You are on page 1of 35



I ntroduction


Formulating a project-specific plan of analysis


Searching data


The cyclical process of analysis


Thick description




Categorizing and conceptualizing






Theory development


What is inductive theory?


Why develop theory?


How to develop theory


Grounding theory


Evaluating quality

. Objectives
After reading this chapter you will:

understand how to apply the tasks in the analytic cycle;

appreciate inductive and deductive elements of analysis;

understand how to make a 'thick description' of data:

be able to identify strategies for comparison;

develop skills for categorizing and conceptualizing data;

understand inductive theory development:

understand how to evaluate the quality of textual data analysis.




__''. ___ ''

I nt ro d uction

_,._.__ ._.. __._ ..___..___.____________.. __.._. ____.. __

In this chapter we continue the process of data analysis following the analytic
cycle (shown at the beginning of Part III, on page

201). We suggest beginning

by mabng an analysis plan to focus your analytic tasks and provide a guide to
review your progress, We then discuss the core tasks in the analytic cycle:
description, comparison, categorization, conceptualization and theory devel
opment. These tasks are closely interlinked and are conducted in a circular
manner, with tasks often repeated and conducted simultaneously.
Description is a core analytic task that provides the foundation of data analy
sis, and uncovers the context of behaviour which can provide clues to social or
cultural meaning. Comparison is the next task that allows you to further define
and explore issues and begin to notice patterns in the data. Categorization is a
task that involves grouping codes with similar attributes into broad categories.
Categorization is closely linked with conceptualization, which involves visualiz
ing your data as a whole to develop a conceptual understanding of the issues.
Both categorization and conceptualization move analysis to a higher level of
abstraction and provide the building blocks for theory development, the final task
in the analytic cycle, which moves qualitative research beyond

description and

into the realm of explanation, to provide a broader conceptual understanding of

the social phenomenon studied. Part of empirical theory development is verify
ing that the theory developed is 'grounded' (or well supported) by data.
In this chapter we highlight both the inductive and deductive elements of
data analysis. Both these elements link the analytic cycle back to the original
design cycle, thus bringing the qualitative research cycle full circle .

. _


.. . . .

, ....

Formulating a project-specific plan of analysis



. .


Embarking on the analysis of qualitative data can seem overwhelming because

of the large volume of data you have and the tangle of issues it contains. Initially
you may feel like you are 'drowning' in data with no dear strategy to navigate
the analysis process. Throughout this chapter we discuss the broad process of
analysis and the analytic tasks that you will use. However, it is useful to begin
by developing a project-specific plan of analysis that will guide you through
analysis of your data and lead to answering your research questions.
Developing a project-specific plan of analysis will provide you with a
'roadmap' of specific issues on which to focus your analysis. There is no format
for a plan of analysis; it simply needs to be a guide that helps you progress
through your analysis. However, it is useful to write down your plan of analy
sis, even though this may change


the analysis progresses; this will enable you

to keep track of the analysis tasks completed, identify areas missed and review






your progress. In developing your plan of analysis you may consider the overall
purpose of your analysis, the level of analysis needed to meet this purpose and
then identify specific topics, questions, areas or codes on which to begin your
analysis. These areas are discussed in the following paragraphs.
Consider the overall purpose of your analysis. For example, are you trying to
. answer a specific research question? Do you want to generally explore a topic?
Are you identifying issues to be used in a survey? Your plan of analysis should
focus towards this goa1. This may seem obvious, but there are often a multi
tude of topics, questions, or interesting aspects you could explore in your data
and it is easy to lose focus of the overall purpose of the analysis. It is good
practice to write down the purpose of your analysis at the beginning of your
plan, so that you stay focused on this broad goal.
Consider the

level of analysis that is needed to meet the purpose of your

research. If your purpose is to write a descriptive account of an' issue or to

identify issues for a survey, then you may only need to use description and
comparison in your analysis (these are the first tasks described in this chapterJ.
However, if the purpose of your study is to develop an explanation or theory
about certain behaviour then you would need to conduct all the tasks related
to developing a grounded theory.
A plan of analysis may simply consist of a series of codes, topics and
questions you wish to explore in the data to answer your research question,
But where do you begin? Which topics or codes should you start with? There
may be a specific question or code that is central to your study and a logical
place to begin your analysis, or you may simply begin with a code that you find
interesting. Start by writing down a question and identifying all the codes from
the codebook that seem relevant to explore in order to answer the question.
This then becomes the first task in your plan of analysis. Continue to identify
questions or codes in the same way until you have a range of core issue to
explore in the analysis. Textual data analysis involves gradually building up the
analysis by focusing on small components in the data (a code, a topic, a
question, etc.), gaining a detailed understanding of each component and then
moving to the next. Also consider practical issues in your plan of analysis; for
example, if you know that a specific topic or issue simply did not yield useful
data then exclude it from the plan. Also, if several researchers are working
to gether, you may each focus on a different question or specific group of codes,
so that the analysis is built up as a group.

Searching data is a fundamental task that is conducted throughout the process

of analysis. It involves selecting a code or series of codes from your codebook






and searching data for each segment of text where the code is mentioned, then
reading these text extracts and developing an understanding of the issue.
Searching data by codes allows you to focus your analysis on one issue at a
time and slowly build up your analysis. It is important to remember that
searching the data by codes will simply retrieve the segments of text you have
coded with a specific code, therefore if data coding was inconsistent or incom
plete this will be reflected in the data searches.
T here are many ways to search the data and several strategies are summa
rized in Table 10.1. Some strategies are used more often with particular
analytic tasks: for example, a code or topic search is most often used during
description, while a search by subgroups is used more during comparison.

Table 10.1

Data search strategies

Search by

A basic search that is continually repeated during analysis. Used most frequently in


description (see section on 'Description').

Select a single code from the codebook and search data to identify each segment of
text where the code is mentioned. Read all retrieved text to get a detailed under
standing of the code. For example, searching for a code 'stress', may reveal how
people talk about stress and different experiences or contexts of stress.

Search by

A search that uses several codes around a specific topic. Useful in description (see


section on 'Thick description').

Select codes from the code book that relate to a specific topic. Search each code
individually (see 'search by code' above) to build a detailed understanding of the topic. For
example, the topic of participants' knowledge of tuberculosis, may be explored using
codes of 'source of knowledge' 'symptoms', 'causes', 'myths' and 'treatmenf. Exploring
each code will lead to a detailed understanding of participants' knowledge of the
causes, symptoms and treatment of tuberculosis, their sources of knowledge and
myths about the disease.

Search by

A search that focuses on a specific subgroup of participants. Used as a basic search


for comparison or categorizing (see sections on 'Comparison' and 'Categorization').

Identify a subgroup of participants by their demographic characteristics (e.g. young
women or married men) or experience (e.g. an illness, miscarriage). Then select only
. this subgroup of participants and search for codes within the subgroup. For example,
within a subgroup of employed men, search for the code 'stress', then select another
subgroup of unemployed men and search for 'stress'. Then compare whether stress
is discussed differently by men with a job and without. This type of search is most
easily done using a qualitative analysis software package.


A search that focuses on a specific question. Used most often in conceptualization


(see section on Conceptualization').

Think of a question that can be answered by the data, then select a range of codes
from the codebook to help you search the data to answer the question. For example,
ask yourself 'Is there a connection between work and stress?', identify codes to help
explore this issue (e.g. 'stress' and 'work' or 'job'). then search data for text labelled with
both codes 'stress' and 'work' or 'stress' and 'job', read the text retrieved to see if the
way participants discuss stress and work suggests a relationship between these codes.



Searching the data is made easier by using a data analysis program (e.g. Atlas.ti,

NVivo, MAXQDA). However, no program will do the analysis for you; these
p rograms will facilitate manipulating the large volume of data in qualitative

research, allow you to do more complex searches and conduct analysis more
qUickly; which is

extremely helpflJl d1Jfing data analysis. Although you


. search data without the assistance of a computer package, it is more cumber

some and time consuming.
During the process of analysis you will conduct many data searches; some

of these will lead to important findings, some will be unfruitful and others

will generate ideas, thoughts and questions for further data searches.
Therefore, it is important to have a system for documenting your searches and

the outcomes of searches as you progress through your analysis so that you

keep an analytic 'trail'. Even unfruitful searches should be documented, not

only so that you do not repeat the same search but also in case what appear
to be unfruitful searches turn out to be significant later in the analysis.
Keeping a trail of your data searches and what you found from each is much

simpler if using a data management program, where you can save, date and

label your searches, keep memos about your analysis and add or refer to these
as your analysis progresses.

__. ....

. __.... __ ..

The cyclical process of analysis_

_.._. ..._. ..._.....____. __ .. _.__. __


Qualitative data analysis comprises a range of core analytic tasks: developing

codes, description, comparison, categorization, conceptualization and theory

development. These tasks are closely interlinked: not only are they conducted

in a circular manner whereby tasks are repeated during data analysis, but they

are also often conducted simultaneously at different points in the analysis.

The circular nature of qualitative data analysis is important as it enables you

to go deeper into the data and develop a more refined understanding of the

issues. The core tasks and circular process of data analysis are reflected in our
analytic cycle.
In addition to the circular nature of data analysis, it may also be viewed as
an analytic spiral (Dey,

1993), as shown in Figure 10.1. As you move through

the analytic tasks in a circular manner, you are simultaneously moving up an

analytic spiral and building up your analysis. You move up the spiral from

description towards explanation of data (or theory development). As you

begin to develop explanations you also validate (or' ground') these explana
tions by returning to the data, moving down the spiral. The process of analysis

involves continuously moving up and down the analytic spiral as you develop

theory and then validate it with the data. In this way qualitative data analysis

may be seen as an inductive conceptual cycle, whereby the process of analysis

leads to the development of inductive concepts and theory that are verified
.. ' ..'.' "

_ . . . . . .-....-.-._--. --_.-._,
























Analytic spiral

Figure 10.1


Adapted from original by Dey (1993).

with the data (as opposed to deductive theory development which is what you
may begin with in the design cycle as you developed your study; see Part I).

_...... _..__


Thick description

_..__ ._.. __

___ .....



. ... _


.. .



, __ .

Description is the first task in the analytic cycle (see diagram at beginning of
Part III, page 201). Description forms the foundation of qualitative data analy..

sis and provides the rich detail that is sought in qualitative research . By devel
oping a detailed description of the issues in your data, you become very famil

iar with each issue, understand how issues are interlinked, the subtle nuances

surrounding issues, and you will begin to understand the issues from the
perspective of your study participants. All of this provides you with unique
and valuable insight into your data, Description is therefore an important

analytic tool that can be used to develop powerful and engaging accounts of
events, processes, or social phenomena in your data. Although description is
the first analytic task, it is used throughout the data analysis process.

Qualitative data analysis typically involves making a thick description of your

data. This term was developed by the cultural anthropologist Clifford Geertz
(1973), who explained that a thick description involves not only describing a
particular behaviour but also the context within which that behaviour occurs.
It is the context that gives the behaviour a social or cultural meaning and helps


us to understand its symbolic importance. In contrast, a thin description would

describe only the behaviour or action. Geertz used the example of a wink to
highlight the importance of including context in description. A wink may have
a sexual meaning if observed between a couple, while in a professional setting
it may indicate a gesture of support, or among friends it may show that
. someone's comment was meant as a joke. Developing a thick description is
therefore fundamental to fully understanding the meaning of the behaviours or
actions in your data and will provide a foundation for later conceptualizing and
explaining behaviour.
Developing a thick description is an inductive activity that involves reading
the data and delving deeper into each issue by exploring its context, meaning
and the nuances that surround it. Description allows you to notice connec
tions and relationships between issues that become useful later when devel
oping explanations. It also helps you to notice patterns of issues' within the
data - for example, some issues may be mentioned repeatedly! and certain
issues may be often mentioned together, suggesting a connection between
specific issues.
Making a thick description typically involves focusing on part of the data,
such as a single code, a broader topic, process, or particular behaviour. You then
search the data (see strategies in Table

1 0.1)

for all related text, read the

extracts of text retrieved and begin to develop a thick description. In making

a thick description of your data, consider the depth, breadth, context and
nuance of each issue, to build up a detailed description about different aspect
of each topic or issue. The follOWing questions will gUide you in making a
'thick description' of a code, issue or topic.

What is the issue? Why is it an issue?

What happens? How is it described?
How can you define/describe the issue?
How does it differ from other issues? What makes it distinct?


Are there different dimensions or variations of the issue?

What is the range of perceptions or opinions about the issue?
How do these dimensions or perceptions vary, and what makes each
Is there any pattern to the issue across the data"
Are some dimensions mentioned more frequently than others?
Do any dimensions cluster by subgroups of participants?


What is the context of the issue? (When, where, how does it happen?)
How do participants discuss the issue? (What words, emotion, expressions
do they use?)
What meaning or explanations are given about the issue?
Are certain dimensions of the issue often discussed together?


: 239

Is the issue discussed differently in different circumstances or contexts?


What are these nuances?

What reasons are given for differences?
Are these nuances linked to particular types of participants?


10. 2 show an example of a thick description of a code called 'time'

from data on access to health services in Malawi. Although this example is

shown in table form, a thick description can also be a written narrative. The
five questions at the top of Table

10.2 were used to develop and structure this

description; however, you can use other questions that suit your project. The
first column shows that there were nine distinct aspects of time in these data
(travel time,


time, etc.), each with a distinct context and meaning. For

example, 'waiting time' refers to the long waits experienced at public health
clinics, compared to short waits at private facilities. How each dimension is
discussed is also important; for example, 'waiting time' was discussed with
anger and despair, 'cost of time' with concern and 'travel time' with compla
cency or acceptance. These subtle insights can provide valuable access to
participants' views and experiences. Furthermore, some aspects are quite
nuanced; for example, 'time of day' highlights the fact that service quality
varies depending on the time of day the service is used, and 'cost of
highlights the non-monetary cost of using the health service, whereby health
conditions often worsened while waiting for service at the clinic. A thick
description of this kind really 'opens up' the data for exploration of the
sions and subtle nuances of each issue, and this level of descriptive detail is
extremely valuable in later stages of analysis when building an explanatory
This thick description also highlights some patterns across the data. For
example 'waiting time' was the most frequently mentioned aspect, 'consulta
tion time' the least frequent; and 'travel time' and Ilost time' were most often
mentioned by rural participants. In addition, the description highlights other
codes that are often mentioned together with various aspects of time. For
example, imother code called

quality' often intersects with some

dimensions of , time' (e.g. consultation time, staff time, time of day and speed
of service). NotiCing these patterns can guide your further exploration of
the data. For example, you may then decide to conduct a thick description
of the code 'service quality' to examine how different aspects of time
influence the quality of service provision.
DeSCription is the first task in the analytic cycle and an important component
of grounded theory. However, description alone does not make a grounded theory
analysis. A grounded theory analysis progresses further around the analytic cycle,'
from description towards theory development. While including description,
theory development involves going beyond description to include interpretation,
explanation, prediction and theory development about the phenomenon studied.

. . .. .


Table 10.2

A thick description of the code 'time'

How often is each

What are the


What is the context

How is each aspect discussed (emotions,

aspect mentioned

What other codes


and meaning?

expressions, examples)?

and by whom?


Longer travel time to private clinic;

Complacency, willingness to travel for quality

Frequently mentioned


cost of travel is a barrier

'if you are able to travel, you can find better

by rural women


Travel time

services', 'we live in the village so we just

travel for every service'
Waiting time

Wait time is short in a private hospital,

Despair due to lack of service choice; anger

Very frequently

Public facility

long in public hospital; if service i s

'some patients lose their temper and


Private facility

free, wait time is long; public hospital

threaten to hit the doctor'

Mentioned by some


is crowed so longer wait time

'Cost' of time

Deterioration of health condition due

Concern, frustration 'some faint in the queue

to wait time, health conditions worsen,

without being attended to', 'some patients

wait time is detrimental to health

get angry and collapse, adding to their


Consultation time

Examinations are too quick, not

Annoyance, dissatisfaction

Least frequent aspect

thorough in public hospital; private

Service quality

hospitals identify illness and monitor

medication before discharge
Diagnosis time

Long time for tests results in public

'they quickly prescribe medicines without

hospital; private clinics do thorough

elaborate diagnosis of the illness'

Mentioned by few


diagnoses and monitor patients before



Table 10.2

How often is each

What a re the

What is the context

How is each aspect discussed (emotions,

aspect mentioned

Whalt other cedes


and meaning?

expressions, examples)?

and by whom?


Frequently mentioned

Service quality

Staff time

Time of day

In a public clinic staff delay opening

Annoyance; frustration; unfairness, anger

the clinic and wait for enough patients

'if you know someone at the government

before starting service so increasing

hospital you can get treatment faster, if you

wait time; they spend too much time

don't you can wait until closing time', 'if

on telephone and in breaks. Staff at

nurses start to attend to patients earlier,

private clinics start service on time; if

then long queues are not seen, but they

you know staff, service is quicker.

start their jobs late and the queues are long'

Service speed depends on time of

'if you get to the health facility when day

day. morning is quicker, staff break

staff are knocking off, they don't attend to

and lunch times slower, end of day

you, only advise to wait for night duty staH'

Staft attitude

Frequently mentioned

Service quality

Frequently mentioned

Service quality

may not be served

Delay/speed of

Frequent delays in public hospital,

'you always gel faster service at a private


speed is always fast in private clinic

clinic, at government clinics there are too

many people, service is slow'

'Lost' time

Time is wasted that could be used to

'we can lose a day in the field, it will serve

Mentioned by rural


work, earn money, prepare crops,

my time better to do other things'


Husband approval

conduct household chores

These explanations may be validated by description, include descriptive details or

use the nuances uncovered in descriptive analysis to refine theory development.
Thus, description is both a precursor to theory development and a component
of it. So, while description does not constitute theory development, description
is an essential comoonent to theorizin from textual data. Strauss and Corbin

{199B: 19J stress that 'although description clearly is


theory, it is basic to

theorizing'. Therefore, in grounded theory researchers need to use description

but go beyond description towards conceptualizing data and theorizing from
data. Perhaps a unique strength of qualitative data analysis, and the grounded
theory approach in particular, is its potential to move beyond description to
develop comprehensive frameworks of concepts that derive from the data,
which are then used to theorize about social phenomena.

_ .. . .. .

' Comparison. .

__ . _..___ .. . ___ ._. __ .. . __ . ___ ..___...

Comparison is the second task in the analytic cycle. Comparison allows you
to further explore issues, identify patterns and begin to notice associations in
the data, as you begin to move up the analytic spiral depicted in Figure 10.1.
Comparison is typically used early in data analysis, often together with
description. While description is used to identify and describe issues in the
data, comparison can further refine these issues by clarifying what makes each
issue distinct from others; it can uncover patterns of each issue in the data and
begin to identify the nature of links between issues. Comparison is also used
in later stages of data aJ;lalysis to define categories of codes and when develop
ing explanations from the data; these applications of comparison will be
described later.
There are many ways to approach comparison. Table 10.3 outlines several
strategies for comparison; these are not exhaustive or mutually exclusive and
are simply intended to provide a starting point for making comparisons in
your data. The first strategy listed is cross-case comparison, which involves
comparing a single issue across the entire data set; for example, comparing
how the issue of 'shame' is discussed in different interviews and what makes
each experience of shame different or similar. This type of comparison is most
commonly used when making a thick description (see section on 'Description'
The other strategies for comparison that are listed in Table 10.3 focus on
comparing an issue by different subgroups of participants. These comparisons
identify whether certain issues are mentioned only by a defined subgroup of
participants or whether the issue is discussed differently by each subgroup.
Comparison by subgroups of participants allows you to identify patterns and
meanings in the data that may not be obvious from simply reading data.
Comparison can therefore facilitate uncovering distinctions in the data that




Strategies for comparison

Type of comparison


Cross"case comparison

Comparing a single code across interviews in the data set. Can be

used to identify the variety of perspectives or experiences of a single

issue. For examp!e, comparing the code 'stress' aCiOSS the data tu
identify similar and unique experiences of stress.
Comparison by deductive

Comparing codes by deductive subgroups developed by the researcher

(e.g. socio-demographic characteristics). For example, comparing how
a code is discussed by men and women, younger or older participants,
single or married, higher or lower socio-economic status.

Comparison by inductive

Comparing codes by inductive subgroups developed from the data (e.g.


participant beliefs, behaviour or experiences). For example, comparing

participants who hold pro-life or pro-choice attitudes towards abortion;
comparing participants who have experienced workplace stigma and
those who have not; comparing labour migrants Who have extra-marital
partners and those who are monogamous.

Comparison across and

Comparing codes across subgroups (e.g. between urban and rural

within subgroups

subgroups) and within subgroups, (e.g. within the rural subgroup and
within the urban subgroup). For example, comparison between urban
and rural subgroups may identify broad differences, while comparison
within the rural subgroup may identify further subgroups (e.g. rural elite
and rural poor).

Comparison by typologies

Comparing codes by typologies developed in the descriptive analysis.

For example, a typology of pill users may include 'regimented',
'haphazard' and 'pill-averse' participants, so comparison would involve
comparing codes by these categories of the typology.

Comparison by study

Comparing by elements of the study design. For example, comparing

design elements

codes by study sites, ethnic groups or participant experiences (I.e.

multiple births, single births) that were part of the study design or
recruitment criteria.

provide the foundation for explaining issues and developing theory. For
example, a. study in the Asian community in the UK found that young Muslim
women were embarrassed to consult with male doctors for f amily planning
services, particularly if the doctor was from the same cultural and religious
background as they were (Hennink et a1., 1999a). This specific finding
emerged after multiple comparisons by subgroups. First, the code 'embarrass
ment' was compared by gender to fmd that the issue was only mentioned by
women. Second, a comparison by age found that it was younger women who
were most concerned about embarrassment. Third, comparison of young women
by their religion found that Muslim women voiced the greatest concerns about
embarrassment. Fourth, comparing the comments from the young, Muslim
women who raised the issue of embarrassment highlighted different nuances




. .

in the issue; for example, if the doctor was of the same cultural and religious
background , the women feared being negatively judged by the doctor for
seeking advice on contraception; if the doctor was from the same community,
they feared that their consultation would be revealed to others in the community;

hovcvcr) if the doctor



. report issues of judgement or confidentiality but would only feel embarrassed

if an examination was required. These patterns and subtle nuances in the code
'em barrassment' are revealed by using constant comparison to fully explore
and define the issue. When conducting comparisons by demographic charac
teristics you need to make sure that you have this information for all partici
pants. Often demographic information is collected separately from the inter
view data, perhaps in a brief survey before an interview (see the methods
chapters in Part II).
Comparison can be between deductive subgroups and inductive subgroups
of participants. Deductive subgroups may have already been identified in the
design cycle when designing the study} and are often socio-demographic
subgroups (e.g. gender, age group, marital status). Comparison by deductive
subgroups is the most commonly used, perhaps because it is easiest to conduct.
However, comparison by inductive subgroups is equally important as it can
often yield fruitful and unexpected discoveries. Inductive subgroups are more
subtle and emerge from the data during the analysis. For example, a study on
b reast cancer prevention in China identified that some participants believed
was the cause of the disease; comparisons were then conducted
between two inductive subgroups of women, those who believed in fatalism
and those who did not. Similarly, a study in Pakistan identified that women's
decisions on contnk:eptive use were strongly influenced by a mother-in-law
who resided in the household. This led the researchers to develop two induc
tive subgroups of women, those who lived with their mother-in-law and those
who did not. Comparisons were then made between these two inductive
subgroups on a range of health-related codes. Comparison by inductive
subgroups can therefore identify important variations and nuances in the data.

Categorizing and conceptualizing

.. _


The third task in the analytic cycle involves categorizing and conceptualizing
data. These tasks are distinct) but in practice they are closely interlinked.
Categorization involves identifying codes with similar characteristics and group
ing these together into meaningful categories. Conceptualization involves then
considering the relationships between these categories) to view the data as

whole and develop a conceptual understanding of the issues.

tasks begin
to move analysis to a higher level of abstraction and prOVide the conceptual



building blocks to begin theory development. Categorization and conceptualiza

tion are closely interlinked, whereby categorization may facilitate the conceptu
alization of data, and conceptualization may lead to category development. Here
we describe categorization first as a precursor for conceptualizing data.

Categorizing involves grouping codes with similar attributes into broad catego
ries. The process of categorizing data involves moving from individual codes
towards broader categories to enable you to gain a more conceptual understand
ing of the data. Individual codes often highlight single issues in the data, while
categories bring together a group of codes that collectively

a broader

concept or topical issue. Therefore, categories represent a higher-order grouping

of data from which you begin to build conceptual frameworks to explain and
develop theory about the research issues. Much social science research consists
of some type of categorizing or conceptual ordering of data that forms a precur
sor for abstraction and theory building (Strauss and Corbin ,


Developing categories involves both induction and deduction. Initially

categories are derived inductively by identifying similarities between codes,
grouping these codes together, and giving the category a name that broadly
indicates the shared issue. However, deduction is also used to develop or refine
categories; for example, a category may be spurred by concepts from research
literature or from the original conceptual framework of the study. Given the
influence of induction and deduction, you need to validate the categories
developed, either with the data or with other members of the research team
to check they are indeed well supported by your data. Categories need to be
well defined, appropriate, and valid as they will become the core components
in theory development. During category development you will again use
description and comparison, to describe the concept represented by each
category and how categories are distinct.
An example of categorization is shown in Figure

10.2 using data on women's

barriers to using health services in Pakistan. In this study, codes were classified
into five broad categories: access, money, knowledge, culture and eligibility.
While some codes could be clearly linked together by a common concept,
others required a more abstract approach to identify their shared attributes.
For example, a range of codes were clearly related to money (or lack of money),
while others signalled access to services as a common concept. However, the
codes relating to eligibility required more conceptual thinking to identify their.
shared characteristics. The five categories were initially labelled by their
common attribute (access, money, knowledge, culture and eligibility), and were
then further refined through comparison with constructs from research
literature. After this comparison the final category labels were refined to




Category label

Common aitribute

Group of codes

Figure 10.2





















Service location

Service cos1

Knowledge 01 clinic

Social norms

Service eligibilily

Opening times

Registraiion fee

P u b lic information

Gender barriers

(e.g. age. gender)

Travel time

Travel cost


Husband decides

Service a rea

No transport

Cost 01 drugs

Understand service

Influence of elders

Permission needed

PhYSical abilily


language'" lileracy

From codes to c ategories in data exploring 'barriers to service use'

physical, economic, cognitive, social and administrative barriers, because the

focus of the study was to identify barriers to service use. However, care should
b e taken not to simply apply concepts from research literature to data, but to
compare categories emerging from your data with those identified in the litera
ture to

or rethink your categories; in this way categorization remains

well rooted in the data. Another example of codes grouped into categories can
be seen later in this chapter in Figure

10.5, where codes are classified into

three core categories (legal, economic and health).

The development of categories continues until saturation, when you can
identify no further caegories in the data. Once category development is
complete, each category needs to be clearly defined and described to fully
. elucidate the concept it captures and how it is differentiated from other
categories. Finally, codes are sorted into the defined categories, and it is these
categories that are used in further analysis towards theory development.

Conceptualizing data is a task that moves analysis to a more abstract level. It

involves seeing your data as a whole to develop a broad conceptual under
standing of what is happening in the data. Conceptualizing data is an essential
precursor to theory development as it involves understanding how the
individual components of the data are linked together into a broad conceptual
framework that can begin to explain or predict the phenomenon under study.
Conceptualizing data involves some

of abstraction however, it is

important to remain close to the data so that you develop an empirically based
conceptual understanding of your data.



.. -


There are many ways to conceptualize data to get a broad overview of what
is happening in the data. Most strategies require some abstraction from the
data to enable you to 'see' the patterns more clearly or understand how data
fit together conceptually. Some strategies will focus your attention on different
parts of the data to highlight

a pattern or trigger


understanding of how an

issue works, while other strategies help to identify the broad structure of data
and how it fits together. The process of categorizing data (described above)
may be seen as part of the conceptualization process as data are classified into

higher-order categories that may begin the conceptual process. We outline a

range of strategies below to help you view your data in different ways, which
may lead to a more conceptual understanding of the data as a whole. The

strategies described below are not exhaustive or mutually exclusive, and some
strategies are similar. We present the strategies as a menu of options and
recommend that you try various strategies to see what works for you and for

different types of data. You may find a single strategy works for a project or
you may use several strategies together to conceptualize your data.

The 'big picture'

A common strategy for conceptualizing data is to consider the 'big picture' of

the data by identifying the central story that accounts for the issues in the data.
Identifying the big picture or central story of the data involves stepping back

from the data to gain a broader perspective of the issues, then synthesizing

these into a concise account that brings together the core issues within the
data. In order to develop a big picture view of the data you first need to have
a detailed underst(mding of individual components of the data and linkages

between these (through description and comparison). The key to developing a

big picture understanding is to look for simplicity to clarify the core issues, key
linkages and overarching explanations, and then identifY how these may
contribute to an overall account of the data. Extracting the central story from

data requires some simplification of the issues; however, this should not be
viewed as diminishing data but rather adding strength by contributing to a
clearer understanding of the relationships within complex data. Furthermore,

there are often several storylines in the data, reflecting the diversity of behav
iour, circumstances or outcomes, and this diversity should be reflected in the
'big picture' overview. The challenge of developing a big picture view of data

is to summarize data yet retain some complexity and nuance that reflects a
comprehensive understanding of the issues.

A telescoping technique allows you to vary your perspective on your data by

moving from a broad overview of the data to a close examination of the detail.

T HE ANALYTIC CYCLE :. _ ---_ ..

Telescoping involves continually switching from 'zooming in' to 'zooming out'

from the data, and essentially conducting two analytic tasks simultaneously.
Zooming out involves gaining distance from the data to develop a broader
perspective on the data as a whole, to identify central issues, core processes or
main lirJcages L'1
. the data. Zooming in enables you to return to the detail in the data, to clarify
links, identify nuances or exceptions and further refine the broad conceptuali
zation process. Telescoping therefore fosters two analytic processes, broad
c onceptualization by moving away from the data and detailed exploration by
moving closer to the data, ensuring that your broad conceptualization still
retains a strong foothold in the data.
Explore links
Exploring the links between individual components of data can help build up
a complete picture of the data as a whole. Exploring relationships between
p arts of the data may begin with follOWing up on any associations you noticed
in the data when describing and comparing codes. You may begin by exploring
links between a particular subset of codes, then move on to explore another
subset of codes and then consider whether the two subsets are linked in any
way. In this way links between codes are built up in an incremental manner to
develop a comprehensive understanding of the network of relationships in the
data. It is important to note that not all issues in the data will be related, and
that the absence of links is just as important as any linkages.
The importance, of fully exploring your data to conceptualize the whole
picture may be highlighted through a nineteenth-century parable by John
Godfrey Saxe called The blind men and the elephant' (Saxe,

1873). This

parable describes how six blind men take turns to touch an elephant and
describe what an elephant is, but each man touches a different part of the
elephant and only that part (e.g. the trunk, tail, tusk, ear, legs or body). The men
then disagree on what an elephant is, because each man has only explored a
part of the elephant and not the whole elephant. The parable states that 'each
was partly in the right, and all were in the wrong', to suggest that although each
b lind man may have correctly described one part of the elephant, no one
described the whole elephant. This underscores the importance of explore all
your data in order to develop a complete picture of what that data say.
Furthermore, exploring links between each element of the data can build up
to a conceptual understanding of the data as a whole. Suppose that the sLx
parts of the elephant described by each of the blind men were codes in your
data, so you would have six codes: a brick walt spear, snake, tree, fan and rope
(as shown in Figure

10.3). Each code first needs to be described and the links

with all other codes explored (see earlier sections on 'Description' and
'Comparison' for strategies). This descriptive detail may identify that there are


--- -, 249


Data components & linkages

Figure 10.3

Conceptualizing linkages into a 'whole picture'

Linking codes towards conceptualizing data

four aspects to the 'tree' code, two variations of the 'fan' and 'spear' codes, but
only one type of 'snake' and 'rope' code. These details will become important
later when understanding the data as a whole. Exploring how each code relates
to each other reveals that the 'brick wall' code is central because it is linked to
three other codes: the fan, rope and tree codes (shown by arrows between these).

However, equally important are the codes that are not linked to the brick wall

(shown by a dashed line and cross): these are the snake and spear codes.

Further exploration of linkages reveals that the rope code is linked only to the
brick wall code but not to any other codes in the data; and the snake code is

also only linked to one other code, the spear. By continuing to explore links in
this way you develop a detailed understanding of the key relationships in the
data (this is shown by the matrix of arrows and dashed lines in Figure

1 0.3). If

you then construct your understanding of the data according to the linkages

found, you will develop a conceptual understanding of the data as a whole (this is

represented by an elephant on the right-hand side of Figure

1 0.3, which resulted

from piecing together the components of data by the linkages between them).
We encourage you to find the 'elephant' in your own data through incremen

tally understanding how different components of data are linked together to

form a comprehensive whole. In conceptualizing data as whole, you may find

that you return to descriptive details of codes to understand the nuances in the
data; for example, to understand how the four aspects of the tree code (legs of

the elephant) and two variations of the fan code (ears of the elephant)
contribute to the whole picture. Exploring linkages is an effective way to

develop a conceptual understanding of your entire data set, which may be

more enlightening than its individual component parts might suggest.

250 :


Drawing up a m atrix of codes is a systematic strategy to show p atterns in the
data more cle arly. It enables you to examine the intersection of several codes
to identify patterns in behaviour, attitudes, actions, outcomes, etc. Highlighting
these patterns may trigger a clearer conceptual understanding of the data or
. lead to further exploration of the data towards l ater conceptualization. This
strategy involves selecting several codes that are somewhat related in the data,
placing these in a matrix and examining the data where the codes overlap, This
helps to identify

patterns in the data and contributes to a more conceptual

understanding of the issues in the data.


1 0 4 shows a m atrix that conceptualizes women's health-seeking

strategies during childbirth in India. The m atrix contains two codes each with
three dimensions: the first code, 'labour characteristics'] has the dimensions of
normal, minor c omplications and major complications; and the second code,
' financial resources ' , has dimensions of poor, intermediate and wealthy.
Examining data that correspond to each cell of the matrix reveals the different
h e alth-seeking strategies adopted according to the conditions of labour and
resources available. For example, women o f intermediate financial means who
had a normal labour gave birth at home, because this is what 'strong' women
do, while wealthy women with a normal labour gave birth at a private clinic.
While drawing up this m atrix it became clear that poor women actually
adopted two strategies due to the availability of a government incentive
p ayment if they delivered at a clinic, so an additional column was added to the
m atrix to accommodate this nuance. In addition, the m atrix revealed that
there were no data on the outcomes for poor women with major labour
complications, and only indirect reports about the strategies of wealthy
women , Therefore] the matrix can also assist in fOCUSing further data collection
on filling specifi c information gaps to further explain behaviour. A matrix can
a ls o be used to identify typologies, for example to define certain types of
behaviour according to different circumstances; for this approach each cell
m ay be used to differentiate a different 'type' of behaViour, outcome, or belief.
Process or path way
A further strategy for conceptualizing data is to consider whether the data
describe a particular process or pathway, Ask whether the data identify steps,
stages] a process or strategy, which may be used to initiate a conceptual under
sta nding of the data , For example, the central focus of your data may be to
outline the p rocess of ma rriage i n a p articular culture, identify j ob-seeking
st rategies] describe the process of migration or docum ent the illn ess histo
r i es of people with a certain disease. Using th e process or pathway approach
to make sense o f the data can highlight and sequence distinct stages, or


Availability of financial resources



(resources can be gathered in

(Resou rce,s available for


(no resource for bribes or fees)

Normal labour

Birth at home due to l ack

Birth at clinic for government's

Birth at home is the choice

of choice

incentive payment

of strong wome n

'[even] if the delivery is normal. . .

'[those I who have no



go immediately to the
private clinic: doctor.'

money are supported by

she is taken to the clinic for the

are problems we g o to the

money ( 1 400 rupees).'

[clinic)' 'Women have babies at

home to save money for
feeding the baby.'

Minor labour

(e.g. need medication)

Continue birth at home

'we all faced many problems

but still gave birth in the

' I f they a r e rich t h e n they

'Birth at home is best. If there

traditional birth attendants'

non-emergency treatment)
Birth at private clinic

Note: participants only

reported experiences of

other women in this cell.

Transfer to government clinic

Transfer to government or

and receive no care

private clinic and receive care ment at private clinic

'They say "if you have no money,

Pay for specialist treat-

'We had to spend 1 200 rupees 'In the private hospital they

come another day, now go

at the hospital. When there is a give you service by how


problem, it is a must, but the

much money you spend.

money comes from household

Only money speaks there:

Major labour

No data

Transfer to regional hospital

(40 km distance)


(e.g. need surgery)

. 'When [government midwife]

fails she says "take her to X

Transfer to regional or
national hospital (1 1 0 km

'If she has

serious case,

and has money, they take

her to X hospital'

Figure 1 0.4

Matrix of women's health-seeking strategies

Source: Class assignment by Sarah Laswell, GH525, 2008.


uncover multiple p athways that are distinguished by certain characteristics or

circumstances. Once the overall process or pathway is conceptualized it can
used as a framework to explain certain paths or describe the outcomes of differ
ent pathways. This approach to conceptualization is sometimes referred to as a
approach, for example by describiIlg a fertility, illness or crime
Some studies are explicitly designed to identify a process or pathway. For
example, a study on migration may be structured to ask about pre-migration,
migration and post-migration experiences, therefore using a clear process
approach from the outset. Therefore, it makes sense to use a process approach
when conceptualizing such data: However, the process approach can also be
used to conceptualize data that were not designed explicitly to uncover a
process. For these types of data the process approach can bring conceptual logic
to data and uncover subtle processes underlying data. To initiate this !>trategy you
m ay consider whether there is a sense of time or timing around any issues in the
data that may uncover a process. For example, consider the timing of behaviour,
the use of time, a change over time, or event sequencing over time in your data .
A study that examined how young people in Pakistan receive information
about sexual health used timing to conceptualize the process of gaining knowl
edge about thi topic (Hennink et

2005) . In this study the researchers

constructed a chronology of when and how information was received by young

p e ople to identify whether there was any pattern by the stage of adolescence.
From this activity it became apparent that information acquisition was
triggered by events during adolescence, for example when a girl begins menses
s he is given information about menstruation, when a boy grows bodily hair he
is given information abo ut puberty, when a marriage is arranged young people
are given information about marital responsibilities. Conversely, when there
was no event to trigger information-giving, young people received no informa
tion about sexual health. Therefore, conceptualizing data around the notion of
time led researchers to understand that knowledge acquisition is related to
specific events during adolescence.
Another study (Hennink et al, 1 999a) used a process approach to conceptual
ize knowledge and use of contraception among Asian women in Britain. The
study developed a timeline of contraceptive histories for each woman in the
study. The patterns evident in this timeline revealed that most women learned
about contraception at marriage from their husband or at their first birth from
health workers, and only women educated in Britain had any knowledge of
c ontraception before they were married. Therefore, using

pathway approach

to conceptualize data can provide an effective synthesis and clarification of data.

Questioning data and analytic puzzles

Another strategy for conceptualizing data involves questioning data by formu
lating analytic questions that prompt your data searches. For example,



researchers in India knew from survey data that the ideal age of marriage in

their state was 18 years, and they wanted to explore whether this was also true

for participants i n their study Village. They questioned their data by asking
'What do our study p articipants say is the ideal age of m arriage?', ' H ow is the
ideal age at marriage decided?' and ( Is it different for boys than for girls? ' . This
led researchers to identify how the ideal age of m arriage is determined by the
study participants. They discovered that marriage was not linked with a certain
age per se, but was determined by whether young women have 'strong bones'
and the physical capacity to bear a child, and whether young men have the
strength of character, m aturity and ability to provide for a family. Questioning
the data in this way prOVided important detail and nuance to understand the
ideal time for marriage in this cultural context.
Silverman (2005) suggests different types of analytic puzzles that m ay be
used to initiate a focused exploration of data. For example:

developmental puzzle examines how an issue or phe nomenon arise s, for

example, 'How does workplace stress develop?'. This t ype of question may lead to
a description of the variou s influences on workplace stress.

mechanical puzzle examines how an issue works or outlines a proce ss, for

example, ' H ow do l abour m igrants seek health care?' or 'What is the proce ss of
health screen ing for breast c ancer?'.

causal puzzle describes the pathway of influence on an issue, for example, 'How

doe s the gender of health providers influence service u se?'.

X-ray view

Taking an X-ray view of data (Richards, 2005) can help to identify their core
structure or 'backbone' . An X-ray view can help to conceptualize complex
data by focusing on the essential framework of the data rather than the details.
Ask yourself what are the central components that hold these data together.
An X-ray view involves looking past the detail of the data to the backbone
or structural aspects. These may be issues in the data such as core b arriers to
service use, or they may be broader organizational or structural issues that hold
together the issues in the data. For exampk a study in Burkina Faso on the
health benefits of women receiving micro-credit loans found that one of the
main issues was the poor organizational structure of the loan institutions, not
the behaviour of the women themselves. Therefore, many of the health issues
in the data related to these structural weaknesses of the lending institution.

A typology is a way to categorize data along a continuum to distinguish differ

ent types of behaviour, beliefs or attitudes. Developing a typology involves




.- - --

- ...

both describing and categorizing data. It is important to remember that a

typology is not a list of issues; it is a classification of variations within a single
issue. Therefore, a typology should have clearly defined categories that are
independent of each other and do not overlap, so that partiCipants can be
,;, hv tvn()1into onlv one 'tvne' . Not all d::lt::l ::lrf' suitabl e for ::In>llv.--categorized
- r -I,J

J ,1"


- '


- .J

ogies and researchers should try not to force data into a typology. The typology
is a versatile tool; it can be used at various stages in analysis, for example as a
way to describe and compare data, as a tool to conceptualize data, as a struc
ture for further analysis or as a way to present study findings in a report. The

following are some examples of typologies. A typology of pill-taking behaviour

may highlight three different types of pill-takers:

regime nted pill-takers who follow instructions without fai l ;

haphazard pill-takers who start but do not com plete the course of treatment;

pi ll-avers e people who prefer to find an alternative solution to pill-taking.

A typology of strategies for payment of household emergencies may include

five strategies:

save money to pay for emergencies ;

sell goods or se rvices to pay for an emergency;

borrow money to pay for the emergency;

ask for credit from the emergency se rvice provide r;

forgo future pu rchases to pay for the em ergency now.

A typology of people's attitudes towards recycling household waste may

include five attitudes:

independent, moti vated people who currently recycle;

incentive-driven people who need an incentive to recycle;

enforceme nt-motivated people who would only recycle if forced by a policy to do so;

convenience-driven people who would recyc le if it were more convenient;

non-supportive people who are against the idea of recycling .

'The music not the dance'

A further strategy for conceptualizing data is to focus attention on 'the music
not the dance' ( Richards,

2005), this means understanding how the background

context influences the focal issues. Focusing attention on the background can
help you recognize any underlying mechanisms that influence the issues you



. . . . .- . . . 255

are examining. For example, an organizational structure, service delivery process,

social context or cultural norms may be background influences that have a
pivotal influence on the c entral issues in the data . Richards (2005) states that
this approach is very effective in analysing situations that we take for granted
in which thE' sodal or cultural structure is familiar. This approach

or contexts

was used in a study on human trafficking in Nepal (Hennink and Simkhada,

2004), where the researchers examined the social backdrop of labour migra
tion and poverty to understand how trafficking exploited the normal economic
migration of women and their vulnerable position in society. By focusing on
the background context you can more fully understand and conceptualize the
data as a whole.
Writing and presenting

One approach to conceptualizing data that is perhaps underused is writing and

presenting. Writing and presenting your study findings are often considered
activities that are conducted only at the completion of research; however, they
can also be extremely beneficial during the analytic process to conceptualize
your data. Writing and presenting are powerful tools for conceptualization
because they involve making sense of data for an audience, which requires core
issues to be distilled into a logical coherent presentation. When you write about
your data or present it verbally you need to identify the core 'headline' and a
'storyline' that concisely captures the issues, outcomes and implications of your
data. This process can help you to more clearly conceptualize your data, and in
addition, audience comments and queries can also refine and strengthen your
conceptualization or spur further analysis of the issues to more fully conceptu
alize the data. You may use this approach by presenting or writing for an actual
audience, or by using a 'three-minute summary' to verbally summarize to your
colleagues the core issues in your data and how they are related.
Social domains

D ata m ay also be conceptualized by considering social domains. Social

domains may b e overarching realms, spheres, arenas or contexts that bring
together a group of issues in the data. Identifying social domains can help to
categorize seemingly unrelated issues that alone may seem insignificant but
collectively become an important component of the data . You may begin by
asking 'What are the social worlds that affect the data? How do these relate to
one another?' There may be several domains in your data, and you may
consider how each is related and whether issues fall under several domains.
Once a number of social domains h ave been identified, these can be repre
sented in a co nceptual diagram that essentially categorizes those issues that fall
under each domain and those that straddle several domains.




Non-use of health clinics due to:

1 . Language barrier at clinics


2. Hostile attitudes of clinic staff to migrants

3. Lack of basic health knowledge on minor

respiratory illness vs TB


Illegal migrants

unfamiliar with
local health

fear bribes from

health workers

,/ Migrants \
,/ face stigma \

;' & fear in health




1 . Migrants fear deportation or

jail if


reported to authorities

by clinic
2. Employers don't register
migrants at health clinics to
avoid tax payments



. ' - , Non-use of health clinics due to:




Non-use of heaith clinics due to: :


Migrants fear

health cost or dismiss workers

with TB

bribes from

2. Migrants reluctant to spend


low wages on TB treatment


3. Migrants prefer to pay for


1 . Employers reduce wages to pay


TB treatment in home

, ,- , ,



F igure

1 0. 5

Domains of influence on labour migrants seeking health care for tuberculosis

(TB), Kazakhstan
Source: Class assignment by Samantha Huffman and Janet Ousley, GH525, 2008.

For example, Figure 1 0 . 5 shows how three domains (health, economic and
legal) were identified to conceptualize the broad range of issues raised in data
on health service use by labour migrants in Kazakhstan . The conceptual
diagram shows the three domains and differentiates codes by those that relate

a single domain, those that straddle two domains, and those that intersect

all three domains. For example, the issue of giving bribes to receive health
services is placed at the intersection of the health and economic domains,
while the practice of bribing police to turn a 'blind eye' to illegal migrants
r elates to both the economic and legal domains. Using a social domain
approach enables a better conceptualization of the three critical domains that
influence migrants' low use of health services and how the seemingly disparate


issues in the data can be b etter understood. The social domain approach to
co nceptualizing data can also be valuable in later analysis when making policy
recommendations aimed at specifi c sectors of society.
Drawing diagrams
Conceptualizing data can also be facilitated by drawing diagrams to visualize
relationships in the data . Drawing diagrams involves data reduction and
simplification to view broad patterns in the data. It is a strategy that can help
to capture relationships and to notice where they are absent. Drawing a
diagram begins by showing relatio nship between codes, which may begin as a
simple sketch with boxes and arrows. This task will quickly identify issues that
are central in the data and those that are m arginal. It will also highlight issues
that are consistently mentioned together, and you will begin to notice a pattern
of linkages in the data. Effective diagrams are inductive, with each issue and
link originating from the data themselves (e.g. there is evidence in the data that
the issues are linked in the.way you are depicting in the diagram). Drawing can
provide a cleqr visual representation of the structure of the data that can spur
further exploration of the data and can be used to discuss initial understand
ings of the data with colleagues. Drawing a diagram may be a means to concep
tualize data or may lead to an inductive conceptual model that captures the
central message of the data.

The fin al task in the analytic cycle involves bringing together all e arlier
components of the analysis to develop an inductive theory about the research
issues. Theory development is actually being slowly advanced during each
stage of data analysis, as you build up a clearer understanding of the issues in
the data and how these fit together. Theory development is therefore closely
linked with conceptualizing data, as here you begin to search for explanations
that form th e framework of a theory. Theory development begins with induc
tion, because you develop a theory which is strongly embedded in the data .
H owever, the inductive theory developed is often then embedded within
broader deductive theories, or compared to existing theory to emphasize new
concepts developed through qualitative research. Theory development is
therefore largely inductive, but does involve an interplay with existing deduc
tive theories. It is the latter deductive tasks that link the analytic cycle b ack .
to the original design cycle (see Part I) . Theory development involves contin
uously moving between three tasks: developing theory, verifying theory and
refining theory.



__ What is inductive theory? _.


______ .___________ .._______. ._. __ . _

To understand theory development, we first need t o clarify what is meant by

'theory' in qualitative analysis. An inductive theory is essentially an explana

tion for how something works as derived from empirical data. However: a
theory is more than a set of findings; it provides a framework for understand
ing, explaining and predicting phenomena, and thus both advances our knowl
edge of a phenomenon and can be used to develop p olicy or practice. Strauss
and Corbin

(1998: 22) define theory as 'a set of well-developed categories (e.g.

themes, concepts) that are systematically inter-related through statements of

relationship to form a theoretical framework that explains some relevant
social , psychological, educational, nursing or other phenomenon' .
Implicit in this definition are the analytic tasks that lead to inductive theory
development . These tasks include identifying codes from the data , grouping
codes into overarching categories (categorizationJ, identifying empirically
supported links between these categories (conceptualizationJ and constructing

an explanatory framework for the phenomenon studied (theory development).

Thus, the process of data analysis moves from description to conceptualization
and builds up to theory development, with each task building on the next, such
that the resulting theory incorporates aspects

of all earlier analytic tasks. These

tasks are all inductive, hence lead to the development of inductive theory.
Theory development can be achieved in different ways. It can involve the
development of entirely new inductive theory through the analytic process
summarized above, or it can be achieved through the modification of
pre-existing theory using new empirical data. Much of this chapter describes
the development of new inductive theory, which involves the scientific discov
ery of new concepts or theoretical frameworks for understanding social
p henomena. However, we wish to highlight that theory development can also
be achieved by extending concepts in pre-existing theory to develop a more
refined understanding of a social phenomenon or by expanding the contexts
to which a theory applies. Snow et a1 .

(2003) describe two approaches to

theory development that involve the modification of pre-existing theory; they

refer to

as 'theoretical extension' and 'theoretical refinement' .

Theoretical extension does not involve the development of new theory per se,
b ut demonstrates the relevance of a pre-existing theory or conceptual frame
work to a different context or social circumstance from that in which the
theory was developed. This is done analytically through examining the 'trans
ferability' of a theory, or concepts within it, benveen two or more contexts.
Theoretical extension uses empirical research to broaden the relevance of an
existing theory by demonstrating how it can be extended to a range of other
social contexts that differ from the one for which the theory was originally
developed or intended to be used.



Snow et a1. (2003) give an example of theoretical extension by describing

empirical research in traditional agrarian societies where the theoretical
construct of 'vengeance' was developed, then other researchers who were conduct
ing research on conflict amongst leaders in private corporations in the USA found
similar constructs of 'vengeance' to that developed in the earlier studies. Even
though these two research contexts were vasdy different (traditionat agrarian
society and modem business environments) the social fonus of 'vengeance' were
found to be very similar; therefore demonstrating the extended relevance of the
theoretical concept of/vengeance' across vastly different social and cultural settings.
Theoretical refinement is another method of theory development that does not
involve constructing entirely new theory, but the elaboration or modification of
pre-existing theory using new empirical material. It may begin with theoretical
extension described above or may be independent of it. Analysis of new empir
ical data may uncover a new theoretical concept or understanding that is not
part of the original theory but contributes to it. The analytic process involves
examining the 'fit' of an existing theory to explain particular phenomena in
empirical data, then examining the components that do not fit the original
theory to identify whether a new concept has emerged that extends the theory.
Theory development that involves the modification of pre-existing theory
directly links the analytic cycle back to the original design cycle within the quali
tative research cycle. This is because it involves embedding the empirical findings
of your study within the original conceptual framework or theoretical constructs
that guided the research design. This task can identify whether or not the empir
ical evidence highlights new theoretical constructs that extend pre-existing
theory. In this chapter we largely focus on developing new empirical theory;
however, we do suggest some strategies that contribute to theoretical refinement.
Why develop theory?

Developing inductive theory moves qualitative research beyond description and

into the realm of explanation, and towards a broader conceptual understanding
of a given social phenomenon. Theory development is important because it leads
to a higher level of abstraction of data, bringing study findings to a more concep
tual level. This conceptual understanding of the phenomenon is what enables a
narrow research topic to relate to broader social processes, and for empirically
developed theory to contribute to understanding and influencing broader social
processes. Without the higher-level abstraction into theory development, study
findings remain limited to description or to the context of a specific research
project. Therefore, theory development enables qualitative research to contrib
ute to the development of broader empirical theories of social behaviour.
The importance of theory development may be depicted graphically, in Figure
1 0.6. The horizontal axis in Figure 1 0,6 depicts the path from description to


Conceptualizing and
Theory Development




Description and

Description ---"-.,--r---,:",-'-





, I


Figure 1 0.6



From description to theory development

Source: based on original diagram by Dahlgren et al. (2007: 1 22).

explanation and the vertical axis from concrete issues to abstract concepts. Data
analysis begins with a description of concrete issues, typically focusing on 'what'
type questions (What are the issues? What are the components of each issue7

What is the context of each issue? What are the problems, processes and
perspec tives in the data7) . Therefore, description provides the critical founda
tion of data analysis. However, description alone cannot explain a given
phenomenon, it only describes it. Data analysis needs to continue with abstrac
tion and conceptualization of data in order to explain the phenomenon and
why it occurs (thus moving to the top right of figure

1 0. 6) . At this point analy

sis can respond to ' how' and 'why' questions (How does it happen? How does
it w ork? How does it influence behaviour? Why does it happen? How can it be
changed?) . Answering how and why questions is a basic reason for conducting
quahtative research (see Chapter 2), so it is important that data analysis


to this point. So theory development is critical because it moves data analysis

beyond descriptive accounts and towards explaining phenomena .
All this may be emphasized by USing the an alogy of a crime scene, super
i mposed on Figure

1 0.6. Stopping analysis at description is akin to describing



a crime scene without solving the crime (Richards,

200S) . A thick description

would involve describing the scene: the dead body, the weapon, muddy
footprints, bullet c asings and fingerprints. However, this description has n o t
solved the crime, identified h o w or w h y it happened o r whether i t is likely

happen again. This re q 1.1ires th eory development, to link the evidence (or
codes) into a n explanation ( or theory) of what happened and why, there
fore m oving beyond description to explain and conceptualize the data .

. _ _ _ __ .

._____ . _

How to develop theory

Although there are many ways to develop inductive theory, what is important
is that it is systematically developed and well supported by data. Theory devel
opment from qualitative data is implicitly inductive as the codes, c oncepts)
categories and conceptualizations arise from the textual data collected. The
overall inductive process for theory development described throughout this
chapter is shown in Figure

1 0. 7 . Below we highlight some inductive approaches

for developing theory (i.e. following the analytic cycle, explicit reasoning), and
refining theory (i.e. comparing explanations, explaining outliers, seeking
negative cases) . We also highlight a range of additional (deductive)


for theory development (i .e. referring to a conceptual framework of the study,

applying deductive logic, inferring or borrOWing an explanation) . These deduc
tive strategies link the analytic cycle back to the original design cycle, to refine
the emerging theory and highlight the contribution of the empirical theory to
the field of study. The following are a range of strategies to assist in developing
and refining empirical theory.
Folfow tasks in the analytic cycle. Following tasks i n t h e a n alytic cycle w i l l b u i l d
up i n d uctive explanations for the p henomenon s t u d i e d a n d lead to theory
development. The p rocess of theo ry development i s s u m m arized vis u al ly i n
Figure 1 0 . 7 .
Identify explicit reasoning. Identify explanations given b y participants themselves to
build a theory. For example, participants m ay say that people do not undergo
regular health screenings for fear of the results, lack of symptoms, the time
involved, etc.
Compare explanations. Compare whether explanations d i ffer by subgroups of
partiCipants to identify diversity and build nuance into theory. For example, opinions
about home birth may d iffer by women having a fi rst o r later birth.
Explain outliers. Identify o u t l i e rs that do not 'fit' the e m e rg i n g theory. Consider
whet h e r t h e theory can exp l a i n these outliers or needs revision to accom mo
date outliers. Do not ignore oulliers ; exp l ai n i n g t h e m may reveal the gem of
you r study.



Develop Theory


Link Categories





Develop Categories




[f [f

Identify Codes


Code Code

Textual Data

F igure 1 0.7

Analytic tasks leading to theory development from textual data

Seek negative cases, Cons ider negative cases, which are cases in the data to r
which you r theory is not valid. These exceptions force you to probe further into data
to refine you r eme'rglTi r theory and i mprove its relevance and validity.
Refer to conceptual framework. Refer back to theory, concepts and explanations in you r
conceptual framework. Compare these with explanations from your data to identify
whether a new theory is emerging or new concepts can be added to exis ting theory.
Use deductive logie, Use deductive logic to identify explanations to add to theory,
For example, you may know from experience that the qual ity of child care at gym
fac i l ities deters women from attending exercise classes, and use this to identify
whether this explanation is also evident in you r data.
Infer an explanation. I nfer a theory by uncovering subtle reasoning, perhaps not
apparent to participants themselves. For example, comparing people who seek free
vaccinations with those who do not may revea l that non-users feel a stigma attached
to free services, Hence. an explanation of stigma is inferred and can refine a theory.
Borrow an explanation. Identify whether explanations given in the research litera
ture are also evident i n you r data, Take care not to force an explanation on 10 data
that is not well su pported.


For more on developing theory from textual data see, for example, Miles and
Huberman (1 994), Denzin and Lincoln (1 998), Strauss and Corbin (1 998),
Wengraf (200 1), Wolcott (200 1 ), Silverman (2005), and Rubin and Rubin (2005).
Grounding theory

How do you know that the theory that you have developed is valid? An impor
tant step in inductive theory development is to verify that a theory or explana
tion is grounded or well supported by data . Grounding a theory means demon
strating how the theory 'emerged' from the data, how the data support the
theory and whether the theory 'fits' the data.
There are three types of strategies that you can use to verify your theory
(see Table 1 0.4] . First, check that your theory is empirically grounded by using
consistency checks, returning to data, and using a concept-indicator model to
validate concepts in the theory. Second, check the 'fit' of your theory with the
data by 'testing' alternative theories to check the robustness of the emerging
theory. Finally, 'real-life' validity may be checked by taking interpretations of
data back to a group of participants to identify whether they can relate to the
emerging theory. Using these tools and strategies may lead to refinement or

Table 1 0 .4

Strategies for verifying empirical theory


Check the consistency of your theory across the data. If explanations are


continuously repeated, internal validity is strengthened.

Return to data

Reread data after your theory is developed to ensure explanations 'fit' and have
a strong foothold in the data. A theory that is distant from data or based on
superficial analysis can lead to misleading explanations.

Use the

Keep a concept-indicator (Strauss, 1 987) list to check that concepts in your


theory are well g rounded in data. For example, a concept of 'stigma' may have

indicator model

developed from indicators in the data of 'exclusion', 'discrimination', 'attitudes',

'community values' and so on.

Apply a

Use a conditional matrix (Glaser and Strauss, 1 967) to delimit a theory by


identiiying conditions in which the theory applies. For example, a theory may


hold true only for a certain subgroup of participants, in a specifiC context or when
a range of conditions are present. Conversely, the theory should not apply when
these conditions are absent.


Consider an alternative theory and identify whether it can be supported by the


data. Ii so, the original theory may not be valid or sufficient. A valid theory can be


well supported with data, and invalidate alter native explanations.


Present the theory to study participants to validate the


i nterpretations and explanations developed. This may identify that further


refinement of a theory is needed.



revision of the emerging theory and identification of nuances in a theory, all of

which strengthen the fit and validity of the final theory. A range of strategies
for verifying empirical theory is shown in Table 1 0.4.
When is your theory adequate? Richards (2005 : 1 3 0) states that a good and
adequate theory should meet the goals of the study and help to answer the
research question. It should go beyond description to develop a new explana
tion or framework to account for, better understand and explain the study
issues. An adequate theory should also offer more than participants could
h ave reported themselves, therefore it involves categorizing, conceptualizing
and theorizing. Finally, a theory should be useable for the intended purpose
of the study, for example to contribute to other theories, policy or social

_ . . ... .

. . ..

Eva I uati ng qual ity _

..... .

.. . .

.. .

_ ....

.... _.

. .. ____

__ .. .


_. ...


. "

'_ . . . . .

How do you evaluate the quality of data analysis in qualitative research? Based
on the type of analysis described in this chapter, consider whether the data
analysis process is transparent and well grounded in the data to validate the
concepts, categories, and ultimately the theory developed. Describing the
depth and nuance in the theory developed can also distinguish well-conducted

Are deductive and inductive analytic techniques used?

Does the analysis go beyond description to explanation?


Is the analytic approach identified?

Are the analytic techniques used made transparent?
Is the process of theory development outlined)
Are concepts clear and well described?


D o explanations follow logically from data?


Are categories and concepts developed well saturated?

G rounded

Are codes, concepts and explanations grounded in the data?

Are in vivo concepts reflected in the theory?
Are inductive conceptual models well grounded in the data)
Are the nuances and context of issues identified?
Does the analysis reflect the 'voices' of study participants?


Are any validity checks described?

How were the codes and concepts validated)
Are analytic interpretation and theory development validated?
Is the theory developed valid)



How do researchers manage subjectivity in analysis?

Is analytic reflexivity described?


Does analysis identify new information that emerged


Key Points

Formul ating a plan of analysis keeps your analysis focused and enab les you to

The analytic cycle comprises the core analytic tasks of description, comparison,

check your progress.

categorization, conceptualization and theory development. These tasks are closely

interlinked and conducted in a circular manner whereby tasks are repeated and
co nducted simultaneously.

Making a 'thick description' is the foundation of data analysis by describing the

Comparison allows you to further explore issues, identify patterns and begin to

context to behaviour which provides social or cultural meaning.

notice associations in the data. Comparisons can be made between i n d uctive and
deductive subgroups in the data.

Categorizing involves grouping codes with similar attributes into broad categories.

Conceptualization involves visualizing your data as a whole to develop a conceptual

understanding of the issues. Categorization and conceptualization are closely
linked, moving analysis to a higher level of abstraction and providing the building
blocks for theory development.

Theory development is largely inductive, but does involve the interplay with existing
deductive theories, which links the analytic cycle back to the original design cycle.

Theory development moves qualitative research beyond description and into the
rea l m of explanation, and towards a broader conceptual understanding of social

An important step in empirical theory development is to veri fy that a theory or expla

nation is 'grounded' or well supported by data. Researchers shou l d check the valid ity
and fi.t of the theory.

Exerc ises ,.,c<

. ......._ .... ..

Develop a 'thick description' of a code in the data. Identify the dimensions of the
code, and the context in which it is discussed. Use examples from the data to illus
trate the description.
Identify whether some codes have similar attributes, group these i nto categories

and label each category.


Try to conceptualize your data as a whole i n an inductive conceptual framework.



. __ ... . _ ..



Review the new theory you have developed from the data to consider the new
information it provides. I s the theory well grounded? Does it offer a more compre
hens ive expl anation of the issues than a single participant could have described?

Further reading on methods


Charmaz! K. (2006) Constructing Grounded Theory: A Practical Guide through

Qualitative Analysis. London: Sage Publications. This is a n excellent book that

synthesises the tasks of grounded theory in an accessible way! useful for both
novice and more experienced analysts.
Corbin! 1.} and Strauss, A (2008) Basics of Qualitative Research: Techniques and

Procedures for Developing Grounded Theory, 3rd edition. Thousand Oaks} CA: Sage
Publications. A landmark volume for outlining the practical tasks of conducting a
grounded theory ana lysis.
Lewins! A arid Silver,

(2007) Using Software in Qualitative Research: A Step-by

Step Guide. London: Sage Publications. This book provides a guide to using quali
tative software to support data analysis, It does not discuss analysis per se but
outlines critical analysis issues related to three leading software packages for analy
sis: Atlas,ti 5} MAXQDA 2 and NVivo 7 .
Richards, 1 . (2005) Handling Qualitative Data: A Practical Guide. London: Sage
Publications. This book provides practical guidance on setting up data in a project!
working with data and making sense of data .

o Further reading on field practice

Polzer, R . and Miles, M . (2007) 'Spirituality in African Americans with diabetes: Self
management through a relationship with God', Qualita tive Health Research, 1 7 (2):
1 76-88. This article is a good example of a well-written grounded theory analysis leading
to the development of an i nductive theoretical mode l .
Rajabiun, S . ; Malli nson, R . , McKoy, K . , Coleman, S . , Drainoni, M . , Rebholz, C . a n d Holbert, T.
(2007) '''Getting me back on track": The role of outreach interventions in engaging and
retaining people living with H I V/AIDS in medical care', A IDS Patient Care and STOs, 21
(supplement 1): S20-9 . This article has a good description of the grounded theory analytic
process, including a discussion on the development of categories a nd the development of
an ind uctive model.
H ay, J"


Cruz, G . a nd Ostroff, J. (2005) 'Thin king t h rough cancer risk:

Characterizing smo kers' process of risk determination', Qualitative Health Research, 15 (8):
10 74-85. This article uses grounded theory to develop a heuristic model.