Violence Reduction in Joliet, Illinois: An Evaluation of the Strategic Tactical Deployment Program

41 

 

Durbin–Watson statistic is close to 2. 

 

The key outcome variables in our assessment of the STD intervention were the monthly 

number of confirmed shots fired and robberies. Since the underlying data were counts, a Poisson 

regression in a log-linear model was selected to analyse the time series data. Poisson regression 

applies where the dependent variable is a count (e.g. crime incidents, cases of a disease) rather 

than a continuous variable. It assumes the response variable has a Poisson distribution whose 

expected value (mean) is dependent on one or more predictor variables. Typically the log of the 

expected value is assumed to have a linear relationship with the predictor variables.  As Crawley 

(2007, p. 527) notes, linear regression is not appropriate for such data since (1) the linear model 

might lead to the prediction of negative counts; (2) the variance of the response variable is likely 

to increase with the mean; (3) the errors will not be Normally distributed, and (4) zeros are 

difficult to handle in transformations. 

 

 

The aim of the time series approach was to isolate and evaluate the direct impact of the 

implementation of the STD only and STD plus interventions on reported offenses in Joliet. To 

accomplish this

 three dummy variables were created to represent the three distinct program 

periods (i.e., pre-STD, STD only and STD + probation/parole) in order to estimate the effects of 

the intervention on the monthly counts of shots fired and robberies. 

It also is important to 

consider that potential reductions in shots fired and robbery associated with the STD intervention 

could be influenced by other factors. Therefore

 

we

 

also included covariates to control for any 

changes in the monthly counts of shots fired and robberies that could be associated with other 

factors such as changes in Joliet’s percent of minority population, the unemployment rate and the 

number of police officers.  Furthermore, the models control for the number of drug arrests across 

the pre and post intervention periods. Over the 7.75 year period there were 5,619 drug arrests 

---

42 

 

(1,567 pre-intervention and 4,052 intervention drug arrests). Controlling for drug arrests may be 

of consequence in light of research on the impact of zero tolerance policing strategies on crime 

reduction. 

Studies have shown that aggressive enforcement activities by uniformed patrol 

officers targeting illicit drug sale locations and illegal drug activity can produce short-term 

reductions in street level drug dealing as well as reductions in many other types of criminal and 

disorderly behavior (Braga et al., 1999; Sherman and Rogan, 1995; Sherman and Weisburd, 

1995; Weisburd and Green, 1995). For example, aggressive patrolling of suspected drug 

locations in Jersey City involving stops and searches by patrol officers resulted in reductions in 

crime not only at the drug locations, but for several surrounding blocks as well (Braga et al., 

1999; Weisburd and Green, 1995).   

Results

 

T

able 5 presents the results of the Poisson regression models for shots fired and robbery. 

Poisson regression models have the defining characteristic that the conditional mean of the 

outcome is equal to the conditional variance. 

Goodness of fit tests based on the Deviance and 

Pearson residuals were used to assess whether the model assumptions have been violated. Each 

should approximately equal its degrees of freedom and so Value/df (value divided by degrees of 

freedom) should be close to one. The data show that the Poisson regression model is a good fit. 

Next, the Omnibus Test, a test to determine if all of the estimated coefficients are equal to zero 

(a test of the model as a whole), was statistically significant.  

 

The parameters for the independent variables were expressed as incidence rate ratios (i.e., 

exponentiated coefficients). Incidence rate ratios are interpreted as the rate at which things occur; 

for example, an incidence rate ratio of 0.65 would suggest that, controlling for other independent 

variables, the selected independent variable was associated with a 35% decrease in the rate at 

---

43 

 

which the dependent variable occurs. To ensure that the coefficient variances were robust to 

violations of the homoskedastic errors assumption of linear regression models, 

Huber/White/sandwich robust variance estimators were used. 

 

Table 5 

Poisson Regression Results for Shots Fired and Robbery 

      

 

 

Shots 

Fired 

       Robbery 

Variable 

B (SE) 

IRR 

p-value 

B (SE) 

IRR 

p-value 

Trend 

.006 

(.006) 

1.01 .170 -.003 

(.007) 

.997 .672 

unemployment 

.000 

(.016) 

1.00 .975 -.025 

(.021) 

.976 .241 

% minority pop. 

.294 (1.07) 

1.34 

.783 

1.63 (1.40) 

5.12 

.245 

# police officers 

.001 (.002) 

1.00 

.477 

.000 (.002) 

1.00 

.920 

Monthly # drug 

arrests 

 

.001 (.002) 

1.00 

.595 

.004 (.003) 

1.00 

.136 

STD only 

-.123 (.126) 

.885 

.329 

.453 (.191) 

1.57 

.018 

STD + 

probation/parole 

 

-.368 

(.235) 

.692 .118 .436 

(.355) 

1.55 .220 

intercept 

2.89 (.133) 

--- 

.000 

1.61 (.236) 

--- 

.000 

 

Shots fired:   Deviance = 205.08, Pearson Chi-square = 202.45 Value/df = 2.41/2.38 

          Likelihood ratio Chi-Square = 22.75, df = 7, p = .002 

 

Robbery: 

Deviance = 166.09, Pearson Chi-square = 154.18 Value/df = 1.95/1.81 

     

       Likelihood ratio Chi-Square = 42.40, df = 7, p = .000 

Controlling for the covariates, the STD and STD+ program interventions were associated 

with a statistically significant decrease in the monthly number of shots fired. According to the 

incidence rate ratios, the STD only and STD + probation and parole components were associated 

with a 11% and 31% decrease in the monthly number of shots fired events, respectively.  

However, these decreases did not differ significantly from the pre-intervention period (p = .329 

---