0

Crime and the Minimum Wage1

Kirstine Hansen* and Stephen Machin**

August 2001 December 2001 - Revised

* Department of Sociology, London School of Economics

** Department of Economics, University College London and Centre for Economic Performance, London School of Economics

Abstract This paper considers the connection between crime and the labour market in a different way to existing work. We focus on a situation where the introduction of a minimum wage floor to a labour market previously unregulated by minimum wage legislation provides substantial pay increases for low wage workers. We argue that this has the potential to alter peoples’ incentives to participate in crime. We formulate empirical tests, based upon area-level data in England and Wales, which look at what happened to crime rates before and after the introduction of the national minimum wage to the UK labour market in April 1999. The minimum wage introduction yielded sizable pay increases to low wage workers. Comparing police force area-level crime rates before and after the minimum wage introduction produces evidence in line with the notion that changing economic incentives for low wage workers can influence crime.

1 We would like to thank Nigel Beaumont, Judith Cotton and David Povey at the Home Office for kindly providing us with some of the data we use in this paper. We would like to thank David Downes, Richard Harries, Marco Manacorda, Steve Pischke, Paul Rock, Jonathan Wadsworth and participants in an LSE criminology seminar, the MIT labor lunch, the Centre for Economic Performance labour markets workshop, the LSE research laboratory opening conference, the British Society of Criminology conference at Leicester and the American Society of Criminology meetings at San Francisco for a number of helpful comments and suggestions. The Editors of this issue of the journal (Tim Hope, Susanne Karstedt and Alan Trickett) also provided us with helpful comments that improved the paper.

1

1. Introduction The way in which the volume of criminal activity moves over time, and what factors lie

behind its evolution, has been an important research and public policy question for many

years. Some of the academic research on this subject studies the role of government in

affecting crime (e.g. the way in which changes to the criminal justice system and its

governance mechanisms impact upon crime). Other work tries to identify why crime rates

fluctuate and why different kinds of individuals are more or less likely to participate in

crime. In this paper we are principally interested in issues of the latter kind. In particular

we consider the way in which one possible determinant of criminal behaviour, the state of

the labour market, impinges upon crime.

We approach this question in a rather different way to existing research on crime

and the labour market. We study what happened to crime before and after a big

regulatory change was made to the UK labour market, namely when a National Minimum

Wage (NMW) was introduced in April 1999 (previously there had been no minimum

wage in operation). If labour market conditions are related in an important way to crime,

or individual’s propensities to commit criminal acts are altered by changing labour

market opportunities, then one may well see changes in crime occur in the time period

surrounding minimum wage introduction.

In this paper we test this hypothesis using police force area level data from

England and Wales. We start the paper by setting the scene in terms of a review of the

relevant accumulated research literature on crime and the labour market. We then move

on to consider theoretical arguments as to why a link between crime and the measure of

the labour market we focus on in this paper, low wages, might exist. This theoretical

discussion is very much lined up to motivate the test of the relation between crime and

low wages that we implement in the empirical part of the paper.

2

How can one test for a link between crime and low wages? We look at crime and

wages but, as already noted above, in a rather different way to that of earlier work. If one

thinks that differential wage opportunities matter for crime then presumably the best way

of testing for the existence of a crime-wage link is to look at a situation where people on

the margins of criminal participation receive a (potentially large) wage increase. Such a

situation is clearly offered when a binding minimum wage floor is introduced to a labour

market that previously was not regulated by minimum wage legislation.

We report a set of findings which rest well with the notion that giving low wage

workers sizable wage increases displays an association with crime rates. Changes in

crime rates before and after the minimum wage introduction in April 1999 are seen to be

lower in areas with more workers affected by the introduction of the NMW. Perhaps not

surprisingly, the negative association is seen to be much stronger statistically for property

and vehicle crimes than for violent crimes.

Moreover, when we look back and benchmark against periods before the

minimum wage introduction we find the negative association between crime and the

incidence of low wages to be much stronger in the period surrounding the minimum wage

introduction. In fact there is a weak negative association between changes in crime and

low pay in the earlier non-minimum wage period (between 1995 and 1998), and a very

strong negative one in the year around minimum wage introduction. As such the

introduction of the minimum wage to the UK labour market does seem to have been

associated with less crime in low wage vis-à-vis high wage areas.

2. Empirical Work On Crime And The Labour Market

Early empirical work in this area tended to focus heavily on links between crime

and unemployment. Surveys of this work by Freeman (1999), Box (1987) and Chiricos

3

(1987) report that the relationship between crime and unemployment appears fragile at

best. Some studies have detected a positive relationship between crime and

unemployment (Land, McCall and Cohen 1990; Levitt 1996, 1998), but this is often more

easily found in studies using individual longitudinal data (see, for example, Thornberry

and Christenson 1984; West 1982 for the UK), The same is true if specific2, rather than

aggregate, unemployment rates are examined. However, other studies have found that, in

data where there exists a statistically significant unconditional correlation between crime

and unemployment, once other variables are taken into account the relationship between

crime and unemployment disappears (examples are Butcher and Piehl 1998 for the US

and Machin and Meghir 1999 for England and Wales). Even stronger than this, others

have found there to be no relationship between crime and unemployment at all (Cullen

and Levitt 1999).

This weak pattern of results is not so surprising when one realises that there are a

number of conceptual reasons why unemployment may not be the most appropriate

labour market variable to examine in relation to crime. Because criminal participation is

unlikely to be something that most individuals enter into lightly crime may well be more

responsive to long term labour market measures than to short run ones such as

contemporaneous unemployment (Gould et al, 2002). Indeed, there is a much larger body

of individuals who, although in employment, are in insecure low paid low skill jobs, or in

part time or temporary work, who are economically and socially marginalized. Moreover,

by the very fact that they are employed and socially connected these people may be in a

2 Usually the unemployment rate of young males in explaining crime amongst young men (as in Freeman and Rogers 1999 and Allan and Steffensmeier 1989 for the US and Reilly and Witt 1996 for the UK).

4

better position to commit crimes than the unemployed (see Box 1987; Fagan and

Freeman 1999; Grogger 1998).3

Because of this a number of studies have looked at broader measures of crime and

the labour market. Sampson and Laub (1993), using the 1939 Boston Cohort, found that

job stability was negatively related to subsequent criminal behaviour. Crutchfield (1989),

looking specifically at violent offences, found that labour instability was a significant

predictor of overall violence, murder, assault and robbery. The link between job stability

and crime has also been highlighted in UK research by West and Farrington (1977) and

Farrington (1986).

Along the same lines, Allan and Steffensmeier (1989) found the quality of work is

important in relation to crime, and for young adults there is a strong association between

individuals who work less hours than they would like to and crime. Hale (1998), in his

UK study, found that changes in the structure of employment are related to crime, in

particular shifts from manufacturing to the service sector, increasing part time and

temporary jobs and changes between male and female employment. Similarly, and again

based on UK data, Farrington et al (1986) found that individuals were more likely to

offend if they worked in low status jobs.

Related to these studies a number of researchers have looked at the role of

economic incentives. The US evidence of Fowles and Merva (1996) and Hsieh and Pugh

(1983) found poverty to be positively related to crime (the latter looked at violent crimes

only). Other US studies link the rise in crime to widening wage inequality, which has

been witnessed since the 1970s as a result of a decline in both relative and absolute wages

at the bottom end of the market (Fowles and Merva 1996; Blau and Blau 1982; Hsieh and

3 In the 1980 wave of the National Longitudinal Study of Youths (NLSY) over half of those working reported that they had committed some crime and one fifth of those working had committed at least 1

5

Pugh 1983). Similarly, Witt, Clarke and Fielding (1999) have looked at police force area

data in England and Wales from 1988 to 1996, finding that wage inequality changes are

positively correlated with changes in crime.4

A more recent body of work has, by looking at the wage rates of low skilled

workers, concentrated on those at the bottom of the wage structure rather than looking at

the gap between the top and the bottom of the wage or income distribution. Gould et al

(2002) look at the relationship between changes in crime and changes in wages across

areas in the US between 1979 and 1995 and report that the falling wages of unskilled men

in this period led to a rise in burglary of nearly 14%, a rise in larceny/theft of around 7%,

a 9% increase in aggravated assault and an 18% rise in robbery. From data on the police

force areas of England and Wales between the mid 1970s and mid 1990s, Machin and

Meghir (1999) looked at cross-area changes in crime in relation to changes in the 25th

percentile of the area wage distribution. They found a negative correlation between the

types of crime they examine (theft and handling, burglary, vehicle crime and total

property crime) and low wages, even after controlling for other variables including

demographic change and measures of deterrence. Finally, Grogger (1998) uses data from

the US National Longitudinal Survey of Youth to look at the relationship between wages

and property crimes for young people. He reports results which show falling real wages

not only offer an explanation of the rise in youth crime in the 1970s and 1980s but also of

the differences in criminal involvement between age and ethnic groups.

These latter findings are clearly in line with the idea that economic incentives are

important for crime. Moreover, they also suggest that wage measures, especially

income producing crime (Fagan and Freeman, 1999; Grogger, 1998). In Fagan’s (1992) study more than 25% of drug dealers were also working. 4 A smaller body of work has looked at other measures of economic activity. For example, Witt and Witte (2000) consider the relation between crime and female labour supply, reporting results based on US time series showing common trends in crime and female labour force participation.

6

measures for workers towards the lower end of the wage distribution, may provide better

measures of the state of the labour market for people on the margins of crime than

unemployment. We also look at crime and wages in this paper, but by adopting a

different methodological approach compared to other work. We consider what happened

to crime before and after the introduction of the National Minimum Wage (NMW) to the

UK labour market in April 1999. This provides a good testing ground for looking at the

impact of a wage change for people deciding whether to participate in crime as the wage

increases received by low wage workers were sizable. Metcalf (1999) estimates that

about 2 million workers would receive wage gains from the imposition of the NMW.

Moreover, the average wage gain for workers paid less than the NMW of £3.60 per hour

(£3.00 for 18-21 year olds) before its introduction was estimated to be of the order of 30

percent.

3. Why Should There Be A Link Between Crime and Low Wages?

Theoretically there are a number of reasons for thinking that low wages should be

related to crime and how the introduction of a minimum wage would affect this

relationship. Firstly, simple choice theoretic models of crime (e.g. Becker 1968 or Ehrlich

1973) formulate that individuals have a choice between crime and work, or more

generally they choose to allocate their time across crime-work space. These decisions are

a function of a number of factors, including expected earnings from crime, expected

earnings from the labour market, and perceptions of the severity of the punishment if one

gets caught. Seen as a simple work/crime decision this explains why people with no work

may decide to partake in crime. But on a more complex level this can also shed light on

how individuals who are employed may also decide to commit crimes and the extent to

which they allocate their time between work and crime (for, as already noted above, we

7

know that many people do both). Thus, an increase in legal wages brought about by the

introduction of the national minimum wage should reduce the incentive to participate in

illegal activities thus bringing the crime rate down. Also by raising wages workers now

have more to lose by getting caught, which should also act to discourage criminal activity

and reduce crime.5

Of course, these simple choice based models of crime have themselves been

called into serious question for their relatively simplistic assumptions about criminal

behaviour. But other theoretical approaches generate a relation between crime and low

wages. Strain theory, for example, predicts that people with low wages are likely to suffer

financial hardship, sometimes in similar ways to those who are unemployed (Merton

1957; Cohen 1955; Cloward and Ohlin 1960). This financial strain may well encourage

individuals to commit acquisitive crimes either for themselves or to sell for cash in order

to obtain the goods they cannot afford. Financial strain may also lead to feelings of

frustration or anger, which may well manifest themselves in violence. Thus, we would

expect low wages to be associated with relatively high rates of both property and violent

crimes. An increase in wages brought about by the introduction of the minimum wage

may ease financial strain, which may well lead to a reduction of both types of crime.

Over and above the financial strain faced by the low paid, their situation is

worsened by the fact that most of them will be in jobs where promotion or career

advancement is hard (if not impossible). Thus their opportunities to have money and

status may be blocked. Unable to achieve success legally these individuals may be forced

to resort to illegal methods. Moreover, such individuals are more likely to live in poorer

areas where it is possible illegal opportunities to achieve goals are more abundant than

5 It is something of an unanswered question as to whether the economic model is only relevant to non-violent crimes for which monetary incentives may alter behaviour or whether it can also be extended to the

8

legal opportunities (Cloward 1959). In these areas there may also be peer pressure to get

involved in crime (Cloward and Ohlin 1960) or increased opportunity for learning

criminal behaviour through associations and interactions with other criminals (Sutherland

1924; Akers 1977). As noted above the introduction of the minimum wage in the UK

raised low wage workers’ wages by a sizable amount and, in doing so, may well have

reduced the need to turn to crime to achieve success or status. It may even eventually

give people the power to migrate to better areas where there is less criminal peer

pressure, but this would be much more long term.

Finally, as employment is one of the major institutions through which social

bonds are formed between individuals and society, social control theory predicts that

employees with low paid jobs may be less attached to society (Hirshi 1969; Box 1971).

Thus, crime rates may be high amongst those in low paid jobs as social controls will be

less able to deter them from breaking the law. If wages are increased due to the

implementation of the National Minimum Wage, this may act as a mechanism for

strengthening the social bonds between the low paid and society. More tied to society and

therefore more constrained by social controls this group will be less likely to commit

crimes.

Thus, there are a number of potential explanations as to why the introduction of

the minimum wage may influence crime and help us try and pin down a link between

crime and the low wage labour market. It is on the basis of these ideas that we developed

our hypothesis that the introduction of the minimum wage may have the potential to

reduce crime. We next turn to the methodology we utilize to formulate tests of this

hypothesis.

case of violent crime. Various researchers have taken different stances upon this (though see Grogger 2000 for an interesting attempt to apply the economic model to violent crime).

9

4. Methodology

Our empirical methodology involves comparing what happened to crime rates

before and after the minimum wage introduction in the police force areas of England and

Wales. We relate changes in various crime rates before and after minimum wage

introduction to the initial proportion of low wage workers (i.e. those paid less than the

minimum wage prior to its introduction) in those areas. This is much the same

methodology as that adopted in some US work (notably Card, 1992) to look at the

relationship between employment and minimum wages in US states before and after the

large federal minimum wage increase of April 1990. Identification of the minimum wage

effect comes from the fact that there are more low wage workers in some areas than other

and therefore the minimum wage should be thought of having more of an effect there

than in areas where there are fewer low wage employees. As Card (1992 p.22) puts it:

‘From an evaluation perspective……a uniform minimum wage is an under-

appreciated asset. A rise in the federal minimum wage will typically affect a larger

fraction of workers in some states than in others. This variation provides a simple natural

experiment for measuring the effect of legislated wage floors, with a “treatment effect”

that varies across states depending on the fraction of workers initially earning less than

the new minimum’.

This approach to looking at crime and the labour market is founded upon the idea

that a sizable change in labour market opportunities has the potential to alter an

individual’s incentive to participate in crime. The theoretical approaches outlined in the

previous section highlight that an individual’s propensity to commit crime, say C6, will

depend on a number of factors such that, in general terms, C = C(Wc, p, S, W, Z) where

6 C may reflect a discrete 0-1 choice between work and crime or could reflect the allocation of hours per week between formal labour market activity and criminal actions (see, for example, Ehrlich, 1973). As we are interested in wages and crime the latter is probably more appropriate.

10

Wc is the earnings from a successful crime, p is the probability of being caught, S is the

punishment, W is the earnings available on the legitimate job market and Z are other

factors relevant for crime. According to the theoretical approaches the C(.) function

depends positively on Wc and negatively on p, S and W. It therefore reveals a clear trade

off between perceived earnings from crime and formal labour market activities. One can

aggregate the C(.) function to area-level so that C(.) becomes the area-specific crime rate

(= the number of people engaging in crime divided by the population). Our empirical

approach, based on looking at the differential impact of the minimum wage introduction

across areas, can be thought of as providing a positive (and sizable) increase in W. As

long as its impact is not offset by coincidental changes in Wc, p, S or Z (which it must be

said seems highly unlikely in the short time period we consider) one should see crime fall

in areas where W has the potential to rise by more.

The main factors in Z, the other determinants of crime, are likely to be those other

factors that influence both the supply and demand for crime. In a simple supply-demand

framework, the demand side can be thought of as being characterised by an inverse

relation between crime and criminal earnings, while the supply side is driven by the wage

and criminal justice system variables. The demand for crime is likely to be shaped by

demographics (e.g. if there are more rich consumers perhaps the pickings from crime

may be more lucrative) and so we also control for demographic changes over the short

time period we consider.7

So, to briefly recap, our empirical approach will be to compare changes in area-

specific crime rates before and after the introduction of the NMW in April 1999. The

quasi ‘natural experiment’ created by the fact that some areas have more low wage

7 These are: change in average age, change in the population share of young (<25) men, change in proportion black, change in population share with no educational qualifications, change in proportion female, change in share of public sector jobs.

11

workers than others will be exploited to see if the minimum wage had the potential to

reduce crime in the time period surrounding the minimum wage introduction.

5. Data

Crime Data

The crime data we use are notifiable offences reported and recorded by the police

force areas of England and Wales. Crime rates were obtained for 41 police force areas8 in

the six-month period (April-September 1999) immediately following the introduction of

the NMW. For most of our analysis we compare and contrast crime in this six-month

period with crime in the six-month period before (October 1998-March 1999) or with the

same six-month period in the preceding year (April-September 1998). In what follows

the choice of comparison group is not that important, though intuitively the analysis of

the same six-month (April-September) periods across years is probably slightly more

attractive.

We focus upon three different crime rates: property crime (defined as burglary

plus theft and handling); vehicle crime (theft of a vehicle, theft from a vehicle,

aggravated vehicle taking, vehicle interference and criminal damage to a vehicle); and a

measure of violent crime (violence against the person). We consider the latter, even

though much of the modelling of crime and the labour market is largely concerned with

the way in which altering economic opportunities affects crime and we would probably

think there is more potential for this to affect non-violent crimes.9

8 There are actually 43 police force areas in England and Wales: in our analysis we aggregate the City of London and Metropolitan police forces and the Gwent and South Wales police forces. The reason for joining together the City and Metropolitan forces is because the small number of residents living in the City produces artificially high crime rates. Gwent and South Wales have been aggregated as a result of a police force boundary change that occurred. As a result we consider 41 consistently defined areas in our empirical work. 9 Indeed, this is what Gould et al. (2002), Machin and Meghir (1999) and May (2000) find in their area-level analyses of crime and wages. Both find strong, negative associations between non-violent crime and

12

One should note that use of official crime statistics may mean our analysis

perhaps overlooks those crimes that are not reported to or recorded by the police, referred

to as the ‘dark figure’ of crime (see McDonald 2001). However, insurance requirements

together with increased communication and public awareness of crime has meant that a

large number of crimes do now appear in the official statistics and that a large percentage

of those that do not are minor or trivial offences. Our focus on very recent time periods is

very much helped by this. Moreover, because we are looking at cross-area patterns of

crime change, if this ‘dark figure’ of unknown crime varies randomly across the areas we

are looking at then it should not bias our results.10 Similarly the fact that most of our

analysis is based upon changes over short time periods means that our results are unlikely

to be contaminated by reporting biases of this kind.11

Labour Market Data

The area-level labour market data is obtained by aggregating the UK Labour

Force Survey (LFS), an individual-level survey, to police force area. The LFS is a

quarterly survey of around 60,000 households in Britain. The data has been set up to

match the six-month crime data, as the LFS reports the month in which people are

interviewed.

The LFS contains information on hourly earnings (derived from separate

questions on weekly wages and hours) and these are used to define our low pay variables.

The NMW was introduced in April 1999 at £3.60 per hour for people aged 22 or higher,

and at £3.00 per hour for those aged 18-21 (inclusive), so we look at the proportion of

low wages, but much less of a link with violent crime. Although others have found a relationship between violent crime and the inequality of wages (see Hsieh and Pugh, 1983 or Land et al, 1990). 10 Furthermore, in England and Wales the official statistics provide the only source of data on crimes by police force area. The British Crime Survey (which as a victim survey some argue captures, at least partially, the ‘dark figure’ of crime) does not have (publicly available) information on areas. 11 Indeed, McDonald (2001) makes the very point that time series analyses may suffer from bias problems because the under-reporting of crime varies systematically with the economic cycle. Our short time period of study is a clear advantage in this regard.

13

workers aged 18 or higher below the appropriate level by area. To consider our key

hypothesis, we are interested in whether the change in crime rates before and after

minimum wage introduction were seen to vary with the initial number of low paid

people.

The LFS contains a great deal of information regarding demographics and job

structure, which has allowed us to additionally set up a number of other variables for our

analysis. One variable of particular interest is the area unemployment rate, a variable to

which we devote some attention later for two reasons. First, as noted earlier, a lot of

work on crime and the labour market has looked at crime-unemployment correlations.

As such we will be interested in exploring whether such a correlation exists here.

Second, and more important, is the possibility that the imposition of the minimum wage

has an effect on unemployment and that this then impinges on crime. We discuss this in

more detail later.

We have also assembled other LFS variables to enter into our crime equations as,

despite the fact that we consider a short time period, we do not want to confound any

minimum wage effects with shifts in demographic structure. These variables, which

theory and past empirical work inform us may be important when examining the

correlates of crime, include variables related to age, education, race, gender and so on

(they are detailed in full in the notes to the Tables and in footnote 5 above). Finally, to

capture any shifts in the probability of detection we consider changes in the crime clear

up rate across the relevant comparison periods.

Descriptive Statistics

Table I presents descriptive statistics on average wages, wage inequality and

average crime rates for three groups of areas, delineated by the proportion low paid into

areas with a lot of low wage workers in the period prior to minimum wage introduction

14

(Most Low Pay), a middle group of areas (Middle Low Pay) and the areas with the fewest

low wage workers (Least Low Pay).12

For each of these groups of areas the Table reports the mean log(hourly wage), a

measure of wage inequality for the bottom half of the wage distribution (the 10-50

log(hourly wage) differential) and mean crime rates (defined per 1000 population).

These are reported for the three six month periods of interest, along with the change

before and after the NMW introduction (with associated standard errors) calculated for

the two possible six month pre-minimum wage periods. Differences in the change

between areas with the most and least low pay are also reported, with the changes over

time in the period surrounding minimum wage introduction given in bold typeface.

The numbers in the Table show that, in terms of wages and wage dispersion, the

introduction of the minimum wage operated largely as one would expect. Average wage

growth was not so different across the groups of areas but wage dispersion at the bottom

end of the wage distribution fell by more in low wage areas. In fact the 10-50 differential

rose by 3 percent in the areas with most low pay, by 2 percent in the middle group of

areas and was more-or-less unchanged in the areas with the least amount of low pay. The

gaps between the areas with most and least low pay are seen to be statistically significant.

As such the imposition of the minimum wage seems to have significantly improved the

relative earnings position of the low paid (abstracting away from any impact on jobs, an

issue we return to later).

Over the pre- and post-minimum wage periods average crime rates do not alter

much for the non-violent crimes, with property crime rising from 30.8 to 31.1 crimes per

1000 population and vehicle crime falling from 14.5 to 14.3 crimes per 1000. Violent

12 The exact cutoffs were found by ordering the data on the proportion of workers paid less than the NMW in the year preceding its introduction. High wage areas had less then 9.7 percent of below NMW workers,

15

crime, on the other hand, shows more of an increase, going from 4.9 to 5.9 crimes per

1000 people. It is, however, of considerable interest that there are differential patterns of

change for the different area groupings. Specifically, property and vehicle crimes decline

in low pay areas, but actually increase in high pay areas compared with either one year or

six months earlier. Violent crime appears to go up in all three groups, but by less in the

low pay areas.

There is therefore clear evidence in the Table that wage inequality fell and crime

fell (or rose by less) in the areas with more low wage workers in the period surrounding

minimum wage introduction. All six of the Most Low Pay Vs. Least Low Pay

comparisons of changes in crime, given in bold, prove to be negative and five of the six

are statistically significant. Thus the descriptive findings appear to be in line with the

hypothesis that crime is likely to have fallen where the minimum wage has had more

impact.

This finding is further illustrated in Figure I which plots changes in the three

crime rates against the initial low pay proportion across police force areas. The graphs

seem to show that crime went up by less in the areas with more low paid workers in the

period before minimum wage introduction.13 This is borne out by the regression lines

fitted through the data points, all of which show a negative relationship between crime

change and the initial low pay proportion. We subject this finding to a more rigorous

analysis in the following Section of the paper.

middle wage areas between 9.7 and 11.7 percent, and low wage areas had over 11.7 percent of workers paid less than the NMW. 13 The violent crime Figure shows a large change in violent crime in the West Midlands over this time period. We checked the Home Office data and contacted the West Midland police to ensure that this is not an error in the data. We are very grateful to Dylan Harthill of the performance review section of the West Midlands police force for confirming that this is a genuine rise and not a data error.

16

6. Statistical Results

Basic Regression Results

Table II reports a set of statistical regressions that relate changes in crime to the

proportion of workers paid less than the NMW in the initial period. Six sets of

specifications are reported for each crime measure. In the first three columns of the Table

the change is defined as the change over the six-month post-minimum wage period April-

September 1999 as compared to the same six-month period in the preceding year. In the

final three columns the April-September 1999 six month period is compared to October

1998-March 1999. For each pre-minimum wage benchmark group, the three reported

specifications differ in terms of variables included. The first is a simple regression of the

change in crime on the initial period proportion of workers paid beneath the minimum.

The second adds in the demographic controls and the change in the crime clear up rate

variable discussed above. The third additionally adds in the change in the unemployment

rate between the pre- and post-minimum time periods.

Considering first the basic specifications, a negative relationship between the

three crime measures and the initial proportion low paid emerges. This is true irrespective

of which benchmark period is used to fix the pre-minimum wage reference time period.

This just re-confirms the raw pattern suggested by the numbers in Table I, namely that

crime has gone up by less in the lower wage areas more affected by the introduction of

the minimum wage.

Adding in the variables measuring changes in the demographic structure14 of

areas and the change in the clear up rate pre- and post-minimum wage introduction tends

to reduce the estimated coefficient on the initial period low pay proportion. This is

14 The estimated coefficients on these variables are not reported as our main concern is with the initial proportion low paid variable. It is, however, worth noting that the coefficient on the change in the clear up rate was usually negative and significant.

17

particularly true for the comparisons based on the same six-month April-September

periods. In these models, in the property and vehicle crime equations the negative

coefficient on the initial proportion remains sizable and strongly significant in statistical

terms. For violent crimes the estimated coefficient is, however, driven to statistical

insignificance. This is probably not that surprising: one would think that shifts in

demographics are more likely to be important for violent crimes.15

As noted above a critical question surrounding the introduction of the NMW is its

likely impact on unemployment. There was much speculation on this before the

introduction of the wage floor as opponents of minimum wage legislation argued that

minimum wages tend to hurt those they set out to initially help as the imposition of a

minimum wage prices workers out of jobs.16 Were this to be true there would be another

mechanism we would need to consider here, namely that there would be more

unemployed workers who could not get jobs who may then turn to crime. In this case the

minimum wage may raise crime rates. For this reason it is important that we also control

for changes in unemployment that may have occurred differentially across areas, in case

we are biasing the coefficient on the low pay proportion by neglecting another route in

which crime may be affected by the labour market.

The final specifications therefore add in changes in the log(unemployment rate).

In almost all cases this has little impact. There is no statistically significant association

between changes in crime and changes in unemployment, and the estimated coefficient

15 See Land et al (1990) for a discussion of the importance of demographic characteristics in predicting violent crime. 16 Of course there has been a lot of (sometimes acrimonious) debate about the economic effects of minimum wages, especially their impact on unemployment. This is not of major concern to us here, but see Card and Krueger’s (1995) book and the symposium in the Industrial and Labor Relations Review of July 1995 for a flavour of the strong views held in the US debate or Metcalf’s (1999) discussion of the UK debate.

18

on the low paid proportion does not shift much at all on its inclusion.17 The only possible

exception to this is the change in the violent crime rate for the April-September year-on-

year comparison. Here the coefficient on the unemployment rate is estimated to be

positive and is right on the margins of significance at the 10 percent level with a p-value

of .106.

Using Other Wage Measures To Gauge The Initial Proportion Variable

The results so far point to a negative association between changes in property and

vehicle crime rates over the period of minimum wage introduction and the incidence of

low pay. We take this as evidence that shifts in the nature of low wage labour markets

have the potential to affect crime. However, there are some issues that should be

addressed about the wage variables we use to compute the initial period low pay

proportion. The first is that so far we have only considered a simple head count measure

of low pay. How far workers’ wages are beneath the minimum may also differ across

areas so one may wish to look at wage gap measures as well. Second we have thus far

looked at all workers in the Labour Force Survey, but men commit most crime. Further,

one may think that it is workers in ‘dead end’ jobs going nowhere, or who are stuck in

certain ‘permanent’ low paid jobs, that may be more movable on the crime-work choice.

For these reasons we have also estimated regression models that refine the nature

of the initial low pay variable. The first three results columns in Table III use a wage gap

measure of low pay, in place of the headcount measure considered earlier. This wage gap

basically measures what share of the area wage bill would need to be paid to bring the

17 This is not a consequence of looking at unemployment rates (see Chamlin and Cochran, 2000, who argue that use of conventional unemployment rates can obscure crime unemployment relations due to a priori measurement choices and that use of other measures may uncover a link with crime). If, instead, one entered the change in the employment rate or the inactivity rate in the area into the equations similar results emerged. For example, for the change in property crimes, the estimated coefficient (standard error) on the initial low pay proportion was -.888 (.289) if the change in the employment rate was added (the employment rate itself had a coefficient -.728 with an associated standard error of .832); if the area-specific

19

low wage workers in the area up to the minimum. The pattern of results using the

headcount in Table II is very much reconfirmed for the wage gap measure specifications

in the Table.18

We have also considered a wage measure for specific low wage workers across

areas. The results using this are given in the remainder of Table III which uses an area

initial low pay measure defined for men working in low skill occupations (defined as

those with an average wage beneath the 25th percentile of the male wage distribution).

We think this is useful as it is low wage men who we would probably think are those on

the margins of crime and who, if the hypothesis advanced about big wage boosts

lowering crime are correct, we should therefore focus upon. The results very much

confirm that property and vehicle crimes went up by less in areas with more low skill

men who were paid beneath the minimum in the period before it was introduced. Again

there is little link with violent crime. We view these findings as strongly supportive of

the notion that the minimum wage gave a sizable boost to people on the margins of crime

thereby shifting them away from crime.

The final issue to do with the wage measures concerns possible measurement

errors. It may be that measurement errors in hourly earnings taken from the Labour Force

Survey do not necessarily produce accurate measures of the size of the area-specific low

wage labour market. Of course if such measurement errors do not vary systematically by

area then this would not be a concern. We have investigated this question by computing

measures of the area low pay proportion from another data source, the employer reported

New Earnings Survey, which is a large sample (1 percent of the working population) of

change in the inactivity rate was included the coefficient (standard error) on the low pay proportion was -.835 (.281) and the coefficient (standard error) on the change in the inactivity rate was .098 (.202). 18 Notice that the scale of the coefficients is sizable. This is because the average wage bill share needed to raise workers to the minimum is a fairly small number (the mean is .0085 when expressed as a proportion, or 0.85 percent of the wage bill).

20

workers carried out in April each year. There is a concern about this data source, namely

that it does tend to undersample low wage workers (as one needs to have weekly earnings

above the National Insurance lower earnings limit to be in the survey). Nevertheless we

also computed the low pay proportion for April 1998 from these data. The measure is

strongly correlated with the LFS measure (correlation coefficient = .89). This gives us

confidence that our LFS based measure is likely to be a good measure of the state of the

low wage labour market in the areas we consider.

As we now have two measures of the initial low pay proportion this opens up the

possibility of using instrumental variable techniques to assess whether measurement error

is a problem. The following Table shows what happens when we instrument the LFS low

pay proportion using the NES low pay proportion as an instrumental variable (IV). The

IV and OLS results are similar but, if anything, there appears to be a stronger link with

changes in crime, for all three crime categories, when we use the IV techniques in an

attempt to purge measurement error.

OLS IV Change in Log(Property Crime Rate) LFS Proportion Paid Less Than The Minimum Wage in Period 1 -.928 (.307)*** -1.153 (.328)*** Change in Log(Vehicle Crime Rate) LFS Proportion Paid Less Than The Minimum Wage in Period 1 -.964 (.316)*** -1.269 (.431)*** Change in Log(Violent Crime Rate) LFS Proportion Paid Less Than The Minimum Wage in Period 1 -1.312 (.655)* -1.872 (.821)**

Notes: Estimated coefficients (standard errors) on LFS initial proportion variables based on specification using (April 1999 – September 1999) – (April 1998 – September 1998) change in crime models as reported in Table II. From full models containing all controls and the change in log(unemployment rate). NES initial low pay wage survey proportion as instrument for LFS initial proportion. *** denotes statistically significant at 1% significance level or better, ** 5%, * 10%.

Benchmarking Against Earlier Time Periods

A potentially very important concern that emerges from considering the results

presented so far is whether we are really identifying any change resulting from studying

the minimum wage period. For example, it might be that crime rates have not been rising

21

as fast in low wage areas in time periods when the minimum wage was not present. Were

this to be the case our results may be spurious.

We have explored this possibility by looking at econometric models specified in

the same way as those considered to date for earlier time periods. In the simplest

specification reported before (in column (1) of Table II) the regression relationship

between changes in property crime and the proportion below the minimum wage in the

initial period for the periods around minimum wage introduction was as follows (standard

error in parentheses):

Change in Log(Property Crime) = -1.026 Proportion Paid Less Than The Minimum Wage in Period 1 (.349)

For earlier periods of change [(April 1996 – September 1996) – (April 1995 –

September 1995)] and [(March 1998 – October 1997) – (March 1997 – October 1996)]19

the regression relationship is:

Change in Log(Property Crime) = -.276 Proportion Paid Less Than The Minimum Wage in Period 1 (.181)

So, in this earlier time period there is a (weak) negative association between

changes in property crime and the initial low pay proportion, but it is nowhere near as

marked as around the minimum wage introduction period. Indeed, the regression line fit

through the points has a slope four times as large (in absolute terms) in the period

surrounding minimum wage introduction.20 This shows a tilting of the crime low pay

19 This sample period is dictated by the availability of the county-level data in the LFS which does not allow us to go back any earlier. Notice also that two possible comparison periods are left out. This is because the Home Office changed their definitions for data collection and recording of crime from the police force areas of England and Wales in April 1998. Any changes that span this period therefore had to be dropped from the analysis. So we omitted the [(March 1999 – October 1998) – (March 1998 – October 1997)] and [(April 1998 – September 1998) – (April 1997– September 1997)] time periods. For more on the nature of the recording change (which affected violent crime definitions by substantially more than property crime definitions) see Home Office (1999). 20 Of course, as the periods not surrounding minimum wage introduction are pooled, the regression slope is the average slope across all periods. However, if each period is taken individually, the slope is always markedly steeper in the period surrounding minimum wage introduction.

22

relationship such that the relationship between changes in crime and low pay becomes

stronger in the period when the minimum wage was introduced.

A more formal way of thinking about this is to explicitly couch the modelling

approach in a “difference-in-differences” framework. Our analysis covers two distinct

time periods, one where the minimum wage raised wages by more in low wage areas

(which we can call period M), and one where no minimum wage legislation was in place

(period NM). We can therefore benchmark our measures of the change in crime from the

period surrounding minimum wage introduction ∆CM against our measure of the change

in crime from the non-minimum wage period ∆CNM. Looking at the relationship between

∆CM - ∆CNM and the initial period low paid proportion then provides a “difference-in-

differences” estimator of the change in crime-low wage relation.

For the case of changes in property crime the estimator is simply the gap between

the coefficients on the initial low pay proportion variable across the two specifications

(for the example considered this is –1.026 – {-.276} = -.750). As column (1) of the upper

panel of Table IV shows, despite this being a stiff test of our hypothesis, this difference-

in-difference estimate remains statistically significant and shows a marked shift in the

relationship between changes in property crime and the initial low pay proportion when

benchmarked against earlier time periods.

The rest of the upper panel of Table IV is devoted to presenting more detailed

specifications for property crimes and the middle and lower panel of the Table report the

same models for the other crime measures. In all cases there seems to have been a shift in

the relationship between changes in crime and the initial low pay proportion across the

time periods considered as the effects are estimated to be much more negative in the

23

period surrounding minimum wage introduction than in the periods before.21 This is true

for all three specifications of property and vehicle crime equations, though not for violent

crime.22

The nature of the data, on the same areas followed through time, means that one

can also adopt an even more stringent test by including area-specific trends in the

estimating equation. The final column of the Table therefore additionally includes 41 area

trend variables. The results are very robust to this. The estimated coefficient on the

minimum wage variable is slightly reduced in absolute terms for property and vehicle

crimes and, of course as one would expect, the standard errors rise. But the coefficient

remains significant at better than the 2 percent level (p-value = .014) for property crimes

and at better than the 12 percent level (p-value = .114) for vehicle crimes. The violent

crime results, probably not surprisingly, become extremely imprecise as the standard

error almost doubles. But the fact that the non-violent crime results remain very resilient

to this strong test seems very reassuring. Our reading is therefore that our results are

strongly supportive of the idea that changes in non-violent crime were significantly lower

in areas where workers’ wages were more affected by the introduction of the NMW.

Discussion of the Size of the Estimated Link Between Crime and Minimum Wages

An important question concerns the magnitude of the estimated links between the

minimum wage introduction and crime. Our results clearly show that areas that were

likely to have been more affected by minimum wage introduction saw lower changes in

crime in the period surrounding the policy introduction. One could think about assessing

21 If the comparison is restricted to April-September comparisons the same pattern of a statistically significant negative relationship between changes in property and vehicle crime and low pay incidence in the minimum wage period remains. For the column (3) models the estimated coefficients (standard errors) on the initial low pay proportion were: property crime -.820 (.330); vehicle crime -.718 (.333). Again the relationship with changes in violent crime was statistically insignificant (coefficient = -.766, standard error = .753).

24

the magnitude of this relationship in a number of ways. One option is to compute the

elasticity of crime changes with respect to the initial low pay proportion so as gauge how

sensitive crime changes were to variations in the proportion of low wage workers across

areas. As the dependent variable is the change in the log(crime rate) and the independent

variable of interest is the level of the initial low pay proportion, then this elasticity can be

computed as the estimated coefficient on the low pay variable multiplied by its mean.23

For property crimes the range of estimates of the coefficient on the initial low pay

proportion variable is from around -.65 in the most stringent difference-in-difference

estimates up to around -.9 in the earlier specifications with control variables included.

The mean of the initial low pay proportion is .10 so this gives an elasticity in the range of

around -.065 to -.090. So an area with a 50 percent higher initial low pay proportion (say

.15 rather than .10) is predicted to have a change in property crime around 3.3 to 4.5

percentage points lower in the period surrounding minimum wage introduction. The

comparable numbers for vehicle crime are much the same as the estimated coefficients

are in the -.6 to -.9 range, corresponding to changes in vehicle crime being around 3 to

4.5 percentage points lower in an area with a 50 percent higher initial low pay proportion.

These predicted changes would shift an area something around one-half to two-thirds of a

standard deviation down the change in property crime distribution (which has a standard

deviation of 6.8 percentage points in the period surrounding minimum wage

introduction), and a little less down the change in vehicle crime distribution (whose

standard deviation is slightly higher at 8.2 percentage points).

22 As in the simple cross-section (single difference) models when one uses the initial share of the low paid in the total wage bill or the low skill males initial low pay proportion instead of the head count initial low pay proportion measure the pattern of estimated effects is very similar. 23 Denote the change in log(crime rate) as ∆logC and the initial low pay proportion as L, and the estimated regression as ∆logC = βL + other variables, then the elasticity is (dC/dL).(L/C) = (dlogC/dL).L = βL.

25

7. Concluding Remarks

In this paper we consider possible links between crime and the labour market by

using the introduction of the UK National Minimum Wage to the UK as a means of

asking what happens to crime when the wages of low wage workers are given a sizable

boost. Our results support the idea that crime and the low wage labour market are

significantly related. In an empirical analysis across the police force areas of England

and Wales, we find that in the period surrounding minimum wage introduction changes in

crime were markedly lower in areas with more low paid workers before the imposition of

the minimum wage. A variety of different empirical approaches confirm this finding.

For example, the estimated effects are unaffected by the possible unemployment effects

of the minimum wage. Nor are we picking up a relationship that was operating in the

same way in earlier periods when a minimum wage was not introduced.

These findings provide an interesting counter-angle to other work in this area

which is increasingly tending to find that wage opportunities matter for crime. By

adopting a rather different methodology we also reach this conclusion. Furthermore, it

seems that the introduction of the minimum wage to the UK labour market went hand in

hand with reductions in crime in areas with low wage workers. Of course, unless the

minimum wage is increased by a sizable amount in future, then this is very much likely to

be a one-off change associated with the sizable wage increases that workers received in

the period when the minimum wage was introduced.

26

REFERENCES

Akers, Ronald L. 1977. Deviant Behaviour. Wadsworth:Belmont, California. Allan, Emilie Andersen and Darrell J. Steffensmeier. 1989. “Youth Underemployment

and Property Crime: Differential Effects of Job Availability and Job Quality on Juvenile and Young Adult Arrest Rates”. American Sociological Review 54: 107-123.

Becker, Gary. 1968. “Crime and punishment: An economic approach”. Journal of

Political Economy 76:175-209. Box, Steven. 1971. Deviance, Reality and Society. London: The Macmillan Press. Box, Steven. 1987. Recession, Crime and Punishment. London: The Macmillan Press.

Butcher, Kristen F. and Anne Morrison Piehl. 1998. “Cross City Evidence on the

Relationship between Immigration and Crime” Journal of Policy Analysis and Management, 17: 457-493.

Card, David. 1992 “Using Regional Variation in Wages to Measure the Effects of the

Federal Minimum Wage” Industrial and Labor Relations Review 46: 22-37. Card, David and Alan Krueger. 1995. Myth and Measurement: The New Economics of

the Minimum Wage. Princeton: Princeton University Press. Chamlin, Mitchell and John Cochran. 2000. “Unemployment, Economic Theory and

Property Crime: A Note on Measurement” Journal of Quantitative Criminology 16: 443-455.

Chiricos, Theodore. 1987. “Rates of Crime and Unemployment: An Analysis of

Aggregate Research Evidence.” Social Problems 34: 187-211. Cloward, Richard A. 1959. “Illegitimate Means, Anomie and Deviant Behaviour”.

American Sociological Review. 24: 164-176. Cloward, Richard A and Lloyd E. Ohlin. 1960. Delinquency and Opportunity. New

York: Free Press. Cohen, Albert K. 1955. Delinquent Boys. New York:Macmillan. Crutchfield, Robert. 1989. “Labour Stratification and Violent Crime” Social Forces 68:

489-512. Cullen, Julie Berry and Steven D Levitt. 1996. “Crime, Urban Flight and the

Consequences for Cities”, Working Paper No 5737 (NBER, Cambridge MA). Ehrlich, Isaac. 1973. “Participation in Illegitimate Activities: A Theoretical and

Empirical Investigation”. Journal of Political Economy 81: 521-563.

27

Farrington, David P, Bernard Gallagher, Lynda Morely, Raymond J. St. Ledger and Donald J. West. 1986. “Unemployment, School Leaving and Crime”. British Journal of Criminology 26: 335-356.

Farrington, David P. 1986. “Stepping Stones to Adult Criminal Careers” In D. Olweus, J

Block and MR Yarrow (eds), Development of Antisocial and Prosocial Behaviour. New York: Academic Press.

Fagan, Jeffrey. 1992. “Drug Selling and Illicit Income in Distressed Neighbourhoods: the

Economic Lives of Street-Level Drug Users and Dealers”, in George E Peterson and Adele V Harrell (eds) Drugs, Crime and Social Isolation: Barriers to Urban Opportunity. Urban Institute Press: Washington DC.

Fagan, Jeffrey and Richard Freeman. 1999. “Crime, Work and Unemployment”, in M.

Tonry (ed.) Crime and Justice: A Review of Research, Volume 25, University of Chicago Press: Chicago.

Freeman, Richard. 1999. “The Economics of Crime” in O. Ashenfelter and D. Card (eds)

Handbook of Labor Economics, Elsevier Science: Amsterdam. Freeman, Richard B, and William M Rodgers III. 1999. “Area Economic Conditions and

Labor Market Outcomes of Young Men in the 1990s Expansion”. Working Paper 7073 (NBER, Cambridge MA).

Fowles, Richard and Mary Merva. 1996. “Wage Inequality and Criminal Activity: An

Extreme Bounds Analysis For The United States, 1975-1990”. Criminology 34: 163-183.

Gould, Eric, Bruce Weinberg and David Mustard. 2002. “Crime Rates and Local Labor

Market Opportunities in the United States: 1979-1995. Review of Economics and Statistics, forthcoming.

Grogger, Jeffrey. 1998. “Market Wages and Youth Crime”. Journal of Labor Economics

16: 756-91. Grogger, Jeffrey. 2000. “An Economic Model of Recent Trends in Violent Crime”, in

Alfred Blumstein and Joel Wallman (eds.) The Crime Drop in America, Cambridge University Press.

Hale, Chris. 1998. “Crime and the Business Cycle in Post War Britain Revisited” British

Journal of Criminology, 38: 681-698. Hirshi, Travis. 1969. Cause of Delinquency. California University Press. Home Office. 1997. “Recorded Crime Statistics England and Wales, April 1998 to March

1999” Home Office Statistical Bulletin 18/99. Hseigh, Ching-Chi and MD Pugh. 1983. “Poverty, Income Inequality and Violent Crime:

A Meta-Analysis of Recent Aggregate Data Studies”. Criminal Justice Review 18: 182-202.

28

Land, Kenneth C, Patricia L McCall and Lawrence E. Cohen. 1990. “Structural

Covariates of Homicide Rates: Are There Any Invariances across Time and Social Space?” American Journal of Sociology 95:922-63.

Levitt, Steven D. 1996. “The Effect of Prison Population Size on Crime Rates: Evidence

from Prison Overcrowding Litigation” Quarterly Journal of Economics 111: 319-352.

------ 1998. “Juvenile Crime and Punishment” Journal of Political Economy, 106: 1156-

1185. Machin, Stephen and Costas Meghir. 1999. “Crime and economic incentives”. University

College London mimeo. May, Sean. 2000. “Wages and youth arrests”. Chapter 1 of MIT PhD thesis ‘Essays on

the Economics of Crime and Econometric Methodology’. McDonald, Ziggy. 2001. “Revisiting the Dark Figure A Microeconometric Analysis of

the Under-reporting of Property Crime and Its Implications”. British Journal of Criminology, 41: 127-149.

Merton, Robert K. 1957. Social Theory and Social Structure. New York: Free Press. Metcalf, David. 1999. “The Low Pay Commission And The National Minimum Wage”.

Economic Journal Features 109: F46-F66. Reilly Barry and Robert Witt. 1996.“Crime, Deterrence and Unemployment in England

and Wales: An Empirical Analysis”. Bulletin of Economic Research 48: 137-159. Sampson, Robert J and John H Laub. 1993. Crime in the Making: Path Ways and

Turning Points Through Life. Harvard University Press: Cambridge MA. Sutherland, Edwin H. 1924. Criminology. Philadelphia: Lippincott. Thornberry, Terence and RL Christenson. 1984. “Unemployment and Criminal

Involvement: An Investigation of Reciprocal Causal Structures”. American Sociological Review 56: 609-627.

West, Donald J and David P Farrington. 1977. The Delinquent Way of Life. London:

Heinemann. West, Donald J. 1982. Delinquency: Its Roots, Careers and Prospects. London:

Heinemann. Witt, Robert, Alan Clarke and Nigel Fielding. 1999. Crime and Economic Activity”,

British Journal of Criminology 39: 391-400. Witt, Robert and Ann Witte. 2000. “Crime, Prison and Female Labor Supply”, Journal of

Quantitative Criminology 16: 69-85.

29

Figure I: Changes in Property, Vehicle and Violent Crime And The Proportion Low Paid,

Between April 1998–September 1998 And April 1999-September 1999

Change log(PC) = .104 - 1.026 Prop Below Min One Year Before

Cha

nge

in L

og(P

rope

rty C

rime

Rat

e

Prop Below Min One Year Before.05 .1 .15 .2

-.2

-.1

0

.1

.2

SURREY

HERTFORD

LONDON

NORTHAMP

THAMES V

ESSEXSUSSEXGLOUCEST

WILTSHIR

WARWICKS

HAMPSHIR

BEDFORDS

NORTH YO

AVON AND

WEST YOR

KENT

CAMBRIDG

LEICESTE

DORSET

CHESHIRE

DERBYSHI

GREATER

WEST MID

MERSEYSISUFFOLKSOUTH YOGWENT SW

WEST MERNORTH WA

STAFFORD

LANCASHI

DURHAMCUMBRIA

NOTTINGH

NORTHUMB

LINCOLNSNORFOLK

HUMBERSI

CLEVELAN

DEVON AN

DYFED-PO

Change log(VC) = .124 - 1.168 Prop Below Min One Year Before

Cha

nge

in L

og(V

ehic

le C

rime

Rat

eProp Below Min One Year Before

.05 .1 .15 .2

-.2

-.1

0

.1

.2

SURREY

HERTFORD

LONDON

NORTHAMP

THAMES V

ESSEX

SUSSEX

GLOUCEST

WILTSHIR

WARWICKS

HAMPSHIR

BEDFORDS

NORTH YO

AVON AND

WEST YOR

KENT

CAMBRIDG

LEICESTE

DORSET

CHESHIRE

DERBYSHI

GREATER

WEST MID

MERSEYSISUFFOLK

SOUTH YO

GWENT SW

WEST MER

NORTH WA

STAFFORD

LANCASHI

DURHAM

CUMBRIA

NOTTINGH

NORTHUMB

LINCOLNS

NORFOLK

HUMBERSICLEVELAN

DEVON ANDYFED-PO

Change log(VioC) = .245 - .869 Prop Below Min One Year Before

Cha

nge

in L

og(V

iole

nt C

rime

Rat

e

Prop Below Min One Year Before.05 .1 .15 .2

-.2

-.1

0

.1

.2

.3

.4

.5

SURREY

HERTFORD

LONDON

NORTHAMP

THAMES V

ESSEX

SUSSEX

GLOUCEST

WILTSHIR

WARWICKS

HAMPSHIR

BEDFORDSNORTH YO

AVON AND

WEST YORKENT

CAMBRIDG

LEICESTEDORSET

CHESHIRE

DERBYSHIGREATER

WEST MID

MERSEYSI

SUFFOLK

SOUTH YO

GWENT SW

WEST MERNORTH WA

STAFFORD

LANCASHI

DURHAM

CUMBRIA

NOTTINGH

NORTHUMB

LINCOLNS

NORFOLK

HUMBERSI

CLEVELAN

DEVON AN

DYFED-PO

Notes: Population weighted regression line fit through data points

30

Table I: Mean Log Wages, Inequality and Crime Rates Broken Down

Across Areas By The Number of Low Wage Workers in the Year Before Minimum Wage Introduction

(1) (2) (3) (4) (5) April 1998 –

September 1998

October 1998 – March 1999

April 1999 – September

1999

Change Relative to Same Six

Month Period A Year Earlier

(3) – (1)

Change Relative to

Previous Six Month Period

(3) – (2)

Log(Hourly Earnings) Most Low Pay 1.79 1.81 1.85 .06 (.01)*** .04 (.01)*** Middle Low Pay 1.86 1.88 1.90 .04 (.01)*** .02 (.01)** Least Low Pay 2.04 2.08 2.10 .06 (.01)*** .03 (.02)** Most Low Pay – Least Low Pay

-.25 (.04)*** -.27 (.03)*** -.25 (.04)*** .00 (.01) .01 (.01)

10-50 Log(Hourly Wage) Differential Most Low Pay -.58 -.57 -.54 .03 (.01)*** .03 (.01)** Middle Low Pay -.60 -.60 -.58 .02 (.01)** .02 (.01)** Least Low Pay -.65 -.66 -.65 -.00 (.02) .01 (.01) Most Low Pay – Least Low Pay

.07 (.02)*** .09 (.02)*** .11 (.02)*** .04 (.01)*** .02 (.01)**

Property Crimes (Per 1000) Most Low Pay 31.08 30.19 29.74 -1.35 (1.27) -.46 (1.17) Middle Low Pay 32.22 32.41 31.92 -.30 (1.08) -.49 (1.15) Least Low Pay 29.53 29.47 31.14 1.61 (.93) 1.67 (.91)* Most Low Pay – Least Low Pay

1.55 (3.25) .72 (3.77) -1.40 (3.66) -2.96 (.88)*** -2.12 (.67)***

Vehicle Crimes (Per 1000) Most Low Pay 13.50 13.97 12.69 -.81 (.65) -1.28 (.77) Middle Low Pay 15.31 16.30 15.44 .13 (.74) -.86 (.91) Least Low Pay 13.49 14.08 14.33 .84 (.74) .25 (.87) Most Low Pay – Least Low Pay

.01 (1.24) -.11 (1.58) -1.64 (1.42) -1.65 (.41)*** -1.53 (.46)***

Violent Crimes (Per 1000) Most Low Pay 4.41 4.40 4.88 .47 (.36) .48 (.35) Middle Low Pay 4.60 4.72 5.59 .99 (.30)*** .87 (.29)*** Least Low Pay 5.47 5.47 6.68 1.22 (.51)** 1.21 (.51)** Most Low Pay – Least Low Pay

-1.06 (1.42) -1.07 (1.46) -1.80 (1.86) -.74 (.53) -.73 (.12)***

Notes: Areas are split into three equal sized groups of police force areas (14 in each, with Gwent and South Wales kept separate here as the boundary change discussed in footnote 7 of the main paper took place before the period considered in this Table). The groupings are based upon the proportion of workers paid less than the minimum wage in the period April 1998 to March 1999. Areas in the Most Low Pay group have over 11.7 percent of workers beneath the minimum wage (average = 14.2 percent). Areas in the Middle Low Pay group have between 9.7 and 11.7 percent of workers beneath the minimum (average = 10.9 percent). Areas in the Least Low Pay group have less than 9.7 percent of workers beneath the minimum wage (average = 7.1 percent). *** denotes statistically significant at 1% significance level or better, ** 5%, * 10%. Standard errors in parentheses.

31

Table II: Estimates of Area-Level Crime Equations

(Proportion Low Paid in Labour Force Survey)

Change in Log(Property Crime Rate) Time Period For Change:

(April 1999 – September 1999) – (April 1998 – September 1998)

Time Period For Change: (April 1999 – September 1999) – (October 1998 – March 1999)

Proportion Paid Less Than The Minimum Wage in Period 1

-1.026*** (.349)

-.839*** (.274)

-.928*** (.307)

-.787*** (.333)

-.459 (.387)

-.454 (.397)

Change in Log(Unemployment Rate) - - .051 (.090)

- - -.013 (.084)

Demographic Changes No Yes Yes No Yes Yes Change in Clear Up Rate No Yes Yes No Yes Yes R-squared .264 .414 .422 .188 .333 .333 Sample Size 41 41 41 41 41 41 Change in Log(Vehicle Crime Rate) Time Period For Change:

(April 1999 – September 1999) – (April 1998 – September 1998)

Time Period For Change: (April 1999 – September 1999) – (October 1998 – March 1999)

Proportion Paid Less Than The Minimum Wage in Period 1

-1.168*** (.348)

-.833*** (.283)

-.964*** (.316)

-1.381*** (.333)

-.966** (.416)

-.977*** (.422)

Change in Log(Unemployment Rate) - - .075 (.127)

- - .029 (.121)

Demographic Changes No Yes Yes No Yes Yes Change in Clear Up Rate No Yes Yes No Yes Yes R-squared .235 .376 .388 .312 .391 .393 Sample Size 41 41 41 41 41 41 Change in Log(Violent Crime Rate) Time Period For Change:

(April 1999 – September 1999) – (April 1998 – September 1998)

Time Period For Change: (April 1999 – September 1999) – (October 1998 – March 1999)

Proportion Paid Less Than The Minimum Wage in Period 1

-.869* (.449)

-.717 (.597)

-1.312* (.655)

-1.257*** (.239)

-1.178*** (.432)

-1.144** (.466)

Change in Log(Unemployment Rate) - - .343 (.206)

- - -.083 (.102)

Demographic Changes No Yes Yes No Yes Yes Change in Clear Up Rate No Yes Yes No Yes Yes R-squared .047 .149 .239 .215 .358 .369 Sample Size 41 41 41 41 41 41

Notes: Coefficients (heteroskedastic consistent standard errors) reported; The demographic controls entered were – change in average age, change in the population share of young (<25) men, change in proportion black, change in population share with no educational qualifications, change in proportion female, change in share of public sector jobs. *** denotes statistically significant at 1% significance level or better, ** 5%, * 10%.

32

Table III: Estimates of Area-Level Crime Equations (Wage Gap And Proportion Low Paid Males in Low Skill Occupations in Labour

Force Survey)

Change in Log(Property Crime Rate) Time Period For Change:

(April 1999 – September 1999) – (April 1998 – September 1998)

Share of Wage Bill Measure Low Skill Males Measure Period 1 Less Than The Minimum Wage Measure

-9.302*** (3.346)

-7.712*** (2.526)

-8.425*** (2.728)

-1.194*** (.430)

-.948*** (.355)

-1.053*** (.390)

Change in Log(Unemployment Rate)

- - .045 (.088)

- - .049 (.091)

Demographic Changes No Yes Yes No Yes Yes Change in Clear Up Rate No Yes Yes No Yes Yes R-squared .243 .406 .412 .254 .399 .406 Sample Size 41 41 41 41 41 41 Change in Log(Vehicle Crime Rate) Time Period For Change:

(April 1999 – September 1999) – (April 1998 – September 1998)

Share of Wage Bill Measure Low Skill Males Measure Period 1 Less Than The Minimum Wage Measure

-11.062*** (3.280)

-8.294*** (2.534)

-9.503*** (2.832)

-1.364*** (.434)

-.946*** (.374)

-1.103** (.382)

Change in Log(Unemployment Rate)

- - .076 (.125)

- - .074 (.126)

Demographic Changes No Yes Yes No Yes Yes Change in Clear Up Rate No Yes Yes No Yes Yes R-squared .235 .387 .399 .228 .367 .378 Sample Size 41 41 41 41 41 41 Change in Log(Violent Crime Rate) Time Period For Change:

(April 1999 – September 1999) – (April 1998 – September 1998)

Share of Wage Bill Measure Low Skill Males Measure Period 1 Less Than The Minimum Wage Measure

-8.732** (4.278)

-6.706 (5.888)

-12.021* (6.585)

-.989* (.581)

-.818 (.743)

-1.552** (.758)

Change in Log(Unemployment Rate)

- - .335 (.206)

- - .345 (.204)

Demographic Changes No Yes Yes No Yes Yes Change in Clear Up Rate No Yes Yes No Yes Yes R-squared .053 .148 .235 .043 .147 .237 Sample Size 41 41 41 41 41 41 Notes: Coefficients (heteroskedastic consistent standard errors) reported; The demographic controls entered were – change in average age, change in the population share of young (<25) men, change in proportion black, change in population share with no educational qualifications, change in proportion female, change in share of public sector jobs. *** denotes statistically significant at 1% significance level or better, ** 5%, * 10%.

33

Table IV: Difference-in-Difference Estimates of Area-Level Crime Equations

Change in Crime Models Estimated With Data Pooled Over Six Month Time Periods Period Surrounding Minimum Wage Introduction: [(April 1999 – September 1999) - (April 1998 – September 1998)] Earlier Time Periods: [ (March 1998 – October 1997) – (March 1997 – October 1996 )], [ (April 1997 – September 1997) – (April 1996 – September 1996)], [ (March 1997 – October 1996) – (March 1996 – October 1995 )] , [(April 1996 – September 1996) – (April 1995 – September 1995)] Change in Log(Property Crime Rate) Proportion Paid Less Than The Minimum Wage in Period 1 X [(April 1999 – September 1999) – (April 1998 – September 1998)]

-.750* (.391)

-.635** (.296)

-.685** (.291)

-.651** (.279)

Change in Log(Unemployment Rate) - - .038 (.028)

.033 (.028)

Demographic Changes No Yes Yes Yes Change in Clear Up Rate No Yes Yes Yes Area-Specific Trends No No No Yes R-squared .369 .452 .457 .609 Sample Size 205 205 205 205 Change in Log(Vehicle Crime Rate) Proportion Paid Less Than The Minimum Wage in Period 1 X [(April 1999 – September 1999) – (April 1998 – September 1998)]

-.875** (.415)

-.566* (.313)

-.604** (.307)

-.592 (.373)

Change in Log(Unemployment Rate) - - .029 (.040)

.026 (.042)

Demographic Changes No Yes Yes Yes Change in Clear Up Rate No Yes Yes Yes Area-Specific Trends No No No Yes R-squared .392 .509 .511 .626 Sample Size 205 205 205 205 Change in Log(Violent Crime Rate) Proportion Paid Less Than The Minimum Wage in Period 1 X [(April 1999 – September 1999) – (April 1998 – September 1998)]

-.810 (.684)

-.842 (.694)

-.953 (.683)

-.907 (.833)

Change in Log(Unemployment Rate) - - .086 (.083)

.093 (.091)

Demographic Changes No Yes Yes Yes Change in Clear Up Rate No Yes Yes Yes Area-Specific Trends No No No Yes R-squared .067 .165 .168 .267 Sample Size 205 205 205 205

Notes: Coefficients (heteroskedastic consistent standard errors) reported; The demographic controls entered were – change in average age, change in the population share of young (<25) men, change in proportion black, change in population share with no educational qualifications, change in proportion female, change in share of public sector job. All equations include dummy variables for time period and the proportion low paid variable. *** denotes statistically significant at 1% significance level or better, ** 5%, * 10%.