U.S. Department of Justice Office of Justice Programs The Local Area Crime Survey: A Companion to the National Crime Victimization Survey Grace Kena Association of State Uniform Crime Reporting Programs 2018 Conference September 5 th BUREAU OF JUSTICE STATISTICS
NCVS Companion Survey Background and Impetus 2
The National Crime Victimization Survey (NCVS) Nationally representative Household (address)-based panel survey Interviews (in-person or via telephone) with non-institutionalized persons age 12 or older Individual and household respondents, very limited proxy response Excludes: persons under age 12, armed forces personnel living in military barracks, the homeless, institutionalized persons, U.S. citizens residing abroad Sample interviewed every 6 months for 7 waves 134,690 households ~ 78% response rate (2016) 224,500 people ~85% response rate (2016) Measures the dark figure of crime, or crimes not reported to police 3
The National Crime Victimization Survey (cont.) Crime types covered Rape/Sexual assault Robbery Aggravated assault Simple assault Pocket picking/purse snatching Burglary Motor vehicle theft Theft Identity theft Types of attributes measured Location of crime and victim presence Attack/threat/injury/medical care Emotional consequences Actions against the offender Offender information/characteristics Attempted and completed thefts Property damage and police
Why an NCVS Companion Survey? Need for local jurisdiction (state, county, city, sub-county/city) estimates of crime victimization National-level NCVS data limited Cost prohibitive to adequately increase NCVS BJS pursued a variety of methods Expanding core NCVS sample Producing model-based estimates Companion Survey: a lower-cost data collection comparable to NCVS and administrable by local areas 5
Companion Survey Goals Support local areas in measuring crime patterns and trends Use statistically reliable instrument as complement to local crime data Produce victimization estimates for smaller set of NCVS crime types Use a simplified mail instrument at a lower cost Produce correlated estimates to the NCVS that can show parallel trends Engage non-victims with community and police perception measures Test items for planned NCVS instrument redesign 6
NCVS Companion Survey (CS) Study Design and Methodology 7
CS Study Design Single household informant, covering up to 4 household adults age 18 or older Twelve-month recall period Simplified set of crime types and fewer victimization correlates Estimates on households and persons touched by crime for violent and serious violent crime, property crime, and motor vehicle theft Not victimization or incident rates Two instruments tested Incident-level survey Person-level survey 8
CS Instruments Incident-level survey Adult household respondent Respondents provided information on up to 4 violent and property victimizations Respondents enumerated and described incidents Household roster included early in instrument, followed by violent crime and property crime questions Provides incident-level information and prevalence estimates 9
CS Instruments (cont.) Person-level survey Completed by each adult in household on most recent violent and property victimization Household not rostered Property crime items early in instrument Victimization experiences of each adult (up to 4) Provides prevalence estimates 10
CS Experiments To test implementation strategies for local areas, experiments included Instrument (Incident vs Person) Placement of community and policing items Incentives and special mailings Overlap sample (included 25% of Year 1 sample in Year 2) Bilingual (Spanish language materials sent to targeted addresses) Interactive voice response vs. postcard reminder 11
CS Comparison to NCVS CS Adults age 18 or older Cross-sectional (annual or other) Household experiences with property crime Experience of household adults with violent crime Reporting to police Offender relationship Experiences with fraud/identity theft Community/policing items NCVS Persons age 12 or older Panel, 7 waves Household property crime victimizations and incidents Personal violent and property crime victimizations and incidents Reporting (and not reporting) to police details Detailed victim and offender characteristics Cost and consequences of crime
CS Study Design Two-wave field test (2015 and 2016) U.S. Postal Service sampling frame, randomized complete block design Sample of 220,000+ households: 40 largest core-based statistical areas (CBSAs) Oversamples in 3 metro areas Philadelphia, Chicago, Los Angeles Initial sample sizes 4,720 for most CBSAs 21,000 in Philadelphia; 16,900 in Chicago and Los Angeles 13
NCVS Companion Survey (CS) Field Test Implementation 14
Atlanta CBSA 2,100 total target completes, allocated across entire CBSA 15
Oversampled CBSAs Philadelphia 9,363 total target completes, allocated within 6 policing divisions (not shown on map) 16
Oversampled CBSAs Chicago 7,500 total target completes, allocated across 3 policing areas (not shown on map) 17
Oversampled CBSAs Los Angeles 7,500 total target completes, allocated across 4 police admin bureaus (not shown on map) 18
CS Year 1 CBSA Overall Response Rates Percent 70.0 Chart Title 60.0 50.0 40.0 30.0 20.0 10.0 0.0 Entire CBSA Major City 19
CS Field Test: General Results Good response rates (~45%), no substantive difference between instruments Relatively high data quality Minimal editing from raw data Good correlation with the NCVS Reasonable correlation with UCR property crime, despite differences Results from Year 1 => some changes for Year 2 including Abandoned form experiment; community items placed at beginning of survey Minor questionnaire revisions including adding content on police efficacy, aligning content across instruments, and wording/formatting enhancements 20
CS Field Test: Instrument Results Quality of reporting similar Similar community/policing estimates for both High correlations with NCVS for both but higher violent crime correlation with person-level Higher touched-by-crime rates (violent and property) on person level More false negatives, victims outside of household on incident level Problems linking crimes to individuals on incident level Better alignment with NCVS content with incident level 21
CS Cost Results (With and Without Incentives) Treatment group Completes Response rate Total cost Estimated cost per complete 0U 9,615 29.1% $ 254,694 $ 26.49 1U 11,249 34.0% $ 279,670 $ 24.86 2U 12,456 37.6% $ 309,520 $ 24.85 0F 11,449 34.3% $ 346,197 $ 30.24 1F 13,077 39.2% $ 362,441 $ 27.72 2F 13,957 41.7% $ 388,823 $ 27.86 0,1,2=Incentive dollar amount; Delivery mode in 3 rd mailing: U=USPS, F=Fed Ex 22
CS Field Test: Incentive and Mailing Results Incentive and mailing effects Incentive increased response rate and as effective as FedEx delivery Small incentive has positive return on investment more cost-effective Observed changes across conditions for some demographic groups (selected for correlation with experiencing victimization) Incentives and mailing options influenced groups differently 23
CS Field Test: Conclusions A mailed crime survey can be a low-cost complement to the NCVS for local areas Moderate base cost compared to the NCVS (~$26 per case) Incentives and special mailings can improve cost-effectiveness Reasonable response rates for mail survey with low evidence of bias Relatively high data quality for this type of instrument Crime estimates correlated with NCVS and, particularly for property crime, UCR Good tool for analyzing data for small areas, including sub-city Useful for assessing impact of local programs and providing indicators of community safety and police performance Useful for making estimate comparisons across time and place 24
NCVS Companion Survey (CS) Next Steps 25
CS Planned Next Steps Detailed methodological report Toolkit on how to use CS methodology to develop a local area survey Content report describing community and police perceptions findings Restricted-use data files and public-use data tool Open to opportunities to share more about this work and its applicability moving forward, collaboration 26
Thank you! Questions? Grace Kena grace.kena@usdoj.gov