This article was downloaded by: [The University of Manchester Library]

On: 22 December 2014, At: 08:45

Publisher: Routledge
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer
House, 37-41 Mortimer Street, London W1T 3JH, UK

Applied Economics
Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/raec20

Youth unemployment and crime in the English

regions and Wales
Fiona Carmichael & Robert Ward
Published online: 04 Oct 2010.

To cite this article: Fiona Carmichael & Robert Ward (2000) Youth unemployment and crime in the English regions and
Wales, Applied Economics, 32:5, 559-571, DOI: 10.1080/000368400322462

To link to this article: http://dx.doi.org/10.1080/000368400322462

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”)
contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors
make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any
purpose of the Content. Any opinions and views expressed in this publication are the opinions and views
of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content
should not be relied upon and should be independently verified with primary sources of information.
Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs,
expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in
connection with, in relation to or arising out of the use of the Content.

This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions
 Applied Economics, 2000, 32, 559 ± 571

Y outh unemployment and crime in the

English regions and W ales
F I ON A C A R MI C H A E L * and R OB E R T WA R D
Department of Economics, University of Salford, Salford M5 4W T , UK
Downloaded by [The University of Manchester Library] at 08:45 22 December 2014

This paper investigates the relationship between unemployment and crime in Eng-
land and Wales taking account of both age and gender in the unemployment meas-
ures. The study is for 1985± 95 and is disaggregated to the regional level. We allow for
di€ erent types of crime and the deterrence e€ ects of detection and punishment. We
also consider the relationship between ethnicity and crime in the light of the con-
tentious public debate on this issue. Our results indicate that there is a systematic
positive relationship between burglary rates and male unemployment regardless of
age. However, we ® nd that while youth unemployment is consistently and positively
related to criminal damage and robbery rates there is no systematic evidence of a
relationship between adult male unemployment and these speci® c crimes. Instead
our evidence supports a positive link between adult unemployment and theft. We
® nd no compelling evidence of a link between ethnicity and crime.

I . I N T R OD U C T I ON the signi® cance and even the direction of the U± C relation-

ship.
Unemployment and crime constitute perhaps the two The theoretical basis of the `consensus of doubt’ relies on
major social problems faced by governments (Hale and two countervailing e€ ects of unemployment on crime ± the
Sabbagh, 1991). It is therefore not surprising that the motivational and opportunity e€ ects. The former is consis-
unemployment ± crime (U± C) relationship has been the tent with the economic theory of crime which originates
focus of much research by economists and other social with Becker (1968 )3 and postulates that an individual will
scientists. 1 However, the strength of this research e€ ort commit a crime if the expected net bene® ts of doing so
has failed to produce clear evidence regarding the existence exceed the expected net bene® ts from legitimate activity.
of a causal link. A number of extensive literature reviews According to this line of reasoning unemployment , by low-
based on work done mainly in the US (Freeman, 1980; ering the expected bene® ts from legitimate activity, will
Long and Witte, 1981) have concluded that the unemploy- have a positive impact on crime rates. However, the oppor-
ment e€ ect, if it exists at all, is relatively weak. This con- tunity e€ ect identi® ed by Cantor and Land (1985 ) is con-
clusion is largely, although not completely,2 echoed by sistent with a negative relationship between unemployment
economic analyses of the U ± C relationship in Britain and crime. Cantor and Land argue that when unemploy-
(Wolpin, 1978; Willis, 1983; Pyle, 1989; Pyle and ment rises there are fewer economic goods (crime targets)
Deadman, 1994). This is the `consensus of doubt’ referred in circulation and those that are in circulation are better
to by Chiricos (1987 ) which calls into question the strength, protected (the guardian e€ ect). Thus, rising unemployment

* To whom all correspondence should be addressed: E-mail [email protected]

1
Although various authors have argued that the analysis of crime is largely excluded from conventional economics (see for example
Burden, 1990).
2
See for example Reilly and Witt (1992 and 1996) and Hale and Sabbagh (1991).
3
Becker’s work has been extended by Ehrlich (1973 ) to take account of economic factors such as earnings and unemployment. Block and
Heincke (1975) additionally allow for moral and ethical constraints in terms of a psychic disadvantage associated with the time spent
participating in illegal acts. The economic theory of crime is discussed more fully in Pyle (1995).
Applied Economics ISSN 0003± 6846 print/ISSN 1466± 4283 online # 2000 Taylor & Francis Ltd 559
http://www.tandf.co.uk/journals/tf/00036846.htm l
 560 F. Carmichael and R. W ard
reduces the opportunity for criminal activity. The inter- young males and crime that is consistent with a stronger
action of these contradictory forces provides the underly- motivational e€ ect from unemployment for this group
ing rationale for the `consensus of doubt’ and appears to (Gillespie, 1978; Long and Witte, 1981). This may be
explain why the U± C relationship has been found to be attributed to a number of factors including lower risk aver-
weak and in some cases negative. sion on the part of younger males, low expectations
However, Chiricos (1987 ) in a review of the empirical (Timbrell, 1990) , greater perceived needs (Britt, 1994 )
evidence, calls into doubt this so called consensus. He and/or a weaker attachment to society and the voluntary
argues that the U± C relationship is conditional on the social controls it imposes due to lack of responsibilities and
type of crime, the level of aggregation of the empirical peer pressure (Allen, 1996 ). According to the last line of
study, the measure of unemployment used and the time argument it is the existence of social controls which prevent
period studied. 4 Taking each of these in turn, property individuals engaging in illegal activities even when they
crimes, and burglary in particular, were much more likely could bene® t from doing so. The breakdown of social con-
than violent crimes to be positively and signi® cantly related trol is therefore a pre-condition for the economic determi-
to unemployment. With regard to the level of aggregation , nants of crime, such as unemployment to take e€ ect
Downloaded by [The University of Manchester Library] at 08:45 22 December 2014

Chiricos found that the U± C relationships were weakest in (Elster, 1989; Field, 1990).
studies conducted at the national level and strongest in In this paper we address two of the issues identi® ed by
those conducted at city, intracity and county levels. He Chiricos by taking account of age and gender in our unem-
suggests that this is due to greater homogeneity at lower ployment measure and allowing for regional di€ erences.
levels of aggregation leading to less aggregation bias. This Speci® cally we di€ erentiate between the unemployment
implies that in empirical tests of the U ± C relationship rate of younger males, or youths, aged between 16 and
regional di€ erences need to be allowed for. However, 25 and that of older males between 26 and 64. We are
Pyle and Deadman (1994, p. 315 ) indicate that there are therefore able to investigate the relationship between
also problems involved in using cross-section data which both youth and adult male unemployment and di€ erent
suggests that pooling time series and cross-section data types of crime whilst taking into account deterrence and
may be a practical alternative. The measure of unemploy- punishment e€ ects. Our principal aim in this project is to
ment was expected to in¯ uence the estimate of the U± C identify any di€ erences that might exist between these two
relationship because of the strong links between crime sets of relationships. More controversially we allow for the
and both age and gender. However, the evidence surveyed possibility of a link between ethnicity and crime which we
by Chiricos was inconclusive in this respect due to the small ® nd to be mostly, but not always, insigni® cant. We take on
number of studies and also because discrete age and gender board the aggregation problem by disaggregating to the
grouped crime data are unavailable. Chiricos also suggests regional level. The result is a panel study conducted over
that the U± C relationship may have strengthened since the 11 years for the 15 English regions and Wales.
1970s when unemployment began to rise substantially. 5 The plan of the paper is as follows. In Section II we
Although the studies included in Chiricos’ survey were describe out data set and discuss some summary statistics.
mainly conducted in the US his conclusions clearly have In Section III we outline our empirical model, the results of
relevance to work done in the UK. And while most studies which are discussed in Section IV. Section V concludes.
of the U± C relationship in Britain ® nd unemployment has
a positive, but small, e€ ect on recorded crime, they have
generally focused on the overall unemployment rate and I I . D A T A A N D S U M MA R Y S T A TI S T I C S
national crime rates.6 Questions therefore remain about
the strength and the nature of the U± C relationship in Two data sources were used in this study; Home O ce
Britain. In particular the role of gender and age speci® c crime statistics and the Labour Force Survey (LFS). The
unemployment has yet to be identi® ed. In the UK this data were collected on a regional basis for the 15 standard
constitutes a major omission given the high rates of male English regions and Wales and relate to the 11 years from
youth unemployment , especially among ethnic minority 1985 to 1995. 7 Tables 1± 4 show sample means for the crime
groups. There is also a presumed relationship between and unemployment data. Tables 1 and 3 show crime rates,

4
An additional factor considered by Elliott and Ellingworth (1998) is functional form.
5
See Dickinson (1995 ) who draws attention to the structural break around 1970 between a period of `full employment’ and mass
unemployment.
6
Some exceptions with respect to the level of aggregation are Pyle and Deadman (1994) , Reilly and Witt (1992 and 1996 ), Witt et al.
(1998) , Timbrell (1990) and Willis (1983 ).
7
Combining the two data sources was not entirely straightforward as the crime data came to us by Police Force Area, while the LFS data
could only be accessed at the regional level. The crime data therefore had to be aggregated up to the regional level using suitable weights
according to area population.
 Y outh unemployment and crime 561
Table 1. Regional crime statistics: 1985± 1995. (Rates per 100 000 population by type of crime, % of those convicted receiving a prison
sentence, average prison sentence (months ) and clear up rates (% ))

Violence Total % Received Average

Criminal against Total clear up prison sentence
Region Burglary damage Robbery Theft person crime rate (%) sentence (months)
East Anglia 1441.81 784.07 21.83 3679. 1 276.63 6512.11 35.54 54.65 19.75
East Mids. 1959.81 1159.74 50.27 4571.79 438.57 8538.02 33.92 51.06 19.22
Greater London 2274.88 1402. 0 283.73 5695.13 448.01 10 744.5 18.20 49.18 23.52
Greater Manchester 3418.18 1295.36 128.36 6204.45 336.82 11 933.64 30.98 52.55 19.36
Merseyside 2780.64 1146.36 131.09 4953.18 414.18 9869.18 39.58 50.82 19.65
Rest of North West 1646.93 596.17 24.38 3578.74 236.68 6448.01 40.5 51.53 17.25
Rest of West Mids. 1590.52 855.43 20.86 3404.96 352.38 6458.35 36.53 53.43 18.52
Rest of North 2292.61 1069. 5 24.95 4878.08 382.64 9055.89 35.35 47.77 19.63
Rest of South East 1429.45 877.25 30.58 3841.03 252.94 6768.22 29.99 50.87 21.58
Rest of Yorkshire
Downloaded by [The University of Manchester Library] at 08:45 22 December 2014

and Humberside 2708.67 1106.02 28.29 4794.82 428.85 9422.94 29.95 46.03 19.04
South West 1638.8 807.27 36.66 4068.44 289.48 7270.73 31.11 45.29 19.18
Tyne and Wear 3852. 0 1630.64 57.36 6107.73 374.55 12 321.55 33.73 45.36 19.02
West Midlands
Conurbation 3051.46 1298.82 159.64 5195.64 358.27 10 431.18 31.50 51.55 20.32
Wales 1764.97 1143.99 17.98 4041.76 368.4 7695.77 38.52 51.0 18.81
West Yorkshire 3177.55 1430.27 83.46 5598. 0 401.46 11 086.09 31.42 44.27 19.52
South Yorkshire 2566.82 1017.45 47.18 4442.82 352.18 8689.91 36.15 43.55 18.63

Source: Home O ce Statistics.

Table 2. Regional unemployment rates (% ): 1985± 1995 London and the West Midlands, although they are also
high in Merseyside. The highest clear up rates are in the
Region Male youths Adult males rest of the North West (which excludes Greater
East Anglia 6.71 4.97
Manchester or Merseyside) , where crime rates are relatively
East Midlands 7.12 6.12 low, and lowest in Greater London. The percentage of
Greater London 8.39 8.01 o€ enders receiving a prison sentence is highest in East
Greater Manchester 9.29 7.94 Anglia (where crime rates are among the lowest) and lowest
Merseyside 12.1 11. 13 in South Yorkshire. The average sentence received is high-
Rest of North West 7.91 5.78
Rest of Midlands 6.63 5.62 est in Greater London and lowest in the rest of the North
Rest of North 9.56 8.45 West. However, except for Greater London, where the
Rest of South East 5.87 4.84 average sentence does appear to be considerably higher
Rest of Yorks and 7.89 7.01 (perhaps re¯ ecting the higher rates of robbery and violent
Humberside
crimes) average sentences do not show much regional
South West 6.43 5.34
Tyne and Wear 11.9 11. 3 variation.
West Midlands Conurbation 9.5 10. 05 The regional unemployment statistics in Table 2 show
Wales 8.43 7.2 that while unemployment was generally higher in the
West Yorks 8.14 6.65 urban regions there is no clear correlation between crime
South Yorks 10.83 9.68
and unemployment. Both adult and youth unemployment
Source: Labour Force Surveys. were highest in Merseyside and Tyne and Wear but unem-
ployment was higher in South Yorkshire than West
prison sentencing rates, average sentences and total clear Yorkshire and youth unemployment was higher in Wales
up rates by region and year respectively. Tables 2 and 4 than in Greater London. In every region except the West
show male unemployment rates for youths (aged 16 ± 25) Midlands youth unemployment was higher over the period
and adults (aged 26± 64) by region and year respectively. 1985 ± 95 than adult unemployment.
Table 1 shows that crime rates vary considerably by type Table 3 shows that while all crime rates increased
of crime and over the regions of England and Wales. Theft between 1985 and 1995, there appears to have been a mod-
is the most common manifestation of crime and robbery est downturn in burglary and theft rates after 1993/4. The
the least common. Total crime rates are highest in the highest percentage increases were in robbery and criminal
urban regions of Greater London, Greater Manchester, damage where the o€ ence rate has more than doubled
Tyne and Wear, the West Midlands and West Yorkshire. (nearly quadrupling in the case of robbery). Total clear
Robbery rates, in particular, are very high in Greater up rates, on the other hand, have been falling. Prior to
 562 F. Carmichael and R. W ard
Table 3. Crime statistics: (all regions ) ; Rates per 100 000 population by type of crime, % of those convicted receiving a prison sentence,
average prison sentence (months ) and clear up rates (% )

Violence Total % Average

Criminal against Total clear up Sentenced sentence
Year Burglary damage Robbery Theft person crime rate (%) to prison (months)
1985 1953.47 703.18 44.52 3838.78 259.52 7120. 3 39.59 52.02 Ð
1986 2107.99 791.66 48.45 4067.79 266.22 7605.13 35.57 51.87 Ð
1987 2041.57 827.86 51.34 4221.78 298.47 7784.99 37.11 52.17 17.89
1988 1829.82 841.38 48.71 3953.14 330.26 7353.38 39.71 49.46 18.01
1989 1822.69 928.7 52.92 4097.22 364.48 7623.39 37.80 47.38 19.13
1990 2158.91 1093.07 58.34 4811.91 375.14 8882.49 36.03 44.28 19.51
1991 2613.85 1244.81 73.46 5531.91 384.03 10 285.33 32.77 44.36 19.23
1992 2906.29 1370.39 88.89 5630.42 405.83 10 838.2 28.19 44.6 19.90
1993 2976.49 1398. 2 100.41 5440.05 406.73 10 758.14 26.42 48.43 20.19
1994 2748. 2 1451.67 102.79 5092.20 422.85 10 234.52 26.99 51.85 20.60
Downloaded by [The University of Manchester Library] at 08:45 22 December 2014

1995 2687.35 1463.05 118.46 4915.59 413.5 9995.81 26.29 55.91 21.18
% change 1985± 1995 37.6 108 166 28 59 40 733.6 7.5 18.4

Source: Home O ce Statistics

Table 4. Annual unemployment rates (% ): (all regions ) ployment and the equally dramatic (in terms of public per-
ception) but proportionatel y lower rise in total crime
Year Male youths Adult males suggests that the two are correlated. However, some
1985 6.35 8.04 crime rates have risen more than others. As already
1986 7.70 7.91 noted criminal damage and robbery rates have risen by
1987 6.97 8.06 most while there was a drop in theft, burglary and total
1988 6.23 7.03 crime rates after 1993/4. This suggests that any link
1989 4.69 5.69 between youth unemployment and crime will not be
1990 4.87 5.21
1991 6.19 6.74 straightforward and may be contingent on type of crime.
1992 13.15 8.81 Similarly, if any relationship exists between adult unem-
1993 13.09 9.13 ployment and crime it is not obvious from simple bivariate
1994 11.10 8.51 comparisons of the data in Tables 1± 4. In order to investi-
1995 12.82 7.45 gate these possibilities further we turn to our multivariate
% change 1985± 1995 102 77
analysis which takes into account both regional di€ erences
Source: Labour Force Surveys. and di€ erent types of crime.

1991, the average clear up rate was over 30%, peaking in

1988 at nearly 40% but by 1995 it had fallen to 26%. These I I I . E MPIR I C A L MOD E L
® gures are consistent with a negative relationship between
detection rates and o€ ence rates. However, both the per- Our basic methodology is to regress crime rates on youth
centage of convicted criminals receiving prison sentences or adult male unemployment rates. We also allow for
and the average sentence received have risen (the latter deterrence and regional ® xed e€ ects8 as well as ethnicity.
more so) which seems to imply that the deterrent e€ ect of We include a measure of the latter in order to test whether
the severity of punishment is weak if it exists at all. This is there is any support for the commonly held belief that
consistent with the evidence of previous studies which have crime is committed disproportionatel y by ethnic minority
indicated that the deterrent e€ ect of certainty of punish- males. However, because the UK ethnic population is
ment is stronger than the deterrent e€ ect of its severity heavily concentrated in particular urban regions which
(Pyle, 1995 ). are also associated with high levels of social deprivation
Table 4 shows that Youth unemployment doubled as well as unemployment , such an e€ ect if it exists may
between 1985 and 1995 with most of this growth taking have more to do with regional characteristics than ethnic-
place after 1992. Adult unemployment actually fell over ity. Allowing for regional di€ erences is therefore particu-
the whole period and the rise between 1992 and 1993 was larly important and in order to clarify the situation we
much less severe. The dramatic increase in youth unem- estimate the OLS model with and without regional ® xed

8
The random e€ ects model was rejected on the basis of a Hausman test.
 Y outh unemployment and crime 563
e€ ects (speci® cations 1 and 2). Because the data are pooled any relationships between crime and ethnicity that depend
over time and region we also need to take into account the on the age pro® le of the male population. 11
possibility that crime and unemployment are moving in Although youth unemployment and adult unemploy-
trend together. However, given that we only have 11 ment do not mirror each other exactly over time they are
years of data, tests for orders of integration are clearly positively correlated. Because of this we ran separate
inappropriate. We therefore additionally estimate a semi- regressions with either Y UNEMP and AUNEMP as an
log speci® cation with a time trend 9 as well as regional ® xed independent variable. For consistency Y W HIT E was only
e€ ects and a ® rst di€ erenced model (speci® cations 3 and 4) included in the regressions with Y UNEMP, and AW HIT E
in order to test for the robustness of the U ± C relationship was only included in the regressions with AUNEMP.
(see Reilly and Witt, 1991, 1996 ; Pyle and Deadman, 1994). Inasmuch as the U± C relationship is positive we expect
positive signs on Y UNEMP and AUNEMP in the regres-
sions. A stronger relationship between Y UENMP and
Dependent variables
speci® c crime rates would be evidence of a stronger moti-
Downloaded by [The University of Manchester Library] at 08:45 22 December 2014

In our estimations the dependent variables are o€ ence rates vational e€ ect of unemployment on younger males.
per 100 000 population for the 15 English regions and However, given the conditional nature of the U± C relation-
Wales over 11 years (1985 ± 95). The o€ ence rates are spe- ship as identi® ed by Chiricos (1987) any such e€ ects may
ci® c with respect to burglary, criminal damage, robbery, well be crime speci® c.
theft, violence against the person and total crime; The probability of detection is proxied by the overall
BURGRT E, CDRT E, ROBRT E, T HRT E, V PRT E, percentage clear-up rate for burglary, criminal damage,
T OT RT E (see appendix for full variable de® nitions and robbery, theft, violence against the person and total
sources ). 10 crime ; BURGCL R, CDCL R, ROBCL R, T HCL R,
V PCL R, T OT CL R. The expected severity of the punish-
ment in the event of being caught depends on the likelihood
Independent variables of a prison sentence and the expected duration of any such
The independent variables take into account male adult sentence. The former is proxied by the percentage of con-
and youth unemployment , ethnicity and deterrence e€ ects. victed criminals who receive a prison sentence, JAIL , and
Youth unemployment is measured by the percentage of the latter by the average sentence in months, SENT .
males aged between 16 and 25 who are classi® ed by the The theoretical arguments predict a negative relationship
LFS as unemployed according to the ILO de® nition, between o€ ence rates and both clear up rates and sentenc-
Y UNEMP. Adult employment is similarly given by the ing. However, the deterrent e€ ect will be obscured if there
percentage of males aged 26 ± 64 who are classi® ed as ILO is a political dimension to punishment such that increases
unemployed, AUNEMP. The use of the ILO de® nition of in crime are met by either higher rates of prison sentencing
unemployment avoids some of the problems that would and/or longer sentences. 12 The greater rise in average sen-
otherwise arise in a time series study, due to the repeated tence duration over the period under consideration sug-
changes that have been made to the o cial UK claimant gests that political considerations might be more of a
count measure of unemployment. Ethnicity is measured by problem in the determination of sentence duration than
the percentage of each region’s population classi® ed as the likelihood of a prison sentence. Inasmuch as this is
white and for consistency with the unemployment variables the case interpretation problems are more likely to arise
we employ two such measures; Y W HIT E is the percentage with SENT than JAIL . 13 A further problem with SENT
of white youths and AW HIT E is the percentage of white is that the data were only available from 1986 which meant
adult males. Making this distinction allows us to identify that 32 observations were lost by employing it. Because of
9
This speci® cation implies that all else equal the growth rate for o€ ences is constant.
10
We also ran regressions with fraud and forgery rates, sex o€ ence rates and `other’ crime rates as dependent variables. We did not ® nd
any signi® cant relationship between these crimes and either youth or adult unemployment. These results are not presented here due to
lack of space but are available on request.
11
The LFS data show that the non-white population is both small and heavily concentrated in the West Midlands, West Yorkshire and
Greater London. The percentage of non-white males between the age of 16 ± 25 is marginally higher in every region than the percentage of
older males. However, the ethnic population has remained fairly constant, in spite of its younger age pro® le, at 6± 7% of the total male
population.
12
The latest British Crime Survey (Hough and Roberts, 1998) actually warns politicians away from populist responses to crime. It says
that politicians have been wrong to `play to the gallery’ by jailing more and more people in order to feed the public’s appetite for tougher
punishment. Being seen to be `tough on crime’ through policing strategies such zero tolerance is another populist response to rising crime
® gures. Downes (1997) discusses this point further and goes so far as to argue that falling crime rates in the US and near-full employment
can both be explained by tougher penal measures.
13
Interpretations problems arise if the political dimension to sentencing implies that sentence duration is endogenously determined (see
Reilly and Witt, 1996).
 564 F. Carmichael and R. W ard
these problems SENT is only included in the regressions ethnic variable is Y W HIT E in table (a) of each pair and
where the total crime rate is the dependent variable. 14 AW HIT E in table (b).
Precise de® nitions of JAIL and SENT and the other vari- In each of Tables 5 (a)± 10 (b) speci® cation 1 is the basic
ables used in the regression analysis, as well as their OLS model which does not allow for regional characteris-
sources, are listed in the appendix. tics, speci® cation 2 is the regional ® xed e€ ects model, spe-
ci® cation 3 is the semi-log version of speci® cation 2 and
includes a time trend. Speci® cation 4 is the ® rst di€ erenced
IV. R ESULTS model. 15 In presenting these di€ erent speci® cations our
purpose is to test the robustness of the U± C relationship.
In Tables 5 (a)± 10 (b) the dependent variables are, respect- Only where the unemployment variable is consistently
ively, BURGRT E, CDRT E, ROBRT E, T HRT E, V PRT E, signed and signi® cant over each of these speci® cations are
T OT RT E. Table (a) in each pair includes Y UNEMP as the any claims made in support of the existence of such a
unemployment measure and investigates the relationship relationship.
Downloaded by [The University of Manchester Library] at 08:45 22 December 2014

between youth unemployment and crime. Table (b) in Most of the regional ® xed e€ ects in speci® cations 2± 3 are
each pair includes AUNEMP and estimates the relation- highly signi® cant (at or above the 5% level) and in each
ship between adult unemployment and crime. Similarly the case À2 and F tests reject the hypothesis that they are the

Table 5 (a). Dependent variable: BURGRTE

Variable Speci® cation 1 Speci® cation 2 Speci® cation 3 Speci® cation 4

Y UNEMP 136.88 (10.27*** ) 54. 74 (5.46*** ) 0.022 (4.81*** ) 1.16 (2.77*** )

BURGCL R 72.97 (70.4) 739. 2 (77.73*** ) 70.014 (76.17*** ) 723.44 (75.06*** )
JAIL 749.78 (75.01*** ) 723. 11 (73.45*** ) 70.006 (71.45) 714.16 (72.18**)
Y W HIT E 27.41 (73.22*** ) 723. 15 (71.26 ) 70.004 (70.49) 78.31 (70.77)
CONST ANT 6247.5 (6.76*** ) ± ± 3.05 (1.88*)
T REND ± ± 0.015 (2.64** ) ±
Regional ® xed e€ ects no yes yes no
R2 (adj) 0.448 0.825 0.83 0.224
F 36.51*** 44. 51*** 43.8*** 12.48***
Log-Likelihood 71395.99 71286.66 79.978 71108.28
Restricted Log-Likelihood 71450.31 71450.31 786.77 71130.61

t statistics in parentheses; *** indicates signi® cance at the 1% level; ** indicates signi® cance at the 5% level; * indicates signi® cance at the
10% level.

Table 5 (b). Dependent variable: BURGRTE

Variable Speci® cation 1 Speci® cation 2 Speci® cation 3 Speci® cation 4

AUNEMP 228.17 (11.86*** ) 103.75 (5.21*** ) 0.044 (5.06*** ) 50.84 (3.84*** )

BURGCL R 730.04 (75.36*** ) 750. 19 (710. 82*** ) 70.017 (77.1***) 723.3 (75.29*** )
JAIL 734.94 (73.76*** ) 723. 95 (73.22*** ) 70.0054 (71.35 ) 716.43 (72.56**)
AW HIT E 74.497 (0.4) 8.72 (0.225 ) 70.027 (1.68) 2.16 (0.1)
CONST ANT 2742.1 (2.29** ) ± ± 55.43 (2.73*** )
T REND ± ± 0.018 (3.25*** ) ±
Regional ® xed e€ ects no yes yes no
R2 (adj) 0.505 0.823 0.833 0.26
F 45.63*** 43. 7*** 44.5*** 14.67***
Log-Likelihood 71386.39 71288.03 81.167 71104.92
Restricted Log-Likelihood 71450.31 71450.31 786.77 71130.61

t statistics in parentheses; *** indicates signi® cance at the 1% level; ** indicates signi® cance at the 5% level; * indicates signi® cance at the
10% level.

14
An additional problem was that the data on sentencing were not available on a crime speci® c basis. Disproportionate increases in
speci® c crimes (such as violent crimes) where a prison sentence and/or a longer sentence is more likely, will therefore lead to increases in
both JAIL and SENT. Thus the disincentive e€ ect of more punitive punishment may be obscured by a positive relationship between
increases in such crimes and average sentencing.
15
This includes a constant term in line with Pyle and Deadman (1994 ) and Reilly and Witt (1992 and 1996).
 Y outh unemployment and crime 565
Table 6 (a). Dependent variable: CDRTE

Variable Speci® cation 1 Speci® cation 2 Speci® cation 3 Speci® cation 4

Y UNEMP 40.03 (7.45*** ) 30. 37 (5.29*** ) 0.017 (3.44*** ) 4.53 (1.80*)

CDCL R 72.74 (78.5*** ) 733. 78 (78.36*** ) 70.03 (79.34*** ) 711.88 (75.31*** )
JAIL 713.18 (73.67*** ) 79.61 (72.39** ) 70.001 (70.24) 75.04 (72.12**)
Y W HIT E 73.29 (70.95 ) 711. 72 (71.16 ) 70.004 (70.49) 1.69 (0.43 )
CONST ANT 2404.3 (6.94*** ) ± ± 65.53 (8.61*** )
T REND ± ± 0.03 (4.98*** ) ±
Regional ® xed e€ ects no yes yes no
R2 (adj) 0.636 0.745 0.82 0.21
F 77.38*** 27. 93*** 41.36*** 11.8***
Log-Likelihood 71218.90 7117.40 79.35 7947.77
Restricted Log-Likelihood 7130.82 71309.82 783.12 7969.05

t statistics in parentheses; *** indicates signi® cance at the 1% level; ** indicates signi® cance at the 5% level; * indicates signi® cance at the
Downloaded by [The University of Manchester Library] at 08:45 22 December 2014

10% level.

Table 6 (b). Dependent variable: CDRTE

Variable Speci® cation 1 Speci® cation 2 Speci® cation 3 Speci® cation 4

AUNEMP 26.50 (31.2*** ) 724.61 (72.02** ) 70.033 (73.7***) 2.29 (0.56 )

CDCL R 739.93 (711.78** ) 747. 33 (712. 32*** ) 70.036 (712.03*** ) 712.76 (75.52*** )
JAIL 79.80 (72.45** ) 71.94 (70.64) 0.0097 (2.52** ) 74.96 (72.04**)
AW HIT E 8.34 (1.62) 14. 46 (0.64 ) 0.022 (1.35) 0.44 (0.05 )
CONST ANT 1512.8 (2.88*** ) ± ± ±
T REND ± ± 0.042 (7.57*** ) ±
Regional ® xed e€ ects no yes yes no
R2 (adj) 0.547 0.707 0.83 0.196
F 54.03*** 23. 18*** 43.79*** 10.73***
Log-Likelihood 71237.92 71191.81 83.61 7949.49
Restricted Log-Likelihood 71309.82 71309.82 783.12 7969.05

t statistics in parenthesis; *** indicates signi® cance at the 1% level; ** indicates signi® cance at the 5% level; * indicates signi® cance at the
10% level.

same. 16 The signs and relative sizes of the ® xed e€ ects are Unemployment and crime
also fairly consistent over the range of crimes considered.
In Tables 5 (a)± (b) the dependent variable is the burglary
Therefore to save space, and because our main concern is
rate and both Y UNEMP and AUNEMP are consistently
with the U± C relationships rather than the regional e€ ects, positive and signi® cant at the 1% level in all four speci® ca-
we only indicate in which of the speci® cations the ® xed tions. In each speci® cation the coe cient on AUNEMP is
e€ ects are included and do not show their individual co- larger indicating that the e€ ect of adult unemployment on
e cients or levels of signi® cance. 17 burglary rates is greater. The coe cients in speci® cation 2
Although the relationships between the speci® c crime imply that 1% increases in youth and adult unemployment
rates and the deterrence and ethnic variables are of will increase the number of burglaries (per 100 000 popula-
interest in their own right, they are not our main concern tion) by 54.74 and 103.75 respectively (or by 2.3% and
and we ® rst consider the signs and signi® cance of 4.4% given a mean burglary rate of 2349.69). The coe -
Y UNEMP and AUNEMP in the estimated equations. cients in the semi-log formulation, speci® cation 3, imply
In particular we are interested in noting any di€ erences similar increases in the burglary rates of 2.2% and 4.4%
in the U± C relationships when the unemployment in response to 1% increases in youth and adult unemploy-
measure changes. Our results suggest that such di€ erences ment respectively. The smaller coe cients on Y UNEMP
exist. and AUNEMP in speci® cation 2, compared with speci® ca-

16
The statistics test the restrictions that (1) there are no group e€ ects on the mean of the dependent variable and (2) there is ® t in the
regression but no group e€ ects (see Greene, 1995: 290).
17
All the ® xed e€ ects are positive and most, but not all, are consistently signi® cant at the 5% level or higher. The smallest e€ ects were
associated with the South West and the rest of the South East and the largest were those of Tyne and Wear, Greater Manchester and the
West Midlands.
 566 F. Carmichael and R. W ard
Table 7 (a). Dependent variable: ROBRTE

Variable Speci® cation 1 Speci® cation 2 Speci® cation 3 Speci® cation 4

Y UNEMP 4.87 (7.1*** ) 5.82 (9.52*** ) 0.048 (6.49*** ) 1.23 (4.14*** )

ROBCL R 72.33 (79.37*** ) 70.77 (71.90* ) 70.026 (75.72*** ) 70.24 (71.37)
JAIL 0.89 (2.15** ) 0.07 (0.16 ) 0.0005 (0.07 ) 0.8 (2.99*** )
Y W HIT E 77.88 (718.97*** ) 76.54 (76.02*** ) 70.018 (71.46 ) 70.897 (1.91* )
CONST ANT 804.69 (19.89*** ) ± ± 6.07 (6.78*** )
T REND ± ± 0.043 (4.87*** ) ±
Regional ® xed e€ ects no yes yes no
R2 (adj) 0.862 0.913 0.925 0.16
F 274.17*** 97. 29*** 109.39*** 8.75***
Log-Likelihood 7838.47 7790.07 3.61 7607.45
Restricted Log-Likelihood 71014.76 71014.76 7235.37 7235.37

t statistics in parentheses; *** indicates signi® cance at the 1% level; ** indicates signi® cance at the 5% level; * indicates signi® cance at the
Downloaded by [The University of Manchester Library] at 08:45 22 December 2014

10% level.

Table 7 (b). Dependent variable: ROBRTE

Variable Speci® cation 1 Speci® cation 2 Speci® cation 3 Speci® cation 4

AUNEMP 3.99 (4.04*** ) 2.06 (1.33 ) 70.008 (70.53) 1.16 (1.93*)

ROBCL R 72.66 (710.23*** ) 73.11 (76.81*** ) 70.0039 (78.18*** ) 70.4 (72.24** )
JAIL 0.97 (2.2** ) 0.18 (0.32 ) 0.015 (2.06** ) 0.87 (2.97** )
AW HIT E 710.21 (17.18*** ) 710. 94 (3.89*** ) 0.015 (0.52) 71.7 (71.91* )
CONST ANT 1062.2 (18.09*** ) ± ± 72.37 (72.41**)
T REND ± ± 0.067 (7.11*** ) ±
Regional ® xed e€ ects no yes yes no
R2 (adj) 0.843 0.862 0.905 0.12
F 235.22*** 58.6*** 84.06*** 6.41***
Log-Likelihood 7850.01 7830.28 717.84 7611.49
Restricted Log-Likelihood 71014.76 71014.76 7235.37 7623.74

t statistics in parentheses; *** indicates signi® cance at the 1% level; ** indicates signi® cance at the 5% level; * indicates signi® cance at the
10% level.

tion 1, re¯ ect the signi® cance of the regional ® xed e€ ects robbery rates is much less clear cut. The coe cients on
and show that unmeasured regional characteristics are an Y UNEMP in speci® cations 2 and 3 imply that a 1%
important determinant of crime. increase in the youth unemployment rate is likely to lead
In Tables 6 (a)± (b) the dependent variable is the criminal to an increase in the robbery rate of between (nearly ) 5%
damage rate and only the in¯ uence of Y UNEMP is con- and 8% (given a mean robbery rate of 71.66 ). A further
sistently positive and signi® cant at the 10% level or higher. consideration is that in the case of robbery the inclusion of
The coe cients on Y UNEMP in speci® cations 2 and 3 the regional ® xed e€ ects in speci® cation 2 actually increases
indicate that a 1% increase in the youth unemployment the strength of the relationship between youth unemploy-
rate is likely to lead to an increase in the criminal damage ment and crime (and the size of the coe cient). This sug-
rate of 2.8± 3%. In contrast to the consistency of the rela- gests that any negative e€ ects of unemployment on robbery
tionship between criminal damage and youth unemploy- rates (due to restricted opportunity) are more to do with
ment, the sign of AUNEMP in Table 6 (b) ¯ uctuates and regional characteristics than unemployment per se.
in speci® cation 4 the e€ ect of adult unemployment is insig- In Tables 8 (a)± (b) the dependent variable is the theft rate
ni® cant. The implication of these results is that rising and the pattern in Tables 6 (a) ± 7 (b) is reversed. While the
youth, but not adult, male unemployment will be associ- in¯ uence of AUNEMP is positive and signi® cant in each of
ated with higher rates of criminal damage. the speci® cations, the in¯ uence of youth unemployment is
In Tables 7 (a)± (b) the dependent variable is the robbery only signi® cant in speci® cations 1± 3. This suggests that
rate and again, only the in¯ uence of Y UNEMP is consis- while rising adult unemployment rates will generate higher
tently positive and signi® cant. The relationship between theft rates, rising youth unemployment will not have the
adult employment and robbery rates is only signi® cant in same e€ ect. Similarly in Tables 10 (a)± (b) where the depen-
speci® cations 1 and 4. Robbery, like criminal damage dent variable is the total crime rate, only AUNEMP is
therefore appears to be positively related to youth unem- consistently signi® cant. Given that the largest category of
ployment, but the link between adult unemployment and crime is theft this is not surprising. The coe cients on
 Y outh unemployment and crime 567
Table 8 (a). Dependent variable: THRTE

Variable Speci® cation 1 Speci® cation 2 Speci® cation 3 Speci® cation 4

Y UNEMP 123.31 (7.76*** ) 50.69 (3.84*** ) 0.008 (2.66** ) 12.17 (1.14 )

T HCL R 732.36 (74.04*** ) 758. 34 (78.65*** ) 70.013 (78.71*** ) 726.1 (73.04*** )
JAIL 781.48 (77.30*** ) 781. 26 (79.07*** ) 70.013 (75.38*** ) 747.13 (74.65*** )
Y W HIT E 741.83 (73.88*** ) 8.85 (0.40 ) 0.003 (0.70) 7.51 (0.45 )
CONST ANT 12 599 (11.77*** ) ± ± 77.23 (2.28** )
T REND ± ± 0.010 (2.83** ) ±
Regional ® xed e€ ects no yes yes no
R2 (adj) 0.535 0.831 0.848 0.16
F 51.23*** 46. 31*** 49.74*** 8.6***
Log-Likelihood 71419.18 71321.91 168.81 71179.75
Restricted Log-Likelihood 71488.51 71488.51 77.527 71195.78

t statistics in parentheses; *** indicates signi® cance at the 1% level; ** indicates signi® cance at the 5% level; * indicates signi® cance at the
Downloaded by [The University of Manchester Library] at 08:45 22 December 2014

10% level.

Table 8 (b). Dependent variable: THRTE

Variable Speci® cation 1 Speci® cation 2 Speci® cation 3 Speci® cation 4

AUNEMP 198.79 (9.04*** ) 88. 43 (3.58*** ) 0.016 (2.99*** ) 62.27 (2.99*** )

T HCL R 761.18 (78.62*** ) 767.67 (711. 84*** ) 70.014 (710.25*** ) 723.33 (72.7** )
JAIL 771.64 (76.7*** ) 781. 59 (9.02*** ) 70.013 (75.5***) 750.75 (75.07*** )
AW HIT E 712.51 (70.89 ) 79. 26 (1.7* ) 0.023 (2.35** ) 26.85 (0.8)
CONST ANT 9880.5 (6.95*** ) ± ± 92.24 (2.79** )
T REND ± ± 0.012 (3.06*** ) ±
Regional ® xed e€ ects no yes yes no
R2 (adj) 0.574 0.83 0.85 0.2
F 59.83*** 45. 82*** 51.11 10.98***
Log-Likelihood 71411.48 71322.7 170.89 71175.83
Restricted Log-Likelihood 71488.5 71488.5 77.527 71195.78

t statistics in parentheses; *** indicates signi® cance at the 1% level; ** indicates signi® cance at the 5% level; * indicates signi® cance at the
10% level.

AUNEMP in speci® cations 2 and 3 in Tables 8 (b) and This suggests that the detection rate has a negative e€ ect on
10 (b) are also comparable. They indicate that a 1% rise crime rates for all crimes except those committed directly
in adult male unemployment will lead to an increase in against the person. However, clear up rates for crimes
the theft rate of just under 2% and an increase in the against the person are anyway higher than for other
total crime rate of just over 2% (given a mean of 4690.97 types of crime, as identi® cation of the perpetrator is
theft o€ ences and 8952.87 total o€ ences per 100 000 popu- much more likely, and the positive in¯ uence of the clear
lation ). up rate in Table 9 may be re¯ ecting only this. Nevertheless,
In Tables 9 (a)± (b) the dependent crime measure is the the evidence can be taken to imply that the higher rate of
rate of violent crimes against the person. In these equations detection has little e€ ect on deterring this type of crime.
neither Y UNEMP nor AUNEMP are consistently signi® - The e€ ect of the likelihood of a prison sentence on bur-
cant, although Y UNEMP is positively signi® cant in all but glary, criminal damage, violent crimes against the person,
the ® rst di€ erenced model. There is therefore no consistent theft and the total crime is largely negative and signi® cant
evidence to suggest that the rise in the violent crime rate (except in speci® cation 3). However, JAIL is positively
over the period of the study can be attributed to higher (although not consistently signi® cantly) related to the rob-
rates of either youth or adult unemployment. bery rate in Table 7. This result may be attributed to the
classi® cation of robbery as a violent crime. Robbery
o€ ences are therefore more likely to attract a prison sen-
Crime and punishment
tence than most other crimes and the positive relationship
With regard to the deterrent variables, the e€ ects appear to between JAIL and robbery rates may be re¯ ecting this
be mixed. Except for the case of violent crimes against the (although the relationship between violent crimes against
person, the in¯ uence of the clear up rate is consistently the person and JAIL does not appear to follow the same
negative. It is also signi® cant at the 10% level or higher pattern ). However, the results in Table 7 do suggest that
in all but speci® cations 1 of Table 5 (a) and 4 of Table 7 (a). the extra deterrent e€ ect of the greater likelihood of a
 568 F. Carmichael and R. W ard
Table 9 (a). Dependent variable: VPRTE

Variable Speci® cation 1 Speci® cation 2 Speci® cation 3 Speci® cation 4

Y UNEMP 10.32 (7.17*** ) 9.61 (9.49*** ) 0.017 (5.37*** ) 70.23 (70.36 )

V PCL R 70.02 (70.02 ) 8.48 (7.18*** ) 0.019 (5.85*** ) 2.07 (3.36*** )
JAIL 74.64 (74.33*** ) 72.79 (72.97** ) 0.001 (0.21) 71.89 (73.02*** )
Y W HIT E 73.53 (73.27*** ) 70.80 (70.35) 70.001 (70.08) 70.71 (70.69 )
CONST ANT 828.15 (8.15*** ) ± ± 15.85 (8.104*** )
T REND ± ± 0.027 (6.32*** ) ±
Regional ® xed e€ ects no yes yes no
R2 (adj) 0.313 0.73 0.78 0.15
F 20.94*** 26. 22*** 32.0*** 6.16***
Log-Likelihood 71005.61 7914.55 129.37 7732.36
Restricted Log-Likelihood 71040.7 71040.7 714.52 7744.17

t statistics in parentheses; *** indicates signi® cance at the 1% level; ** indicates signi® cance at the 5% level; * indicates signi® cance at the
Downloaded by [The University of Manchester Library] at 08:45 22 December 2014

10% level.

Table 9 (b). Dependent variable: VPRTE

Variable Speci® cation 1 Speci® cation 2 Speci® cation 3 Speci® cation 4

AUNEMP 9.49 (3.49*** ) 9.10 (3.28*** ) 0.0012 (0.16 ) 0.63 (0.50)

V PCL R 0.40 (0.39) 11. 01 (7.90*** ) 0.21 (6.13*** ) 2.13 (3.45*** )
JAIL 74.02 (73.39*** ) 71.34 (71.21) 0.008 (2.61*** ) 71.92 (73.06*** )
AW HIT E 74.35 (72.53** ) 18. 20 (3.47*** ) 0.031 (2.4** ) 70.75 (70.36 )
CONST ANT 868.29 (5.78*** ) ± ± 15.79 (8.24*** )
T REND ± ± 0.038 (8.88*** ) ±
Regional ® xed e€ ects no yes yes no
R2 (adj) 0.165 0.623 0.75 0.113
F 9.61*** 16. 23*** 27.06*** 6.07***
Log-Likelihood 71022.85 7944.72 117.68 7732.53
Restricted Log-Likelihood 71040.70 71040.70 714.52 7744.17

t statistics in parentheses; *** indicates signi® cance at the 1% level; ** indicates signi® cance at the 5% level; * indicates signi® cance at the
10% level.

prison sentence for robbery is weak, if it exists at all. The the coe cients on AW HIT E and Y W HIT E are in general
average sentence duration , SENT , is only included in Table only consistently negative and signi® cant in speci® cation 1
10 (see Section III). It is consistently insigni® cant in Table which does not allow for regional characteristics. The
10 (a) and has a positive and signi® cant e€ ect in two of the exception is Table 7 where the dependent crime variable
speci® cations in Table 10 (b) . 18 These results are consistent is the robbery rate and the in¯ uences of AW HIT E and
with those of previous studies implying as they do that the Y W HIT E are both negative and signi® cant in speci® ca-
severity of the punishment has a much smaller, if any, e€ ect tions 1, 2 and 4. But neither is signi® cant in speci® cation
on crime than either the detection rate or the certainty of 3 where AW HIT E even takes a positive sign. The in¯ uence
punishment. of AW HIT E is similarly positive as well as signi® cant in
speci® cations 2 and 3 in Tables 8 (b) and 9 (b) where the
dependent variables are the theft rate and the violent
Ethnicity and crime crime rate. These results are at best mixed. There is cer-
Lastly we consider the in¯ uence of the ethnic variables, tainly no evidence of a systematic positive relationship
Y W HIT E and AW HIT E. A signi® cant, negative sign on between crime in general and the proportion of (young )
either Y W HIT E or AW HIT E would be consistent with males who are of non-white ethnic origin. There is some
the view that, even after allowing for higher unemployment evidence of a positive relationship between robbery rates
among ethnic minorities, crime rates are higher in regions and nonwhite ethnic origin but it is not compelling (given
where the non-white ethnic community is relatively large. A the lack of signi® cance of AW HIT E and Y W HIT E in spe-
positive relationship would suggest the opposite. However, ci® cation 3).
18
Regressions for speci® c crime rates were run with SENT but it was only signi® cant in speci® cation 1. Although the signi® cance of the
other variables was only marginally a€ ected the loss in degrees of freedom resulting from the inclusion of SENT meant that the results
obtained were statistically less signi® cant and they are not reported here.
 Y outh unemployment and crime 569
Table 10 (a). Dependent variable: TOTRTE

Variable Speci® cation 1 Speci® cation 2 Speci® cation 3 Speci® cation 4

Y UNEMP 285.59 (7.71*** ) 110.48 (3.96*** ) 0.0099 (2.97*** ) 32.1 (1.56 )

T OT CL R 795.98 (73.95*** ) 7136.41 (77.93*** ) 70.016 (77.85*** ) 789.72 (75.09*** )
JAIL 7130.29 (75.0*** ) 7110.00 (75.88*** ) 70.009 (73.4***) 780.0 (74.23*** )
SENT 7109.69 (71.03 ) 118.51 (1.66 ) 0.011 (1.32) 758.0 (71.18)
Y W HIT E 793.18 (73.64*** ) 723. 12 (70.45) 70.009 (70.5) 723.97 (70.69 )
CONST ANT 27 070 (6.73*** ) ± ± 2.75 (0.01 )
T REND ± 0.0086 (2.12** ) ±
Regional ® xed e€ ects no yes yes no
R2 (adj) 0.549 0.87 0.865 0.225
F 35.77*** 48. 89*** 44.62*** 2.85***
Log-Likelihood 71266.93 71168.99 138.37 71013.04
Restricted Log-Likelihood 71326.77 71326.77 717.23 71040.31
Downloaded by [The University of Manchester Library] at 08:45 22 December 2014

t statistics in parentheses; *** indicates signi® cance at the 1% level; ** indicates signi® cance at the 5% level; * indicates signi® cance at the
10% level.

Table 10 (b). Dependent variable: TOTRTE

Variable Speci® cation 1 Speci® cation 2 Speci® cation 3 Speci® cation 4

AUNEMP 477.39 (9.17*** ) 212.93 (4.22*** ) 0.023 (3.59*** ) 124.22 (3.03*** )

T OT CL R 7141.15 (76.49*** ) 7145.37 (78.93*** ) 70.016 (78.37*** ) 779.15 (74.57*** )
JAIL 797.21 (73.99*** ) 7107.62 (75.87*** ) 70.01 (73.76*** ) 785.47 (74.6*** )
SENT 10.48 (0.14) 213.21 (3.1*** ) 0.022 (2.77*** ) 724.66 (70.5)
AW HIT E 78.91 (70.27 ) 111.89 (1.03 ) 0.022 (1.76* ) 20.02 (0.29 )
CONST ANT 15 723 (3.35*** ) ± ± 227.66 (3.35*** )
T REND ± ± 0.006 (1.52* ) ±
Regional ® xed e€ ects no yes yes no
R2 (adj) 0.59 0.87 0.87 0.317
F 42.89 49. 73 46.64 12.76
Log-Likelihood 71259.27 71167.89 141.19 71013.38
Restricted Log-Likelihood 71326.78 71326.78 717.23 71040.31

t statistics in parentheses; *** indicates signi® cance at the 1% level; ** indicates signi® cance at the 5% level; * indicates signi® cance at the
10% level.

Summary those of any other type of crime (see Table 3). The rise in
each case is so great that, although the o€ ence rates are in
Our results indicate that youth unemployment and adult fact much lower for robbery and criminal damage than
unemployment are both signi® cantly and positively related either theft or burglary, it is plausible that public concern
to burglary rates. Youth unemployment is also consistently with rising crime rates may have been almost entirely
and signi® cantly related to robbery and criminal damage fuelled by increases in the incidence of criminal damage
rates. Adult unemployment has a consistently signi® cant and robbery.
and positive in¯ uence on theft and the total crime rate An explanation for the di€ erences found between the U±
(which disproportionatel y re¯ ects the theft rate) but is C relationships must lie in the nature of the speci® c crimes
not signi® cantly related to either robbery or criminal themselves as well as di€ erences in the psychological make
damage rates. This evidence is consistent with an overall up of adult males and youths. With respect to the speci® c
positive U± C relationship in the case of burglary. However, crimes, criminal damage , which includes vandalism, is a
for other crimes as well as the total crime rate the relation- crime that does not appear to have much, if any, reward
ship is conditional on the measure of unemployment used. in economic terms. Robbery is classi® ed as a violent crime
It appears that while youth unemployment motivates rob- and most of the o€ ences are street-crimes or muggings for
bery and criminal damage, rising adult unemployment is which the penalties are high and the rewards potentially
more likely to lead to a higher incidence of theft. It may, or quite low. Both these types of crime seem to lend them-
may not be a coincidence that the o€ ence rates for robbery selves well to spontaneous behaviour while acts of theft are
and criminal damage have risen considerably more than likely to require planning (especially with respect to the
 570 F. Carmichael and R. W ard
disposal of the stolen goods) and the rewards should be dence supports a positive link between adult unemploy-
easier to predict. 19 Inasmuch as these characteristics accu- ment and theft as well as total crime (of which around
rately re¯ ect the crimes concerned, criminal damage and 50% is accounted for by theft alone). This suggests that
robbery can be seen as crimes committed by those who the motivational e€ ects from unemployment lead youths
are more risk loving and perceive themselves as having and adults into di€ erent patterns of criminal behaviour.
less to lose from committing a criminal act. In this light a Younger unemployed men appear more likely to commit
positive link between youth unemployment and both rob- acts of criminal damage or robbery than theft while the
bery and criminal damage might be expected given the opposite is true for older men. Furthermore, while burglary
more headstrong nature of younger men and their possibly rates are related positively to both youth and adult un-
weaker attachment to society (see Akerlof, 1998, who employment the in¯ uence of the latter is stronger. The
makes a similar argument). explanation for these results must lie in the particular nat-
This interpretation is consistent with the conditional ure of the crimes concerned as well as di€ erences in the
nature of the U± C relationship that we have identi® ed psychological make up of youths and older males. One
and implies that acts of robbery and criminal damage are interpretation is that younger males are less risk averse
Downloaded by [The University of Manchester Library] at 08:45 22 December 2014

disproportionatel y committed by young, unemployed and have weaker attachment s to society than older males.
males. However, the nature of this study, which relies on They therefore have less to lose from committing risky and
unemployment data at the regional level and crime data more socially unacceptable criminal acts.
which is neither age nor gender-speci® c, precludes a ® rm
conclusion of this kind. All we can say is that our evidence
supports the existence of a positive U± C relationship that is
conditional on both type of crime and the measure of A C K N OW L E D G E ME N T S
unemployment. Whether it is the unemployed themselves Material from the Labour Force Survey, made available
who are committing the crimes is another question. The through the O ce of Population Censuses and Surveys
economic approach to crime implies that this is the case and the ESRC Data Archive, has been used by permission
but the econometric analysis can only support this propo- of the Controller of H.M. Stationary O ce. Crime statis-
sition and does not provide conclusive proof. However, for tics made available through the Home O ce. We would
policy makers it may be more important to establish also like to thank Christine Oughton for helpful comments
whether or not a positive U± C relationship exists rather in the initial stages of this study.
than whether it is the unemployed who are the actual per-
petrators of the criminal acts. Our evidence suggests that a
positive U± C relationship does exist subject to the quali® - R EFER ENCES
cations we have identi® ed relating to age and gender. Akerlof, G. A. (1998 ) Men without children, Economic Journal,
108, 287± 309.
Allen, R. C. (1996 ) Socioeconomic conditions and property crime:
a comprehensive review and test of the professional litera-
V . C ON C L U S I ON ture, American Journal of Economics and Sociology, 55 (3),
293 ± 308.
This study has investigated the relationship between unem- Becker, G. S. (1968) Crime and punishment: an economic
ployment and crime in England and Wales taking account approach, Journal of Political Econom y, 82, 169 ± 217.
Block, M. K. and Heincke, J. M. (1975 ) A labor theoretic analysis
of both age and gender in the unemployment measures. We of the criminal choice, T he American Economic Review, 65 (3),
additionally allowed for di€ erent types of crime, the deter- 314 ± 25.
rence e€ ects of detection and punishment and regional Britt, C. L. (1994) Crime and unemployment among youths in the
® xed e€ ects. A further consideration was ethnicity. United States, 1958± 1990: a time series analysis, American
However, we found no evidence of a consistent (positive ) Journal of Economics and Sociology, 53 (1), 99± 109.
Burden, T. (1990 ) Crime: why don’t economists have much to say
relationship between non-white origin and crime. about it? T he Journal of Interdisciplinary Economics, 3, 209±
Our results indicate that there is a systematic positive 21.
relationship between burglary rates and male unemploy- Cantor, D. and Land, K. C. (1985 ) Unemployment and crime
ment regardless of age. However, we found that while rates in the post-world War II United States ; a theoretical
youth unemployment is consistently and positively related and empirical analysis, American Sociological Review, 50,
317 ± 32.
to criminal damage and robbery rates there was no sys- Chiricos, T. G., (1987) Rates of crime and unemployment: an
tematic evidence of a relationship between adult male analysis of aggregate research evidence, Social Problems,
unemployment and these speci® c crimes. Instead our evi- 32 (2) , 189± 212.

19
More than half of recorded thefts are thefts of or from a vehicle. After vehicles, shops are the biggest target for thefts, accounting for
about one in ten recorded thefts (Home O ce Criminal Statistics, 1995).
 Y outh unemployment and crime 571
Criminal Statistics England and W ales (various issues) , The Home Hough, M. and Roberts, J. (1998 ) Attitudes to Punishment: 1996
O ce, London. British Crime Survey, Home O ce, London.
Dickenson, D. (1995 ) Crime and Unemployment, New Economy, Long, S. K. and Witte, A. D. (1981) Current economic trends:
115 ± 20. implications for crime and criminal justice, in Crime and
Downes, D. (1997 ) Prison does wonders for the jobless ® gures, Criminal Justice in a Declining Econom y (Ed. ) K. N.
Guardian, Tuesday, 25 November, 17. Wright, Oelgeschlager, Gunn and H. Medo€ , Cambridge,
Ehrlich, I. (1973 ) Participation in illegitimate activities: a theor- MA, pp. 69± 143.
etical and empirical investigations, Journal of Political Pyle, D. J. (1989 ) The economics of crime in Britain, Economic
Economy, 81, 521 ± 65. A€ airs, 9, 6 ± 9.
Elliott, C. and Ellingworth, D. (1998 ) Exploring the relationship Pyle, D. J. (1995) Cutting the costs of crime: the economics of
between unemployment and property crime, Applied crime and criminal justice, Institute of Economic A€ airs
Economics L etters, 5, 527± 30. Hobart Papers, 129.
Elster, J. (1989 ) The Cement of Society, Cambridge University Pyle, D. J. and Deadman, D. (1994) Crime and unemployment in
Press, Cambridge. Scotland: some further results, Scottish Journal of Political
Field, S. (1990 ) Trends in crime and their interpretation: a study Economy, 41 (3) , 314± 24.
of recorded crime in postwar England and Wales, Home Reilly, B. and Witt, R. (1992 ) Crime and unemployment in
Downloaded by [The University of Manchester Library] at 08:45 22 December 2014

O ce Research Study, 119, Home O ce, London. Scotland: an econometric analysis using regional data,
Freeman, R. B. (1980) Crime and Unemployment, in Crime and Scottish Journal of Political Econom y, 39 (2), 213 ± 28.
Public Policy (Ed. ) J. Q. Wilson, ICS Press, San Francisco, Reilly, B. and Witt, R. (1996 ) Crime, deterrence and unemploy-
pp. 89± 106.
ment in England and Wales: an empirical analysis, Bulletin of
Gillespie, R. W. (1978 ) Economic factors in crime and delin-
Economic Research, 48 (2), 3307± 78.
quency: a critical review of the empirical evidence, in
Timbrell, M. (1990 ) Does unemployment lead to crime? T he
Unemployment and Crime: Hearings before the
Subcommittee on Crime of the Committee of the Judiciary, Journal of Interdisciplinary Economics, 3, 223 ± 42.
House of Representatives, Washington, DC: US Willis, K. G. (1983 ) Spatial variations in crime in England and
Government Printing O ce, pp. 601± 26. Wales: testing an economic model, Regional Studies, 17 (4),
Greene, W. H. (1995 ) L imdep Users Manual, Econometric 261 ± 72.
Software, Inc, Bellport, NY. Witt, R., Clarke, A. and Fielding, N. (1998) Crime, earnings
Hale, C. and Sabbagh, D. (1991 ) Testing the relationship between inequality and unemployment in England and Wales,
unemployment and crime: a methodological comment and Applied Economics L etters, 5, 265 ± 7.
empirical analysis using time series data from England and Wolpin, K. I. (1978 ) An economic analysis of crime and punish-
Wales, Journal of Research in Crime and Delinquency, 28 (4), ment in England and Wales, 1894± 1967, Journal of Political
400 ± 17. Economy, 86 (5) , 815± 40.

A PPE N D I X : V A R I A B L E D E F I N I T I ON S A N D D A T A S OU R C E S

Variable De® nition Source

BURGRT E Burglaries per 100 000 population Home O ce Criminal Statistics

CDRT E Criminal damage o€ ences, excluding those £20 or under, per 100 000 population Home O ce Criminal Statistics
ROBRT E Robberies per 100 000 population Home O ce Criminal Statistics
T HRTE Theft and handling stolen goods o€ ences per 100 000 population Home O ce Criminal Statistics
V PRT E Violent crimes against the person per 100 000 population Home O ce Criminal Statistics
T OT RT E All noti® able o€ ences, excluding criminal damage £20 or under, per 100 000 Home O ce Criminal Statistics
population
Y UNEMP Percentage of males aged between 16 and 25 recorded as unemployed according Labour Force Survey
to International Labour O ce de® nition
AUNEMP Percentage of males aged over 25 and under 65 recorded as unemployed Labour Force Survey
according to the International Labour O ce de® nition
Y W HIT E Percentage of white males aged between 16 and 25 Labour Force Survey
AW HIT E Percentage of white males aged over 25 and under 65 Labour Force Survey
BURGCL R Percentage clear-up of burglary o€ ences Home O ce Criminal Statistics
CDCL R Percentage clear-up of criminal damage o€ ences excluding those £20 or under Home O ce Criminal Statistics
ROBCL R Percentage clear-up of robbery o€ ences Home O ce Criminal Statistics
T HCL R Percentage clear-up of theft and handling of stolen goods Home O ce Criminal Statistics
V PCL R Percentage clear-up of violent o€ ences against the person Home O ce Criminal Statistics
T OT CL R Percentage clear-up of all noti® able o€ ences excluding criminal damage of Home O ce Criminal Statistics
£20 or under
JAIL Percentage of convicted criminals receiving a custodial sentence Home O ce Criminal Statistics
SENT Average sentence duration in months Home O ce Criminal Statistics
T REND Time trend Internal to data