St. Ambrose University Dubuque Police Traffic Stop Analysis Copyrighted
July 17, 2017
City of Dubuque Consent Items # 6.
ITEM TITLE: St. Ambrose University Dubuque Police Traffic Stop
Summary
SUMMARY: City Manager providing the St. Ambrose
University Dubuque Police Traffic Stop Summary
presented at the July 12, 2017 City Council work session.
SUGGESTED DISPOSITION: Suggested Disposition: Receive and File
ATTACHMENTS:
Description Type
St. Ambrose University Dubuque PD Traffic Stop Supporting Documentation
Executive Summary
Dubuque Police Traffic Study
Executive Summary
Prepared by:
Chris Barnum
Jackson Boland
Austin Mack
Marc McClory
St. Ambrose University
July, 2017
Dubuque Police Traffic Study
In 2016,the City of Dubuque partnered with St. Ambrose University to develop and
implement an analysis of the Dubuque Police Department's traffic stop activity. The
investigation focused on evaluating stops made by the DPD between January 1st,2015 and
December 31st 2015 and centered on two broad categories of discretionary police conduct: (i)
racial disparity in vehicle stops instantiated as racial differences in the likelihood of being
stopped by the DPD and (ii) dissimilarities across racial demographics in the outcome or
disposition of a stop.
To evaluate the racial demographics of stops, our research team utilized driver-population
benchmarks fashioned from roadside observations and census data. A benchmark should be
thought of as the proportion of minority drivers on the roads in a given location. In plain terms,
the benchmark is a standard that can be used to judge the percentage of minority drivers that
should be stopped by the police when no bias is occurring. In Dubuque, the population
characteristics of individual neighborhoods vary across the city. Consequently, it was necessary
to form separate benchmarks for these distinct areas. The first step in the process involved
strategically placing traffic observers in these neighborhoods or `observation zones.' To facilitate
this, we first isolated two key features of the city: (i) locations where the police tended to make a
lot of stops and (ii) areas characterized by comparatively high concentrations minority member
residents. Interestingly, in Dubuque these areas also happen to roughly parallel 2010 US Census
Tracts. This was fortuitous because it enabled us to use the boundaries of US Census tracts as
observation zones borders. The two figures below illustrate this by showing the locations and
frequency-density of traffic stops made by the DPD and then the US Census Tract locations.
Figure 1
M11-
u �
ower Densiry HigM1er Density - � � �.
Dubuque City Limits DUB F 'II
1
Figure 2
cit:y of DU
2001 .4"raucts
R � pc p4M.
n L .•
p n�
Once the boundaries of the observation zones were determined,roadside surveyors were
deployed to monitor traffic at several locales within selected zones. The observers watched
traffic at various times of the day ranging from 7:00 am until 2:00 am. They surveyed vehicles
seven days a week from mid-January, 2017 through mid-May, 2017. The observers logged more
than 55,000 observations from fifty-three locations across the city. Analyses show that the traffic
observers saw similar percentages of minority drivers on the roads both day and night.'
Additionally, results show a high degree of consistency between observers across all zones.2
These findings suggest that the benchmarks used for the analyses were a valid and representative
sampling of the drivers on the roadways in Dubuque.
The process of comparing police data to benchmarks is straight forward. It centers on
identifying differences between the demographic percentages from DPD traffic stop data and
benchmark information. Any positive difference between benchmark values and police data
signifies dfsproporfionafity—or an over representation of minority drivers in the data. Although,
' Roadside observers' night and day observations values did not significantly differ within zones:t=0.057, p=0.95,
12 of.
'Correlation rfor inter-surveyor observations within each location equaIs0.63
2
disproportionality can indicate bias or discrimination, it does not necessarily indicate bias. It is
possible for disproportionality to occur for a number of legitimate reasons.3
The central feature of this methodology is a disparity index or weighted average. For
each observation zone, the disparity index equals the difference between the percentage of
minority drivers stopped and the benchmark value, weighted by the number of traffic stops made
in the zone. When summed across all observation zones, this index gives a summary measure of
disproportionality.
The methodology makes it possible to track disproportionality by area, by time of day, by
duty assignment and by individual officer. While the disparity index serves as a useful tool in
assessing disproportionality, please keep in mind that the index value is only an estimate of
disproportionality in police activity, not a cast-iron fact. This is because the index is predicated
on benchmarks which are formed from samples of the drivers on the roads in a given area and
time. And consequently like any sample, a benchmark is associated with a degree of sampling
error.
Stop Analyses Results: The disparity index value for the DPD indicates that on average, the
2015 level of disproportionality in traffic stops for the entire department was low at about 2%
points above benchmark values. 4 Table 1 below gives the information used to calculate the
disparity index.5
Table 1 information for calculating disparity index
Department
zone min % bench stops
1.1 0.13 0.11 312
1.2 0.11 0.11 320
1.3 0.18 0.15 633
4 0.03 0.05 266
5 0.24 0.2 680
6 0.14 0.12 174
7.1 0.25 0.12 103
7.2 0.12 0.1 143
9 0.12 0.08 346
11 0.05 0.05 523
99 0.08 0.05 1684
The differences in table 1 between the percentages of minority drivers stopped by the
DPD (column 2) and corresponding benchmark values (column 3) are consistently small. Only in
'Including differences between racial groups in driving behavior,vehicle condition,drivers' license status and so
forth.
°T-tests show that census and final benchmark values did not significantly differ t=0.99;p=0.33;df=18.
However, in comparison to other areas, zone 1 census values tended to run substantially higher than roadside
observations.Consequently,we broke census tract 1 into three smaller census block group areas and then
modified the benchmark.
'Zone 99 represents areas of Dubuque that were not included in our observation zones. These areas are generally
well outside the city-center. Additionally, these areas are not included in the high density traffic stop areas of
Dubuque (represented by blue in figure 1.).Analysis of US Census data indicates that the mean aggregate minority
population in these areas is roughly 5%.This value was used as the benchmark in zone 99 areas.
3
zone 7.1 is there a substantial difference. However, comparatively few stops were made in this
zone, so this difference did not substantively impact of the overall level of disproportionality.
Comparisons of mean-differences show that the overall level of disproportionality is not
significantly different from zero.' In plain terms, the average level of disproportionality of 2%
points above benchmarks falls within the margin of error and is likely not meaningful.7
Table 2 information for calculating disparity index based on time of day.
Days Nights
zone min% bench stops zone min% bench stops
1.1 0.12 0.11 170 1.1 0.15 0.11 142
1.2 0.08 0.11 219 1.2 0.17 0.11 101
1.3 0.17 0.15 393 1.3 0.20 0.15 240
4 0.02 0.05 214 4 0.06 0.05 52
5 0.24 0.2 330 5 0.23 0.2 350
6 0.11 0.12 101 6 0.19 0.12 73
7.1 0.23 0.12 64 7.1 0.28 0.12 39
7.2 0.11 0.1 106 7.2 0.14 0.1 37
9 0.10 0.08 176 9 0.15 0.08 170
11 0.03 0.05 349 11 0.07 0.05 174
99 0.07 0.05 1024 99 0.09 0.05 660
Table 2 above, breaks out information by time-of-day. The findings indicate that roughly 60% of
all DPD stops were made during daylight hours. Additionally, calculations for the disparity index
show virtually no disproportionality for stops made during the day. Here the disparity index
value is less than 1%point above benchmarks and is well within the margin of error. The index
value for stops made during nighttime hours is higher at roughly 4%points above benchmarks.
Although this finding is significant in a statistical sense, substantively this value is generally
lower than comparative disparity index values for other police departments in Eastern Iowa.
Stop Outcomes Results: We used an examination of stop outcomes to assess disproportionality
in citations, arrests and consent searches. As the name implies, a stop outcome gives information
about the consequence of a stop. An example would be whether or not a driver received a ticket
as a result of the stop. In what follows we measure disproportionality using an estimator called
an odds ratio. This estimator is a measure of effect size and association. It is useful when
comparing two distinct groups and summarizes the odds of something happening to one group to
the odds of it happening to another group.
Univariate odds ratio analyses shown in table 3 indicate that Dubuque officers were
slightly more likely to ticket minority ticket minority drivers than others but were also
significantly more likely to arrest minority drivers and to ask for consent to search their vehicles.
e Unweighted t-test(t=1.11,p=0.27, df=20).
Although there are a number of ways to calculate margin of error, the one used for this analysis is simple and
conservative-an unweighted t-test of comparison between means.
4
Table 3 department outcomes and odds ratios
Outcome Odds-ratio
citations 1.21
arrests (raw) 3.49
arrests (adjusted) 2.66
search requests 6.08
Care should be exercised when evaluating findings for search requests. Although the
odds ratio of 6.08 is large, Dubuque officers asked drivers for permission to search very
infrequently. In fact, this occurred only twenty times during the entire study period (or roughly
once every 260 stops). This small sample size makes it virtually impossible to draw meaningful
conclusions and should be taken into consideration when interpreting results.
In addition, we were not able to exclude several potentially important alternative
explanations when analyzing arrest data using univariate odds-ratios.8 For instance, raw arrest
information from table 3 indicates that DPD officers were roughly three-and-one-half times more
likely to arrest minority members as others on traffic stops (OR=3.49). However,the limited
nature of the data make it impossible to evaluate whether this level of disproportionality was a
product of bias, other factors or a combination of both. We did however have access to one piece
of information that may hint at an answer. When we excluded from the analyses stops that were
made for warrant checks, we found that the odds ratio for arrests meaningfully decreased. This
finding suggests multiple factors may account for arrest disproportionality. Finally, we also
conducted multivariate logistic regression analysis to control for several important contextual
variables including time-of-day, age and gender of driver as well as stop location. Results of
these analyses were substantively similar to those from univariate findings. This congruence
suggests the univariate findings are reasonable. Findings from multivariate analyses also suggest
that DPD officers stopped minority and non-minority drivers in similar proportions for moving
violations and equipment violations. 9
Conclusions
This study examined the traffic stop behavior of the Dubuque Police Department using data
from 2015 more than 5000 stops. The investigation focused on two broad categories of
discretionary police conduct: (i) racial disparity in vehicle stops and (ii) disparity in the outcome
or disposition of a stop. Findings from the examination of disproportionality in vehicle stops
show little or no evidence of disparity. Given that the term racial profiling is generally applied to
contexts where the police use race as a factor in deciding whether to stop a vehicle, and that
racial profiling routinely shows up as disparity in traffic stops,the results from this portion of the
analyses provide no evidence that the Dubuque Police Department systematically engaged in
racial profiling in 2015. However, in addition to profiling, racially biased policing can take other
forms, including disparity in stop outcomes. Notably, the results of the analyses for stop
outcomes do indicate some racial disproportionality in certain outcomes specifically, moderate
amounts in arrests and lesser amounts in citations. These findings may signify bias, however
'These factors include things like whether the arrest was discretionary, potential differences in driving status or
offending rates,and subject demeanor, etc. These factors could serve as competing explanations for the observed
results.
9 The complete list of control variables include:time of day,gender of driver,age of driver, location of stop and
type of violation (moving or equipment violation). Results of logistic regression are available upon request.
5
further work is needed to know for sure. Any future analyses should focus on assessing the
likelihood of potential alternative explanations for the findings. Consequently, we suggest that
the City of Dubuque conduct one additional round of analyses. The proposed work could utilize
DPD traffic stop data from 2016 and 2017, with no updates to observational zone benchmarks
needed. The additional analyses should take a two-prong approach by: (i) looking for trends or
changes in disproportionalitylo and (ii) by developing and evaluating supplementary measures of
arrests and counts of outcomes." The new measures should focus on identifying whether arrests
were the result of discretionary police decisions and include more nuanced measures for counts
of outcomes.
"A good way to gain confidence in the truthfulness of a statistical finding(including the disparity index) is to look
for changes in values across time rather than concentrating on single year values.
"The current data set indicates whether a given outcome (such as a citation or arrest)occurred on a traffic stop
but does not count the number of such events.
6