2011 Imputation Methodology document

Attachment J - Documentation of the Imputation Methodology for the 2013 CJRP.pdf

Census of Juveniles in Residential Placement (CJRP)

2011 Imputation Methodology document

OMB: 1121-0218

⚠️ Notice: This form may be outdated. More recent filings and information on OMB 1121-0218 can be found here:

Document [pdf]

Download: pdf | pdf

Page 1 of 37

June 15, 2015
MEMORANDUM FOR THE RECORD
From:

Suzanne M. Dorinski
Public Sector Statistical Methods Branch
Economic Statistical Methods Division

Subject:

Documentation of the Imputation Methodology for the 2013 Census of Juveniles in
Residential Placement

This memorandum presents a streamlined version of the imputation methodology for the 2013 Census
of Juveniles in Residential Placement (CJRP).
This document uses intentional white space to improve readability. The document is available in PDF
format, because the reader’s computer might not reproduce the original formatting.
This document also serves as a guide to the programmer who works on the 2015 CJRP. Notes to the
programmer are in brackets. The 2013 CJRP system is reusable for the 2015 CJRP. [The programs are in
the \\govs011fs\cjrp\2013 subdirectory.]
The imputation methodology for the 2001 CJRP and earlier censuses used the section and question
numbers as variable names. It is very easy to make a typing mistake while using that convention. The
naming convention also makes it more difficult to read the program code and debug it. For the 2013
CJRP, we assigned variable names that are more descriptive. Table 1 shows the naming conventions
used in the programs in the imputation system. The section and question number for each item are in
parentheses. The final record layout for the 2013 CJRP also includes descriptive variable names, but
those variable names do not always match the variable names used in the imputation system. The
variable names as assigned in the final record layout for 2013 are also included in parentheses.

Page 2 of 37

Table of Contents
Table 1. Naming Conventions in the 2013 CJRP Imputation System ............................................... 4
I.

Introduction....................................................................................................................... 5

II.

Summary of Changes during 2013 Data Collection ................................................................... 6

III.

Summary of the file ......................................................................................................... 8

Table 2. Summary of the facilities on the 2013 final imputed file .................................................. 8
Table 3. 2013 CJRP counts (nontribal facilities in the 50 states and the District of Columbia)............. 8
IV.

Records Eligible for Imputation.......................................................................................... 8

Table 4. Values for the 2013 Status Flag.................................................................................... 9
V.

Critical Item Facilities .......................................................................................................... 9
Table 5. Values for Critical Items Field .....................................................................................10

VI.

Item response rates ........................................................................................................10

Table 6. 2013 CJRP Item Response Rates..................................................................................11
VII.

Questionnaire Items Eligible for Imputation .......................................................................12

Table 7. Items Eligible for Imputation ......................................................................................12
VIII.

Imputation Rates............................................................................................................13

Table 8. CJRP Section I Item Imputation Rates by Year................................................................13
Table 9. 2013 CJRP Imputation Rates for Items in Section I Not Previously Imputed........................13
Table 10. CJRP Section II Item Imputation Rates by Year.............................................................14
IX.

Collapsed Facility Type Codes ...........................................................................................15

Table 11. Hierarchy Used to Assign Collapsed Facility Type Code .................................................16
X.

Imputation Methodology for Section I Data ...........................................................................16
Missing data for check box questions....................................................................................16
Table 12. Section I Check Box Questions over Time....................................................................17
Missing data for the population counts .................................................................................17

XI.

Imputation Methodology for Item Nonresponse in Section II Data.........................................18
Changes in methodology over time ......................................................................................18

Table 13. How We Impute Missing Offense Codes .....................................................................18
Randomly imputing the day of birth, month of birth, or day of admission ..................................19
Age and stay calculations ....................................................................................................19
Hierarchical hot-deck imputation for item nonresponse ..........................................................19

Page 3 of 37

XII.

Imputation Methodology for Section II Data for Critical Item Facilities....................................20
Changes in methodology over time ......................................................................................20
Background.......................................................................................................................21
Hierarchical hot-deck for critical item facilities .......................................................................22

XIII.

Quality Checks Performed After Imputation .......................................................................22

XIV.

Other Programming Notes ...............................................................................................23

XV.

Caution When Comparing State Data over Time..................................................................24

XVI.

Issues to Consider for the Next CJRP Collection ...................................................................25

XVII.

Kid_offense_location revision after sponsor review .............................................................26

XVIII.

References.................................................................................................................27

Attachment A: Percentage of Juvenile Offenders Held in Critical Item Facilities by State of Facility.......28
Attachment B: Section II Item Imputation Rates by State of Facility .................................................29
Attachment C: Data Quality Issues in CJRP over Time ....................................................................31
Attachment D: 2013 CJRP Section II Nonresponse Patterns ............................................................35

Page 4 of 37

Table 1. Naming Conventions in the 2013 CJRP Imputation System
Concept
Persons assigned to
beds
Persons assigned to
beds age 21 or older
Persons under age 21
assigned to beds
Persons under age 21
assigned to beds due to
offenses
Persons assigned to
beds for reasons other
than offenses
Juvenile offender ID
Juvenile offender’s sex
Juvenile offender’s
birth date

Juvenile offender’s race
Agency that placed the
juvenile offender in
facility
Juvenile offender’s
most serious offense
Juvenile offender’s
adjudication status
Juvenile offender’s date
of admission

2013 CJRP
total_2013
(Number_Assigned)
(S1Q14b)
adults_2013
(Number_21_Older)
(S1Q15)
kids_2013
(Number_21_Under)
(S1Q16b)
kid_offenders_2013
(Number_Offense)
(S1Q17b)
kid_nonoffenders_2013
(Number_Non_Offense)
(S1Q18b)
kid_id
(S2Q1)
kid_sex
(S2Q2)
kid_birth_month
(S2Q3)
kid_birth_day
(S2Q3)
kid_birth_year
(S2Q3)
kid_race
(S2Q4)
kid_placed_by
(S2Q5)
kid_offense
(S2Q7)
kid_adjudication_status
(S2Q9)
kid_admitted_month
(S2Q10)
kid_admitted_day
(S2Q10)
kid_admitted_year
(S2Q10)

2012 JRFC
total_2012_jrfc
(S1Q5b)

2011 CJRP
total_2011
(S1Q10b)

adults_2012_jrfc
(S1Q6)

adults_2011
(S1Q11)

kids_2012_jrfc
(S1Q7b)

kids_2011
(S1Q12b)

kid_offenders_2012_jrfc
(S1Q8b)

kid_offenders_2011
(S1Q13b)

kid_nonoffenders_2012_jrfc
(S1Q9b)

kid_nonoffenders_2011
(S1Q14b)

Page 5 of 37

I. Introduction
First conducted in 1997, the CJRP is a mail canvass census. The 2010 CJRP was the first collection to give
facilities the option to respond online. The CJRP asks juvenile residential custody facilities in the U.S. to
describe each youth assigned to a bed in the facility on the last Wednesday of October. Adult facilities,
or facilities exclusively for drug or mental health treatment, or facilities for abused or neglected children
are not included in the census. Normally conducted in odd-numbered years, the CJRP collection
scheduled for 2005 occurred in early 2006, and the collection scheduled for 2009 occurred in early 2010.
The reference date for the 2013 CJRP was Wednesday, October 23, 2013.
In 1997, CJRP replaced the Census of Public and Private Juvenile Detention, Correctional, and Shelter
Facilities, also known as the Children in Custody census, which began in the early 1970s. Previous
censuses collected data on the facilities and the juvenile offenders held in the facilities.
CJRP collects an individual record on each offender less than 21 years of age held in the residential
facility, with information on the juvenile’s sex, date of birth, race, agency or authority placing the
offender there, most serious offense, court adjudication status, and date of admission to the facility.
The National Center for Juvenile Justice, the research division of the National Council of Juvenile and
Family Court Judges, maintains the CJRP databook online. The databook contains a set of pre-defined
tables detailing the characteristics of juvenile offenders in residential placement facilities. Tables are
currently available for 1997, 1999, 2001, 2003, 2006, 2007, 2010, and 2011.
The National Archive of Criminal Justice Data holds the previous data files, where they are part of the
restricted access collection. For more information, see http://www.icpsr.umich.edu/icpsrweb/NACJD/.
The project sponsor is the Office of Juvenile Justice and Delinquency Prevention (OJJDP). The Public
Sector Statistical Methods Branch (PSSMB) of Economic Statistical Methods Division is responsible for
imputation in the CJRP, while the Criminal Justice Branch (CJB) of Economic Reimbursable Surveys
Division is responsible for data collection and editing.
A facility is temporarily out-of-scope when they do not hold juveniles on the reference date.
A facility is permanently out-of-scope for one of several reasons:
•
•
•
•
•

The facility is no longer a residential facility (might have converted to day treatment only).
The facility is a duplicate of a record already on the data file.
The facility has changed from public to private, or private to public. When this happens, the
facility ID changes, and the previous facility ID is out-of-scope.
The facility no longer holds any juveniles (only handles adults).
The facility no longer holds any offenders (juveniles are all voluntary placements, or in the
facility because of neglect, abuse, dependency, or abandonment).

Page 6 of 37

There were 2,293 in-scope facilities on reference day. 2,111 of the 2,293 facilities responded to the
2013 CJRP, for a 92.1 percent unit response rate. 182 refused to participate in the 2013 CJRP, but we
imputed records for the nontribal facilities in the 50 states and the District of Columbia.
The 2013 questionnaire has two sections: Section I, which collects general information about the
facility, and Section II, which collects individual person data for juvenile offenders held at the facility.
This document includes the response rates and describes the imputation methodology that we used to
make complete data sets for analytical purposes. Chapter II gives a brief description of the changes
during the 2013 data collection cycle. We summarize the numbers of facilities and records on the 2013
imputed file in Chapter III. Chapter IV describes the records eligible for imputation. In Chapter V, we
discuss facilities, referred to as critical item facilities, which were only able to respond to a critical subset
of the requested data. Chapter VI lists the item response rates. We did not impute for every item in
the questionnaire, and Chapter VII covers the items that were eligible for imputation.
The discussion of imputation rates is in Chapter VII. As described above, the unit response rate is high
and leads to low imputation rates for Total Persons, Total Adults, Total Juveniles, Total Juvenile
Offenders, and Total Juvenile Nonoffenders, as seen in Chapter VII.
Exercise caution when using State data over time. There was a marked increase in high imputation rates
for 2007. The exact imputation rates by State for 2007 are in Attachment G of the 2007 imputation
documentation. Attachment B of this document shows the exact imputation rates by State for 2013.
Highlighted Items have imputation rates that exceed 30 percent; exercise caution when using these
data. Attachment C shows the items within each State with imputation rates of more than 30 percent
for the period from 1997 through 2013.
A description of the imputation methodology follows in Chapters IX through XII. Chapter XIII discusses
quality checks performed after the imputations. Chapter XIV discusses other programming notes, which
will be useful when running the imputations for the next collection. Chapter XV discusses issues about
comparing the data over time. Chapter XVI summarizes issues to consider for the next collection.
References are included in Chapter XVII.

II. Summary of Changes during 2013 Data Collection

There were quite a few changes for the 2013 CJRP:
•
•
•
•

General facility information questions from the Juvenile Residential Facility Census were added
to Section I of CJRP.
Records for facilities that are temporarily closed, permanently closed, or out-of-scope are now
included on the final imputed file.
CJRP variable names are more descriptive.
The way that facility type is assigned in CJRP imputation processing now more closely matches
the way that facility type is assigned for JRFC imputation processing.

Page 7 of 37

•
•
•
•

More questions in Section I are imputed.
A “special handling” critical item status was used for the 2013 CJRP.
The values of the status flag for the 2013 CJRP are different than they were in the 2011 CJRP.
The values of the imputation flags have changed, to provide more details on the methods used.

Historically, the general facility information questions in Section I of CJRP have been a smaller set than
the general facility information questions on the Juvenile Residential Facility Census (JRFC). The JRFC is a
mail canvass census that occurs in alternate years from the CJRP. JRFC collects more information about
the facilities that hold juvenile offenders. General facility information questions have been added to
Section I of the 2013 CJRP.
OJJDP and juvenile justice researchers want to be able to track juvenile residential facilities over time.
Adding more general facility information questions to CJRP will help researchers track changes over
time.
To help with longitudinal analysis, the 2013 CJRP final imputed file now includes records for facilities
that were temporarily closed, permanently closed, or out-of-scope during the 2013 data collection.
As mentioned on the first page of this memorandum, the 2013 CJRP variable names are more
descriptive. See the final record layout for more details.
The way that the facility type question is handled in CJRP has differed from how it was handled in JRFC.
For 2013, the way that facility type is assigned in CJRP imputation processing now more closely matches
the way that facility type is assigned for JRFC imputation processing. More detail is provided in a later
section of this document. Census provides data tables with each year’s final file. The CJRP table
package has been updated to include tables that refer to the new questions.
Historically, the only questions in Section I of CJRP that were imputed were the check box questions
associated with the population counts, the population counts, and the facility type question. The CJRP
imputation system has been updated to impute more of the general facility information questions in
Section I, in a manner similar to how those questions are imputed in JRFC.
The CJRP has used a hot-deck imputation methodology since 2003. Facilities were classified as noncritical item facilities if they answered most survey items. The hot-deck handled item nonresponse on
the juvenile roster in Section II. Facilities were classified as critical item facilities if they only answered
critical item questions, which are discussed in a later section of this document. Critical item facilities
provided very little information about the offenders held, so a separate set of hot-deck programs was
used to impute juvenile offenders, taking care not to impute multiple identical offenders in the same
facility. Critical item facilities may have reported that they only held one sex, or all the offenders had
the same adjudication status, or all the offenders were status offenders. In 2013, the analysts asked
some critical item facilities to report racial distributions for the juvenile offenders, or percentage of
juvenile offenders by offense code. Special hot-deck programs were written to deal with these “special
handling” critical item facilities.

Page 8 of 37

The status flag on the 2013 CJRP final imputed file uses the same coding scheme as the status flag on the
2012 JRFC final imputed file. Please note that the meaning of the values of the status flag has changed
since the 2011 CJRP final imputed file.
The values of the imputation flags have changed, to provide more details on the methods used. The
imputation flag values are also similar to those used for the JRFC file.

III. Summary of the file
Table 2. Summary of the facilities on the 2013 final imputed file
101
22
67
19
11
2,263
2,483

permanently closed facilities
temporarily closed facilities
out-of-scope facilities
tribal facilities
territorial facilities
nontribal facilities in the 50 states and District of Columbia
facilities on the 2013 CJRP final imputed file

Table 3. 2013 CJRP counts (nontribal facilities in the 50 states and the District
of Columbia)
65,756
367
65,389
54,148
11,241
2,263

people in residential placement
adults
juveniles
juvenile offenders
juvenile nonoffenders
nontribal facilities in the 50 states and District of Columbia

[The program summarize_imputed_file_counts.sas produces the counts in this section.]

IV. Records Eligible for Imputation
In previous versions of CJRP, we imputed missing data for all facilities and all offender records. Starting
with the 2010 CJRP, OJJDP requested that we not impute missing data for tribal facilities or for offenders
in tribal facilities.
[Tribal facilities face special challenges, which is one reason that we do not impute them. The reader is
directed to “From Broken Homes to a Broken System”, written by Sari Horwitz and published in the
Washington Post on November 28, 2014, and accessible online at
http://www.washingtonpost.com/sf/national/2014/11/28/from-broken-homes-to-a-broken-system/.]

Page 9 of 37

The 2010 CJRP was the first cycle to attempt to collect data from facilities in American Samoa, Guam,
Northern Marianas Islands, Puerto Rico, and the Virgin Islands. OJJDP and Census agreed that we would
not impute missing data for territorial facilities or for offenders in territorial facilities.
The 2013 CJRP has a 2013 status flag in column 1205. Table 4 shows the values of the status flag.
Records eligible for imputation in the 2013 CJRP have the 2013 status flag set to 1, 2, or 3.

Table 4. Values for the 2013 Status Flag
2013 status flag value
1
2
3
4
5

Description
State-operated public facility in the 50 states or District of Columbia
Locally-operated public facility in the 50 states or District of Columbia
Private facility in the 50 states or District of Columbia
Tribal facility (missing data are not imputed)
Territorial facility (missing data are not imputed)

V. Critical Item Facilities
In follow-up interviewing, CJB attempted to collect as much data as possible to fill in both sections of the
questionnaire. The following data items were critical:
Section I:
•
•
•
•
•

Question 9 (type of facility)
Question 14a and 14b (total persons assigned to beds in the facility)
Question 15 (number of persons age 21 or older assigned to beds in the facility)
Question 17a and 17b (number of offenders under age 21 assigned to beds in the facility)
Question 18a and 18b (number of nonoffenders under age 21 assigned to beds in the facility)

Section II:
•
•
•
•
•
•

Question 2 (whether facility is all-male, all-female, or holds both sexes)
Question 5 (placement agency)
Question 6 (placement agency’s government level)
Question 7 (offense code)
Question 8 (state or territory where offense committed)
Question 9 (adjudication status)

Table 5 shows the values for the critical items field.

Page 10 of 37

Table 5. Values for Critical Items Field
Critical items
field value
blank
0
1
2
3
4

Meaning
Not eligible for imputation.
Facility is neither a critical item facility nor a refusal.
Facility responded only to the critical items.
Facility responded to Section I of the questionnaire, but date of birth, offense, and
date of admission are missing for all records in Section II of the questionnaire, while
the reported characteristics are the same for every juvenile offender in the facility.
Facility is a refusal; we imputed all data on the file for that facility.
Special handling is required to impute juvenile offenders in these facilities. CJB
collected percentage distributions on selected offender characteristics.

The critical items field is column 1193 on the data file.
Facilities with the critical items field sent to 1 have only one record per facility on the edited file, and the
information in Section II for those facilities is used to generate the juvenile offender roster for each
facility.
Facilities with the critical items field set to 2 are imputed in the same manner as facilities with the
critical items field set to 1. When the date of birth, offense, and date of admission are missing for every
juvenile offender in the facility, and the reported characteristics are the same for every juvenile offender
in the facility, we have to impute for every juvenile offender in the facility, and we want to avoid using
the same donor repeatedly within the facility.
Refusal facilities are also imputed in the same manner as facilities with the critical items field set to 1, to
minimize the number of records within a facility imputed using the same donor. The critical items field
value of 3 is the way to readily identify refusals on the file that were eligible for imputation.
Facilities with the critical items field set to 4 require special handling. The analysts collected percentage
distributions of juvenile offenders by sex, or by race, or by offense, or by adjudication status. The
programs developed to perform the hot-deck imputation were designed based on the nonresponse
patterns observed in the 2001 CJRP file. The 2001 CJRP file had two types of nonresponse patterns:
items missing within a reported juvenile offender record, or the roster in Section II is completely missing
(critical items field = 1 or the facility is a refusal). As the options for nonresponse patterns increase,
more and more special handling is required for each data collection.

VI. Item response rates
We calculated the item response rates for the 2013 CJRP by looking at the flag values after imputation.
2,263 nontribal facilities in the 50 states and the District of Columbia held juveniles on reference day.
The calculations consider skip patterns. If a facility was skipped out of a question on the form, they

Page 11 of 37

were not counted as an item nonrespondent. For example, if the facility answered No to the question
asking if they held juveniles on reference day for reasons other than offenses (question 18a in Section I),
they were not considered an item nonrespondent for question 18b because they were instructed to skip
question 18b.

Table 6. 2013 CJRP Item Response Rates
Variable

Response rate

Variable

Response rate

Larger_Agency

92.4

Non_Offense

92.4

Own

88.4

Number_Non_Offense

92.4

Own_Level

88.4

Kid_Sex

86.8

Operator

92.2

Kid_Birth_Month

83.4

Operator_Level

92.2

Kid_Birth_Day

83.4

Treatment

90.7

Kid_Birth_Year

83.3

Treatment_Type

90.7

Age

83.2

Foster_Care

90.5

Kid_Race

81.5

Independent_Living

90.3

Kid_Placed_By

86.0

Facility

93.5

Kid_Offense

78.7

Overflow

91.2

Kid_Adjudication_Status

79.4

Locked_Room

90.6

Kid_Admitted_Month

82.9

Locked_Reason

90.6

Kid_Admitted_Day

82.9

Security

91.1

Kid_Admitted_Year

82.1

Facility_Locked

90.9

Stay

82.0

Locked

90.9

Locked_Outside

90.9

Any_Beds

92.4

Number_Assigned

92.4

Number_21_Older

92.4

Any_21_Under

92.4

Number_21_Under

92.4

Under_21_Offense

92.5

Number_Offense

92.5

[The program CJRP_response_rates.sas
produces the data in this section.]

Page 12 of 37

VII. Questionnaire Items Eligible for Imputation
The following items were eligible for imputation in the 2013 CJRP. See the notes at the end of Table 7
for descriptions of the codes used in the imputation methods column.

Table 7. Items Eligible for Imputation
Question
Questions 1a, 1b
Questions 2a, 2b, and 3
Questions 4a, 4b, and 5
Questions 6a and 6b
Question 7
Question 8
Question 9
Question 10
Questions 11a and 11b
Question 12
Questions 13a, 13b, and 13c
Questions 14a and 14b
Question 15
Questions 16a and 16b
Questions 17a and 17b
Questions 18a and 18b
Question 2
Question 3
Question 4
Question 5
Question 7
Question 9
Question 10

Concept
Section I
Facility part of larger agency
Who owns this facility
Who operates this facility
Treatment provided at facility
Foster care
Independent living arrangements
Type of facility
Overflow detention population
Locked sleeping rooms
Security features
Locked outside doors
Total persons assigned to beds in the facility
Number of persons age 21 or older assigned to beds
Number of persons under age 21 assigned to beds
Number of offenders under age 21 assigned to beds
Number of nonoffenders under age 21 assigned to beds
Section II
Juvenile offender’s sex
Juvenile offender’s birth date
Juvenile offender’s race
Placement agency
Juvenile offender’s most serious offense code
Juvenile offender’s adjudication status
Juvenile offender’s date of admission to the facility

NOTES:
A: Data derived from response to other variables (flag=2).
B: Data pulled forward from prior year JRFC (flag=20).
C: Data imputed using growth rate applied to prior year CJRP data (flag=21).
D: Data imputed using growth rate applied to prior year JRFC data (flag=22).
E: Data pulled forward from prior year CJRP (flag=25).
F: Data imputed using hot-deck (flag=26).
G: Month or day was randomly assigned (flag=27).
H: Data imputed using mean value (flag=28).

Imputation
methods
E
B
B
B, E
B, E
B, E
B, E
A, B, E
B, E
B, E
B, E
A
C, D, H
A
C, D, H
C, D, H
F
F, G
F
F
F
F
F, G

Page 13 of 37

VIII.

Imputation Rates

Table 8 and Table 9 show the facility imputation rates for Section I. The facility imputation rate is
𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁 𝑜𝑜𝑜𝑜 𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓 𝑤𝑤𝑤𝑤𝑤𝑤ℎ 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 𝑓𝑓𝑓𝑓𝑓𝑓 𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 𝐼𝐼 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖
× 100.
𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁 𝑜𝑜𝑜𝑜 𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓 𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 𝑓𝑓𝑓𝑓𝑓𝑓 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑖𝑖𝑛𝑛 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓

The facility imputation rate is not necessarily the same as the facility nonresponse rate.

Table 8. CJRP Section I Item Imputation Rates by Year
Item

2003

Total persons
Adults
Juveniles
Juvenile offenders
Juvenile nonoffenders

0
0
0
0
0

Percent Imputed By Year
2006
2007
2010
2011
0
0.3
7.1
4.6
0
0.3
7.1
4.6
0
0.3
7.1
4.6
0
0.3
7.1
4.6
0
0.3
7.1
4.6

2013
7.6
7.6
7.6
7.5
7.6

Table 9. 2013 CJRP Imputation Rates for Items in Section I Not Previously
Imputed
Item
Larger agency
Own
Own level
Operator
Operator level
Treatment
Treatment type
Foster care
Independent living
Facility type
Overflow
Locked room
Locked reason
Security
Facility locked
Locked
Locked outside

Percent imputed 2013
5.0
8.0
8.0
5.9
5.9
9.1
9.1
9.2
9.5
6.5
8.6
9.1
9.1
5.6
8.9
8.9
8.9

Page 14 of 37

Table 10 shows the item imputation rates for Section II. The item imputation rate is:
𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁 𝑜𝑜𝑜𝑜 𝑗𝑗𝑗𝑗𝑗𝑗𝑗𝑗𝑗𝑗𝑗𝑗𝑗𝑗𝑒𝑒 𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜 𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟 𝑤𝑤𝑤𝑤𝑤𝑤ℎ 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 𝑓𝑓𝑓𝑓𝑓𝑓 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖
× 100.
𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁𝑁 𝑜𝑜𝑜𝑜 𝑗𝑗𝑗𝑗𝑗𝑗𝑗𝑗𝑗𝑗𝑗𝑗𝑗𝑗𝑗𝑗 𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜 𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟 𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 𝑓𝑓𝑓𝑓𝑓𝑓 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 𝑖𝑖𝑖𝑖 𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓

The item imputation rates for Section II are the same as the item nonresponse rates.

Table 10. CJRP Section II Item Imputation Rates by Year
Item
Sex
Birth month
Birth day
Birth year
Race
Placed by
Offense
Adjudication status
Admitted month
Admitted day
Admitted year

2003
2.9
10.2
10.2
10.1
11.0
2.4
12.5
4.9
9.8
9.8
9.6

Percent Imputed By Year
2006
2007
2010
2011
3.5
7.5
7.1
6.1
10.8
20.3
11.5
11.6
10.8
20.3
11.8
11.7
10.8
20.3
11.4
11.6
11.1
20.6
11.3
11.6
2.5
2.8
8.0
7.0
13.1
23.2
14.0
15.0
5.7
8.5
9.1
8.9
10.8
20.3
11.3
12.4
10.9
20.3
11.6
12.5
10.8
20.3
11.3
12.4

2013
13.2
16.6
16.6
16.7
18.5
14.0
21.3
20.6
17.1
17.1
17.9

There are 54,148 juvenile offender records eligible for imputation (offenders held in nontribal facilities
in the 50 states and District of Columbia) on the imputed file. 14,574 (26.9 percent) of those offender
records have at least one Section II item imputed.
Each item has an imputation flag on the imputed file. See the record layout for the explanation of the
imputation flag values.
PSSMB calculates the age of the juvenile offender. The age imputation flag is set to 26 or 27 when any
part of the birth date (month, day, or year) has been imputed.
PSSMB calculates the length of stay of the juvenile offender. The length of stay imputation flag is set to
26 or 27 when any part of the admission date (month, day, or year) has been imputed.
[The program tabulate_nonresponse_over_time.sas produces the tables for Section I and Section II item
imputation rates.]

Page 15 of 37

IX. Collapsed Facility Type Codes
For imputation purposes, we need to assign a collapsed facility type code (Cat) to every facility. Cat is
the variable on the 2013 CJRP file that contains the collapsed facility type code (column 1207). We only
assigned Cat codes to nontribal facilities in the 50 states and District of Columbia in the 2013 CJRP file,
because we did not impute the tribal facilities or the territorial facilities.
Please note that the method of assigning collapsed facility types for 2013 is different than it was in
previous years. The author has been the mathematical statistician responsible for imputing missing data
in CJRP beginning with the 2003 collection. The author had the same role for the 2012 JRFC. While
working with the 2013 file, the author asked why CJRP and JRFC handle the facility type question
differently. Unfortunately, there does not seem to be any surviving documentation explaining the
reason for the differences.
The following procedure assigned the 2013 Cat code:
Check to see if the facility checked at least one facility type box for the 2013 CJRP. If so, the
2013 answers were used to assign the Cat code.
If none of the 2013 facility type answers were checked, pull forward the 2012 JRFC answers, if at
least one box was checked.
If none of the 2013 facility type answers were checked, and none of the 2012 JRFC answers
were checked, pull forward the 2011 CJRP answers, if at least one box was checked.
We assigned the collapsed facility type based on the hierarchy shown in Table 11. If the facility
checked more than one box, the box listed highest in the table determines the collapsed facility
type code. If an agency checks boxes that indicate it is both a reception center and a training
school, the assigned code is training school, since training school is higher up in the hierarchy
than reception center is.
The CJB analysts provided guidance in assigning the collapsed facility type if only the 10th box,
Facility_Other, was checked.

Page 16 of 37

Table 11. Hierarchy Used to Assign Collapsed Facility Type Code
Cat

Collapsed Facility Type

3
0
2
5

Training School
Detention Center
Reception / Diagnostic Center
Ranch, Camp, or Farm

Shelter

Halfway House / Group Home

Check box on 2013 form
[Section 1 Question 9]
02 (Facility_Training_Long)
01 (Facility_Detention)
03 (Facility_Reception_Diagnostic)
06 (Facility_Boot_Camp),
07 (Facility_Ranch_Camp_Wilderness)
08 (Facility_Runaway_Homeless),
09 (Facility_Other_Shelter)
04 (Facility_Group_Halfway),
05 (Facility_Residential)

[The program assign_collapsed_facility_type.sas creates the Cat code.]

X. Imputation Methodology for Section I Data
Section I contains both check box questions and questions about population counts. The methods used
to impute Section I data are described in more detail below.
Missing data for check box questions
If the question was not answered for the 2013 CJRP, but was asked on either the 2012 JRFC or the 2011
CJRP, we pulled forward answers if they were provided on the prior year file. If no prior year data
existed for the question, we left the answer as refusal or don’t know and set the imputation flag to show
that the value is refusal or don’t know.
We analyzed the patterns of nonresponse for the check box questions before doing the imputations.
That helped us to simplify the programming. Note that if the nonresponse patterns change for the 2015
CJRP, the program that imputes the answers for the check box questions will need to be updated
appropriately.
Table 12 shows the 2013 check box questions imputed, and whether or not the item was on the prior
JRFC or prior CJRP. Note that we imputed some questions as groups, so that the skip patterns would be
preserved. Also, note that as the questions asked on JRFC and CJRP change over time, the imputation
program that imputes the answers for these questions will need to change.
The 2013 CJRP processing system did not contain flags for the check box questions. Most of the check
box questions were set so that the value of 1 indicates that the box was checked. On the “mark all that
apply” questions, we assume that if at least one box was checked, the question was answered
completely. If none of the boxes are checked, we attempt to impute an answer.

Page 17 of 37

Table 12. Section I Check Box Questions over Time
2013 CJRP Question
1a – Facility part of larger agency?
2a, 2b, 3 – Who owns this facility?
4a, 4b, 5 – Who operates this facility?
6a, 6b – Treatment provided inside facility
7 – Facility provides foster care?
8 – Facility provides independent living arrangements?
10 – Facility house overflow detention population?
11a, 11b – Juveniles locked into sleeping rooms?
12 – Facility have security features to confine juveniles?
13a, 13b, 13c – Outside doors locked?

On 2012 JRFC?
YES
YES
YES
YES
YES
YES
YES
YES
YES

On 2011 CJRP?
YES
YES
YES
YES
YES
YES
YES
YES

Missing data for the population counts
We calculated average 1-year growth rates by imputation cell for facilities that reported the population
counts (persons age 21 or older assigned to beds, offenders under age 21 assigned to beds, and
nonoffenders under age 21 assigned to beds) in both the 2013 CJRP and the 2012 JRFC.
We calculated average 2-year growth rates by imputation cell for facilities that reported the population
counts in both the 2013 CJRP and the 2011 CJRP.
The imputation cell is all facilities within a given state and Cat (collapsed facility type) code. If there are
fewer than 15 respondents or less than 70 percent response in the imputation cell, we collapse the
imputation cell to the national level.
If the population count is missing in the 2013 CJRP, but the facility has a value for the item in the 2012
JRFC, we impute the 2013 value by applying the 1-year growth rate to the 2012 value and then rounding
to a whole number.
If the population count is missing in the 2013 CJRP, and the facility does not have a value for the item in
the 2012 JRFC, but does have a value for the item in the 2011 CJRP, we impute the 2013 value by
applying the 2-year growth rate to the 2011 value and then rounding to a whole number.
If the population count is missing in the 2013 CJRP, and the facility does not have a value for the item in
either the 2012 JRFC or the 2011 CJRP, we impute the 2013 value by rounding the mean value reported
in the imputation cell in 2013 to a whole number.
If the number of persons under age 21 assigned to beds is missing, we derive the value by summing the
values of the number of offenders under age 21 assigned to beds and the number of nonoffenders
under age 21 assigned to beds.
If the total persons assigned to beds is missing, we derive the value by summing the values of the
number of persons age 21 or older assigned to beds and the number of persons under age 21 assigned
to beds.

Page 18 of 37

In all cases, the imputation flags are set to show the method used to impute the value.
[The program impute_counts.sas does the Section I population imputations. The program
impute_section_1_check_boxes.sas does the Section I imputations for the check box questions.]

XI. Imputation Methodology for Item Nonresponse in Section II Data
Changes in methodology over time
The basic methodology for dealing with item nonresponse is still the hierarchical hot-deck, used in the
2003, 2006, and 2007 CJRP collections. In 2007, we imputed juvenile offenders in tribal facilities
separately from juvenile offenders in all other facilities. Starting In 2010, we do not impute juvenile
offenders in tribal facilities or in territorial facilities. Juvenile offenders in tribal facilities or territorial
facilities are not eligible to be donors for juvenile offenders in nontribal facilities in the 50 states and
District of Columbia.
If the offense code is missing, the imputation system fills in the missing offense based on the code
provided in the juvenile offender record. Code 97 indicates an unknown offense for both underage
persons and adults, code 98 indicates an unknown offense for underage persons only, and code 99
indicates unknown offense. In previous years of CJRP, we used that missing offense code to guide the
acceptable imputed offense code imputations. See Table 13 to understand how we impute missing
offense codes.

Table 13. How We Impute Missing Offense Codes
Missing
offense
code
97
98
99

Acceptable imputed offense code
Offenses against property, offenses against persons, drug-related offenses, offenses against
the public order, or probation or parole violation (offense codes 10 through 50)
Offenses for underage persons only (offense codes 60 through 69)
Any valid offense code (offense codes 10 through 69)

Status offenders are juveniles who have committed offenses for underage persons only. The Juvenile
Justice and Delinquency Prevention Act prohibits the placement of status offenders in secure facilities in
most instances. For the purposes of CJRP, we consider any facility that checks either the detention
center or training school facility type box a secure facility. We have imputed some status offenders in
secure facilities in previous CJRP collections, but will no longer do so. To prevent this from happening in
the 2010 CJRP and future collections, we review the missing offense codes for offenders in facilities that
checked either the detention center or training school facility type boxes, and we force the missing
offense code to be 97, which means that the imputed offense will not be a status offense.

Page 19 of 37

We note that secure facilities have reported juveniles with status offenses. We do not have an edit that
verifies that a secure facility holds status offenders. In discussions with OJJDP and various stakeholders,
we decided that we would accept the reported data as is. 175 facilities in the 2013 CJRP final imputed
file reported 534 status offenders.
Randomly imputing the day of birth, month of birth, or day of admission
The first missing items that we impute are:
•
•
•

kid_admitted_day if both kid_admitted_month and kid_admitted_year are reported,
kid_birth_month if kid_birth_day is missing but kid_birth_year is reported, and
kid_birth_day if both kid_birth_month and kid_birth_year are not missing.

When a facility reports both kid_admitted_month and kid_admitted_year, but kid_admitted_day is
missing, we impute kid_admitted_day by randomly selecting a day based on kid_birth_month. This
prevents the imputation of days that do not exist, such as February 30th.
Age and stay calculations
The reference date of the questionnaire is October 23, 2013. Some facilities may report based on an
alternative reference date. If an alternative reference date is used, the alternative reference date field
contains that date (columns 1194 through 1201).
We calculate an age for all records where it is possible to do so. If the facility is reporting based on an
alternative reference date, we calculate the age of the juvenile offender as of the alternative reference
date; otherwise, we calculate the age of the juvenile offender as of October 23, 2013.
End users calculate a length of stay variable, based on the date that the juvenile offender enters the
facility. If a facility is reporting based on an alternative reference date, we calculate the length of stay as
of the alternative reference date; otherwise, we calculate the stay as of October 23, 2013. We calculate
length of stay for all records where it is possible to do so.
[The imputation of kid_admitted_day when we have reported kid_admitted_month and
kid_admitted_year, the imputation of kid_birth_month and / or kid_birth_day when kid_birth_year is
reported, and the calculation of age and stay when possible is performed in the program
create_flags.sas]
Hierarchical hot-deck imputation for item nonresponse
The imputation methodology for item nonresponse in Section II data is hierarchical hot-deck. We match
the record requiring imputation to a pool of records where none of the information is missing, and then
we select a donor record. We replace the missing values in the record requiring imputation with the
values from the donor record. We first try the match on all available information. If we do not find a
match, we make the match less restrictive until we find a donor record.
The definition of records where none of the information is missing includes those records for which we
only imputed kid_birth_month, kid_birth_day, or kid_admitted_day. These records are considered

Page 20 of 37

eligible donors because if kid_birth_year is not imputed, we have a good idea how old the offender is,
and if kid_admitted_month and kid_admitted_year are not imputed, we have a good idea how long the
offender has been held in the facility.
The available information for matching is the Cat code, the state where the facility is located, and any
reported data for kid_sex, age, kid_race, kid_placed_by, kid_offense, kid_adjudication_status, and
length of stay.
When imputing kid_adjudication_status, those records with kid_adjudication_status = 08 (convicted in
adult criminal court) are never part of the pool of potential donors. We confirmed with the sponsor that
there should not be imputed values of 08 (convicted in adult criminal court) on the final data file.
The advantage of the hierarchical hot-deck method is that imputed values should be consistent with the
rest of the juvenile offender record, because the donor record is a juvenile offender record that has
passed the edits.

XII. Imputation Methodology for Section II Data for Critical Item
Facilities
Changes in methodology over time
The basic methodology for dealing with nonresponse in critical item facilities is the same as it was in the
2003 and 2006 CJRP collections. For the 2007 CJRP, we imputed juvenile offenders in tribal facilities
separately from juvenile offenders in non-tribal facilities. Beginning with the 2010 CJRP, Census was
instructed not to impute juvenile offenders in tribal facilities. The 2010 CJRP is also the first time we are
collecting data from territorial facilities. It was decided that we would not impute juvenile offenders in
territorial facilities.
We introduced a new classification of critical item facility in 2007. We noticed in the 2006 CJRP that
some facilities would provide a roster of juvenile offenders, but not much information about the
individual offenders. If the date of birth, offense, and date of admission are missing for all the juvenile
offenders in a facility, we really do not have much information to work with.
If we try to impute those records as merely having item nonresponse, we run the risk of using the same
donor repeatedly within the facility, creating what looks like duplicate records in the facility. To
minimize that risk, we now handle such facilities like critical item facilities, and have assigned them a
code of 2 (Facility responded to Section I of the questionnaire, but date of birth, offense, and date of
admission are missing for all records in Section II of the questionnaire) in the critical item field.
We introduced a new classification of critical item facility in 2013. Some facilities refused to fill out
Section II for their juvenile offenders. Instead, the analysts collected percentage distributions for some
characteristics (percentage of offenders by sex, percentage of offenders by race, percentage of
offenders by offense code, etc.) of the juvenile offenders. Unfortunately, the imputation system was

Page 21 of 37

not designed to impute missing data this way. The analysts had to generate records that matched those
percentages, and then we created special programs to impute these records appropriately. These
records have the critical item field set to 4. If the current imputation system is used for the 2015 CJRP,
we recommend that the number of “special handling” facilities be kept as small as possible, because
they are difficult to deal with.
Background
The edited file has one record per critical item facility if the critical item field is set to 1 (Facility
responded to only the critical items) or 3 (Facility is a refusal; all data on the file for that facility has been
imputed if the facility is a non-tribal facility in the 50 states or District of Columbia).
If the critical item facility holds juvenile offenders, the Section II data on the record refers to all the
juveniles held by that facility. The analysts in CJB tried to find out as much as possible about the types of
juveniles held in critical item facilities.
If kid_sex = 1 in Section II of the critical item facility record, that means that the facility only holds males,
while kid_sex = 2 means that the facility only holds females, and kid_sex = 3 means that the facility holds
both males and females.
Some critical item facilities were unable to indicate for which types of offenses they held offenders, so
kid_offense = 88 or 99 for those critical item facilities. Some critical item facilities were able to indicate
that they held offenders for offense codes applicable to both underage persons and adults, so
kid_offense = 97 for those critical item facilities. Some critical item facilities were able to indicate that
they held offenders for those offense codes applicable to underage persons only, so kid_offense = 98 for
those critical item facilities.
We generate the required number of juvenile offender records for each critical item facility and assign
kid_id to each juvenile offender record for the critical item facility. Kid_id is a 15 character juvenile
identifier. We number the records sequentially within each critical item facility, starting with
000000000000001. We also replicate the available reported information for each juvenile offender
record within the critical item facility.
If we know that the facility only holds males or only holds females, we do not consider kid_sex imputed.
The edited file may have multiple records if the critical item field is set to 2 (Facility responded to
Section I of the questionnaire, but date of birth, offense, and date of admission are missing for all
records in Section II of the questionnaire). For example, the facility may have two sets of offenders
placed in the facility by two different types of authorities. If date of birth, offense, and date of
admission are missing for all the offenders in that facility, we do not have much information to work
with. We handle these facilities as critical item facilities rather than item nonresponse facilities to
minimize the amount of duplication in the imputed data.

Page 22 of 37

The edited file may have multiple records if the critical item field is set to 4 (Special handling is required
to impute juvenile offenders in these facilities, CJB collected percentage distributions on selected
offender characteristics). The juvenile offender records in these facilities are missing date of birth and
date of admission, and thus look similar. Because the reported values are not distinct, these records
require special handling, so that we do not impute multiple records that look like duplicates within the
facility.
Hierarchical hot-deck for critical item facilities
We modified the hierarchical hot-deck methodology used for item nonresponse for critical item
facilities. Instead of finding a matching donor pool for an individual juvenile offender record, we find a
donor pool for the critical item facility and then randomly select donors from the pool without
replacement. This modified version of the hierarchical hot-deck requires that the donor pool have at
least as many juvenile offenders as the critical item facility. This requirement ensures we do not
duplicate the imputed juvenile offender records for the critical item facility within the facility.
The available information for matching is the Cat code, the state where the facility is located, and any
reported data for kid_sex, age, kid_race, kid_placed_by, kid_offense, kid_adjudication_status, and stay.
When imputing kid_adjudication_status, those records with kid_adjudication_status = 08 (convicted in
adult criminal court) are never part of the pool of potential donors. OJJDP does not want any imputed
values of convicted in adult criminal court on the imputed file.
The advantage of the hierarchical hot-deck method is that imputed values should be consistent with the
rest of the juvenile offender record, since the donor record is a juvenile offender record that has passed
all the edits.
In the 2003 and 2006 CJRP files, we only used the top two levels of the hierarchical hot-deck for critical
item facilities. For the 2007 CJRP, we used as many as four levels of the hierarchical hot-deck for critical
item facilities. We had to use four levels in states with large numbers of juvenile offenders held in
critical item facilities where the facility reported a relatively uncommon value for who placed the
juvenile in the facility. In the 2010, 2011, and 2013 CJRP files, we only used the top two levels of the
hierarchical hot-deck for critical item facilities.

XIII.

Quality Checks Performed After Imputation

The program final_qc_check.sas runs after the imputation system, to check that all flags are properly set
and that all imputed fields have valid values. Specifically, the program checks the following:
•
•
•
•

We assigned collapsed facility type. (We use collapsed facility type as a matching variable in the
hierarchical hot deck.)
No juvenile offender is 21 or older on the final file.
Every juvenile offender has a nonnegative value for length of stay.
All character variables have valid values.

Page 23 of 37

•
•

None of the juvenile characteristics eligible for imputation is missing.
The number of juvenile offender records is the same as the number of juvenile offenders given
in Section I of the questionnaire.

•

The number of juveniles plus the number of adults is the same as the total given in Section I of
the questionnaire.
The number of juvenile offenders plus the number of juvenile nonoffenders is the same as the
number of juveniles given in Section I of the questionnaire.
The answer to “Any persons assigned to beds?” is consistent with the total assigned to beds.
The answer to “Any persons under age 21 assigned to beds?” is consistent with juvenile count.
The answer to “Any persons under age 21 assigned to beds because they were charged with or
court-adjudicated for an offense?” is consistent with juvenile offender count.
The answer to “Any persons under age 21 have assigned beds for reasons other than offenses?”
is consistent with juvenile nonoffender count.
All facility records have values for total, adults, juveniles, juvenile offenders, and juvenile
nonoffenders.
We did not impute an adult adjudication status when the offense is one for underage persons
only.

•
•
•
•
•
•
•

XIV.

Other Programming Notes

The imputation system for the 2003 through 2011 CJRP files resided on a Linux server. There were 167
SAS programs and 1 Perl program for 2011. The 2011 CJRP imputation system was run with a KornShell
Script, and took 9 minutes.
The author’s account on the Linux server was terminated in September 2013, and all files were moved
to a Windows server. The author now uses SAS in the Census Bureau’s Virtual Desktop Infrastructure
(VDI). The 2011 CJRP system was converted for use on Windows, and the author noticed that it took 30
minutes to run late in the day. If the author needs to rerun the imputations during core business hours,
the system takes at least an hour to finish.
The Perl script was used on the Linux server to automatically check all the log files and report any
problems. Perl is not available on the SAS servers in VDI, so the author had to quickly find other
solutions. The author created a SAS driver program to replace the KornShell script. The driver calls 232
SAS programs in sequence. 65 new programs were written for the 2013 CJRP, to handle new patterns
of nonresponse in Section II that had not been observed in previous cycles, or to do the special handling
of critical item facilities. Attachment D shows the nonresponse patterns for the juvenile roster data.
Note there are some patterns where only part of the birth date or part of the admission date is missing.
The SAS driver program routes all the output to an HTML file, and routes the log to an alternate file. The
author uses UltraEdit or Notepad++ to search the log quickly for any occurrences of “ERROR” or
“WARNING”. The author would like to point out that Notepad++ was extremely useful in editing

Page 24 of 37

multiple SAS programs at once to convert them from the Linux environment to the Windows
environment.
Control.sas sets up a SAS data set that stores the values for macro variables used in the find___.sas and
match___.sas programs. This arrangement makes the hierarchical hot-deck programs much easier to
use over time. Instead of hard coding the survey year or missing values for each item in Section II in the
hot-deck programs, the programs get the macro variable values from the SAS data set.
Control.sas also includes the seed for the random number generator. By storing the seed in a SAS data
set, we can rerun the imputation system at any time and get the same results. The SAS programs that
use the seed for the random number generator also update the seed and store it, so we use a different
seed in each program that needs random numbers.
The programs that create the data sets for the current year CJRP, the prior year CJRP, and the prior year
JRFC are specific to each year, so we have to edit them for each census. 2013_edited_qa.sas checks the
edited file for any unusual values before imputation. The program lists problems that need to be
resolved before imputation, such as the number of juvenile offender records for a given facility not
matching the reported number of juvenile offenders in Section I for that facility.
Juvenile_offender_item_nonresponse_patterns.sas creates a listing showing the nonresponse patterns
for juvenile offender records in facilities that reported more than the critical items.
Juvenile_offender_item_imputation_report.sas opens the file
juvenile_offender_item_imputation_report.txt. The text file shows the results of the hierarchical hotdeck from each find___.sas program.
Critical_item_kid_imputation_report.sas opens the file critical_item_kid_imputation_report.txt. The
text file shows the results of the hierarchical hot-deck from each match___.sas program.
Impute_critical_item_kids.sas generates the correct number of juvenile offender records for each critical
item facility and creates a listing showing the nonresponse patterns for juvenile offender records in
critical item facilities.
Three SAS programs run checks on the final imputed file to ensure that the imputation processing
system has successfully completed.

XV. Caution When Comparing State Data over Time
In the 2003 CJRP documentation, we noted that critical item facilities held 84.6 percent of all juvenile
offenders in DC, which meant that we imputed an unusually large percentage of the data in DC for 2003.
In 2007, critical item facilities held 95.7 percent of all juvenile offenders in DC. We do not recommend
comparing juvenile offenders held in DC facilities across the 2003, 2006, 2007, and 2010 data
collections, due to extreme levels of missing data for the juvenile offenders.

Page 25 of 37

Attachments A and B show some high levels of juvenile roster item imputation for 2013. New
Hampshire had 80 percent of their juvenile offenders in facilities that only reported critical items. States
with 30 percent or more imputation by item included Arizona, Arkansas (kid_sex, kid_birth_date,
kid_race, kid_offense, kid_admitted_date), Colorado, Florida (kid_birth_date, kid_race, kid_offense,
kid_adjudication_status, kid_admitted_date), Illinois, Nebraska, New Hampshire, Rhode Island, South
Carolina (kid_sex, kid_birth_date, kid_race, kid_offense, kid_adjudication_status, kid_admission_date),
Utah, West Virginia, and Wyoming (kid_sex).
Users should be aware the differences in DC data from 2003 to 2006 might be due in part to the high
levels of imputation for DC in 2003, and from 2006 to 2007 due to high levels of imputation for DC in
2007, and from 2007 to 2010 due to high levels of imputation for DC in 2007. Similarly, the differences
in Colorado, Illinois, Rhode Island, and Wyoming data from 2003 to 2006 may be due in part to high
levels of imputation for those states in 2006.
The differences in Arizona, Arkansas, Colorado, Connecticut, Florida, Illinois, Iowa, Maine, Mississippi,
Montana, Nevada, New Mexico, New York, South Carolina, Utah, West Virginia, and Wyoming from 2006
to 2007 may be due in part to high levels of imputation for those states in 2007.
The differences in Colorado, Florida, Illinois, Vermont, and West Virginia from 2007 to 2010 may be due
in part to high levels of imputation for those states in 2010.
The differences in Colorado, Florida, Illinois, Nevada, and West Virginia from 2010 to 2011 may be due in
part to high levels of imputation for those states in 2011.
The differences in Arizona, Arkansas, Colorado, Florida, Illinois, Nebraska, New Hampshire, Rhode Island,
South Carolina, Utah, West Virginia, and Wyoming from 2011 to 2013 may be due in part to high levels
of imputation for those states in 2013.
Attachment C shows items by state for the 1997 through 2013 CJRP data collections. If the item
imputation rate was 30 percent or more for a given year, we show the year in the cell of the table.
[The program nonresponse_by_state.sas produces Attachments A and B, while the program
2013_data_quality_concerns.sas produces Attachment C.]

XVI.

Issues to Consider for the Next CJRP Collection

Previous CJRP data collections included a field called “s2 line number”. That field was helpful to refer to
individual juvenile offender records within a facility. While the questionnaire does ask the facility to
provide an identifier for each juvenile offender record, some facilities do not provide an identifier. We
were able to use the s2 line number in previous collections to ensure that the imputation system did not
overwrite any reported data. Since the s2 line number field does not exist for 2013, we were unable to
perform that check. The analysts and respondents also find s2 line number helpful when an identifier is
not provided for the juvenile offender. We recommend including s2 line number for the 2015 CJRP.

Page 26 of 37

Attachment D shows the nonresponse patterns on the juvenile offender roster. We suspect that some
facilities may be trying to anonymize their data by withholding a part of the birth date or a part of the
admission date. It is tricky to use a hierarchical hot-deck to impute when only part of the date is
missing.
There should be some discussion of whether or not to accept percentage distributions for roster data in
the future. The CJRP imputation system was originally designed to handle item nonresponse on a
reported juvenile roster, or to impute an entire facility’s roster when the facility only reported critical
items. Special handling is required to deal with other patterns of missing data, and the system already
consists of 232 SAS programs. At some point, the imputation system becomes unmanageable.
The increase in missing data, whether due to percentage distributions, critical item facilities, or missing
parts of dates, is unsettling. Respondents may not understand the purpose of the data collection, nor
how the data are used. There seems to be understandable concern on the part of respondents to
protect the data about the juvenile offenders they house. However, respondents may not be aware that
quite a bit of policy analysis is done at an aggregate level. For example, see “Juvenile prison populations
fall as states’ changes take effect”, by Reid Wilson, published February 1, 2015 in the Washington Post
and available online at http://www.washingtonpost.com/blogs/govbeat/wp/2015/01/29/states-seemarked-drop-in-juvenile-prison-populations-as-reforms-take-hold/. We recommend doing more
outreach with the 2015 CJRP, letting the facilities know how CJRP data has been used and why it’s
important every time it’s collected.
Future CJRP collections should have a dashboard during data collection, so that the analysts can monitor
the amount of critical item facilities and the amount of missing data on the juvenile offender roster.
Substantial missing data at the state level is an issue for any sub-national analysis performed on the final
imputed file.

XVII.

Kid_offense_location revision after sponsor reviewCJB

OJJDP and the National Center for Juvenile Justice (NCJJ) reviewed the imputed file, and noticed that
kid_offense_location was missing more often on the 2013 CJRP file than in previous collections. They
directed CJB to call some facilities to see if the data could be reported. 32 facilities with some missing
kid_offense_location data reported that they only hold offenders who committed offenses in the state
where the facility is located. In addition, facilities in Florida only hold offenders who have committed
offenses in Florida. We updated the edited file and reran the imputations and the tables on June 11,
2015. While doing that work, we discovered that two offender records in Florida facilities were coded
with kid_offense_location of Hawaii, so we reset kid_offense_location on those records to Florida.
We did not impute kid_offense_location on the 2013 CJRP file, so there is still some missing data.

Page 27 of 37

XVIII.

References

Census of Juveniles in Residential Placement Databook, currently online at
http://www.ojjdp.gov/ojstatbb/cjrp/.
Dorinski, Suzanne M. “How to impute section I data,” memo dated January 8, 2004.
Dorinski, Suzanne M. “How to calculate age on individual juvenile data records in 2003 CJRP,” memo
dated February 25, 2004.
Dorinski, Suzanne M. “How to impute for item nonresponse on juvenile offender records in 2003 CJRP,”
memo dated April 1, 2004.
Dorinski, Suzanne M. “Documentation of the Imputation Methodology for the 2003 Census of Juveniles
in Residential Placement FOR INTERNAL USE ONLY,” memo dated December 1, 2004.
Dorinski, Suzanne M. “Documentation of the Imputation Methodology for the 2006 Census of Juveniles
in Residential Placement FOR INTERNAL USE ONLY,” memo dated March 22, 2007.
Dorinski, Suzanne M. “Documentation of the Imputation Methodology for the 2007 Census of Juveniles
in Residential Placement FOR INTERNAL USE ONLY,” memo dated January 15, 2009.
Dorinski, Suzanne M. “Documentation of the Imputation Methodology for the 2010 Census of Juveniles
in Residential Placement,” memo dated May 13, 2011.
Dorinski, Suzanne M. “Documentation of the Imputation Methodology for the 2011 Census of Juveniles
in Residential Placement,” memo dated June 7, 2013.
Horwitz, Sari . “From Broken Homes to a Broken System”, published in the Washington Post on
November 28, 2014, and accessible online at
http://www.washingtonpost.com/sf/national/2014/11/28/from-broken-homes-to-a-broken-system/
Wilson, Reid. “Juvenile prison populations fall as states’ changes take effect”, published February 1,
2015 in the Washington Post, available online at
http://www.washingtonpost.com/blogs/govbeat/wp/2015/01/29/states-see-marked-drop-in-juvenileprison-populations-as-reforms-take-hold/.

Page 28 of 37

Attachment A: Percentage of Juvenile Offenders Held in Critical Item
Facilities by State of Facility
(Percentages of 30 or more are highlighted)

Facility
State

Juvenile
Offenders

Percentage in
critical item
facilities

Facility
State

Juvenile
Offenders

Percentage in
critical item
facilities

Alabama

945

15.6

Montana

173

15.6

Alaska

192

7.3

Nebraska

597

38.5

1,284

42.8

Nevada

624

11.7

Arkansas

752

36.3

New Hampshire

80.0

California

8,011

11.5

New Jersey

888

1.6

Colorado

1,189

34.7

New Mexico

407

17.2

Connecticut

281

3.2

New York

1,770

8.5

Delaware

146

0.0

North Carolina

545

0.4

D.C.

199

25.1

North Dakota

171

2.3

Florida

2,799

50.8

Ohio

2,247

1.4

Georgia

1,553

4.3

Oklahoma

553

8.9

Hawaii

0.0

Oregon

1,104

1.5

Idaho

498

10.0

Pennsylvania

3,741

25.8

Illinois

1,650

46.4

Rhode Island

156

60.3

Indiana

1,593

0.0

South Carolina

672

63.4

Iowa

970

0.0

South Dakota

336

9.8

Kansas

856

5.3

Tennessee

749

11.9

Kentucky

776

0.0

Texas

4,355

3.6

Louisiana

781

1.7

Utah

698

56.7

Maine

151

0.0

Vermont

0.0

Maryland

736

10.9

Virginia

1,568

9.0

Massachusetts

509

20.6

Washington

1,015

5.9

1,779

5.2

West Virginia

654

41.9

Minnesota

985

13.1

Wisconsin

843

1.8

Mississippi

245

16.3

Wyoming

189

19.6

1,039

1.9

54,148

15.9

Arizona

Michigan

Missouri

TOTAL

Page 29 of 37

Attachment B: Section II Item Imputation Rates by State of Facility
(Imputation rates of 30 percent or more have been highlighted)
Note that these item imputation rates are also the item nonresponse rates.
Kid_birth_
Facility
State

Kid_admitted_

Kid_
Kid_
Kid_
Kid_ placed_ Kid_ adjudication_
Offenders sex month day year race
by
offense
status
month day

year

Alabama

945

8.6

15.7

15.6

10.2

13.0

10.6

16.2

Alaska

192

0.0

7.3

0.0

7.3

1.6

7.3

1,284

42.8

45.7

45.8

42.8

42.9

43.8

44.4

45.7

Arkansas

752

35.2

36.3

36.4

8.2

37.8

24.5

37.8

California

8,011

12.7

11.6

11.8

13.8

13.2

20.3

18.5

13.0

16.5

Colorado

1,189

33.9

35.3

87.0

75.6

89.7

89.0

34.7

Connecticut

281

0.0

3.2

0.0

3.2

0.0

3.2

Delaware

146

0.0

D.C.

199

20.1

25.1

22.1

25.1

22.1

25.1

Florida

2,799

9.5

54.0

51.7

11.3

58.9

51.1

56.0

Georgia

1,553

4.3

4.9

4.3

8.8

4.3

Hawaii

0.0

1.4

0.0

4.1

0.0

Idaho

498

8.6

10.0

10.2

8.6

10.0

10.4

10.0

Illinois

1,650

47.3

46.5

46.6

46.5

46.4

46.7

47.4

46.4

Indiana

1,593

0.0

1.3

0.7

0.0

4.1

9.5

1.1

Iowa

970

0.0

1.0

0.9

0.2

3.1

4.7

0.4

0.5

0.4

Kansas

856

0.8

9.0

8.9

5.6

0.6

7.5

8.8

5.6

5.4

5.5

Kentucky

776

0.0

0.6

1.9

0.0

0.6

0.4

0.1

Louisiana

781

1.7

1.9

2.0

1.8

1.7

2.6

5.2

1.7

5.2

Maine

151

0.0

0.7

0.0

Maryland

736

8.7

11.0

14.5

8.7

14.5

9.4

11.5

11.7

Massachusetts

509

13.8

20.6

16.7

21.4

25.0

21.6

1,779

5.2

5.3

8.7

8.5

10.1

15.6

9.5

9.4

8.8

Minnesota

985

13.2

13.7

13.6

13.2

14.9

16.0

14.1

14.0

14.5

Mississippi

245

16.7

16.3

16.7

16.3

21.6

16.3

1,039

1.9

2.0

2.1

4.0

2.9

5.5

11.5

1.9

2.1

Arizona

Michigan

Missouri

Page 30 of 37

Kid_admitted_

Kid_
Kid_
Kid_
Kid_ placed_ Kid_ adjudication_
Offenders sex month day year race
by
offense
status
month day

year

Montana

173

15.6

17.9

15.6

Nebraska

597

36.5

38.5

37.2

36.5

39.0

42.7

38.5

Nevada

624

11.7

15.2

11.7

80.0

81.3

100.0

80.0

New Jersey

888

1.8

1.9

1.8

2.7

1.7

2.5

2.0

2.3

17.0

New Mexico

407

17.2

17.4

17.7

18.7

17.2

1,770

6.5

8.6

8.2

6.5

19.6

15.3

8.9

8.8

8.9

North Carolina

545

0.4

0.6

0.4

6.6

7.3

0.4

North Dakota

171

2.3

2.9

2.3

6.4

2.9

2,247

2.1

1.4

1.7

9.1

1.7

8.4

2.4

1.7

1.8

553

9.8

11.8

8.9

9.2

8.9

14.8

10.8

11.8

Oregon

1,104

1.6

2.4

2.5

2.3

1.6

4.3

4.7

4.1

Pennsylvania

3,741

25.8

25.9

26.1

25.8

26.8

27.5

26.5

26.6

Rhode Island

156

60.3

60.9

60.3

63.5

60.3

65.4

South Carolina

672

37.1

67.7

63.4

13.5

63.4

59.5

63.4

South Dakota

336

9.8

12.8

9.8

10.1

Tennessee

749

11.1

12.0

11.9

11.3

11.1

13.5

12.1

12.0

13.2

4,355

3.6

5.7

6.9

5.9

8.5

10.6

4.7

698

48.9

56.7

50.7

63.0

50.7

58.9

0.0

28.0

0.0

16.0

0.0

Virginia

1,568

9.0

12.3

11.7

15.1

19.0

9.9

Washington

1,015

5.9

6.0

5.9

6.0

6.1

7.5

8.0

6.8

West Virginia

654

41.9

43.0

43.1

44.2

42.2

43.3

45.7

42.7

Wisconsin

843

1.8

1.9

1.8

3.9

7.1

2.6

2.7

Wyoming

189

40.2

16.4

15.9

16.9

15.9

54,148

13.2

16.6

16.7

18.5

14.0

21.3

20.6

17.1

17.9

New Hampshire

New York

Ohio
Oklahoma

Texas
Utah
Vermont

TOTAL

Page 31 of 37

Attachment C: Data Quality Issues in CJRP over Time
1997 through 2013 data collections
(Year is shown in table cell when imputation rate [item nonresponse rate] is 30 percent or more)
Birth
State

Sex

month

day

Admitted
year

Race

Placed
by

Offense

Adjudication
status

Alabama

month

day

year

2001

Alaska
Arizona

1999
2007

2007

2013

2007

2013

Arkansas
2013

2013

2007

2013

2007

2013

2007

2013

2001

California
Colorado

1999
2001
2006
2007

2007

2007
2010
2011

2013
Connecticut

2013

2001

2013

2001
2007

D.C.

1999
2003

2003

2007

2011

2013

1999

Delaware

1999

Florida

2007
2010
2011

2011

2013

Georgia
Hawaii

1999
2011

2013

Page 32 of 37

Attachment C: Data Quality Issues in CJRP over Time
1997 through 2013 data collections
(Year is shown in table cell when imputation rate [item nonresponse rate] is 30 percent or more)
Birth
State

Sex

Admitted
Placed
by

month

day

year

Race

Offense

2006

2007

2010

2011

2013

Adjudication
status

month

day

year

2006

2007

2010

2011

2013

2007

Idaho
Illinois

Indiana
Iowa

2007

Kansas
Kentucky
Louisiana
Maine

2007

Maryland

2007

2001
2003

Massachusetts

2001

1997

1999

2007

2013

Michigan
Minnesota
Mississippi

Missouri
Montana
Nebraska

2013

Nevada
2007

New Hampshire

2013

2007

2013

2007

2013

1997

2007

2011

2013

Page 33 of 37

Attachment C: Data Quality Issues in CJRP over Time
1997 through 2013 data collections
(Year is shown in table cell when imputation rate [item nonresponse rate] is 30 percent or more)
Birth
State

Sex

month

day

Admitted
year

Race

Placed
by

New Jersey

Offense

Adjudication
status

month

day

year

1999

2007

1997
1999

New Mexico
2007

1999

2007

New York

2007

1999
2007

2007

North Carolina
North Dakota

1999

Ohio
Oklahoma
Oregon
Pennsylvania
Rhode Island

1999
2006

2006

2013

2007

2013

1997

1999

South Carolina
2013
South Dakota
1999

2013

Tennessee

2006

2013

2007

2013

1997

1999

2007

2013

1999
2001

Texas
Utah

1999

2013
Vermont
Virginia
Washington

2007

2013

2007
2013

2013

2013
2010

Page 34 of 37

Attachment C: Data Quality Issues in CJRP over Time
1997 through 2013 data collections
(Year is shown in table cell when imputation rate [item nonresponse rate] is 30 percent or more)
Birth
State

Sex

West Virginia

Admitted
Placed
by

month

day

year

Race

2007

2010

Adjudication
status

month

day

year

2007

2011

2013

2006

2007

Wisconsin
Wyoming

Offense

1999
2006

2006

2007

2013

2006

Page 35 of 37

Attachment D: 2013 CJRP Section II Nonresponse Patterns
0 indicates a reported value, 1 indicates a missing value.
Kid_birth_
Kid_
sex

Kid_admitted_
Kid_
race

Kid_
placed_
by

Kid_
Kid_
adjudication_
offense
status

month

day

year

month

day

year

Count

Percent

39,586

73.107

454

0.838

0.004

0.009

0.002

0.020

0.146

1,367

2.525

0.006

0.015

1,219

2.251

0.006

0.002

142

0.262

280

0.517

193

0.356

0.028

115

0.212

0.046

0.139

0.018

306

0.565

0.004

0.098

128

0.236

0.002

178

0.329

Page 36 of 37

Attachment D: 2013 CJRP Section II Nonresponse Patterns
0 indicates a reported value, 1 indicates a missing value.
Kid_birth_
Kid_
sex

Kid_admitted_
Kid_
race

Kid_
placed_
by

Kid_
Kid_
adjudication_
offense
status

month

day

year

month

day

year

Count

Percent

0.061

500

0.923

116

0.214

0.083

0.004

0.002

0.006

0.002

0.013

0.002

0.004

0.002

0.028

142

0.262

0.066

0.004

0.018

0.070

0.020

0.022

0.054

0.159

0.057

Page 37 of 37

Attachment D: 2013 CJRP Section II Nonresponse Patterns
0 indicates a reported value, 1 indicates a missing value.
Kid_birth_
Kid_
sex

Kid_admitted_
Kid_
race

Kid_
placed_
by

Kid_
Kid_
adjudication_
offense
status

month

day

year

month

day

year

Count

Percent

0.011

0.004

357

0.659

1,148

2.120

0.024

0.161

0.170

0.148

0.011

0.006

0.002

138

0.255

288

0.532

0.044

6,511

12.024

54,148

100.000

File Type	application/pdf
Subject	196
Author	Suzanne Marie Dorinski
File Modified	2016-01-15
File Created	2015-06-15