An Investigation into the Relationship between Criminal Behaviour ...

15 downloads 116 Views 185KB Size Report
w ithin the group there is a value system that supports this w ay of sorting out .... based survey (PAPI), a sm all num ber of random ly selected interviews were ..... In Table 8, how ever, w e present the full set of estim ated coefficients for the first.
O FFEN D ERS A S V ICTIM S O F C RIM E ? A N IN V ESTIG A TIO N IN TO TH E R ELA TIO N SH IP B ETW EEN

C RIM IN A L

B EH A V IO U R A N D V ICTIM ISA TIO N $ D erek D eadm an and Ziggy M acD onald* Public SectorEconom ics Research Centre D epartm entofEconom ics U niversity ofLeicester O ctober2001 A bstract In this paperw e considerthe association betw een victim isation and offending behaviourusing data from the Y outh Lifestyles Survey. W e consider the im pact of violent, non-violentand persistentoffending on the probability of being a victim of violentand non-violentcrim e and find a positive association betw een these using univariate probit estim ates.H ow ever,taking into account the endogenous nature of offending and victim isation via a bivariate probit m odel, w e find that univariate estim ates understate the association.W e suggest that policy recom m endations should only be based on the bivariate analysis of the association betw een offending and victim isation.

K eyw ords:V ictim s ofCrim e,O ffenders,Bivariate Probit JEL C lassification: K 42

$

W e are gratefulto the H om e O ffice for providing access to the Y outh Lifestyles Survey,and to Ian Bradley, K evin Lee,Steve Pudney and PSERC sem inar participants at Leicester for helpful com m ents and suggestions. A llerrors and om issions are entirely ourresponsibility. * Corresponding author:em ailziggy.m acdonald@ le.ac.uk.

1

1.Introduction In this paper w e consider som e relatively unexplored factors relating to the determ inants of crim e victim isation. The identification of characteristics of individuals or firm s that suffer disproportionate risks of being victim s of crim e is a long established area of research.O ne defect of this literature is that it overw helm ingly portrays victim s and offenders as separate groups from w ithin the population. H ow ever, there has recently been a sm all num ber of studies of violent offenders w hich have challenged this overly sim plistic view (Jensen & Brow nfield,1986;M ayhew & Elliott,1990;Sam pson & Lauritsen,1990,1994;W ittebrood & N ieuw beerta, 1999; and Pedersen, 2001), and w hich have dem onstrated that offenders also run a greater risk of being victim s of violence than non offenders. W hether this finding generalises to victim s of non violentcrim es is an im portantconsideration,notleastfor policy issues relating to both policing and victim support. A dditionally, one group of victim s, nam ely those w ho have experienced repeat or m ultiple victim isation, have been seen increasingly as a particularly im portant group for policing (Pease,1998) and it is of special interestto considerthe victim /offenderrelationship forsuch persons. In addressing these questions, this paper com plem ents the literature in a num ber of w ays. Firstly w e have explicitly considered the influence of individual crim inality on the probability ofbeing a victim ofeitherviolentand/ornon violentcrim e.Previously,m odels of victim isation have included covariates to capture socio-dem ographic characteristics of the individual and the area in w hich the individual resides (e.g. inner city area), w hich m ay or m ay notactas proxies for crim inality.G iven the nature of our data w e are able to notonly controlfor these characteristics,butalso for self-reported crim inalbehaviour.To explore the resulting issues of victim /offender relationships, this paper uses a rich and inform ative dataset, the 1998 Y outh Lifestyles Survey, which has hitherto not been used to study the process ofcrim e victim isation. The balance of the paper is as follow s.In the nextsection w e consider the factors that are likely to influence the probability of being a victim of crim e, as discussed in the recent literature. Follow ing this w e describe our data set and then proceed to present som e prelim inary analysis. In Section 5 w e present the results of our m ain analysis and our discussion ofthese results.Section 6 concludes.

2

2.Victim isation and O ffending B ehaviour There have been at least three reasons advanced in the literature to explain w hy one m ight observe offenders as running an enhanced risk ofbecom ing a victim ofcrim e.The firstdue to W olfgang and Ferracuti(1967) is related to the purported existence of violentsubcultures in society for w hom retribution for harm done to them as m em bers of this culture is seen as a legitim ate response. V ictim s becom e offenders and, in turn, offenders becom e victim s, as w ithin the group there is a value system thatsupports this w ay ofsorting outdisagreem ents. M ore generalroutine activity and lifestyle theories due to H indelang etal.(1978)and Cohen et al. (1981) are outlined by W ittebrood and N ieuw beerta (1999) and by Pedersen (2001) to explain observed associations betw een levels of offending and victim isation (not necessarily relating to the sam e persons). Sim ply put, routine activity or lifestyle theory suggests that an association w ill be observed if victim s and offenders share sim ilar general lifestyles. It is assum ed that certain lifestyle factors enhance the risk of being an offender. People w ho live in the sam e area and have sim ilar socialand dem ographic characteristics to the offenders they encounteron a day-to-day basis w illrun a higherrisk ofbecom ing a victim of violence than those w ho do notshare these lifestyle features.If this accurately portrays the situation facing offenders,then,as W ittebrood and N ieuw beerta (1999) suggest,an observed general positive correlation betw een victim isation rates of violent crim e and rates of offending is essentially a spurious relationship. A s an exam ple,consider tw o districts in a tow n thatdiffer w ith respectto crim e rates. D istrict one is a poor inner city area w ith high crim e rates and districttw o is a relatively prosperous suburban area w ith low crim e rates.A sam ple of persons from these tw o districts w ould revealboth a higherproportion ofoffenders and victim s in those sam pled from district one com pared w ith district tw o. The apparent positive relationship betw een offending and victim isation is spurious in this case as both are linked to the lifestyle factor ‘district’ and does not im ply that an offender is either m ore or less likely to be a victim once one has controlled for‘district’. This theory is to be distinguished from that w hich asserts that crim inal conduct in itselfexerts an extra and directreason foran observed association.The conductofthe violent offender increases the risk of being a victim of violent crim e ‘because of the m otives, vulnerability or culpability of people involved in those activities’ (Jensen and Brow nfield, 1986).O ffenders are seen as putting them selves m ore frequently atrisk of violence tow ards

3

them than non offenders w ho otherw ise share the sam e socio-dem ographic profiles.They w ill tend to m eet w ith other offenders or engage in activities w ith other offenders, so m aking them selves m ore vulnerable to violent crim e. U sing the exam ple above, in this case the conditionalprobability of being a victim given a districtand being an offender w illbe higher than the conditional probability of being a victim just given district. A positive correlation betw een victim isation and offending should stillexisteven w hen ‘district’ characteristics are controlled for. A dditionally, it m ay also be reasonable to think that offenders w ho are also victim s m ay be less prepared than non offenders to reportto the police any violentcrim inal acts carried out against them . Such a finding w ould be indirect evidence in favour of this theory com pared to the theory based on routine activity orlifestyle as outlined above. This evidence is of particular interest w hen repeat or m ultiple victim s of crim e are considered.The H om e O ffice definition ofrepeatvictim isation (Bridgem an and H obbs,1997) is ‘w hen the sam e person orplace suffers from m ore than one incidentovera specified period of tim e’. Repeat victim isations have becom e recognised as im portant because they account fora disproportionately high num beroftotalvictim isations.Pease (1998,p.3),using evidence from four British Crim e Surveys,indicates thatbetw een 1982 and 1992,on average 41% of property victim isations (excluding vehicle offences) w ere associated w ith the 2% of respondents w ho reported 4 or m ore victim isations. In this sam ple, 84% of respondents reported no property offences against them . For personal crim e (largely violent crim e), the corresponding figure w as 59% oftotalvictim isations suffered by just1% ofrespondents,w ith 92% of respondents reporting no experience of personalcrim e. Pease (1998,p3) states that ‘The im portantconclusions justified by the research to date are thatvictim isation is the best single predictor of victim isation;thatw hen victim isation recurs ittends to do so quickly;that a m ajor reason for repetition is thatoffenders take later advantage of opportunities w hich the first offence throw s up; and that those w ho repeatedly victim ise the sam e target tend to be m ore established in crim e careers than those w ho do not’.Som e evidence in supportof these conclusions is given in Ellingw orth etal.(1995),Ratcliffe and M cCullagh (1998)and O utlaw etal.(1999). The conclusions ofO utlaw etal.(1999) are of particular interest,as they suggestthat single,repeat(the person suffers a repeatof the sam e crim e in a given period) and m ultiple (the person suffers from m ore than one type of crim e in a given period) victim isation are distinct phenom ena that should be considered separately. Repeat property victim isation relates to the com m only held im pression thata property w hich has been burgled m ay w ellbe burgled again (probably by the sam e burglar) once goods have been replaced or w here 4

inform ation about the property (e.g. the existence of som e unusual possessions) has been handed on to other crim inally interested parties. M ultiple victim isation w as found to be a function of individual lifestyle factors (such as being young m ales taking part in dangerous activities) and did not reflect neighbourhood-level variation. The latter w as found to be particularly im portantfor repeatproperty victim isation how ever,along w ith individuallevel predictors (such as ethnicity,sex,and incom e). A s the research above indicates, victim isation and repeat victim isation studies have both concentrated on the individual and local area socio-dem ographic factors to explain outcom es.Clearly,such factors m ustbe allow ed for if one w ishes to isolate a separate effect forthe offending nature orotherw ise ofvictim s.The range and variety ofsuch factors thathas been considered in the victim isation literature is extrem ely large,and is prim arily constrained by the particular features of the data set available. Research in this area has tended to em phasise the role of area characteristics (seen as indicators of social deprivation) upon property crim e victim isation (forexam ple see O sborn etal.,1992;Trickettetal.,1993,1995). Individual or household characteristics have usually been found to be of less im portance in ‘explaining’the incidence ofproperty crim e,although O sborn etal.(1992)and O utlaw etal. (1999) suggest that repeat victim isation is associated w ith key characteristics at the m icro level. A com m on finding in these studies is that less affluent areas are m ost likely to be targeted by burglars,although itm ay be w ealthierpeople in these areas thatbecom e victim s.

3.The D ata Previous em piricalanalysis of property crim e victim isation in the U K has tended to focus on a single year of the British Crim e Survey (Budd,1999),or in som e cases the British Crim e Survey supplem ented w ith area characteristics taken from m atched Census data (O sborn et al.,1992 and Trickett etal.,1995).O ther papers have either used specific household surveys (Fishm an et al., 1998), or in one study, the G eneral H ousehold Survey (M acD onald and Pudney,2000).In this paperourdata are from the 1998 Y outh Lifestyles Survey (Y LS).This is a rich source of inform ation, as it contains inform ation on victim isation and crim inal behaviour.The Y LS is conducted by the N ationalCentre forSocialResearch on behalfofthe H om e O ffice, and is based on a nationally representative sam ple of 4,848 12-30 year olds living in private households in England and W ales. The core sam ple for the Y LS w as achieved by revisiting eligible households w ho w ere interview ed for the 1998 British Crim e

5

Survey.This provided a sam ple of3,643 young people.In addition to this core sam ple a ‘topup’ sam ple w as achieved through focused enum eration and screening of neighbouring addresses. The top-up sam ple resulted in an additional 1,205 interview s, giving a com plete sam ple of 4,848 observations. For m ore details of the survey and the sam pling fram e see Stratford and Roth (1999). In the survey, inform ation on offending behaviour (and other sensitive subjects) is collected via self-com pletion questionnaires, and in m ost cases through Com puter-A ssistedSelf-Interview ing (CA SI). To allow a com parison betw een CA SI and the traditional paperbased survey (PA PI), a sm all num ber of random ly selected interview s w ere based on the latter. For our analysis, because CA SI responses have been found to be m ore accurate (see Flood-Page et al., 2000), w e have chosen to exclude those based on PA PI. D ropping these observations and any w ith m issing values yields a finalsam ple of3,956 observations.

4.Prelim inary analysis To address the questions posed earlier,w e splitour sam ple into those w ho have offended in the pastand those w ho have notusing a H om e O ffice derived variable thatindicates w hether a respondent has adm itted to ever having com m itted any one of 27 core offences covered. These offences relate to crim inaldam age (tw o),property offences (fifteen),fraud (four) and violent offences (six),but exclude ‘low level’ or trivial offences.Q uestions w ere w orded to resem ble the legal definition of offences as far as possible and w ere intended to relate to incidents w here the respondent intended harm or dam age. Theft, outside of shoplifting, related only to incidents w here the w orth of stolen item s w as in excess of £5.Tw o of the six questions pertaining to violentoffences related to incidents w here the victim required m edical attention.D rug and sexualoffences w ere notcovered.Based on these classifications,in our sam ple 1,798 individuals can be broadly defined as offenders and 2,158 as non-offenders. W ith respect to victim s of crim e,there are three victim isation questions in the Y LS, butw e concentrate on the follow ing tw o:1

1

The third m ain victim isation question concerns robbery, but the num bers reporting to being a victim of this offence are too sm allfor our analysis.In addition,respondents under the age of 16 are asked w hether they have been a victim of sex crim e, but w e exclude this from our analysis, as there are obvious questions about the reliability ofresponses to this question.

6



In the last12 m onths w hen you w ere out(notathom e),has anyone STO LEN anything of yours thatyou had leftsom ew here (e.g.from school,a cloakroom ,an office,a car oranyw here else you leftit)?



In the last 12 m onths w hen you w ere out (aw ay from your hom e), has anyone deliberately done any of the follow ing: kicked you, hit you w ith their fists or w ith a w eapon ofany sort,slapped orscratched you,orused force orviolence againstyou in any otherw ay?

Respondents answ ering yes to question 1 are defined as being a ‘victim of theft from the person’, w hilst individuals responding yes to question 2 are defined as being ‘a victim of assault’. O f the 1,798 respondents defined as offenders, 592 (32.9% ) have been a victim of either assault or theft or both, w hereas 415 (19.2% ) of the 2,158 non-offenders have been victim s. This significant difference in victim isation (t = 9.84) suggests a strong association being offending behaviour and victim isation.In Table 1 w e break these figures dow n further. H ere w e report the num bers of offenders and non-offenders w ho have been victim s of only assault,ofonly theft,and ofboth assaultand theft.

Table 1.G eneralvictim isation rates foroffenders/non-offenders (% )$ N everO ffended

O ffended Ever

7.0

13.5

(0.551)

(0.805)

10.0

14.2

(0.645)

(0.823)

2.2

5.3

(0.318)

(0.528)

2158

1798

V ictim ofonly assault V ictim ofonly theft V ictim ofassaultand theft O bservations $

N ote:Standard errors in parenthesis

7

Table 1 illustrates thatthose in the sam ple w ho adm itted to having evercom m itted one of the nam ed crim inalacts w ere disproportionately m ore likely to also be a victim of assault, theftor both.In each case,the difference in the proportion of the sam ple victim ised betw een offenders and non-offenders is statistically significant at the conventional 5% level of significance.Forassaultonly,the t-value is 6.7,fortheftonly the t-value is 4.1 and forassault and theftthe t-value is 5.1. This pattern of victim isation in relation to offending behaviour is one thatappears to be established relatively early in life.The Y LS sam ple can be furtheranalysed to include only those in the sam ple currently at school (including sixth form students). Table 2 reports the findings for assault,theftand both assaultand theftfor this group.In each case victim isation rates for schoolchildren are statistically significantly greater for those adm itting to crim inal offences than for those w ho did not.The t-values here are 3.0 for assaultonly,2.3 for theft only and 2.3 for assaultand theft.Taken together,201 outof 757 non-offenders w ere victim s of the nam ed crim es (26.5% ) w hereas 173 out of 429 offenders (40.3% ) w ere also victim s. The overallt-value forthe difference in these proportions is 4.9.

Table 2.V ictim isation rates forschoolchildren (% )$ N everO ffended

O ffended Ever

9.2

14.9

(1.054)

(1.722)

13.3

18.4

(1.237)

(1.874)

4.0

7.0

(0.710)

(1.233)

757

429

V ictim ofonly assault V ictim ofonly theft V ictim ofassaultand theft O bservations $

N ote:Standard errors in parenthesis

Section 2 reported on som e ofthe published w ork thathad identified an increased risk of being a victim of violent crim e w ith being an offender of violent crim e.It seem s useful, therefore,to exam ine the evidence in the Y LS relating explicitly to those in the sam ple w ho

8

self reported violentoffences. Prelim inary analysis of the Y LS data for those w ho adm itted being offenders ofassaultadds supportforthese earlierfindings relating to violentcrim e.For instance,G ottfredson (1984)w orking w ith an early sw eep ofthe British Crim e Survey,found that of those in the sam ple w ho had com m itted at least one violent crim e, 42% w ere also victim s of violent crim es. This could be contrasted w ith those people w ho had never com m itted a violentcrim e ofw hom only 6% had been victim s ofviolentcrim e. H ow ever,w hathas received very little attention in the literature is the com plim entary enhanced risk of violent (and non-violent) offenders being victim s of non-violentproperty crim e (specifically theft). Table 3 illustrates this point. The Y LS sam ple w as split for selfreporting offenders betw een those w ho reported violent offences (som e of w hom w ill also have reported to non-violent offending) and those w ho reported only non-violentoffences. Both violentand non-violentoffenders w ere significantly m ore likely to be victim s ofviolent crim e than non-offenders (line 1 in Table 3).Interestingly,both groups w ere also m ore likely than non-offenders to be victim s of theft,or of both assaultand theft(lines 2 and 3 in Table 3).

Table 3.V ictim isation rates forviolent/non-violentoffenders and non-offenders (% )$

V ictim ofonly assault V ictim ofonly theft V ictim ofassaultand theft O bservations $

N ever

N on-violent

V iolent

O ffended

O ffender

O ffender

7.0

10.2

19.2

(0.551)

(0.893)

(1.553)

10.0

13.5

15.3

(0.645)

(1.001)

(1.420)

2.2

3.2

9.0

(0.318)

(0.519)

(1.127)

2158

1153

645

N ote:Standard errors in parenthesis

A lso noted in Section 2 w as the grow ing interestshow n to the problem ofm ultiple and repeatvictim isations.The Y LS survey data is broadly in line w ith the British Crim e Survey

9

figures reported in Section 2 for repeat victim isation. For assault, 57% of offences w ere suffered by the 2% of respondents w ho reported 4 or m ore assaults on them in the previous year.Fortheft,21% ofoffences w ere on the 0.8% ofrespondents w ho selfreported 4 orm ore property offences in the year.Table 4 indicates thatviolentoffenders are substantially m ore likely than non-violentornon offenders to be repeatvictim s ofboth assaultand theft.A s w as the case forTable 3,violentoffenders m ay also have adm itted to non-violentoffences.

Table 4.Single and RepeatV ictim isation(% )$

V ictim ofonly one assault V ictim ofonly one theft V ictim ofm ore than one assault V ictim ofm ore than one theft O bservations $

N ever

N on-violent

V iolent

O ffended

O ffender

O ffender

3.9

6.0

9.5

(0.416)

(0.699)

(1.153)

7.8

10.1

9.9

(0.578)

(0.886)

(1.178)

3.8

5.0

13.8

(0.409)

(0.644)

(1.359)

3.2

5.1

10.5

(0.373)

(0.644)

(1.186)

2158

1153

645

N ote:Standard errors in parenthesis

W e have seen in this section thatthere appears to be an association betw een offending behaviour and victim isation. These sim ple descriptive statistics provide m otivation for studying the factors thatinfluence the probability of being a victim in m ore detail.W hether this evidence supports either the lifestyle or the crim inal conduct theories of victim isation above, or neither, needs to be addressed through a statistical analysis that controls for the lifestyle factors ofvictim s explicitly.In the nextsection w e consideran em piricalapproach to the current sam ple that provides results from m ultivariate m odels that help clarify this problem .

10

5.R esults 5.1 U nivariate Probits The probability of the discrete eventof being a victim of crim e is m ostnaturally m odelled as a probit(or logit) relation.W e denote an individual’s propensity to be a victim of crim e w ith the latent variable vi* , w hich is related to the observed individual and area characteristics through the structuralm odel: vi* = X ib 1 + cid + e1i

(1)

w here X i is a vectorofpersonal,dem ographic and lifestyle attributes forindividuali,ci is an indicator variable for w hether the individualhas engaged in crim inalbehaviour,b and δ are the param eters to be estim ated,and e1i is a norm ally distributed errorterm w ith m ean zero and variance one,thatcaptures the unobserved determ inants of victim isation.The latentvariable vi* drives the observed outcom e ofbeing a victim , vi,through the m easurem entequation:

⎧1 if vi* > 0 vi = ⎨ * ⎩0 if vi ≤ 0

(2)

Estim ation of(1)as a probitm odelis straightforw ard,and provides us w ith directm easures of the im pactofthe various explanatory variables on the likelihood ofbeing a victim ofcrim e. In Tables 5 and 6 w e presentthe results forourestim ated m odels forvictim isation and repeatvictim isation respectively.In each case w e estim ate m odels forvictim s ofassaultonly, theft only and assault and theft (m ultiple victim isation). W e control for personal characteristics (e.g. age, gender, ethnicity, having children, m arital status, etc), area characteristics (including region and m easures of social deprivation),risk factors related to being outside the hom e (e.g. participation in sport and social activities), and offending behaviour.The base categories are:single,fem ale,‘other’ ethnic origin,w ith no children,not born in U K ,in w ork and having qualifications,living in non ow ner-occupied property in an inner city area of London that is not considered deprived. D escriptive statistics for all the variables used in this analysis are given in A ppendix Table A 1.

11

Table 5.Probitestim ates ofthe probability ofbeing a victim Covariate PersonalC haracteristics Age M ale H ave atleastone child H as currentpartner W hite origin Black origin A sian origin N ative born U nem ployed N o qualifications A tschool O w neroccupier A rea C haracteristics N orth ofEngland Y orkshire/H um berside N orth W estEngland EastM idlands W estM idlands EastA nglia South EastEngland South W estEngland W ales U rban area Ruralarea A corn 17 m ostdeprived People w ish to leave area R isk Factors A ctive in com m unity Sports participation Socialactivities H angouton street W as bullied atschool G oes outalone atnight Carries personalalarm Thinks judges outoftouch O ffending behaviour N on-violentoffender V iolentoffender Persistentoffender Intercept Log Likelihood Chi-squared (d.f.) O bservations

A ssaultO nly β t-value

TheftO nly A ssaultand Theft β t-value β t-value

-0.032 0.411 -0.058 -0.070 0.480 0.370 0.312 0.149 -0.018 -0.120 -0.214 -0.044

3.70 5.95 0.61 1.06 1.73 1.09 0.96 0.94 0.13 0.93 2.04 0.66

-0.012 -0.027 0.017 0.083 -0.077 0.018 0.163 0.039 0.216 0.196 0.275 -0.041

1.48 0.44 0.21 1.32 0.42 0.08 0.76 0.32 1.78 1.82 2.73 0.67

-0.056 0.219 0.347 0.077 -0.410 -0.285 -0.395 -0.370 0.276 0.089 -0.095 0.010

4.19 2.26 2.47 0.80 1.87 0.98 1.39 2.31 1.55 0.54 0.66 0.11

0.230 0.237 0.260 0.045 0.109 0.220 -0.061 0.241 0.213 0.135 0.285 -0.322 0.099

1.66 1.89 2.12 0.32 0.82 1.37 0.50 1.70 1.45 1.74 2.07 2.19 1.37

-0.254 0.000 0.033 -0.178 -0.035 0.008 0.042 -0.205 0.007 0.080 -0.022 0.190 0.142

1.87 0.00 0.31 1.42 0.31 0.06 0.42 1.52 0.05 1.14 0.16 1.32 2.14

-0.117 0.144 -0.195 0.066 -0.224 0.081 0.049 -0.998 -0.072 -0.157 -0.181 0.125 0.014

0.59 0.92 1.12 0.38 1.20 0.38 0.32 2.63 0.35 1.51 0.80 0.53 0.14

0.006 -0.023 -0.098 0.121 0.357 0.056 0.281 0.176

0.07 0.33 1.00 1.56 5.94 0.79 2.14 2.59

0.078 0.180 0.044 0.035 0.089 0.190 -0.049 0.123

1.12 2.86 0.48 0.47 1.56 2.99 0.38 1.95

0.190 0.175 -0.043 0.096 0.403 0.018 -0.215 0.103

1.86 1.70 0.31 0.91 4.72 0.18 0.85 1.04

0.145 2.04 0.201 3.18 0.201 1.88 0.316 3.80 0.212 2.67 0.537 4.74 0.393 3.80 -0.165 1.45 0.260 1.91 -1.856 4.91 -1.527 5.37 -0.582 1.45 -1149.91 -1390.76 -528.66 265.21 (36) 102.73 (36) 173.01 (36) 3956 3956 3956

12

Table 6.Probitestim ates ofthe probability ofbeing a repeatvictim Covariate PersonalC haracteristics Age M ale H ave atleastone child H as currentpartner W hite origin Black origin A sian origin N ative born U nem ployed N o qualifications A tschool O w neroccupier A rea C haracteristics N orth ofEngland Y orkshire/H um berside N orth W estEngland EastM idlands W estM idlands EastA nglia South EastEngland South W estEngland W ales U rban area Ruralarea A corn 17 m ostdeprived People w ish to leave area R isk Factors A ctive in com m unity Sports participation Socialactivities H angouton street W as bullied atschool G oes outalone atnight Carries personalalarm Thinks judges outoftouch O ffending behaviour N on-violentoffender V iolentoffender Persistentoffender Intercept Log Likelihood Chi-squared (d.f.) O bservations

A ssaultO nly β t-value

TheftO nly A ssaultand Theft β t-value β t-value

-0.057 0.459 0.225 -0.142 -0.035 -0.088 -0.225 0.010 0.033 0.155 -0.173 -0.044

5.05 5.38 1.87 1.74 0.14 0.26 0.70 0.05 0.20 1.11 1.40 0.54

-0.025 0.133 0.184 0.069 -0.256 -0.111 -0.098 -0.241 0.280 0.196 0.347 0.018

2.14 1.57 1.54 0.80 1.21 0.41 0.38 1.64 1.78 1.39 2.53 0.21

-0.033 0.277 0.736 0.012 -0.362 -0.542 -0.234 0.021 -0.277 0.494 0.484 0.231

1.28 1.50 2.90 0.07 0.97 0.95 0.49 0.06 0.63 1.96 1.71 1.27

0.287 0.265 0.135 0.158 -0.014 0.159 0.032 0.137 0.223 0.113 -0.052 0.113 0.057

1.75 1.77 0.89 0.96 0.09 0.81 0.22 0.78 1.27 1.22 0.27 0.57 0.66

-0.332 0.054 -0.133 0.048 -0.172 0.054 -0.005 -0.618 -0.235 -0.052 -0.241 0.293 0.223

1.72 0.38 0.90 0.31 1.08 0.29 0.04 2.60 1.21 0.55 1.08 1.27 2.61

-0.131 0.025 -0.558 -0.031 -0.680 -0.192 -0.561 0.012 -0.050 0.210 0.314

0.43 0.10 1.70 0.12 1.89 0.53 1.96 0.07 0.11 0.43 1.82

0.088 -0.085 -0.054 0.028 0.415 0.116 0.257 0.139

0.93 1.01 0.47 0.31 5.75 1.36 1.55 1.67

0.090 0.070 -0.200 0.019 0.305 0.109 -0.032 0.148

0.95 0.80 1.79 0.19 4.02 1.24 0.17 1.72

0.015 0.005 0.253 -0.161 0.396 0.209 0.316 0.020

0.08 0.03 0.92 0.80 2.54 1.14 0.96 0.11

0.130 1.45 0.287 3.15 0.001 0.01 0.443 4.55 0.531 5.23 0.456 2.24 0.436 3.78 0.062 0.47 0.541 2.50 -1.179 3.01 -1.279 3.53 -2.665 3.44 -753.67 -682.92 -151.59 236.53 (36) 158.55 (36) 68.95 (34) 3956 3956 3428

13

The figures in Table 5 are quite revealing about the association betw een offending behaviour and victim isation,once other lifestyle factors have been controlled for.Regardless of how victim isation is defined, there appears to be a positive and statistically significant association betw een offending behaviour and the risk of being a victim . W ith respect to victim s of assault only, it appears that violent or persistent offending are m ore statistically significantpredictors of violentvictim isation than non-violentoffending.For victim s of theft only, non-violent and violent offending appear m ore im portant than persistent offending, w hereas violentoffending is the m oststatistically significantfactorassociated w ith the risk of being a m ultiple victim ofassaultand theft. Before w e consider the results for repeatvictim isation itis w orth m entioning som e of the other factors that are significantly associated w ith the probability of being a victim . Considering personalcharacteristics,these only appearim portantin the firstand third m odels (assault only or assault and theft). For these tw o m odels there is a statistically significant negative association betw een age and victim isation (in the theftonly m odelthe coefficienton age is negative butnotsignificant),and m ales appear m ore likely than fem ales to be victim s of assault only or assault and theft. Interestingly, individuals at school are less likely than those not currently at school to be victim s of assault only,but m ore likely to be victim s of theft only. W ith respect to factors that indicate an individual’s exposure to risk, those w ho w ere bullied at school appear m ore likely to be victim s of assault w hen com pared to those w ho w ere never bullied. It also appears that individuals w ho think judges are out of touch w ith ordinary people tend to have a higherprobability ofbeing a victim ofeithertheftonly or assault only (although this variable is potentially endogenous), w hilst individuals w ho actively engage in sport or w ho go out alone at night are m ore likely to be victim s of theft only. G enerally, regional or area characteristics are not significant. This m ay be due to the relatively w ide m easures used in the analysis, w hich fail to capture the essentially local effects thatm ay affectbehaviourofthe relatively young sam ple underinvestigation. It is im portant to note that w hen the offending variables are excluded from all three m odels reported in Table 5, not m uch changes in term s of the lifestyle and personal characteristics that are associated w ith victim isation (these results are not reported in detail here). For the assault only m odel, the exclusion of offending variables results in only one further lifestyle factor (hanging outin the street) becom ing statistically significant,w hilstfor the theft only m odel being involved in sport becom es significant, and for the m ultiple victim isation m odel (assault and theft) the estim ated coefficients on sports participation and hanging outin the street,becom e statistically significantatthe 10% levelorless. 14

The results for repeat victim isation given in Table 6 also support the strong association betw een offending behaviour, particularly violent offending, and an increased likelihood of victim isation.In addition,having atleastone child and having been bullied at school appear as statistically significant factors determ ining repeat victim isation. W hen com pared to non-offenders,violentoffenders are m ore likely to be repeatvictim s of assault, theft,or assaultand theft.Interestingly,non-violentoffending is only significantly associated w ith being a repeat victim of theft only, w hilst persistent offending appears to have a significant im pact on the risk of being a repeat victim of assault only and m ultiple victim isation. 5.2 B ivariate Probits The results presented above provide a strong case in supportofthe theory thatthere is a direct link betw een offending behaviour and the risk of victim isation,once lifestyle characteristics are controlled for. U nfortunately, there is a potential bias in the univariate probit estim ates due to the likely overlap in unobserved characteristics that determ ine both offending behaviour and the likelihood of being a victim . This potential for unobserved heterogeneity w ill result in the error term , e1i in (1), being correlated w ith the explanatory variable(s) capturing offending behaviour.If this is the case,offending w ill not be exogenous, and the coefficients on the offender variables in the probitm odels w illbe biased,capturing notonly the true effect of being an offender but also the effect on victim isation of having this unobservable characteristic.Previous studies have failed to address this potentialbias. Estim ating the relationship betw een victim isation and offending as a bivariate probit can overcom e this problem (G reene,1997).The em piricalspecification ofthe bivariate m odel is as follow s, vi* = a 1 + X ib1 + cid + e1i

(3)

ci* = a 2 + X ib 2 + Z ix + e2i

(4)

w here the error term s e1i and e2i are jointly distributed as bivariate norm alw ith m eans zero, unitvariances,and correlation r.The variables vi,ci and Xi are as before, Zi is a vector of identifying restrictions,and b1,b2,d and x are the param eters of interest that w e w ish to 15

estim ate.O ne practicaldifficulty w e face in trying to estim ate the bivariate probitis finding a set of identifying restrictions that are significant determ inants of the endogenous variable(s) butalso orthogonalto the residuals ofthe m ain equation (i.e.notsignificantly associated w ith the probability of being a victim ).In order to estim ate the bivariate probit,w e have included the follow ing in Zi: expulsion from school and truancy,pacifism ,excessive drinking,drug use,view s on the courts,contactw ith people in trouble,and having no fatherw hen a teenager (13 variables in total).2 In table 7, in order to save space w e present a sum m ary of the key results from the bivariate m odels w e have estim ated, alongside the equivalent univariate estim ates. In this table w e only consider the im pact of estim ating the bivariate m odel on the coefficient for offending behaviour, plus w e provide the estim ated value for the correlation betw een error term s (r).In Table 8,how ever,w e presentthe fullsetof estim ated coefficients for the first tw o of these m odels (assaultonly-violentoffender and theftonly-non-violentoffender).Full results are available from the authors. The results reported in Table 7 show thatthe univariate estim ates ofthe coefficienton offending behaviourare quantitatively sm allerthan the bivariate estim ates.In addition,forall the m odels estim ated, there appears to be a significant negative correlation at the 10% significance levelor less betw een the error term s of the tw o equations (3)-(4).This suggests thatthe unobserved heterogeneity influencing the probability ofbeing a victim is significantly and negatively associated w ith the unobserved influences on the likelihood of being an offender.Thatis,there are unobserved factors (possibly personalcharacteristics) w hich both raise the probability of an individualbecom ing a victim (and a repeatvictim ) w hilstlow ering the probability of being an offender, or vice versa. This negative correlation explains the increase in the m agnitude of the coefficientestim ates for offending behaviour in the bivariate probit m odels com pared w ith those for the univariate probit analysis,and suggests that any policy recom m endations com ing from this type ofw ork should only be based on the bivariate analysis. Looking at the figures in Table 8 to com pare the results of the univariate and bivariate m odels,itis clear thatare very few changes in term s of significantcoefficients.In m any cases there is a slightreduction in the size of the t-values in the bivariate m odels,such thatfor assaultonly,hanging outin the streetbecom e only m arginally significant(t= 1.67). 2

Likelihood ratio tests w ere conducted forallthe m odels reported in Table 5.In fouroutofsix cases there w as no significantdifference (atthe 5% level)in the log likelihood betw een the m odels w ith and w ithoutidentifying restrictions in the victim isation equation.In only tw o cases (assaultand theft/any offence,repeattheftonly/nonviolentoffender)w ere the identifying restrictions rejected.In allotherrespects,how ever,the results forthese tw o cases are com pletely consistentw ith the otherresults reported.

16

The only other difference is thatage becom es significantin the bivariate estim ate of the theft only m odel,as do being unem ployed and having no qualifications,w hich w ere previously of m arginal significance. A dditionally, one m ay note sm all differences betw een the univariate estim ates in Table 8 and those reported earlier in Table 5 because the m odels reported in the form eronly have one offendervariable,ratherthan three.

Table 7.Sum m ary ofunivariate and bivariate estim ates$ vi= assaultonly

vi= theftonly

vi= assaultortheft

ci= violentoffender

ci= non-violentoffender

ci= any offence

U nivariate Bivariate U nivariate



$

Bivariate

0.627

0.134

0.772

0.347

0.939

(4.56)

(3.31)

(2.31)

(3.13)

(7.22)

(9.31)

-0.189

-0.384

-0.427

(1.69)

(2.45)

(5.67)

vi= repeatassaultonly

vi= repeattheftonly

vi= repeatassaultortheft

ci= violentoffender

ci= non-violentoffender

ci= any offence

U nivariate Bivariate U nivariate



U nivariate

0.327





Bivariate

Bivariate

U nivariate

Bivariate

0.475

1.079

0.088

0.874

0.411

0.869

(5.81)

(5.48)

(1.11)

(2.96)

(6.46)

(6.42)

-0.375

-0.454

-0.329

(3.13)

(2.55)

(3.50)

N ote:A bsolute t-values in parenthesis

17

Table 8.Fullresults forunivariate and bivariate estim ates A ssaultO nly U nivariate Bivariate β |t| β |t|

Covariate PersonalC haracteristics Age -0.033 3.82 M ale 0.425 6.19 H ave atleastone child -0.042 0.44 H as currentpartner -0.056 0.85 W hite origin 0.471 1.72 Black origin 0.366 1.10 A sian origin 0.269 0.84 N ative born 0.175 1.12 U nem ployed -0.010 0.07 N o qualifications -0.114 0.89 A tschool -0.248 2.38 O w neroccupier -0.042 0.64 A rea C haracteristics N orth ofEngland 0.226 1.64 Y orkshire/H um berside 0.232 1.86 N orth W estEngland 0.254 2.08 EastM idlands 0.058 0.41 W estM idlands 0.120 0.92 EastA nglia 0.212 1.33 South EastEngland -0.074 0.61 South W estEngland 0.236 1.68 W ales 0.205 1.42 U rban area 0.117 1.52 Ruralarea 0.246 1.80 A corn 17 m ostdeprived -0.296 2.03 People w ish to leave area 0.102 1.43 R isk Factors A ctive in com m unity -0.003 0.04 Sports participation -0.023 0.33 Socialactivities -0.079 0.81 H angouton street 0.173 2.26 W as bullied atschool 0.351 5.86 G oes outalone atnight 0.076 1.09 Carries personalalarm 0.272 2.07 Thinks judges outoftouch 0.191 2.83 O ffending behaviour N on-violentoffender V iolentoffender 0.327 4.56 Intercept -1.810 4.84 rˆ Log Likelihood -1160.33 Chi-squared (d.f.) 244.36 (34) O bservations 3956

TheftO nly U nivariate Bivariate β |t| β |t|

-0.030 0.377 -0.052 -0.070 0.461 0.334 0.265 0.155 -0.016 -0.126 -0.217 -0.029

3.43 5.10 0.55 1.05 1.70 1.00 0.84 0.99 0.11 0.99 2.06 0.44

-0.012 -0.002 0.023 0.091 -0.072 0.034 0.174 0.050 0.208 0.197 0.268 -0.050

1.55 0.03 0.27 1.46 0.40 0.15 0.81 0.41 1.72 1.83 2.66 0.83

-0.016 -0.001 0.007 0.070 -0.085 0.046 0.180 0.001 0.222 0.220 0.277 -0.047

2.03 0.01 0.09 1.14 0.48 0.20 0.85 0.01 1.87 2.09 2.82 0.80

0.232 0.231 0.248 0.054 0.118 0.196 -0.070 0.245 0.219 0.123 0.263 -0.291 0.094

1.69 1.85 2.04 0.39 0.91 1.23 0.58 1.75 1.51 1.60 1.93 2.01 1.32

-0.263 -0.003 0.036 -0.180 -0.039 0.020 0.039 -0.213 -0.002 0.079 -0.033 0.192 0.146

1.94 0.02 0.33 1.44 0.34 0.14 0.39 1.58 0.02 1.11 0.23 1.34 2.22

-0.241 0.002 0.042 -0.180 -0.025 -0.005 0.045 -0.197 0.009 0.096 0.025 0.143 0.134

1.82 0.02 0.41 1.47 0.22 0.03 0.47 1.49 0.07 1.39 0.18 1.01 2.07

-0.007 -0.035 -0.080 0.133 0.347 0.052 0.276 0.171

0.08 0.52 0.82 1.67 5.81 0.74 2.12 2.51

0.076 0.183 0.051 0.050 0.093 0.205 -0.051 0.132

1.09 0.099 2.93 0.183 0.56 0.015 0.69 0.005 1.64 0.074 3.26 0.160 0.40 -0.053 2.10 0.111

1.44 2.99 0.16 0.07 1.32 2.50 0.42 1.79

- 0.134 2.31 0.627 3.31 -1.836 4.94 -1.513 5.32 -0.189 1.69 -2522.87 -1394.59 862.26 (80) 95.07 (34) 3956 3956

18

0.772 3.13 -1.456 5.20 -0.384 2.45 -3606.01 434.65 (80) 3956

6.C oncluding R em arks In this paper we have used data from the Y outh Lifestyles Survey (Y LS) to explore the determ inants of crim e victim isation.W e have considered the relationship betw een offending behaviour and being a victim of crim e, and found that sim ple cross-tabulations suggest a strong association betw een these variables. In particular, w e found that violent and nonviolent offenders w ere significantly m ore likely to be victim s of violent crim e than nonoffenders (see Table 3),and thatboth groups w ere also m ore likely than non-offenders to be victim s oftheft,orofboth assaultand theft. To explore these associations further w e estim ated univariate probit m odels, w hich indicated a range of personal,area and risk characteristics w hich influence the probability of being a victim (or repeatvictim ) of violence,theftor both.The m odels w hich also included self reported offending variables consistently indicated the enhanced probability of being a victim for those w ho adm itted to som e type of offending in the past.In so far as lifestyle and otherfactors have been controlled forby the othervariables included in these equations,these results provide strong evidence in favour of there being an additional risk to offenders of becom ing a victim through the conductofthe offenders themselves.The observed association betw een offending and victim isation is not a spurious relationship, therefore. O ne potential w eakness in interpreting the results in this w ay is that the offending variables m ight them selves be endogenously determ ined by,in part,the sam e lifestyle and otherfactors w hich determ ine victim isation.This w ould bias the coefficientvalues on allvariables,including the offending variables,in the univariate probit. In order to address this potential problem ,w e estim ated bivariate probit m odels for victim isation and offending.Rather than reduce the estim ated effect of offending behaviour on victim isation, the bivariate results are even m ore strongly in favour of there being an increased probability of being a victim of either violentor non violentcrim e of an individual w ho has adm itted to offending behaviourin the pastthrough the individualbehaviourofthose persons. The separation ofthe young population betw een those w ho are victim s ofcrim e and those w ho are offenders is nota separation thatcan be supported by this analysis.

19

R eferences Bridgem an,C.and H obbs,L.(1997), Preventing RepeatVictim isation; The Police O fficer’s G uide,London:H om e O ffice. Budd,T.(1999), Burglary of D om estic D wellings: Findings from the British Crim e Survey, H om e O ffice StatisticalBulletin,4/99,H om e O ffice,London. Cohen,L.E.,K luegel,J.R.and Land,K .C.(1981),‘SocialInequality and Predatory Crim inal V ictim ization: A n Exposition and Test of a Form al Theory’, Am erican Sociological Review,vol.46(5),pp.505-524. Ellingw orth,D .,Farrell,G .and Pease,K .(1997),‘A V ictim is a V ictim is a V ictim :Chronic V ictim isation in Four Sw eeps of the British Crim e Survey’, British Journal of Crim inology,vol.35,pp.360-365. Fishm an,G .,H akim ,S.and Shachm urove,Y .(1998),‘The U se of H ousehold Survey D ata – The Probability of Property Crim e V ictim isation’, Journal of Econom ic and Social M easurem ent,vol.24,pp.1-13. Flood-Page, C., Cam pbell, S., H arrington, V . and M iller, J.(2000), Youth Crim e,Findings from the 1998/99 Youth Lifestyles Survey,H om e O ffice Research Study 209,London: H om e O ffice. G ottfredson,M .R.(1984),Victim s ofCrim e:The D im ensions ofRisk,London:H om e O ffice. G reene,W .(1997),Econom etric Analysis,London:Prentice H all. H indelang, M ., G ottfredson, M .R. and G aofalo, J. (1978), Victim s of Personal Crim e: An Em piricalFoundation for a Theory ofPersonalVictim ization,Cam bridge:Ballinger. Jensen, G .F. and Brow nfield, D . (1986), ‘G ender, Lifestyles, and V ictim ization: Beyond Routine A ctivity’,Violence and Victim s,vol.1,pp.85-99.

20

M acD onald, Z. and Pudney, S. (2000), The V ictim s of Property Crim e, In: M acD onald, Z. and Pyle, D . (eds) Illicit Activity: the Econom ics of Crim e, D rugs and Tax Fraud, A shgate Publishers,pp.151-172. M ayhew ,P.and Elliott,D .(1990),‘Self-Reporting O ffending,V ictim ization,and the British Crim e Survey’,Violence and Victim s,vol.5,pp.83-96. O sborn,D .R.,Trickett,A .and Elder,R.(1992),‘A rea Characteristics and RegionalV ariates as D eterm inants of A rea Property Crim e Levels’,JournalofQ uantitative Crim inology, vol.8,pp.265-85. O utlaw ,M .,Ruback,R.B.and Britt,C.(1999),Repeatand M ultiple Victim izations: The Role ofIndividualand ContextualFactors,M im eo,(Pennsylvania State U niversity). Pease,K .(1998),RepeatVictim isation: Taking Stock,Crim e D etection and Prevention Series Paper90,London:H om e O ffice. Pederson,W .(2001),‘A dolescentV ictim s of V iolence in a W elfare State’,British Journalof Crim inology,V ol.41,pp.1-21. Ratcliffe, J.H . and M cCullagh, M .J. (1998), ‘Identifying Repeat V ictim ization w ith G IS’, British JournalofCrim inology,V ol.38,pp.651-662. Sam pson, R.J. and Lauritsen, J.L. (1990), ‘D eviant Lifestyles, Proxim ity to Crim e,and the O ffender-V ictim link in Personal V iolence’, Journal of Research in Crim e and D elinquency,V ol.27,pp.110-139. Sam pson,R.J.and Lauritsen,J.L.(1994),V iolentV ictim ization and O ffending:Individual-, Situational-, and Com m unity-levelRisk Factors. In: Reiss, A .J. and Roth, J.A . (eds), U nderstanding and Preventing Violence: Social Influences, W ashington: N ational A cadem y Press,vol.3,pp.1-114. Stratford, N . and Roth, W . (1999), The 1998 Youth Lifestyles Survey: Technical Report, London:N CSR. 21

Trickett,A .,O sborn,D .R.and Ellingw orth D .(1993),Sim ple and RepeatVictim isation: The Influences ofIndividualand Area Characteristics,D iscussion PaperES239,D epartm ent ofEconom etrics and Statistics,U niversity ofM anchester. Trickett, A ., O sborn, D .R. and Ellingw orth, D . (1995), ‘Property Crim e V ictim isation: The Roles of Individualand A rea Influences’,InternationalReview ofVictim ology,vol.3, pp.273-95. W ittebrood, K . and N ieuw beerta, P. (1999), ‘W ages of Sin? The Link betw een O ffending, Lifestyle and V iolent V ictim isation’, European Journal on Crim inal Policy and Research,vol.7,pp.63-80. W olfgang,M .and Ferracuti,F.(1967),The Subculture of Violence: Toward an Integrated Theory in Crim inology,London:Tavistock.

22

A ppendix Table A 1.V ariable m eans

PersonalC haracteristics Age M ale H ave atleastone child H as currentpartner W hite origin Black origin A sian origin N ative born U nem ployed N o qualifications A tschool O w neroccupier A rea C haracteristics N orth ofEngland Y orkshire/H um berside N orth W estEngland EastM idlands W estM idlands EastA nglia South EastEngland South W estEngland W ales U rban area Ruralarea A corn 17 m ostdeprived People w ish to leave area

21.191 0.470 0.234 0.511 0.910 0.027 0.040 0.940 0.048 0.070 0.278 0.624 0.072 0.110 0.119 0.082 0.099 0.045 0.196 0.072 0.061 0.567 0.176 0.124 0.211

R isk Factors A ctive in com m unity 0.175 Sports participation 0.609 Socialactivities 0.885 H angouton street 0.214 W as bullied atschool 0.320 G oes outalone atnight 0.555 Carries personalalarm 0.052 Thinks judges outoftouch 0.259 A dditionalvariables for offender equation Expelled from school 0.097 Persistenttruant 0.084 N evertem pted to hitsom eone 0.177 Frequentdrinker 0.054 Started drinking early in life 0.254 O nly taken softdrugs in pastyear 0.194 Taken hard drugs in pastyear 0.038 Evertaken any drug 0.167 Think courts too lenient 0.511 Think courts too tough 0.047 Fam ily in trouble w ith police 0.019 Friends in trouble w ith police 0.152 N o fatherw hen teenager 0.190 O ffending behaviour A ny offence 0.454 N on-violentoffender 0.163 V iolentoffender 0.291 Persistentoffender 0.062

23