Non-response in survey research: a methodological discussion and development of an explanatory model K. Louise Barriball BA RGN Lecturer and Alison E. While BSc MSc PhD RGN RHV Cert Ed Professor of Community Nursing, Research in Nursing Studies Section, The Florence Nightingale School of Nursing and Midwifery, King’s College London, London, England Accepted for publication 26 November 1998 BARRIBALL BARRIBALL K.L. K.L. & WHILE A.E. (1999) WHILE A.E. (1999) Journal of Advanced Nursing 30(3), 677–686 Non-response in survey research: a methodological discussion and development of an explanatory model This paper demonstrates that the different sources of non-response in survey research must be considered by researchers in order to minimize the potential for bias, and that careful planning and management during sample selection, sample recruitment and data collection can control the extent of response failure. The findings of an interview survey of nurses and nurses’ aides illustrate, however, that the success of strategies designed to reduce the loss of data in survey research depends upon researchers acknowledging the complex relationships that exist between non-response phenomena throughout the research process. Keywords: data collection, data loss, non-response, nursing research, response failure, sample recruitment, sample selection, survey research INTRODUCTION The aim of this paper is threefold: to examine some of the implications of non-response in survey research; to discuss how non-response can be contained and managed; and to highlight the complexity of non-response. The literature addressing non-response is reviewed prior to a discussion of the strategies employed by a research team to reduce non-response in an interview survey exploring the views and perceptions of continuing professional education (CPE) among nurses (including midwives and health visitors) and nurses’ aides. An explanatory model of data loss in survey research based upon the study’s findings is presented. The significance of non-response While the survey offers a data collection method which may yield greater validity through the use of more representative samples, the potential for non-response is a major disadvantage (Miller 1991). If response rates are low and/or non-response is systematic and in some way correlated with the variables under investigation, the sample from which data are collected becomes unrepre- sentative (Hartman et al. 1985). In such instances external validity is threatened and valid conclusions from the data cannot be drawn (Williamson 1981, Denzin 1989). The methodological interest in non-response therefore arises from its potential to introduce bias into the research process since: … the chances are good that those subjects who find their way into the research will differ appreciably from those subjects who do not. (Rosenthal & Rosnow 1969 p. 69) Correspondence: K. Louise Barriball, Research in Nursing Studies Section, The Florence Nightingale School of Nursing and Midwifery, King’s College London, James Clerk Maxwell Building, Waterloo Road, London SE1 8WA, England. Ó 1999 Blackwell Science Ltd 677 Journal of Advanced Nursing, 1999, 30(3), 677–686 Methodological issues in nursing research
10
Embed
Non���response in survey research: a methodological discussion and development of an explanatory model
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Non-response in survey research:a methodological discussionand development of an explanatory model
K. Louise Barriball BA RGN
Lecturer
and Alison E. While BSc MSc PhD RGN RHV Cert Ed
Professor of Community Nursing, Research in Nursing Studies Section,
The Florence Nightingale School of Nursing and Midwifery,
King's College London, London, England
Accepted for publication 26 November 1998
BARRIBALLBARRIBALL K.L.K.L. & WHILE A.E. (1999)WHILE A.E. (1999) Journal of Advanced Nursing 30(3), 677±686
Non-response in survey research: a methodological discussion
and development of an explanatory model
This paper demonstrates that the different sources of non-response in survey
research must be considered by researchers in order to minimize the potential for
bias, and that careful planning and management during sample selection, sample
recruitment and data collection can control the extent of response failure. The
®ndings of an interview survey of nurses and nurses' aides illustrate, however,
that the success of strategies designed to reduce the loss of data in survey
research depends upon researchers acknowledging the complex relationships
that exist between non-response phenomena throughout the research process.
Keywords: data collection, data loss, non-response, nursing research,
response failure, sample recruitment, sample selection, survey research
INTRODUCTION
The aim of this paper is threefold: to examine some of the
implications of non-response in survey research; to
discuss how non-response can be contained and managed;
and to highlight the complexity of non-response. The
literature addressing non-response is reviewed prior to a
discussion of the strategies employed by a research team
to reduce non-response in an interview survey exploring
the views and perceptions of continuing professional
education (CPE) among nurses (including midwives and
health visitors) and nurses' aides. An explanatory model
of data loss in survey research based upon the study's
®ndings is presented.
The signi®cance of non-response
While the survey offers a data collection method which
may yield greater validity through the use of more
representative samples, the potential for non-response is
a major disadvantage (Miller 1991). If response rates are
low and/or non-response is systematic and in some way
correlated with the variables under investigation, the
sample from which data are collected becomes unrepre-
sentative (Hartman et al. 1985). In such instances external
validity is threatened and valid conclusions from the data
cannot be drawn (Williamson 1981, Denzin 1989).
The methodological interest in non-response therefore
arises from its potential to introduce bias into the research
process since:
¼ the chances are good that those subjects who ®nd their way into
the research will differ appreciably from those subjects who do not.
(Rosenthal & Rosnow 1969 p. 69)
Correspondence: K. Louise Barriball, Research in Nursing Studies Section,
The Florence Nightingale School of Nursing and Midwifery,
King's College London, James Clerk Maxwell Building, Waterloo Road,
London SE1 8WA, England.
Ó 1999 Blackwell Science Ltd 677
Journal of Advanced Nursing, 1999, 30(3), 677±686 Methodological issues in nursing research
As Moser and Kalton (1971) cautioned, one should
never assume that differences between respondents and
non-respondents do not exist.
Towards a de®nition of non-response
Perhaps the most familiar understanding of non-response
is the degree to which a researcher does not succeed in
obtaining the co-operation of all potential respondents
included in the net sample (i.e. excluding individuals not
meeting the sampling criteria and containing only those
from whom one intends to obtain information). However,
this commonly used and apparently straightforward
de®nition is ambiguous. It also belies the complexity of
non-response phenomena. There are many potential
sources of non-response and it is important that these
are fully considered by researchers (Kviz 1977).
Steeh (1981), for example, identi®ed two major
components of non-response: refusals of potential
respondents to co-operate in the research process and
other reasons why data are not collected from all units in
the sample such as non-contact. In his paper examining
non-response, Aiken (1988) also addressed the problem of
refusals but included in his de®nition the problem of
non-coverage and the potential bias which the exclusion
of individuals from the sampling frame can introduce. In a
comprehensive review of non-response for the Of®ce of
Population Census and Surveys in Britain, Elliot (1991
p. 3) distinguished between three types of non-response:
· Non-coverage: when the sampling frame omits some
units of the survey population either accidentally or
deliberately.
· Unit non response: when no information is collected
from a sampled unit due to, for example, refusal or
non-contact.
· Item non-response: when the sampled unit agrees to
participate in the study but information on all the areas
under investigation is not collected because, for
example, the sampled unit refuses or is unable to
answer a particular question or the researcher fails to
ask the question by mistake.
Using Elliot's threefold typology as the basis for a
standard de®nition of non-response phenomena, it can be
seen that non-response can occur at different stages of the
research process, namely during sample selection, sample
recruitment and data collection. These three sources of
response failure, however, can be manipulated by
researchers to minimize the potential for non-response
bias (Figure 1) and strategies to achieve this are discussed
together with the usefulness of a linear conceptualization
of non-response.
THE STUDY
Two main approaches to address the issue of non-
response phenomena were adopted in the study. First, a
range of strategies were employed which the research
team anticipated would reduce non-response during the
course of the study. Secondly, the impact of the loss of
data at each stage of the research process on the
representativeness and the generalizability of the data
were assessed.
Figure 1 Literature based framework of non-response data loss in survey research.
K.L. Barriball and A.E. While
678 Ó 1999 Blackwell Science Ltd, Journal of Advanced Nursing, 30(3), 677±686
Non-coverage
It is usually not possible to obtain data from an entire
population which means that survey researchers need to
select and recruit samples. However, including some
individuals in a sampling frame while excluding others
runs the risk of introducing bias. A number of approaches
to ensuring the selection of representative samples have
been developed (e.g. strati®ed, random, cluster and census
sampling techniques, see Moser & Kalton 1971) and it is
the responsibility of all survey researchers to consider
carefully the approach or combination of approaches to
sample selection which will minimize non-coverage bias
within a given research situation.
The study designA census sample of randomly selected clinical units in
two district health authorities (Trust status had not been
granted at the time of data collection) was considered to be
the best study design to enhance the representativeness of
the sample and to generate a suf®cient sample size. The
clinical units were randomly selected by the research team
to represent the specialities of paediatrics, general med-
icine and surgery, care of the older adult, mental health
and learning disabilities as well as maternity and com-
munity care. While this is not an exhaustive list of the
clinical specialities available within each study site (with
peri-operative, accident and emergency and outpatients
departments, critical care and specialized units being
excluded from the sampling frame), there is evidence that
different clinical specialities employ staff with different
demographic, professional and employment pro®les, e.g.
age group, ethnicity, type of registration, number and type
of hours worked (Department of Health 1995).
It was important that the sampling frame consisted of a
heterogeneous group of potential respondents since the
evidence in the United Kingdom (UK) suggests that nurs-
ing staff with different pro®les gain differential access to
continuing professional education activities (e.g. Buchan &
Seccombe 1991, Larcombe & Maggs 1991, Chapman & Hall
1992, Rashid 1992). Therefore, the range of specialities for
inclusion in the study were purposively selected in order
to enhance the representativeness of the sample in the light
of the heterogeneity of the total population.
The limitations associated with purposive sampling,
notably the introduction of bias, are acknowledged.
However, decisions regarding the inclusion of different
specialities were informed by empirical evidence (e.g.
Department of Health 1995) and not judgement alone. The
potential for bias was also reduced by the random selection
of units representing each speciality (Moser & Kalton 1971).
In addition, it was appropriate to utilize strategies which
would contribute to the recruitment of a sample represen-
tative of the total population. Purposively selecting clinical
specialities was considered to be one such strategy.
Creating the sampling frame
A census sample approach (excluding students) was
selected for several reasons. For example, to enhance the
reliability of the sampling frame and to make sample
recruitment more manageable. Cartwright (1978) noted
that one of the major dif®culties facing researchers work-
ing with nurses is obtaining lists of staff working in the
areas to be surveyed. The research team was also con-
cerned about maintaining accurate records over several
months of data collection from potential respondents
consisting of an occupational group notorious for mobil-
ity. Following discussions with senior nurses it was
decided that the most accessible records of nursing staff
on each clinical unit were the duty rotas. As these tend to
be amended on a weekly basis it was also felt that the rotas
would provide the most reliable list of potential respon-
dents as well as indicate when respondents could be
visited in person about the study. Each rota was therefore
reviewed regularly by the research team so that, for
example, respondents planning to take annual leave or
commence maternity leave could be recruited to the study
prior to their departure and respondents due to return
from maternity or sick leave during data collection could
be included in the study.
Unit non-response
Several research texts have set out a number of detailed
strategies for minimizing the level of non-contact and
refusals in surveys using both questionnaire and interview
techniques such as the use of pre-noti®cation, personal-
ized and follow-up letters as well as face-to-face contact
with potential respondents (Champion & Sears 1969,
Dillman & Frey 1974, Gordon 1975, Bradburn & Sudman
1979, Miller 1991). Such texts are largely based upon the
assumption that there is much that researchers can do to
reduce unit non-response in survey research.
Non-contact
Contacting the sampleInitial contact with the study population was made
through a personalized letter to each potential respondent
which introduced the research team, explained the pur-
pose of the study, and gave assurances of anonymity and
con®dentiality for study sites and individuals. Secondly,
each clinical unit was given a poster about the study
which included photographs of the two nurse researchers
responsible for the collection of data so that all staff
(including those not in the sample such as student nurses,
ward clerks, doctors) and patients and visitors were
informed that a research study was in progress.
In addition to the personalized letters and posters, the
research team wanted to establish face-to-face contact
Methodological issues in nursing research Non-response in survey research
Ó 1999 Blackwell Science Ltd, Journal of Advanced Nursing, 30(3), 677±686 679
with potential respondents so that the purpose of the
study could be explained in more detail. It was anticipat-
ed that `being in the ®eld' on different days (including
weekends) and at different times (including during the
night) would ensure that personal contact was made with
the majority of the sample; however, relying solely on the
two nurse researchers visiting clinical units in the hope
that potential respondents were on duty would be inef®-
cient in terms of their time and effort. The duty rotas were
therefore used not only to identify individuals to be
included in the sampling frame but also to plan and guide
sample recruitment. The duty rotas provided a detailed
and up-to-date record of when staff might be available and
were adapted by the research team to create record sheets
of contacts with potential respondents, arranged inter-
views (e.g. dates and times), and dif®culties with data