Observational Surveys: Implementation and Analysis

Observational Surveys: Implementation and

Analysis

2

Observational Surveys - Speakers

William W. Stenzel, D.Sc.– Associate Director, Center for Public Safety

(Management Consulting)

Roy E. Lucke– Director, Research and Development for the

Center for Public Safety

3

Notebook Materials

A hardcopy of all of the PowerPoint slides for this session (“Observational Surveys”) can be found in the seminar notebook.

4

Why Use Observational Surveys?

For the Illinois Traffic Stop Statistics Study, “disparities” are going to be calculated by comparing:– the racial composition of traffic stops– the racial composition of the driver population.

The racial composition of the driver population is going to estimated by using “adjusted census data” at the city and county level.

5

Why Use Observational Surveys?

What options does an agency have if there is concern that the adjusted census data will not provide accurate information about the racial composition of the driver population in its jurisdiction?

One option is to obtain a better estimate of the racial composition of the driver population with the use of observational surveys.

6

Center for Public Safety Experience with Observational Surveys

The Center for Public Safety has conducted three observational surveys for agencies in Illinois:

– Highland Park (Sept-Oct 2001 – Stenzel)– Hinsdale (May 2004 – Lucke)– Schaumburg (August 2004 – Lucke)

7

Observational Survey Topics

Observational Surveys:

– Topic 1: Data Collection the nuts and bolts of how to conduct a survey

– Topic 2: Data Summarization putting the survey data into a format suitable for

review and analysis

– Topic 3: Data Analysis comparing and assessing the survey and traffic

stop data

8

Observational Surveys - Topic 1

Topic 1: Data Collection

The nuts and bolts of how to conduct a survey

9

Conducting Observational Studies

Once the decision is made to do an observation

study, there are three major tasks:

Determining what data to collect Identifying data collection sites Recruiting and training observers

10

Conducting Observational Studies

In addition to the three major tasks, other steps include:

Scheduling data collection Equipping the data collectors Developing forms, data entry and data

analysis procedures

11

Preliminary Information

It is only possible to obtain good driver demographic information from stopped vehicles

Observations can only be made at intersections with traffic signals or stop signs– Efforts to observe drivers on controlled-access

roadways were not successful

12

Site Selection

Primary Criteria

– Conduct observations at or near intersections that are among the agency’s high traffic stop locations.

– Agency should try to identify locations for all stops, not just where citations are issued

– Also identify times of day and days of week for stops so observation times can be matched as well as possible

13

Site Selection, Continued

Site must provide a good view of stopped vehicles– Steep shoulders may raise observers too high– “Sweeping” right turn lanes might keep observers

too far from lanes to see– No limit on number of lanes – observers need only

check lanes they can clearly see Site must be safe for observers

– There must be a shoulder or sidewalk – curbs are desirable

– Observers must be free from potential harassment

14

Survey Sessions

Session is a 2 or 3 hour observation period Sessions should be distributed across all days of

the week, according to traffic stop information Sessions can be done

– Mornings– Afternoons– Early evenings– Again, dependent on stop information and available

daylight Surveys should be done in both directions of

travel

15

The Survey Team

Three individuals are needed for each session– Observer– Recorder– Counter

16

The Survey Team, Continued

Team members can be recruited from a number of possible sources. – Agency volunteers or auxiliaries– College students (e.g., criminal justice students)– Crossing guards– Temporary labor pools– Etc.

17

The Survey Team, Continued

Training must be provided to survey team members– Classroom instruction covering the nature of the

project and what they will be expected to do– Practice sessions under guidance of project leaders

Team members must be scheduled in groups of threes at dates and times identified for surveys. – Have substitutes available– Project leaders should oversee all observation sessions

18

Sample Agenda for Observer Training

Agenda Review Agenda Complete Forms Driver Survey Vehicle Counter Field Work Schedule/Signups

19

Survey Team Equipment

Safety vests

Traffic counting devices (or digital cameras)

Clipboards and pencils

Rain gear (ponchos, umbrellas, “writing pouches”)

20

Data Items to be Recorded

Each agency must decide what data items they believe are important to capture. Candidates items include:– Driver race/ethnicity– Driver gender– Driver age– Number of passengers in vehicle– Driver residency– Type of vehicle

21

Data Collection

Paper “check mark” form Scantron form Palm or other hand-held device Tablet-type personal computer

Each session should be stored as a separate file, either in a physical packet or data file

22

Sample Data Collection Form

Date: ______________________ Start Time: ______________ End Time: _____________

Location: ____________________________Travel Direction: _______ Session #: __________

Survey Conducted by: ____________________________________________________________

Notes: _________________________________________________________________________

# ofObs # M F U W B H O U Y M E O Pas.

Sheet No. ___________ of ____________

Gender Race/Ethnicity Age

Hinsdale Driver Observation Study

23

Number of Observations

The number of data collection sessions can be affected by:– Traffic volume– Roadway configuration (number of lanes)– Stop signs or traffic signals– Number of data items to be collected

General observations:– Higher capture rates (75% - 100% of drivers) at

stop signs, but usually lower traffic volumes– Lower capture rates (20% - 75%) at signalized

intersection depending on volume, number of lanes, and signal timing

24

Observation Limitations

Can be done only during daylight hours Glare from windows (or window tinting) can

affect observation Weather (rain, snow, excessive heat) Subjective decisions be observers Cost of doing surveys (labor intensive

activity)

25

Topic 2: Data Summarization

Data Summarization:– Recordkeeping – Data Entry – Data Base Software

26


Recordkeeping - additional information that should/can be added to each observation (record):– Location (should)– Day of the week (should)– Time of day (should)– Direction of traffic (optional)– Data collectors (optional)

27


Data Entry - getting the data into an electronic format– “The old fashion way” – keying the data in– Use machine-readable data collection forms

(e.g., Scantron)– Download from a file created at the time of

data collection (e.g., from a Palm Pilot or a PC tablet)

28


Data Base Software - a computer program that can be used to:– Manipulate the data (i.e., sort and filter)– Display the data (i.e., print summary tables

and charts)– Describe the data (i.e., compute various

descriptive attributes): number of observations Average value Minimum and maximum values

– Examples: (Access, EXCEL)

29


Example of a printout

Survey Site

White

Black

Hispanic

Other

Number of Observations*

Intersection 1

85.8% 3.2% 6.1% 5.0% 2,191

Intersection 2

78.0% 6.7% 9.9% 5.4% 1,241

Intersection 3

72.7% 5.1% 10.5% 11.6% 1,421

Intersection 4

84.8% 2.2% 6.1% 6.9% 1,200

Intersection 5

83.6% 2.7% 7.4% 6.3% 1,218

Intersection 6

80.5% 2.2% 8.3% 9.0% 873

Total** 80.8% 4.2% 8.0% 6.9% 8,144

* - The percentages shown in the table are based on the adjusted number of observations (i.e., the total number of observations less the number of “Unknowns” recorded for Race). ** - The percentages shown for all 6 survey sites are based on weighted averages of the percentages for each survey site. The percentages are weighted using the average traffic volume for each site.

30

Observational Surveys - Topic 3

Topic 3: Data Analysis

Comparing and assessing the survey and traffic stop data

31

Topic 3: Data Analysis

Data analysis consists of comparing two sets of data:– Traffic stop data – Driver survey data

And addressing the question: Are differences between the two sets of data important?

32

Data Analysis: A Sample Comparison

White Black Hispanic Other Total Traffic Stops Number of 237 52 98 21 408 Percent 58.1 12.7 24.0 5.1 100.0 Driver Survey Number of 1,165 250 349 117 1,881 Percent 61.9 13.3 18.6 6.2 100.0

Question: Are the differences in the percentages between the traffic stop data and the driver survey data in the each racial category important?– Are differences due only to natural variation, or– Are differences due to the some outside influence on

the officer’s decision about whom to stop (e.g., race)?

Two data sets

5

33

Statistical Benchmarking

Statistical benchmarking consists of:

1. Comparing two sets of data: Encounter data: the racial composition

of drivers in traffic stops, and Survey data: the racial composition of

drivers who are potential participants in a traffic stop

2. A procedure for assessing the significance of differences in the percentages between the two data sets

and

34

Statistical Benchmarking

Highland Park

– Statistical benchmarks were used to assess the importance of the differences in the percentages in the driver survey and traffic stop data.

– The benchmarks were determined using a statistical procedure called “confidence intervals.”

35

Confidence Interval Example


Example: Is the difference between the two percentages for Hispanics (i.e., 24.0% and 18.6%) important?

One way to address this is to determine a range of values (i.e., a confidence interval) for the expected number of traffic stops involving Hispanic drivers.

36



Example: The confidence interval for the number of Hispanics in the traffic stop data shown above is:

[64, 96]. This interval can be interpreted as follows:

– If the decision about who to stop is not influenced by race, then the expected number of Hispanics stopped, due to normal variation, should fall between 64 and 96.

37

Confidence Intervals Example

The upper and lower limits for the confidence interval can be interpreted as statistical benchmarks for the number of Hispanics stopped.

The limits are determined based on:– Total number of traffic stops (408)– Estimated number of Hispanics in the driver

population– Selected confidence level

38



Example: The statistical benchmarks for the expected number of Hispanics, [64, 96], is based on a confidence level of 95%.

The 95% confidence level means the margin of error is 5%. A 5% margin of error means that there is 5% chance that

even with normal statistical variation the number of Hispanics stopped could fall below 64 or above 96.

1

39

Statistical Benchmark Example

The upper and lower benchmarks for each racial category are shown at the bottom of the table.

These benchmarks are compared with the actual number of encounters in each racial category.

Except for Hispanics, the actual number of stops within each category falls within the benchmark limits.

White Black Hispanic Other Total Driver Survey Number of Encounters 1,165 250 349 117 1,881 Percentage 61.9 13.3 18.6 6.2 100.0 Traffic Stops Number of Encounters 237 52 98 21 408 Percentage 58.1 12.7 24.0 5.1 100.0 95% Confidence Interval (Statistical Benchmarks) Upper Benchmark 269 68 96 34 Lower Benchmark 231 41 64 16

2

40

Statistical Benchmark Example

The number of Hispanics stopped in this example, 98, is outside the statistical benchmarks of [64, 96].

THIS DOES NOT PROVE RACIAL PROFILING.

It indicates that further investigation is needed to determine what special circumstances might be present that are influencing the number of Hispanics that are stopped.

White Black Hispanic Other Total Driver Survey Number of Encounters 1,165 250 349 117 1,881 Percentage 61.9 13.3 18.6 6.2 100.0 Traffic Stops Number of Encounters 237 52 98 21 408 Percentage 58.1 12.7 24.0 5.1 100.0 95% Confidence Interval (Statistical Benchmarks) Upper Benchmark 269 68 96 34 Lower Benchmark 231 41 64 16

41

Why Use 95%?

Use of 95% for the confidence interval is a conservative approach that assumes that racially motivated policing is not occurring unless there is significant evidence to the contrary.

Justification for a conservative approach is appropriate in view of the many uncertainties associated with the data:

– Difficulty in identifying race– Different driver behaviors by race– Different driver behaviors by gender and age– Unknown mix of drivers by gender and age by

race

42

How Can I Use Statistical Benchmarks?

The benchmarking procedure described is based on statistical procedure called the “two-sample test for proportions.”

Its use requires a basic understanding of applied statistics. (Note: “Statistical Benchmarks for Police Traffic Stops” in seminar notebook.)

To help departments that may want to use statistical benchmarking based on this procedure, the Center for Public Program has put an easy-to-use spreadsheet on its website that can be used to find statistical benchmarks.

43

Statistical Benchmarking Spreadsheet

Northwestern University Center for Public Safety

Note: Only enter data in the colored (shaded) cells.

Confidence Level (%) 95

Traffic Stops White Black Hispanic Other Total Number of Stops 237 52 98 21 408

Percentage 58.1% 12.7% 24.0% 5.1% 100.0%

Driver Survey White Black Hispanic Other Total Number of Stops 1165 250 349 117 1881

Percentage 61.9% 13.3% 18.6% 6.2% 100.0%

Pooled Est. of Percent 61.2% 13.2% 19.5% 6.0% 100.0%

White Black Hispanic Other TotalLower Limit 231 41 64 16

Upper Limit 269 68 96 34

Benchmarking Limits On the Number of Traffic Stops*

- Statistical Benchmarks -

44

Statistical Benchmark Spreadsheet

The statistical benchmarking spreadsheet can be found on the website for the Center for Public Safety:

www.northwestern.edu/nucps

Select “Links”Select “Racial Profiling”At bottom of page under “Recent Articles” find:

“Benchmarking Spreadsheet”

45

Contact Information

William Stenzel– 847/491-8995– [email protected]

Roy Lucke– 847/491-3469– [email protected]

Observational Surveys: Implementation and Analysis

Documents