37
Table 4 presents the number of Joliet shots fired and robbery incidents between January
2005 and September 2012. After the implementation of the STD strategy, the mean monthly
count of shots fired and robbery incidents increased by 12.81% and 25.24%, respectively, from
pre-test monthly means of 31.08 (shots fired) and 11.25 (robbery) incidents to post-test monthly
means of 35.06 and 14.09 incidents. The data in Figure 4 illustrate the monthly counts of
confirmed shots fired across the 93-month study period.
Figure 4
Confirmed Shots Fired, January 2005 – September 2012
The data in Figure 5show the monthly counts of robberies across the study period. Although
the average monthly counts of robberies increased from the pre-to-post periods, there was a
18.3% decline in average monthly robberies between the STD and STD+ periods.
38
Figure 5
Robberies, January 2005 – September 2012
Research Design
To further assess and clarify the potential impact of the STD strategy on shots fired and
robbery, we conducted an interrupted time series quasi-experiment.
Time series designs
attempt to detect whether an intervention has had an effect significantly greater than the
underlying trend. They are useful in program implementation research for evaluating the
effects of interventions when it is difficult to randomize or identify an appropriate control
group. Data are collected at multiple time points before and after the intervention. The
multiple time points before the intervention allow the underlying trend to be estimated,
the multiple time points after the intervention allow the intervention effect to be
estimated accounting for the underlying trend. Time series designs increase the
confidence with which the estimate of effect can be attributed to the intervention,
although the design does not provide protection against the effects of other events
39
occurring at the same time as the study intervention, which might also improve
performance.
Typically, a time series has three unobserved components, namely:
Trend (T) – the long-term upward or downward movements of the time series due to
influences such as population growth or general economic development.
Seasonal fluctuations (S) – a regular periodic pattern that repeats from year to year
(e.g.,
the number of shots fired increases in summer months).
Irregular component (I) – represents the non-systematic movements of the series caused
by events of all kinds not captured by the other components (e.g., random error).
If a time-series model does not account for these sources of error, the intervention analysis
will be confounded. To account for trends in the time series, we included a simple trend variable
for linear trends and a trend-squared variable for curvilinear trends. The trend variable was
simply the month number from the start to the end of the time series (i.e., for the January 2005
through September 2012 series, the trend variable ranged from 1 to 93. The trend-squared
variable was calculated by taking the square of the trend variable. However, the trend-squared
measure did not improve the fit of the model to any of the pre-intervention time series;
consequently, this variable was dropped from the analytic models.
The issue of seasonal fluctuations (i.e., seasonality) is particularly relevant for both the shots
fired and robbery data in Joliet. Not surprising, the peak months for shots fired were the summer
months (May – August). Forty-one per cent of the confirmed shots fired incidents occurred
during these months. Similarly, the peak months for robbery were May-September accounting
for 46.4% of robbery incidents. Seasonal fluctuations in data make it difficult to analyse whether
changes in data for a given period reflect important increases or decreases in the level of the data,
40
or are due to regularly occurring variation. To search for the measures that are independent of
seasonal variations, statistical methods have been developed to remove the effect of seasonal
changes from the original data to produce seasonally adjusted data. The seasonally adjusted data
provide more readily interpretable measures of changes occurring in a given period, reflecting
real movements without misleading seasonal changes.
In the current study, seasonal adjustment of the time series data was carried out with the
SPSS seasonal decomposition program (X-12-ARIMA). This method is based on a moving-
average technique and is more sophisticated and able to provide adjustments customized to the
characteristics of individual series. The components of a time series are estimated using weighted
moving averages, or filters, on the series. Trend filters smooth the short-term fluctuations out of
a time series, leaving the long-term movements -- the trend. By definition, the seasonally
adjusted estimates will contain an estimate of the underlying trend and the irregular components.
To identify whether there was a serial autocorrelation component, we analyzed the pre-
intervention time series. We used Auto Regressive Integrated Moving Average (ARIMA) models
to detect whether the monthly counts of shots fired and robbery events were serially auto-
correlated (i.e., the number of events made in January 2005 was significantly correlated with the
number of events in February 2005, and so on). The pre-intervention time series data did not
show significant serial autocorrelation; therefore we did not estimate an autoregressive
component in our model. We also ran an OLS model on the pre-intervention time series for shots
fired and robberies (seasonalized monthly shots fired counts = constant + trend + trend
2
+ error;
seasonalized monthly robbery counts = constant + trend + trend2 + error) and analyzed the
residuals using the Durbin–Watson Test (shots fired result = 1.78; robbery result = 1.). The
Durbin–Watson Test ranges from 0 to 4. First-order serial correlation does not exist when the
Dostları ilə paylaş: |