Document Name: Resource Manual for Customer Surveys Part 4
Title: Resource Manual for Customer Surveys Part 4
Date: October 1993
4. SOME FURTHER CONSIDERATIONS
Most of the 12 steps that are described in Section 3 are relevant
to all surveys, and if you have conducted surveys on other
topics, they are familiar to you. In conducting customer
surveys, however, you may encounter some issues that you have not
faced before, even if you have conducted many surveys.
This section contains short descriptions of these problems, in an
attempt to forewarn you of potential difficulties as you progress
through the survey process. Because of the need to prepare this
manual quickly, there has been time to discuss only a few of
these even briefly (see Appendix). Future updates of the manual
will have additional appendices discussing these and other issues
in greater depth.
Repeated Surveys of Similar Design Facilitate Measures of Change
in Levels of Satisfaction
The real leap in information about your agency's relationships
with its customers comes by seeing how customer satisfaction
changes over time in response to management decisions.
Measurement of change in levels over time involves sample designs
that are compatible at both times, a questionnaire that has
measures that are comparable, levels of participation in the
survey that are similar, and statistical analyses that directly
measure change in the same statistics. All of these issues, on
first experience, contain knotty problems. Many of the problems
involve finding techniques that allow you to improve the quality
of the second survey by learning from your mistakes in the first,
while retaining design features that make the two surveys
comparable. Researchers who are experienced in using surveys to
measure change can help you with these issues.
Building Customer Lists is Difficult for Some Agencies
Agencies that don't have direct contact with their customers may
have difficulty even identifying their customers. They may find
that because there are intermediaries between the agency and the
ultimate customers, the agency has no natural way of contacting
them. This is both a conceptual problem and a logistical
problem. You have to decide how important to the agency are the
customers who are difficult to enumerate. If you include them in
the target population, you may have to develop ways to measure
their satisfaction that are unusual. Here survey methodologists
with long experience in doing surveys on populations without
conveniently available sampling frames can be helpful.
Using Multiple Questions to Measure Satisfaction is Important
Because satisfaction held by a customer cannot be observed
objectively but is an internalized state containing several
components, most private sector customer surveys ask the customer
many different questions, all of which are viewed as slightly
different indicators of the same overall concept of satisfaction.
This "multiple indicator" approach has been found to improve the
reliability of satisfaction levels measured by surveys. Indeed,
most models of measurement suggest that the more questions used
to measure satisfaction, the more stable or reliable the results
will be. Two problems arise from a multiple indicator approach:
first, how many different questions should be used to measure
satisfaction (with each added question, the length of the
questionnaire increases) and how do you combine statistically the
various questions into a useful measure of satisfaction?
Statisticians, especially those experienced in attitudinal
surveys, scale construction, and multivariate modeling with
multiple indicators, can be helpful in guiding decisions on these
Respondents Tend to Answer Positively to
Customer Satisfaction Questions
Private sector customer surveys often find that customers tend to
overstate their levels of satisfaction. This produces skewed
distributions for satisfaction measures, with the vast majority
of respondents giving positive ratings. Analysis on skewed
variables cannot use traditional normal distribution theories for
statistical inference to the full customer population. You need
to use techniques that are sensitive to these distributional
The tendency to overreport satisfaction also produces a problem
for the measurement of change in satisfaction levels over time.
Because of the tendency to overestimate positive sentiments, it
becomes increasingly difficult for levels of customer
satisfaction to show large increases over time. A law of
diminishing returns affects estimates of change. Private sector
researchers have found that this issue is ameliorated somewhat by
use of many different questions on satisfaction that vary in
their tendency to achieve very high ratings (the multiple
indicator approach mentioned above).
A related phenomenon to the tendency to answer positively is the
problem of a "halo" effect, whereby customers answer each
individual question by giving their overall impression of the
agency rather than assessing the particular attribute of the
product or service measured by a question. This produces
inflated correlations among different satisfaction items, that
decreases the value of any one measure.
There are Many Choices for Response Scales in Satisfaction
Some customer surveys ask the respondents to report their
satisfaction on a 5 point scale with each point labelled (e.g.,
"Not at all satisfied" to "Very satisfied"). Others ask the
respondent to use a scale from 0 to 10, with 0 meaning
"Completely dissatisfied" and 10 "Completely satisfied." Others
use 7 point scales, or 100 point scales. The two decisions, on
the number of scale points to use and how to label the points,
affect the answers the respondents give and the subsequent
overall measures of satisfaction.
There is a large literature on how to construct attitude scales
that can alert you to how to avoid unintended biasing of
responses (see Bibliography). No matter what response scale is
chosen, however, you cannot safely use the answers from a single
question on, say, a 7 point scale, as if it were a simple count
of some uniform "satisfaction unit." The mean rating on a single
7 point scale, for example, is likely to have very low
reliability (another reason to use the techniques of scale
construction and multiple indicator analysis mentioned above).
High Response Rates for Customer Surveys Improve the
Credibility and Usefulness of Results
When customers sampled in a satisfaction survey do not
participate in the survey, the survey results are threatened.
This is particularly troublesome when nonresponse is higher among
certain types of customers (e.g., if those who are mildly
positive to the agency choose not to respond but those who are
very dissatisfied or very satisfied do respond). The results of
the survey may give a very distorted picture of agency
performance among its current customers. No survey achieves a
100% participation rate, but efforts to assure that customers in
different important groups (e.g., by frequency of contact with
the agency, by demographic characteristics, by type of
product/service used) participate at the same rate are important.
In doing this, use non-threatening follow-up and
"respondent-friendly" solicitation of participation (both for
ongoing relationships with valued customers and to avoid
distortion of answers to the satisfaction questions).
New Developments in Customer Surveys Are Occurring Rapidly
Methods to craft satisfaction measurements, conduct the surveys,
and analyze the results are undergoing rapid developments, as the
private and academic sectors learn how to improve techniques.
Continuous improvements in your agency's customer surveys will
require your staff to keep up with these developments. Updates
of this manual will highlight those developments and training
opportunities on survey techniques can instruct your staff in how
to implement new methodologies.
5. EFFICIENTLY MANAGING THE
REVIEW OF SURVEY PLANS
5.1 Public and Private Sector Surveys
The primary differences between government surveys and private
sector surveys are the standards and the oversight process
imposed by the Paperwork Reduction Act (PRA). In the private
sector, decisions concerning surveys are made by individual
entrepreneurs and any specific discipline imposed on the decision
process varies considerably within the overall discipline imposed
by the marketplace ("what the market will bear").
The discipline imposed by the PRA tends to limit the number
and/or size of surveys and to set certain thresholds on the
quality of survey design and implementation (depending on their
use). This discipline is enforced by agency and OMB review prior
to data collection, and by close scrutiny of many results after
the fact by the General Accounting Office, the Congress, and the
This discipline has produced some important advantages --
significantly higher response rates often are achieved in
government surveys -- but the oversight requirements have, on
occasion, tested the ability of the Federal government to make
surveys a responsive tool for public policy. Without some
attention to assuring an efficient review process, PRA review
could become a significant barrier to the rapid development of
customer surveys. This section spells out several proven methods
to make the process more efficient and describes a new option
developed to support surveys called for by E.O. 12862.
Congress provided one mechanism in the PRA to streamline the
review process -- the assumption of more substantial review
responsibilities by individual agencies through the delegation
authorized in the Act. This mechanism has not been attractive to
most agencies for several reasons:
- The duplication and coordination objectives of the PRA are
difficult to achieve outside the centralized environment of
- Agencies have found it difficult to justify the commitment of
scarce resources for an independent and sometimes highly
technical in-house review.
- Congress designed the PRA delegation process with several
difficult hurdles (specific determinations and a Notice and
Comment rulemaking procedure) to protect the public's
interests in minimizing burden.
5.3 Less Difficult Alternatives
In cooperation with Federal agencies, OMB has devised other
methods to streamline the review process. The most successful of
these devices have been "bundled," "contingency," and "generic"
clearances. Each of these approaches has been in use for more
than a decade, but they have been continuously refined under the
- The bundled clearance involves combining many similar data
collections into a single clearance package. Such bundled
packages have been negotiated with agencies in cases where
combined review reduced demands on both OMB and agency
resources (where similar data collections present common
- A contingency clearance is an approved plan for a data
collection that is justified by specific events -- the plan
is approved in advance of the events and can be implemented
immediately if those events take place, e.g., a survey to
track consequences of a strike. Advance review and approval
permits agencies to respond quickly to the need for data.
- Generic clearance also involves advance approval, but of a
well-defined class of low-burden data collections that are
not fully documented until they are actually used. A
generic clearance typically includes a set of agreements
negotiated between the sponsoring agency and OMB, covering
limitations on methods and usage, a burden cap, a periodic
reporting requirement to update the OMB Docket, and a
commitment by OMB to review any specific application
Many excellent customer surveys have been reviewed and approved
through the existing clearance process (some examples are
mentioned in the NPR report). However, OMB expects a substantial
increase in the number of customer surveys sponsored by Federal
agencies within the next few years, as a result of Executive
In order to make customer surveys more responsive tools for
agency management, OMB has proposed several types of generic
clearance to expedite the data collection tasks involved in a
program of customer surveys.
Each agency subject to the PRA has a senior agency official and
associated staff responsible for functions specified in the PRA,
including internal review of clearance requests and coordination
with OMB. These resources are available to assist you in
preparing proposals for generic clearance and any other aspects
of PRA review.
The remainder of this section describes generic clearance models
that may be useful for particular types of studies needed at
various stages of the customer satisfaction measurement process.
5.4 Generic Clearance for Qualitative Studies
This model has been used by the Census Bureau for questionnaire
development and testing, by the Bureau of Labor Statistics for
cognitive laboratory experiments, and by the Internal Revenue
Service for a program of focus groups. The terms of the
agreements worked out with these agencies have proven workable
for both the agencies and OMB. The terms include --
- A burden cap. The agency proposes a total annual burden
that will be imposed by studies conducted under the generic
clearance. Individual applications are tracked against the
- Specified methods. The agency proposes the type(s) of data
collection(s) to be performed and the method(s) to be used,
with particular attention to those features and commitments
that assure consistency with the guidelines of the Paperwork
Rule (5 CFR 1320).
- A periodic reporting requirement. This allows tracking of
performance relative to the burden cap and updates the
public docket by demonstrating actual results achieved. The
frequency of such reports is negotiable.
- Quick-turnaround OMB review of specific applications. The
agency submits information on each specific application to
update the public docket prior to each actual data
collection. OMB agrees to a quick- turnaround review of each
submission (this varies from same day in the simplest cases
to two weeks in more complex cases.)
5.5 Generic Clearance for Quantitative Surveys
Quantitative surveys usually must meet more stringent standards
than qualitative studies and are more likely to be tailored to a
specific task. For these reasons, the models for generic
clearance are fewer and vary from agency to agency. Two examples
that have been operating for several years were developed with
the National Park Service and the Internal Revenue Service. Both
examples include a burden cap as described above, but they differ
in most other details.
The National Park Service model is built around a catalog of
tested questions covering a broad spectrum of issues involved in
operating a national park. The clearance also includes several
approved methods (sample designs) for administering the questions
to respondents. The components of this scheme were developed
with considerable effort and extensive consultation with OMB to
provide flexibility to the local managers of national parks.
Using this tool, managers can quickly assemble surveys in "kit"
form to address current problems and charge the reporting burden
against the burden cap of the generic clearance.
The IRS model supports the agency's program of customer
satisfaction measurement. It stipulates specific methods,
including professional design, adequate follow-up, and a
commitment to high response, that assure high quality statistics.
The model covers opinion questions only and includes steps to
ensure that response is perceived as purely voluntary. The other
features are identical to the qualitative clearance model
described above. (In fact, both qualitative and quantitative data
collections are managed in a single generic clearance.)
5.6 Simplified Generic Clearance for Voluntary Customer Surveys
In response to the recommendations of the National Performance
Review, OMB has developed a new simplified generic clearance
model specifically for voluntary customer surveys. This
simplified approach is possible because the conditions proposed
by NPR eliminate many of the issues that might otherwise require
a more extensive review.
This form of generic clearance is available only for strictly
voluntary collections of opinion information from clients who
have experience with the program that is the subject of each data
This option may not be used, for example --
- by regulatory agencies to survey regulated entities;
- in any situation where the respondent may perceive risks to
his interests, either through potential penalties or loss of
- for collecting factual information (other than simple
identifying information, where needed); or
- for collecting data from the general public.
Surveys of former customers or discouraged customers may well be
useful for discovering sources of customer dissatisfaction, but
such surveys may also involve some difficult statistical issues
(e.g., the adequacy of coverage) that require more extensive
review; hence, they are not included here.
Agency proposals for this simplified generic clearance should
include a description of the kinds of customer surveys the
clearance will cover, as well as the agency program(s) they will
address. The clearance request should cite the authority of
Executive Order 12862, and request a three year expiration date
(since these data collection programs are expected to support a
process of repeated measurement).
The request should propose a maximum number of burden hours (per
year) against which burden will be charged for each survey
actually used. It should also include an arrangement to submit a
brief summary of objectives, specific burden estimates, and all
final or near final survey instruments (focus group scripts, test
questions, etc.) covered by the generic clearance for inclusion
in the OMB public docket prior to their use.
The proposal should specify an adequate internal review process
to ensure that individual applications are consistent with the
PRA, the Paperwork Rule, and the terms of the generic clearance.
This requires qualified reviewers who are independent of the
sponsoring programs. Review by a professional statistician may be
needed in some cases (e.g., if the generic clearance will include
quantitative surveys). This review must also assure that
material submitted for the public docket is accurate, timely, and
Finally, the agency should propose an appropriate progress
reporting schedule (e.g., at one year intervals) for summarizing
actual burden, reporting results achieved, and addressing any