Bachelor Script I e Socio Logie 2

Running head: PREDICTIVE PATTERNS OF SEX OFFENDERS 1
Concept Version
Predictive Patterns of Sex Offenders
Crime Trajectory Analysis Prior to the First Sex Offence
Diane Groothuysen (3788156)
Utrecht University
Abstract
Bachelor thesis Sociology (201100018)
Thesis supervisor: dr. J. E. H. Beijers
Course coordinator: prof. I. Maas
Special thanks to: prof. dr. A. Blokland

PREDICTIVE PATTERNS OF SEX OFFENDERS 2
1.Introduction
Sex offending is a serious offence that leads to a lot of physical and psychological
damage among its victims. In the Netherlands, numbers of police-registered sexual offences
have declined from 9.000 to 2.000 per year between 2010 and 2014 (Politie.nl, 2017).
However, the number of reported sex offences is higher than the registered number (7.039 in
2014) and it may be assumed that the actual numbers of sex offences are even higher since
many offences never get reported. It is important for the wellbeing of potential victims as well
as the society as a whole to decline the number of sex offences as much as possible using
policy instruments. Dutch policies regarding sex offenders are different from regular
offenders. For example, when a person wants to apply for a certificate of conduct, the
applicant will be checked for offences he or she committed in the past 2-4 years (with some
special cases for specific jobs that can take up to 30 years). This certificate of conduct is a
necessary document for many jobs in the Netherlands. However, sex offences are treated
differently. These offences will be taken into account for the rest of the offender’s life for the
application of a certificate of conduct. This means the sex offender will have lifelong troubles
receiving the certificate of conduct. ("Terugkijktermijnen", n.d.)
This policy was made for several reasons (Boone, 2011); one of them was that policy
makers aimed at protecting the public from sex offenders by not letting them occupy a job
based on any kind of dependency. However, this policy could also entail some negative
consequences. A lot of criminological research shows that employment is negatively
associated with recidivism (Uggen & Staff, 2001). So by excluding sex offenders from
employment, policy makers could be increasing their recidivism (rates). It is therefore
important to develop accurate policy instruments that reduce sex offending and do not entail
such negative unexpected consequences. For example, policy could be made effectively if one
could find predictive patterns for sex offending, which is the main topic of the present study.
In the present study, the ten-year period prior to the first sex offence will be studied, to find
out whether we can find predictive factors in crime frequency and/or type of crime. Policy
instruments could be made more accurately when we can predict when the onset of sexual
offending takes place, since the risk of the occurrence of a sex offence will be reduced when
we can predict when this offence will occur and by who. The main theme of the present study
will be the extent to which the occurrence of the first sex offence can be predicted.
First of all, it will be studied how the crime frequency of sex offenders up to their first
sex offence differs from other types of offenders. For studying the criminal trajectory prior to
the first sex offence, we can take studies and theory into account that study the difference
between the entire criminal career of sex offenders versus nonsex offenders, since we may
assume that if the entire criminal career is not distinct, this would also account for the period
up to the first sexual offence. Some studies have found that sex offenders tend to have a
specialised and persistent criminal career: they will persistently commit specialised (sex
offending in this case) crimes throughout their lives (Harris, Smallbone, Dennison & Knight,
2009; Jennings, Piquero, Zimring & Reingle, 2015). This view that sex offenders are a
distinct type of offenders and are more dangerous than other types of offenders is also the
popular view that is held in societies (Zimring, 2004). This view resonates in many policy
interventions treating sex offenders differently than nonsex offenders, such as the policy in
the Netherlands (Boone, 2011). However, a vast amount of research has also found support
for the fact that the criminal careers of sex offenders are just like the ones of other types of
offenders (Jennings, Piquero, Zimring & Reingle, 2015). Thus there is controversy in the way
that the scientific as well as the societal world views sex offenders; are they distinct of similar
to other types of offenders? Since the main subject of the present study is predicting the onset
of sexual offending, it is important to first investigate whether sex offenders differ from
nonsex offenders in their crime frequency trajectory up to this first sex offence. If they differ,
namely, this would be a predictive factor in determining who will commit a sex offence. If
they do not differ, we can look further in how type of offences can predict a subsequent sex
offence. Thus, the first part of the present study focuses on the following question;
1. To what extent do sex offenders have distinctive criminal trajectories based on crime
frequency prior to the first registered sex offence compared with age-matched nonsex
offenders?
To answer this question, sex offenders will be age-matched to nonsex offenders. Then,
the criminal trajectories (with a duration of 10 years) of a) sex offenders up to the year of their
first sex offence; will be compared with those of b) nonsex offenders up to the year that their
age-matched sex offenders commit their first sex offence. For simplication reasons, this year
is called year zero. A criminal trajectory (also called criminal career/criminal history) is the
sequence of crimes committed by an individual offender (Blumstein et al., 1986), the present
study will only look at crime frequency in comparing the trajectories. The dataset I will use is
a subsample from the Criminal Career and Life-Course Study (CCLS). This study was
conducted by the Netherlands Institute for the Study of Crime and Law Enforcement (NSCR).
This dataset is appropriate for the present study since it contains criminal records of about 800
sex offenders, and their criminal behaviour was followed for most of their lives. There have
been very few to zero other studies on this subject using a Dutch sample (Blokland, Nagin, &
Nieuwbeerta, 2005).
After addressing the extent to which criminal trajectories of sex offenders are distinct
from nonsex offenders, I will investigate how types of crimes could predict the occurrence of
the first sex offence while controlling for crime frequency. Research on the predictive types of
crimes for sex offenders has barely been conducted yet, so using the data I will address a
second research question;
2. To what extent do types of crime in the criminal trajectory predict the likelihood of a
subsequent sex offence?
The types of crime under scrunity are property, damage and violent crimes that are
committed during the 5 years prior to the first sex offence. The present study is unique in the
sense that little to no research has been done yet on the predictive offences of sex offenders.
The large dataset that will be used provides an opportunity to conduct this unique research.
For addressing the first research question group based trajectory modelling analysis will be
used. For the second research question a multinomial probit regression will be performed.
I will first discuss relevant theories and research on the topics. From this information,
hypotheses will be derived which will be tested with the methods mentioned above. The
results of these tests will be useful to make suggestions regarding the literature on this topic as
well as the policy for reducing sex offending. These results, implications, strengths and
limitations of the study will be discussed at the final part of the study.
2. Theoretical Framework and Literature Review
2.1 Comparing Criminal Trajectories
In explaining how crime frequency differs between different types of offenders one
can apply a general as well as a crime-specific approach (Piquero, 2000). The general
approach implies that mechanisms that cause crime can cause any type of crime, which thus
suggests that criminal trajectories based on frequency do not differ between different types of
offenders (Piquero, 2000). In this section I will discuss some general theories of crime that
will explain how crime emerges without distinguishing between types of crime. These
theories are applicable to all types of offenders and do not suggest that one crime-specific
offender is more persistent than another. On the other hand, the crime-specific approach
implies that mechanisms differ for different types of offences which thus suggests that
criminal trajectories may differ between different types of offenders. Therefore, I will also
discuss some crime specific theories. Since the criminal trajectories up to the first sex offence
will be studied, it may be assumed that if criminal trajectories of different types of offenders
are expected to be similar throughout the whole criminal career based on the theory, this
would also hold for the period up to the first sex offence in comparing sex offenders to nonsex
offenders.
The first major theory that could predict whether sex offenders differ in their criminal
career from regular offenders is Terrie Moffitt’s dual taxonomy theory (1993). The main
argument of this theory is that offenders can be classified in two types of offenders: the
adolescence-limited offender [AL] (offending only occurs during adolescence) and the life-
course-persistent offender [LCP] (offending occurs during the whole life course). She
suggests that most offenders are adolescence-limited offenders. The crime committed by this
group of offenders is caused in two ways: first of all, Moffitt argues that adolescents
experience a gap between the extent to which their body has developed into maturity and the
way society views the adolescents as mature. More specifically, the adolescent body has
already developed into maturity for a great deal while society still regards these adolescents as
children. This sense of strain will lead to antisocial behaviour (Moffitt, 1993).
The second group of offenders explained by Moffitt, life-course-persistent offenders,
show criminal behaviour throughout their whole lives due to certain characteristics of their
brain and due to social characteristics of the offender. This could be tragic events that have
happened in the offender’s life such as abuse and neglect, or because of the socioeconomic
status of the offender or deviant behaviour of the parents of the offender (Moffitt, 1993). The
theory of Moffitt can hold different interpretations in predicting how distinctive the criminal
trajectories of sex offenders are. Firstly, based on the typology of LCP offenders, a sex
offender would not have a different criminal career regarding persistence than a nonsex
offender since both sex offenders and nonsex offenders can become life-course-persistent
offenders due to the factors mentioned above. However, one could also strive for the opposite
idea: sex offenders may have a higher chance in being a life-course persistent offender since it
may be more common that people who end up committing a sex offence have tragic events
that have happened in their lives. Also, the theory of Moffit implies that serious criminals
have a higher chance of being a LCP offender. Since sex offending accounts for serious
crime, this would suggest that sex offenders would follow a more persistent criminal
trajectory than less serious offenders. However, this last argument would not hold in
comparing between various serious types of offenders, such as comparing murderers with sex
offenders.
Another general theory of crime is the self-control theory of Gottfredson and Hirschi
which explains that crime occurs when people that have low self-control come into the
opportunity to commit a crime (1990). They state that this low self-control develops during
childhood by factors such as bad parenting, and usually this self-control increases with age
because of processes such as socialization and biological maturation (Gottfredson & Hirschi,
1990). If we assume that the amount of crime that an offender commits is directly linked to
the extent to which a person has self-control, then this would account for any kind of crime. If
a person has no self-control in general, it can be assumed that this low self-control accounts
for any type of crime he or she wants to commit. Thus, based on the self-control theory we
could predict that sex offenders do not have distinctive criminal trajectories based on crime
frequency.
However, there are some theories that imply that crime-specific mechanisms for
explaining criminal behaviour do exist. The routine activity theory, for example, does not
agree with the idea that crime frequency is generated by general mechanisms that cause any
type of crime instead of specific crimes (Cohen & Felson, 1979). Instead, this theory builds
on the idea that crimes happen in situational contexts where the circumstances of offending a
specific crime are ‘favourable’. Routine activity theory states that activities that happen every
day provide offenders with certain targets. Daily activities such as going to work may put
material or personal targets in accessible places for offenders (Cohen & Felson, 1979). This
theory could be interpreted in such a way that property offences are committed more
frequently than sex offences since many people leave their houses empty when they go to
work so offenders have access to the material goods of the property owners. However, the
theory could also be interpreted in such a way that sex offences can happen frequently as well
because parents may leave their children alone at home when they go to work, which may
make the children targets of child molestation. One can thus interpret this theory in different
ways for the research question under study.
Thus, there are theories that explain crime-specific behaviour as well as general-crime
behaviour which may both be used in explaining how sex offenders are or are not distinct
from nonsex offenders in their crime frequency curve up to the first sex offence. Prior
literature on this topic will now be discussed to derive a fitting prediction.
2.2 Prior Research on Comparing Criminal Trajectories
The general thoughts on sex offenders have for a long time been that they are more
specialised and persistent offenders compared with nonsex offenders (Harris, Smallbone,
Dennison & Knight, 2009; Jennings, Piquero, Zimring & Reingle, 2015). Sex offenders are
not only treated as a separate group of offenders in literature, but also by the society as a
whole and public policy (Zimring, 2004). As described in the introduction, the specialised
treatment of sex offenders in public policy of the Netherlands implies that policymakers base
policies on the assumption that sex offenders are distinct from nonsex offenders. However,
results from some recent studies showed support for the idea that criminal careers of sex
offenders are not more specialised or persistent than those of nonsex offenders (Jennings,
Piquero, Zimring & Reingle, 2015). In light addressing the extent to which sex offenders have
distinctive criminal trajectories up to their first sex offence compared with nonsex offenders
based on crime frequency, I will discuss the issue of general recidivism. General recidivism is
the extent to which an offender commits any type of crime throughout their criminal
trajectory, thus how persistent they are. We could assume that if sex offenders have
distinctive crime frequency curves up to their first sex offence, this crime frequency curve and
thus their pattern of general recidivism is also distinct from nonsex offenders throughout their
whole life span. Therefore, the results of previous research regarding the general recidivism of
sex offenders compared with nonsex offenders will now be discussed.
The available research that compares the general recidivism of sex offenders with
nonsex offenders shows mixed results, which is probably due to the differences in
methodology of the studies (Jennings et al., 2015). Caldwell has summarized available
research that mostly indicates that the general recidivism rates of sex offenders is lower than
that of nonsex offenders (2002). In addition, the results of Sipe et al. indicate that the juvenile
nonsex offenders had a higher rate of adult general recidivism (1998). Langan and Levin have
conducted a large study on recidivism of prisoners released in 1994 (2002). Their results
suggest that the general recidivism rates were not higher for sex offenders than for nonsex
offenders. More specifically, they found that, within 3 years, sexual assaulters showed a
general recidivism rate of 41%, rapist showed 46%, murderers showed 41%, and property
offenders showed 74% general recidivism rates. Another study of general recidivism of
prisoners also shows that the general recidivism of sex offenders was 25% lower than those of
nonsex offenders (Langan, Schmitt, & Durose, 2003). In addition, Sample and Bray (2003)
have shown that sex offenders are not more dangerous than other types of offenders based on
their general recidivism. In a follow up period of 5 years, they found that sex offenders
showed general recidivism rates of 45%. This recidivism rate was one of the lowest compared
with the other groups analysed in this study, only murderers (44%) and property damagers
(39%) showed lower recidivism rates. Other groups, such as robbers (75%), burglars (58%)
and nonsexual assaulters (58%) showed higher general recidivism rates than sex offenders.
Another study that used data from released prisoners in 1994 is a study directed
towards the persistence of sex offenders by Miethe et al. (2006). The results of this study
suggests that sex offenders have low levels of persistence in absolute numbers and also
compared with other types of offenders, namely violent offenders, property offenders and
public-order offenders. Zimring et al. (2007, 2009) have conducted two large birth cohort
studies to examine the sexual and nonsexual reoffending rates of juvenile sex offenders. Their
results show similar results compared with Miethe et al. (2006) and most of the other
literature discussed above, namely, sex offenders are similar to nonsex offenders in their
criminal career patterns based on frequency. If sex offenders show similar patterns to nonsex
offenders, this can also imply that the crime frequency curve up to the first sex offence is also
similar to that of age-matched nonsex offenders. However, there might be other
characteristics of the criminal trajectory of sex offenders that do differ from nonsex offenders.
For example, certain types of crime may predict a subsequent sex offence. This will be the
focus of the second part of the present study.
2.3 Predicting a Sex Offence
Since prior research on predictive patterns in criminal trajectories is limited, there is
also a lack of theoretical knowledge about how criminal histories can predict future offences.
However, it is useful to explain the criminal career paradigm by itself before discussing
applicable theories for deriving hypotheses for the present study.
The criminal career paradigm emerged in the 1980s, when new policy strategies were
needed to reduce the quickly expanding crime rates in the United States (Blokland & Lussier,
2015). This paradigm focuses on individual criminal activity of an individual offender instead
of the aggregate numbers of crimes per capita. The specific definition of a criminal career
according to the Panel of Criminal Career Research is ‘the characterization of the longitudinal
sequence of crimes committed by an individual offender’ (Blumstein et al., 1986: p. 12). The
four key dimensions that characterize a criminal career are participation (who engages in
crime?), frequency (how much criminal activity do offenders show?), seriousness (how
serious are the offences that are committed) and career length (for how long is an individual
active as an offender?). The criminal career paradigm has been expanding ever since
(Blokland & Lussier, 2015).
In 2004, Soothill et al. express their disapproval to the extent to which the criminal
career paradigm had developed up until then. According to them, the criminal career
paradigm had focused more on the quantity of crime instead of the types of crimes committed.
In their paper, they suggest a new typology of criminal activity, for example by focusing more
on type of criminal activity than quantity and by examining shorter trajectories rather than a
whole life course. They propose this typology because they see a chance that previous
offences can serve as indicators that can predict a certain offence (Soothill et al., 2004).
Soothill et al. also come with a distinction between indicators and precursors of subsequent
serious offences. Indicators are offences characterize the present situation, whereas precursors
can identify how the future will unfold regarding offences and are thus related to future
behaviour (2002). Soothill et al. state that the use of criminal trajectories in predicting
offences has been neglected by criminological theory in general (2004).

Despite the fact that a theoretical framework regarding predictive patterns in criminal
careers has not developed yet, a couple of existing theories could be useful to predict a
hypothesis for the present study. First of all, Sampson and Laub explain how escalation over
the criminal career can occur (1997). This term refers to the idea that the crimes in the
criminal career increase with seriousness. One theory that explains how escalation can occur
is called cumulative disadvantage. Cumulative disadvantage is a phenomenon where criminal
acts can cause more criminal acts by the mechanism that being convicted for a crime can lead
to a decrease in social bonds. A decrease in social bonds, in turn, can enhance an increase in
the seriousness of offending behaviour. This can be explained by Hirschi’s social control
theory (1969), which suggests that a lack of social bonds to actors that prohibit antisocial
behaviour, such as family or friends, can increase criminal activity (Hirschi, 1969). To
illustrate, suppose that a conviction leads to a loss of job or a divorce. This decrease of social
bonds may deteriorate criminal behaviour because these social bonds prohibit deviant
behaviour and a decrease in social bonds results in the idea that the individual has less to
loose when he or she commits a more serious offence. It may also be that the criminal will
have a hard time coming back into the marriage market or labour market in general after this
conviction because of bad reputation caused by the conviction. This will lead more serious
criminal behaviour, and it can thus be expected that the criminal career will escalate in
seriousness because of negative side effects of each crime (Sampson & Laub, 1997). Thus,
based on this theory, we might predict that sex offences happen in a period of escalation.
However, if we would assume that committing a sex offence happens in a period of
escalation, we still do not know when during this escalation the sex offence will occur. Also,
this escalation in seriousness does not mean it will only lead to sex offending, serious
offences such as homicide could also be the result of cumulative disadvantage.

The theory of self-control brought forward by Gottfredson and Hirschi are not only
useful in explaining crime frequency, but types of crimes as well (1990). Since they state that
people that have low self-control and come into contact with illegal opportunities will commit
certain crimes. Thus another way to speculate on how predictive patterns for sex offences
look is focusing on where in the criminal career the opportunity for such a crime could take
place (Gottfredson & Hirschi, 1990). For example, it could be reasoned that a burglary can
bring offenders into the opportunity of committing a sex offence, since they have thresspassed
the property first. This opportunity of committing a sex offence in combination with low self-
control may result in the occurrence of a sex offence.
Building on this last argument, one could argue that people may commit a sex offence
after being convicted for specific types of other offences that are directly related to sex
offences. For example, a person that will commit a sex offence might be stalking his potential
victim first. Similarly, as mentioned above burglaries might take place at the property of the
victim before the offender commits a sex offence. Some studies support this idea and have
found types of crimes that predict other types of crimes. These studies will be discussed in the
following paragraph.
2.3 Prior Research on Predictive Offences
To my knowledge, there has been little to no studies on predictive patterns in criminal
behaviour prior to a sex offence. The present study will thus be one of the first to direct this
particular question. Despite the fact that there has been little to no research done in predicting
a sex offence based on the criminal history of the offender, there has been limited research
aimed at finding predictive patterns in the criminal careers prior to other serious offences.
Most of the research done in this topic has been done by Keith Soothill. I will now discuss his
and his colleagues’ work in chronological order.

In 2000, Soothill and Francis have conducted a study where they researched around
7000 sex offenders that were convicted for a sex offence in England and Wales in 1973. They
investigated whether these sex offenders were more likely to commit a homicide than the
general population. The answer is yes. The sex offenders had a chance of 1/400 compared
with a chance of 1/3000 for the general population to commit a subsequent homicide. This
study was the first study of Soothill where the criminal history of an individual was used to
predict a subsequent offence (Soothill & Francis, 2000).
In 2002, Soothill et al. have conducted a study where they searched for the relationship
between the criminal history of an individual and the risk to commit a subsequent serious
sexual assault of an adult female. This is thus one of the few studies that was aimed at
predicting a sex offence. The sample consisted of 1057 males under the age of 45 who were
convicted for a serious sexual assault or rape for the first time in 1995-1997. The findings
suggested that being convicted for a prison sentence for committing the crimes ‘other
wounding’, robbery, stealing in a dwelling, arson, kidnapping and cruelty to children,
increased the risk of a subsequent serious sexual assault (Soothill, Francis, Ackerley &
Fligelstone, 2002).
In 2008, Soothill et al. addressed the issue whether one can predict when a homicide
will take place in a criminal trajectory based on four preliminary serious crimes (arson,
blackmail, kidnapping and threats to kill). Specifically, they firstly examined how
specialisation of one of the four crimes and escalation may predict the homicide. Afterwards,
they looked for certain combinations and sequences of the four crimes per case, to see how
these combinations and sequences can form risk factors for the subsequent homicide. They
used large datasets of offenders that have been convicted between 1979 and 2001 in England
and Wales for one of the four focused crimes. They found that, first of all, the type of the first
serious offence was a significant predictor for the subsequent homicide. For example, a
person whose first offence was a kidnap had a 48% higher risk in committing a homicide, and
a person who started with threats to kill had a 55% higher risk of committing a homicide over
an arson offender. The results of their study also show that the offenders who offended
multiple different types of serious offences were more likely to commit a homicide than those
specialising in their first serious offence. A person who committed two distinct crimes was
nearly twice as likely to commit a subsequent homicide than a person who only committed
one type of serious offence (Soothill et al., 2008).
The literature review mentioned above shows that research aimed at finding predictive
patterns in criminal behaviour for a specific type of offence is scarce, and that there are almost
no studies that investigate predictive patterns prior to a sex offence, except for the study of
Soothill et al. (2002). However, studies aimed at predicting other serious offending, such as
the study that predicted how types of crimes increase the risk in homicide (Soothill et al.,
2008), do show results that imply crimes can be predicted using the offenders’ criminal
history. After controlling for crime frequency, the present study will use Soothill’s and
colleagues’ study as an inspiration for predicting sex offences instead of homicides, and will
hereby meet the shortcoming of information in criminological literature regarding predicting a
sexual offence. The expectations based on the literature review and theoretical framework
mentioned above are as follows;
H1: Sex offenders do not have distinctive criminal trajectories based on crime
frequency prior to their first sex offence compared with age-matched nonsex offenders
H2: Types of crimes can predict the likelihood of a subsequent sex offence
To confirm the first prediction, the trajectory groups found in the group based
trajectory modelling should constitute about the same proportion of the total sex offenders as
nonsex offenders. If this hypothesis would be false, the sex offenders would constitute an
entire trajectory on their own or be highly concentrated in one or more trajectories. To
confirm the second prediction, the multinomial probit regression should show that certain
types of crimes increase the likelihood of a subsequent sex offence. The two hypotheses shall
be tested in the next sections.
3. Data and Methods
3.1 Original Sample
The sample used to test the hypotheses is based on the Criminal Careers and Life-
Course Study (CCLS) dataset. The CCLS project was conducted by the Netherlands Institute
for the Study of Crime and Law Enforcement (NSCR) (Blokland, Nagin & Nieuwbeerta,
2005). The sample is a representative sample of 4% of all persons who were tried by a judge
for a serious offence in 1977 or who were decided on by a public prosecutor. The sample was
weighted by offence types to gain an accurate representation of all individuals tried in 1977.
Retrospective and prospective information such as life-course information (e.g. marriage,
fertility history and death records) and conviction data was used to describe the people in the
sample for which this information was available. Since this information was not available for
all persons in de sample, this step reduced the sample somewhat. The people in the sample
used in the present analysis were followed until 2006 or death and the sample consists of an
age range of 12 to 91. The General Documentation Registry of the Ministry of Justice Court
Documentation Service was used to be able to register the entire criminal career of the
persons in the sample. This registry provided all criminal cases of the individuals from 1977
on registered by public prosecutors. These only include crimes committed in the Netherlands.
The sample consists of 4167 individuals, from which 11 percent were female offenders
(Blokland, Nagin & Nieuwbeerta, 2005). The data is a person-year file, which means that
each case represents a year for a person.

3.2 Final Sample
Using the original sample, several restrictions and measures were made for the present
study. Since only 5 sex offenders were women, all women were deleted from the sample. A
variable was constructed to indicate who in the sample is a sex offender (everybody who
committed at least one sex offence in their criminal career) and who was a nonsex offender
(everybody who has not committed for any sex offences in their criminal career). Since the
group of sex offenders is relatively small (about 20% of the total sample), there was not made
a distinction between different type of sex offenders. Next, an age-matched group was created
of sex offenders and the nonsex offenders. This was done by creating birth cohorts and
matching multiple nonsex offender to each sex offender based on the age of the sex offender
at their first sex offence. For the present study, the criminal careers are restricted to ten years
up to the first sex offence of the sex offenders, and for the nonsex offenders up to the year
prior to at which an age-matched sex offender committed his first sex offence. For
simplification purposes, this year will be referred to as year zero. It is thus important to note
for the present study that the years represented in the criminal trajectories refer to the amount
of years prior to year zero. These years may, for each individual, refer to different calendar
years, ages, or both. The final sample used in the first analysis consisted of 3365 nonsex
offenders and 777 sex offenders (4142 in total). Since for the second analysis (multinominal
probit regression) the goal was to predict what crime one was going to commit in year zero,
the 12 people who had died before year zero were deleted from the sample and thus the
sample for the second analysis consisted of 3353 nonsex offenders and 777 sex offenders
(4130 in total).
3.3 Operationalization
There were multiple dependent and independent variables used for the analysis. The
descriptive statistics of these variables are given in Table 1.

Table 1
Descriptive statistics of used variables

Variable N Minimum Maximum Mean STDEV
First Analysis
Total convictions
in 10-year period 4142 0 101 3.10 5.62
Proportion free per year 4142 0 1 .94 .18
Second Analysis
Age at year zero 4130 13 66 24.94 9.77
Crime at year zero 4130 1 3 2.39 .78
Birth cohort 4130 1 11 8.53 2.10
Trajectory group 4130 1 5 2.44 .88
At least 1 violent offence 4130 0 1 .10 .31
At least 1 property offence 4130 0 1 .30 .46
At least 1 damage offence 4130 0 1 .11 .31
Note. 12 people died before year zero and were not taken into the second analysis
For the first analysis, crime trajectories were estimated over the period of 10 years
prior to year zero (ytosex = -10 to ytosex = -1). This was done with the variable vftotal which
was the sum of total crimes of each year. Since it should be taken into account that people
cannot commit crimes while in prison, this analysis was controlled by the variable free, which
is the proportion of freedom of each year. For the second analysis, it was tested whether types
of offences can predict a subsequent sex offence using a multinomial probit regression. The
dependent variable, delict0, is the crime they committed at year zero. This variable had three
outcomes; a sex offence, a nonsex offence or no offence. The predictors were violent, damage
and property, which are variables that indicate how many people have committed at least 1
violentdamage- or property offence within the 5-year period before year zero. This
prediction was controlled for the age people had at year zero, the birth cohort they were in and
the trajectory group they were in (thus for crime frequency). This last variable was determined
by the previous analysis (trajectory analysis).
3.4 Method
The data was analysed in several steps using STATA. First, semi parametric group-
based trajectory modelling was conducted by using the traj plugin to estimate the underlying
distinct groups of offenders that show similar patterns of conviction rate over time (Nagin,
1999). This is a method that estimates the underlying crime curves distributed over a number
of distinct groups that follow the same trajectory based on frequency. It is thus a good
procedure to check how different types of people are distributed over different crime curves.
For the analysis, the zero-inflated Poisson model was used. This is a model that makes sure
that short periods of non-offending do not result in changes in the offending trajectory. The
model used a cubic relationship for the crime curve over time based on the following formula;
log( λ ¿j ¿=β 0j + β 1j Time ¿ + β 2j Time 2¿ + β 3j Time 3¿
where λ ¿j indicates the expected number of convictions of person i at time t given his
membership in group j. The time over which the curve was estimated was the 10 years up to
year zero. Time ¿ indicates the amount of years prior to year zero at time t, for Time 2¿ this is
squared and for Time 3¿ this is cubed. β 0j , β 1j , β 2j∧¿ β 3j are the coefficients that determine the
shape of each trajectory. The denotation of j indicates that the coefficients vary across the
groups. The crime trajectories were controlled for the proportion that the offenders were in
prison.
A key step in this analysis is determining how many distinct groups can be identified
in the data. To determine this number, the Bayesian Information Criterion (BIC) is used. The
higher the BIC value, the better the model fits. However, in determining the right number of
distinct groups, the BIC number is not the only important criterion since the BIC number in
the used sample gets higher every time another group is added to the model. Since estimating
a trajectory model is done for simplification of the data, the maximum number of distinct
groups to be tried in the model was 6. The BIC is calculated by the following formula:
BIC = log(L) – 0.5log(n)*(j),

where n is the sample size, L the value of the maximized likelihood of the model, and
k is the number of parameters (Nagin, 1999).
After the trajectory groups are formed, individual probabilities of group membership
were determined. This posterior probability of group membership was determined with the
following equation:
^ (Y i∨ j) π^ j
P
^
P ( j|Y i ) =
∑ P^ (Y i∨k ) π^ j
j
where ^P ( k|Y i ) is the estimated probability that we can observe individual i is actual
crime curve (Y i ¿, given that this individual is a member of group j, and π^ j indicates the
estimated proportion of the population in that group j. This posterior probability of group
membership is used to be able to assign the individuals to the right crime curve. This
probability should not be lower than .70 for each trajectory in the estimated model. The
analysis described up until now will answer the first research question.
After defining how sex offenders are distributed over the different trajectory groups by
the posterior probability, this group assignment was added to the original sample data to
perform the remaining analysis for research question two. For this analysis, a multinomial
probit regression was performed using STATA, since aim of the second research question is
predicting a nominal dependent variable. The analysis was used to test whether having
committed at least one violentdamage- or property offence in the 5 years up to zero
predicted which crime was going to be committed in year zero (a sex offence, a nonsex
offence or no offence). This prediction was controlled for age at year zero, birth cohort and
the trajectory group people belonged in (thus crime frequency over time). The probit
regression was used instead of the multinomial logistic regression since the assumptions on
this latter model were violated. The syntax (SPSS as well as STATA) for the data measures
and the analyses are shown in Appendix A.

4. Results
The results of the analyses are as follows. First of all, the crime curve up to year zero
averaging all individuals in the sample is plotted in Figure 1. This figure indicates that the
mean number convictions increase gradually over time, with about 0.5 mean convictions in
the year prior to zero. However, this average crime curve may obscure underlying distinct
groups of offenders who follow similar trajectories. This possibility was addressed in the
group based trajectory analyses.
Figure 1
Average Crime Curve for Entire Sample (N=4142)
0.5
0.45
0.4
mean number of convictions per year
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
-10 -9 -8 -7 -6 -5 -4 -3 -2 -1
years to 'zero'
After conducting the group based trajectory analysis, it became clear that there were
indeed distinct underlying groups that followed similar crime curves. The response variable
was the number of convictions every year up to the year before zero. Multiple models were
tested to find which model of trajectories fitted best. I only tested up to 6 group models for the
trajectories since more than 6 groups models take away the main function of group based
trajectory modelling: simplifying the data into homogeneous groups. For each model I tested
the curves to be first order, second order and third order curves. For each model, I checked
whether the visual representation of the graph was in line with the average numbers of crimes
for each year within each group. In addition to visual criteria for finding the best fitting
model, another criterion was used; the Bayesian Information Criterion (BIC) score. The
higher this score, the better the model. Since for the present dataset the BIC score gets higher
and higher, visual criteria are important also. In the end, the five-group cubic model came out
to be the best fitting model. Even though the six-group model model had a better BIC score,
the individual groups were too similar in this model. The BIC score of the five-group model
was -18339.47. The visual representation of the estimated lines of the best fitting model are
represented in Figure 2. The numerical values of this graph are represented in Table 2. The
posterior probabilities for group assignment are represented in Table 3.
Table 2
Numerical Values of Parameters Estimates for the Five-group Model
Group 1 Group 2 Group 3 Group 4 Group 5

Intercept -0.05190 5.21644*** 1.00159*** 3.84286*** 1.62597***
Linear -1.49985*** 0.37003* 0.37769* 0.02325 -0.00016
Quadratic -0.31517*** 0.03477 0.10495** -0.12088** 0.00674
Cubic -0.01564*** 0.00073 0.00702*** -0.00915** 0.00078
free_1 -1.45072*** -7.52201*** -1.71187*** -3.77114*** -1.26216***
Note. *=p<.05; **=p<.01; ***=p<.001

Figure 2
Estimated Trajectories for Five-group cubic model
1.8
1.6
mean number of convictions per year
1.4
1.2
0.8
0.6
0.4
0.2
0
-10 -9 -8 -7 -6 -5 -4 -3 -2 -1
years to 'zero'
Group 1 (3.3%) Group 2 (64.7%) Group 3 (11.5%) Group 4 (17.2%)

Group 5 (3.3%)
Table 3
Mean assignment posterior probability
Assigned group Group 1 Group 2 Group 3 Group 4 Group 5

Group 1 .79 .00 .05 .04 .04
Group 2 .00 .88 .06 .10 .00
Group 3 .07 .04 .79 .09 .05
Group 4 .07 .07 .07 .76 .01
Group 5 .07 .00 .03 .01 .91
Note. All posterior probabilities of the accurate group are higher than the threshold of .7
Figure 2 shows that Group 1 and Group 5 make up a small proportion of the
population, Group 3 and 4 a larger propertion and Group 2 the largest. Group 1 shows a
constant line with little to no crime frequency in the ten years prior to zero. Group 5 seems to
do the opposite; this group of offenders show a more or less constant line of relatively a lot of
offences during the 10 years prior to zero. Group 1 shows a steep increase in crime frequency
up until 3 years prior to zero, where after they show a steep decline up until the year prior to
zero. Group 4 shows a gradual increase in crime frequency, and finally Group 3 shows an
increase, a decline and another increase during the 10 years prior to zero. Since Group 2 is the
largest group, most people in the population are expected to show little to no crime frequency
in the ten years up until zero. This fact is not very surprising, since the average age of the
offenders at year zero is relatively young (25 years old). Age and other characteristics of the
groups are discussed later.
Table 3 shows the posterior group probabilities. These are all above the threshold
of .70, which means that this trajectory model assigns the individuals to the accurate crime
curves and thus seems to be a fitting model. The group that each individual belongs to, which
became clear in this first analysis, were matched to the data to be able to conduct the second
analysis which is discussed later.
First, to be able to answer the first research question, it is explored how the sex
offenders are distributed over the trajectory groups. This distribution is indicated in Table 4.
The percentages are the percentages of the total group of (non)sex offenders.
Table 4
Distribution of sex offenders and nonsex offenders over the trajectory groups
Sex offenders (n=777) nonsex offenders (n=3353) Total population

Group 1 20 (2.6%) 95 (2.8%) 3.3%
Group 2 544 (70%) 2394 (71.4%) 64.7%
Group 3 67 (8.6%) 276 (8.2%) 11.5%
Group 4 109 (14%) 502 (15%) 17.2%
Group 5 37 (4.8%) 86 (2.6%) 3.3%
As shown in Table 4, the percentages of the total type of offenders are similar within
each group, which indicates that sex offenders do not have distinct criminal trajectories in the
ten years prior to their first sex offence compared with age-matched nonsex offenders. This
confirms the first hypothesis; a statistical significance test is not necessary since the
distribution of the sample is clear in showing an even distribution. The third column of Table
4 shows that the percentages of the distribution over the groups of two types of offenders are
in line with the distribution over the trajectories of the entire population.
Table 5
Average Age of Sex offenders and nonsex offenders within each trajectory
Average age at ‘zero’ Average birth year

Sex nonsex Sex nonsex
Group 1 28 25 1951 1953
Group 2 24 24 1948 1949
Group 3 30 30 1948 1949
Group 4 23 24 1950 1953
Group 5 32 29 1953 1952
Since only the whole sample was age-matched but not the trajectories itself, it is
informative to compare the average age of sex offenders with nonsex offenders within each
trajectory. These average ages at year zero and average birth years are given in Table 5. The
average ages at year zero and the average birth years do not differ much between the two
types of offenders, thus age does not seem to influence the notion that the crime trajectories of
sex offenders are not distinct from nonsex offenders.
Now we go to the second analysis of the present study; test whether certain types of
crimes can predict a subsequent sex offence. Therefore, we need to explore what types of
offences are committed in year 0. A distinction was made between people who committed a
sex offence, a nonsex offence or no offence at all. For this prediction, the years under scrunity
were 5 years prior to zero. It was found that the most common offences in these years were
violent offences, property offences and damage offences. Therefore, only these three types of
offences were used as predictors for a subsequent sex offence, since the other types of
offences did not have enough hits to be used in an analysis. Table 6 shows, for each
trajectory, the type of offences followed by the percentage of individuals who subsequently
are convicted of a sex offence, a nonsex offence or no offence in year zero. It thus indicates
the likelihood of subsequently committing a sex offence given that the individual has
committed at least one violence, property or damage offence in the 5 preceding years.
Table 6
Type of offences subsequent sex offence, nonsex offence and no offence
N sex offence (%) nonsex offence (%) none (%)

Group 1
At least 1 property 105 19 (18.1) 57 (54.3) 29 (27.6)
At least 1 damage 61 10 (16.4) 38 (62.3) 13 (21.3)
At least 1 violence 31 6 (41.9) 12 (38.7) 13 (41.9)
Group 2
At least 1 damage 85 10 (11.8) 18 (21.2) 57 (67.1)
Group 3
At least 1 damage 53 12 (22.6) 17 (32.1) 24 (45.3)
Group 4
At least 1 damage 180 32 (17.8) 73 (40.6) 75 (41.7)
Group 5
At least 1 damage 60 20 (33.3) 26 (43.3) 14 (23.3)
At least 1 violence 50 19 (38) 22 (44) 9 (18)
Based on Table 6, a multinomial probit regression was conducted using STATA. The
dependent variable was type of crime committed in year zero, which could be a sex offence, a
nonsex offence or no offence. The predictors were having committed at least 1 violence, 1
damage or 1 property offence in the 5 years up to zero. The analyses were controlled for age
at year zero, birth cohort and the trajectory that people belonged in. The results of the
analyses are displayed in Table 7.

Table 7
Parameter Estimates for Multinominal Probit Regression (N=4130)
Wald Chi² (18) 435***

Log likelihood -3759.8085
Crime at zero Coeff. SE z-score

Sex offence Intercept -.559* .238 -2.34
Damage .011 .119 0.09
Property .343*** .088 3.92
Violent .090 .114 0.79
Group1 .360 .229 1.57
Group3 .211 .128 1.65
Group4 .115 .113 1.02
Group5 .879*** .214 4.10
Age -.011* .004 -2.54
Cohort -.029 .019 -1.53
Nonsex offence Intercept -2.108*** .271 -7.78
Damage .127 .109 1.17
Property .649*** .083 7.77
Violent .045 .107 0.42
Group1 .961*** .202 4.77
Group3 .522*** .122 4.29
Group4 .481*** .105 4.58
Group5 .911*** .205 4.44
Age -.009 .005 -1.85
Cohort .129*** .021 6.10
Note. The reference category is ‘no offence’

*=p<.05; **=p<.01; ***=p<.001
The results of the multinominal probit regression that are displayed in Table 7 can be
interpreted as follows. First of all, the overall fitted model has a Log likelihood of -3759.8085
and is statistically significant (Wald Chi² (18) = 435, p<.001) which means that the overall
predicted model tested here is better than a model with no predictors. Now let us proceed to
the individual coefficients. As displayed in Table 7, only property has a significant effect.
Thus, keeping trajectory group, age, and birthcohort constant, having committed at least 1
property crime during the 5 preceding years increases the predicted probability of committing
a subsequent sex offence versus not committing any offence (B=.343, z=3.92, p<.001). This
result suggests that hypothesis 2 is confirmed (H2: Types of crimes can predict the likelihood
of a subsequent sex offence). However, this confirmation should be interpreted with caution;
having committed at least 1 property crime namely also increases the predicted probability of
a nonsex offence (B=.649, z=7.77, p<.001). Thus, while hypothesis 2 can be confirmed thanks
to the increase of predicted likelihood after a property offence, we can not say anything about
the difference in likelihood of a subsequent sex offence versus a subsequent nonsex offence.
Having committed at least 1 property offence namely increases the likelihood of a subsequent
sex offence as well as a subsequent nonsex offence versus no offence in year zero.
To be able to say a little more about the likelihood of subsequent offences, the actual
predicted probabilities can be computed using STATA. Therefore, the predicted probability of
a subsequent sex offence for a person who has committed at least 1 property offence (person
A) is compared with a person who has not committed a property offence in the 5 preceding
years (person B). The same is done for predicting the probability of a subsequent nonsex
offence. The other predictors are held constant for this comparison (both persons are in
trajectory group 2 and have not committed a violent or damage offence). These predicted
probabilities are computed using F, which is the cumulative distribution function of the
standard normal.
Predicted probability of a subsequent sex offence;
F (-.559 + property *.343 + age*-.011 + cohort * -.029)
F (-.559 + 1*.343 + 25*-.011 + 9*-.029) =
F (-0.752) = .22602553 [Person A]
F (-.559 + property *.343 + age*-.011 + cohort * -.029)
F (-.559 + 0*.343 + 25*-.011 + 9*-.029) =
F (-1.095) = .13675832 [Person B]

Predicted probability of a subsequent nonsex offence;
F (-2.108 + property * .649 + age*-.009 + cohort * .129)
F (-2.108 + 1*.649 + 25*-.009 + 9*.129) =-225 + 1.161 – 2.108
F (-0.523) = .30048713 [Person A]
F (-2.108 + property * .649 + age*-.009 + cohort * .129)
F (-2.108 + 0*.649 + 25*-.009 + 9*.129) =
F (-1.172) = .12059853 [Person B]
Thus, for a person who has committed at least 1 property offence in the 5 preceding
years (Person A), the predicted probability of a subsequent sex offence is .23, while the
predicted probability of a subsequent nonsex offence is .30. For a person who has not
committed a property offence in the 5 preceding years (Person B), the predicted probability of
a subsequent sex offence is .14, while the predicted probability of a subsequent nonsex
offence is .12.
After having more insight in the actual predicted probabilities, it may still be said that
the the likelihood of a subsequent sex offence can be predicted by the type of crime.
However, this is not to say that this prediction can be made more easily for sex offenders, nor
that a subsequent sex offence is more likely than a subsequent nonsex offence after
committing a property crime.
5. Conclusion and Discussion
In investigating the extent to which the first sex offence can be predicted, the
following conclusions can be made. First of all, the criminal trajectories based on crime
frequency up to the year of the first sex offence does not differ between sex offenders and
age-matched nonsex offenders. This confirms the first hypothesis. This implies that the
occurrence of a sex offence can not be predicted based on crime frequency. However, this
does confirm that the popular view that sex offenders are distinct and more dangerous in their
persistence, which exists within society and among policy makers (Zimgring, 2004), is not
based on accurate information. Instead, sex offenders seem to be similar to nonsex offenders
based on their crime frequency curve up to their first sex offence. Even though only the 10
years up to zero were under scrunity, it is likely that this confirmation accounts for the entire
criminal career, however this is not certain.
By confirming the idea that sex offenders have similar crime frequency patterns as
nonsex offenders, we could use this information for the second research question, which
addressed the extent to which types of crimes can predict a subsequent sex offence. While
controlling for this crime frequency, it was found committing at least 1 property offence in the
5 years under scrunity, increased the predicted probability of committing a subsequent sex
offence in year zero versus not committing any offence. This confirms the second hypothesis.
However, this confirmation does not go without saying that it does not say anything about
comparing sex offenders with nonsex offenders; even though a property offence increases the
predicted probability of a subsequent sex offence, it does also increase the predicted
probability of a subsequent nonsex offence. Thus, also in the second part of the study sex
offenders do not seem to be special when comparing them with nonsex offenders.
[terugkoppeling probleemstelling/theory/bestaande literatuur]
5.1 Policy Implications
5.2 Limiations and Research Suggestions
- nog vergelijken nonsex offence ses offence in predicted probability
- grotere sample voor meer specifieke delicten
- niet alleen naar eerste sex offence kijken, maar naar alle sex offences
- kijken naar combinaties van delicten

References
Berg, C.J.W. van den (2015). From Boys to Men: Explaining Juvenile Sex Offenders'
Criminal Careers (Ph.D.). Vrije Universiteit Amsterdam.
Blokland, A., Nagin, D., & Nieuwbeerta, P. (2005). Life Span Offending Trajectories of a
Dutch Conviction Cohort*. Criminology, 43(4), 919-954. http://dx.doi.org/10.1111/j.1745-
9125.2005.00029.x
Blokland, A., & Geest, V. (2015). Life-Course Transitions and Desistance in Sex Offenders
(ZEDENBOEK)
Blokland, A., & Lussier, P. (2015). The Criminal Career Paradigm and Its Relevance to
Stuyding Sex Offenders (ZEDENBOEK).
Blumstein, A., Cohen, J., Roth, J., Y Visher, C. (1986). Criminal Careers and Career
Criminals. Washington, DC: National Acadamy Press.
Boone, M. (2011). Judicial Rehabilitation in the Netherlands: Balancing between safety and
privacy. European Journal of Probabion, 3(1), 63-78.
Caldwell, M. F. (2002). What We Do Not Know About Juvenile Sexual Reoffence Risk.
Child Maltreatment, 7(4), p. 291-302.
Cohen, L. E., & Felson, M. (1979). Social Change and Crime Rate Trends: A Routine
Activity Approach. American Sociological Review, 44(4), p. 588-608.
Gottfredson, M., & Hirschi, T. (1990). A general theory of crime. Stanford, CA: Stanford
University Press.
Harris, D. A., Smallbone, S., Dennison, S., & Knight, R. A. (2009). Specialisation and
versatility in sexual offenders referred for civil commitment. Journal of Criminal Justice,
37, 37–44
Hirschi, T. (1969). Causes of Delinquency. Berkeley: University of California Press.
Hirschi, T., & Gottfredson, M. (1995). Control theory and the life‐course perspective. Studies
on Crime and Crime Prevention, 4, 131–142.
Jennings, Piquero, Zimring, Reingle. (2015): Assessing the continuity of sex offending over
the life course. Sex Offenders, 129-142. http://dx.doi.org/10.1002/9781118314630.ch6
Jones, B. & Nagin, D. (2013). A Note on a Stata Plugin for Estimating Group-based
Trajectory Models. Sociological Methods & Research, 42(4), 608-613.
http://dx.doi.org/10.1177/0049124113503141
Langan, P. A., & Levin, D. J. (2002). Recidivism of Prisoners Released in 1994. Federal
Sentencing Reporter, 15(1): 58-65.
Langan, P., Schmitt, E., & Durose, M. (2003). Recidivism of sex offenders released from
prison in 1994. Washington, DC: US Department of Justice.
Levenson, J. S., Brannon, Y. N., Fortney, T., & Baker, J. (2007). Public perceptions about sex
offenders and community protection policies. Analyses of Social Issues and Public Policy,
7, 1–25.
Liu, J., Francis, B., Soothill, K. (2008). Kidnapping offenders: Their risk of escalation to
repeat offending and other serious crime. The Journal of Forensic Psychiatry &
Psychology, 19(2): 164-179.
Lussier, P., & Cortoni, F. (2008). The development of antisocial behavior and sexual aggres-
sion: Theoretical, empirical, and clinical implications. In B. K. Schwartz (Ed.), The sex
offender: Offender evaluation and program strategies, Vol. 6 (pp. 2/1 ‐ 2–26). Kingston,
NJ: Civic Research Institute.
Lussier, P., LeBlanc, M., & Proulx, J. (2005). The generality of criminal behaviour: A
confirmatory factor analysis of the criminal activity of sex offenders in adulthood. Journal
of Criminal Justice, 33, 177−189.

Miethe, T., Olson, J., & Mitchell, O. (2006). Specialisation and persistence in the arrest
histories of sex offenders: A comparative analysis of alternative measures and offence
types. Journal of Research in Crime and Delinquency, 43, 204−229.
Moffitt, T. (1993). Adolescence-limited and life-course-persistent antisocial behavior: A
developmental taxonomy. Psychological Review, 100(4), 674-701.
http://dx.doi.org/10.1037//0033-295x.100.4.674
Nagin, D. (1999). Analyzing developmental trajectories: A semiparametric, group-based
approach. Psychological Methods, 4(2), 139-157. http://dx.doi.org/10.1037//1082-
989x.4.2.139
Nagin, D. & Odgers, C. (2010). Group-Based Trajectory Modeling in Clinical Research.
Annual Review Of Clinical Psychology, 6(1), 109-138.
http://dx.doi.org/10.1146/annurev.clinpsy.121208.131413
Nieuwbeerta, P., Blokland, A., Piquero, A., & Sweeten, G. (2011). A Life-Course Analysis of
Offence Specialisation Across Age: Introducing a New Method for Studying Individual
Specialisation Over the Life Course. Crime & Delinquency, 57(1), 3-28.
http://dx.doi.org/10.1177/0011128710376336
Piquero, A. (2000). Frequency, Specialisation, and Violence in Offending Careers. Journal of
Research in Crime and Delinquency, 37, 392-418.
Politie.nl (2015). Klassieke Zedenmisdrijven Dalen, Online Zedendelicten Nemen Toe.
Retrieved 10-04-2017 via https://www.politie.nl/nieuws/2015/april/7/misdaadmeter-
zedenmisdrijven.html
Politie.nl (2017). Publicaties en Cijfers. Retieved 10-04-2017 via
https://www.politie.nl/themas/thema/publicaties-en-cijfers.html
Rubenstein, M., Yeager, C. A., Goodstein, C., & Lewis, D. O. (1993). Sexually assaultive
male juveniles: a follow up. The American Journal of Psychiatry, 150(2): 262-265.
Sample, L., & Bray, M. (2003). Are Sex Offenders Dangerous? Criminology and Public
Policy, 3(1): 59-82.
Sampson, Rl, & Laub, J. (1993). Crime in the Making. Cambridge, Mass.: Harvard University
Press.
Sampson, Robert J. and John H. Laub. 2003. Life-course desisters? Trajectories of crime
among delinquent boys followed to age 70. Criminology 41(3): 555-592.
Sampson, R. J., & Laub, J. H. (1995). Understanding variability in lives through time:
Contribution of life‐course criminology. Studies on Crime and Crime Prevention, 4,
143–158.
Sipe, R., Jensen, E. L., & Everett, R. S. (1998). Adolescent sexual offenders grown up:
Recidivism in young adulthood. Criminal Justice and Behaviour, 25: 109-124.
Soothill, K., Francis, B. (2000). Does sex offending lead to homicide? The Journal of
Forensic Psychiatry, 11(1): 49-61.
Soothill, K., Francis, B., Sanderson, B., & Ackerley, E. (2000). Sex offenders: Specialists,
generalists, or both? A 32‐year criminological study. British Journal of Criminology, 40,
56–67.
Soothill, K., Francis, B., Ackerley, E., & Fligelstone, R. (2002). Murder and serious sexual
assault: what criminal histories can reveal about future serious offending. Policing and
Reducing Crime Unit: Police Research Series.
Soothill, K., Francis, B, Fligelstone, R. (2004). Identifying Patterns and Pathways of
Offending Behaviour: A new Approach to Typologies of Crime. European Journal of
Criminology, 1(1): 47-87.
Soothil, K., Francis, B., Liu, J. (2008). Does Serious Offending Lead to Homicide? Exploring
the Interrelationships and Sequencing of Serious Crime. The British Journal of
Criminology, 48(8): 522-537.

Terugkijktermijnen. Justis.nl. Retrieved 15 February 2017, from
https://www.justis.nl/producten/vog/vog-aanvragen/naar-welke-gegevens-wordt-gekeken/
terugkijktermijnen.aspx
Uggen, C., & Staff, J. (2001). Work as a Turning Point for Criminal Offenders. Corrections
Management Quarterly, 5(4), 1-16.
Zimring, F. E. (2004). An American travesty: Legal responses to adolescent sexual offending.
Chicago, IL: University of Chicago Press.
Zimring, F. E., Piquero, A. R., & Jennings, W. G. (2007). Sexual Delinquency in Racine:
Does Early Sex Offending Predict Later Sex Offending in Youth and Young Adulthood?
Criminology & Public Policy, 6(3), 507-534.
Zimring, F. E., Jennings, W. G., Piquero, A. R., & Hays, S. (2009). Investigating the
Continuity of Sex Offending: Evidence from the Second Philadelphia Birth Cohort. Justice
Quarterly, 26(1), 58-76.

Appendix A: Syntax data restrictions - analyses
1. SPSS
* Encoding: UTF-8.
get file ='F:\Final\SOperson_yearFINAL.sav'.
*variabele die aangeeft hoeveeel unieke personen in de data zitten.
compute person = 0.
if (idnew = lag(idnew)) person = 1.
exe.
freq var person.
*variabele die voor elk jaar aangeeft of er sprake is van een sexual offence.
compute sexoffence = rape + sexass + sexab + flash.
exe.
*age sex offence.
if (sexoffence ge 1) agesex = age.
exe.
*variabele die voor elke respondent het aantal zeden delicten weergeeft en onsetage sex
offence.
AGGREGATE
/OUTFILE=* MODE=ADDVARIABLES OVERWRITEVARS=yes
/BREAK=idnew
/sexoffences = sum(sexoffence) / onsetagesex = min(agesex).
*variabele die per respondent weergeeft of die respondent een zedendader is.
recode sexoffences (0=0) (1 thru hi = 1) into sexoffender.
exe.
*variabele die per respondent weergeeft of die respondent een zedendader is.
recode sexoffences (0=0) (1 thru hi = 1) into sexoffender.
exe.
temporary.
select if person = 0.
freq var sexoffender onsetagesex sexoffences.
*slechts 5 sex offenders zijn vrouw.
temporary.
select if (sexoffender = 1) and (person = 0).
freq var geslp.
*daarom analyses alleen op mannen.
select if geslp = 1.
exe.
*alleen mannen.
temporary.
freq var sexoffender onsetagesex sexoffences.
temporary.
select if (sexoffender = 1) and (person = 0).
freq var gebjp.
temporary.
select if (sexoffender ne 1) and (person = 0).
freq var gebjp.
*geboortejaar in 5-jaar categorieen.
recode gebjp (1960 thru 1964 = 11)

(1955 thru 1959 = 10)
(1950 thru 1954 = 9)
(1945 thru 1949 = 8)
(1940 thru 1944 = 7)
(1935 thru 1939 = 6)
(1930 thru 1934 = 5)
(1925 thru 1929 = 4)
(1920 thru 1924 = 3)
(1915 thru 1919 = 2)
(1910 thru 1914 = 1) into gebjp_c.
*aantal sex offenders per geboortecohort.
temporary.
select if person = 0 and sexoffender = 1.
freq var gebjp_c.
temporary.
select if person = 0 and sexoffender = 0.
freq var gebjp_c.
*voor elk geboortecohort aantal verschillende onset age sex.
temporary.
select if person = 0 and sexoffender = 1 and gebjp_c = 1.
freq var onsetagesex.
*in geboortecohort 1 zitten 2 sex offenders, respectievelijk met onset leeftijd sex van 26 en
62.
*de helft van het corresponderende cohort nonsex offenders krijgt onsetagesex 26, de andere
62.
*zo kunnen zij als controle personen dienen voor de sex offenders uit dit cohort.
if (person = 0 and sexoffender = 0 and gebjp_c = 1) hulp = 1.
exe.
sort cases by hulp (d) idnew.
if (hulp = 1) and (lag(hulp) ge 1) hulp = lag(hulp)+1.
EXECUTE.
freq var hulp.
*geeft aantal controle personen.
*dit aantal opdelen in aantal relevante onsetagecategorieen.
if (hulp lt 5) onsetagecon = 26.
if (hulp ge 5) onsetagecon = 62.
exe.
delete var hulp.
temporary.
freq var onsetagecon.
temporary.
*in cohort 2 zitten 10 sex offenders.
exe.
EXECUTE.
freq var hulp.
if (hulp le 3) onsetagecon = 18.
if (hulp gt 3 and hulp le 6) onsetagecon = 19.
*onsetagesex 59 komt twee keer voor in dit cohort.
*je houdt nog 4 controle personen over in dit cohort.
*die verdelen over de hoogste onsetagesex categorieen ivm grootste kans op uitval door
overlijden.
if (hulp = 31) onsetagecon = 45.
exe.
delete var hulp.
temporary.
temporary.
exe.
EXECUTE.
freq var hulp.

*onsetagesex 53 komt twee keer voor in dit cohort.
*nog 12 over.
exe.
delete var hulp.
temporary.
temporary.
exe.
EXECUTE.
freq var hulp.
*nu zijn er nog 7 over.

exe.
delete var hulp.
temporary.
temporary.
exe.
EXECUTE.
freq var hulp.
*138 controle personen verdelen over 33 sex offenders, dus eerst 4 per groep.

exe.
delete var hulp.
temporary.
temporary.
exe.
EXECUTE.
freq var hulp.

exe.
delete var hulp.
temporary.
temporary.
exe.
EXECUTE.
freq var hulp.

exe.
delete var hulp.
temporary.
temporary.
exe.
EXECUTE.
freq var hulp.

exe.
delete var hulp.
temporary.
temporary.
exe.
EXECUTE.
freq var hulp.

exe.
delete var hulp.
temporary.
temporary.
exe.
EXECUTE.
freq var hulp.

exe.
delete var hulp.
temporary.
temporary.
exe.
EXECUTE.
freq var hulp.

exe.
delete var hulp.
temporary.
*Uiteindelijk is het totaal van onsetagecon 3386, dit is een minder dan de aantal nonsex
offenders maar dat is omdat hij buiten de cohort categorien dus het is geen probleem.
sort cases by idnew age.

*variabele die voor elke respondent de controle onsetleeftijd weergeeft.
AGGREGATE
/BREAK=idnew
/onsetagecon = max(onsetagecon) .
if (sexoffender = 0) onsetagesex = onsetagecon.
temporary.
temporary.
freq var onsetagecon sexoffender onsetagesex.
*aanmaken variabele year to sex offence.
compute ytosex = age-onsetagesex.
exe.
freq var ytosex.
*alleen jaren voorafgaand aan jaar sex offence meenemen.
SELECT IF ((ytosex < 0) & (ytosex >= -10)).
exe.
SORT CASES BY idnew(A) ytosex(D).
exe.
EXECUTE.
save outfile= 'F:\Final\SOyearFINAL.sav' .
exe.
get file= 'F:\Final\SOyearFINAL.sav' .
exe.
CASESTOVARS
/ID=idnew
/GROUPBY=VARIABLE.
save outfile= 'F:\Final\SOyearFINAL1.sav'
/ keep = idnew vftot.1 to vftot.10 free.1 to free.10 ytosex.1 to ytosex.10.
exe.
get file= 'F:\Final\SOyearFINAL1.sav'.
exe.
SAVE TRANSLATE OUTFILE='F:\Final\SOyearFINALSTATA.dta'
/TYPE=STATA
/VERSION=11
/EDITION=SE
/MAP
/REPLACE.
*geeft foutmeldingen mbt variabele namen.
*die kun je negeren.
*in STATA gekeken of vftot.1-vftot.10 extreme waarden lijken te hebben, dat is neit het
geval, dus ik ga er niet van uit dat er outliers zijn die de analyse in de weg zitten.
*TRAJECTANALYSE UITVOEREN (zie STATA syntax).

*resultaten traject analyse.
*vanuit STATA, group probabilities opslaan als .cvs bestand.
*vervolgens in excel alle punten vervangen door comma's.
*daarna excel bestand in lezen in spss.
*koppelen group probabilities aan SPSS file.
*Eerst een SPSS file maken die nog wel ytosex=0 erbij heeft (want die had ik er uitgefilterd
en het gaat om wat er wordt gepleegd in het jaar ytosex=0.
*Deze SPSS file maak ik even hier buitenom en die noem ik SOyearFINAL2.
get file= 'F:\Final\SOyearFINAL2.sav'.
exe.
sort cases by idnew year.
MATCH FILES /FILE='F:\Final\SOyearFINAL2.sav'
/TABLE='F:\Final\SOtrajecten3.sav'
/BY idnew.
EXECUTE.
SAVE OUTFILE = 'F:\Final\SOyearFINALMATCHED.sav'.
GET FILE = 'F:\Final\SOyearFINALMATCHED.sav'.
*Nu ga ik zorgen dat ik de beschrijvende statistieken kan gaan invullen.
GET
FILE='F:\Final\SOyearFINALMATCHED.sav'.
*hier zitten dus alle jaren tot en MET het sexoffence of nonsex offence of het nonoffence.
*Alleen nu zie ik allemaal missings bij de probabilities, hoe kan dat?
freq group.
*het zijn er maar 14 missing, misschien omdat hun onsetagesex 12 was?
freq onsetagesex.
*Ja, dat klopt. Dus mensen die onsetagesex 12 hebben moeten er uit.
select if onsetagesex > 12.
freq group.
*Nu moet ik per groep kijken hoeveel mensen een sexoffence, nonsexoffence of nonoffence
plegen in jaar 0.
*Daarvoor moet ik eerst een variabele aanmaken die per case aangeeft welk delict ze plegen
in het jaar 0.
IF (ytosex = 0 & sexoffence > 0) delict0=1.
IF (ytosex = 0 & sexoffence = 0 & vftot > 0) delict0=2.
IF (ytosex = 0 & sexoffence = 0 & vftot = 0) delict0=3.
EXECUTE.
freq delict0.
compute persona = 0.
if (idnew = lag(idnew)) persona = 1.
exe.
freq var persona.
temporary.
select if persona=0.
freq var delict0.
AGGREGATE
/BREAK=idnew
/delict0 = max(delict0) .
temporary.
select if delict0=sysmis & ytosex=0.
freq persona.
*hierbij is 1 sexoffence, 2 nonsex offence en 3 nonoffence.
TEMPORARY.
select if group = 1 and persona= 0.
freq delict0.
TEMPORARY.
select if group = 2 and persona=0..
freq delict0.
TEMPORARY.
freq delict0.
TEMPORARY.
freq delict0.
TEMPORARY.
freq delict0.
*Nu zou ik binnen elke trajectory (alleen die 10 jaar) het aantal sex, nonsex en non offenders
moeten hebben.
*Ik rond in de tabel op hele getallen af.
temporary.
select if group = 1 & delict0 = 2 and persona=0.
DESCRIPTIVES age gebjp.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
*Nu moet ik per traject gaan kijken naar het percentage sex offenders/nonsex/none offenders
dat in de 5 jaar voorafgaand tenminste een property-violence etc hebben gepleegd.
IF ((ytosex = - 1 & vfviol > 0) | (ytosex = - 2 & vfviol > 0) | (ytosex = - 3 & vfviol > 0) |
(ytosex = - 4 & vfviol > 0) | (ytosex = - 5 & vfviol > 0)) violent=1.
EXECUTE.
IF ((ytosex = - 1 & vfprop > 0) | (ytosex = - 2 & vfprop > 0) | (ytosex = - 3 & vfprop > 0) |
(ytosex = - 4 & vfprop > 0) | (ytosex = - 5 & vfprop > 0)) property=1.
EXECUTE.
IF ((ytosex = - 1 & vfdam > 0) | (ytosex = - 2 & vfdam > 0) | (ytosex = - 3 & vfdam > 0) |
(ytosex = - 4 & vfdam > 0) | (ytosex = - 5 & vfdam > 0)) damage=1.
EXECUTE.
IF ((ytosex = - 1 & vfovsr > 0) | (ytosex = - 2 & vfovsr > 0) | (ytosex = - 3 & vfovsr > 0) |
(ytosex = - 4 & vfovsr > 0) | (ytosex = - 5 & vfovsr > 0)) othercriminallaw=1.
EXECUTE.
IF ((ytosex = - 1 & vfdrugs > 0) | (ytosex = - 2 & vfdrugs > 0) | (ytosex = - 3 & vfdrugs >
0) |
(ytosex = - 4 & vfdrugs > 0) | (ytosex = - 5 & vfdrugs > 0)) drugs=1.
EXECUTE.
IF ((ytosex = - 1 & vfguns > 0) | (ytosex = - 2 & vfguns > 0) | (ytosex = - 3 & vfguns > 0) |
(ytosex = - 4 & vfguns > 0) | (ytosex = - 5 & vfguns > 0)) guns=1.
EXECUTE.
IF ((ytosex = - 1 & vftraf > 0) | (ytosex = - 2 & vftraf > 0) | (ytosex = - 3 & vftraf > 0) |
(ytosex = - 4 & vftraf > 0) | (ytosex = - 5 & vftraf > 0)) traffic=1.
EXECUTE.
IF ((ytosex = - 1 & vfoth > 0) | (ytosex = - 2 & vfoth > 0) | (ytosex = - 3 & vfoth > 0) |
(ytosex = - 4 & vfoth > 0) | (ytosex = - 5 & vfoth > 0)) othernoncriminallaw=1.
EXECUTE.
compute person = 0.
if (ytosex=0) person = 1.
exe.
freq var person.
*oke dus bij ieder unieke persoon is person=1.
freq violent.
AGGREGATE
/BREAK=idnew
/violent = max(violent) .
temporary.
select if person=1.
freq violent.
temporary.
select if person=1.
DESCRIPTIVES violent.
*volgensmij klopt het zo, dus nu voor alle nieuwe variabelen.
AGGREGATE
/BREAK=idnew
/property = max(property) .
AGGREGATE
/BREAK=idnew
/damage = max(damage) .
AGGREGATE
/BREAK=idnew
/othercriminallaw = max(othercriminallaw) .
AGGREGATE
/BREAK=idnew
/drugs = max(drugs) .
AGGREGATE
/BREAK=idnew
/guns = max(guns) .
AGGREGATE
/BREAK=idnew
/traffic = max(traffic) .
AGGREGATE
/BREAK=idnew
/othernoncriminallaw = max(othernoncriminallaw) .
*Percentage SO/NSO/NO per traject dat in de 5 jaar voorafgaan ten minste 1 violence delict
heeft gepleegd.
temporary.
select if person = 1 & group = 1 & delict0 = 1.
freq violent.
temporary.
freq violent.
temporary.
freq violent.
temporary.
select if person = 1 & group = 1.
freq violent.
temporary.
freq violent.
temporary.

freq violent.
temporary.
freq violent.
temporary.
freq violent.
temporary.
freq violent.
temporary.
freq violent.
temporary.
freq violent.
temporary.
freq violent.
temporary.
freq violent.
temporary.
freq violent.
temporary.
freq violent.
temporary.
freq violent.
temporary.
select if person = 1 & delict0 = 1.
freq violent.
temporary.
freq violent.
temporary.
freq violent.
*Percentage SO/NSO/NO per traject dat in de 5 jaar voorafgaan ten minste 1 property delict
heeft gepleegd.
temporary.
freq property.
temporary.
freq property.
temporary.
freq property.
temporary.
freq property.
temporary.
freq property.
temporary.
freq property.
temporary.
freq property.
temporary.
freq property.
temporary.
freq property.
temporary.
freq property.
temporary.
freq property.
temporary.
freq property.
temporary.
freq property.
temporary.
freq property.
temporary.
freq property.
temporary.
freq property.
temporary.
freq property.
temporary.
freq property.
*Percentage SO/NSO/NO per traject dat in de 5 jaar voorafgaan ten minste 1 damage delict
heeft gepleegd.
temporary.
freq damage.
temporary.
freq damage.
temporary.
freq damage.
temporary.
freq damage.
temporary.
freq damage.
temporary.
freq damage.
temporary.
freq damage.
temporary.
freq damage.
temporary.
freq damage.
temporary.
freq damage.
temporary.
freq damage.
temporary.
freq damage.
temporary.
freq damage.
temporary.
freq damage.
temporary.
freq damage.
temporary.
freq damage.
temporary.
freq damage.
temporary.
freq damage.
*Percentage SO/NSO/NO per traject dat in de 5 jaar voorafgaan ten minste 1

othercriminallaw delict heeft gepleegd.
temporary.
freq othercriminallaw.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
*Percentage SO/NSO/NO per traject dat in de 5 jaar voorafgaan ten minste 1 drugs delict
heeft gepleegd.
temporary.
freq drugs.
temporary.
freq drugs.
temporary.

freq drugs.
temporary.
freq drugs.
temporary.
freq drugs.
temporary.
freq drugs.
temporary.
freq drugs.
temporary.
freq drugs.
temporary.
freq drugs.
temporary.
freq drugs.
temporary.
freq drugs.
temporary.
freq drugs.
temporary.
freq drugs.
temporary.
freq drugs.
temporary.
freq drugs.
temporary.
freq drugs.
temporary.
freq drugs.
temporary.
freq drugs.
*Percentage SO/NSO/NO per traject dat in de 5 jaar voorafgaan ten minste 1 guns delict heeft
gepleegd.
temporary.
freq guns.
temporary.
freq guns.
temporary.
freq guns.
temporary.
freq guns.
temporary.
freq guns.
temporary.
freq guns.
temporary.
freq guns.
temporary.
freq guns.
temporary.

freq guns.
temporary.
freq guns.
temporary.
freq guns.
temporary.
freq guns.
temporary.
freq guns.
temporary.
freq guns.
temporary.
freq guns.
temporary.
freq guns.
temporary.
freq guns.
temporary.
freq guns.
*Percentage SO/NSO/NO per traject dat in de 5 jaar voorafgaan ten minste 1 traffic delict
heeft gepleegd.
temporary.
freq traffic.
temporary.
freq traffic.
temporary.
freq traffic.
temporary.
freq traffic.
temporary.
freq traffic.
temporary.
freq traffic.
temporary.
freq traffic.
temporary.
freq traffic.
temporary.
freq traffic.
temporary.
freq traffic.
temporary.
freq traffic.
temporary.
freq traffic.
temporary.
freq traffic.
temporary.
freq traffic.
temporary.
freq traffic.
temporary.
freq traffic.
temporary.
freq traffic.
temporary.
freq traffic.
*Percentage SO/NSO/NO per traject dat in de 5 jaar voorafgaan ten minste 1

othernoncriminallaw delict heeft gepleegd.
temporary.
freq othernoncriminallaw.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
*Volgende stap is kijken hoeveel SO/NSO/NO in de afgelopen jaar alleen een violence, alleen
een property, alleen een damage, of een combinatie van die drie hebben gedaan.
IF (violent = 1 & damage = 0 & property = 0) violentonly=1.

EXECUTE.
IF (violent = 0 & damage = 1 & property = 0) damageonly=1.

EXECUTE.
IF (violent = 0 & damage = 0 & property = 1) propertyonly=1.
EXECUTE.
IF (violent = 1 & damage = 0 & property = 1) violentproperty=1.
EXECUTE.
IF (violent = 1 & damage = 1 & property = 0) violentdamage=1.
EXECUTE.
IF (violent = 0 & damage = 1 & property = 1) damageproperty=1.
EXECUTE.
IF (violent = 1 & damage = 1 & property = 1) violentpropertydamage=1.
EXECUTE.
AGGREGATE
/BREAK=idnew
/violentonly = max(violentonly) .
AGGREGATE
/BREAK=idnew
/damageonly = max(damageonly) .
AGGREGATE
/BREAK=idnew
/propertyonly = max(propertyonly) .
AGGREGATE
/BREAK=idnew
/violentproperty = max(violentproperty) .
AGGREGATE
/BREAK=idnew
/violentdamage = max(violentdamage) .
AGGREGATE
/BREAK=idnew
/damageproperty = max(damageproperty) .
AGGREGATE
/BREAK=idnew
/violentpropertydamage = max(violentpropertydamage) .
temporary.
freq violentonly damageonly propertyonly violentproperty violentdamage damageproperty
violentpropertydamage.
temporary.
select if person = 1 & delict0 = 1 & group = 1.

temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
freq violentpropertydamage.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
temporary.
*NA AL DEZE ANALYSES BLIJKT DAT ER NIET GENOEG HITS ZIJN OM

COMBINATIES ETC TE DOEN, DUS IK GEBRUIK ALLEEN DE VARIABELEN
VIOLENT, DAMAGE EN PROPERTY GEBRUIKEN (DUS AT LEAST 1
VIOLENCE/PROPERTY/DAMAGE DELICT IN AFGELOPEN 5 JAAR)
*Nu ga ik naar de volgende stap; een multinominale logistische regressie.

*Hiervoor ga ik regressies uitvoeren die de kans op SO/NO/NSO berekenen wanneer iemand
at least een violence/property of damage offence hebben gepleegd in afgelopen 5 jaar.
*Dus dependent var = delict 0 (met 3 uitkomsten, vandaar de multinominale logistische
regressie).
*indep var is violent damage en property.
*controle variabele is cohort en age at agetosex = 0.
*controle variabele is ook group, daar aparte dummies voor maken omdat het categoriale
variabelen zijn (dit bleek na regressies per traject uit voeren bleek dat er niet genoeg hits
waren dus het moest als controle variabele.
*daarnaast bleek dat er een probit utigevoerd moest worden omdat assumpties waren violated
in de normale multinominale logistische regressie, er is dus een stata bestand aangemaakt
want dit moest in STATA.
TEMPORARY.
save outfile= 'F:\Final\LOGREGSAMEN.sav'
/keep = delict0 age gebjp_c violent damage property group.
exe.
get file = 'F:\Final\LOGREGSAMEN.sav'.

RECODE violent damage property (SYSMIS=0).
EXECUTE.

recode group (1=1) (else=0) into group1.
SAVE TRANSLATE OUTFILE='F:\Final\PROBIT.dta'

/TYPE=STATA
/VERSION=11
/EDITION=SE
/MAP
/REPLACE.
*Nu gaat de analyse in STATA verder.
*Hieronder nog een syntax die ik had gedraaid voor nog wat descriptives tabellen.
GET
DATASET NAME DataSet9 WINDOW=FRONT.
AGGREGATE
/BREAK=idnew
/totalconvictions = sum(vftot).
freq group.
plegen in jaar 0.
in het jaar 0.

EXECUTE.
freq delict0.
exe.
freq var persona.
temporary.
freq var totalconvictions.
des totalconvictions.
AGGREGATE
/BREAK=idnew
/proportionfreeperyear = sum(free).
temporary.
des proportionfreeperyear.
AGGREGATE
/BREAK=idnew
temporary.
select if ytosex = 0.
freq totalconvictions.
temporary.
select if ytosex=0.
des ytosex vftot free age gebjp.

des violent property damage group delict0 age gebjp_c.
GET
DATASET NAME DataSet9 WINDOW=FRONT.
AGGREGATE
/BREAK=idnew
/totalconvictions = sum(vftot).
freq group.
plegen in jaar 0.
in het jaar 0.

EXECUTE.
freq delict0.
exe.
freq var persona.
temporary.
freq var totalconvictions.
des totalconvictions.
AGGREGATE
/BREAK=idnew
/proportionfreeperyear = sum(free).
temporary.
des proportionfreeperyear.
AGGREGATE
/BREAK=idnew
temporary.
select if ytosex = 0.
2. STATA
net from http://www.andrew.cmu.edu/user/bjones/traj

net install traj, force
help traj
*P&W 2014.
*single trajectory, no splines.

*delinquency frequency, zip model.
*Het beste model blijkt het 5 keer derde order model, deze staat hier bovenaan, de rest wat ik
heb geprobeerd staat er onder.
use "\\Client\F$\Final\SOyearFINALSTATA.dta"
traj, var(vftot_1-vftot_10) indep(ytosex_1-ytosex_10) tcov(free_1-free_10) model(zip)
order(3 3 3 3 3)
trajplot, xtitle(ytosex) ytitle(totcase) ci
sort _traj_Group
by _traj_Group: summarize _traj_ProbG*
sort idnew
matrix list e(plot1)
format idnew %15.0f
format _traj_ProbG* %15.13f
format _traj_Group %4.0f
outsheet idnew _traj_ProbG* _traj_Group using "\\Client\F$\Final\SOtrajecten3.cvs"
export delimited _traj_ProbG* idnew using "\\Client\F$\Final\SOtrajecten4.txt",
novarnames replace datafmt
outfile idnew _traj_ProbG* using "\\Client\F$\Final\SOtrajecten3.txt"
clear
export delimited _traj_ProbG* idnew using "G:\TH3\CCLS 2\CCLS data\Data Diane\final\

SOtrajecten4.txt", novarnames replace datafmt
type data.txt
*nog even een average age crime curve voor entire sample
use "G:\TH3\CCLS 2\CCLS data\Data Diane\final\SOyearFINALSTATA.dta"
order(1)
sort _traj_Group
sort idnew
*outfile idnummer _traj_ProbG* _traj_Group using "\\Client\E$\Final Sample\
SOtrajecten.txt"
clear


order(1)
sort _traj_Group
sort idnew
*outfile idnummer _traj_ProbG* _traj_Group using "G:\TH3\CCLS 2\CCLS data\
Data Diane\SOtrajecten.txt"
clear

order(1 1)
sort _traj_Group
sort idnew
clear

order(1 1 1)
sort _traj_Group
sort idnew
clear

order(1 1 1 1)
sort _traj_Group
sort idnew
clear

order(1 1 1 1 1)

sort _traj_Group
sort idnew
clear

order(1 1 1 1 1 1)
sort _traj_Group
sort idnew
clear

order(2)
sort _traj_Group
sort idnew
clear

order(2 2)
sort _traj_Group
sort idnew
clear

order(2 2 2)
sort _traj_Group

sort idnew
clear

order(2 2 2 2)
sort _traj_Group
sort idnew
clear

order(2 2 2 2 2)
sort _traj_Group
sort idnew
clear

order(2 2 2 2 2 2)
sort _traj_Group
sort idnew
clear

order(3)
sort _traj_Group
sort idnew

clear

order(3 3)
sort _traj_Group
sort idnew
clear

order(3 3 3)
sort _traj_Group
sort idnew
clear

order(3 3 3 3)
sort _traj_Group
sort idnew
clear

order(3 3 3 3 3)
sort _traj_Group
sort idnew

clear

order(3 3 3 3 3 3)
sort _traj_Group
sort idnew
clear
*Nu even vanaf mijn memory stick werken, wel vanaf zelfde bestand.

order(3 3 2 3 2 3)
sort _traj_Group
sort idnew
clear

order(3 3 2 2 2 3)
sort _traj_Group
sort idnew
clear

order(3 3 3 3 2 3)
sort _traj_Group
sort idnew

clear

order(3 3 3 3 2)
sort _traj_Group
sort idnew
clear
*die groep 5 ziet er nog steeds niet goed uit, misschien met startwaarden werken?

order(2 2 2 2 2)
sort _traj_Group
sort idnew
clear

order(1 1 1 1 1)
sort _traj_Group
sort idnew
clear

order(3 3 3 3 1)
sort _traj_Group
sort idnew

clear

order(3 3 3 3 1 3)
sort _traj_Group
sort idnew
clear

order(3 3 3 3 2 3)
sort _traj_Group
sort idnew
clear
*Het lijkt toch alsof er outliers zijn, bij groep 5. Misschien eerst kijken of ik met startwaardes
dit kan oplossen, anders moet ik outliers gaan bekijken

order(3 3 3 3)
sort _traj_Group
sort idnew
clear
*hij pakt nooit de bovenste groep, lijkt dus echt aan outliers te liggen. Ik ga het proberen met
een nieuw databestand zonder outliers
*Als je tot 10 convictions per jaar doet wordt wel de helft van het bestand verwijderd, maar
ga het voor de zekerheid proberen
use "\\Client\F$\Final Sample\SOyear5aOUTLIERS.dta"


order(3 3 3 3 2 3)
sort _traj_Group
sort idnew
clear
*probabilities veel te laag en hij pakt de bovenste lijn alweer niet, lijkt dus niet te werken om
de outliers weg te halen, nog een paar proberen
use "\\Client\F$\Final Sample\SOyear5aOUTLIERS.dta"

order(3 3 3 3 3)
sort _traj_Group
sort idnew
clear
*probabilities goed maar hij pakt de bovenste lijn weer niet. ik ga weer terug naar de originele
data en toch het probleem anders oplossen.
*beste model tot nu toe 333323

order(3 3 3 3 2 3)
sort _traj_Group
sort idnew
clear

order(3 3 3 3 3 3)
sort _traj_Group
sort idnew

clear
*startwaardes van 333323 gebruiken om 333333 te schatten

matrix strt = (1.72, 0.24, 0.07, 0.77, 0.26, 0.08, -0.32, -1.76, -0.38, 5.23, 0.38, 0.03, 0, 0, 0,
3.85, 0.05, -0.11, 4.21, 11.5, 2.94, 63.83, 0.5, 17.11)
order(3 3 3 3 3 3) start (strt)
sort _traj_Group
sort idnew
clear
*moet blijkbaar 36 startwaardes hebben. Oke dit lukt voor nu niet met die
startwaardes. En misschien 5 groepen toch beter gezien het plaatje, de onderste paar lijken
namelijk veel op elkaar Startwaardes lijken ook niet nodig want hij pakt nergens de bovenste
lijn, dat is het enige probleem wat ik nog heb

order(3 3 3 3 3 3)
sort _traj_Group
sort idnew
clear

order(3 3 3 3 3)
sort _traj_Group
sort idnew
clear

order(2 2 2 2 2)
sort _traj_Group
sort idnew
clear

order(2 2 3 2 3)
sort _traj_Group
sort idnew
clear

order(3 2 3 2 3)
sort _traj_Group
sort idnew
clear
*COMMAND VOOR DE REGRESSIE
use "\\Client\F$\Final\PROBIT.dta"
mprobit delict0 damage property violent group1 group3 group4 group5 age gebjp_c
*predicted probabilities berekenen
display normal(0)
display normal(-0.752)
display normal(0)
display normal(0)
display normal(0)

Bachelor Script I e Socio Logie 2

Uploaded by

Copyright:

Available Formats

Bachelor Script I e Socio Logie 2

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Bachelor Script I e Socio Logie 2

Uploaded by

Copyright:

Available Formats

Running head: PREDICTIVE PATTERNS OF SEX OFFENDERS 1

Predictive Patterns of Sex Offenders

Crime Trajectory Analysis Prior to the First Sex Offence

Diane Groothuysen (3788156)

Bachelor thesis Sociology (201100018)

Thesis supervisor: dr. J. E. H. Beijers

Course coordinator: prof. I. Maas

Special thanks to: prof. dr. A. Blokland

receiving the certificate of conduct. ("Terugkijktermijnen", n.d.)

consequences. A lot of criminological research shows that employment is negatively

employment, policy makers could be increasing their recidivism (rates). It is therefore

second research question;

subsequent sex offence?

2. Theoretical Framework and Literature Review

2.1 Comparing Criminal Trajectories

The second group of offenders explained by Moffitt, life-course-persistent offenders,

ways for the research question under study.

literature on this topic will now be discussed to derive a fitting prediction.

2.2 Prior Research on Comparing Criminal Trajectories

sex offenders compared with nonsex offenders will now be discussed.

similar to that of age-matched nonsex offenders. However, there might be other

focus of the second part of the present study.

2.3 Predicting a Sex Offence

Since prior research on predictive patterns in criminal trajectories is limited, there is

applicable theories for deriving hypotheses for the present study.

(Blokland & Lussier, 2015).

offences has been neglected by criminological theory in general (2004).

is called cumulative disadvantage. Cumulative disadvantage is a phenomenon where criminal

However, if we would assume that committing a sex offence happens in a period of

offences such as homicide could also be the result of cumulative disadvantage.

control may result in the occurrence of a sex offence.

2.3 Prior Research on Predictive Offences

To my knowledge, there has been little to no studies on predictive patterns in criminal

and his colleagues’ work in chronological order.

predict a subsequent offence (Soothill & Francis, 2000).

wounding’, robbery, stealing in a dwelling, arson, kidnapping and cruelty to children,

one type of serious offence (Soothill et al., 2008).

hereby meet the shortcoming of information in criminological literature regarding predicting a

mentioned above are as follows;

entire trajectory on their own or be highly concentrated in one or more trajectories. To

be tested in the next sections.

3. Data and Methods

3.1 Original Sample

Retrospective and prospective information such as life-course information (e.g. marriage,

each case represents a year for a person.

3.2 Final Sample

descriptive statistics of these variables are given in Table 1.

Descriptive statistics of used variables

by the previous analysis (trajectory analysis).

log( λ ¿j ¿=β 0j + β 1j Time ¿ + β 2j Time 2¿ + β 3j Time 3¿

BIC = log(L) – 0.5log(n)*(j),

k is the number of parameters (Nagin, 1999).

and the analyses are shown in Appendix A.

group based trajectory analyses.

Average Crime Curve for Entire Sample (N=4142)

posterior probabilities for group assignment are represented in Table 3.

Numerical Values of Parameters Estimates for the Five-group Model

Group 1 Group 2 Group 3 Group 4 Group 5

Note. *=p<.05; **=p<.01; ***=p<.001

Note. =p<.05; =p<.01; =p<.001

F (-.559 + property .343 + age-.011 + cohort * -.029)

F (-.559 + 1.343 + 25-.011 + 9*-.029) =

F (-.559 + property .343 + age-.011 + cohort * -.029)

F (-.559 + 0.343 + 25-.011 + 9*-.029) =

F (-2.108 + property * .649 + age-.009 + cohort .129)

F (-2.108 + 1.649 + 25-.009 + 9*.129) =-225 + 1.161 – 2.108

F (-2.108 + property * .649 + age-.009 + cohort .129)

F (-2.108 + 0.649 + 25-.009 + 9*.129) =