Research Design

Research design is analogous to construction of a building. Before starting the procedures, making estimates, purchasing materials, setting up dates for completion of project, it is important to understand what sort building do I need. Similarly research needs a design or a structure before data collection or analysis can commence. A research design is not just a work plan.

The function of research design is to ensure that the evidence obtained enables us to answer the initial question as unambiguously as possible (Vaus, 2001: 9).

Obtaining relevant evidence entails specifying the type of evidence needed to answer the research question, or to test a theory, to evaluate a programme, or to accurately describe some phenomenon. In other words, when designing research we need to ask: given this research question (or theory), what type evidence is needed to answer the question (or test the theory) in a convincing way?

Section 1.01 Research question:

Research design begins with framing research question ends with analysing the data and presenting reports.

Research questions are fundamentally of two types:

What is going on? (descriptive research question)
Why is it going on? (explanatory research question)

(a) Descriptive research:

Descriptive research involves accurate description of a phenomenon. Good description is needed before preparing for explanatory research question. For example, before asking Why the gap between rich and poor is increasing, it is important first to understand whether the gap is actually increasing or not.

Much of the evaluation programmes, government sponsored researches are descriptive in nature. Questions such as “how secular is the society?” “How much poverty is there in the community?” are descriptive but significantly abstract.

(b) Explanatory research:

Explanatory research focuses on why questions. For example, after determining that the crime rate is increasing in the society, one might search for causes of such increase. This is then explanatory research.

Causation or establishing Causal connection is fundamental to explanatory research. Establishing causal connection requires stating factor X (gender) affects factor Y(Income).

The causal connection may be deterministic or probabilistic in nature.

A causation is deterministic where X is said to cause Y if, and only if X invariably produces Y. That is when X is present then Y will necessarily, inevitably and infallibly occur (Cook and Campbell, 1979). In social science however, because of complexity of human behaviour and the subjective, meaningful and voluntaristic components of human behaviour mean that it will never be possible to arrive at causal statements of the type.

(c) Framing a research question:

Research begins with a research question. At first, it is important to make the research question as unambiguous as possible.

(i) Descriptive research questions: points to remember

Scope of the core concept (What aspect of the concept the research tends to focus on)
Time frame of the description (Whether contemporary or change over time)
Geographical location
Degree of generalisability (whether pattern identification is the aim)
Degree of abstraction (whether the description indicates to some abstract concepts, divorce rate with secularism, isolation, etc.)
Unit of analysis (individual, family, society, phenomena)

(ii) Explanatory research questions: points to remember

Explanatory research questions demand further specifications of the focus “so the research question must be clear about the style of explanatory research and identify which causes or consequences it will investigate” (Vaus, 2001: 19). Some terms are frequently used in explanatory researches, these are:

a) Dependent variable:

It is a variable which depends on something else, usually treated as the effect in the causal model. It is usually designated as Y variable.

b) Independent variable:

Presumed cause is the independent variable. It is also called as predictor variable, experimental variable or the explanatory variable. It is designated as X. (Education (X) – income (Y)).

c) Intervening variable:

Variables that come between dependent and independent variables are intervening variables (Z). They are the means by which X produces Y.

d)

e) Extraneous variables:

Two variables may be correlated without being causally related. This correlation may be due to the two factors being outcomes of a third factor. This is also symbolised as Z.

f)

g)

h)

i) Searching for causes or effects:

Changes in divorce rate since World War II (X)

The least focus explanatory research starts with identification of a core phenomenon and then searches for possible causes. The core phenomenon may be changes in divorce rate after World War II and the possible causes may be changing values, decline in religion, changing population mix, economic changes, legal reforms etc. Alternatively one may search for consequences of rising divorce rate.

Article II.

The causal propositions can be Simple – an outcome of highly focused research question, or can be Complex which is an outcome of more exploratory research.

Taking different, competing approaches for explaining phenomena is another way of framing a research question. Different approaches are compared to find out which one is best fitted to facts.

a) Ideographic and Nomothetic explanations:

Ideographic explanation involves examination of relatively a few (hence partial) causal factors in large number of cases. Nomothetic explanations on the other hand focus on a particular case or a few of cases to draw nearly complete explanations.

Section 2.02 Range of research designs:

There are four types of research designs:

(i) Experimental designs:

One pretest
Two groups: experimental groups (exposed to the treatment), control group (not exposed to the treatment)
Random allocation to the groups before the pre-test
One intervention (test/treatment)
One pos-intervention measure on the outcome variables.

(ii) Longitudinal design:

One group
One pre-intervention measure on the outcome variable
One intervention where everyone receives the treatment
One post intervention measurement on the outcome variable.

(iii) Cross sectional design:

Instead of interventions the cross sectional design relies on existing variations in the independent variables in the sample.
At least on independent variable with at least two categories is present.
Data are collected at one point of time
There is no random allocation of groups.

(iv) Case studies:

Less depends on comparison more on exhaustive analysis of individual cases.
Might consist of single case study or multiple case studies which test a theory from multiple angles.
Treat a single case study as a single experiment.

Section 2.03 Research method and research design:

For Yin (1989) design deals with a logical problem and not a logistical problem. Research design concentrates on nature of evidence and not the methods of data collection. Design is often confused with methods of data collection. For instance Case study is a research design where as methods of data collection may include participant observation, questionnaire, interview etc. Research design should not equate to either quantitative or qualitative method. Yin (1993) for example strongly recommends researcher not to equate case study with qualitative method. Marsh (1982) similarly points out the possibility of quantitative survey in providing information and explanations that are adequate at the level of meaning.

Section 2.04 Importance of research design:

The need for research design stems from a sceptical approach to research and a view that scientific knowledge must always be provisional. The purpose of research design is to reduce the ambiguity of much research evidence. The prime importance of doing social scientific research is to seek for alternative ways of explaining a particular phenomenon. A proper research design helps in minimising the chance of drawing incorrect causal inferences from data.

(i) Identifying plausible rival hypothesis (Threats to the conclusions):

It is important to identify rival ways of accounting for the phenomena under study before carrying out the research. Vaus (2001) identifies two main types of rival hypothesis, first, theoretical and substantive rivals, and second, technical or methodological rivals.

1) Theoretical and substantive rivals:

Theoretical literature helps in identifying possible theoretical explanations of a particular phenomenon. It requires a close reading of available theories and then asking oneself how different theories would explain the research question. (How feminists explain this issue? What might conflict theorists say?)

Other researchers’ account must be incorporated for having a clear understanding of the competing explanations.

Taking practitioners’, key informants’, policy makers’, advocates’ perspective is crucial as ‘insiders’ practical knowledge of the field is invaluable.

Self experience is also very important.

Lateral thinking, i.e. looking beyond the purview of literature solely devoted to the topic often brings fresh perspectives (Page – 24)

2) Technical or methodological rivals:

Several technical or methodological factors may undermine conclusions drawn from a particular research, a good research design can minimise this threat.

Goldenberg (1992) outlines the types of methodological rivals. These are Situations including the context of data collection, attributes of the researcher and research participant, subjectivities of the researcher and research participant, sampling of items i.e. whether concepts are well measured, nature of sample, i.e. degree of generalisability, format of data collection – appropriate methods, analysis errors.

a) Operationalisation:

Most social science research involves use of concepts (tap concepts) upon which observations are made. If the research is on effect of marital breakdown we need first to work out what is meant by marital breakdown.

We use concepts that are not readily observable. Therefore, we need to translate concepts to something observable and recordable. A clear definition of concepts we use is an important component of research design, which in turn requires developing nominal and operational definitions. Because of the abstract nature of concepts, e.g. social class, marital happiness, it becomes necessary to develop indicators of them.

b) Clarifying concepts – nominal definitions:

Several concepts do not have a fixed or correct meaning. For example the concept of marriage breakdown can be seen from legal, emotional and logistics perspectives. Nominal definition specifies the meaning of the concept but remains abstract. Since, different definitions produce different findings, consequently defining concepts is crucial.

In order to arrive at nominal definition of the concept following steps can be followed:

Obtain a range of definitions from literature
Decide on a definition either by selecting a definition or by selecting relevant components from several existing definitions.
Delineate dimensions and subdimensions of the definition. For example the concept of child well being may have economic, psychological, physical, educational, social etc. Furthermore, the social dimension of child well being might have further subdimensions like relationship with peer, mother, father, etc. However, determining the dimensions and sub dimensions depends on the research question and nature of enquiry (Page 25 – 26).

c) Clarifying concepts – Operational definition:

Operational definition is the observations to measure the concept. Operational definition is the way in which concepts are measured. For example, marriage breakdown as a concept can be measured in terms of quality of relationships. It is then studied on the basis of levels of conflict, types of communication, signs of lack of affection and level of cooperation or lack of cooperation.

Once the operational definition of the concept is developed we come to final stage of operationalisation. This entails the precise way in which indicators will be measured. This might involve developing questions for a questionnaire or identifying what and how observations will be made.

Section 2.05 A Brief understanding of different research designs:

1. Experimental designs:

Apart from the physical and biological sciences, it is difficult for social sciences and humanities to conduct experiments. However, despite of several difficulties, social sciences often make experiments.

Classic experimental design deals with an independent and a dependent variable. Whether we can infer the impact of intervention over the dependent variable depends on how well we can make experimental designs.

The designs involve two groups, viz. control group, i.e. where no intervention is given, and experimental group, i.e. where intervention is done. People are allotted randomly to these groups. One pre-test – intervention (to the experimental group) – one post test. The pre-test and post-test variation between the two groups is attributed to the intervention.

There are three contexts in which experiments are possible, these are, a) laboratory context, b) field context (usually development related inputs and then impact assessment), and c) natural context (ongoing experiments in the social world)

Analysis of experimental data:

Choices of statistics

There are several statistical techniques for the analysis of data coming out of experimental designs. The core of experimental analysis is the comparison of groups. The difference between control group and experimental group is the primary focus of experiments.

There are two major types, viz. descriptive statistics, and inferential statistics.

DESCRIPTIVE STATISTICS	INFERENTIAL STATISTICS
Summarising the data	Extrapolation of the findings to wider population / generalisation
	Probability sample is necessary pre-requisite

Level of selection:

Level of measurement of variables is critical in selecting particular statistical analysis. There are three main levels at which variables are measured:

Interval or ratio variable:

Categories can be ranked from low to high in some meaningful way.
It is possible to specify the amount of differences between the categories.

Example: measuring age in terms of years, IQ test, income in rupees, etc.

Ordinal variable:
Categories can be ranked from low to high.
Can not specify the exact difference between categories.

Example: age in social categories, like childhood, adolescence, young adult, middle aged and elderly. We cannot specify exact differences between these categories.

Nominal variable:

Different categories do not have a set rank order.

Example: religious affiliation

2. Longitudinal designs:

There are different types of longitudinal designs which ultimately aim at measuring change over time. In consequence, it requires taking measurements in at least two time points.

To make longitudinal designs, it requires making certain critical decisions.

Should the same cases be followed?

It requires to make choices between trend studies and panel studies.

Trend studies: require taking data from comparable samples and not the same people over different period of time.

Panel studies: require repeated survey over the same people.
How many times do I need to collect data?

Although panel and trend studies require collecting data on two or several occasions. However, retrospective panel design does not require collecting data at different occasions. It depends on the memories of the informants.

Purposes of longitudinal designs:

Describing patterns of change and stability
Establishing temporal order.

Longitudinal designs help in tracking the order in which events take place. While cross sectional study can establish the link between mental state and employment, longitudinal study can track the range of emotional state before, immediately after and then in specific intervals of employment.
Establishing developmental (age) effects: effect of age over political, religious conservatism
Establishing historical (period) effects: while doing a research on political or religious conservatism, it is important to not the effects of historical events. Studying people at different time can look at the impact of historical events over people.
Life course or career analysis.

Types of longitudinal designs:

Simple prospective panel design:

Requires collection of data at two points of time from the same sample.

Multiple point prospective panel designs:

Everything is same as above except that it involves multiple points of time. It is useful in:

Examine short and long term effects
Track when changes occur.
Plot the shape of any change.
Factors that precede any change.

Retrospective designs:

It depends on the possibility of reconstructing data over time by collecting all information at one point of time.

It may be done by:

Retrospective panel design
Record linkage designs

Analysis of experimental data:

Since longitudinal design strives at understanding the change over time, measuring change is the most important component of its measure. There are certain issues to be kept in mind before going for the analysis:

Differentiate between aggregate and individual change:

A change may be at aggregate level, or individual level. Even when there is no change in over all aggregate level, there may be individual level changes. For example, in a study of employment and unemployment, over time same number of people may change their status from employed to unemployed and from unemployed to employed. Thus, there will be change at aggregate, but at individual level significant change is found.

To measure aggregate level change, we need to take aggregate measures, such as group mean, group variance, etc. To measure individual level change we need individual panel data to measure.
Qualitative and quantitative change:

The type of analysis to be undertaken depends on the type of change being considered. If the study involves qualitative change (Nominal variable, for example) the analysis will be different from analysis which looks into quantitative change (Continuous variable, for example)

Raw change score:

Quantitative dependent variable.
Calculate change scores for each person by substracting wave 1 scores from wave 2 scores and treating the difference as reflecting change.

Residual change scores:

Raw change score cannot identify the amount of change resulted from initial score. For example we can expect more change among the people who had initial extreme scores. One way to remove this effect is:

Residualise or regressed change scores
Identify cases where a person has changed more than would have been expected on the basis of their initial scores.
Regression analysis is done to predict wave 2 scores from wave 1 scores on the basis of a correlation of wave 1 and wave 2 scores.
Subtraction of predicted wave 2 score from actual wave 2 score. The remaining part is the residual gain score – the amount of gain that is not due to the influence of the initial wave 1 score.

Percentage change:

Calculate the percentage change in the initial score.
The formula is the following

Wave 2 score – wave 1 score

X100

Wave 1 score

Distinguishing between real change and lack of reliability:

Replicate the analysis using a separate measure to look for consistency

3. Cross sectional designs:

Three distinctive features characterise Cross sectional designs: a) no time dimension, b) reliance on existing differences rather than change following intervention, and c) groups based on existing differences rather than random allocation.

Therefore:

Cross sectional designs essentially depends on the differences between groups rather than a change.
Groups are constructed on the basis of existing differences in the sample. The sample is divided up into groups according to the category of the independent variable to which they happen to belong.
Obtaining time dimension is possible if the design involves collecting information at a number of different time points but from a different sample at each time point.

4. Case Study Designs:

Case study has been ignored or treated as soft options.
Some believe that case studies must be treated as a method for generating hypothesis to be tested eventually by rigorous research design.
It is important to understand that case studies have been fundamental to the substantive and methodological development of the social sciences. In social anthropology studies of tribes has been case studies.
Yin (1989) points out that unlike other research strategies, the potential catalogue for developing a research design for case studies is yet to be developed.

What is a case?

Units of analysis:

A case can be treated as unit of analysis. It is the unit that we seek to understand as a whole.
The unit may be individuals, events, place, organisations, decisions or even a time period. (220)

Holistic embedded units of analysis:

It is important to distinguish between holistic and embedded units.

Example:

School = as a whole

School having different subunits or elements:

Staff
1. Teaching staff
2. Non teaching staff.
Students
Guardians (220)

A well designed case study will avoid examining just some of the constituent element. It will build up a picture of the case by taking into account information gained from many levels. The final case study will tell us more than, and something qualitatively different from, that which any constituent element of the case could tell us. While doing research with all the subunits of a case, we must be able to tell a much fuller, more complex understanding of the whole than would the perspective provided by any particular element of the case (221).

Case study and theory:

Case studies are essentially theoretical. It may be used to test or build a theory.

Explanatory case study:

The case studies are different from other designs as they seek to achieve both fuller and detailed explanations of phenomena. They seek to achieve both ideographic (fuller explanation and understanding of a case) and nomothetic (explanation of a phenomena by studying different cases) explanations.

Theory testing:

This approach is seen by Yin (1989) as being at the heart of case studies, begins with a theory, or a set of rival theories, regarding a particular phenomenon.

For example, the study may begin with the following question:

“What is the effect of devolved, school based control of staffing on the quality of education in a school?”

A devolved system means, that the school has the authority to recruit, control the salary and other incentives of its staffs.

We may take a school as a case where the devolved system has been introduced. We then will thoroughly investigate the school, its embedded elements and also would try to use the historical data to see if there is any link between devolution and quality of education. The point of this case study is to see if the theory works in a practical situation. What modification is needed? Does the theory require any refinement? Is the theory applicable only to specific circumstances?

Theory building case studies:

Using a theory building approach to case studies we select cases to help develop and refine the propositions and develop a theory that fits the cases we study.

For example, we might begin with simple proposition that devolve system improves the quality of education. Then we may select a school where devolve system is introduced. If it indeed yields good result, then the theory is true. We can compare two schools where the same system has resulted differently. We may then try to find out the reasons for such variation.

The differences between a theory testing and building approach is that in the former we begin with a set of quite specific propositions and then see if these work in real world situations. In the theory building model we begin with only a question and perhaps a basic proposition, look at real cases and end up with a more specific theory or set of propositions as a result of examining actual cases.

Clinical case studies:

Try to build up a full picture of the case so that we can evaluate which explanation best fits the facts of the case.

Descriptive case studies:

Good description is essential prerequisite of good explanation.
Description cannot be atheoretical. We always select and organise that which we describe. Descriptions will highlight aspects of the case. It will be more like a painting of a landscape than a photograph. It will be an interpretation rather than a mirror image.
Descriptive case studies may consist of a single or multiple cases. One way of reporting multiple case studies is to use typologies and ideal types. Typologies may be theoretically or empirically derived. A theoretically derived typology is one that is logically or theoretically possible. For example, Merton (1968) develops a typology of types of deviants based on the notions of cultural goals and institutionalised means of achieving those goals.

Accepts means of achieving the goals		Acceptance of cultural goals
		Accepts	Rejects
	Accepts	Conformist	Rituallists
	Rejects	Innovator	Retreatists

Another way of achieving typologies is Inductive typologies, in which we start with a question and then examine cases in the light of the question. A comparison of cases can then highlight clusters of similar cases.

Dimensions of case study design:

Single or multiple – usually strategically selected multiple cases give a tough test to a theory.
Parallel or sequential – parallel design is one where all case studies are done simultaneously, usually by a number of researchers. In a sequential design, one case study follows another. Sequential design is effective for finding out issues to probe deeper.
Retrospective or prospective – retrospective design uses historical report, i.e., events are studied or data is collected afterwards. A prospective design goes on a considerable period of time, usually participation is needed.

Types of case studies:

Descriptive or explanatory
Theory testing or theory building
Single or multiple case
Holistic nor embedded units of analysis
Paralle or sequential case studies
Retrospective or prospective.

Explanatory?			Descriptive				Explanatory
Time			Retrospective		Prospective		Retrospective		Prospective
Case order			Parallel	Sequential	Parallel	Sequential	Parallel	Sequential	Parallel	Sequential
Cases	Units	Theory	Parallel	Sequential	Parallel	Sequential	Parallel	Sequential	Parallel	Sequential
Single case	Embedded units	Testing
	Embedded units	Building
	Holistic	Testing
	Holistic	Building
Multiple cases	Embedded units	Testing
	Embedded units	Building
	Holistic	Testing
	Holistic	Building

Variations of case study designs (Vaus, 2001: 229)

Case study analysis

Methods for analysing case study is less systematically developed than are the techniques for analysing data collected with other types of research designs.

Statistical analysis:

Case studies use statistics in a different way. Since, the aim of a case study is fuller and complex understanding of a phenomena, case study do not count number of occurrences of a given variable. However, a case may be statistically described. For example if we take a place as a case and population as embedded elements, we might need to describe the population statistically.
Meaning and context:

Many social scientists argue for the importance of incorporating contextual meaning in analysis. Case studies provide lucrative options of incorporating contextual factors in research design.

Analysing descriptive case studies:

Theoretical dimension:

Several scholars argue for the possibility of doing research without any theoretical orientation. This view suggests that facts must be allowed to speak for itself. However, Vaus (2001) in this book strongly denies any existence of such an approach.

He states that

“such an approach is both undesirable and impossible. Any description of any case always involves a selection of facts. This selection will be based on what we see as relevant and important. The very act of selecting means that we are making decisions about what is relevant. This selection process will be heavily influenced by our implicit theories. Furthermore, any reporting of a case will involve ordering the selected facts. The inevitable selectivity and ordering will mean that all descriptions are our descriptions, rather than the description of the case.” (2001:251)
Ideal type analysis:

Constructing ideal types based on theoretical and literature review is a way of describing cases. The analytic strategy is to use this ideal type as a template to guide the analysis of an actual case. We can use the template to see how closely our actual case fits the template.

The use of the ideal type provides a way of looking at and organising the analysis for the descriptions of actual cases. Using this approach we avoid description that simply describes whatever we happen to find out about the case or simply reports the features that catch our attention.
Typologies:

Typology is a set of types. We may construct types of different personalities, forms of government, types of organisational structure or types of marriages. This set of type may be based on ideal types or can be derived empirically. This approach helps in constructing an overall picture of a case taking a wide range of characteristics in account and not just the traits.
Cluster analytical approaches:

These approaches involve identifying a set of variables we want to use as the basis for our typology. For example, in a study of an organisation we might have collected from out case studies information about the way decisions are made, the way in which rules and regulations are used, the degree of hierarchy, degree of rationality etc. In a set of case studies we could then group cases that had similar constellations of characteristics.
Time ordered descriptions:

Time can be taken as a variable on the basis of which a case may be described. Histories of events, organisations, policies, or whatever the unit of analysis might be, represent a way of describing a case where the emphasis is on the sequence of events.

Explanatory case studies:

Theory plays an even more crucial role is explanatory case studies. Explanatory case studies are used either in theory building or theory testing.

When multiple case studies are used we may go for a two step analysis. First, we need to understand each case as a whole, and second, we can go for comparing these cases (Yin, 1989; Stake, 1994).

Theory testing analysis:

Theory testing can be done by a number of different ways. Yin (1989) outlines two approaches for doing this, a) Pattern matching and b) time series analysis.

Pattern matching:

Pattern matching is a form of theory testing in which we establish a detailed set of predictions before the case study is conducted. These predictions stem from a theoretical model and therefore represent a clear theory testing approach. The analysis could proceed by establishing a set of alternative patterns we would predict on the basis of competing theories.

The basic principal is that the more elaborate the predicted pattern (so long as it still follow logically from theory) the tougher the test of a theory. After the prediction of particular patterns we need to conduct the case study to see if the case does, in fact, match the predicted pattern. If the case matches the predicted pattern then the case supports the theory in the same way that a successful experiment supports a theory. If, however, the case does not match the predicted pattern the theory requires modification. Yin (1989) argues that on the basis of the degree of complexity of variables, i.e. with the increasing number of dependent and independent variables there are different types of pattern matching.

Simple patterns:

Simplest level pattern matching involves one independent variable with two values (e.g. male and female) and one dependent variable with two possible values (behaves in one of the two particular manners).

Example:

If we site the example of possible impact of devolve management over school’s performance, we have the devolve management system Xa and school’s better performance Yb.

Our prediction is

When Xa (devolve system/local based staffing system) exists then Yb (Better school performance/ high level of teachers’ commitment) will follow.

We would also expect that when Xb (centralised system) exists then Ya (lesser school performance/ low level of teachers’ commitment) will follow.

Dependent variable	Independent variable
		Xa	Xb
	Ya	Pattern 1	Pattern 2
	Yb	Pattern 3	Pattern 4

Table 1 Simple patterns: pattern matching for two variables each with two categories.

Additionally we may test alternative theories through case study. With the same issue of types of school administration we can have two theories.

Theory A: Local control will lead to higher commitment to work because effort and ‘fit’ is seen and rewarded (and lack of effort and not fitting in with school needs is punished)
Theory B: local control makes people feel more demoralised and vulnerable to local politics and prejudices, and does not recognise wider professional development etc. This leads to a lack of commitment and a lack of professionalism and to playing politics to win favour rather than fostering performance.

Complex patterns with multiple variables:

A more complex set of predictions is possible while using multiple variables. For example, if we use different variables of parental authority style and child’s level of anxiety.

Dependent variables
Independent variables	Parental authority style		Authoritarian		Authoritative		Permissive
	Child anxiety level		Low	High	Low	High	Low	High
	Child behaviour	Compliant	Pattern 1	Pattern 2	Pattern 3	Pattern 4	Pattern 5	Pattern 6
		Rebellious	Pattern 7	Pattern 8	Pattern 9	Pattern 10	Pattern 11	Pattern 12
		Child sets own limits	Pattern 13	Pattern 14	Pattern 15	Pattern 16	Pattern 17	Pattern 18
		Tests parental limits	Pattern 19	Pattern 20	Pattern 21	Pattern 22	Pattern 23	Pattern 24

Table 2 logical patterns with two independent variables and one dependent variable

Theory testing:

For theory testing we need to specify theoretical proposition to test it. If the proposition is not supported by a case study then the next step is to refine the theory so that it can take account of the exception provided by the case. In this way the proposition vocers a wider and wider set of cases and becomes more powerful/ This process is called analytic induction.

Time series analysis:

Vaus (2001) argues that even though Yin (1989) treats time series analysis to be different from pattern matching, its logic is same. For Vaus, treating time series analysis as a variation to the pattern matching is more logical. The analytic strategy involves predicting a particular pattern of change over time. This type of pattern analysis can take one of two forms: trend analysis and chronological analysis.

Trend analysis:

Trend analysis is an examination of the direction of change in a particular variable or a set of variables. We address the question of whether the trend is upward (steep or gradual), shows no change, variable (up and down) or downward (steep or gradual).

Based on number of dependent and independent variables, predicted trends can range from the simple to the highly complex.

Chronological analysis:

Chronological analysis involves predicting a sequence of events involving a number of different events or variables. Examples of a staged version of chronological analysis might be models that propose predictable stages in becoming a marijuana user (Becker, 1966), stages in the disintegration of intimate relationships (Vaughn, 1986), stages in the process of adjustment of retirement (Atchley, 1976) or changes in the relationships between adults and their parents as parents age (Marsden and Abrams, 1987). Yin (1989) indicates four types of ways in which events might be predicted to change in relation to each other.

Some events must always occur before other events, with the reverse sequence being impossible
Some events must always be followed by other events, on a contingency basis
Some events can only follow other events after a specified passage of time;
or certain time periods in a case study may be marked by classes of events that differ substantially from those of other time periods [stages].

Replication and generalisation:

Generalisation from the case study depends on the replication logic of experiments rather than the statistical logic of surveys. We gain confidence in experimental results not just from the elegance of the experiment but from out capacity to predictably replicate results and to predictably fail to replicate results.

We gain confidence in case study findings when we can accurately predict which types of cases will display particular patterns and which cases will not display specific patterns. Before strive for generalisability we need to ask the following questions:

Does the full set of outcome characteristics occur when the presumed causal factor is present? If so we have confirmation of our theory.
We would then find another case where the presumed causal factor is present and see whether the full set of outcomes is also present in that case. If so we have a literal replication of the previous case and further confirmation of out theory.
Do we get cases where the presumed causal factor is present but only some of the predicted outcome characteristics are present? If we find such cases then we have failed to replicate the theory and we would either reject or modify the theory. If we could find no cases where the cause was present and the full set of outcomes was not present then we have a theoretical replication.
We would then seek to find a case in which the presumed causal factor is not present. We would expect that the full set of outcomes would not occur when the cause was not present. That is, we should not find cases where we have the effects without our presumed cause. If we fail to find any such cases we have achieved further theoretical replication. (Vaus: 262 – 263)

Analysis for theory building: analytic induction

Analytic induction is ‘a strategy of analysis that directs the investigator to formulate generalisations that apply to all instances of the problem’ (Denzin, 1978:191). It is a method that can used to achieve descriptive generalisations or to arrive at causal explanations. It is a strategy that moves from individual cases and seeks to identify what the cases have in common. The common element provides the basis of theoretical generalisation. The modes of generalisations practiced by other research designs are based on statistical and probability calculations. WE estimate whether one group is more likely than other groups to behave in a particular way. Analytic induction, however, seeks to achieve universal generalisations.

Denzin (1978:192) summarises six key steps in the process of analytic induction:

Specify what it is you are seeking to explain (the dependent variable)
Formulate an initial and provisional possible explanation of the phenomenon you are seeking to explain (Your theory)
Conduct a study of a case selected to test your theory.
Review (and revise if necessary) your provisional theory in the light of the case or exlude the case as inappropriate.
Conduct further case studies to test the (revised) proposition and reformulate the proposition as required.
Continue with case studies (including looking for cases that might disprove the proposition) and revise the proposition until you achieve a causal proposition that accounts for all the cases.