Annual Jail Collection Part_B_11_20_15

B. Collection of Information Employing Statistical Methods

1. Universe and Sampling Design

Deaths in Custody Reporting Program

The DCRP jail data collection has two parts: individual death forms and jail annual summary forms (ASF). The universe for the death forms includes inmates who died in all local jails in the United States in the past year. The universe for the ASF includes all county or city jail jurisdictions and all regional jails in the country (about 3,000) except for jails in Indian country. A jail jurisdiction is a county (parish in Louisiana) or municipal government that administers one or more local jails and represents the entity responsible for managing the jail facilities under its authority. A regional jail is one administered by two or more governments (or a board composed of representatives from two or more governments).

The respondents of DCRP are jail administrators in the U.S., who fill out a death form for each inmate who died under their custody and one ASF for jails they operate. BJS established procedures for collecting data directly from the approximately 3,000 local jails in 2000 and has collected this data without interruption since then. In doing this, BJS built upon established contacts and procedures for collecting data from these entities for other BJS collections.

All jails report population data and death counts, including zeroes, as required by the original and the 2013 DICRA legislation (referenced in Part A, Item 1) on the ASF. The ASF is necessary to capture the population data needed to calculate mortality rates. The DCRP annually conducts the only complete enumeration of the U.S. jail population.

The passage of the 2013 DICRA provides a Congressional mandate for all jails to submit data about deaths occurring in their facilities. Sampling jails to obtain mortality data is not viable because jails not sampled for data collection will fail to meet the 2013 Congressional mandate to submit custodial death data to the Attorney General.

Annual Survey of Jails sampling design

The ASJ has been a sample survey since its inception in 1982. Its universe is the same as for the DCRP, i.e., all county or city jail jurisdictions and all regional jails in the country.

Jail jurisdictions may contain multiple facilities and/or multiple facility operators, called reporting units. For example, four reporting units in Allegheny County (PA) represent a single jail jurisdiction. The ASJ sample is drawn at the jurisdiction level. When a jail jurisdiction is sampled, all reporting units within that jail jurisdiction are included in the sample. BJS collapses the reporting units into jail jurisdictions and reports statistics at the jurisdiction level.

The ASJ uses a panel design. The 2006-2014 Annual Survey of Jails (ASJ) sample was drawn in 2006, based on population measurements from the core items in the 2005 Census of Jails. This panel was surveyed annually from 2006 to 2014 until the next round of sample refresh.

Statistical precision is a priority of the ASJ sampling design. Survey costs and respondent burden also played significant roles in determining the sample size and selection criteria. About a third of the sample included certainty jail jurisdictions (typically large jails) and accounted for two-thirds of the jail inmate population, therefore reducing the burden to sample a greater number of smaller jails. All combined, the sample size and selection criteria played a significant role in maintaining small standard errors. Year-to-year comparison of standard errors are available in the appendix tables in Jail Inmates at Midyear 2014.¹

Over the past several years the statistical precision of the ASJ has declined due to jail closures, openings, and changing populations. In an effort to maintain optimal statistical precision, we will use data from the census of jails (OMB clearance number 1121-0305) to refresh the ASJ sample, as was originally planned. The 2015-2017 ASJ will adopt the same sampling strategy from the 2006-2014 ASJ, but will refresh the panel when the next census is conducted (planned for 2019).

As with the previous round of ASJ, 2015-2017 ASJ will use a stratified sampling design. The sampling frame consists of 2,870 jail jurisdictions nationwide, as identified in BJS’s 2013 Census of Jails. In the sampling design, the jail jurisdictions are grouped into 10 strata. Two of the 10 strata are certainty strata; all jurisdictions in certainty strata are included in the sample. Specifically, Stratum 12 consists of all regional jails, and Stratum 1 consists of all California jail jurisdictions as well as jail jurisdictions that met one of the following two criteria:

• held juvenile inmates at yearend 2013 and had an ADP of 500 or more inmates during 2013, or

• held only adult inmates and had an ADP of 750 or more during 2013.

All California jails were added to the ASJ sample in 2012 to capture changes in the jail population after the California Public Safety Realignment policy (See Part A, Section 6). The 2015-2017 ASJ sample design has kept this feature. All certainty jails have a sampling weight of 1.

The remaining jail jurisdictions are divided into 8 strata based on 2013 ADP and the presence of juveniles at yearend 2013, and a random sample of jail jurisdictions was selected from each stratum. The 2015-2017 ASJ sample includes 875 jail jurisdictions (or 938 reporting units), allocated to 12 strata (table 4). The sample sizes of the various strata are determined by Neyman allocation method to minimize the variance for a fixed total sample size. The formula for Neyman allocation is

where n_h is the sample size of stratum h, n is the total sample size, N_h is the total number of units (or the population) in stratum h, and S_h is the square root of the variance in stratum h.

Table 6. Sample design for DCRP jails sampled for the ASJ

	Description		Number of jurisdictions in census	Number of sampled jurisdictions	Number of sampled reporting units	Weight
1	Non-regional certainty jails and California jails*		317	317	364	1.000
2	Holding at least one juvenile on Census day	ADP 264 to 499	95	35	36	2.714
3		ADP 141 to 263	94	20	20	4.700
4		ADP 69 to 140	93	8	8	9.000
5		ADP 0 to 68	68	12	12	7.417
7	Holding adults only on Census day	ADP 227-749	266	208	221	1.279
8		ADP 103-226	407	83	85	4.904
9		ADP 40-102	567	63	63	9.000
10		ADP 0-39	894	60	60	14.900
12	Regional Jails and jails with ADP 750 or greater		69	69	69	1.000
Total			2870	875	938

Survey of Jails in Indian Country

Universe

The universe for the SJIC is all Indian country correctional facilities. The SJIC universe is identified through the Bureau of Justice Assistance’s Correctional Facilities on Tribal Lands program, as well as through informal means such as internet searches, Google alerts and word-of-mouth via established respondents.

Currently, BJS’s universe of jails in Indian country consists of 90 facilities. Prior to the 2014 survey collection, eight facilities were closed, not operating, or out of scope, resulting in a survey universe of 82 facilities. During collection year 2014, BJS determined that four additional facilities were holding cells and were, therefore, out of scope. One facility was added to the list, resulting in a final universe of 79 Indian country jails. BJS received responses from 74 facility administrators in 2014, for a 94% response rate. Year-to-year comparison of the SJIC universe is available in the methodology in Jails in Indian Country, 2014.²

Sampling from this universe would not be practical given the small universe of jails and the variety of purposes these facilities hold offenders, which range from short-term holding to persons convicted of misdemeanors with sentences usually of a year or less. With the passage of the Tribal Law and Order Act of 2010 (TLOA), expanding tribal court sentencing authority could result in offenders serving potentially longer sentences (up to 3 years) in Indian country correctional facilities. Additionally, producing a representative sample would be impractical considering jails in Indian country may be unique to tribal criminal codes and traditions, and many jails in Indian country are not operated based on county location, but by tribal affiliation. For example, the Navajo Nation operates 8 facilities located in Arizona and New Mexico. Given the high response rate for the SJIC, it’s most practical to continue as a census. In the case of survey or item non-response, data are estimated based on previously reported data. Additional information on estimation procedures are available in the methodology section in Jails in Indian Country, 2014.

2. Procedures for Information Collection

Procedures for the DCRP-Jails and ASJ collections. During November of the year prior to the upcoming calendar year collection (i.e., November of 2015 for the 2016 data collection cycle), BJS’ data collection agent (RTI International) notifies jails of the annual collection cycle. The notification occurs during annual verification calls. In the annual verification calls, RTI verifies that the current collection year respondent will be the respondent for the upcoming collection year and if not, will obtain the new respondent’s contact information. RTI also verifies that the jail facilities listed on the current roster within a jail jurisdiction are complete and their information is accurate.

Using the contact information obtained from the November verification calls, RTI mails to each jail respondent an information packet announcing the start of the current year’s collection cycle and requesting annual summary data for the previous year (see Appendix C, attachment L). In addition to the announcement letter, the packets contain year-specific envelopes that distinguish the DCRP current year’s surveys (CJ-9 or CJ-10 forms) from the DCRP and ASJ annual summary forms (CJ-9A or CJ-10A for the DCRP; CJ-9/5A and CJ-10/5A for those in the ASJ sample).

The packets include a reporting instructions sheet and a DCRP-ASJ fact sheet. The instruction sheet directs the respondents to the DCRP-ASJ website and provides them with login information to use in submitting data via the secure web-based reporting tool. In addition, respondents are reminded to submit an annual summary form and any outstanding death record forms from the preceding year by March 1^st of the current year.

To maximize response rates of local jail respondents, BJS provides multiple modes of data submission, with web as the primary mode. The primary mode of collection, the web-based reporting tool, allows respondents to submit individual death record forms on an ongoing basis. Secondary modes of submission of records, including email, fax and mail-in, are provided to accommodate a small number of local jail respondents with no or limited access to the web. A postage-paid business reply envelope is provided for respondents who choose to submit their data via U.S. mail. After a shift to a paperless mode, (see Part A, section 4 of this package for more information), 93% of DCRP respondents submitted the annual summary form online, and 73% submitted death forms online.

Since ASJ respondents are a sample of the DCRP universe, we expect the majority of ASJ respondents will also use the web-based reporting tool as well, but the same pencil and paper reporting available to DCRP respondents will also be available to the ASJ sample.

All data for a calendar year are due within 90 days following the end of the year. Follow-up contacts begin in the spring following the collection year to solicit late responses. RTI contacts nonrespondents via phone, fax, email, and mail (for respondents for whom RTI does not have an email address). Concentrated data quality follow-up with respondents occurs during the summer (see appendix C, attachment M).

RTI International monitors the DCRP data and its respondents throughout the active data collection period. This includes tracking the time to complete, number of contacts made to each agency, and other information (e.g., which form(s) have and have not been returned to date). This information is then used for tailored follow-up prompts to nonrespondents. A similar approach is employed when there are missing or conflicting data within the returned forms. For such outreach activities, agencies are assigned and managed by project team members, each of whom serves as that agency’s liaison for DCRP.

The DCRP data collection methods will extend to DCRP respondents sampled for the ASJ.

Due to the rural nature of Indian Country jails, an online-based system is not feasible at this time. As with previous iterations, SJIC respondents can submit data by mailing their reply in a pre-paid envelope, by fax, or by telephone.

3. Methods to Maximize Response

BJS and contract staff have been and will continue to market the jail data collections to garner support and generate interest among stakeholders to maximize response. We seek input from stakeholders to review and comment on the survey. Typically feedback is provided to BJS through email, at jail related conferences and meetings, or through comments provided by survey respondents during the data collection. BJS staff also present at conferences about the needs and uses of jails data, including workshops at the American Jails Association’s annual conference. BJS will also promote the jail collections through articles in the AJA magazine. For the 2013 Census of Jails, the American Jails Association included a blurb in their email “AJAlert” about BJS’s initiative in 2013 to expand the DCRP-ASF to act as the Census of Jails. We will continue to utilize various methods to reach out to stakeholders.

BJS has consistently achieved high rates of response for DCRP and the ASJ jail collections. Since BJS initiated the DCRP jail collection in 2000, it had an average annual response rate of 98% from approximately 3,000 local jail jurisdictions. The response rate in 2013 decreased slightly, but recovered in 2014 (table 7). The response rate for the ASJ was 100% in 2006 and 2007 and 99% from 2008 to 2010. In 2011 the response rate declined to a low of 91.3% in 2013, but recovered in 2014 (94.4%) (table 8).

Table 7. Response rates for eligible jails in the DCRP frame, 2010-2014

	2010	2011	2012	2013	20141
# of jails in the frame	2,813	2,895	2,878	2,949	2,997
# jails responding	2,892	2,797	2,869	2,732	2,869
% of jails responding	97.3%	96.6%	99.6%	92.8%	95.7%

¹ 2014 response rates are as September 14, 2015. The final data file will be delivered by the end of the month.

Table 8. Response rates for sampled jails in the ASJ, 2010-2014

	2010	2011	2012	2013	2014
# of jails in the frame1	930	926	920	941	930
# jails responding	922	888	861	859	878
% of jails responding	99.1%	95.9%	93.6%	91.3%	94.4%

¹Includes active respondents. The sample panel starting in 2006 consisted of 936 respondents. Between 2006 and 2014, the survey universe was adjusted to account for out-of-scope facilities and facility closures, either permanently or temporarily. The ASJ surveys without replacement, except for certainty jail jurisdictions. Starting in 2013, all jail jurisdictions in California were included with certainty, resulting in an additional 21 respondents.

Survey of Jails in Indian Country

BJS has consistently maintained high rates of response to the SJIC, typically between 90% and 100% survey response and item response rate. It should be noted that the SJIC is a voluntary collection. BJS believes that our efforts to minimize the reporting burden have been integral to achieving high rates of voluntary response. In the previous SJIC OMB clearance cycle, BJS eliminated a number of survey items for which the data quality was questionable. This resulted in a significant reduction in respondent burden hours. The rate to validate and revise reported data for the eliminated items was nearly triple that of critical items, such as the midyear inmate population and characteristics, average daily population (ADP), and admission and releases from jail. The benefit for eliminating those items included a reduction in the data collection time, validation attempts, and collection agent resources.

The data collection plan allows for the jail respondents to submit data by mailing their reply in a postage-paid envelope, by fax, or by telephone response. Follow-up telephone calls and faxes to non‑respondents are planned to encourage a 100 percent response rate. The response rate for the SJIC in 2014 was 94%.

Nonresponse Adjustments

ASJ

As previously mentioned, BJS has consistently maintained very high survey and item response rates for the DCRP, ASJ and SJIC. Data collection staff not only use follow-up emails, telephone calls, or faxes to promote high response rates, but also conduct out-of-range analysis of critical items to check data quality and follow up with respondents. These methods have proved effective in reaching the nearly 100% unit response rate, minimizing item nonresponse, and increasing data validity.

Typically for the ASJ, item response rates are nearly 100% for the following critical data items from responding jail jurisdictions: confined persons at midyear, average daily population, total rated capacity, the adult male and female population, and for the number of juvenile male and female inmates confined at midyear. The response rate ranges between 90% and 95% for conviction status and the number of inmates by race and Hispanic origin.

Despite BJS’s best efforts to collect complete data from all respondents, there is still a small number of nonrespondents. When unit nonresponse increased in 2011, BJS implemented nonresponse weight adjustment procedures to account for unit nonresponse. Detail information on imputation procedures are available in the methodology section of Jail Inmates at Midyear, 2014.

To deal with item nonresponse, BJS has developed a number of appropriate procedures of missing data imputation. Utilizing its panel design, missing data on critical items of midyear inmate population, ADP, and rated capacity are often imputed as the most recent value reported from the same reporting units. For example, missing data on rated capacity for 9 reporting units in 2014 ASJ were imputed as their reported values in 2013 ASJ.

Another imputation strategy BJS uses is borrowing from a similar (highly correlated) item. For example, missing data on 2014 ADP for 9 reporting units were imputed as the value of their confined population at midyear 2014; missing weekly release counts for 5 reporting units were imputed as their reported weekly admission counts in 2014. The correlations between ADP and midyear confined population, and between admission/release counts are very high.

Missing data on inmate characteristics are not imputed. Based on the 2014 ASJ, between 90% and 95% of the 878 reporting units provided valid data on sex, age, race/Hispanic origin, and inmate conviction status. National counts of inmates of a particular characteristic are estimated as the total midyear confined population multiplied by the percent of inmates with that characteristic among jurisdictions that reported that characteristic.

During the next wave of ASJ, BJS staff will conduct non-response analysis by comparing the characteristics of the ASJ respondents and non-respondents using data from the 2013 Census of Jails. Through this analysis, BJS staff will understand what characteristics predict unit and item non-response and investigate additional strategies of adjusting for non-response bias.

SJIC

Typically the unit and item response rates for the survey are nearly 100%. Since 2004, data were imputed for the seven facilities in 2004, four facilities in 2007, two facilities in 2013, and five facilities in 2014 that did not respond to the surveys. Typically imputations for the midyear jail population, average daily population, peak population, and admission during the month of June used the most recent available data.

In some cases, facilities responding to the survey report incomplete data. For example, the rated capacity for midyear 2014 was imputed for two facilities; one was based on 2013 data and one was based on 2012 data. The imputed value for the peak population for four facilities during June 2014 was based on their midyear or ADP in 2014, whichever was greater.

Prior to 2014, data on inmate demographic characteristics, conviction status, and offense, and facility staff characteristics were analyzed based on reported data. Starting in 2014, BJS estimated data for these populations and produced estimates for previous years. Estimation procedures were based on the most recent reported data or the ratio of the total number of inmates in their jail to the reported number of inmates by the inmate/staff demographic or characteristic. Detail estimation procedures by year are available in Jails in Indian Country, 2014.

4. Test of Procedures or Methods

BJS obtains feedback from local and tribal jail administrators and other experts when significant changes to existing surveys are implemented or when a new survey instrument is introduced to the field. BJS is concerned about placing additional burden on respondents while simultaneously understanding how important it is to collect and disseminate current and critical information on jail and jail inmates.

As described in part A, Item 2 of this package, BJS proposes to drop six items from the DCRP-ASJ CJ9A/5 and CJ10A/5 forms. The proposed deletions will be offset by three proposed additions, based on feedback received during the 2013 Census and earlier iterations of the ASJ.

The SJIC was last revised in 2013. The current SJIC was tested in 2013 and revised based on comments received. Since then the survey has been successfully administered and completed without significant difficulty.

The jail portion of the DCRP was last revised in 2010. The newer items (holds for other correctional authorities; flags indicating inmates had a stay in a mental health facility; additions to location of death) have been administered and completed without difficulty.

Through discussions with the varied stakeholders and respondents for SJIC and DCRP, BJS has found that there is general satisfaction with the current surveys, that the questions cover important topics in the field, and that the accompanying instructions for completing the survey are clear.

6. Contact Information

The Corrections Statistics Unit of BJS takes responsibility for the overall design and management of the activities described in this submission, including data collection procedures, development of the questionnaires, and analysis of the data.

BJS contacts for the Annual Jail Collections include -

Anastasios ‘Tom’ Tsoutis

Acting Chief

Corrections Statistics Unit

Bureau of Justice Statistics

810 Seventh Street, N.W.

Washington, DC 20531

(202) 514-1062

[email protected]

Margaret Noonan

Statistician and DCRP Project Manager

Corrections Statistics Unit

Bureau of Justice Statistics

810 Seventh Street, N.W.

Washington, DC 20531

(202) 353-2060

[email protected]

Todd Minton

Statistician, ASJ and SJIC Project Manager

Corrections Statistics Unit

Bureau of Justice Statistics

810 Seventh Street, N.W.

Washington, DC 20531

(202) 305-9630

[email protected]

1¹ Minton, T. and Zeng, Z. (2015). “Jail Inmates at Midyear 2014”. (NCJ 248629). U.S. Department of Justice, Office of Justice Programs, Bureau of Justice Statistics.

2² Minton, Todd D. (2015) Jails in Indian Country, 2014, Bureau of Justice Statistics, Washington, DC. NCJ 248974.

8