analyse qualitative data in applied policy research

52 NURSERESEARCHER 2011, 18, 2

issues in research

Qualitative data analysis: the framework approach

Introduction

The framework approach was developed in the 1980s by social policy research-

ers at the National Centre for Social Research as a method to manage and

analyse qualitative data in applied policy research. In this context, the research

brief is commissioned; aims and objectives are highly focused and the research-

ers work with structured topic guides to elicit and manage data. This approach

contrasts with entirely inductive approaches, such as grounded theory, where the

research is an iterative process and develops in response to the data obtained

and ongoing analysis. More recently, the framework approach has been gaining

in popularity as a means of analysing qualitative data derived from healthcare

research because it can be used to manage qualitative data and undertake

Abstract

Qualitative methods are invaluable for exploring the complexities of health

care and patient experiences in particular. Diverse qualitative methods are

available that incorporate different ontological and epistemological

perspectives. One method of data management that is gaining in popularity

among healthcare researchers is the framework approach. We will outline

this approach, discuss its relative merits and provide a working example of

its application to data management and analysis.

Authors

Joanna Smith MSc, BSc(Hons) RSCN, RGN is lecturer in children

and young people’s nursing, School of Nursing and Midwifery,

University of Salford, UK

Jill Firth RGN, PhD is a senior research fellow at the School of

Healthcare, University of Leeds, UK

Keywords

Qualitative research, framework approach, patient experiences

NURSERESEARCHER 2011, 18, 2 53

analysis systematically. This enables the researcher to explore data in depth

while simultaneously maintaining an effective and transparent audit trail, which

enhances the rigour of the analytical processes and the credibility of the findings

(Ritchie and Lewis 2003). This article will provide an overview of the framework

approach as a means of managing and analysing qualitative data. To illustrate its

application, we will draw on a study undertaken by one of the authors (JS) as

part of her programme of doctoral research investigating parents’ management

of their children’s hydrocephalus and shunt.

Context

Delivering health care that is responsive to individual needs is an integral part

of the modernisation agenda of the UK’s NHS. Policy directives for people with

long-term conditions emphasise actively involving patients in the management of

their conditions, valuing their expertise and working collaboratively with patients

(Department of Health (DH) 2001, 2005, 2007). When the patient is a child, this

includes understanding the views and experiences of their parents. The potential

benefits of this involvement include: empowering patients to take control of their

health needs, better mutual understanding for patients and healthcare profession-

als, and patients influencing the healthcare agenda (Simpson 2006). Qualitative

approaches are appropriate for exploring the complexities of health and wellbeing

and can help in creating an in-depth understanding of the patient experience.

Debates about the epistemological and ontological perspectives underpinning

qualitative methods can overshadow the need to ensure that qualitative studies

are methodologically robust. Published qualitative research often lacks transpar-

ency in relation to the analytical processes employed, which hinders the ability of

the reader to critically appraise the studies’ findings (Maggs-Rapport 2001). This is

not to say that the research is not of good quality, but there are sometimes weak-

nesses in reporting that make evaluation problematic. For the novice researcher,

a framework to guide the stages of the data analysis has the potential to assist in

developing the skills required to undertake robust qualitative data analysis.

Approaches to qualitative data analysis

Methods for undertaking qualitative data analysis can be divided into three

categories:

54 NURSERESEARCHER 2011, 18, 2

issues in research

n Sociolinguistic methods, such as discourse and conversation analysis, that

explore the use and meaning of language.

n Methods, typified by grounded theory, that focus on developing theory.

n Methods, such as content and thematic analysis, that describe and interpret

participants’ views.

Despite the diversity of qualitative methods, data are often obtained through

participant interviews. The subsequent analysis is based on a common set of

principles: transcribing the interviews; immersing oneself in the data to gain

detailed insights into the phenomena under investigation; developing a data-

coding system; and linking codes or units of data to form overarching categories

or themes that can lead to the development of theory (Morse and Richards

2002). Analytical frameworks such as the framework approach (Ritchie and

Lewis 2003) and thematic networks (Attride-Stirling 2001) are gaining in popu-

larity because they systematically and explicitly apply the principles of undertak-

ing qualitative analysis to a series of interconnected stages that guide the process.

Generating themes from data is a common feature of qualitative methods and

a widely used analytical method. Thematic analysis is an interpretive process in

which data are systematically searched for patterns to provide an illuminating

description of the phenomenon (Tesch 1990). This results in the development

of meaningful themes without explicitly generating theory. Thematic analysis

can provide rich insights into complex phenomena, be applied across a range

of theoretical and epistemological approaches, and expand on or test existing

theory (Braun and Clarke 2006). However, thematic analysis has been criticised

for lacking depth (Attride-Stirling 2001). The thematic analysis approach can

result in sections of data being fragmented from the original, which can result in

data being misinterpreted. As a consequence findings are subjective and lacking

transparency in how themes are developed.

An overview of the framework approach

The framework approach has many similarities to thematic analysis, particularly

in the initial stages when recurring and significant themes are identified. However,

analytical frameworks, such as thematic networks and the framework approach,

emphasise transparency in data analysis and the links between the stages of the

analysis (Pope et al 2000, Ritchie and Lewis 2003, Braun and Clark 2006). Central

NURSERESEARCHER 2011, 18, 2 55

to the analytical processes in the framework approach is a series of interconnected

stages that enables the researcher to move back and forth across the data until a

coherent account emerges (Ritchie and Lewis 2003). This results in the constant

refinement of themes that may aid the development of a conceptual framework.

Application of the framework approach

We independently chose the framework approach to underpin data analysis for

a range of reasons. First, the framework approach is particularly suited to analys-

ing cross-sectional descriptive data, enabling different aspects of the phenomena

under investigation to be captured (Ritchie and Lewis 2003). Second, research-

ers’ interpretations of participants’ experiences are transparent (Ritchie and Lewis

2003). Third, moving from data management to developing the analysis suffi-

ciently to answer research questions can be a daunting and bewildering task for

novice researchers. The interconnected stages in the framework approach explic-

itly describe the processes that guide the systematic analysis of data from initial

management through to the development of descriptive to explanatory accounts.

In the example we use to illustrate the stages of analysis, JS conducted inter-

views to elicit parents’ perceptions of living with a child with hydrocephalus.

Shunts are the main treatment for hydrocephalus but they are problematic in that

they are prone to malfunctions, which for some children can be life-threatening.

Detecting shunt failure is not straightforward because the signs and symptoms are

variable, subtle and often idiosyncratic to the individual child. Common symp-

toms, such as headache, vomiting and drowsiness, are presenting symptoms of

many childhood illnesses, particularly viral infections. An interview topic guide

enabled the interviewer to explore parents’ perceptions of living with their child

and examine their decision making in relation to identifying shunt malfunction

and seeking healthcare advice. The interviews were conducted face-to-face, either

with individual parents or jointly. Interviews were transcribed verbatim. The next

stage of the research applied the framework approach described by Ritchie and

Lewis (2003). Briefly these stages are:

n Data management – becoming familiar with the data (reading and re-read-

ing); identifying initial themes/categories; developing a coding matrix; assigning

data to the themes and categories in the coding matrix.

n Descriptive accounts – summarising and synthesising the range and

56 NURSERESEARCHER 2011, 18, 2

issues in research

diversity of coded data by refining initial themes and categories; identify

association between the themes until the ‘whole picture’ emerges; devel-

oping more abstract concepts.

n Explanatory accounts – developing associations/patterns within concepts

and themes; reflecting on the original data and analytical stages to ensure

participant accounts are accurately presented and to reduce the possibil-

ity of misinterpretation; interpreting/finding meaning and explaining the

concepts and themes; seeking wider application of concepts and themes.

Data management using a case and theme-based approach

Codes and categories were developed by considering each line, phrase or para-

graph of the transcript in an attempt to summarise what parents were describing.

The process initially involved using printed versions of the transcripts with key

phrases highlighted and comments written in the margins to record preliminary

thoughts. Key phrases were summarised using participants’ own words (‘in-vivo’

codes). In-vivo codes are advocated in the framework approach as a means of

staying ‘true’ to the data (Ritchie and Lewis 2003). Initial thoughts began to

develop into more formal ideas from which a coding matrix was generated. The

coding index enabled changes to be tracked and progress to be recorded. Table

1 gives an example of the coding matrix, highlighting the processes involved in

identifying codes and categories.

Identifying and testing a thematic framework

The coding matrix was developed from four family interview transcripts, which

appeared to represent a range of experiences. These parents had different

experiences in relation to the frequency of shunt complications, including one

family whose older child had not had any problems with the shunt. As part of

the measures taken to ensure rigour, two experienced researchers reviewed the

coding matrix and transcripts from which the matrix was derived. Changes were

tracked by maintaining a research journal and adding notes to the margins of

the matrix. Each in-vivo code initially formed a potential category but as coding

progressed and the number of categories developed they were grouped together

into broader categories. Similar categories were eventually brought together to

form initial themes. These categories and themes formed a ‘coding index’ that

NURSERESEARCHER 2011, 18, 2 57

was used as a means of organising the whole dataset. However, the coding index

was constantly refined throughout the data analysis as new insights emerged.

Table 2 includes an example of the coding index.

Unlike policy-driven research, the interview topic guide in healthcare research

may be less tightly focused and a qualitative software package such as NVivo can

aid data retrieval when searching for patterns in the data. Initial data manage-

ment used written notes and memos, but these were subsequently transferred

to an NVivo database. As data management progressed, NVivo was used more

intuitively, with the tagging of data into relevant categories shifting from a paper-

based exercise to directly coding data in NVivo. Data management using the

coding retrieval and search facilities in NVivo was the first stage of more in-depth

analysis because it enabled researchers to start thinking about the links between

the initial categories and themes, while retaining links to the original data.

Table 1. Example of the coding matrix used to identify codes and categories

Interview transcript: Family 11, child five years, many hospital

admissions, three shunt revisions

Description (in-vivo codes)

Preliminary thoughts

(what is this about?)

Initial categories*

‘You know if your child is being sick whether they are poorly or not.’ Dad

‘Out of hours that they tend to keep her in… I think it is the out of hours that we find difficult, if we are unsure… If we go out of hours service we know that we will be admitted. So we tend to wait a bit longer.’ Mum

‘know… your child’.

‘Unsure’ whether to access out of hours services ‘know we will be admitted’ ‘wait a bit’.

Knowing something is wrong.

Uncertainty access out of hours services.

Experience/ views of out of hours services.

Trying to decide what to do.

Recognising when the child is ill.

Uncertainty: when to access services.

Views about services.

Making decisions.

*Some of the initial categories became themes (for example, recognising when the child is ill) or core concepts (for example, uncertainty)

58 NURSERESEARCHER 2011, 18, 2

issues in research

Table 2. An example of the coding index

Initial themes Initial categories

Uncertainty n Immediate impact of the condition.

n Long-term effects of the condition.

n Child becoming independent.

n Child’s development.

n Embarking on family activities.

Responding to the child’s needs

n Recognising when the child is ill.

n Experiences of shunt complications.

n Beliefs about the signs of shunt malfunction.

n Recognising when the child’s illness is due to shunt malfunction.

n Feelings relating to the possibility of child’s shunt malfunctioning.

n Seeking help for child.

n Taking precautions to protect child because of having a shunt.

n Making allowances for child because of hydrocephalus.

n Explaining hydrocephalus to child.

n Supporting child to develop.

Making decisions

n Making choices about treatment options.

n Beliefs about involvement in healthcare decisions.

n Deciding if illness is due to shunt problem or not.

n Deciding when to access healthcare services.

n Lifestyle choices.

n Family activities.

n Factors that influence decision making.

n Feelings about making decisions.

Coding matrices can be created using Word or Excel spreadsheets but the process

can be unwieldy and problematic when large volumes of data are involved. Since

this research was undertaken, the National Centre for Social Research (www.

natcen.ac.uk) has developed a computer-aided, qualitative data package to assist

in the application of the framework approach. The package can be used to sum-

marise data in a series of matrices from which it is possible to conduct case-based

and thematic analysis. This may overcome some of the inherent difficulties faced

NURSERESEARCHER 2011, 18, 2 59

when trying to manage large volumes of data using spreadsheets. However, the

software and training provided by the centre have cost implications that need to

be factored into applications for research funding.

Development of descriptive and explanatory accounts

Descriptive accounts involve summarising and synthesising the range and diver-

sity of coded data by refining initial themes and categories. Crucial elements in

qualitative analysis are the critical thinking that occurs in relation to: how par-

ticipants’ descriptions are coded; links between codes and categories; and links

between categories and themes (Ritchie and Lewis 2003). Remaining true to

participants’ descriptions is a fundamental principle in the framework approach

and central when developing more abstract concepts.

For the novice, the movement from in-vivo codes and initial categories and

themes to more abstract concepts can seem contradictory. Two linked proc-

esses were undertaken to reconcile these tensions. First, data were synthesised

by refining the initial themes and categories until the ‘whole picture’ emerged.

To ensure the themes were grounded in participants’ descriptions, the author

constantly referred to the original transcripts and checked meaning across inter-

views using Nvivo’s search functions. Second, abstract concepts were developed

by identifying key dimensions of the synthesised data, and making associations

between themes and concepts. Table 3 (page 61) gives an example of moving

from the initial themes and categories in the coding index, and the links between

the refined categories and final themes from which the core concepts emerged.

To ensure the experiences and beliefs of parents were accurately reflected and

to minimise misinterpretation, explanatory accounts began with reflection on the

original data and on the analytical stages. Through using the framework approach,

three core concepts were developed that appeared to reflect parents’ accounts of

living with a child with hydrocephalus: ‘uncertainty’, ‘becoming an expert’ and ‘liv-

ing a normal life’. In the remainder of this article, the development of ‘uncertainty’

will be used to illustrate the application of the framework approach.

The final stages involved making sense of the concepts and themes in terms

of participants’ lives and experiences. This was achieved by exploring the rela-

tionship between the core concepts, the established literature and theoretical

perspectives relating to living with a child with a long-term condition.

60 NURSERESEARCHER 2011, 18, 2

issues in research

Once the natures of the phenomena have been described and concepts identi-

fied, typologies may emerge that explain how concepts operate (Ritchie and

Lewis 2003). The way in which parents respond to illness in their children was

considered by linking their accounts to ‘uncertainty’. Penrod (2007) identified

four possible typologies that reflect individuals’ perceptions of their levels of con-

fidence and control when faced with uncertainty: ‘overwhelming uncertainty’;

‘role uncertainty’; ‘pervasive uncertainty’; and ‘minimal uncertainty’. Parents’

lack of control over shunt malfunction positioned them in ‘overwhelming uncer-

tainty’. This may explain why parents’ accounts were dominated by the possibil-

ity that their children’s shunts could malfunction at any time.

Our experiences of undertaking qualitative data analysis share similarities with

the experiences of other novice qualitative researchers (Li and Seale 2007). The

first challenge related to the process of attaching labels to preliminary codes,

which were initially abstract in nature and did not fully represent the extracts

from which they were derived. Although our enthusiasm remained undiminished,

we grossly underestimated the time required to undertake the early stages of the

analysis. Yet these stages are essential if the findings are to be credible. In our

separate studies, we valued working with experienced researchers who were will-

ing to challenge assumptions and decisions at each stage of the analysis, adding

to the rigour of the research. Sufficient time needs to be allocated to evaluating

initial thoughts and reflecting on the relationships between ideas and participants’

accounts. Asking the question, ‘What are participants really trying to describe?’

when considering sections of the data and using participants’ own words when-

ever possible assisted in ensuring that labelling reflected participants’ accounts.

For part-time students, other work commitments can make it difficult to

re-engage with the data after a period away, although it can prevent over-immer-

sion. Returning to data analysis after time away and re-reading all the transcripts

to consider the phenomena as a whole resulted in data analysis becoming much

more meaningful. Forward and backward movement between the data, par-

ticipants’ accounts and links with initial categories resulted in the emergence of

the final categories and the development of the final conceptual framework that

describes parents’ accounts. This iterative process resonates with the central tenet

of the framework approach that the interconnected stages are not linear, but a

scaffold that guides the analysis (Ritchie and Lewis 2003).

NURSERESEARCHER 2011, 18, 2 61