Loading...
The URL can be used to link to this page
Your browser does not support the video tag.
Home
My WebLink
About
Staff Report 180-07
City of Palo Alto City Manager’s Report TO:HONORABLE CITY COUNCIL FROM:CITY MANAGER DEPARTMENT: POLICE DATE: SUBJECT: APRIL 9, 2007 CMR:180:07 ANALYSIS OF DEMOGRAPHIC DATA FROM THE FIRST AND SECOND QUARTERS OF FISCAL YEAR 2006-07 This is an informational report. No Council action is required. BACKGROUND Since July 1, 2000, the Police Department has been proactively collecting demographic data on all enforcement contacts. On September 10, 2001, the Police Department submitted its first quarterly report on this data to Council (CMR:350:01). Since then, 12 additional informational reports have been submitted (CMR:223:02, CMR: 186:03, CMR:391:03, CMR:491:03, CMR:231:04, CMR:387:04, CMR 484-04, CMR:173:05, CMR: 298:05, 381:05, 443:06, and 395:06). The collection of demographic data is just one part of a comprehensive approach that the City of Palo Alto Police Department has undertaken to enhance police-community relations and ensure policing activities are conducted without racial bias. , The Palo Alto Police Department continues in its effort to provide statistical information on demographic data for enforcement contacts by police officers. Since 2000, the Department has continued to evolve and refine the collection and reporting of this information; changes include the types of information captured, when demographic information is captured, the manner with which the data is recorded and how it evaluates and reports this information. The overarching goal is to provide more meaningful information to the City Council and public. To ensure the Department meets its goal, during this past year the Department has collaborated with the City Auditor to review the data that is being presented in this report. The past quarter, the Department also had a group of Stanford University students who are majoring in Public Policy review and analyze previous demographic reports. CMR:180:07 Page 1 of 8 DISCUSSION Update on Department’s Activities Improving Communi _ty Relations COMMUNITY ACCESS LrNE-- Since implementing the Community Access Line in 2004, the Department has had little activity. Since 2004, only 12 calls have been received. During the past quarter, there were no phone calls received. The line continues to be accessible, is marketed on the Department’s website and is translated into Spanish. Staff continues to inform the community of the existence of this telephone line and encourages members of the public to share its concerns and commendations regarding interactions with the Palo Alto Police Department. FAST FRIENDS PROJECT -- The Police Department continues to work with Stanford University Professor Jennifer Eberhardt on the Fast Friends Project. This police/community relations program involves pairing up officers and community members to do a series of structured activities designed to create a bond between the participants. Pre and post tests are taken by all participants to measure the increase in positive attitudes towards police and towards community members. The Fast Friends Project was part of the 19th Citizens Police Academy, held September 20 -November 15, 2006. The Fast Friends workshop was conducted on October 11th"nd final surveys were completed by email with the participating students. INDEPENDENT REVIEW -- In its continuing efforts to seek input and have an independent review of demographic information, Staff worked with students from the Stanford University Public Policy Program in analyzing data from prior demographic data reports. The project completed by the Stanford students, assessed the Department’s racial analysis profiling methodology, discussed benchmarking, provided an independent analysis of data using a basic statistical technique (probit regression) and ran it with a nested regression model. Finally, the students provided some recommendations to the Department related to racial profiling, demographic data collection and analysis. A copy of the student’s report submitted to their professor is provided in Attachment G. Staff will be reviewing the group’s recommendation for feasibility of implementation. AUTOMATED DATA COLLECTION--During the first quarter, the Department implemented its automated demographic data collection application and officers now input demographic information into the systems via a mobile data computer. The only manual entry process that is required is for traffic officers who have no immediate access to the Department’s automated system. Contact Data In previous Council reports, the data collection process was carefully outlined. For reference purposes, a summary of this information has been provided in Attachment A. A total of 4,647-initiated contacts related to vehicle, pedestrian and bicycle stops were made by officers during the first quarter. During the second quarter, officers made 3,412 self-initiated contacts. CMR:180:07 Page 2 of 8 Of the contacts in both quarters, the Department is reporting demographic information for 98 percent of the contacts. During the first quarter of FY06-07, contacts by officers are up more than 30 percent during the past year (from 3,466 in the fourth quarter FY2005-06 to 4,647 contacts during the first quarter FY2006- 07). The Department saw a return to more consistent numbers of contact in the second quarter FY2006-07 (3,412 contacts). The increase in contacts is directly related to the trend of burglaries in the City and increased criminal activity in East Palo Alto during the same time period. A compilation of various demographic data tables has been provided in Attachment C. RACE - Table 1 provides the summary by race for each of the contacts made during the first and second quarter. GENDER-- Of the 4,647 contacts made during the first quarter, 31 percent (1,429 contacts) were female and 69 percent (3,217 contacts) were male. There was one contact during the first quarter where the gender was not specified. During the second quarter, of the 3,412 contacts, 32 percent (1,084 contacts) were female and 68 percent (2,328 contacts) were male. AGE-Of the 4,647 contacts during the first quarter, nearly half of the contacts (48 percent) were with individuals between the ages of 22 and 40. The second quarter shows a slightly higher percentage of ages 22 to 40 year old individuals (53 percent) of the 3,412 contacts made. Attachment C - Table 2 shows the breakdown by age and race for each contact during both the first and second quarters. There were instances in both quarters where the age of the individual was not captured. CONTACTS DURING NIGHTTIME AND DAYTIME HOURS (TABLE 3) - Officers typically do not know the race, gender, etc., of drivers when they stop an individual for a vehicle code or other violations. This is especially true during hours of darkness. To show the demographic information from an alternative perspective, during the first and second quarters the Department broke down several categories of demographic information by time of day. A further breakdown of various categories by time of day (daytime or nighttime hours) information has been provided in Attachment D. During the first quarter, of the total number of contacts made by officers (4,647 contacts), 44 percent (2,039 contacts) were made during nighttime hours (between 5:01 PM and 6:59 AM) and 66 percent (2,608 contacts) were made during the daytime. For the second quarter, of the total number of contacts made by officers (3,412 contacts), slightly higher than 50 percent (1,721 contacts) were made during nighttime hours (between 5:01 PM and 6:59 AM) and 50 percent (1,691 contacts) were made during the daytime. Throughout the report, the information on day and nighttime comparative data will be referenced in the various category areas. RESIDENCE OF INDIVIDUALS CONTACTED -- Palo Alto serves as a destination point for many individuals traveling into/through the City to work, study (e.g. Stanford University) and for leisure activities. This CMR:180:07 Page 3 of 8 has been verified by the information provided in Table 5. Any comparison to demographic census data should be done with the information from at least the four adjacent counties. Table 4 provides summary percentages of eac-h race taken from the 2000 census data for the four Bay Area counties compared to percentage of enforcement contacts during the third quarter. It is important to recognize that the census data is now over six years old and many changes have occurred in the interim. This table represents updated calculations from prior Council reports (CMR 395:06). These formulas utilize the same methodology used by the City Auditor in the Service Efforts and Accomplishments (SEA) Report. As highlighted in the following sections of the report, there are a number of factors that contribute to why an officer makes contact with an individual. Attachment C - Table 5 shows that for those contacts made as a result of pre-existing knowledge or information, during the first quarter 27 percent of those types of contacts were African-Americans. During the second quarter 29 percent of African-American contacts were for pre-existing information. Pre-existing information includes suspect description from crime reports (Attachment C- Table 6), as well as other all-points bulletins on subjects being sought by other law enforcement agencies or other crime trend information received from law enforcement or court agencies. Table 4 provides detailed information on the census data. To help account for the diversity in its contacts, the Department recorded the city of residence of the people arrested or cited in Palo Alto as part of its demographic collection process. Table 6 provides a summary of those persons contacted who reside in Palo Alto, as well as other adjacent communities within Santa Clara County (Mountain View, Los Altos, Los Altos Hills and Stanford) and San Mateo County (East Palo Alto, Menlo Park and Redwood City). Other city and county areas within San Mateo and Santa Clara Counties include people who live either in other parts of California, those who reside outside of California or abroad. In addition, it provides the percentage for each city of residence broken down by race. Because the Department does not keep track of contacts byname of individuals, it is impossible to determine whether some individuals were contacted on more than one occasion during the quarter. SUSPECT DESCRIPTION -- Officers will make contact with individuals for a number of reasons. One reason is pre-existing information (they match the suspect description in prior or recently reported crimes). Of the 4,647 contacts made by officers during the first quarter and the 3,412 contacts during the second quarter, three percent were a result of pre-existing information. Information on suspects is captured in the Department’s Records Management System (RMS). During the first quarter, citizens reported 208 suspects. Of those, 184 included descriptions of the perceived race of the individual. During the second quarter, citizens reported 245 suspects. Of those, 195 included descriptions of the perceived race of the individual. Table 7 outlines the description for suspects byrace obtained during both the first and second quarter offense reports. In the first quarter, out of the 184 incidents with suspect description including race, the largest majority was 35 percent reported as being White. In the second quarter, of the 195 reported suspects with perceived race, 27 percent were Hispanic. An important caveat is these numbers may reflect the same suspect for multiple offenses, but without the CMR:180:07 Page 4 of 8 actual identity of suspects it is impossible to determine. The Department continues to attempt to gather additional suspect description data from other area law enforcement agencies, as contacts occur as a result of pre-existing information from local area bulletins or crime reports from outside agencies; however, there were not enough other departments with consistent, reliable data to provide a meaningful comparison. It is important to note that citizens, not police officers, typically generate a suspect description. The racial distribution of suspects presented in this report is compiled from these citizen-provided descriptions in police reports. REASON FOR CONTACTS -- In addition to pre-existing information, other reasons officers make contact with subjects include Penal Code violations, equipment or other non-hazardous violations, moving violations or other traffic related hazards, and other Code violations (Municipal, Health and Safety Code). All violations were observed by the officers. It is not surprising that well more than half the contacts (63 percent) are related to moving violations or other traffic-related hazards since traffic has been one of the Council’s priorities for several years. Table 8 provided in Attachment C includes a breakdown by the reason for the contact by race and the associated percentages. For example, there were 393 contacts with Hispanics for Equipment Violations in the first quarter (25 percent). In the first table showing percentages, the numbers reflect those percentages by the reason for contact with officers. For example, of all penal code violations, 18 percent of the contacts were with African Americans and 60 percent were White. The second chart in Table 5 shows the percentage breakdown for each racial group for both the first and second quarters. For example, for all African American contacts during the first quarter, 51 percent were for equipment violations and four percent were for penal code violations. For the second quarter, moving hazards represented the largest number of reason for contacts by police officers (51 percent). In Attachment C, Table 5, the change in the number of contacts and the corresponding percentage changes between the first and second quarters are provided. For example, equipment violation contacts were up for African-Americans by six percent from the first quarter to the second quarter, while they declined by five percent for Hispanics during the same time period. Table 6 provides a comparison of first and second quarter reason for contact information. As previously noted, during the first quarter, the City continued to experience increased burglaries. Although during the second quarter the number of burglaries appeared to return to more consistent numbers, the majority of suspect descriptions provided by a reporting party identified either White or African-American males in the first and second quarters. LOCATION OF CONTACT-- Since the fourth quarter of 2005-06, the City has been using its Geographical Information System (GIS), to break down contacts in four quadrants (Beat 1 through Beat 4). The CMR:180:07 Page 5 of 8 boundaries for each of the four beat areas are provided on a map in Table 8. Of the 4,647 contacts for the first quarter, the Department was able to plot 4,443 of the contacts on a map. The remaining 204 contacts could not be plotted because they were geographically outside the city limits or the location had some other anomaly (e.g., an incomplete or unrecognizable address) that made it impossible to map. During the second quarter, the Department was able to plot 3,238 of the contacts on a map by beat. Beat 2 continues to show the largest numbers of contacts. This is consistent with the overall higher police activity levels in the Downtown and Beat 2 area. Table 9 shows contacts byBeat area. A map of Beats 1 through Beat 4 for each race is provided in Attachment F. DISPOSITION OF THE CONTACT -- In Attachment C, Table 9 provides the disposition for each contact where demographic information was captured during first and second quarters. Citations issued continue to represent the highest percentage of final action taken by officers. Of the 4,647 contacts during the first quarter, 1,200 stops (52 percent) resulted in citations being issued by police officers. Arrests are up from last quarter, two percent in the fourth quarter FY 05-06 to four percent during the first quarter and five percent in the second quarter. Although citations still continue to represent the highest disposition during the first quarter (49 percent) and the second quarter (43 percent), overall the number of citations issued by officers has dropped from 2,726 in the fourth quarter of FY05-06 to 2,287 in the first quarter and 1,477 in the second quarter. Attachment C, Table 10, details the changes in disposition of .contacts and the corresponding percentages between quarters three and four. In Attachment D, the tables show that a majority of contacts (65 percent) result in citations being issued during the daytime hours (1,707 citations) during the first quarter. The Disposition Table goes on to know that for citations issued during the first quarter, White drivers represented the highest percentage of citations issued (54 percent) and of White individuals contacted by officers, 72 percent received a citation. Searches Of the 4,647 contacts during the first quarter, approximately seven percent resulted in a search being conducted (342 searches). For the second quarter, the number of searches rose to eight percent (275 of 3,412 contacts). Table 15 provides a summary of search information for the first and second quarters, and corresponding percentages, are provided in Attachment C. Attachment C, Tables 11 and 12 shows that of the 342 searches that were conducted in the first quarter, 49 percent were required by Department policy as a result of either an officer making an arrest CMR:IS0:07 Page 6 of 8 (24 percent) or vehicle impound inventory (25 percent). Probation/parole situations resulted in 26 percent of all searches. SEARCHES RELATED TO PROBATION/PAROLE -- During the first quarter, 90 searches (26 percent) were the result of conditions of parole/probation for the individuals contacted. In the second quarter, 53 individuals (19 percent) were subjects searched as a result of conditions of their probation or parole. CONSENT AND PROBABLE CAUSE SEARCHES -- Twenty-six percent of all searches conducted were a result of consent or probable cause searches in the first quarter and 25 percent during the second quarter. In past reports, the Department has conducted reported detailed reviews of each probable cause and consent search conducted by officers to provide the Council with the rationale that officers typically use when making contact. In this report, and in future reports, the Department will report these in a table that denotes the nature of the contact. Probable cause searches are the result of reasonable suspicion, officer safety, or plain view of possible contraband. These searches differ from the consent searches because the officers had legal justification to conduct the search even without the consent of the subject. In Table 16 below, probable cause and consent searches are broken down by race and gender for the first and second quarter. In Table 13 (Attachment E), the probable cause and consent searches are broken down by race and by disposition of the contact with the individual for both the first and second quarter. RESOURCE IMPACT Although the process has been streamlined, demographic data collection continues to be labor intensive. Approximately 100 hours of administrative staff time at an approximate cost of $5,000 is spent reconciling the data and preparing analysis of the statistical information, review of the demographic data cards, citations and arrest reports. Staffcontinues to hope the workload will reduce in the future as processes are automated, benchmark criteria are standardized and workload reallocations are considered. ATTACHMENTS Attachment Attachment Attachment Attachment Attachment Attachment Attachment A - Data Collection Process B - 2000 Census Data C - Demographic Data Tables D - Demographic Data Tables by Time of Day E - Demographic Data Tables Consent!Probable Cause F - Maps G - Analysis of Data and Methods Used to Assess Presence of Racial Profiling in Palo Alto Police Department CMR:180:07 Page 7 of 8 PREPARED BY: DEPARTMENT HEAD: SHERYL A/CONTOIS Director, Police Technical Services BURNS C~i, Field ~ervi~I Divisi°~ CITY MANAGER APPROVAL: City Manager CMR:180:07 Page 8 of 8 ATTACHMENT A DEMOGRAPHIC DATA COLLECTION PROCESS Demographic information is captured on all self-initiated enforcement vehicle, pedestrian and bicycle stops. The Department’s Police Records Management System (RMS) captures the following information: race age gender location of stop reason for the contact action taken by the officer (disposition) city of residence for the individual contacted and, whether or not a search of the individual was conducted o If a search is conducted, the officer notes the reason and outcome of the sea}ch An officer must make a reasonable determination of the individual’s race during the contact in lieu of asking the person. The following race categories being used for data collection purposes are consistent with those used by other law enforcement agencies. White African American / Black Hispanic Asian (Includes Other Asian, Chinese, Cambodian, Filipino, Japanese, Korean, Laotian, and Vietnamese) Other (Includes Guamanian, American Indian, Mid-Eastern, Pacific Islander, Samoan, Hawaiian and Unknown) ATTACHMENT B TABLE 4- SUMMARY 2000 CENSUS DATA IAfrican-°1 Hispanic White ! Asian ~o~ Other!i American..................................................... ~ff ....................................r .....................................r ................................~ .........................r ...........................AL~EDA !! 14.6% ]I 19.0% [ 40.9% ! 20.3% ]~ 5.2% 7.6% 3.4% 2.6% 7,3% 12% SAN FRANCISCO SAN MATEO SANTA CLARA PERCENT OF TOTAL PALO ALTO ENFORCEMENT CONTACTS FIRST QUARTER PALO ALTO ENFORCEMENT CONTACTS SECOND QUARTER ![ 21.9% ][ 24.0% [20.4% ii 19% 12% I [~A i 46% 49.8%~.19.8% 44.2%25.4% 44.0%23.8% 47%12% 12% .8% lO% iL .......... ATTACHMENT C TABLE 1- NUMBER OF CONTACTS -- FIRST QUARTER TOTAL 4,647 564 867 2,198 538 480 PERCENT OF TOTAL 100%12%19%47%12%10% NUMBER OF CONTACTS -- SECOND QUARTER ~i African-ti 11 .................... _T_ota__l_ ....... !!..~A~___e..ri_c_~__...~_His__p__a_n~._c_ ._Whi_t.~ ...... ~[’_A.Ii _si_a}q______J_2t_h_e_r____ TOTAL 3,412 424 611 1,568 415 394 PERCENT OF TOTAL 100%12%18%46%12%12% TABLE 2 - AGE OF CONTACTS FIRST QUARTER ...........Total I Under i 22:40 41-60 ...... ........Reported21 ....= ..... ..... .....6a~and ~ NoAge African-1 i i 564 ~61 I 239 i 235 ~ 24.~erica.n .... i ’ : Hlspamc -~-~-~.15.1 .......... ...........54~~ .......155 .................13 ........ [ White 266 ~ 908 =[............................................................................................... :~ ......816 .................[, ..........................199 [Asian...................I ........................................................538 i 4z 282 ~, ...............182 ,~ ...................................26 5 Other ~ 480 ~55 ~ 277 ~ 133 ~, 14 ~ TOTAL [4,647 :i 576 ~ 2250 ~ 1521 ~ 276 TABLE 2 - (CONT’D) AGE OF CONTACTS SECOND QUARTER .... :above:Reported African-i I American ! 424 [44 215 145 16 i Hispanic i 611 ’97 i 403 100 9 2 White 1,568 i[163 i 684 561 146 Asian i 415 !18 i 259 1 115 16 i 7 ~Other ...........394 1 ....38.i _ 238 99 i 16 3 ....... .............TOTAL 3,412 360 i 1799 ,1020 ~ 203 TABLE 3 " CONTACTS MADE DURING NIGHTTIME HOURS - BY RACE FIRST QUARTER TOTAL PERCENT Total 2,039 100% American" li Hispanic ti White ]i Asian I! Other CONTACTS MADE DURING DAYTIME HOURS - BY RACE SECOND QUARTER ]Total I American Hispanic ilWhite 1Asian ~ Other TOTAL 1,721 269 320 771 206 155 PERCENT 100%12%19%45% 12% 9% 00 oo 00 ~ua o~O TABLE 6 - RESIDENCE OF CONTACTS FIRST QUARTER African- Total American Hispanic ~ White Asian PALO ALTO 1,137 118 59 % OF TOTAL 10%5% ADJACENT ~ COMMUNITIES 1,490 i 268 I 419 ’ 557 106 % OF TOTAL i 18%! 28%.................................. OTHER SANTA CLARA COUNTY 1,149 83 234 ~524 165 % OF TOTAL ............7~ .....20%i 46% ....14% ALLOTHER i 871 95 155 406 i 111 °,/o OF TOTAL [~11%18%47%I 12% TOTAL 4,647 .564 -i 867 i 2,198 538 o~ OF TOTAL STOPS BY RACE 100%~.12%19%.47%,12% Other 92 140 ~ 144 13% 104 10% 480 RESIDENCE OF CONTACTS SECOND QUARTER African- ! ~ Total American, Hispanic i White i Asian i Other PALOALTO r 901 72 l 44 i 567 i 128 i 90 : ADJACENT -[ {COMM~ITIES ~1,130 207 325 ~401 {69 ~127 % OF TOTAL ~18%~2 o ~o {o o9~~35~~6~,llVo ~ CLA~CO~TY ~835 81 t 148 ~357 ~150 {99 ~ % OFTOTAL ~, 10%} 18%{43% ....... r..............................~{ ...............~ ..........~ ................r ..............AL~ OTHER 566 ~85 { 83 276 66 ~ 56 ~ % OF TOTAL ~16%"12% [ 17%~ 45% [TOTAL {3,412 424;611 { 1,568~ ....,~ 415 394 ~% OF TOTAL ~~ ~ STO~SBV~ACE ~:~ 100%12%18% , 46%: 0 0 ILl o 0 0 t- ~ ° r- 0 00 w 0 w 0 0 ~- 0 0 TABLE 7 -- PERCENT OF SUSPECT DESCRIPTION BY RACE FIRST QUARTER TOTAL I PERCENT OF TOTAL SUSPECT DESCRIPTION :i African- [ [ Total American:Hispanic ,i White Asian Other i PERCENT OF ~TOTAL ’CONTACTS ............. 184 40 100%~ ~22% t00% 54 65 12 13 29%35% i 7%7% 19%ii 47% 12% 10% PERCENT OF SUSPECT DESCRIPTION BY RACE SECOND QUARTER African- i..........................................T0ta! ................~erlcan.} Otheri ~span!~ ...Wh!te Asian ................ 13i TOTAL f PERCENT OF i TOTAL SUSPECT i’ DESCRIPTION I PERCENT OF TOTAL 195 56 100%29% 100%12% ! .COST, CTS ............ i ................................................... 52 i 65 9 27%i 33%5% 18%46%12% 7% 12% TABLE 8 - CONTACTS BY BEAT FIRST QUARTER BEAT 1 PERCENT BEAT 2 PERCENT BEAT 3 PERCENT Total]!. Americanll Hi~a~n~_j[ White 564 56 87 ’ 309 12% 2,23648% ~l1321 .!I. 443 ii~ 1,020 957 .![ 93 ][ 192 li 43521% _j. Asian j[ Other 64 I[ 48 236 i[ 216 123 i 114. BEAT4 686 t 62 ][95 i[ 349 ti 97 l[ 83 PERCENT 15% ]L~,i!!jl 1 STOPS MADE OUTSIDE CITY BOUNDARIES PERCENT TOTAL PERCENT OF: TOTAL CONTACTS 204 32 50 85 18 19 4% ..... r ~~ CONTACTS BY BEAT SECOND QUARTERAgree. il_~ Total j[American Hispanic White Asian j[ Other ................... t PERCENT 9 PERCENTBEAT2 t,769 { 2431’ 348 ][ 786~[ 203 52%_ .~ i " 189 BEAT3 739 ~ 72 ~ 127 ]~ 344 ]~ 100 i" 96 .......................................................BEAT 4407~{ .....................34 Ft ...............57 ]I~ .................205 ~lC ................60 I[~ ................51 PERCENT 12%~ STOPS MADE OUTSIDE C~TY 174 36 29 70 17 22 BO~DARmS 5% PeRCeNT TOTAL PERCENT OF : o 00 © Ol 0 © 0IE~n TABLE 11 - SEARCH VS. No SEARCH FIRST QUARTER SEARCH No SEARCH TOTAL TotaloAfrican-American i! Other 342 95 12 4,305 469 564 i! Hispanic j[ White li Asian 126 99 10 741 2,099 528 867 468 SEARCH VS. NO SEARCH SECOND QUARTER .........................SEARCHNo SEARCH TOTAL 00 0~ 00 < ~ << ATTACHMENT D I’~00 00 < o o~ 00 w0 Zw o 0I~ 0 00 00 00 W 0 o o ~o ._~ I 00 00 c) oo 00 w W W E~ 0 0 ~0 0 00 0 ~ .~ o~~<oxo~ 00 w rn W c (D ~- .~’© I~- 00 ATTACHMENT E 00 © 00 ~o--e. o--S. o-e. o’.s. o--0. ,~ ,=, o o--, o ~ < 00 << < BEAT 2 ~ ATT+ACHMENT F © Paio Xito Palo Alto Police Department This ~ap is a p~oducl el the The City of Palo Alto Palo Alto Police Department Demographic Data July-September 2006 Beat 1 Race=African American This map is a product of the City of Pale Alto GIS rcervan, 2006d 0-24 1235:51 ]’his document is a graphic roprosen(ation only el bast available sources.Police Beat 1 for ,CA (\~cc~aps~gis$~gis~dmin\Per~nalWeRead.mdb)The City of Pale Alto assumes no responsibiIity for any errors. @1989 to 2006 city of Pale Alto The City of Palo Alto ¢cervan. 2006-10-24 12:56:39 Police Beat I for Hispanic (~l~*maps~is$\gis~adminWersonaRrcewan,mdb) Palo Alto Police Department Demographic Data July-September 2006 Beat 1 Race=Hispanic This map is a product of the City of Palo Alto GIS 0~2067’ This document is a graphic representa0on orgy of best available ~urces, The city of Pale Alto assumes no responsibility for any errors. @1989 to 2006 City of Pale Alto The City of Palo Alto Palo Alto Police Department Demographic Data July-September 2006 Beat 1 Race=White This map is a product of the City of Palo Alto GIS O’2O67’ rce~an, 2006-10-24 13:09:07 This document is a graphic fepresen~t}on only of best avails hie sources.Po~ice Beat 1 for White (l\c¢÷ma ps\gis$~gls~dminWer~na~cervan mdb)The City of Palo Alto assumes no responsibility for any errors, @1989 to 2006 City ef Palo A~Io XN!~X :~N ! /: ........City of palo Alto GIS tt kk ~ ]J )),: ~’~Demogr@hio Data ~~-~’July-September 2006 -.~ Z Rac =As~an _Palo Alto 0~~,, 111 The City or" Palo Alto Palo Alto Police Department Demographic Data July-September 2006 Beat 1 Race=Other This map isa product of the City of Pale Alto GIS 2667’ ~ce~an, 2006-10-24 13:02:08 This document is a graphic repre senta6en only of best available sources.Police Beat f for Other (t’~c=m a p s~jis$\gis~a dmintP e r sona lVce rva n.m db )The City of Pale Alto assumes no responsibility for any errors. ©1989 to 2006 City of Palr~ ABo The City of Pale Alto Pale Alto Police Department Demographic Data July-September 2006 Beat 2 Race=African American This map is a product of the City of Palo Alto GIS O’297T This document is a graphic repre~ntation only of best available ~urces.The City of Pate Alto assumes no responsibility for any errors @1989 to 2006 City of Pale Alto The City of Palo Alto Palo Alto Police Department Demographic Data July-September 2006 Beat 2 Race=Hispanic This map is a product of the City of Palo Alto GIS 0’323ff rcerqan, 2006-10-24 13:40:00 This document is a graphic representation only of best available s, ou~cesPolice Beat 2 Hispanic (\tac-maps~jis$\gis~dmin\PersonaIVce~an.mdb)The Criy of Pale Alto assumes no responsibility for any errors. ©1989 ta 2006 City of Pale Alto L ( P a 1 o A 1 t o r~an. 2006-10-24 13:51:0BPolice Beat B Mite (\~c-maps\gis$~gis~dmin\Per~nal~rce~an>mdb) Pale Alto Police Department Demographic Data July-September 2006 Beat 2 Race=White This map is a product of the City of Pale Alto GIS 0’3230’ This document is a graphic repre~ ntabon only of best available ~urces.The City of Pate Alto assumes no responsibility for any errors. ©1989 to 2006 City of Pale Alto The City of Pale Alto Pale Alto Police Department Demographic Data July-September 2006 Beat 2 Race=Asian This map is a product of the City of Pale Alto GIS 0’323O’ rce~an, 2006+10~24 13:34:27 This document is e graphic representa~on only of best available source.=Police Beat 2 Asian (\\cc÷ma ps\gisStgis~dmin\Pe r ~nal~cervan.mdb)The City of Pale Alto assumes no responsibilit~ [or any errors. @I 989 to 2006 City of Pale Alto The City of Palo Alto Police Beat 2 Other O\cc~laps\gisStgis\adminlPer~naRr ~lvan.mdb Palo Alto Police Department City of PaloAIto GIS Demographic Data B~~ July-September 2006 Beat 2 Race=Othero The City of Pale Alto Pale Alto Police Department Demographic Data July-September 2006 Beat 3 Race=African American This map is a product of the City of Pale Alto GIS 0’2827’ dt~vate, 2006-10-24 13:28:15 This do~ment is a graphic re presentation only of best available ~urce(\~cc-rn a ps~gls$~gis~a dmin\P e r so na P~d ~a va r e mdb)The City of Pale A~to assumes no responsibillby for any errors. ©1909 to 2006 City of Pale Alto The City of Palo Alto Palo Alto Police Department Demographic Data July-September 2006 Beat 3 Race=Hispanic This map is a product of the City of Pale Alto GIS O’2827’ dtavare. 2006-10-2,1 12:35:43 This document is a graphic representation only of best available ~urce s,(\\cc-maps~is$\gis~dmin\motalView, Mdb)The C~ty of Pale Alto a~umes no responsibility for any errors. @1989 to 2006 City of Pale ,aJto The City of Palo Alto dt~vare, 2006-10-24 12:48:36 Palo Alto Police Department city of Palo Alto GIs Demographic Data o~~. July-SeptemberRace=WhiteBeat 3 2006 This document is a graphic repre~ent~on only of best available sources. The City of Polo Alto assumes no responsibility for any enors ©t 989 to 2006 City of Polo Alto The City of Palo Alto Palo Alto Police Department Demographic Data July-September 2006 Beat 3 Race=Asian This map is a product of the City of Palo Alto GIS 0’2827’ dtmva[e, 2006~10~24 1 ~_02:39 This docummn[ is a graphic representation only of b~st available ~urees,{\\cc-ma ps~gis$~Jis~a d minkmeta \View. Md b )3"ha Cib/of PmlO Alto assumes no responsibiIRy for any errors. ©1989 to 2006 Cib/of Palo Alto The City of Palo Alto Palo Alto Police Department Demographic Data July-September 2006 Beat 3 Race=Other This map is a product of the City of Palo Alto GIS O’2827’ drayage; 2006-10-24 12:44:48 This document is a graphic representaSon only of best available sources( \\cc-m a ps~gis$~gls~a dmin’~P e rson a I~d tava re mdb )The City of Pale Alto assumes no responslbi/~ty for any er~ors, ,01989 to 2006 City of Pale Alto The City of Pale Alto This map is a product of the City of Palo Alto GIS Pale Alto Police Department Demographic Data July-September 2006 Race=African American dtavar e. 2006-10-24 14:02:09 This document is a graphic represe ntafion only of best available sources.(\~c-maps\gis$~gis\admin~nle~a\Vmw.Mdb)The City of Pate AJte assumes no responsibility for any errors. ©1989 to 2006 City of Pale PJto The City of , P ale Alto dtavare, 2006-10-24 14;09:26(\\cc-maps\gis$~gis~dmin~ne ta\View.Mdb ) This map is a product of the City of Palo Alto GIS Pale Alto Police Department Demographic Data ~ July-September 2006 Race=Hispanic 254t’ This document is a graphic representa6on only of best available sources, The Cib/of Pale Alto assumes no responsibility for any e#ors, @1989 to 2006 City of Pale Alto The City of Palo Alto This map is a product of the City of Palo Alto GIS Palo Alto Police Department Demographic Data July-September 2006 Race=White 2541’ dtavare. 2006-10-24 14:22107 This document is a graphic representation onty ef best available ~urce$.(\~cc-ma p s~gis$~gis~admin’~rn e ta\View.Mdb)The City of Pa{o Alto assumes no responslbi}ity for any errors. ’#1989 to 2006 City of Palo Alto The City Pale Alto Pale Alto Police Department Demographic Data July-September 2006 Race=Asian This map is a product of the City of Pale Alto GIS O’2541’ dtavare. 2006-18*24 14:05:02 This document +s a gfaphio re preeer~tatJon ordy of best available ~urce s.{\\cc-m a ps\gis$~jis\a dminlme ta \View.M d b )The City of Pale Alto assumes no responsibility for any errors, ’c~t 989 to 2006 City of Pale Alto The City of Palo Alto (\\ccqna ps\gis$\gis~dmin\meta\View.Mdb) Palo Alto Police Department Demographic Data July-September 2006 Race=Other This map is a product of the City of Pale Alto GIS 2541’ This document is a graph}c representation only of best available sources.The City of Pale Alto assumes no responsibility for any errors @1989 to 2006 City of Pale Alto Legend Beat 1 By Race African American The City of Pale Alto Pale Alto Police Department October 1 to December 31,2006 Beat 1 Race: African American This map is a product of the City of Palo Alto GIS O’2292’ tear,an, 2007411-09 13:52:43 This document is a graph© representation only o[ basl available The city of Pale Alto assumes no respnnsibili~ tot any errors ©1989 to 21107 City of Pa© Legend @ Beat i By Ra~e Hi~pani~ The City of Palo Alto This map is a product of the City of Palo Alto GIS Palo Alto Police Department October 1 to December 31,2006 Beat 1 Race: Hispanic 2292’ rcervan, 20074)1439 14:00:13 This document is a graphic representa~on only of best available sources. The City of Pale AJto ~ssumes no responsibility for any errors ©1989 to 2007 City of Pale ,adlo Legend Beat 1 By Race White The City of Palo Alto Palo Alto Police Department October 1 to December 31,2006 Beat 1 Race: White This map is a product of the City of Palo Alto GIS 2292’ rce~’an, 20074)1-09 14:0&01 This document is a graphic represen~fon only of best available sources ]’he City of Palo Alto assumes no responsibility for any errors ©1989 to 2007 City of PaSo Alto Legend i Beat ~ By Race Asian The City of Palo Alto Palo Alto Police Department October 1 to December 31,2006 Beat 1 Race: Asian This map is a product of the City of Pale Alto GIS 2292’ rce~an, 2007-O1-09 13:56:45 This document is a g Fa phic representation only of best available sources. The Cit’/of Pale Alto assumes no responsibility for any errors ©1989 to 2007 City of Pate Alto The City of Palo Alto Palo Alto Police Department October 1 to December 31,2006 Beat 1 Race: Other This map is a product of the City of Palo Alto GtS 2292’ rcervan, 2007-01 ~09 14:04;24 TNs document Is a g~aphic repre~ntatJon only of best available sources The CiV of Palo Alto assumes no r espons~bilr~y [or ~ny errors ©1989 to 2007 City of P~{o Alto Legend Beat 2 By Race African American The City of Palo Alto rcer~an, 2007-01=09 14:16:02 Palo Alto Police Department October 1 to December 31,2006 Beat 2 Race: African American This map is a product of the City of Palo Alto GIS This document is a graphic rep~esenmSon only of best availaNe sources. The City of Pale Alto assumes no respensibil{ty for any errors O1989 to 2007 Cib/of Pale Alto 0’2938’ Legend Beat 2 By Race Hispanic The City of Palo Alto Palo Alto Police Department October 1 to December 31,2006 Beat 2 Race: Hispanic This map is a product of the City of Palo Alto GIS 0’2938’ rce~an, 2007~t ~9 14;22:31 This document is a graphic repre~ntatJon only of best available sources. The City of Pale Alto a~umes no responsibility [or any errors ©1989 to 2007 City of Pale Nto The City of Pale Alto rcer~an, 2007~31-09 14:28:58 Pale Alto Police Department October 1 to December 31,2006 Beat 2 Race: White This map is a product of the City of Pale Alto GIS 0*2938’ This do~ment is a graphic representahon on~y of best available sources The City of Pale Alto assumes no responsibility for any errors ©1989 to 2007 City of Pale Alto Legend o Beat 2 By Race Asian The City of Palo Alto Palo Alto Police Department October 1 to December 31,2006 Beat 2 Race: Asian This map is a product of the City of Palo Alto GIS 2938’ reagan, 20074)14)9 14:19:05 This document is a graphic repre sentat~on only of best availaNe sources. The city of Pale Alto assumes no responsibility for any errors @1989 to 2007 City of Pale Alto Legend Beat 2 By Race Other The City of Palo Alto Palo Alto Police Department October 1 to December 31,2006 Beat 2 Race: Other This map is a product of the City of Palo Alto GIS O’ 2938’ rcewan, 20074~1419 14:25:59 This document is a graphic representaUon only of best available sources, The City of Palo Alto assumes no respons~bilrly for any errors @1989 to 2007 City of Pa!o Alto Legend Beat 3 By Race African American The City of Palo Alto Palo Alto Police Department October 1 to December 31,2006 Beat 3 Race: African American This map is a product of the City of Palo Alto GIS 0’2938’ rcewan, 20074)149 14:55:57 This document is a graphic representabon only of best available sources. The CibJ of PaSo Alto a ssumes no responsibility for any effo~s @1989 to 2007 City of Palo Alto Legend Q Beat 3 By Race Hispanic The City Palo Alto Palo Alto Police Department October 1 to December 31,2006 Beat 3 Race: Hispanic This map is a product of the City of Palo Alto GIS 2938’ tcer~an, 2007~)1~9 15;05:46 This document is a 9raphic ~eptesent~ion only of best available ~urces. The Ci~/of Palo Alto assumes no responsibility for any orrors ©1989 to 2O07 City of Palo Alto Legend Beat 3 By Race White The City of Palo Alto rcervan, 2007-0~ -139 15:12:23 Palo Alto Police Department October 1 to December 31,2006 Beat 3 Race: White This map is a product of the City of Palo Alto GIS 2938’ This document is a graphic rehresentallon only of best available sources. The City of Palo Alto assumes no ~esponsibility for any errors ©t 989 to 2007 cily of Palo Alte Legend ® Beat 3 By Race Asian The City of Palo Alto Palo Alto Police Department October 1 to December 31,2006 Beat 3 Race: Asian This map is a product of the City of Palo Alto GIS rcew~n, 2007~31439 15:03:03 This document is a g~a p hic ~epresent~tion only of best availa hie source s. The City of Palo Alto assumes no ~esponsibility for any errors ©1989 to 2007 City of Palo Alto Beat 3 By Race Other The City of Pale Alto Pale Alto Police Department October 1 to December 31,2006 Beat 3 Race: Other This map is a product of the City of Pale Alto GIS O’2938’ rceman, 2007~1J,)9 15;08:54 This document is a graphic represent~5on only of best available sources. The City of Pale Alto assumes no re sponsibilrty for any errors ©1989 to 2007 City of Pale Alto 172 Legend Beat 4 By Race African American 18 Pale Alto mervan, 2007431 ~9 15:23:18 Pale Alto Police Department October 1 to December 31,2006 Beat 4 Race: African American This map is a product of the City of Pale Alto GIS O’2350’ This document is a graphic representation only o1 best available sources. The City of Paid Alto assumes no responsibility for any errors ©1989 to 2007 City of Pale Alto Legend Beat 4 By Race Hispanic The City of Palo Alto Palo Alto Police Department October 1 to December 31,2006 Beat 4 Race: Hispanic This map is a product of the City of Palo Alto GIS O’2350’ rcewan, 2007~1 ~9 15:25:54 This aocument is a graphic representa~on only of Pest availabte sources. The Cib/of Pale Alto assume s no re sponsibility for an)* enors ©1989 to 2007 C gy of Pale Alto The City of Pale Alto Pale Alto Police Department October 1 to December 31,2006 Beat 4 Race: White This map is a product of the City of Palo Alto GIS 2350’ rce~an, 2007~11 -O9 15:28:48 This document is a graphic represent;t0on only of best availaNe ~uree s. The City of Pale Alto assumes no responsibility for any errors @1989 to 2007 City of Pale Alto Legend Beat 4 By Race Asian The City of Palo Alto Palo Alto Police Department October 1 to December 31,2006 Beat 4 Race: Asian This map is a product of the City of Palo Alto GIS O’2350’ rcervan, 20(]7~01-09 15:24:42 This document is a graphic repre~nta’~on only of best avaflaNe sourc~ s. The Cr~ of Palo Alto assumes no responsibility for any errors ©1989 to 2007 City of P~lo ~to Legend Beat 4 By Race Other The City of Pale Alto Pale Alto Police Department October 1 to December 31,2006 Beat 4 Race: Other This map is a product of the City of Pale Alto GtS 2350’ tee rvan, 20074}14}9 15:27:35 This document is a gfaphic reptesenta~on only of best available seuices. The City of Pale Alto as~mes no ~esponsibllity for any errors ©1989 to 2007 city of Pale Alto ATTACHMENT G AN ANALYSIS OF THE DATA AND METHODS USED TO ASSESS THE PRESENCE OF RACIAL PROFILING IN THE PALO ALTO POLICE DEPARTMENT PREPARED FOR THE PALO ALTO POLICE DEPARTMENT SARAH KAOPUIKI, AUSTIN PHELPS, UDEME UDOFIA MARCH 14, 2007 TABLE OF CONTENTS TITLE PAGE NUMBER EXECUTIVE SUMMARY 2 INTRODUCTION 1. ]PAPD’s Current Racial Profiling Methodology 4 5 RECOMMENDATIONS FOR IMPROVING PAPD’S DATA AND METHODOLOGY FOR ASSESSING RACIAL PROFILING 2. ] Background on the Racial t~rofiling Experts 2. 2 Critique of Current Analysis 2.2J The Importance of Accurate and Complete Raw Data 2. 22 Using Stop Data Strategically-- The Importance of Benchmark Design 2.221 Designing Appropriate Benchmarks 2.2211 Choosing a Benchmark 2.3 Analyzing the Data and Obtaining Results 2.31 Data Interpretation Solution: The Odds-Ratio AN INDEPENDENT ANALYSIS OF THE CURRENT DATA 3.1 The Specifics of our Proposed Model .... 3.2 Conclusions Based on Current Data 3.3 Searches vo Citations 6 7 8 8 11 13 15 16 18 20 22 24 24 OTHER RECOMMENDATIONS RELATED TO RACIAL PROFILING ISSUES 4.1 Recommendations for Community Relations 4.12 Survey Results 4.13 Electronic data collection units 26 26 28 30 5.REPORT CONCLUSIONS AND SUMMARY OF RECOMMENDATIONS 3O 6.WORKS CITED AND CONSULTED 32 APPENDICES: Appendix 1. Summary of results for citation analysis Appendix 2." Graphical depiction of results for citation analysis Appendix 3." Graphical depiction of results for search analysis Appendix 4." Stanford Survey Results 33 34 35 36 ATTACHED: Supplement." Full Results for Citation Analysis EXECUTIVE SUMMARY The Palo Alto Police Department (PAPD) has been collecting and analyzing a substantial amount of data to assess whether its officers have been engaged in racia! profiling. In this report, we first examine ways to strengthen both their data collection methods and the analysis itself. After analyzing the available literature, primarily reports by Lamberth Consulting and Northeastern Institute of Race and Justice, we come to the conclusion that many aspects of the PAPDs data collection scheme could be improved upon. Suggestions include improving the benchmark to which they compare their data by using survey data and including more data that captures officer discretion. We also find that PAPD’s statistical analysis would benefit greatly from using the odds-ratio method, which is outlined in detail in the full report and utilized by Lamberth Consulting. Although we believe the department’s current data is insufficient for making definitive statements about whether racial profiling is occurring, we chose to analyze the data using a different method from PAFD’s method to provide an independent conclusion regarding whether race was a factor in police activity. More specifically, we ran a nested model probit regression to see whether race affected the likelihood that someone would receive a citation. A probit is a predicting model, and the best fit equation, achieved at the end of the process, is able to tell you the likelihood that someone with a particular set of demographic factors will be given a citation. In this case, the variables that were tested were race, age, sex, and area of residence: The results indicated that race was not correlated with the likelihood of receiving a citation, thereby supporting the PAPD’s claim that the current data show a lack of racial profiling within the department. 2 In the last section of the report, we address other issues related to racial profiling. These include actions the police department can undertake to improve community perceptions of the police department on many issues including racial profiling. The suggestions are drawn from multiple sources including a small survey we conducted of Stanford students regarding the Palo Alto Police Department. 1. INTRODUCTION ¯Since 2000, the Palo Alto Police Department, in an effort to determine whether its officers were participating in racial profiling, has been engaged in an extensive program to collect and assess demographic data of the persons with whom they come into contact. Despite the efforts, too many people are unclear on whether or not racial profiling is actually taking place within the Palo Alto Police Department (PAPD), and many community members view the officers with hostility as a result. 1 Although it was not possible to accurately measure the percent of" Palo Alto residents who believe that racial profiling is an issue within the PAPD, nationwide data shows that many Americans believe that racial profiling is a very real issue plaguing a number of" policing efforts. As reported by the American Civil Liberties Union, "A July 2001 Gallup poll reported that 55 percent of‘ whites and 83 percent of blacks believe racial profiling is wideswead".2 It is reasonable to assume that Palo Alto residents share similar sentiments regarding the occurrence of" racial profiling in their community. In fact, in a small, informal survey we conducted of‘ 20 Stanford students, we found that over one-quarter of‘the students believe that the PAPD engages in racial profiling. In order to help increase the community’s faith in the PAPD, it is necessary to analyze the department’s data and methodology used to assess whether racial profiling is occurring in the police force. The results of" this analysis will help PAPD determine what I Through out this report, we rely on the U.S. Department of Justice’s definition of racial profiling. They define it as: "any police-initiated action that relies on the race, ethnicity or national origin rather than the behavior of an individual or information that leads the police to a particular individual who has been identified as being, or having been, engaged in criminal activity." (Ramirez, Deborah. Resource Guide on Racial Profiling Data Collection Systems: Promising Practices and Lessons Learned. November, 2000.) <http ://www.ncjrs.gov/pdffiles I/bia/184768.pdt~ additional steps it should take both to ensure that it is accurately measuring whether racial profiling exists and to relay its findings to the public. To this end, we first discuss the data and methodology employed by the Palo Alto Police Department to assess racial profiling. We then make suggestions for changes in their data collection and their use of benchmarks which wil! improve the accuracy of their analysis; these suggestions are supported by racial profiling experts from Lamberth Consulting and Northeastern University’s Institute on Race and Justice. Following these recommendations, a clear method for data .interpretation will be described. We feel that the combination of better data, benchmarks, and data interpretation will strengthen the PAPD’S quarterly racial profiling reports and will help improve the department’s image within the community. Despite our belief that the ideal way to strengthen the PAPD’S current analysis is to implement these changes, we will also analyze the durrent data using an alternative method to come to an independent conclusion regarding whether race is a factor in policing activities. This is merely a supplemental analysis and does not reflect what we believe to be the ideal model for racial profiling analysis. Lastly, we will make a few other recommendations related to racial profiling, including fostering better community relations. 1.1 PAPD’s Current Racial Profiling Methodology The current methodology employed by PAPD department officials to investigate the issue of racial profiling relies on a large amount data collected by officers, which is used to generate percentages and ratios that illustrate the demographic make-up a ACLU.org. Racial Profiling: Old and New. 6 March 2007. <http ://www.aclu.org/racialj u stice/racialprofiling/index.html> of police contacts. These numbers are then compared to the benchmark--demographic data of Palo Alto gathered from the Census. Using comparison as the primary analysis technique, the police department concludes that there is no racial profiling if racial minorities are not over represented in the contact sample with respect to the minority’s representation in the area’s demographic data. Racial profiling expert, Ron Davis, has reviewed the Palo Alto Police Department’s Report and concluded that the department is not guilty of racial profiling. 2. RECOMMENDATIONS FOR IMPROVING PAPD’s DATA AND METHODOLOGY FOR ASSESSING RACIAL PROFILING In our assessment of previous quarters’ reports prepared by the PAPD, we conclude that the methodology used to provide the evidence against racial profiling in the department does not represent an accurate analysis. Essentially, improvements of raw data as well as benchmarks need to be made. The implementation of these changes will enable the department to utilize a much more accurate means of determining whether racial profiling is a problem within the PAPD and if so, which locations need to be monitored in order to address the issue. "~ In order to validate our initial responses to the past quarters’ reports, we examined methods employed by other police departments ~nonitoring racial profiling. The experts in racial profiling, primarily Lamberth Consulting Company as well as Northeastern University’s Institute on Race and. Justice, have addressed the issues that we found problematic in the PAPDs reports with similar suggestions to those we have devised. Thus, the following analysis is in step with recommendations that have been implemented by racial profiling experts in numerous police departments across the country facing similar issues as the PAPD. 2.1 Background on the Racial Profiling Experts Lamberth Consulting is a professional group dedicated solely to racial profiling "assessment, training, and communication services to universities, states, counties, cities, civil rights groups, litigators, and communities’’3. John Lamberth was the first to design a specific methodology for measuring and analyzing profiling activity in 1993 and has been involved in statistical litigation consulting since 1973. Throughout the past decade, Lamberth has devoted himself to the cause of racial profiling and has perfected his methods for determining whether racial profiling exists, implementing changes in policing methods and improving community-policing relations. On many occasions, Lamberth Consulting works closely with Northeastern University’s Institute on Race and Justice (IRJ), another well known racial profiling resource. The IRJ specializes in utilizing "strategic social science research methodologies to assist government agencies, educational institutions, and members of the community in the development of policy changes that advance the cause of social justice.’’4 It also maintains an extensive and informative online Racial Profiling Data Resource Center. The expertise of these two grdups, has provided significant direction for the following report and analysis of the current practices of the Palo Alto Police Department. 3 Lamberthconsulting.com. Lamberth Consulting. <http://www.lamberthconsulting.com!about- us/index.asp>4 Irj.neu.edu. Northdastern University Institute of Race and Justice. <http ://ww~v.racialprofilinganalysis.neu.edu/index.php> 7 2.2 Critique of Current Analysis From our analysis of literature on analyzing racial profiling, primarily reports by Lamberth Consulting and Northeastern Institute of Race and Justice, we concluded that there are many ways to improve the current methodology employed by the Palo Alto Police Department. The three primary areas that will be addressed in order to improve the accuracy of racial profiling studies in the PAPD are completeness of raw data, accurate benchmarks, and methods used to interpret this data to draw appropriate conclusions. 2.21 The Importance of Accurate and Complete Raw Data At the time of last quarter’s report, the Palo Alto Police Department held each officer individually responsible for gathering the personal information of those that they came into contact with by filling out the "Data Collection Card" provided by the department. Although there was a high rate of compliance, and the majority of the cards were returned, the primary problem with this type of data collection was that it required additional processing to transcribe the officers’ notes into a database that compiles the collected information. This two-step process left a significant amount of room for human error that invalidated otherwise accurate data. The data collected by the PAPD depended on officers to accurately record the information, while making educated guesses about ethnicity, and to have handwriting legible to those conducting the data entry. It was clear that human error could potentially account for a number of inaccuracies in the demographic data of the PAPDs contacts. The department, however, acknowledged this shortcoming and within the past three weeks has transitioned to an electronic method in which the demographic data is 8 automatically transferred to the central data recording system--this method increases the precision of the raw data and significantly eliminates the potential of human error. Ensuring the .accuracy of this stop data is crucial in order to obtain accurate results from the benchmark comparison. Our group is confident that with the transition to automated collection methods, the department’s data will be accurate. Although it is important to h~ive accurate data, it is equally important to have complete information. As a study of the Ann Arbor, Michigan Police Department emphasizes, highly detailed stop data is critical in conducting valid and accurate comparisons with demographic data when determining whether or not racial profiling is taking place. As the report argues, generalizations are not enough when collecting stop data because the specific cross streets and times of day are crucial in malting comparisons with the benclm~arks. Since the transient population varies greatly from street to street depending on types of business, time of day, and proximity to other community services or residential areas, stop data must be detailed in order to conduct the appropriate comparisons. Table 1, which illustrates the extent of the data currently required by the PAPD, contains much of the information necessary to conduct a complete analysis of possible racial profiling activity, but additional informatioh on searches should also be collected. The data collected on searches is currently limited to whether or not one was conducted, for what reasons, and the end result of the search; however, additional data would be helpful in determining whether or not the officer’s actions could have been a result of profiling. 9 Table 1: Information Currently Collected by PAPD Incident #: Data Collection Card Reason for the stop (check one) __ Vehicle code - Moving Hazard __ Vehicle code--Equipment/Reg Violation Penal Code Other Criminal Code __ Pre-Existing Knowledge/Info Location of the stop: Gender of the subject: Age of the subject: M F Race of the subject: (check one) White African American/Black __ Hispanic __ Asian (includes Asian, Chinese, Cambodian, Filipino, Japanese, Korean, Loation and Vietnamese) Unknown City in which subject resides: Disposition of the stop: Cite __ Warning Arrest No action Other Was a search conducted:Yes No If yes, why was a search conducted: (check one) __ Consent (note reason on back) __ Probable cause (note reason on back) Parole/Probation Incident to arrest , Vehicle impound inventory Disposition of search: (was anything located?) Due to the fact that searches are particularly complicated issues because of the number of variables that must be taken into account when determining whether or not profiling is taking place, it is crucial to collect as many variables as possible. John Lamberth emphasized this fact during the interview and suggested a number of variables 10 that prove to be helpful during analysis.5 These included the length of the search, the scope of the search (ie, vehicle, person and/or passengers), and the officer’s action if consent is denied. Finally, it is important that the data reflect whether or not the subject had a criminal record. Since it is mandated that subjects on parole be pulled over and searched, police officers cannot exercise discretion in choosing whether to conduct a search in those cases. Recording this information will prove useful in determining whether racial profiling was used in conducting searches. 2, 22 Using Stop Data Strategically--The Importance of Benchmark Design RegardIess of the accuracy and completeness of stop data, an accurate picture of whether or not racial profiling exists within a police department is not possible without a conclusive standard of comparison. Thus, it is crucial to have an accurate benchmark from which to compare the stop data if an accurate analysis is to be conducted. Currently, the Palo Alto Police Department uses a combined 2000 census data from Alameda, San Francisco, San Mateo and Santa Clara as the comparative benchmark.6 Although last quarter’s report acknowledges the fact that ’°the census data is now over six years old and many changes have occurred in the interim," it justifies this discrepancy because it is ’°the same methodology used by the City Auditor in the Service Efforts and Accomplishments (SEA) Report.’’7 This, however, does not address the fact that using this information as a benchmark does not allow for an accurate interpretation of the stop 5 Lamberth, John. Personal Interview. 5 March 2007. 6 If census data continues to be used as a benchmark, PAPD should consider whether their census benchmark would be more accurate if there were a different grouping or weighting of counties. Such changes might improve the benchmarks since the minority populations and proximity to Palo Alto vary greatly among the counties currently included in the measure. 11 data collected by the officers. Table 2, a chart constructed by Lamberth Consulting for the San Antonio Police Department, illustrates why census data is a fundamentally flawed benchmark by comparing census data to data obtained through the observational method. Essentially, the observational method represents an accurate account of the driving population in a specific area by physically surveying the driving population at specific intersections. This surveying requires individuals to manually record demographic statistics of drivers passing through the intersection and thus provides an accurate account of the transient population. This method will be elaborated on in subsequent sections. Table 2: Traffic versus Census Data 1-410 & Harry Wurzbach S.W. Military &Tacoma; S.W. Military & Pleasanton 9.3%26.1%64.4% 33.7% 76.1%55.7% 5.3%0.7%-371.4%88.4%86.9%-1.7% S.W. Military & Zarzamora 2.4%0.9%-166.7%73.9%91.4%19.1% Guadelupe &S. Zarzamora 2.2%0.8%-175.0%89.3%.96.9%7.8% Wheatley Courts 65.1%61.3%-6.2%33.0%35.1%6.0% This chart shows the discrepancies between the statistics estimated by census data and the demographic statistics collected through the observational survey method, as 7 City of Palo Alto. City Manager’s Report. October 10, 2006. CMR:395:06, page 6 12 represented by the computed disproportion column, ’Comp. Disp’. Not only are there instances of severe overestimating, as seen in the 1-410 & Harry Wurzbach location, but dramatic underestimating as well, as seen in the data from the intersection of Guadelupe& S. Zarzamora. Thus, it is obvious that benchmarks other than pure census data must be designed in order to accurately interpret stop data in a given area. 2.221 Designing Appropriate Benchmarks. Racial profiling experts nation-wide recognize the fact it is difficult to obtain accurate benchmarks and despite continuing efforts, an optimal method has yet to be named. As expressed in the Rhode Island Traffic Stop Statistics Data Collection Study Final Report, "There are several alternatives for benchmarks that researchers have employed to determine racial disparities in traffic stops, but no consensus exists about the most effective and valid benchmark for every type of community".9 For example, there are two types of benchmarks that courts deem valid for litigation purposes, the observational method and internal data. In addition, in an attempt to revolutionize the methods of obtaining an accurate benchmark, a third method has been developed, the Driving Population Estimate. On the assumption that it is possible to obtain accurate driving population information from census data, Northeastern University designed the Driving Population Estimate (DPE).S° It takes into account variables that might attract individual drivers out 8 www.sanantonio.gov/SAPD. San Antonio Police Department. Larnberth Consulting, San Antonio Racial Profiling Data Analysis Study. 15 January 2004. <http://www.smlantonio.gov/SAPD/Prot~lin gReport.asp>9 Rhode Island Traffic Stop Statistics Data Collection Study 2004-2005 Final Report. page 33 10 Farrell, A. The Rhode Island Driving Population Estimate, Confronting Racial Profiling in the 21 st Century: Implications for Racial Justice, Boston, MA; Northeastern University Institute on 13 of surrounding communities into the area in question such as business opportunities and the average commute time of workers as well as "the percent of State employment, percent of State retail trade, percent of State food and accommodation sales, and percent of State average daily road volume." It also incorporates statistics on eligible drivers, average commute times of the population in surrounding areas as well as average travel times. Using a combination of this type, the DPE is then computed through statistical analysis. In theory, this method would be ideal due to its accessibility and minimal costs, but it only provides a citywide DPE rather than "traffic patterns at the neighborhood tevel".1~ Essentially the DPE is adjusted census data and has yet to be perfected for department level analysis. As the DPE has yet to accurately depict neighborhood driving demographics, two traditional methods are widely used due to their acceptability as evidence in litigation. The first is an internal benchmark, which is gathered from officer-level statistics. The officer-level benchmark relies on individual officers’ stop demographic data "as the baseline against which to compare stops over time.’’12 Thus, by comparing individual officer data from quarter to quarter, discrepancies will not only become visible between officers during a specific quarter but wil! also show fluctuations in an individual officer’s statistics over a number of quarters. As a result, the officer-level statistics monitor the possibility that officers are engaging in racial profiling by observing their individual trend over time in addition to comparing it to the statistics of their peers. Obvious strengths of Race and Justice: Boston, MA, 2003.11 Lamberth Consulting and Northeastern University. Practitioners Guide for Addressing Racial Profiling. Spring, 2005. Page 17 <http://www.lamberthconsulting.com/about-racial- profi ling/research-articles .asp> 14 this method are minimal cost, efficiency and officer management. It does not, however, provide department-level information and many officers object to having their data published as a benchmark thus making it uncertain whether or not this benchmark can be used at all. The second option is to utilize an observational survey method, which yields strong measures of the transient driving population and thus serves as a very accurate benchmark. First developed by Dr. John Lamberth in 1993, the "observation method [has not only been] well established in the social sciences community as a valid way to gather data," but has also ~been validated in court cases in New Jersey, Arizona, and Maryland". 13 Although this method does require surveyors to physically collect the demographic data of the driving population, it truly is the only method currently available to obtain an accurate measure of the demographics of the driving population. 2~ 22] l_ Choosing a Benchmark." Ensuring an accurate benchmark requires additional time and resources, but it is impossible for profiling analysis to be informative without them. For example, a report compiled by an outside consulting firm for the Ann Arbor, Michigan Police Department, acknowledges that census data is hardly reliable as a benchmark due to the fact that it is a static measure of household demographics and thus is an inaccurate measure of roadway traffic. The report recommends that direct observations are the most accurate and reliable 12 Lamberth Consulting and Northeastern University. Practitioners Guide for Addressing Racial Profiling. Spring, 2005. Page 19 <http://www.lamberthconsulting.com/about-racial- profiling/research-articles.asp>13 Lamberth Consulting and Northeastern University. Practitioners Guide for Addressing Racial Profiling. Spring, 2005. Page 24 <http://www.lamberthconsulting.com/about-racial- profiling/research-articles.asp> 15 means of obtaining an appropriate benchmark. Essentially, the report underscores the importance of identifying and recording the "racial and ethnic mix of individuals traveling through a locality [by developing a sclaedule] to survey carefully chosen locations according to a randomly selected time schedule.’’14 It is only after sufficient data has been collected by this means that the Lamberth Consulting Group feels an appropriate benchmark, representative of the area’s true. demographic, can be reached and subsequent comparisons can yield insightful results of a department’s racial profiling activities. Ultimately, if the benchmark to which stop data is being compared is not accurate, the entire analysis is inconclusive. Thus, it is our recommendation that the PAPD adopt the observational survey method in order to obtain accurate benchmarks. It is only after the implementation of this benchmark design that the community as well as the police department can be confident that officer’s actions are being accurately monitored. 2.3 Analyzing the Data and Obtaining Results Regardless of the accuracy of a department’s stop data and benchmarks, if an appropriate method of analysis and data interpretation is not employed, the study will fail to yield conclusive results. Currently, the PAPD uses a simple method of statistic comparison. The stop data is turned into percentages based on race and other variables, while the same variables are broken out of the combined 2000 census data. Ignoring the flaws in the data as wel! as the benchmark, this simple comparison method fails to take 14 Larnberth, John C. Ann Arbor Police Department Traffic Stop Data Collection Methods and Analysis Study. February, 2004. Page 8 <http://www.lamberthconsulting.com/about-racial- profiling/research-articles .asp> 16 into account specific locations within the jurisdiction and consequently loses a plethora of valuable information. Additionally, Ron Davis, a racial profiling expert, concluded that %ased on statistics for daytime and nighttime stops, he does not believe Palo Alto police are guilty of racially profiling".15 In addition to several other factors, this is deemed to be a valid benchmark comparison because of the argument that police officers are unable to determine the race and gender of the driver when simply driving behind them, particularly at night. Because there is no discrepancy between the day and night time data, this would support the idea that there is no racial profiling within the department. This reasoning is questionable however, because a fundamental distinction has been omitted--the distinction between whether an officer can determine the race of the driver and whether the officer can determine the race of the driver if he wants to. Dr. Lamberth addr.essed a very interesting point during our interview in relation to this topic of whether it is possible for an officer to determine the race of the drivers when driving behind them at night or during the day. Through his own statistical research, Lamberth has proven that 80% to 90% of the time, a police officer is capable of determining the race of the driver if he wants to andtries.16 It is not only Lamberth’s research that has proven this fact as the New Jersey Court also found similar results from determining whether turnpike troopers could determine the race of drivers despite high speeds and darkness~7. These findings undermine the claim that the small differentiation between the PAPD’s daytime and nighttime stops provide additional support for the idea of no racial 15 Abramson, Mark. Racial profiling report scrutinized. Palo Alto Daily News. Oct 18, 2006 <h ttp://www.paloaltodaitvnews.corn/al~icle/2006-10-18-pa-demographics>i6 Lamberth, John. Personal Interview. 5 March 2007. 17 Lamberth, John. Personal Interview. 5 March 2007. 17 profiling. They also provide further reasoning why alternative methods of analysis are necessary in order to accurately interpret the stop-data collected by the PAPD. 2.31 Data Interpretation Solution: The Odds-Ratio We will offer a simple yet effective way to interpret the data collected by the PAPD in light of the current methodology’s flaws outlined above. For the purposes of this explanation of the Odds-Ratio, assume that the PAPD now has accurate data and benchmarks after the implementation of the previous suggestions. It is best to understand the odds-ratio by the completion of the following sentence: "If you are a Black motorist/pedestrian, you are __ times as likely to be stopped as if you are not a Black motorist/pedestrian",la If no racial profiling is taking place within a department, the odds-ratio will equal 1.0, which can be interpreted as "Black motorists are no more likely to be stopped than nonminority motorists".19 A scale is used to determine whether or not action must be taken in specific locations to address the possibility of racial profiling. As Lamberth Consulting suggests, the scale must be adjusted in certain areas to accurately reflect the demographic make-up of the area, but the general system of analysis is as follows. If the odds-ratio is between 1.0 and 1.5, the conclusion is that there is no racial profiling occurring in that specific location and thus the results are positive. If is Lamberth Consulting. Data Collection and Benchmarking of the Bias Policing Project. Final Report for the Metropolitan Police Department in the District of Columbia. September, 2006. Page 45. < http://www.~amberthc~nsu~ting.c~m/ab~ut-racia~-pr~fi~ing/research-ai~tic~es.asp>19 Lamberth Consulting. Data Collection and Benchmarking of the Bias Policing Project. Final Report for the Metropolitan Police Department in the District of Columbia. September, 2006. Page 45. < http://w~vv.~amberthc~nsu~ting.c~m/ab~ut-racia~-pr~ing/research-artic~es.asp> 18 the odds-ration falls between 1.6 and 2.0, a review of stops is recommended because there is a possibility that unacceptable activity is taking place. It is also possible however, that there are other variables contributing to the higher odds-ratio that will be discovered upon further analysis. Finally, if the odds-ratio is greater than 2.0, a detailed review of stops is highly recommended because this large statistic raises a red flag and represents a strong possibility of foul play. Table 3 represents an example of how Lamberth Consulting effectively organizes this information in a data interpretation that clearly displays policing activities in different locations of the area in question. Table 3: Calculating the Odds-Ratio2° Location = N 1-10 & Hildebrand 1-35 & US 90 1-410 & Perrin Beitel 1-10 & Woodlawn 1-410 & Broadway 1215 276 1567 1215 929 Bench i Stop: Black % " N : 3.9% 2004 4.0% 1912 Stop Diff Odds Black% :- % : Ratio 6.8%2.9% 1.8 4.8%0.8% 1.2 12.8%1714 2!.1%8.3% 1.8 3.8%1652 7.1%3.3%1.9 9.5%422 13.5%4.0% 1.5 The odds-ratio is highly descriptive and transparent while also easy to calculate As evidence by Figure 3, the odds-ratio is simply the percent stopped divided by the percent benchmark. Thus, if the PAPD should choose to revamp their racial profiling 20 www.sanantonio.gov/SAPD. San Antonio Police Department. Lamberth Consulting, San Antonio Racial Profiling Data Analysis Study. 15 January 2004. <http ://wv,~v.sanantonio. aov/SAPD/ProfilingReport.asp> 19 efforts and obtain complete stop data as well as accurate benchmarks, the odds-ratio will prove to be a very clear and accurate method of interpreting their raw data. 3. AN INDEPENDENT ANALYSIS OF THE CURRENT DATA As we discussed, the current data is not sufficient to make definitive statements about whether racial profiling is occurring in Palo Alto. Nonetheless, we used the current data in our own statistical analysis, which relies on a completely new and unrelated framework. Our objective was to come to a separate conclusion that would either help strengthen the Police Department’s claims that it is not racially profiling or raise a red flag that would encourage the speedy implementation of our earlier recommendations. Since these conclusions are only as strong as the data on which they are based, we still urge the PAPD to begin diversifying data collection methods to improve their data. For our analysis of the data, we utilized a basic statistical technique called a probit regression and ran it within a nested regression model. Probits are predicting regressions. They allow you to see the potential outcomes of a certain combination of variables. For example, they can tell you the likelihood of getting a citation if you are a 45 year old white male compared to the likelihood if you are a 45 year old black male. A nested regression model evolves in many steps. You first run each individual variable (e.g. race) to see the effect it has on the dependent variable, the thing you want to predict (e.g. whether an individual receives a citation when stopped by the police). Using this data, you can then add variables one by one and create the ’°best fit" model, the regression that predicts the outcome of a particular situation with the most accuracy. 20 This model takes into consideration interaction between variables (e.g., if age and race are correlated) as well as each variable’s individual effect. It also allows us to see the different weights each variable holds in determining the final outcome (e.g. is age more likely than race to affect whether you receive a ticket). For our regression, we first decided to use citations (i.e. whether or not someone who stopped was given a citation) as the dependent variable in the regression. Our initial decision was based on the PAPD’s claim that it is difficult to identify the demographic factors of an individual when they are driving or biking. Accepting this claim as true, racial profiling could not explain any discrepancies found in the stop rates of different races. In addition, there would be no variation in the dependent variable since everyone would have been stopped, and thus probit analysis could not be used. In regards to searches, we did not believe that we currently had the right data to run an effective and accurate regression model on this variable. After our interim meeting with Chief Johnson, however, we did rerun the model with searches as will be discussed below. The first step to analyzing the data on the instance of citations is to run a regression on just the variable for citations, assuming that no other factor alters the way that officers give tickets or arrest individuals. That basic probit regression will give us a log-likelihood number that’we will use as a basis for comparison when running subsequent regressions. We will then seek to run the rest of the previously identified variables individually and in combination to ultimately find the most accurate model, the model that does the best job of predicting whether someone will receive a citation. Each regression will produce its own individual log-likelihood number. We will then compare this number, and the number of the previous model, to see if the difference 21 holds significance on a Chi-squared table. If it does, this new regression equation is a better fit than the previous one. If it does not, the added variables do not have descriptive power (i.e., they do not help to explain whether someone received a citation), so it is advisable to return to using the previous, simpler equation. Ultimately, through this process the regression that most accurately describes the variables that affect a situation will be identified. Each regression also identifies a weight factor, describing how much each variable affects whether you receive a ctation. This number is the coefficient of the variable within the regression function. The resulting regression will explain, to the best of the ability of the current data, which demographic factors, if any, predict whether or not an individual will receive a citation. The final regression will then allow you to enter demographic data about an individual and find the likelihood of that individual being cited. For example, this regression equation would allow you to see if the likelihood that a 22 year old white male who lives in Palo Alto would receive a citation was higher or lower than that of a 22 year old black male who lives in Palo Alto. 3. l The Specifics of our Proposed Model Using the methodology outlined above, our proposed model for analyzing racial profiling includes race, sex, age and residence as independent variables while using citation as the dependent variable. Although additional variables would potentially prove to be informative, the following analysis is limited by the scope of the data provided by the department as discussed above. Future analyses would ideally include more accurate and complete data. The variables of race, sex, age, and area of residence were selected 22 for a variety of reasons. As the primary concern of the study, race was determined to be the most important factor considered, and is represented in the model by the dummy variable NWHtTE, where NWHITE = 1 if the subject was not white and 0 if he or she were. However, the remaining variables were identified as those that .may have significant impact .on both the distribution of the racial makeup of the area and the likelihood of getting a citation. Gender was included in the model because of the generally accepted notion that women are able to talk their way out of tickets more easily than men in comparable situations. If this were true, the number of women receiving tickets should be lower than that of men. This is important to our study if there are larger percentages of women in certain racial groups that live in the area. This variable is represented in our analysis by the dummy variable MALE, where male -- 1 if the contact was with a male and equals 0 if it was with a female. Age was considered for similar reasons as to why gender was determined to be important. Age can alter the analysis of the issue if a racial group has a larger number of younger or older individuals, and people of a certain age group are more likely to be ticketed by the department. In our regression AGE, as well as AGE^2 were represented as continuous variables. For the residential area variable in this regression, Palo Alto and East Palo Alto are considered to be one unit and their data will be compared to the data for all other areas of residence. If an individual lives in the area, the frequency with which they drive through the area will obviously be much larger than an individual who does not. Therefore, it seems as though it will be more likely for them to be ticketed. The racial makeup of Palo 23 Alto and East Palo Alto will determine which groups are more likely to be in the area and available to be ticketed. The variable RES = 1 for individuals who lived in Palo Alto and East Palo Alto and zero for everyone else. 3.2 Conclusions Based on Current Data Following the methodology outlined above, the regression results showed that the best fit model for the current data is Pr(citation=l) = qS(0.0313AGE - 0.0002AGE^2 - 0.2118MALE - 0.1733RACE - 0.1615RES -0.1468). (For a list ~of the models we considered and a summary of their results, see Appendix 1 .) After substituting values for the variable names, we found that a white male of average age (in this case 38.3-years- old) and average a^2 (1675) who lives in Palo Alto is approximately 63% likely to get a citation if stopped. A non-white person whose demographic data is the same (i.e., non- white male of average age and average a^2 who lives in Palo Alto) is less likely to receive a ticket (57%). For a graphical depiction of these results, see Appendix 2. While we believe in the quality of our statistical analysis and the power of the probit as a predictive mode!, we are by no means recommending this as the optimal form of data collection and analysis when looking at racial profiling. This model can only work within the limitations of the data, and therefore, is only as strong as the data on which it is based. The implementation of our suggestions in the previous sections does not just work toward improving the quality of the available data and methodology but has several positive side effects such as increased visibility in the public view and the ability to pinpoint problem individuals within the department. In this section, we merely set out 24 to provide an alt.ernate way to analyze the data that in the end serves to strengthen the PAPD’s case when presenting to the greater public. 3.3 Searches vs. Citations At our second consultation meeting, Chief Johnson expressed her desire to see the regression equations using searches of a vehicle instead of citations given as the dependent variable. This process also requires some officer discretion, as a search of the vehicle is not always necessary after a traffic stop. The intricacies of required searches made us initially reluctant to take on this additional analysis. However, after both the meeting with Chief Johnson and speaking with John Lamberth, we decided to attempt to run the search data. The potential for racial profiling in searches exists after a stop has already occurred when an officer must decide if he is going to search the vehicle. In the case of parole/probation searches and some kinds of probable cause searches, there is little or no officer discretion in this decision. But in all other types of searches, officers may allow themselves to believe that there is more cause to search the cars of particular kinds of individuals. This opens the door for racial profiling and is what we wanted to address by examining these data. For this analysis we used the same methodology outlined above but substituted searches for citations. We found that only the variables of race and sex were even close to being statistically significant. None of the variables had an effect that we could be 95% confident was not just due to chance (i.e., p_<.05), but race and sex were close enough for us to consider. The final model was Pr(search) = ~(-0.1109MALE - 0.1344NWHITE - 1.5608). Thus the likelihood of being searched if you are male and 25 white is about 4.7% and drops to 3.5% if you are non-white. (For a graphical depiction of these results, see Appendix 3.) Although this is only a 1.2 percentage point difference, it is a significant amount considering that only a small percent of stops result in searches. 4. OTHER RECOMMENDATIONS RELATED TO RACIAL PROFILING ISSUES Improving the data and methodology used to determine whether racial profiling is occurring in PAPD is an essential step in addressing racial profiling concerns among the public. In addition, other actions can be taken to address racial profiling concerns among the public. Improve~t community relations are critical to the process of improving community trust and disseminating evidence that the police department is not racially profiling. 4. l Recommendations for Community Relations An important step for improving community relations regarding racial profiling issues is to provide the punic with a clear analysis of the evidence that they have gathered on the topic. One of the main criticisms by the members of the HRC is that the information given to them is not only overwhelming in volume but also difficult to understand. In order to solve this problem, a 2-3 page summary of the information gathered by the department should be created, given to the HRC, and made available to the public. This report should include a brief discussion of the methodology used by the department, the criteria used, and the main findings of the analysis. A condensed form of the data would make it easier for the public to not only understand the findings of the 26 report but also to read the report in its entirety..A shorter report would hopefully appeal to more individuals and increase the ability to have a productive discussion on the findings. More generally, it is important for the PAPD to be proactive in fostering a positive relationship between the department and Palo Alto as a community. As Dr. John Lamberth expressed in an interview, "policing efforts are most effective and efficient when the community and the police force work together as a team". This is difficult to achieve when there is mutual skepticism and distrust, thus efforts must be made for each side to express their concerns and improve the relationship. Ideally, this improvement will result in the cooperative effort that Dr. Lamberth refers to as ~community oriented policing." The first step that Dr. Lamberth suggests in order to achieve this collaboration and mutual respect is to hold a community police workshop. This workshop allows each side to express their hostilities to an objective observer, like Dr. Lamberth himself. Lamberth expressed that the majority of the time, the community as well as the police department both want the same thing from the other--respect, protection and the ability to do their work in the best manner possible.21 Thus, it is possible to begin a conversation between the two sides explaining why they act the way they do and why they feel the way they do. Opening up this type of discussion in a non-threatening environment makes it possible for the community to better understand the police officers and vice-versa. Following the workshop, an ongoing group is formed to ensure that the two sides continue open dialogues, which helps avoid hostility from occurring. Dr. Lamberth has witnessed the positive impact of these types of discussions in a number of jurisdictions 27 and highly recommends that all police departments engage in open conversations with their community. It is only after each side is understood that relations will start to improve. Therefore, we suggest that the Palo Alto Police Department take similar steps to foster increased trust and collaboration within its community. 4.12 Survey Results We surveyed 20 Stanford students to get a sense of how a small segment of the community views PAPD in general and specifically with respect to racial profiling. We also asked them for suggestions to improve the department’s image overall and specifically with regard to racial profiling. The students live in the same undergraduate dorm and are split evenly by gender. We recognize that our sample is extremely small and unrepresentative but believe the results still provide some interesting and informative findings regarding the relationship between the community and PAPD. The survey was comprised of four questions: whether or not the PAPD racially profiles, the degree to which the student believes the department profiles, how much the student believes the PAPD is interested in the community, and an open-ended question regarding what the department could do in order to improve relations in the community and prove they do not conduct racial profiling. Appendix 4 contains a list of the four questions asked on the written survey and the results for the three closed-ended questions. In response to the yes or no question of whether the PAPD racially profiles, six of the 20 students (30%) answered yes. In response to the second question of how often racial profiling occurs, these six respondents were split among answering "all the time" _~1 Larnbdrth, John. Personal Interview. 5 March 20137. 28 (2 students), "on a regular basis" (3 students) and "sometimes" (1 student). Of the 14 respondents who answered that racial profiling does not exist in the yes or no question, 13 indicated that the occurrence of profiling was "never" or "rare" while the remaining students said it "sometimes" occurs but wrote a note that stated that the department cannot help the fact. These responses make clear that a sizeable portion The third question asked how interested PAPD is in the community of Palo Alto. Respondents could choose from five numeric values indicating a range of interest. Seven of the 20 students (35%) indicated that PAPD either has no interest or is disinterested in the community. The same number of students indicated that PAPD is "involved" in or "very dedicated" to the community. The remaining six students chose the middle response, "interested." Regarding the open-ended question, the students had interesting suggestions for how PAPD could show that’they do not conduct racial profiling. Some students said that the department should write a letter apologizing for previous incidents and stating they will make a renewed effort to make sure racial profiling does not take place. Another idea was for the PAPD to allow members of the community to follow officers while they are on-duty in order to physically show officers’ actions on the job, and how they react to situations that race factors into. Also, students said that if the police department became more amicable to them, they would be more apt to believe that they do not racially profile. Students also had ideas on how the Paid Alto Police Department could improve community relations in general. A majority of students just wanted to have a friendlier officer who they come in contact with. One student wrote that "people do not want to go 29 to the police sometimes because of the vibe they get from them." Students also called for small meetings between officers and students or members of the community to talk about problems. 4.13 Electronic data collection units A more specific way community relations may be able to be improved is through changes in the use of the use of the new electronic data collection units used by the PAPD. The Mobile Audio Visual systems are a great addition to the department and help to show that the interactions between the department’s officers and civilians are within guidelines. However, the cameras are only recording during set period of times (when the sirens are turned on, and when the officer manually has the camera recording). During all other times, the cameras are put on a loop back at the server at the PAPD which is constantly overwriting the videotape. If the department was willing to invest more funding to increase the data storage of the servers, there would be constant surveillance of the officers and their activities, which would further alleviate the demands of the HRC and the public for evidence that the police are not guilty of racial profiling. 5. REPORT CONCLUSIONS AND SUMMARY OF RECOMMENDATIONS As outlined in this report, there are three primary areas of the current racial profiling methodology utilized by the Palo Alto Police Department that must be addressed. First, the accuracy and completeness of raw stop data must be guaranteed in 30 order to obtain valid results from data interpretation efforts. Second, it is imperative the benchmark used as a baseline comparison of the stop data is an accurate representation of the transient population driving in Palo Alto. The only way to achieve a truly accurate measure is by employing an observational survey method. Finally, once the data and benchmark recommendations have been instituted, this report suggests that odds-ratios are used in order to interpret t.he data. While our statistical analysis does support the claim of the department that there is not currently a problem with racial profiling, the quality of the data on which the statistics are based must be improved to ensure accurate results. However, there is still value for the department in the analysis in that it provides an alternative way to look at the data for both the PAPD and its critics. The recommendations that we have outlined in this report are in no way all encompassing, but they are an important step towards informing the greater community of the lack of racial profiling within the PAPD. The dialogue between the police and the community, the increased efficiency of MAV, and the simplification of the reports to the general public will all help the department move in that direction. By following the steps outlined in this report, the Palo Alto Police Department will be better equipped to improve relations with the community as well as provide accurate and clear data to the community about the issue of racial profiling. 31 6. WORKS CITED AND CONSULTED ACLU.org. Racial Profiling: Old and New. 6 March 2007. <http://w’a~w.aclu. org/racialj ustice/racialprofiling/index.html> Abramson, Mark. Racial profiling report scrutinized. Palo Alto Daily News. Oct 18, 2006 <http :/!w~w.paloaltodailvnews. co~rdarticle/2006-10-18-pa-demo graphics> City of Palo Alto. City Manager’s Report. October 10, 2006. CMR:395:06, page 6 Farrell, A. The Rhode Island Driving Population Estimate, Confronting Racial Profiling in the 21 st Century: Implications for Racial Justice, Boston, MA; Northeastern University Institute on Race and Justice: Boston, MA, 2003. Irj.neu.edu. Northeastern University Institute of Race and Justice. <http://www.racialprofilinganalysis.neu.edu/index.php> Lamberthconsulting.com. Lamberth Consulting. <http://www.lanaberthconsutting.com/about-us/index.asp> Lamberth, John. Personal Interview. 5 March 2007. Lamberth, John C. Arm Arbor Police Department Traffic Stop Data Collection Methods and Analysis Study. February, 2004. Page 8 <http://www.~amberthc~nsu~ting.c~m/ab~ut-racia~-pr~fi~ing/research-artic~es.asp> Lamberth Consulting. Data Collection and Benchmarking of the Bias Policing Project. Final Report for the Metropolitan Police Department in the District of Columbia. September, 2006. Page 45. < http://www.lamberthconsulting.com/about-racial- profiling/research-articles .asp> Lamberth Consulting and Northeastern University. Practitioners Guide for Addressing .Racial Profiling. Spring, 2005. Page 17 <http://www.lamberthconsultin~.com/about- racial-profiling/research-articles.asp> Ramirez, Deborah. Resource Guide on Racial Profiling Data Collection Systems: Promising Practices and Lessons Learned. November, 2000. <http://wavw.ncirs. gov/pdffiles 1/bj a/184768.pdf> Rhode Island Traffic Stop Statistics Data Collection Study 2004-2005 Final Report. page 33 ww~v.sanantonio.gov!SAPD. San Antonio Police Department. Lamberth Consulting, San Antonio Racial Profiling Data Analysis Study. 15 January 2004. <http ://~w~v.sanantonio. gov/SAPD/ProfilingReport.asp> 32 Appendix 1: Summary of results for citation analysis Prob(Citation= 1) Variables Log Li keli hood Test LRT DF P-value from ChiSquare Significantly different? Model 0 age, age2 -2488.02 Model 1 age, age2, sex -2476.6 Ivs, 0 22.84 0.0000018 Yes Coefficients from Model 3 Age 0.0313 Age2 -0.0002 Sex -0.2118 Race -0.1733 Res -0.1615 InLercept -0.1468 Value of Z Index Probability of Citation Model 2A age, age2,sex, race -2467.8 2A vs. 1 17.6 0.0000273 Yes Mean Age, White1 Resident 38.253 1675 1 0 1 1 0,3422189 0.633906919 Model 2B age, age2,sex, res -2469 2Bvs. $ 15.2 1 0.0000967 Yes Mean Age, Non- white, Male & Resident 38.253 1675 1 1 1 1 0.1689189 0.567069784 Hodel 3 age, age2,sex, race, res -2460.6 3 vs 2A 14.4 0.0001478 Yes FINAL MODEL 33 Appendix 2: Graphical depiction of results for citation analysis Probability of receving a citation White Prob ~ ~ Non-White ProbI >,0.5 ~ 0.4 0.2 0.1 1 7 13 19 25 31 37 43 49 55 61 67 73 79 85 91 97 Age 34 Appendix 3: Graphical depiction of results for search analysis 0.05 0.045 0.04 0.035 0.03 0.025 0.02 0.015 0.01 0.005 0 Probability of being searched Irqwhite male ~.i~ non-white maleI :..: .:...::<.~:-: .:~.:.~:..~..::..::...:. ============================= 1 35 Appendix 4: Stanford Survey Results Respondents were 20 Stanford undergraduate students (10 men, 10 women) from one dormitory. The survey was conducted in March 2007. 1.Does the department racially profile? Yes - 6 No - 14 2. On a scale of 1-5 what do you believe the scale of racially profiling lies within the Palo Alto Police Department? Never - 10 Rare - 3 Occurs sometimes - 2 Occurs on a regular basis - 3 Occurs all the time - 2 3. What do you believe the interest level of the PAPD is concerning the community of Palo Alto on a scale of 1-5? No interest - 2 Disinterested - 5 Interested - 6 Involved - 4 Very dedicated - 3 4. How do you think the PAPD can rectify these situations within the community and show they do not racially profile and they care about community relations? 36 The SAS System Variables: The NWHITE 16:59 Wednesday, CORR Procedure MALE AGE February 21, 2007 RES CITE Variable NWHITE MALE AGE RES CITE Simple Statistics Mean Std Dev 4135 0.46360 0.49873 4135 0.64522 0~47850 4135 38.25127 14.56525 4135 0.39927 0.48981 4135 0.68585 0.46423 S ~]m 1917 2668 158169 1651 2836 Minimum o o o o o Maximum 1.00000 1.00000 99.00000 1.00000 1.00000 NWHITE MALE AGE RES CITE NWHITE Pearson 1.00000 0.05383 0.0005 -0.17371 <.0001 Correlation Coefficients, N Prob > Irl under H0: Rho=0 MALE AGE 0.05383 -0.17371 0.0005 <.0001 1.00000 -0.03430 0.0274 -0.03430 1.00000 0.0274 -0.00338 -0.06942 0.05931 0.8282 <.0001 0.0001 = 4135 RES -0.00338 0.8282 -0.06942 <.0001 0.05931 0.0001 1.00000 -0.09798 -0.07825 0.19910 .-0.04611 <.0001 <.0001 <.0001 0.0030 CITE -0.09798 <.0001 -0.07825 <.0001 0.19910 <.0001 -0.04611 0.0030 i.ooooo Probit Procedure Model Information Data Set Dependent Variable Number of Observations Name of Distribution Log Likelihood WORK.SWITCHED oppCITE 4135 Normal -2553.712999 Number of Observations Read Number of Observations Used Class Level Information Name Levels Values oppCITE 2 0 1 4135 %135 RESponse Profile Ordered Total Value oppCITE Frequency 1 0 2836 2 1 1299 PROC PROBIT is modeling the probabilities of levels of oppCITE having LOWER Ordered Values in the RESponse profile table. Algorithm converged. Effect NWHITE Type III Analysis of Effects Wald DF Chi-Square Pr > ChiSq 1 39.5770 <.0001 Parameter Intercept NWHITE Analysis of Parameter Estimates Standard 95% Confidence Chi- DF Estimate Error Limits Square Pr > ChiSq 1 0.6072 0.0285 0.5514 0.6630 454.71 <.0001 1 -0.2569 0.0408 -0.3369 -0.1769 39.58 <.0001 Probit Procedure Probit Model in Terms of Tolerance Distribution MU SIGMA 2.36342709 3.89248109 MU SIGMA Estimated Covariance Matrix for Tolerance Parameters MU SIGMA 0.095356 0.184632 0.184632 0.382834 Probit Procedure Model Information Data Set Dependent Variable Number of Observations Name of Distribution Log Likelihood WORK.SWITCHED oppCITE 4135 Normal -2542.398093 Number of Observations Read Number of Observations Used Class Level Information Name Levels Values oppCITE 2 0 1 4135 4135 RESponse Profile Ordered Total Value oppCITE Frequency 1 0 2836 2 1 1299 PROC PROBIT is modeling the probabilities of levels of oppCITE having LOWER Ordered Values in the RESponse profile table. Algorithm converged. Type III Analysis of Effects. Wald Effect DF Chi-Square NWHITE 1 36.5246 MALE 1 22.4715 Pr > ChiSq <.0001 <.0001 Parameter Intercept NWHITE MALE Analysis of Parameter Estimates Standard 95% Confidence DF Estimate Error Limits Chi- Square Pr > ChiSq 1 0.7378 0.0399 0.6596 0.8160 342.03 <.0001 1 -0.2475 0.0410 -0.3278 -0.1672 36.52 <.0001 1 -0.2055 0.0433 -0.2904 -0.1205 22.47 <.0001 4 Probit Procedure Model Information Data Set Dependent Variable Number of Observations Name of Distribution Log Likelihood WORK.SWITCHED oppCITE 4135 Normal -2478.201198 Number of Observations Read Number of Observations Used 4135 4135 Class Level Information Name Levels Values oppCITE 2 0 1 RESponse Profile Ordered Total Value oppCITE Frequency 1 2 0 1 2836 1299 PROC PROBIT is modeling the probabilities of levels of oppCITE having LOWER Ordered Values in the RESponse profile table. Algorithm converged. Type III Analysis of Effects Wald Effect DF Chi-Square NWHITE 1 18.5728 AGE 1 144.6177 Pr > ChiSq <.0001 <.0001 Parameter Intercept NWHITE AGE Analysis of Parameter Estimates Standard 95% Confidence DF Estimate Error Limits Chi- Square Pr > ChiSq 1 -0.1095 0.0653 -0.2374 0.0184 2.82 0.0933 1 -0.1800 0.0418 -0.2618 -0.0981 18.57 <.0001 1 0.0183 0.0015 0.0153 0.0213 144.62 <.0001 Probit Procedure Model Information Data Set Dependent Variable Number of Observations Name of Distribution Log Likelihood WORK.SWITCHED oppCITE 4135 Normal -2549.357143 Number of Observations Read Number of Observations Used Class Level Information Name Levels Values oppCITE 2 0 1 4135 4135 RESponse Profile Ordered Total Value oppCITE Frequency 1 0 2836 2 1 1299 PROC PROBIT is modeling the probabilities of levels of oppCITE having LOWER Ordered Values in the RESponse profile table. Algorithm converged. Type III Analysis of Effects Wald Effect DF Chi-Square NWHITE 1 39.5372 RES 1 8.7215 Pr > ChiSq <.0001 0.0031 Parameter Intercept NWHITE RES Analysis of Parameter Estimates Standard 95% Confidence DF Estimate Error Limits Chi- Square Pr > ChiSq 1 0.6569 0.0332 0.5919 0.7220 391.92 <.0001 1 -0.2570 0.0409 -0.3371 -0.1769 39.54 <.0001 1 -0.1227 0.0416 -0.2041 -0.0413 8.72 0.0031 Probit Procedure Model Information Data Set Dependent Variable Number of Observations Name of Distribution Log Likelihood WORK.SWITCHED oppCITE 4135 Normal -2560.687608 Number of Observations Read Number of Observations Used 4135 4135 Class Level Information Name Levels Values oppCITE 2 0 1 RESponse Profile Ordered Total Value oppCITE Frequency 1 0 2836 2 1 1299 PROC PROBIT is modeling the probabilities of levels of oppCITE having LOWER Ordered Values in the RESponse profile table. Algorithm converged. Effect MALE Type III Analysis of Effects Wald DF Chi-Square Pr > ChiSq 1 25.5016 <.0001 Parameter Intercept MALE Analysis of Parameter Estimates Standard 95% Confidence Chi- DF Estimate Error Limits Square Pr > ChiSq 1 0.6275 0.0352 0.5586 0.6964 318 24 <.0001 1 -0.2180 0.0432 -0.3026 -0.1334 25.50 <.0001 Probit Procedure Probit Model in Terms of Tolerance Distribution MU SIGMA 2.87872223 4.58763286 Estimated Covariance Matrix for Tolerance Parameters MU SIGMA MU 0.201078 0.398408 SIGMA 0.398408 0.825295 Probit Procedure Model Information Data Set Dependent Variable Number of Observations Name of Distribution Log Likelihood WORK.SWITCHED oppCITE 4135 Normal -2476.146721 Number of Observations Read Number of Observations Used 4135 4135 Class Level Information Name Levels Values oppCITE 2 0 1 RESponse Profile Ordered Total Value oppCITE Frequency 1 2 0 1 2836 1299 PROC PROBIT is modeling the probabilities of levels of oppCITE having LOWER Ordered Values in the RESponse profile table. Algorithm converged. Type III Analysis of Effects ¯ Wald Effect DF Chi-Square MALE 1 22.5163 AGE 1 160.9896 Pr > ChiSq <.0001 <.0001 Parameter Intercept MALE AGE Analysis of Parameter Estimates Standard 95% Confidence DF Estimate Error Limits Chi- Square Pr > ChiSq 1 -0.0893 0.0659 -0.2186 0.0399 1.84 0.1755 1 -0.2080 0.0438 -0.2939 -0.1221 22.52 <.0001 1 0.0191 0.0015 0.0161 0.0220 160.99 <.0001 9 Probit Procedure Model Information Data Set WORK.SWITCHED Dependent Variable oppCITE Number of Observations 4135 Name of Distribution Normal Log Likelihood -2555.174865 Number of Observations Read Number of Observations Used 4135 4135 Class Level Information Name Levels Values oppCITE 2 0 1 RESponse Profile Ordered Total Value oppCITE Frequency 1 2 0 2836 1 1299 PROC PROBIT is modeling the probabilities of levels of oppCITE having LOWER Ordered Values in the RESponse profile table. Algorithm converged. Type III Analysis of Effects Wald Effect DF Chi-Square MALE 1 27.7420 RES 1 ii.0361 Pr > ChiSq <.0001 0.0009 Parameter Intercept MALE RES Analysis of Parameter Estimates Standard 95% Confidence DF Estimate Error Limits Chi- Square Pr > ChiSq 1 0.6903 0.0401 0.6118 0.7689 296.88 1 -0.2282 0.0433 -0.3131 -0.1433 27.74 1 -0.1382 0.0416 -0.2198 -0.0567 11.04 <.0001 <.0001 0.0009 ]0 Probit Procedure Model Information Data Set Dependent Variable Number of Observations Name of Distribution Log Likelihood WORK.SWITCHED oppCITE 4135 Normal -2487.487671 Number of Observations Read Number of Observations Used 4135 4135 Class Level Information Name Levels Values oppCITE 2 0 1 RESponse Profile Ordered Total Value oppCITE Frequency 1 2 0 1 2836 1299 PROC PROBIT is modeling the probabilities of levels of oppCITE having LOWER Ordered Values in the RESponse profile table. Algorithm converged. Effect AGE Type III Analysis of Effects Wald DF Chi-Square Pr > ChiSq 1 163.7657 <.0001 Parameter Intercept AGE Analysis of Parameter Estimates Standard 95% Confidence Chi- DF Estimate Error Limits Square Pr > ChiSq 1 -0.2307 0.0588 -0.3459 -0.1155 15.41 <.0001 1 0.0192 0.0015 0.0163 0.0221 163.77 <.0001 ]] Probit Procedure Probit Model in Terms of Tolerance Distribution MU SIGMA 12.0145378 52.0753963 MU SIGMA Estimated Covariance Matrix for Tolerance Parameters MU SIGMA 4.866535 -7.838626 -7.838626 16.559306 Probit Procedure Model Information Data Set Dependent Variabl°e Number of Observations Name of Distribution Log Likelihood WORK.SWITCHED oppCITE 4135 Normal -2480.871073 Number of Observations Read Number of Observations Used 4135 4135 Class Level Information Name Levels Values oppCITE 2 0 1 RESponse Profile Ordered Total Value oppCITE Frequency 1 2 0 1 2836 1299 PROC PROBIT is modeling the probabilities of levels of oppCITE having LOWER Ordered Values in the RESponse profile table. Algorithm converged. Type III Analysis of Effects ¯ Wald Effect DF Chi-Square AGE 1 167.9678 RES 1 13.2453 Pr > ChiSq <.0001 0.0003 Parameter Intercept AGE RES Analysis of Parameter Estimates Standard 95% Confidence DF Estimate Error Limits Chi- Square Pr > ChiSq 1 -0.1790 ~0.0605 -0.2975 -0.0605 8.77 1 0.0195 0.0015 0.0165 0.0224 167.97 1 -0.1534 0.0422 -0.2360 -0.0708 13.25 0.0031 <.0001 0.0003 ]3 Probit Procedure Model Information Data Set Dependent Variable Number of Observations Name of Distribution Log Likelihood WORK.SWITCHED oppCITE 4135 Normal -2569.161258 Number of Observations Read Number of Observations Used 4135 4135 Class Level Information Name Levels Values oppClTE 2 0 1 Ordered Value 1 2 RESponse Profile Total oppCITE Frequency 0 2836 1 1299 PROC PROBIT is modeling the probabilities of levels of oppCITE having LOWER Ordered Values in the RESponse profile table. Algorithm converged. Effect RES Type III Analysis of Effects Wald DF Chi-Square Pr > ChiSq 1 8.7619 0.0031 Parameter Intercept RES Analysls of Parameter Estimates Standard 95% Confidence DF Estimate Error Limits Chi- Square Pr > ChiSq 1 0.5339 0.0265 0.4820 0.5858 406.13 1 -0.1225 0.0414 -0.2037 -0.0414 8.76 <.0001 0.0031 ]4 Probit Procedure Probit Model in Terms of Tolerance Distribution MU SIGMA 4.3568705 8.16017505 Estimated Covariance Matrix for Tolerance Parameters MU SIGMA MU 1.805914 3.676241 SIGMA 3.676241 7.599747 Probit Procedure Model Information Data Set Dependent Variable Number of Observations Name of Distribution Log Likelihood WORK.SWITCHED oppCITE 4135 Normal -2460.143432 Number of Observations Read Number of Observations Used 4135 4135 Class Level Information Name Levels Values oppCITE 2 0 1 RESponse Profile Ordered Total Value oppCITE Frequency 1 2 0 1 2836 1299 PROC PROBIT is modeling the probabilities of levels of oppCITE having LOWER Ordered Values in the RESponse profile table. Algorithm converged. Type III Analysis of Effects. Wald Effect DF Chi-Square Pr > ChiSq NWHITE 1 16.2201 <.0001 MALE 1 23.0139 <.0001 AGE 1 147.1631 <.0001 RES 1 15.3630 <.0001 Parameter Intercept NWHITE MALE AGE RES Analysis of Parameter Estimates Standard 95% Confidence DF Estimate Error Limits Chi- Square Pr > ChiSq 1 0.0830 0.0732 -0.0604 1 -0.1689 0.0419 -0.2511 1 -0.2116 0.0441 -0.2981 1 0.0185 0.0015 0.0155 1 -0.1663 0.0424 -0.2495 0.2264 1.29 0.2567 -0.0867 16.22 <.0001 -0.1252 23.01 <.0001 0.0215 .147.16 <.0001 -0.0832 15.36 <.0001 Probit Procedure Model Information Data Set Dependent Variable Number of Observations Name of Distribution Log Likelihood WORK.SWITCHED oppCITE 4135 Normal -2536.985892 Number of Observations Read Number of Observations Used 4135 4135 Class Level Information Name Levels Values oppCITE 2 0 1 RESponse Profile Ordered Total Value oppCITE Frequency 1 2 0 1 2836 1299 PROC PROBIT is modeling the probabilities of levels of oppCITE having LOWER Ordered Values in the RESponse profile<table. Algorithm converged. Type III Analysis of Effects Wald Effect DF Chi-Square Pr > ChiSq NWHITE 1 36.3253 <.0001 MALE 1 24.5566 <.0001 RES 1 10.8353 0.0010 Analysis of Parameter Estimates Standard 95% Confidence Parameter DF Estimate Error Limits Intercept 1 0.7999 0.0443 NWHITE 1 -0.2471 0.0410 MALE 1 -0.2155 0.0435 RES 1 -0.1374 0.0417 Chi- Square Pr > ChiSq 0.7132 0.8867 326.65 <.0001 -0.3274 -0.1667 36.33 <.0001 -0.3008 -0.1303 24.56 <.0001 -0.2193 -0.0556 10.84 0.0010 ]7 The SAS Variable nwhite male age res search nwhite male System 5 Variables: 17:07 Monday, The CORR Procedure nwhite male age res Simple Statistics Mean Std Dev Sum 4136 0.46349 0.49873 1917 4136 0.64531 0.47848 2669 4136 38.25411 14.56464 158219 4136 0.39918 0.48979 1651 4136 0.04570 0.20885 189.00000 March 5, 2007 I Pearson Correlation Coefficients, N = 4136 Prob > Irl under H0: Rho=0 nwhite male age res 1.00000 0.05365 -0.17386 -0.00319 0.0006 <.0001 0.8374 0.0423 0.05365 0.0006 1.00000 -0.03416 -0.06956 0.0281 <.0001 age -0.17386 -0.03416 1.00000 0.05914 <.0001 0.0281 0.0001 res search -0.00319 -0.06956 0.05914 1.00000 0.8374 <.0001 0.0001 -0.03158 -0.02653 0.01248 -0.00105 0.0423 0.0880 0.4223 0.9461 search Minimum o o o o o Maximum 1.00000 1.00000 99.00000 1.00000 1.00000 search -0.03158 -0.02653 0.0880 0.01248 0.4223 -0.00105 0.9461 i.oo0o0 ]8 Probit Procedure Model Information Data Set Dependent Variable Number of Observations Name of Distribution Log Likelihood WORK.SWITCHED osearch 4136 Normal -765.7357824 Number of Observations Read Number of Observations Used 4136 4136 Class Level Information Name Levels Values osearch 2 0 1 Response Profile Ordered Total Value osearch Frequency 1 2 0 1 189 3947 PROC PROBIT is modeling the probabilities of levels of osearch having LOWER Ordered Values in the response profile table. Algorithm converged. Effect nwhite Type III Analysis of Effects Wald DF Chi-Square Pr > ChiSq 1 4.1292 0.0421 Parameter Intercept nwhite Analysis of Parameter Estimates Standard 95% Confidence Chi- DF Estimate Error Limits Square Pr > ChiSq 1 -1.6274 0.0443 -1.7143 -1.5405 1346.88 <.0001 1 -0.1397 0.0688 -0.2745 -0.0050 4.13 0.0421 ]9 Probit Procedure Probit Model in Terms of Tolerance Distribution MU SIGMA -11.646307 7.15634956 MU SIGMA Estimated Covariance Matrix for Tolerance Parameters MU SIGMA 35.294164 -20.904785 -20.904785 12.402608 2O Probit Procedure Model Information Data Set Dependent Variable Number of Observations Name of Distribution Log Likelihood WORK.SWITCHED osearch 4136 Normal -766.3926291 Number of Observations Read Number of Observations Used Class Level Information Name Levels Vakues osearch 2 0 1 4136 4136 Response Profile Ordered Total Value osearch Frequency 1 2 0 1 189 3947 PROC PROBIT is modeling the probabilities of levels of osearch having LOWER Ordered Values in the response profile table. Algorithm converged. Effect male Type III Analysis of E~fects Wald DF Chi-Square Pr > ChiSq 1 2.8767 0.0899 Parameter Intercept male Analysis of Parameter Estimates Standard 95% Confidence Chi- DF Estimate Error Limits Square Pr > ChiSq 1 -1.6149 0.0541 -1.7209 -1.5089 891.32 <.0001 1 -0.1177 0.0694 -0.2537 0.0183 2.88 0.0899 2] Probit Procedure Probit Model in Terms of Tolerance Distribution MU SIGMA -13.723203 8.49803868 HU SIGMA Estimated Covariance Matrix for Tolerance Parameters MU SIGMA 71.477524 -42.335796 -42.335796 25.104394 Probit Procedure Model Information Data Set Dependent Variable Nimber of Observations Name of Distribution Log Likelihood WORK.SWITCHED osearch 4136 Normal -767.5032442 Number of Observations Read Number of Observations Used 4136 4136 Class Level Information Name Levels Values osearch 2 0 1 Response Profile Ordered Total Value osearch Frequency 1 2 0 1 189 3947 PROC PROBIT is modeling the probabilities of levels of osearch having LOWER Ordered Values in the response profile table. Algorithm converged. Effect age Type III Analysis of Effects Wald DF Chi-Square Pr > ChiSq 1 0.6342 0,4258 Parameter Intercept age Analysis of Parameter Estimates Standard 95% Confidence Chi- DF Estimate Error Limits Square Pr > ChiSq 1 -1.7584 0.0950 -1.9445 -1.5722 342.76 <.0001 1 0.0018 0.0023 -0.0027 0.0063 0.63 0.4258 23 Probit Procedure Probit Model in Terms of Tolerance Distribution MU SIGMA 965.099801 548.853212 MU SIGMA Estimated Covariance Matrix for Tolerance Parameters MU SIGMA 1353338.5616 801668.60963 801668.60963 475000.47693 24 Probit Procedure Model Information Data Set Dependent Variable Number of Observations Name of Distribution Log Likelihood WORK.SWITCHED osearch 4136 Normal -767.8164573 Number of Observations Read Number of Observations Used 4136 4136 Class Level Information Name Levels Values osearch 2 0 1 Response Profile Ordered Total Value osearch Frequency 1 2 0 1 189 3947 PROC PROBIT is modeling the probabilities of levels of osearch having LOWER Ordered Values in the response profile table. Algorithm converged. Effect res Type III Analysis of Effects Wald DF Chi-Square Pr > ChiSq 1 0.0046 0.9461 Parameter Intercept res Analysis of Parameter Estimates Standard 95% Confidence Chi- DF Estimate Error Limits Square Pr > ChiSq 1 -1.6862 0.0436 -1.7717 -1.6008 1496.00 <.0001 1 -0.0047 0.0691 -0.1402 0.1308 0.00 0.9461 Probit Procedure Probit Model in Terms of Tolerance Distribution MU SIGMA -360.72291 213.92206 MU SIGMA Estimated Covariance Matrix for Tolerance Parameters MU SIGMA 28525751.192 -16898169.63 -16898169.63 10010206.542 Probit Procedure Model Information Data Set Dependent Variable Number of Observations Name of Distribution Log Likelihood WORK.SWITCHED osearch 4136 Normal -764.4762154 Number of Observations Read Number of Observations Used 4136 4136 Class Level Information Name Levels Values osearch 2 0 1 Response Profile Ordered Total Value osearch Frequency 1 2 0 1 189 3947 PROC PROBIT is modeling the probabilities of levels of osearch having LOWER Ordered Values in the response profile table. Algorithm converged. Type III Analysis of Effects Wald Effect DF Chi-Square nwhite 1 3.8009 male 1 2.5396 Pr > ChiSq 0.0512 0.iii0 Parameter Intercept nwhite male Analysis of Parameter Estimates Standard 95% Confidence DF Estimate Error Limits Chi- Square Pr > ChiSq 1 -1.5608 0.0603 -1.6791 -1.4425 668.90 <.0001 1 -0.1344 0.0689 -0.2694 0.0007 3.80 0.0512 1 -0.1109 0.0696 -0.2472 0.0255 2.54 0.Iii0 Probit Procedure Model Information Data Set Dependent Variable Number of Observations Name of Distribution Log Likelihood " WORK.SWITCHED osearch 4136 Normal -765.6360508 Number of Observations Read Number of Observations Used 4136 4136 Class Level Information Name Levels Values osearch 2 0 1 Response Profile Ordered Total Value osearch Frequency 1 2 189 3947 PROC PROBIT is modeling the probabilities of levels of osearch having LOWER Ordered Values in the response profile table. Algorithm converged. Type III Analysis of Effects Wald Effect DF Chi-Square Pr > ChiSq nwhite 1 3.7058 0.0542 age 1 0.2000 0.6547 Parameter Intercept nwhite age Analysis of Parameter Estimates Standard 95% Confidence Chi- DF Estimate Error Limits Square Pr > ChiSq 1 -1.6697 0.1048 -1.8752 -1.4643 253.77 <.0001 1 -0.1344 0.0698 -0.2712 0.0024 3.71 0.0542 1 0.0010 0.0023 -0.0035 0.0056 0.20 0.6547 Probit Procedure Model Information Data Set Dependent Variable Number of Observations Name of Distribution Log Likelihood WORK.SWITCHED osearch 4136 Normal -765.7318282 Number of Observations Read Number of Observations Used 4136 4136 Class Level Information Name Levels Values osearch 2 0 1 Response Profile Ordered Total Value osearch Frequency 1 0 189 2 1 3947 PROC PROBIT is modeling the probabilities of levels of osearch having LOWER Ordered Values in the response profile table. Algorithm converged. Type III Analysis of Effects Wald Effect DF Chi-Square nwhite 1 4.1325 res 1 0.0079 Pr > ChiSq 0.0421 0.9292 Parameter Intercept nwhite res Analysis of Parameter Estimates Standard 95% Confidence DF Estimate Error Limits Chi- Square Pr > ChiSq 1 -1.6249 0.0523 -1.7275 -1.5223 963.52 <.0001 1 -0.1398 0.0688 -0.2746 -0.0050 4.13 0.0421 1 -0.0062 0.0693 -0.1419 0.1296 0.01 0.9292 29 Probit Procedure Model Information Data Set Dependent Variable Number of Observations Name of Distribution Log Likelihood WORK.SWITCHED osearch 4136 Normal -766.1153911 Number of Observations Read Number of Observations Used 4136 4136 Class Level Information Name Levels Values osearch 2 0 1 Response Profile Ordered Total Value osearch Frequency 1 2 0 1 189 3947 PROC PROBIT is modeling the probabilities of levels of osearch having LOWER Ordered Values in the response profile table. Algorithm converged% Type III Analysis of Effects Wald Effect DF Chi-Square male 1 2.7992 age 1 0.5571 Pr > ChiSq 0.0943 0.4554 Parameter Intercept male age Analysis of Parameter Estimates Standard 95% Confidence DF Estimate Error Limits Chi- Square Pr > ChiSq 1 -1.6818 0.1051 -1.8878 -1.4758 256.09 1 -0.1161 0.0694 -0.2522 . 0.0199 2.80 1 0.0017 0.0023 -0.0028 0.0062 0.56 <.0001 0.0943 0.4554 3O Probit Procedure Model Information Data Set Dependent Variable Number of Observations Name of Distribution Log Likelihood WORK.SWITCHED osearch 4136 Normal -766.3743621 Number of Observations Read Number of Observations Used 4136 4136 Class Level Information Name Levels Values osearch 2 0 1 Response Profile Ordered Total Value osearch Frequency 1 2 0 1 189 3947 PROC PROBIT is modeling the probabilities of levels of osearch having LOWER Ordered Values in the response profile table. Algorithm converged. Type III Analysis of Effects Wald Effect DF Chi-Square male 1 2.9088 res 1 0.0365 Pr > ChiSq 0.0881 0.8485 Parameter Intercept male res Analysis of Parameter Estimates Standard 95% Confidence DF Estimate Error Limits Chi- Square Pr > ChiSq 1 -1.6090 0.0622 -1.7309 -1.4871 669.18 1 -0.1187 0.0696 -0.2550 0.0177 2.91 1 -0.0133 0.0694 -0.1493 0.1228 0.04 <.0001 0.0881 0.8485 3] Probit Procedure Model Information Data Set Dependent Variable Number of Observations Name of Distribution Log Likelihood WORK.SWITCHED osearch 4136 Normal -767.4946839 Number of Observations Read Number of Observations Used 4136 4136 Class Level Information Name Levels Values osearch 2 0 1 Response Profile Ordered Total Value osearch Frequency 1 2 0 1 189 3947 PROC PROBIT is modeling the probabilities of levels of osearch having LOWER Ordered Values in the response profile table. Algorithm converged. Type III Analysis of Effects Wald Effect DF Chi-Square age 1 0.6466 res 1 0.0171 Pr > ChiSq 0.4213 0.8959 Parameter Intercept age res Analysis of Parameter Estimates Standard 95% Confidence DF Estimate Error Limits Chi- Square Pr > ChiSq 1 -1.7557 0.0972 -1.9461 -1.5653 326.56 1 0.0018 0.0023 -0.0027 0.0063 0.65 1 -0.0091 0.0694 -0.1451 0.1269 0.02 <.0001 0.4213 0.8959 32 Probit Procedure Model Information Data Set Dependent Variable Number of Observations Name of Distribution Log Likelihood WORK.SWITCHED osearch 4136 Normal -764.3619792 Number of Observations Read Number of Observations Used Class Level Information Name Levels Values osearch 2 0 1 4136 4136 Response Profile Ordered Total Value osearch Frequency 1 2 0 1 189 3947 PROC PROBIT Is modeling the probabilities of levels of osearch having LOWER Ordered Values in the response profile table. Algorithm converged. Type III Analysis of Effects Wald Effect DF Chi-Square Pr > ChiSq nwhite 1 3.4191 0.0644 male 1 2.5532 0.ii01 age 1 0.1881 0.6645 res 1 0.0551 0.8144 Parameter Intercept nwhite male age res Analysis of Parameter Estimates Standard 95% Confidence DF Estimate Error Limits Chi- Square Pr > ChiSq 1 -1.5951 0.1156 -1.8217 -1.3684 1 -0.1293 0.0699 -0.2664 0.0078 1 -0.1115 0.0698 -0.2482 0.0253 1 0.0010 0.0023 -0.0036 0.0056 1 -0.0164 0.0697 -0.1531 0.1203 190.25 <.0001 3.42 0.0644 2.55 0.ii01 0.19 0.6645 0.06 0.8144 33