2009 ASA Paper--Improving Response Rates for the 2007 Census of Government Employment

Attachment 7

GOVERNMENTS DIVISION REPORT SERIES
(Research Report #2009-8)

Improving Response Rates for the 2007 Census of Government Employment

Kerstin Edwards

Kenneth Long, Jr.

Carma Hogue

Grace O’Neill

U.S. Census Bureau

Washington, DC 20233

CITATION: Edwards, Kerstin, Kenneth Long, Jr., Carma Hogue, Grace O’Neill. 2009.
Improving Response Rates for the 2007 Census of Government Employment. Governments
Division Report Series, Research Report #2009-8

____________________________________
Report Completed: September 24, 2009
Report Issued: October 2, 2009

Disclaimer: This report is released to inform interested parties of research and to encourage discussion of work in
progress. The views expressed are those of the authors and not necessarily those of the U.S. Census Bureau.

Attachment 7

Improving Response Rates for the 2007 Census of

Government Employment

Kerstin Edwards, Kenneth Long, Jr., Carma Hogue, Governments Division

U.S. Census Bureau1, Washington, DC

Grace O’Neill2

Energy Information Administration, Washington, DC

Abstract
In March 2007 the U.S. Census Bureau conducted the 2007 Census of Government
Employment, a voluntary census of state and local governments measuring public civilian
employment and payroll. The following methods were used in an effort to bolster
response rates: training in non-response follow-up methods, targeted contact with chronic
non-respondents prior to mail-out, mail-out of a reminder letter to non-respondents one
month after the initial mail-out, substantial increase in site visits to chronic or large nonresponding governments, and use of the National Processing Center’s telephone center to
conduct non-response follow-up calls. The unit response rate improved from 76.5 percent
in 2002 to 89.3 percent in 2007.

Keywords: Response rates, survey methods, establishment surveys, non-response
1. Introduction
For the 2007 Census of Government Employment, the U.S. Census Bureau decided to
employ various methods in order to increase response rates. It is necessary to have a good
response to the Census of Government Employment in order to publish high quality data
in both the census and intervening years since census data are vital to the quality of the
imputation, estimation, and sample design of the Annual Survey of Government
Employment.
This paper will outline the methods used to increase the response rates as well as the
equations used to measure the response over time.
In Section 2, we cover the background of the Census of Government Employment.
Section 3 details the data collection methods for the 2002 Census of Government
Employment. Section 4 gives an overview of the methods for increasing response rates
for the 2007 Census of Government Employment. Section 5 outlines the methodology for

1

This report is released to inform interested parties of research and to encourage discussion of
work in progress. Any views expressed on statistical, methodological, or operational issues are
those of the authors and not necessarily those of the U.S. Census Bureau.

2

Ms. O’Neill was a member of the Census Bureau’s Establishment Survey Methods Staff at the
time this research was conducted.

Attachment 7

calculating response rates. Section 6 covers our conclusions and Section 7 suggests
further research.

2. Background
The U.S. Census Bureau conducts a Census of Governments every 5 years (years ending
in ‘2’ and ‘7’) as authorized under Title 13, United States Code, Section 161. The census
covers three major subjects: (1) Government Organization, (2) Government Employment,
and (3) Government Finance. The organization phase of the census compiles a universal
list of governments, classified by type of government (i.e., counties, municipalities,
townships, special districts, and school districts). This list of governments is then used as
a universe for the government employment and government finance phases of the census.
The government employment phase collects data on government employment and payroll
and the government finance phase collects data on government revenue, expenditure,
debt and assets.
The Census Bureau has been conducting a Census of Government Employment every
five years since 1957. In order to reduce the reporting burden on governments, in the
intervening years, the Annual Survey of Government Employment is used to collect
identical government employment and payroll data. This is sent to a sample of
approximately 1/8th of the governments in the universe. A new sample is drawn every
five years (in years ending in ‘4’ and ‘9’) and is based on the most recent census data.
The population of interest for the census and annual survey includes the civilian
employees of all Federal Government agencies (except the Central Intelligence Agency,
the National Security Agency, and the Defense Intelligence Agency), all agencies of the
50 state governments, and all local governments (i.e., counties, municipalities, townships,
special districts, and school districts) including the District of Columbia. For the 2007
Census of Government Employment, this amounted to 89,476 local governments (i.e.,
counties, municipalities, townships, special districts, and independent school districts).
There were 90,986 local governments when including dependent school districts.
There are five different types of local governments: (1) counties, (2) municipalities, (3)
townships, (4) special districts, and (5) school districts. Counties, municipalities, and
townships are known as general-purpose governments and provide general government
services (e.g., administration, corrections, parks and recreation, etc). Special districts
provide either one (single-purpose special districts) or more (multi-purpose special
districts) services and have enough administrative and fiscal autonomy to qualify as
independent governments. An example of a single-purpose special district is a library
district and an example of a multi-purpose special district is a water and sewer district.
School districts provide public elementary, secondary and/or higher education services.
School districts can be dependent on general-purpose governments or can have enough
administrative and fiscal autonomy to qualify as independent governments.
The Census of Government Employment is a voluntary census that collects information
on the number of full-time and part-time employees and their corresponding gross payroll
amounts for the pay period that includes March 12. For general-purpose governments,
colleges and universities, school districts and multi-purpose special districts, the data are
collected by government function. For example, data for school districts are collected for
instructional employees and non-instructional employees separately.

Attachment 7

This is a mail-out/mail-back self-administered paper and pencil survey with an Internet
reporting option. In addition, some governments have developed alternative reporting
arrangements, known as central collection, wherein a central source reports data for
multiple agencies in an electronic file format. See Table 1, for the response modal
distribution for state and local governments that reported to the Census of Government
Employment and for the governments that reported to the Local Government Directory
Survey (G-30), which will be discussed in Section 4.1.

Table 1: Modal Distribution for 2007 Census of Government Employment
State Governments Local Governments G-30 Respondents
Web
21.3%
29.5%
20.0%
Paper
19.8%
50.4%
79.3%
Central Collection
55.0%
1.2%
0.0%
Other
4.0%
18.9%
0.7%

3. The 2002 Census Data Collection Methods
This section gives a general overview of the methods that were used for the 2002 Census
of Government Employment in an attempt to achieve an acceptable response rate. These
methods can be contrasted with the methods that were used for the 2007 Census of
Government Employment.
The initial mail-out of the 2002 Census of Government Employment was completed on
March 19, 2002. Follow-up mail-outs were conducted on June 14, 2002 and
August 14, 2002. Survey forms were included in each mail-out.
During the organization phase of the 2002 Census of Governments, limited employment
data were collected on the Local Government Directory Surveys. Total full-time
employees, part-time employees and gross annual payroll were collected. While these
data items are slightly different from the data items on the 2002 Census of Government
Employment surveys, the data were used for the 2002 Census of Government
Employment survey in the following circumstances. On May 29, 2002, before the first
follow-up mail-out, special district governments reporting no employees on the Local
Government Directory Survey were checked in as received for the 2002 Census of
Government Employment. Also, near the end of the processing cycle, after all nonresponse follow-up efforts were exhausted, any data from the Local Government
Directory Surveys that were deemed usable, were used for the Census of Government
Employment instead of imputing data using prior year or hot-deck methods. Data were
deemed usable if they were consistent with any prior year reported data after being
adjusted to a monthly gross payroll and allocating full-time and part-time employees to
the appropriate government functions.
Although a cost estimate was obtained from the Census Bureau’s National Processing
Center (NPC) in Jeffersonville, Indiana for creating a computer assisted telephone
interview (CATI) system for non-response follow-up, it was beyond the resources
available and was therefore not used. Instead, headquarters’ staff, including a total of five
analysts and one statistical assistant, and Governments Division clerks at NPC began
making non-response follow-up telephone calls on June 7, 2002 and continued until
September 30, 2002. No automated non-response system was available at the time, and

Attachment 7

calls were made from lists printed from a database of non-respondents. Analysts trained
the Governments Division NPC clerks on appropriate non-response follow-up
procedures.
While in previous years Governments Division employed field staff to visit large nonresponding governments, in 2002 the division no longer maintained a field staff.
Therefore, site visits were not conducted with large non-responding governments.
Approximately four months prior to the close-out of the 2002 Census of Government
Employment, the supervisory staff was re-structured, thus resulting in a one-month delay
in closing out the Census and requiring analysts to focus their efforts on things other than
non-response follow-up. Overall, the unit response rate decreased from 82.2 percent in
1997 to 76.5 percent in 2002.

4. Methods for Increasing Response Rates for 2007 Census
This section describes the methods used for the 2007 Census of Government
Employment to bolster response rates: (1) collected data for special district governments
on the G-30 form, (2) targeted contact with chronic non-respondents prior to initial mailout, (3) mailed an endorsement letter, (4) mailed a reminder letter, (5) conducted nonresponse follow up training, (6) conducted site visits, and (7) contracted with the National
Processing Center’s Telephone Center to conduct non-response follow-up calls.

4.1 Collected Data for Special District Governments on the G-30 form
During the organization phase of the Census, the Local Government Directory Survey
form (G-30) was mailed to special district governments. In addition to the government
organizational data that were collected, employment and payroll data were collected. The
employment and payroll data items collected were identical to those collected on the
2007 Census of Government Employment forms (E-3 and E-7). The E-3 is mailed to
single-purpose special district governments while the E-7 is mailed to multi-purpose
special district governments. Of the 38,185 special district governments in the universe,
72.8 percent are considered single-purpose special districts and 27.2 percent are
considered multi-purpose.
The only difference between the G-30 form and the E-3 and E-7 forms was the time
period for the requested data. The G-30 form requested monthly data for October 2006
while the E-3 and E-7 requested data for March 2007. In addition, because the G-30
form’s primary purpose was to collect government organizational data, the G-30 form
collected only total employment and payroll data while the E-7 form requested
employment and payroll data by government function. For this reason, data for multipurpose special district governments that completed the G-30 form were allocated to the
appropriate government functions based on the prior year distribution.
Instructions on the G-30 form informed the governments that if they completed the
employment portion of the G-30 form, they would not receive an E-3 or E-7 form. This
was done to reduce respondent burden. In March 2007, the E-3 and E-7 forms were
mailed to all special district governments in the 2007 universe that had not provided the
employment data to the organization phase of the Census. Units reporting employment
data on the G-30 form are identified in the data file released to the public because of the
difference in reference date and the data collection instrument.

Attachment 7

4.2 Targeted Contact with Chronic Non-Respondents Prior to Mail-Out
A few months prior to the initial mail-out of the 2007 Census of Government
Employment, large, chronic non-respondent governments were identified using the
following considerations: the last year it responded, whether or not it was affected by the
2005 weather disasters in Mississippi and Louisiana, its prior year total employment, and
its sample weight from the previous sample. For example, governments in the hurricane
affected areas of Mississippi and Louisiana that have not responded since the hurricanes
were included on the list.
Starting February 7, 2007 many of these large, chronic non-respondent governments were
contacted by phone or email by analysts to inform the governments about the survey and
the importance of the census year data collection. The point of contact was established by
using any prior year contact information that was available or targeting specific
departments such as payroll or human resources. Where appropriate, an endorsement
letter was attached to the emails. In addition, analysts informed the governments of the
year of the most recent response and the current mailing address on record. This allowed
many governments the opportunity to either verify or correct the mailing address prior to
the initial mail-out.
This early contact of large, chronic non-respondent governments raised awareness of the
survey, allowing the governments to watch for the survey after it was mailed. It also
helped to improve the accuracy of the database used to pull mailing addresses.

4.3 Endorsement Letter
The initial mail-out of the 2007 Census of Government Employment was completed on
March 26, 2007. Included in this mail-out was an endorsement letter. The endorsement
letter was a letter to all governments signed by 16 directors of various government
associations such as the National Association of Counties and the United States
Conference of Mayors. This letter stated the importance of the successful completion of
the census and requested each government to participate. The hope was that either the
government itself or the employee completing the form would be a member of one of the
associations and therefore more likely to complete the survey form. Before mail-out, this
letter received a methodological expert review to ensure it was as effective as possible.

4.4 Reminder Letter
On April 26, 2007, a reminder letter was mailed to governments that had not yet
responded. The purpose of this letter was to remind non-responding governments to
complete the survey. The letter was signed by the Census Bureau director and contained
no reference to whether the survey was voluntary or mandatory. No survey form was
mailed with the letter. The letter politely requested the government to complete the
previously mailed paper survey form or to complete the survey online at a web address
provided. Each government’s UserID was provided on the letter for use in completing the
form online. In addition, the letter contained a brief paragraph explaining the purpose of
the survey and a toll-free phone number for governments to call for additional assistance.
To keep the survey fresh on the minds of the governments, the reminder letter was mailed
one month after the initial mail-out, a few days before the original deadline for
completing the survey form. After the initial mail-out period, addresses were updated
from Geography Division’s database. In cases where the initial mailing address was
incorrect, the early mailing of the reminder letter allowed for contact with the

Attachment 7

governments earlier in the cycle than in prior years. All follow-up mailings were
completed earlier in the survey cycle in an attempt to solicit responses more quickly.
For a few weeks following the reminder letter mail-out, analysts received hundreds of
phone calls per day from governments who had received the reminder letter. The calls
consisted of governments requesting new forms, questions about how to complete the
survey form, questions about how to complete the web forms, and questions on other
various needs. For governments that were leaning towards not responding to the survey
or who have a policy of not completing voluntary surveys, this provided analysts the
opportunity to explain why the data are important and answer any questions about what
data should or should not be included. Many small governments opted to report their data
over the phone and analysts were able to obtain data that otherwise might not have been
reported.

4.5 Non-Response Follow-up Training
Staff members who routinely contacted governments received formal refusal avoidance
training that better prepared them to work with reluctant governments. For the 2007
Census of Government Employment, two main groups were trained on refusal avoidance:
(1) the clerks at NPC who primarily answer incoming telephone calls about missing
forms or answer frequently asked questions about the form and (2) the analysts who
primarily place outgoing telephone calls concerning data errors and to conduct nonresponse follow-up.
Refusal avoidance training was specifically selected because it aims to increase survey
cooperation and improve data quality by focusing on basic telephone skills along with
techniques to identify and address reluctance on the part of respondents. Refusal
avoidance training is based on the principles of tailoring an interviewer’s response to a
particular respondent’s concern and maintaining interaction with the survey respondent.
The training is based on research by Groves and McGonagle (2001) which outlined four
main steps for conducting refusal avoidance training: assembling respondents’ concerns,
developing responses to those concerns, training interviewers to classify concerns, and
training interviewers to provide quick and appropriate responses. The key to this training
was to provide interviewers with several responses to the same concern and the skills
needed to apply them quickly and accurately. For more information on Refusal
Avoidance Training, see O’Neill (2007).
While the analyst training is similar to the clerk training, analysts do not typically receive
specialized training on telephone call mechanics and often benefit from the telephone
skills discussion in addition to the refusal avoidance skills. After receiving the one-day
training, both groups said that the training gave them more confidence in their jobs,
helped them better prepare for telephone calls, and helped them to build rapport with
other clerks and analysts and bond as a unit. It also provided them with the opportunity to
learn from each other and disseminate information across the whole group of clerks and
analysts involved. Clerks especially felt the training allowed for open communication
with survey managers and staff from Census Bureau headquarters, with whom they often
have little personal contact.

4.6 Site Visits
Between March and September 2007, 13 staff members (including Governments Division
employees, contracted employees, and employees from other divisions in the Census

Attachment 7

Bureau) visited 676 governments throughout the country to encourage their response.
Visit locations were selected around large, chronic non-respondent governments.
Efforts were made to schedule appointments with the appropriate government contact to
discuss possibly creating an alternative reporting arrangement or to identify a way in
which the Census Bureau could help them complete the survey form. This involved
identifying the appropriate department to contact (i.e., human resources, payroll,
personnel, etc) and requesting that an information technology (IT) staff member be
present to discuss the feasibility of setting up an automated system of reporting. When
the analyst had time between scheduled visits, drop-in visits were conducted with other
governments in the area. We found that for the most part, governments responded
positively to our visits and felt that their response was important.
Staff members prepared folders to leave with the government offices. Included in the
folder were a copy of the survey form, the director’s letter, endorsement letter, prior year
survey forms printed out for reference (if available), information on alternative reporting
arrangements (where applicable), a business card of the staff member visiting, and a selfaddressed, postage paid FedEx envelope for the easy return of the completed survey.
Of the 676 governments visited, 448 responded to the survey resulting in a cooperation
rate of 66.3 percent. Of the 676 governments visited, 184 were considered chronic nonrespondent governments. After the visits, 83 of the chronic non-respondent governments
responded to the survey resulting in a cooperation rate of 45.1 percent. The cooperation
rate is the percentage of governments that were checked-in as complete.

4.7 National Processing Center’s Telephone Center
For the 2007 Census of Government Employment, Governments Division contracted
with the NPC’s Jeffersonville Telephone Center (JTC) professional interviewers to
conduct non-response follow-up calls to most of the remaining non-respondent
governments from May 3, 2007 to July 15, 2007. The non-response follow-up calls began
soon after the initial survey completion deadline of April 30, 2007 to keep the survey
fresh on the minds of the governments and to attempt to make any necessary corrections
to addresses before the final form follow-up mail-out on June 1, 2007. Due to other
obligations of JTC, the interviewers needed to finish by July 15, 2007.
Analysts created training manuals and worked with the JTC management to modify the
existing headquarters telephone follow-up application to be used by the interviewers.
This application provided a list of governments to call and allowed the telephone
interviewers to record notes about the call and the action that was taken. It also enabled
survey managers to generate reports to assist in managing the project. Analysts spent
several days training more than 100 telephone interviewers, researchers, and supervisors
on the details of the survey and the telephone follow-up procedures.
The interviewers were instructed to call to find out if the government unit received a form
and when they might be able to complete it. If multiple calls to a government went
unanswered, then the JTC researchers searched for a new phone number. If a government
unit had no employees or would only report total line data, interviewers were instructed
to take data over the phone. Interviewers continued calling until the unit reached a
resolution. A unit was considered resolved if the data were received, the government unit
representative refused to complete the form, or the unit was referred to analysts.

Attachment 7

Interviewers were able to email, fax, or mail requested information to the governments as
needed.
Interviewers called school districts first in an attempt to contact them before summer
break. After the school districts were completed, they called special district governments
and then general-purpose governments.

5. Calculation of the Response Rates
In addition to employing various methods of increasing response to the 2007 Census of
Government Employment, a more accurate response rate calculation methodology was
used than had been used in 2002. This section shows the new equations used to calculate
the response rates for the 2007 Census of Government Employment and the differences
between the response rates calculation methodology from the 2002 Census of
Government Employment.
The unit response rates for the 2007 Census of Government Employment were calculated
as described in Census Bureau Standard S18-0_v1.4, Response Rates Definitions. In
particular, we used the response rate definitions given in Supporting Document B, S182_v1.6_Requirements_Economics. In Supporting Document B, three response rate
definitions are given for use in calculating response rates on Census Bureau surveys of
establishments: the Unit Response Rate, the Quantity Response Rate, and the Total
Quantity Response Rate. For this census only the Unit Response Rate and Total Quantity
Response Rate were published, so we will only cover those two rates in this paper.
The Unit Response Rate (URR) is the proportion of reporting units based on unweighted
counts, that responded and were eligible, or of unknown eligibility, for the census
(expressed as a percentage).
URR = [R/(E+U)] * 100
where
R

Number of reporting units selected for the sample that were eligible for
data collection and classified as a respondent

E

Number of reporting units selected for the sample that were eligible for
data collection

U

Number of reporting units selected for the sample for which eligibility
could not be determined

For the 2007 Census of Government Employment, a unit was classified as a respondent if
it responded to full-time employees, full-time pay, part-time employees, and part-time
pay for at least one government function on the questionnaire.
The Total Quantity Response Rate (TQRR) for data item t is the part of the estimated
(weighted) total (T) of data item t that was reported by tabulation units in the sample or
from sources deemed to be of equivalent-quality-to-reported data (expressed as a
percentage). [Note: Since the value of economic data items can be negative, the absolute
value must be used in the numerators and denominators in all calculations.]

Attachment 7

⎡
nT
⎢ ∑ wi ×
(
rti +
qti )
×
t i
TQRR
=
⎢ i =1
T

⎢
⎢
⎣

⎤

⎥
⎥ × 100

⎥
⎥
⎦

where
wi

weight of the i-th unit in the sample

rti

indicator variable for whether unit i provided a response to item t

qti

indicator variable for whether unit i was equivalent-quality-to-reported
data for item t

ti

value of variable t for unit i

In an effort to become compliant with the response rates standard in the dissemination of
response rates for the 2007 Census, new edit and imputation flags for this survey were
assigned to distinguish exactly how the data were processed. More edit situations were
considered in the new assignment of flags. Some analyst corrections, which are actually
imputations, previously carried edit flags that made it impossible to distinguish the
imputation from a response. Also, between 2002 and 2007, the definition of a respondent
changed. In 2002, a unit was a respondent if it responded with any usable data.
Consequently, response rates for 2002 are not comparable to the response rates for 2007.
The overall unit response rate, URR, for 2007 was 88.5 percent. TQRR, which is item
specific, is available on the website for the 2007 Census of Government Employment for
local governments data at http://www.census.gov/govs/apes/index.html.
In our publications, we display the URR and TQRR for 2007, but for this paper we also
calculated a response rate using the 2002 response rate definition (a unit was a respondent
if it responded with any usable data) to show the improvement in response rates from
2002 to 2007. For response rates that can be compared over time from 1997 to 2007, see
Table 2. These response rates rose from 76.5 percent in 2002 to 89.3 percent in 2007 with
the greatest gain being recorded for special district governments.

Attachment 7

Table 2: Response Rates for the Census of Government Employment from 1997 to 2007
Type of
Government Total

2007

2002

Number of Response
respondents
rate

Total

1997

Number of Response
respondents
rate

Total

Number of Response
respondents
rate

Total with
Dependent
School
Districts
Dependent
School
Districts

90,986

81,214

89.3%

89,021

68,128

76.5%

88,905

73,043

82.2%

1,510

1246

82.5%

1,496

1,094

73.1%

1,452

1,131

77.9%

Total without
Dependent
School
Districts
Counties

89,476

79,968

89.4%

87,525

67,034

76.6%

87,453

71,912

82.2%

3,033
19,492
16,519

2,590
17,524
13,007

85.4%
89.9%
78.7%

3034
19,429
16,504

2268
15,048
11,807

74.8%
77.5%
71.5%

3,043
19,372
16,629

2,473
16,490
13,128

81.3%
85.1%
78.9%

Special
Districts

37,381

35,283

94.4%

35,052

27,307

77.9%

34,683

28,520

82.2%

Independent
School
Districts

13,051

11,564

88.6%

13,506

10,604

78.5%

13,726

11,301

82.3%

Cities
Townships

Source: U.S. Census Bureau, Census of Government Employment, 1997-2007. For
additional information, see: < http://www.census.gov/govs/apes/index.html >

6. Results
Because the methods were implemented in succession, without associated control groups,
the individual effect of each method cannot be isolated. For example, for the 2007 Census
of Government Employment non-response follow-up calling began one week after the
mail-out of the reminder letter. In the month following, 15,921 forms were returned. It is
impossible to say whether those forms were returned due to the reminder letter, the nonresponse follow-up calling, both, or neither.
However, it is possible to track when each unit was returned and compare the percentage
of returned forms over time for 2002 and 2007. Figure 1 shows the percentage of returns
over time along with the corresponding survey activities over time. Note that due to the
difference in how special district governments and many state agencies were collected
between the 2002 Census of Government Employment and the 2007 Census of
Government Employment these units were excluded from the data in Figure 1.

Attachment 7

In 2002, the early return rate was very good but it levels off between 37 days and 86 days
after the initial mail-out. It is interesting to note that no non-response follow-up activities
were conducted until 79 days after the initial mail-out in 2002. While in 2007, all the
planned non-response follow-up activities were started within 66 days after the initial
mail-out and instead of leveling off, the percentage of returned forms continued to climb
through-out the observation period. In addition, in 2007, 154 days after the initial mailout of the survey form, 99.2 percent of forms that were going to be returned were
returned. However, at the same point in 2002, only 91.2 percent of forms that were going
to be returned were returned. This suggests that repeated contact with non-respondents
using multiple follow-up methods is effective in gaining responses (see Dillman et al.,
2009, for a summary of literature on multiple contacts).

7. Conclusion
For the 2007 Census of Government Employment, the following methods were used in an
effort to bolster response rates: collected data for special district governments on the G30 form, targeted contact with chronic non-respondents prior to mail-out, mail-out of an
endorsement letter with the initial mail-out, mail-out of a reminder letter to nonrespondents one month after the initial mail-out, training in non-response follow-up
methods, substantial increase in site visits to chronic or large non-responding
governments, and use of the National Processing Center’s telephone center to conduct
non-response follow-up calls. The unit response rate improved from 76.5 percent in 2002
to 89.3 percent in 2007.

Attachment 7

With the exception of collecting data for special district governments on the G-30 form,
each of the methods described in this paper could be implemented by other economic
surveys to offset declining response rates. It should be noted, that while each of these
methods has an associated cost, it is our opinion that the substantial increase in response
rates was worth the cost of implementing the procedures. In addition to raising the overall
response rate, we were able to obtain data from 58.5 percent of the governments that
were previously identified as chronic non-respondents (including governments that we
did not have the opportunity to visit). Since imputed data are less precise the longer the
time period from the last year of reported data, we believe that obtaining data from these
specific governments has improved the quality of our estimates.

8. Further Research
The new sample for the 2009 Annual Survey of Government Employment has fewer
township and special district governments than in the prior samples. This will allow for
more time and effort to be spent in increasing the response rates for these types of
governments in the intercensal survey years. In the future, we plan to examine the state
by type of government item response rates more closely to determine where questionnaire
item wording may need to be changed to boost item response. We will also be able to
target non-response efforts by state and type of government. In addition, we will do a
non-response bias study if time and resources permit.

References
Dillman, D.A., J.D. Smyth, and L.M. Christian (2009). Internet, Mail, and Mixed-Mode
Surveys: The Tailored Design Method, John Wiley & Sons, New York.
Groves, R.M., McGonagle, K.A., (2001). “A Theory-Guided Interviewer Training
Protocol Regarding Survey Participation.” Journal of Official Statistics, 17, 249265
O’Neill, Grace E. (2007) “Improving the Effectiveness of Interviewer Administered
Surveys through Refusal Avoidance Training.” Proceedings of the 3rd
International Conference on Establishment Surveys, American Statistical
Association, Alexandria, VA.
U.S. Census Bureau. “Standard on Response Rate Definitions,” S18-2_v1.4,
, June 21, 2007.
U.S. Census Bureau. “Standard on Response Rate Definitions, Supporting Document B –
Variables, Rates, and Formulae for Calculating Response Rates and Reporting
Requirements: Economic Surveys and Censuses,” S182_v1.6_Requirements_Economic,
, December 23, 2008.