Free Essay

Regression Analysis

In: Business and Management

Submitted By kanikam14
Words 6399
Pages 26
Unemployment and crime: New evidence for an old question
Kerry L. Papps
Victoria University of Wellington

Rainer Winkelmann*
IZA and Centre for Economic Policy Research, London December 1999

This paper uses panel data techniques to examine the relationship between unemployment and a range of categories of crime in New Zealand. The data cover sixteen regions over the period 1984 to 1996. Random and fixed effects models are estimated to investigate the possibility of a causal relationship between unemployment and crime. Hypothesis tests show that two-way fixed effects models should be used. The main result of the paper is that there is some evidence of significant effects of unemployment on crime, both for total crime and for some subcategories of crime.

We are grateful to Rachel Bambery, New Zealand Police National Headquarters, for her assistance in obtaining crime and population statistics. The staff of the University of Canterbury Library also gave invaluable help in unraveling the complexities of New Zealand unemployment and income data. The paper has benefited from useful comments by two anonymous referees, Simon Kemp, Jacques Poot and participants of the CEPR conference on “Metropolitan Economic Performance”, Lisbon, October 1998. *Corresponding author: IZA, P.O. Box 7240, 53072 Bonn, Germany;

“I know only of three ways of living in society: one must be a beggar, a thief, or a wage earner.” HONORÉ de MIRABEAU (1749-1791)

1. Introduction
It is a common observation for many countries that unemployment rates and crime rates are positively associated. A more contentious issue is whether this association means that unemployment causes crime, crime causes unemployment or third factors cause both. Only the first of the three possibilities would imply that the effects of unemployment on crime deserve to be counted among the “non-pecuniary” costs of unemployment that should be taken into account in a cost-benefit analysis of potential unemployment-reducing policies. The theoretical underpinning of the causality notion was developed some thirty years ago by Becker (1968), Stigler (1970) and Ehrlich (1973), among others. In Ehrlich’s model, individuals divide their time between legal activities and risky illegal activities. If legal income opportunities become scarce relative to potential gains from crime, the model predicts that crime will become more frequent. Increased unemployment could be one such factor. Numerous subsequent empirical papers have attempted to test the predictions of the BeckerEhrlich model and to find out whether the magnitude of the unemployment effect is quantitatively important. The hallmark of this literature is its failure to reach consensus as to whether higher levels of unemployment lead to a greater incidence of crime. In a survey of the literature, Box (1987) reports 35 reliable studies on the topic, 20 of which find a positive relationship between unemployment and crime, with the remainder unable to find any such relationship. In the context of New Zealand, a previous econometric study by Small and Lewis (1996), based on time-series techniques and Granger causality tests, lends “strong support to the idea that crime and unemployment are linked in some way” and that unemployment causes crime more often than vice versa. The objective of this paper is to revisit the issue of whether unemployment has a causal effect on various categories of economic and anti-social crime. For this purpose, we analyse New Zealand regional panel data, regressing crime rates on unemployment rates using fixed and random effects models. Our approach solves several of the problems that have been characteristic of previous empirical papers. In particular, we cannot reject the hypothesis that unobservable period specific effects are correlated with the unemployment rate. This finding suggests that time series regressions will likely be affected by omitted variable bias. Still, we find some evidence that unemployment matters even after time, region, deterrence, and income effects are accounted for.


The paper is structured as follows. Section 2 commences with an eclectic review of some previous empirical studies. Section 3 gives a discussion of some general data issues as well as a description of the data that were actually used in this study. The results from the various models are presented in Section 4. Section 5 concludes with commentary on the implications of these findings and possible improvements that could be made to the analysis.

2. Previous Studies
As mentioned in the introduction, a consensus as to whether higher levels of unemployment lead to a greater incidence of crime has not yet been reached. Differences in the results may be related to a variety of factors: differences in the type of data used and differences in the definition of crime being two of them. The empirical literature on the topic of crime and unemployment typically is based on one of four types of data: aggregate (national) time series data, aggregate cross-section data, regional panel data or individual level data (cross-sections or panel). Studies of the first two types often affirm the existence of a causal relationship. Ehrlich (1975), Leveson (1976), Chapman (1976) and Brenner (1978) are some early examples. While these studies to varying degrees attempt to control for other factors, they still are likely to be affected by omitted variable bias. The availability of regional panel data can ameliorate this problem and, indeed, when such data are used, the evidence is much less supportive of a causal relationship. For example, Entorf and Spengler (1998) conducted a regional panel study for Germany and found unemployment to have “small, often insignificant and ambiguous signs”. With individual level data, one observes the labour market status of a particular offender at the time of committing a crime. Studies include Myers (1983), Schmidt and Witte (1984), Trumball (1989), Tauchen et al. (1994), and Grogger (1991). There are several advantages of such data: the number of observations is large, these datasets usually provide a large number of controls and it becomes possible to focus on particular sub-populations, such as the socially lessadvantaged, where an effect might be more likely to occur. With individual level panel data, one can solve the most common omitted variable problems by comparing the crime propensities of the same individuals in different employment states. Unfortunately, such individual level panel data are rarely available and, in their absence, the use of regional panel data is arguably the “second-best” methodological option. An independent issue is the definition of crime, with the distinction between economic and anti-social crime. Economic crimes are those where the motivation is pecuniary gain, while anti-


social crimes are committed for some other reason. Most economic studies have focused on the link between unemployment and economic crimes since such a relation is supported by economic theory, which predicts that potential offenders compare the costs and benefits associated with crime. Nevertheless, the notion that increases in unemployment lead to increases in anti-social crime has been proposed by sociologists and others. As a consequence, the total costs of unemployment may be higher than some studies predict. Finally, the issue of whether reported crime accurately reflects the actual number of committed crimes arises. Using reported crime might be a misleading indicator of the total amount of crime in society as not all crimes committed are reported to the police. Hence, this measure is dependent on the public’s proclivity to report crimes to the police. This would be of minor empirical importance unless the likelihood that crimes are reported has changed significantly over time. There is some evidence, however, that this was the case. For instance, surveys conducted by Market Research Limited (M.R.L.) in 1993 and 1995 found that the proportion of victims who had reported the most recent crime to the police increased from 67% to 77% between these years. The use of regional panel data will give estimates that are unaffected by this measurement issue as long as the reporting propensity changes in different regions at the same rate over time.1

3. Data
Annual data on the level of crime were obtained from the New Zealand Police for the period 1984-1996 for 16 police districts. This included the number of offences reported to police in each police district for seven offence groups and the total number of crimes, which they collectively comprise. Numbers are transformed into crime rates by division with the regional population size (in thousands). The reported crime rate for each category is denoted by o1 to o7, respectively, while the overall crime rate is represented by o (for “offence”). The groups used by the New Zealand Police are: violent offences; drug and anti-social offences; dishonesty offences; property damage offences; property abuse offences; sexual offences; administrative offences. One should thus notice that the overall crime rate aggregates very different types of crimes. Even within each category, there is a substantial amount of heterogeneity (see Appendix). While it would be desirable to consider more homogeneous groups, such data were not available to us. The heterogeneity means that any unemployment effect needs to be interpreted as an average effect that may differ from the effect of

This is an assumption we have to make and that we cannot verify given the available data.


unemployment on any component crime. The heterogeneity is moreover likely to drive up the residual variation, and thereby the standard errors, of the estimates. There are a variety of measures of unemployment in New Zealand. The official measure is derived from the Household Labour Force Survey (HLFS), a quarterly survey by Statistics New Zealand. The HLFS provides estimates which are internationally comparable and are not subject to changes in the definition of “being unemployed”. Unfortunately, the series only includes subnational estimates since 1990. The quinquennial Census of Population and Dwellings provides the most complete survey of unemployment in New Zealand. However, drawbacks are the infrequent observations provided and the fact that different definitions of unemployment have been used over time. Because of these shortcomings, the measure of unemployment selected for this study was the number of people registered as unemployed with the Department of Labour, hereinafter denoted by UN.2 Annual averages of this series were obtained for each of 21 employment districts for the same period from Statistics New Zealand’s INFOS and were then matched to the 16 police districts.3 Unemployment rates were obtained by division with a population estimate and were denoted by un.4 We acknowledge that this is a less than perfect measure of legal employment opportunities. Part of unemployment is related to job search, the length of which depends on many factors apart from job availability, including benefit levels and eligibility. Ideally, we would have liked to include a measure of long-term unemployment but such a statistic was not available for the full time period and the required regions. As will be explained later, possible determinants of the crime level other than unemployment are also investigated in the study. Firstly, the clearance rate for each offence group was obtained from the New Zealand Police. This is given by the ratio of the number of crimes cleared by police to the total number of crimes reported for each region and crime sub-category. The overall clearance rate is denoted p while the clearance rates for each offence group are denoted p1-p7, where the index matches the crime sub-category number. Secondly, information on the average level of income for each region was obtained. Since there is no annual sub-national data for income in New Zealand, information from the 1986, 1991
This is also the series Small and Lewis (1996) used in their study on the subject. Registered unemployed are those who have chosen, as one of their methods of job search, to enrol with the New Zealand Employment Service and who are available for employment.
3 4 2

Details on the matching protocol are available from the authors.

An estimate of the population of each region was used since the only adequate data were derived from the Census and, hence, involved interpolation. Since the labour force participation rate fluctuates over the business cycle, it was not appropriate to calculate a labour force estimate in this manner.


Figure 1 New Zealand Unemployment Rate and Crimes per 1000 Residents un o

7 Unemployment rate (%) 6 5 4 3 2 1 0 1978

150 140 130 120 110 100 90 80 1996 Total crimes per 1000 residents




1986 Year





Source: New Zealand Police (o), Statistics New Zealand (un)

and 1996 Censuses on mean personal income for each police district was used. To obtain a complete panel, the income of each district relative to the national average was calculated, using a linear time trend to extrapolate the missing observations. An annual index of real GDP per capita for all New Zealand was derived from the New Zealand National System of Accounts for 19841996, based on 1984 = 1000. The income series used in this study, denoted y, was the product of this series and the estimate of the relative income of each region for the appropriate year.

4. Results and Analysis
Figure 1 plots the national unemployment rate and the total crime rate between 1978 and 1996.5 There is visual evidence that the two series move closely together over time, and the correlation coefficient, r, can be calculated as 0.41. However, as mentioned in the introduction, this is by no means an indication of causality from unemployment to crime. While the hypothesis of reverse causation (from crime to unemployment) seems implausible a priori (although there may be such a feedback process in individual cases), third variables are likely to affect both crime and unemployment.

Information for the years 1978 to 1983 is available at the aggregate level. However, it could not be used in the econometric analysis, as changes in the definition of police districts made a comparable disaggregation impossible.



In order to evaluate and control for the potential influence of third factors, we employ regression analysis. Our general strategy is to start with a very simple model and successively generalize. Following Entorf and Spengler (1998), we use a log-log specification of the unemployment-crime relation in all cases. This gives rise to an estimated coefficient that has the interpretation of an elasticity. A log-log model is also consistent with Ehrlich (1973), who suggested a multiplicative form for the supply-of-offences function when variables are in levels. One extreme possibility is that unemployment is the sole determinant of the crime rate and that the parameters of the model are identical, regardless of what region or year an observation is drawn from. This implies that a pooled regression can be applied to the data as follows: ln oit = α + β ln unit + ε it .


Here the subscript i indicates the region of the observation and t the year of the observation. εit denotes the residual associated with observation it. The error term, εit, is assumed to have mean zero. Moreover,

σ 2 for i = j and t = s E (ε it ε js ) =  i .  0 else
Therefore, the model allows for region-specific heteroskedasticity that is accounted for in the computation of the variance-covariance matrix of the OLS estimator for β. ˆ The first column of Table 1 reports the parameter estimates, β , obtained from pooled regressions of the logarithmic crime rates on the logarithmic unemployment rates. These are based on a total of eight separate regressions (one for each type of crime, plus the total). The first ˆ row of Table 1 gives the value for the total crime rate. β is positive and statistically different from zero. The value of 0.144 indicates that a 10% increase in the unemployment rate is associated with an increase in the crime rate of about 1.4%.


Table 1 OLS Regression Results: Estimated Unemployment Elasticities (1) Total offences Violent offences Drug and anti-social offences Dishonesty offences Property damage offences Property abuse offences Sexual offences Administrative offences Time effects (13) Region effects (16) 0.144* (0.023) 0.364* (0.042) 0.264* (0.041) 0.058 (0.036) 0.096* (0.030) 0.212* (0.039) 0.410* (0.042) 0.805* (0.064) no no (2) 0.038 (0.048) 0.065 (0.068) -0.126 (0.080) 0.350* (0.068) -0.044 (0.060) -0.025 (0.090) 0.185* (0.091) 0.393* (0.121) yes no (3) 0.163* (0.014) 0.414* (0.035) 0.331* (0.033) 0.003 (0.026) 0.122* (0.015) 0.254* (0.028) 0.453* (0.029) 0.881* (0.058) no yes (4) 0.080 (0.048) 0.109 (0.064) -0.043 (0.118) 0.188* (0.093) 0.031 (0.047) 0.064 (0.083) 0.300* (0.107) 0.569* (0.148) yes yes

Notes: The number of observations is 208. Reported coefficients comprise β in Model (1). Estimated standard errors are in parentheses (covariance matrix estimator with regional heteroskedasticity). An asterisk indicates statistical significance at the 10% significance level.

Browsing through the remaining coefficients of the first column reveals that for most (five out of seven) subclasses of crime, the estimated elasticity is substantially larger than in total. In the case of administrative offences, the elasticity reaches almost unity. The elasticity is significantly different from zero in six out of seven subclasses of crime. Thus, there appears to be evidence for a substantial effect of unemployment on crime. Of course, this conclusion is conditional on the validity of the model. We next investigate how appropriate such a regression is in the first place, given the data used in this study. Figures 2a and 2b provide a first indication. Figure 2a is a scatter plot of the average crime rate over the


Figure 2a Scatter Plot of Time-Averaged Crime and Unemployment Rates


Mean total crime rate, o

Rotorua Gisborne

140 Auckland


Palmerston Nth Hamilton Christchurch Dunedin Timaru Invercargill


Greymouth Whangarei

New Plymouth




5 6 Mean unemployment rate, un


sample period for each region, oi, against the time-averaged unemployment rate, uni. Figure 2b is a scatter plot of annual observations of the national value of the crime rate, ot, against annual observations of the national value of the unemployment rate, unt.6 There is evidence that the model is incorrectly specified. The graphs reveal no strong relationship between oi and uni (if anything, this relation is negative) but indicate that ot and unt are positively correlated. Either there are region-specific factors present which inhibit the ability to report a positive relation in Figure 2a or there are time-specific factors which create the appearance of a link between unemployment and crime over time or a mixture of both situations exists. Potential misspecifications of this kind can be addressed by the inclusion of region- or timespecific fixed effects (or both) in the regression. Formally, let ln oit = α + µ i + λ t + β ln unit + ε it ,



National observations can be considered weighted averages of the regional observations, thus giving convenient estimates of o and un that have a time dimension only.


Figure 2b Scatter Plot of Region-Averaged Crime and Unemployment Rates
1992 1993 1991 1994 1990

140 National total crime rate, o

1996 1995

1985 1986 1987



1982 1981 1980

1984 1983



80 0 1 2 3 4 5 National unemployment rate, un 6 7

where µi is a region-specific fixed effect and λt is a time-specific effect, while εit is a white noise error term as before. This is an identical specification to the pooling case, only that now µi and λt are considered to be parameters to be estimated, whereas before they were part of the error term. A possible factor that might feature in µi is the degree of urbanisation. Figure 2a shows that major metropolitan centres like Auckland, Wellington and Christchurch tend to have low values of uni and high values of oi. It is quite possible that more densely populated regions offer more employment opportunities but also support higher levels of criminal activity. The age and ethnic structure of the various regions may also impact on µi. The period specific effects λt capture, for instance, any change in macroeconomic conditions, such as inflation or oil-price shocks that might be expected to lead to higher levels of crime (assuming that they affect all regions equally). In addition, these effects account for changes in the propensity to report crimes over time as long as these are uniform across regions. The estimation results are given in Columns (2)-(4) of Table 1. Column (2) includes time effects but no region effects. Column (3) includes region effects but no time effects. Column (4) includes both time and region effects. As we would expect from Figures 2a and 2b, the estimated


Table 2 Testing the Model F-tests Time Region effects effects Total offences Violent offences Drug and anti-social offences Dishonesty offences Property damage offences Property abuse offences Sexual offences Administrative offences Model includes time fixed effects Model includes region fixed effects Distribution under H0 Critical value (α=0.05) 4.87 40.85 4.38 3.98 8.51 20.33 6.39 10.92 n.a. yes 31.84 22.99 7.74 18.10 69.89 31.29 19.08 6.06 yes n.a. Hausman tests Time effects 4.82 6.57 20.99 13.57 4.60 5.87 5.66 6.40 n.a. no χ2(1) 3.84 1.97 4.42 6.91 2.66 2.05 2.58 1.38 1.60 n.a. yes χ2(1) 3.84 Region effects 0.40 2.93 14.63 4.53 0.26 1.28 0.97 4.40 no n.a. χ2(1) 3.84 0.08 0.05 0.17 0.53 0.11 0.16 0.21 0.43 yes n.a. χ2(1) 3.84

F(12,179) F(15,179) 1.83 1.67

unemployment elasticities are more affected by the inclusion of period fixed effects than by the inclusion of region fixed effects. In all but one case, the magnitude of the unemployment elasticity drops when period effects are included and increases when region effects are included. In the most general model, with both region and time effects, an overall drop in the elasticity is observed, since the time effect dominates the region effect. The elasticity for the total crime rate is about halved in magnitude, relative to the pooled model without fixed effects. In terms of statistical significance, only three of the eight estimated elasticities remain significant (dishonesty, sexual, and administrative offences) once time and region fixed effects are included in the model. The significance of the two types of fixed effects can be tested by means of various F-tests. Table 2 reports the relevant test statistics. In the first column, the two-way fixed effects model is compared to a model with region effects only, i.e., the null hypothesis is that λt = 0 for all t. A comparison of the F-statistic with the critical value of 1.83 shows that the time effects are jointly


significant. A similar conclusion may be drawn from the second column of Table 2, with regard to the region effects. From this point of view, the two-way fixed effects model is the superior model. However, as far as estimation of the unemployment elasticity of crime is concerned, efficiency gains may be realised by excluding some of the fixed effects and modeling them as random effects instead. After all, 27 degrees of freedom are lost in the two-way fixed effects model relative to the simple model. As a consequence, standard errors increase, in the case of the total crime rate by more than 100%. Those degrees of freedom could be preserved if the time and region effects were modeled as random effects. In order to do so, one needs to assume that the unemployment rate is uncorrelated with the time and region effects, respectively. If this assumption is valid, the fixed effects estimator is inefficient and a random effects estimator should be used. Otherwise, the random effects estimator for the unemployment elasticity is biased and inconsistent. In essence, the decision is whether to make inferences conditional on the effects observed in the sample or unconditional (marginal) inferences with respect to the population characteristics. The validity of the assumption of no correlation can be tested using Hausman’s (1978) test. This test involves comparing the estimated parameter values for the unemployment elasticity under both random and fixed effects specifications. Under the null hypothesis that unemployment is uncorrelated with the random effects, the coefficients estimated by either model are consistent but only the random effects estimator is efficient. However, if the null hypothesis is false, the use of random effects produces an inconsistent estimator whereas the fixed effects estimator remains consistent. Columns 3-6 of Table 2 report the results of these tests. The first hypothesis is that the time effects are uncorrelated with the unemployment rate. This test can be based either on a model that does not control for region fixed effects (Column 3) or on a model that does include such effects (Column 4), both under the null hypothesis and under the alternative. In the former case, the hypothesis of no correlation, and thus the random effects specification, is always rejected. If region fixed effects are included as well, the evidence is less clear-cut. This may reflect the lower power of the test, stemming from the reduced degrees of freedom of the model. In two cases (violent and drug and antisocial offences), the Hausman test remains significant. The Hausman test results for the region effects are weaker. Based on the model without time effects (Column 5) correlation between unemployment rate and the error component is detected in three out of seven cases. If time fixed effects are included as well, the null hypothesis is always accepted.


Based on these various tests, one can argue whether or not the regional effects should be modeled as random or as fixed. While the regional effects are jointly highly significant (Column 2 of Table 2), the evidence for a correlation with unemployment is not strong. Nevertheless, the two-way fixed effects model seems to be justifiable in order to hedge the results against this type of omitted variable bias for all subclasses of crime. An augmented model Of course, even the two-way fixed effects model can suffer from further omitted variable bias with respect to the estimated unemployment elasticity if correlated variables are omitted that are neither constant over time nor constant over regions. It is not difficult to come up with examples. For instance, the proportion of non-married young men is likely to vary both over time and over regions. Elsewhere, this proportion has been shown to be correlated both with unemployment and with criminal activity (e.g., Akerlof, 1998). While we suspect that the variation of such a variable over the 13 year observation period is small relative to its variation over regions, it is clear that subsuming this effect under the region-fixed effect is only an approximation that introduces error and, potentially, bias. To avoid bias, such variables should then be included as regressors in the model along with the unemployment rate in a multiple regression framework. Unfortunately, lack of data availability imposes serious limitations on our ability to model any additional effects. A rich set of regionally disaggregated data is simply not available for New Zealand. Hence, we restrict our attention to two additional effects, deterrence and income. Both variables are part of the traditional Becker-Ehrlich model of rational crime. It is generally hypothesised that the deterrence rate has a negative effect on the crime rate. As the likelihood of potential criminals being caught increases, the expected penalty resulting from crime increases.7 This decreases the probability that a rational agent will choose to commit an offence. The effect of the regional income level is less clear cut. Increases in income can be thought of as reflecting an increase in the benefits derived from legal activities, thus a negative relation may be posited. However, as income increases, the potential gains from economic crimes also increase. This would increase the attractiveness of crime relative to legal work and a positive relation between o and y may be observed. Which effect will dominate is unclear. Entorf and
The expected penalty is given by the probability of capture and conviction times the severity of the sentence imposed. Buchanan and Hartley (1996) note that in New Zealand although the severity of prison sentences remained roughly constant for part of the period 1983-1992 the “penalty probability” fell due to decreases in the conviction rate over this period. As long as this fall is uniform across regions, however, it will not affect the results, but, rather, be subsumed into the time fixed effects.


Spengler (1998) believed that the unemployment rate could be interpreted as a measure of legal income opportunities, while the absolute level of income represented illegal income opportunities.8 As described in Section 3, a suitable proxy variables for deterrence is available, namely the clearance rate, p, as is an estimate of the regional income level in year t, y. Table 3 reports the results of two-way fixed effects estimation of the following model: ln oj it = α + µ i + λ t + β ln unit + γ ln pj it + δ ln y it + ε it .


Model (3) offers a clear improvement over Model (1), as the F-test for joint significance of deterrence and income rejects the restricted model in each case. The results show that the clearance rate has a significant effect on the crime rate for each of the offence groups, except for administrative offences. However, in half of the cases a positive relation is found, contrary to the prior hypothesis. This is puzzling, as the clearance rate is given by the ratio of the number of crimes cleared to the total number of crimes reported and, hence, potentially biased downwards. For instance, an overestimate of the number of crimes due to measurement error leads to an underestimate of the clearance rate, thereby introducing a negative correlation between the error and the clearance rate. Another reason for endogeneity could be that a high crime rate may lead to a higher clearance rate as the police act more aggressively to counter the increase in crime. In an attempt to overcome either type of endogeneity, we also estimate a model in which the current clearance rate is instrumented by its first lag. The last column of Table 3 shows the results of a Hausman test, where the model reported in Table 3 is tested against the instrumented model.9 The test does not reject the null hypothesis, with one exception, and we take this as a confirmation of Model (3) with current clearance rate included, although we admit that this conclusion might be sensitive to the limitations of the instrument (see, for example, Levitt (1997)). Apart from the mixed effect of deterrence, the estimates appear quite plausible. The level of income has a negative effect for every offence group and the overall crime rate. This suggests that the effect of an increase in legal income opportunities outweighs the effect of a corresponding increase in illegal income opportunities. The introduction of p and y has little impact on the

They also included a measure of relative income (or income inequality), which was assumed to have a positive effect on crime.

The full set of regression results is available from the authors upon request.


Table 3 Extended Model (two way fixed effects model) un Total offences Violent offences Drug and anti-social offences Dishonesty offences Property damage offences Property abuse offences Sexual offences Administrative offences Distribution under H0 Critical value (α=0.05) 0.088* (0.047) 0.075 (0.059) -0.037 (0.104) 0.156* (0.091) 0.037 (0.044) 0.093 (0.072) 0.326* (0.104) 0.576* (0.144) p -0.019 (0.100) 0.584* (0.244) 0.286* (0.109) 0.701* (0.307) -0.162* (0.073) -0.404* (0.084) -0.586* (0.171) 0.222 (0.152) y -1.048* (0.380) -2.002* (0.584) -4.409* (0.849) -0.581 (0.804) -1.251* (0.374) -0.572 (0.578) -1.011 (0.811) -5.182* (1.370) F-test 8.14 22.01 30.79 5.57 14.15 24.09 12.12 18.06 F(2,177) 3.00 Hausman test 0.186 0.327 0.259 0.101 6.748 0.183 3.873 0.549 χ2(1) 3.84

Notes: The number of observations is 208. Estimated standard errors are in parentheses (covariance matrix estimator with regional heteroskedasticity). An asterisk indicates statistical significance at the 10% significance level. The null hypothesis is that deterrence and income are jointly insignificant. The null hypothesis of the Hausman test is that the clearance rate is uncorrelated with the regression error.

estimated effect of the unemployment rate on crime. Most importantly, though, the significance ˆ of β has increased for the total crime rate, leading to a rejection of the null hypothesis of no effect at the 10% level. Thus, there is some evidence that increased legal income opportunities, be they through lower unemployment levels or higher income, lead to a substitution away from criminal activities.


5. Conclusions
So, having muddied further the already turbid waters of research into the unemploymentcrime relationship, what has this study contributed? Evidence has been found that tends to confirm earlier conclusions reached by Small and Lewis for New Zealand. Results indicate that the total rate of crime remains significantly affected by the unemployment rate, once complicating factors are controlled for. In particular, unemployment was found to have a significant relationship to the number of dishonesty crimes committed. This is the category that includes the economic crimes of theft, fraud, car conversion, receiving and burglary that much of the previous literature, including Small and Lewis (1996), has focussed on. Furthermore, this study indicates several possibilities for further research. In particular, the introduction of additional regressors that may explain crime, for example income inequality, may alleviate any remaining omitted variable bias. The unemployment-crime relationship is an old issue. No consensus has been reached by economists during the past three decades, nor does one seem likely to emerge in the near future. Perhaps an observation by McDowell and Webb (1995) on this area of research has particular relevance: that it is an “urge to achieve a certainty which simply does not exist”. Or perhaps with superior data and superior techniques, a conclusive model of the crime decision may be found and, in the process, bring credibility to application of economic principles to social issues.


Appendix: Definitions of offence groups
Violent offences (O1) Homicide Grievous assaults Minor assaults Group assemblies Robbery Kidnapping/abduction Serious assaults Intimidation/threats Drug and anti-social offences (O2) Drugs (not cannabis) Drugs (cannabis only) Gaming Liquor offences Disorder Vagrancy offences Family offences Dishonesty offences (O3) Theft Fraud Car conversion Receiving Burglary Property damage offences (O4) Destruction of property Endangering Property abuse offences (O5) Firearms offences Littering Post, rail and fire abuses Animals Trespass Sexual offences (O6) Sexual attacks Abnormal sex Sexual affronts Immoral behaviour Immoral behaviour/miscellaneous Indecent videos Administrative offences (O7) Against justice Against national interest Births, deaths and marriages Immigration Race relations By-law breaches


Akerlof, G.A. (1998) Men without children, Economic Journal, 108, 287-309. Baltagi, B.H. (1995). Econometric analysis of panel data. Chichester: Wiley. Becker, G.S. (1968). Crime and punishment: An economic approach. Journal of Political Economy, 76, 169-217. Box, S. (1987). Recession, crime and punishment. London: Macmillan. Brenner, H. (1978). Review of Fox’s “Forecasting Crime”. Journal of Criminal Law and Criminology, 70, 273-274. Buchanan, C. and Hartley, P.R. (1996). Controlling crime in New Zealand. Wellington: New Zealand Business Round Table. Chapman, J.I. (1976). An economic model of crime and police. Journal of Research Crime and Delinquency, 13, 48-63. Ehrlich, I. (1973). Participation in illegitimate activities: A theoretical and empirical investigation. Journal of Political Economy, 81, 521-565. Ehrlich, I. (1975). Deterrent effect of capital punishment. American Economic Review, 65, 397417. Entorf, H. and Spengler, H. (1998). Socio-economic and demographic factors of crime in Germany: Evidence from panel data of the German states. Mannheim: Centre for European Economic Research (Discussion Paper #98-16). Grogger, J. (1995). The effect of arrest on the employment and earnings of young men. Quarterly Journal of Economics, 110, 51-72. Hausman, J.A. (1978). Specification tests in econometrics. Econometrica, 46, 1251-1271. Hsiao, C. (1986). Analysis of panel data. Cambridge: Cambridge University Press. Leveson, I. (1976). The growth of crime. Croton on Hudson: Hudson Institute. Levitt, S.D. (1997). Using Electoral Cycles in Police Hiring to Estimate the Effect of Police on Crime. American Economic Review, 87, 270-290. McDowell, M. and Webb, D. (1995). The New Zealand legal system: Structures, processes and legal theory. Wellington: Butterworths. M.R.L. Research Group (1993; 1995). Public attitudes towards policing. Wellington: M.R.L. Research Group.


Myers, S. (1983). Estimating the economic model of crime: Employment versus punishment effects. Quarterly Journal of Economics, 98, 157-166. Schmidt, P. and Witte, A.D. (1984). An economic analysis of crime and justice. New York: Academic Press. Small, J. and Lewis, C. (1996). Economic crime in New Zealand: Causation or coincidence? Auckland: University of Auckland (Working Paper #158). Statistics New Zealand (1981; 1986; 1991; 1996). New Zealand census of population and dwellings. Wellington: Statistics New Zealand. Statistics New Zealand (1996). New Zealand now: Crime. Wellington: Statistics New Zealand. Statistics New Zealand (1996). New Zealand now: Crime tables. Wellington: Statistics New Zealand. Stigler, G.J. (1970). The optimum enforcement of laws. Journal of Political Economy, 78, 526536. Tauchen, H., Witte, A.D. and Griesinger, H. (1994). Criminal deterrence: Revisiting the issue with a birth cohort. Review of Economics and Statistics, 76, 399-412. Trumball, W. (1989). Estimation of the economic model of crime using aggregate and individual level data. Southern Economic Journal, 56, 423-439.


Similar Documents

Premium Essay

Regression Analysis

...Regression analysis is a statistical process for estimating the relationships among variables. It includes many techniques for modeling and analyzing several variables, when the focus is on the relationship between a dependent variable and one or more independent variables. More specifically, regression analysis helps one understand how the typical value of the dependent variable (or 'criterion variable') changes when any one of the independent variables is varied, while the other independent variables are held fixed. Local Government Engineering Department (LGED) is a public sector organization under the ministry of Local Government, Rural Development & Cooperatives. The prime mandate of LGED is to plan, develop and maintain local level rural, urban and small scale water resources infrastructure throughout the country. Here, I considered LGED as the organization and considering a projects eight districts “available fund” as Independent variable and “development (length of development of road in km)” as dependent variable. The value of the variables are- Districts Fund, X (lakh tk) Development,Y (km) Panchagar 450 10 Thakurgaon 310 6.8 Dinajpur 1500 32 Nilphamari 1160 24.5 Rangpur 1450 31 Kurigram 450 9 Lalmonirhat 950 16 Gaibandha 1550 33 For the two variables “available fund” and “development”, the regression equation can be given as: Y= a + bX Where, Y = Development X = Fund b = rate of change of development a...

Words: 365 - Pages: 2

Premium Essay

Regression Analysis

...electricity, property taxes, advertising, accounting, janitors, cleaning supplies, distribution costs, legal fees, interest, inspectors, human resources department, etc, etc, etc. Life would be too easy if it were just that simple.  There is one wrinkle.  There is a distinction between between overhead and manufacturing overhead. Factory Overhead is not a financial statement account It is a  “suspense account” for capturing and reallocating overhead costs Factory Overhead is debited for actual overhead costs incurred Factory Overhead is credited to allocate overhead to production Regression Analysis Interpretation of output summary The regression model like that, Here,    Y= Cost of production A= Constant b1,b2 &b3= Regression coefficient X1= Direct Materials X2 = Direct Labor X3= Factory overhead From the co-efficient table, the values of a, b1,b2& b3 are found out & the regression model can be written as follows: Y= a+b1x1+b2x2+b3x3 = -6537089.828+.248×1+38.489×2+12.326×3 This equation indicates that if taka of direct materials increases by 1taka, the cost production will increases by .248 taka and other things remain constant. Again, if taka of direct labor increases by 1 taka, the cost of production will increases by 38.489 taka and other things remain constant. On the other hand, if taka of factory overhead increases by 1taka, the cost of production will increases by 12.326 taka and other things remain constant. The relationship among......

Words: 578 - Pages: 3

Premium Essay

Regression Analysis

...Case Study: Locating New Pam and Susan‘s Stores Professor Demetra Paparounas Lisa Chan MGSC 6200- Information Analysis July 3, 2014 Introduction The purpose of this study to is to determine a new store location for Pam and Susan Stores. This discount department store chain has 250 stores that are primarily in the South. Expansion is important to their strategic success. A multiple regression model will be used to determine which location has the highest sales potential and projections. It will also be used to help see how strong of a relationship sales has to the other independent variables. Data For this model, the wealth of census data that was used to compute this model contained 250 observations, 33 variables and 7 additional dummy variables were created from the main comtype variable, taking values of zero or one depending on level of competitiveness for a particular store. This data set contained economic and demographical data, population type, sales numbers, store size and the competitive types. The amount of sales and selling square feet variables are given in thousands of dollars. Results and Discussions In analyzing the data on the 250 Pam and Susan’s stores, we first created a scatter plot of the competitive types in the horizontal axis against sales (in thousands) on the vertical axis. The competitive types were identified as follows: * Type 1- Densely populated area with relatively little direct competition. * Type 2 –High income areas with little......

Words: 1892 - Pages: 8

Premium Essay

Regression Analysis

...Ben Leigh American Intercontinental University Unit 5 Individual Project BUSN311-1301B-10: Quantitative Methods and Analysis Instructor Leonidas Murembya April 23, 2013, Abstract This paper will be discussing regression analysis using AIU’s survey responses from the AIU data set in order to complete a regression analysis for benefits & intrinsic, benefits & extrinsic and benefit and overall job satisfaction. Plus giving an overview of these regressions along with what it would mean to a manager (AIU Online).   Introduction Regression analysis can help us predict how the needs of a company are changing and where the greatest need will be. That allows companies to hire employees they need before they are needed so they are not caught in a lurch. Our regression analysis looks at comparing two factors only, an independent variable and dependent variable (Murembya, 2013). Benefits and Intrinsic Job Satisfaction Regression output from Excel SUMMARY OUTPUT Regression Statistics Multiple R 0.018314784 R Square 0.000335431 The portion of the relations explained Adjusted R Square -0.009865228 by the line 0.00033% of relation is Standard Error 1.197079687 Linear. Observations 100 ANOVA df SS MS F Significance F Regression 1 0.04712176 0.047122 0.032883 0.856477174 Residual 98 140.4339782 1.433 Total 99 140.4811 Coefficients Standard Error t......

Words: 830 - Pages: 4

Premium Essay

Regression Analysis

...Introduction Regression analysis was developed by Francis Galton in 1886 to determine the weight of mother/daughter sweet peas. Regression analysis is a parametric test used for the inference from a sample to a population. The goal of regression analysis is to investigate how effective one or more variables are in predicting the value of a dependent variable. In the following we conduct three simple regression analyses. Benefits and Intrinsic Job Satisfaction Regression output from Excel SUMMARY OUTPUT Regression Statistics Multiple R 0.616038 R Square 0.379503 Adjusted R Square 0.371338 Standard Error 0.773609 Observations 78 ANOVA df SS MS F Significance F Regression 1 27.81836 27.81836 46.48237 1.93E-09 Residual 76 45.48382 0.598471 Total 77 73.30218 Coefficients Standard Error t Stat P-value Lower 95% Upper 95% Lower 95.0% Upper 95.0% Intercept 2.897327 0.310671 9.326021 3.18E-14 2.278571 3.516082 2.278571 3.516082 X Variable 1 0.42507 0.062347 6.817798 1.93E-09 0.300895 0.549245 0.300895 0.549245 Graph Benefits and Extrinsic Job Satisfaction Regression output from Excel SUMMARY OUTPUT Regression Statistics Multiple R 0.516369 R Square 0.266637 Adjusted R Square 0.256987 Standard Error 0.35314 Observations 78 ANOVA ...

Words: 684 - Pages: 3

Premium Essay

Regression Analysis

...A Term Paper On BUSINESS STATISTICS 1 Submitted To Dr. Md. Abul Kalam Azad Associate Professor Department of Marketing University of Dhaka Submitted By Group Name: “ORACLES” Section: B Department of Marketing (17th Batch) University of Dhaka Date of Submission: 12- 04-2012 Group profile “ORACLES” | Roll No. |NAME | |42 | Imran Hosen | | | | |74 |Zerin Momtaz Chowdhury | | | | |106 |Toufiqul Islam | | | ...

Words: 1445 - Pages: 6

Free Essay

Unit 5 - Regression Analysis

...Abstract This paper describes the application of regression analysis for the workplace. Three sets of variables are investigated - benefits and intrinsic job satisfaction, benefits and extrinsic job satisfaction, and finally benefits and overall job satisfaction. The regression analysis is performed using Excel and the results are shown in this paper, along with a graph for each set. The results are analyzed for recommendation to the company.   Introduction Regression analysis is performed on three sets of variables – benefits and intrinsic job satisfaction, benefits and extrinsic job satisfaction, and finally benefits and overall job satisfaction. The results of the regression analysis are used to determine whether any relationship exists for the three sets of variables and the strength of the relationship. Benefits and Intrinsic Job Satisfaction Regression output from Excel Regression Statistics Multiple R 0.069642247 R Square 0.004850043 Adjusted R Square -0.004718707 Standard Error 0.893876875 Observations 106 ANOVA df SS MS F Significance F Regression 1 0.404991362 0.404991 0.506863 0.478094147 Residual 104 83.09765015 0.799016 Total 105 83.50264151 Coefficients Standard Error t Stat P-value Lower 95% Upper 95% Lower 95.0% Upper 95.0% Intercept 5.506191723 0.363736853 15.13784 4.79E-28 4.784887914 6.227496 4.784888 6.227496 Benefits -0.057165607 0.080295211 -0.71194 0.478094 -0.216394019 0.102063 -0.21639 0.102063 Graph ......

Words: 770 - Pages: 4

Premium Essay

Voting Tendencies: a Regression Analysis

...Introduction Every four years, the United States presidential election becomes one of the country’s biggest news stories for a large chunk of that year. During this time there are a great many words both spoken and written about how people are going to vote and what factors supposedly lead them to vote a certain way. It seems that much of the time the analysis of these issues is of a qualitative nature, and with that in mind this paper will attempt to approach the problem from a statistical point of view. That is not to say there is a lack of quantitative research on how specific demographics tend to vote; indeed, some of the good data and information out there will be used to inform this paper’s hypotheses and econometric model. This study will focus on the 2012 U.S. presidential election Barack Obama and Mitt Romney (there were several other candidates on the ballot, but none received a significant portion of the votes). Pundits threw around seemingly countless factors in an attempt to analyze and predict how people would vote in this election; this paper will focus on a select group of those factors, in hopes of drawing some firm and well-grounded conclusions as to whether they actually played a statistically significant role. Countywide data will be utilized. The recent widespread availability of election results on a county-by-county level, combined with countywide information from the U.S. Census Bureau, allows for a very large number of observations (at least......

Words: 3259 - Pages: 14

Premium Essay

Pam & Sue Regression Analysis

...Multiple Regression Project: Forecasting Sales for Proposed New Sites of Pam and Susan’s Stores I. Introduction Pam and Susan’s is a discount department store that currently has 250 stores, most of which are located throughout the southern United States. As the company has grown, it has become increasingly more important to identify profitable locations. Using census and existing store data, a multiple regression equation will be used to forecast potential sales, and therefore which proposed new site location will be more profitable. II. Data The data set has 37 independent variables. This includes 7 categorical variables for competitive type and 30 numerical categories. There are 250 stores, meaning the sample size is 250. As the sales are given in $1,000’s of dollars it is best to remember that a unit change in x will correspond to that coefficient of x multiplied by 1,000. III. Results and Discussion Building a multiple regression model requires a step-by-step approach. Failure to follow such methodology could ultimately lead to incorrect and inaccurate forecasting for the dependent variable of interest. Below I will outline the process and findings used to obtain a multiple regression equation to forecast potential sales at newly proposed site of Pam and Susan’s discount department stores. The initial step in building a multiple regression model is to look for outliers and non-linear relationships between your dependent (predicated sales) and......

Words: 1847 - Pages: 8

Premium Essay

Mid City Regression Analysis

...Assignment 2 Mid City Regression Analysis November 6, 2012 Question 1 – Do buyers pay a premium for a brick house, all else being equal? According to Model 1 above, a premium indeed is paid for a brick house if no other factor is considered. The reference being of a non-brick house shows an average price of $121,958 for house of all sizes, all locations and any number of rooms and bedrooms. A brick house, for all of the same criteria shows a premium of $25,810 being paid. Question 2 – Is there a premium for a house in neighborhood 3, all else being equal? Going by Model 2 above, a premium can indeed be observed for houses purchased in Neighborhood 3. Setting neighborhood 3 as the reference point, we observe an average house price of $159,294 with houses in the neighborhoods 1 & 2 showing average prices being $49,140 & $34,063 lower, respectively. Question 3 – Is there an extra premium for a brick house in neighborhood 3, in addition to the usual premium for a brick house? According to model 3, a premium can indeed be observed for brick houses in neighborhood 3 as opposed to brick houses in all other neighborhoods. We test this by adding an interaction variable of brick houses within neighborhood 3 and testing this against the reference which is a brick house and a house within neighborhood 3. With these reference parameters set, we get an average price for the reference group of $148,230 & a premium for the houses meeting the interaction variable......

Words: 449 - Pages: 2

Premium Essay

Regression Analysis

...Acts 430 Regression Analysis In this project, we are required to forecast number of houses sold in the United States by creating a regression analysis using the SAS program. We initially find out the dependent variable which known as HSN1F. 30-yr conventional Mortgage rate, real import of good and money stock, these three different kinds of data we considered as independent variables, which can be seen as the factors will impact the market of house sold in USA. Intuitively, we thought 30-yr conventional mortgage rate is a significant factor that will influences our behavior in house sold market, which has a negative relation with number of house sold. When mortgage rate increases, which means people are paying relatively more to buy a house, which will leads to a decrease tendency in house sold market. By contrast, a lower interest rate would impulse the market. We believe that real import good and service is another factor that will causes up and down in house sold market. When a large amount of goods and services imported by a country, that means we give out a lot of money to other country. In other words, people have less money, the sales of houses decreased. Otherwise, less import of goods and services indicates an increase tendency in house sold market. We can see it also has a negative relationship with the number of house sold. Lastly, we have money stock as our third impact factor of house sold. We considered it has a positive relationship with the number of...

Words: 723 - Pages: 3

Premium Essay

Regression Analysis

...ANALYSIS OF REGRESSION Jessica Cain American InterContinental University Abstract The world today uses statistics in many different ways to understand numbers and possible outcomes. One way that this is by using regression analysis. The regression analysis which is based on a correlation between two variables can help us to better understand the relationship between the two variables. The process which is a valuable one has helped researchers, and businesses to grow based on information obtained from a regression analysis that contains a linear regression. Introduction The purpose of a regression analysis is to help show a linear regression of certain variables. This helps to understand the correlation of the variables being tested. Correlation does give reason to suspect that the relationship between two variables is not die to chance or other hidden variables (Editorial Board, [EB], 2012). This is done by utilizing excel to show how the variables match up, and if one is causing the other or if there are outliers that are affecting the outcome. This is important as it will allow for a company to see and eliminate these unnecessary variables and continue their growth. Benefits and Intrinsic Job Satisfaction Regression output from Excel |SUMMARY OUTPUT | | | | | |Intrinsic |-0.08484 |4.844477 ......

Words: 589 - Pages: 3

Premium Essay

Multiple Regression Analysis of Rb in Bangladesh

... A Report on “Multiple Regression Analysis of Determinants of Dividend Payout Ratio of Reckitt Benckiser” Acknowledgement It is a great honor for us to submit this report to our respected teacher. At first we want to convey our thanks and gratitude to her for assigning us to prepare report entitled, “Reckitt Benckiser”. It would not have been possible for us to complete the report, but for his help. All of the efforts ended at a desired point for the cooperation and hard work, Sincerity and seriousness of our group members. So, all of them as well as our group members are worth of pure compliment. Letter of Transmittal February 14, 2015 Dear Sir, Subject: Submitting the report on “Determinants of dividend payout ratio of Reckitt Benckiser”. We are submitting a well-structured and comprehensive report on Reckitt Benckiser”. Despite many constraints like scope and access to information, we have tried to create something satisfactory. We have tried to follow your guideline in every aspects of preparing this report. We have concentrated on the most relevant and logical areas to make our report coherent as well as practical. We hope this report will entice your kind appreciation. Sincerely, ________________ Executive Summery Reckitt Benckiser is a global leader in household, health and personal care sectors and one of the fast growing multinationals. In our report we mainly deal with Multiple regression analysis of determinants of......

Words: 3639 - Pages: 15

Premium Essay

Significance and Advantages of Regression Analysis

...Significance of Regression Analysis In statistics, regression analysis includes many techniques for modeling and analyzing several variables, when the focus is on the relationship between a dependent variable and one or more independent variables. More specifically, regression analysis helps one understand how the typical value of the dependent variable changes when any one of the independent variables is varied, while the other independent variables are held fixed. Most commonly, regression analysis estimates the conditional expectation of the dependent variable given the independent variables — that is, the average value of the dependent variable when the independent variables are held fixed. Less commonly, the focus is on a quantile, or other location parameter of the conditional distribution of the dependent variable given the independent variables. In all cases, the estimation target is a function of the independent variables called the regression function. In regression analysis, it is also of interest to characterize the variation of the dependent variable around the regression function, which can be described by a probability distribution. Regression analysis is widely used for prediction and forecasting, where its use has substantial overlap with the field of machine learning. Regression analysis is also used to understand which among the independent variables are related to the dependent variable, and to explore the forms of these relationships. In restricted......

Words: 784 - Pages: 4

Premium Essay

Unit 5 – Regression Analysis

...In this document, there will be discussions and data showing the regression analysis. There are charts and graphs show the regression analysis using intrinsic, extrinsic dependent variable and benefits as the independent variable. Benefits and overall job satisfaction is discussed and represented in the charts, graphs and data.   Introduction There is data, charts and graphs representing job satisfaction of Intrinsic, Extrinsic and overall. There are discussions on the slop, y-intercept, equation and r^2 using intrinsic, extrinsic and overall components of each regression output. Benefits and Intrinsic Job Satisfaction Regression output from Excel Benefits Intrinsic 5.4 5.5 6.2 5.2 2.3 5.3 4.5 4.7 5.4 5.5 6.2 5.2 2.3 2.1 4.5 4.7 5.4 5.4 6.2 6.2 6.2 5.2 2.3 5.3 4.5 4.7 5.4 5.4 6.2 5.5 6.2 5.2 5.4 5.3 6.2 4.7 2.3 5.5 2.3 4.7 4.5 5.3 2.3 4.7 4.5 4.7 5.4 5.5 6.2 5.2 2.3 2.1 4.5 4.7 5.4 5.4 6.2 6.2 2.3 5.2 4.5 5.3 5.4 4.7 6.2 5.4 6.2 6.2 4.5 5.2 5.4 5.3 6.2 4.7 2.3 5.2 4.5 5.3 5.4 5.3 SUMMARY OUTPUT Regression Statistics Multiple R 0.468795174 R Square 0.219768915 Adjusted R Square 0.199236518 Standard Error 0.713005621 Observations 40 ANOVA df SS MS F Significance F Regression 1 5.44142339 5.44142339 10.70352 0.002279584 Residual 38 19.31832661 0.508377016 Total 39 24.75975 Coefficients Standard Error t......

Words: 1225 - Pages: 5