.
Shye, S. (1985a). Nonmetric multivariate models for behavioural action systems. In D. Canter (Ed.), Facet
Theory Approaches to Social Research. Springer Series in Social Psychology (pp. 97 /148). New York:
Springer Verlag.
ARSON ACTION SYSTEM MODEL 15

Shye, S. (1985b). Multiple Scaling: the Theory and Application of Partial Order Scalogram Analysis.
Amsterdam: Elsevier Science.
Shye, S., Elizur, D. and Hoffman, M. (1994). Introduction to Facet Theory: Content Design and Intrinsic Data
Analysis in Behavioral Research . Applied Social Research Methods, vol. 35. Thousand Oaks: Sage
Publications.

You might also like