The Marketing Review, 2001, 1, 373-397

www.themarketingreview.com

Michael J. Baker
Monash University, Melbourne

Selecting a Research Methodology

Having identified a research issue or question one must select a research strategy and appropriate methodology for collecting information that will illuminate the problem. In this paper we look first at the research strategy and the scientific approach to problem solving. The primary research methods of Observation, Experimentation and Survey are discussed in some detail. The paper concludes with a look at qualitative methodologies and the differences between cross-sectional and longitudinal research.

Keywords:

research strategy, scientific methods, observation, experimentation, survey, qualitative methods, cross-sectional and longitudinal research designs.

Introduction
This paper is a follow-up to two earlier papers that have appeared in The Marketing Review Developing a Research Proposal and Writing a Literature Review. In the latter paper we developed the theme that knowledge accumulates over time and that the purpose of research is to extend and improve both our knowledge and understanding. It follows that the first step in any research project must be to establish what is already known about the topic or problem in which we are interested and secondary research, or a review of the existing literature, is the place to begin. In the course of conducting a literature review it is possible that we will discover the answer to our research question and so have no need to undertake any new research as a solution to our problem already exists. Usually, however, we will find that while earlier work throws light on our problem it does not provide a precise solution to it. A gap exists between what is known and what we need to know to solve our problem. However, the literature review will enable us to define more clearly what the nature of this gap is and what additional information is required to eliminate it. Once we have defined our information needs then it becomes possible to determine the most appropriate method or methods for collecting this information. The purpose of this paper is to review these methods, There are several different methods that we will examine in some detail. First, however, we need to consider the more basic issues raised in the first paper Developing a Research Proposal concerning approaches to the conduct of research and research strategy. Next, we explore the three primary research methods Observation, Experimentation and Survey ISSN 1472-1384/2001/010373 + 24 4.00/0 Westburn Publishers Ltd.

374

Michael J. Baker

Research with a short discussion of qualitative methods. Finally, we review the difference between cross-sectional and longitudinal procedures.

Research Strategy
In the first paper in this series we suggested that there are two broad approaches to research positivistic and interpretivistic. This distinction rests basically on ones personal philosophy concerning the conduct of research with positivists emphasising an inductive or hypothetico-deductive procedure to establish and explain patterns of behaviour while interpretivists seek to establish the motivations and actions that lead to these patterns of behaviour. The selection of a research strategy will be strongly influenced by ones preferred approach. An excellent book that provides extensive advice on preparing research proposals is Designing Social Research by Norman Blaikie (Polity Press, Cambridge: 2000. This book was published after the first paper in the series was written and is strongly recommended). This book is based on many years experience of teaching courses on social science research and contains a wealth of advice for persons wishing to develop a research design. In it Blaikie discusses research strategy and methodology in detail. To begin with Blaikie makes a clear distinction between methodology and other aspects of the research process that are often subsumed under this umbrella term. Thus Blaikie uses the term research design for the planning aspect of a research project, research strategy for the logic of enquiry and methods for the execution of the project. Methodology, on the other hand, includes a critical evaluation of alternative research strategies and methods. (p.9) He continues: Research strategies provide a logic, or set of procedures, for answering research questions (p.24) and identifies four distinct alternatives Inductive, deductive, retroductive and abductive. These are defined as follows: The inductive research strategy is the commonsense view of how scientists go about their work. According to this view, meticulous and objective observation and measurement, and the careful and accurate analysis of data, are required to produce scientific discoveries. (p.102) Deductive research, also known as the hypothetico-deductive method, or falsificationism was developed by Popper to overcome the deficiencies of Positivism and the inductive strategy.(p.104). Deductive research involves the statement of a hypothesis and the conclusion drawn from it, the collection of appropriate data to test the conclusion and the rejection or corroboration of the conclusion. The retroductive strategy is the logic of enquiry associated with the philosophical approach of Scientific Realism(p.108). Like deductive research it also starts with an observed regularity but seeks a different type of explanation. In this strategy, explanation is achieved by locating the

Selecting a Research Methodology

375

real underlying structure or mechanism that is responsible for producing the observed regularityRetroduction uses creative imagination and analogy to work back from data to an explanation. (P.25) The fourth research strategy, identified by Blaikie as abduction, is associated with a range of interpretivistic approaches. These are particular to the social sciences whereas the three preceding strategies are common to both the natural and social sciences. Blaikie identifies himself as only one of three social scientists who have used the concept of abduction. In his words The idea of abduction refers to the process used to generate social scientific accounts from social actors accounts for deriving technical concepts and theories from lay concepts and interpretations of social life. (p.114) The basic distinction between the different schools of thought are spelled out clearly by Blaikie: Positivists are concerned with establishing the fundamental patterns or relationships in social life and Critical Rationalists are concerned with using such patterns to form explanation arguments. However, Interpretivists argue that statistical patterns or correlations are not understandable on their own. It is necessary to find out what meaning (motives) people give to the actions that lead to such patterns. (P.115) Clearly, seeking to understand behaviour is at the very heart of the marketing discipline and it is for this reason that interpretivism has attracted an increasing following among those undertaking research in the field. However, many researchers feel uncomfortable with what they perceive as a subjective and relativist view of the world by comparison with the objective and absolutist perspective of the physical scientist that underpins the scientific approach to problem solving. Thus the emphasis in both research and publication in marketing tends to positivist if for no other reason than that it yields results that can be readily tested and validated. By contrast the relativist view is that there is no single social reality but a multiplicity each of which is real to the persons sharing a given point of view and basing their actions upon it. Accordingly, the emphasis of this paper is on the positivist or scientific approach.

The Scientific Approach to Problem Solving

In the opening paragraph of the first chapter of what is widely considered to be one of the definitive texts for research students in the social sciences Selltiz et al (2nd Ed 1959) state unequivocally: The purpose of research is to discover answers to questions through the application of scientific procedures. These procedures have been developed in order to increase the likelihood that the information gathered will be relevant to the question asked and will be reliable and unbiased. To be sure, there is no guarantee that any given research undertaking actually will produce relevant, reliable, and

376

Michael J. Baker

unbiased information. But scientific procedures are more likely to do so than any other method known to man. To begin with then it will be helpful to define the nature of the scientific method itself. At the outset it is necessary to recognise that while social science research in general, and marketing research in the particular, follows the same basic process as pure scientific research and is based upon the same procedures of observation and experimentation, it differs significantly in that, as well as information, social science research is concerned with subjective, human values. The point about any research is that it is triggered by a desire to know more about a phenomenon and it appears possible that more information on this may be obtained through observation and experimentation. In the pure sciences where one is dealing with objective data then the cycle of scientific research would follow the sequence illustrated in Figure 1 which is derived from Popper (1968).
1. PROBLEM Rebuttal of existing theory or expectation 5. PREFERENCE Established between old and new theory A new theory or hypothesis 2. PROPOSED SOLUTION

4. TESTS

Attempted refutation including observation and experimentation

Of testable propositions from the new theory

3. DEDUCTION

Figure 1.
1968)

The Cycle of Scientific Research (derived from Popper

In the pure sciences the process of research is predominantly intellectual and prompted by the desire to improve and increase knowledge and understanding of the world in which we live. Intellectual curiosity also underlies applied and social science research but here the dominant motivation is usually practical - the desire to improve on something or even create something new with a specific end use or purpose in mind. That said, the process is largely the same albeit that the objects of social science research are often less amenable to research than other natural phenomena. In part this intractability may be due to the relative youth of modern social

Selecting a Research Methodology

377

science compared with the modern natural sciences (in the Golden Age of the Greek Philosophers and for many centuries thereafter there was no distinction between them!) In the natural sciences research has uncovered the existence of natural laws such that it has been possible to develop theoretical explanations not just of how things are but also how, given observance of the qualifying conditions, they will be in the future. While some progress has been made in the social sciences the theoretical foundations are less secure. The ability to replicate exactly tests of ones theory and hypotheses make it difficult, if not impossible, for the social scientist to predict outcomes with the same accuracy and reliability as the pure scientist. Of course, pure science doesnt stand still either which is why the research cycle is common to them both. At the outset one recognises the existence of a problem because ones existing knowledge and/or experience is insufficient to guide or inform one exactly how to proceed. Faced with a problem one is most likely to speculate or hypothesise as to the answer by drawing on past knowledge or experience and extrapolating from it. In cases where one has strong feelings about the likely outcome then one may formulate positive hypotheses while, where one is uncertain and can anticipate several possible outcomes then one would formulate a null hypothesis. Selltiz et al. (1959) cite Cohen & Nagel (1934) who stated: We cannot take a single step forward in any inquiry unless we begin with a suggested explanation or solution of the difficulty which originated it. Such tentative explanations are suggested to us by something in the subject matter and by our previous knowledge. When they are formulated as propositions, they are called hypotheses. The function of a hypothesis is to direct our search for the order among facts. The suggestions formulated in the hypothesis may be solutions to the problem. Whether they are, is the task of the inquiry. No one of the suggestions need necessarily lead to our goal. And frequently some of the suggestions are incompatible with one another, so that they cannot all be solutions to the same problem. Given the research hypothesis (or theory) then one will proceed to deduce a series of further working hypotheses (testable propositions) as the basis for determining precisely what information is required to enable one to test them. It is these Propositions which will largely govern the approach taken to data collection and the precise questions asked, and of whom if the data cannot be obtained by observation alone. Data collection and the question of how we analyse and interpret data in order to decide whether we can draw acceptable conclusions and come to a decision on our original problem are beyond the scope of the current paper. First, one must develop a research design and to do so will have to decide between observation, experimentation or a sample survey (or some combination of them) as the most appropriate means of securing the data necessary to test our hypotheses and, hopefully, solve the problem, which initiated the cycle. It is to this we turn in the succeeding sections.

378

Michael J. Baker

Observation
It is generally accepted that while marketing research is not a sufficient condition for competitive success, in the medium to long term it is certainly a necessary condition for continued survival. This conclusion may be reached by considering the explanations offered for marketing failures where a lack or absence of marketing research is most frequently cited as the primary cause. Alternatively, the reasons advanced for marketing success often cite the identification of a marketing opportunity unrecognised by competitors. Either way, keeping ones finger on the pulse of the marketplace is seen as an essential and continuing concern of the firms management. Observation may be both informal and formal. When undertaken as part of ones normal daily routine it is usually informal and loosely structured often referred to as scanning e.g. reading the Financial Times. Observation has a critical role to play in this activity. By contrast, observation within marketing research is a scientific technique - indeed it is the basis of the scientific method - which is characterised by a much more structured and systematic approach than that called for in a scanning mode. Fundamentally, the distinction between scanning and observation as a scientific technique is that scanning is only partly structured and is intended to maintain an awareness of information, actions and events which may have a bearing upon the decision makers judgement and/or action. On the other hand, observation consists of the systematic gathering, recording and analysis of data in situations where this method is more appropriate - usually in terms of objectivity and reliability - and able to yield concrete results (e.g. the flow of persons in a shopping centre) or provide formal hypotheses about relationships which can then be tested by experimentation or survey analysis. Hence, scanning is often a precursor of observation and may result in the formulation of tentative hypotheses leading to formal observation and the development of conclusions or formal hypotheses for further testing. Although observation may be regarded as a technique in its own right it is probably used most often as an element of one or other of the other two methods. In experimentation, observing and recording behaviour may well be the single most important technique while in survey analysis interviewers will frequently record factual information both to reduce the burden on the respondent (e.g. type of property, make of consumer durable) and to ensure accuracy where the respondent may be unsure about the correct answer (make of tyres on your car). A simple inspection will immediately provide an accurate factual answer. Similarly, a meter can record precisely what channel your TV is tuned to while questions asking you to recall this at some later date could result in considerable response error. A major advantage and disadvantage of observation as a research method is that it is very largely a real-time activity. With the advent of low cost video cameras it is true that one can record events for later analysis but this is certainly not possible for participant observation where the direct involvement of the observer is an essential part of the methodology.

Selecting a Research Methodology

379

Participant observation is particularly suited to the gathering of qualitative data where one is seeking to establish the behaviour of the subjects in a particular context (family decision making, board meeting, etc.). However, unless the observer is unknown to those they are observing, which is often difficult to achieve, there is always a danger of influencing the very behaviour one is seeking to monitor due to the control or Hawthorne effect. Similarly, another danger is that by becoming a participant the observer will change their own attitudes and/or behaviour and so introduce bias into their observation. (For an extended review of the uses and problems which relate to participant observation the reader should consult Moser & Kalton (1971)). Stephen Brown provides an example of the value of an observation study in his article Information seeking, external search and Shopping behaviour (Journal of Marketing Management, 1988, 4, No.1, 33-49). As the author notes in his introduction Few topics in consumer research have generated as much discussion as pre-purchase information seeking; but, that said, there is a clear difference between the research findings which show that consumers do not shop around and the retailers belief that they do. Brown suggests that both interpretations may be correct but reflect a different perspective. Thus, retailers observe how consumers behave, while the great majority of academic researchers survey consumers and invite them to recall and reconstruct their actual behaviour. But, in doing so, most respondents focus solely upon the purchase decision and ignore or forget their antecedent behaviour when they were acquiring information and evaluating it prior to making a shopping expedition. To overcome this deficiency Brown undertook a weeklong observation study in a shopping centre and observed the behaviour of 70 groups of shoppers from their time of entry to their time of departure. While only regarded as an exploratory study Brown identified three main types of shopping behaviour, leisure shopping (17%), chore or purposeful shopping (41%), and mixed activity (42%) from which it was inferred that consumers use the occasion of shopping trips to gather information on products which they are not purchasing, presumably for future reference should it be required, i.e. shopping around in the retailers sense is incidental to the main purpose of the trip. To test this hypothesis Brown proposes a much more rigorous design incorporating both observation and face-to-face interviewing underlining the importance of using observation to help formulate hypotheses and then combining it with other approaches to test those hypotheses. In sum, observation is usually the first step in the scientific method. Having identified a problem observation (successive focussing) allows one to define those areas or issues whose detailed examination may provide a solution to the problem.

Experimentation
Because of the dynamic and complex nature of most marketing problems few

380

Michael J. Baker

are readily amenable to the experimental approach used so widely in the physical sciences. However, some specific marketing problems are suited to an experimental design which, if properly controlled, will yield better and often less expensive data than can be obtained from a survey. In an experiment the researcher usually seeks to control all the variables so that by varying one while holding the others constant they can determine the effect of the input or independent variable upon the output or dependent variable. It follows that a basic requirement for the conduct of an experiment is that one must be able to specify all the relevant variables. It is also implied that one has a theory, which can be stated as a hypothesis or hypotheses about the nature of the relationship between the variables, e.g. if two identical objects carry different prices then prospective buyers will perceive the higher priced object to be of higher quality. Alternatively, one may hypothesise that the colour of a products packaging will influence the consumers perception of it without any specific hypothesis about what colour will have what effect. In a nutshell, experiments are usually undertaken to determine if there is a causal relationship between the variables under investigation. Moser & Kalton (1971) avoid a detailed philosophic discussion of the nature of causality but provide a useful guide to its determination. If A is a cause of the effect B then normally there would be an association between A and B. Thus, if smoking causes lung cancer one would expect that more smokers than non-smokers would contract the condition. All the statistical evidence points to this conclusion but some will seek to dismiss or ignore it on the grounds that smoking is neither a necessary nor sufficient condition for contracting lung cancer. A sufficient condition would mean that all smokers would invariably get lung cancer while a necessary condition would exist if people only got lung cancer after smoking. Clearly, neither of these conditions obtain, as there are many non-smokers who catch lung cancer. That said, the degree of association between smoking and lung cancer is so high, i.e. we can measure how many smokers and non-smokers get lung cancer and will discover that the likelihood of a smoker getting the disease is many times as great as a non-smoker, that the Government now insists that tobacco products should carry a formal health warning. Evidence of causality is also to be found in the sequence in which events occur, it being obvious that a subsequent event cannot be the cause of an antecedent event. That said, while an antecedent event may be the cause of a subsequent event we will have to apply other tests of causality to determine whether in fact this is the case. In addition to tests of association we will also have to determine if there are any intervening variables, which may influence the apparent relationship (or even disguise it by masking the relationship or making it disappear). Moser & Kalton (1971) exemplify the problem in terms of the observed association between the income of the heads of households (I) and their conservatism (C) as measured by some suitable index. While C may increase or decrease in parallel with I (or vice versa!) and we may measure the degree of association through the calculation of the coefficient of correlation we cannot impute any causality to the relationship without also

Selecting a Research Methodology

381

determining the time sequence - did I precede C or vice versa and whether or not their are any intervening variables such as education, occupation, age or whatever, that might explain the relationship better. The more the possible number of intervening variables, the more complex the task of determining whether or not they have any bearing upon the relationship. It follows that in designing experiments one must have a particular concern to determine the degree of association, the sequence of events and the possible effect of intervening variables. To address these problems 3 broad kinds of experiment are available, all of which require the establishment of a control group against which to compare the experimental group, namely: 1. After-only design 2. Before-after design. 3. Before-after design with control group. In an after-only design the experimental group is exposed to the independent variable which it is hypothesised will cause a particular effect and their subsequent behaviour or condition is compared with that of the control group which has not been so exposed. For example, a new cold remedy is administered to 100 volunteers and the progress of their cold compared with that of the 100 members of the control group who must be as closely matched as possible with the experimental group in terms of age, sex, physique, general state of health, etc. If the experimental group show a dramatic improvement then, assuming no intervening factors, we may assume there is a causal relationship between remedy and cure and express this in quantitative terms of the degree of association established. Before-after designs are commonplace in marketing where the experimenter is seeking to determine the effect of a specific factor on peoples attitudes and/or behaviour. In these cases the same people are used as both the control and experimental group. Suppose one wishes to assess the effect of a change in a marketing mix variable - price, packaging, product performance, advertising, etc. etc. - upon a group of consumers, the first step must be to establish a benchmark of their current attitudes/ behaviour. Given the benchmark the group is then exposed to the modified marketing mix (always bearing in mind you can only vary one factor at a time) and their attitudes/behaviour measured again. Because of the experimental effect (see below) it would be dangerous to assume that all change detected is a direct consequence of the variation in the mix variable - that said, if change occurs it is reasonable to assume that, in part, this is due to the change in the independent variable. The third design - before-after with control group - is generally recognised as the best of the three options in that it combines elements of both the preceding types and, most importantly, has established a benchmark for the control group before exposing the experimental group, i.e. one can quantify the similarities between control and experimental groups and not merely

382

Michael J. Baker

assume their similarity as is the case with the after-only design. In all experiments the primary concern must be for validity and Moser & Kalton (1971) cite Campbell & Stanleys (1966) checklist of the 12 most frequent threats to validity. The first 8 of these refer to internal validity and are: 1. History. Here the problem is to ensure that some extraneous factor, including prior experience, is not the cause of an observed outcome. The establishment of a control group and clear definition of the antecedent beliefs and status is the major defence against this source of invalidity. Even so, the fact that many experiments are not instantaneous but involve the passage of time may allow some external event to influence the experiment. 2. Maturation. Independent of any specific environmental event the passage of time is likely to influence respondent performance, e.g. fatigue, boredom or, in the case of experiments designed to measure long-term effects, ageing itself. Before-after designs with control groups offer the best protection against this source of invalidity. 3. Testing. Given that we learn from experience it is unsurprising that the mere act of participating is likely to result in modification of our attitudes/behaviour if for no other reason that it gives salience to something, which previously had been of little or no importance. Once again the before-after with control design enables us to assess the effect of this factor. 4. Instrumentation. The conduct of experiments frequently involves the use of more than one test instrument, e.g. questionnaire, or the interpretation of test scores, which involves judgement. The potential for bias is obvious and can only be guarded against by careful testing of instruments and the establishment of objective measures wherever possible. 5. Statistical Regression. For a variety of reasons discussed in some detail by Campbell & Stanley (1966) there is a natural tendency for test subjects to gravitate towards the mean score on a given test when tested more than once. Whether this is because poor performers try harder and high performers become complacent or, as in the case of say the Delphi forecasting technique deviants gravitate towards the security of the average is a matter of speculation - the existence of the phenomenon is not! 6. Selection. Experimental design demands that experimental and control groups should be as near identical as possible. Patently, to claim that the

Selecting a Research Methodology

383

inhabitants of Slough (a favoured location for test markets) are the same as the inhabitants of, say, Aberdeen because the demographic structure of the towns populations are similar (an after-only design) is a travesty of reality as anyone familiar with the two towns will appreciate. Before-after designs help overcome some of the problems of bias due to selection as the same persons are being studied throughout. That said, many of the other threats to validity - maturation, testing, etc. - will be more acute. Specialist texts, such as Moser & Kalton (1971) provide extensive advice on how to minimise these problems, which are beyond the scope of a practitioner orientated article of this kind. 7. Experimental mortality Quite simply means that some of the subjects will die or drop out during the course of the experiment and so may threaten its validity. 8. Selection maturation This is a problem that occurs when persons volunteering for an experiment may possess a quality absent in non-volunteers and it is this which leads to the observed effect or outcome. In addition to the 8 sources of internal invalidity cited Campbell & Stanley (1966) also give examples of 4 sources of external invalidity, i.e. the extent to which the findings can be generalised and applied to other persons. One example of this is Reactivity. Reactivity describes the interaction effect of testing as a result of which the subjects behave differently than they would otherwise have done. A classic example of this is the so-called Hawthorne Effect in which the performance of both experimental and control groups improved largely because of the awareness of the groups that they were being monitored. According to Moser & Kalton (1971) Often in designing studies the demands of internal and external validity compete; the stronger the design is made in internal validity, the weaker it becomes in external validity. In general, surveys are strong on external validity but weak on internal, while for experiments it is the other way round. The authors provide extensive advice for controlling the effects of extraneous variables (p220224) but these are of primary interest to the specialist and will not be considered further here.

Survey Research
While observation and experimentation both have an important role to play in marketing research it is the survey which is the best known source of primary data collection, not only in marketing but the social sciences in general. Undoubtedly this owes a great deal to their widespread use in polling opinion on political issues or other matters of current interest and concern such as health and food, or the effects of environmental pollution.

384

Michael J. Baker

What then is a Survey?

Moser & Kalton (1971) devote over 500 pages to the subject of Survey Methods in Social Investigation but decline to offer a definition on the grounds that any such a definition would have to be so general as to defeat its purpose, since the terms and the methods associated with it are applied to an extraordinarily wide variety of investigations, ranging from the classical poverty surveys of 100 years ago to Gallup Polls, town planning surveys, market research, as well as the innumerable investigations sponsored by research institutes, universities and government. In the Macmillan Dictionary of Marketing and Advertising Baker is less restrained and asserts that a survey is The evaluation, analysis and description of a population based upon a sample drawn from it. In the Marketing Handbook (1965) Mayer adds to this attempt at a definition when he states The essential element in the survey method is that the data are furnished by an individual in a conscious effort to answer a question. Thus Mayer sees the essence of surveys as posing questions (the questionnaire technique) and goes on to add, The survey method is the most widely used technique in marketing research. Some people go so far as to regard the questionnaire technique as being synonymous with marketing research. Unfortunately, the method is so universally employed in this field that many researchers use the survey technique when one of the other methods, the observational or experimental, is more appropriate. Tull & Albaum (1973) support all three of the preceding views when they write, Survey research is a term that is susceptible to a variety of interpretations. As most often used, it connotes a project to get information from a sample of people by use of a questionnaire. The question may be designed to obtain information that is retrospective, concurrent, or projective with regard to time. They may be asked in a personal interview, by telephone, or sent to the respondent by mail. Tull & Albaum stress that surveys are concerned with understanding or predicting behaviour and offer as their definition: Survey research is the systematic gathering of information from (a sample of) respondents for the purpose of understanding and/or predicting some aspect of the behaviour of the population of interest. Consideration of these definitions indicates that surveys are concerned with Fact finding By asking questions Of persons representative of the population of interest To determine attitudes and opinions; and To help understand and predict behaviour.

The design and execution of surveys and the methodology and techniques available is the subject for another paper. Here we must be content with a broad overview of the survey method covering:

Selecting a Research Methodology

385

The Purposes for which surveys are used The Advantages and Disadvantages of Survey Research Issues and Topics suited to surveys.

The Purposes of Survey Research

Mayer identifies 3 kinds of survey, which he classifies as factual, opinion and interpretive, each of which is seen as having a distinctive purpose. As the name implies factual surveys are concerned with securing hard, quantitative data on issues such as usage, preference and habits, e.g. How much beer to you drink? What is your preferred brand? Where do you normally drink beer? etc etc. In other words, such surveys are concerned with actual behaviour while in opinion surveys the objective is to get respondents views upon the topic under consideration. Such opinions are almost always qualitative and may or may not be based upon actual experience. For example, a teetotaller may have quite strong perceptions about brands of beer without ever having consumed them. However, consumers attitudes and beliefs based upon past knowledge and experience are of particular value in helping to plan future strategy, e.g. in designing a new product, developing a copy platform or selecting a distribution channel. As with factual surveys, a major purpose is to try and quantify the strength and direction of opinion as a basis for future decision-making. By contrast interpretive surveys are used in circumstances where the respondent is asked to explain why they hold particular beliefs or behave in a particular way rather than simply state what they do, how, when and where, etc. Interpretive surveys are often the first step in primary data collection when the researcher is trying to get a feel for the topic under investigation and will often involve the use of projective techniques such as picture and cartoon tests, word and object association tests, sentence completion tests, and role playing. Depth interviewing and focussed group interviews are also widely used, often to define questions to be used in a formal questionnaire for use in the factual or opinion survey. Most market research surveys have a practical objective - to provide information on which to base decisions - while surveys of other populations are usually undertaken for theoretical reasons to enhance understanding. While such information may well be used as an input to decision making this is not its primary purpose. Table 1 summarises the multiplicity of surveys identified by Moser & Kalton (1971), the great majority of which are undertaken from a background of intellectual curiosity rather than to provide inputs to the solution of particular problems.

386

Michael J. Baker

Table 1. Taxonomy of Surveys

1. 2. 3. 4. 5. Classical Poverty Surveys (Booth, Rowntree, Bowley, Ford) Regional Planning Surveys Government Social Surveys Market, audience and opinion research Miscellaneous:

- Population (Census) - Housing - Community studies - Family life - Sexual behaviour (Kinsey) - Family expenditure - Nutrition - Health - Education - Social mobility - Occupations and special groups - Leisure - Travel - Political behaviour - Race relations and minority groups - Old age - Crime and deviant behaviour Source: Adapted from Moser & Kalton (1971).

Advantages and Disadvantages of Survey Research

In an article The use of the survey in industrial market research (Journal of Marketing Management, 1987, 3, No.l, 25-38) Susan Hart confirms that the survey is the most usual form of primary research undertaken and attributes its popularity to the following factors: (i) The objectives of most research require factual, attitudinal and/or behavioural data. Survey research provides the researcher with the means of gathering both qualitative and quantitative data required to meet such objectives. (ii) One of the greatest advantages of survey research is its scope: a great deal of information can be collected from a large population, economically. (iii) Survey research conforms to the specifications of scientific research: it is logical, deterministic, general, parsimonious and specific Alreck & Settle (1985) consider that the main advantages of surveys are that they are:

Selecting a Research Methodology

387

Comprehensive Customised Versatile Flexible Efficient By comprehensive Alreck and Settle mean that the method is appropriate to almost all types of research (cf. Mayers factual, opinion and interpretive categories of survey). The other four advantages are closely interrelated and boil down to the fact that one can design surveys to suit all kinds of problems and budgets. Naturally, dipstick or quick and dirty research where a limited budget and time pressures dictate only limited sampling using judgemental methods will lack the credibility (and validity) of the properly designed survey in which carefully designed and tested questionnaires are administered professionally to a statistically representative sample of the population. That said, virtually any research is better than none and there is considerable evidence to show that diminishing returns set in early in terms of the insights gained from research and also in terms of the confidence one can attribute to ones findings. Given then that the essence of marketing is that one should seek to determine the precise needs of prospective customers, a pragmatic approach which seeks to acquire additional information consistent with available resources is to be preferred to no research on the grounds that it lacks the rigour called for in the experimental sciences. While such rigour is attainable, the question is whether it is possible, necessary and/or worthwhile. In most student and practical projects it is not! Of course surveys also have their disadvantages and Hart (1987) cites the following: (i) The unwillingness of respondents to provide the desired data. The overriding concern here is of the non-response error, which can invalidate research findings. (ii) The ability of respondents to provide data. In studying managerial decisions, it is important to target individuals in the organisation with the knowledge and experience of the subject under examination. (iii) The influence of the questioning process on the respondents. Respondents may give the answers they think the researcher will want to hear, thus distorting the accuracy of the data. Response errors, accidental or deliberate, may be reduced significantly through careful design and execution of the test instrument and were discussed by Webb (2000).

Issues and Topics Suited to Surveys

388

Michael J. Baker

From the above discussion, and particularly Table 1, it is clear that surveys can be used to gather data on virtually any problem which involves the attitudes and behaviour of people in either their individual capacities or as members of various kinds of social and organisational groupings. Within the domain of marketing research Alreck & Settle (1985) distinguish eight basic topic categories, namely: Attitudes Images Decisions Needs Behaviour Lifestyle Affiliations Demographics Of course, these categories are neither mutually exclusive nor independent but to the extent that they often require different treatment and measures the classification provides a useful guide to survey planning. Alreck & Settle subscribe to the cognitive-affective-conative (CAC) model of attitudes and see attitudes as a predisposition to act. They also assert that attitudes precede behaviour. In other words if you can define and measure attitudes you should be in a good position to predict behaviour. However, many researchers prefer the expectancy-value (EV) model developed from the work of Heider, Rosenberg and Fishbein which does not seek to establish a link between attitude and behaviour and so can accommodate problems of the kind touched on earlier when examining the purposes of survey research - namely, the teetotaller who may have strong positive attitudes towards brands of beer but with no intention of translating their attitude into behaviour. The point we are seeking to make is that one must be careful not to assume a causal relationship between attitude and behaviour albeit that the notion of consistency in both CAC and EV models indicates that if behaviour does occur it is most likely to be consistent with the pre-existing attitude - if one exists! Either way an understanding of underlying attitudes will clearly be of great value to the marketing planner. Image is defined in the Macmillan Dictionary as Consumer perception of a brand, company, retail outlet, etc. Made up of two separable but interacting components, one consisting of the attributes of the object, the other consisting of the characteristics of the user. The most important word in this definition is perception. Objectively it is possible to list all the attributes of a product or service but the importance assigned to these attributes is likely to vary from individual to individual. It is for this reason that in personal selling the seller will invite the potential buyer to list and rank order the specific attributes they are looking for in the intended purchase so that, in turn, the seller can focus on those elements of importance to the particular buyer. But in mass markets, or when composing copy in support of personal selling, one

Selecting a Research Methodology

389

must seek to determine the image of the object under consideration and then define clusters of attributes which correspond to worthwhile segments in the market place. Surveys offer this potential, particularly through the use of scaling devices; another topic for a later issue! Given knowledge of peoples attitudes and some insight into their image of different products or services the next thing the marketing planner would like to know is how they actually choose between alternative courses of action. Elsewhere (Marketing, Strategy and Management (2000) Chapter 9) I have proposed a simple, composite model of buyer behaviour, which incorporates elements of the economic, psychological and sociological models and argues that, when faced with a need, the consumer will consult their store of past experience (learned behaviour) to see if they have an acceptable solution. If not, the decision maker will seek to acquire relevant information on performance attributes and cost- benefit data on possible solutions to their need. This information will then be evaluated in terms of the decision makers own perceptions and preferences and a choice made of the item, which best satisfies these. If marketers are to influence this process then they require knowledge of the information used by consumers and the evaluative criteria applied in order to arrive at a preferred solution. Survey research offers a means to acquire this information. Needs is a word often used loosely by the marketer and as if it were synonymous with wants, desires, preferences, motives and goods. The Macmillan Dictionary makes it clear that only if there is no choice will needs and wants become synonymous otherwise needs represent basic requirements such as food, shelter, clothing, transportation, entertainment, etc. (c.f. Marketing Myopia by Ted Levitt) while wants comprise highly specific means of satisfying these basic needs. Desires are largely synonymous with wants but the term implies rather more commitment to a given solution than preference. Motives have a connotation of action in that people are seen as being motivated to do something usually to achieve a goal, which is satisfaction of the need through behaviour. Scaling provides a means to documenting all of these dimensions through survey research. According to Alreck & Settle (1985) The measurement of behaviour usually involves four related concepts: what the respondent did or did not do; where the action takes place; the timing, including past, present, and future; and the frequency or persistence of behaviour. Such data is factual and readily acquired by standardised questionnaires using what, when, where and how often questions, usually with multiple-choice answers provided to speed up completion. Of course such questions are frequently combined with others designed to measure the other categories of topics and may be regarded as a benchmark against which to assess them. While it is convenient for purposes of data collection and analysis to identify clear-cut factors or variables such as Attitudes, Images, etc. actual behaviour is the outcome of the interaction of all these factors. In order to capture this complexity many researchers prefer to use composite measures, which describe behaviour in the round. Lifestyle is such a construct - and is

390

Michael J. Baker

concerned with Distinctive or characteristic ways of living adopted by communities or sections of communities, relating to general attitudes and behaviour towards the allocation of time, money and effort in the pursuit of objectives considered desirable (Macmillan Dictionary). Validated batteries of items for inclusion in lifestyle research are widely available and usually cover four main dimensions - activities, interests, opinions (hence AIO research) and possessions. The identification of subgroups or segments with similar lifestyles has been greatly improved with the development of multivariate analysis techniques utilising the full capabilities of modern computers. Behaviour is strongly mediated by the social context in which it occurs. It follows that an understanding of the nature of social interaction is central to an understanding of actual behaviour. In this context affiliation in terms of the membership of both formal and informal groups is a rich source of data of particular value in predicting likely behaviour. For example, membership of a referent group may exercise a significant influence on what is considered an acceptable dress code (IBM and blue suits, students and jeans), place in which to live, type of holiday, car, etc. Similarly, an understanding of the composition of a group and relationships within it may provide useful guidance when introducing a new product to the market place by helping to identify the opinion leaders (those to whom other group members turn for information and advice), and concentrating ones selling-in effort on these individuals. Survey methods are very appropriate to research of this kind. Alreck & Settles final category demographics is perhaps the most obvious and easily measured dimension associated with consumer behaviour. In addition most governments collect copious data on the structure and composition of the population, which provides a reliable and inexpensive base line for other survey research.

Qualitative Research Methodology

When considering the issues involved in defining problems and selecting possible approaches to solve them earlier we touched on the distinction between quantitative and qualitative research. While these two approaches are often presented as if they are opposing and mutually exclusive research methodologies it was pointed out that they are in fact complementary and supportive approaches to the conduct of research. Our discussion of the research cycle and the main techniques of observation, experimentation and sample survey in the preceding pages should have confirmed this view. However, some further elaboration will be useful here before turning to a more detailed examination of sampling and data collection techniques in future papers. Perhaps the main reason why many practitioners are suspicious of the value of qualitative research is because of its identification with motivation research and particularly motivation research of the kind developed by Ernest Dichter in the 1950s. It was Dichters approach to motivation research and its

Selecting a Research Methodology

391

application to business problems which was the subject of Vance Packards The Hidden Persuaders in which he claimed that marketing researchers were using psychological and psychoanalytic techniques in order to discover the hidden or subconscious motivations of consumers and then using this information to persuade these consumers to purchase particular goods or services. Following on McCarthyism and the Red scare of the early 1950s anything, which smacked of brainwashing was clearly to be avoided. On mature reflection (and given many numerous election failures) if is clear that short of coercion you cannot make people do what they dont want to do. Thus, while one may debate the social implications and consequences of materialism and the acquisitive society, it is usually those who are materially well-off who seem concerned that others may be persuaded to aspire to the same level of material comfort who are in the van of the critics. Fortunately, technological innovation makes it seem increasingly likely that the majority of mankind will be able to aspire to higher levels of material comfort and so share a common concern for responsible consumption. However, this is to digress from the main point, which is that qualitative methods were brought into some disrepute by criticisms of some of Dichters methods and it has taken a generation for a more balanced approach to emerge. Experience has shown that qualitative research is particularly useful in a number of specific situations, which may be summarised as: 1. 2. 3. 4. 5. Traditional preliminary exploration Sorting and screening ideas Exploring complex behaviour Developing explanatory models of behaviour Enabling the decision maker to experience the world as consumers see it 6. To define unfilled needs and means of satisfying them. Most of these uses are self-explanatory and address the issue raised at the beginning of this paper, namely, that social science research is concerned not only with facts but also with values. Through using qualitative research it becomes possible to discover what some of these underlying values are for, while one may seek to infer them from observing actual behaviour, the only real way one can establish why? people behave as they do is to ask them. Even then it is not easy to get respondents to give you the real reason for we all have a tendency to rationalise behaviour (hence the emphasis upon price as an acceptable reason for not making a purchase rather than by giving offence and saying the article in question was useless, ugly or what have you). There is also the well-known human tendency to want to please and so give the researcher the answer you think they are looking for. To overcome these difficulties qualitative researchers have developed a whole battery of projective techniques so that the respondent is invited to speculate how someone else would behave in a given situation, e.g. by completing a sentence, by filling in the dialogue in a cartoon, etc. Of course,

392

Michael J. Baker

the only real basis we have for such speculation is our own knowledge and experience, attitudes and opinions, so it is hardly surprising if the projected behaviour is similar to how we would behave in the given situation. Gordon & Langmaid (1988) devote a whole chapter to Projective and Enabling Techniques, which they classify into 5 categories: Association Completion Construction Expressive Choice-ordering Within the Association procedures are to be found traditional word association tests such as Tell me the first thing that comes to mind when I say detergent, through the classic Rorschach ink-blot test, to the construction of brand personalities, e.g. could you imagine Fosters Lager as a person and describe them to me. A further refinement is to provide the respondent with a pile of words and pictures and ask the respondent to choose those they associate with a brand name or product. Completion procedures invite respondents to complete a sentence such as People who drive Porsche motor cars are... or the missing dialogue in a conversation between two persons. Brand mapping, in which respondents are invited to group like brands/products according to various criteria, is also regarded as a completion technique and enables the researcher to determine how consumers see the products competing with each other in the market place. Construction procedures also invite respondents to construct a story from a picture (Thematic Apperception Tests) or in response to projective questions, through bubble procedures (you write in what you think the character in the drawing/cartoon is thinking) to the classification of stereotypes, e.g. you define a category of consumer and ask respondents to specify their consumption behaviour. Expressive procedures also involve the use of drawings and the invitation to the respondent to describe their perceptions of the person and/or context. They may also include role-playing in which the respondent is asked to act out a particular activity, e.g. purchasing a product or playing the part of a named brand. Finally, choice ordering is just that and asks respondents to rank order objects in terms of specific criteria - a technique which is very useful in determining what alternatives will be provided in multiple choice questions or for coding the answers to open-ended questions. However, projective techniques are only one approach to qualitative research and probably used less frequently than the group discussion and depth interview. Both topics for another paper.

Selecting a Research Methodology

393

Cross-Sectional and Longitudinal Studies

There can be little doubt that when people speak of marketing research they are concerned primarily with a specific, one-off or cross-sectional study designed to address a particular marketing problem, which is facing them. Given my own preferred definitions that marketing is selling goods that dont come back to people who do and that marketing is concerned with mutually satisfying exchange relationships it is clear that one salient feature which distinguishes marketing from the production or sales orientation is that it is concerned with developing lasting relationships with specific customers over time. This being so it would seem sensible that a significant part of the firms research effort should be devoted to such longitudinal studies - a view which is considerably reinforced when it is recognised that it costs approximately 5 times as much to create a new customer as it does to retain an existing one. In this section we shall examine the nature of longitudinal research by comparison with ad hoc or cross-sectional studies and then review its use in both the industrial and consumer setting.

What is it?
According to Wancevich & Matteson (1978) Longitudinal organisational research consists of using techniques and methodologies that permit the study, analysis and interpretation of changes that occur over a time period sufficiently long to assess meaningful change in the variables of interest, as well as to facilitate researcher and managerial understanding about causality. In other words, longitudinal analysis is concerned with the collection of data by means of observation, experimentation or over time with the primary purpose of describing and explaining change in these objects or respondents over the period of the research. In the social sciences generally there are numerous examples of the use of longitudinal methods particularly in the case of cohort or tracking studies in which the researchers have traced the development and behaviour of the subjects over extended periods of time. Clearly the benefit of such studies is that it provides insight and understanding of the process by which change occurs which is quite different from the large scale cross- sectional study of a population as a consequence of which one subdivides it into a set of major segments, e.g. psychographic, life style or benefit segmentation. In the latter case one can define distinct sub-groups in terms of their present behaviour but can say little about how this developed or about what would be required to change it. Similar weaknesses may also exist in the case of data collected by continuous methods, e.g. Neilsen Audit, Attwood, where the primary emphasis is upon what rather than why. In brief, cross-sectional studies involve the investigation of one or more variables as they are at a particular point in time, while longitudinal studies require the measurement of the same variables or factors on a number of successive occasions over a defined period of time. Because of the

394

Michael J. Baker

sequential nature of longitudinal research it is possible to avoid the greatest weakness of the snapshot approach of the cross-sectional study, namely that any explanation of the data collected is retrospective. As an SSRC Working Party reported (1975) The hazards incurred when any survey shifts from description of the contemporary situation of a population sample to an overt or implicit attempt to identify causal influences retrospectively are heightened by the time-span of recall and by the salience of the subject matter. Small wonder then that helpful respondents may well provide misleading explanations of why they behave in a particular way when faced with requests to recall past decisions of comparatively minor importance in their lives as a whole such as the difference between Brand A and Brand B. Inevitably, cross-sectional or ad hoc studies are unable to deal adequately with dynamic behavioural changes and properly designed longitudinal surveys are necessary to monitor such change. By properly designed is meant, research, which consists of measurement of the same objects /respondents at a series of different points in time as opposed to what often passes for a longitudinal study, which is a series of cross-sectional studies at specified points in time. While this may appear to be a counsel of perfection it is one which is felt to be justified, particularly where one is seeking to establish causality and when concerned with organisational as opposed to individual behaviour where the consequences of misunderstanding are likely to be larger and longer lived. A good example of the problems and importance of establishing causality is to be found in Gordon et als (1975) study of the adoption of innovations by hospitals. It was hypothesised that increased centralisation of authority in a hospital would result in a reduction in its willingness to innovate in the purchase of new equipment but in order to test this hypothesis it is clear that one must discriminate between two quite distinct possibilities. First, where authority is decentralised it may be that innovative individuals will purchase new equipment and introduce it to the system. On the other hand, an equally plausible causal priority is that where a hospital (as a system) has acquired new equipment then it will be more dependent upon individual physicians and their expertise and so involve them. A cross-sectional study to determine the relationship between centralisation and responsiveness to change would have failed to discriminate causal priority between the two variables. Similarly, the longitudinal study helps to reduce the possibility of making incorrect causal inferences, which, as Likert (1967) has demonstrated, may occur due to the time lag between certain classes of causal and end-result variables. Longitudinal research also helps overcome what Seiler (1967) has termed the single cause habit in which, based on a single data set from a cross- sectional study, one imputes all the observed effects to single causes. In reality behaviour is invariably the outcome of a time-related process and one should seek to monitor the causal chain which results in given outcomes and so avoid unduly simplistic interpretations of events recorded by the snapshot of a cross-sectional study. In summary, and as Kennedy (1982) has pointed out, (the) literature

Selecting a Research Methodology

395

would appear to advocate the use of longitudinal methods on the basis of the facts that this particular research methodology addresses the issue of causality; identifies time lags between cause and effect; permits analysis of dynamic processes; minimises the probability of hidden third-factor error and enables the researcher to monitor the actual occurrence of changes in dependent and independent variables. Longitudinal research would appear to afford the potential to eradicate problems of causality, produce more exact conceptualisation of process, develop better predictive models of growth and change and facilitate the identification of contextual constraints. Unfortunately, few academic research projects are sufficiently well resourced to permit the adoption of a longitudinal approach.

Summary
In this paper we have taken a broad look at the issues involved in selecting a research design, i.e. the method and approach by which we seek to secure the additional information necessary to solve our problem. From this review certain broad principles emerge. The first principle is that research is usually initiated because our current knowledge and experience appear insufficient or unsatisfactory to explain an issue of importance to us. Irrespective of whether this perception is prompted merely by intellectual curiosity or is the response to a practical problem, which needs to be resolved, the most usual approach is to follow the socalled scientific method. In essence the scientific method requires us to formulate hypotheses as to the causes and possible solution to our problem and develop these into testable propositions. The second principle is that in formulating hypotheses and developing these into testable propositions we would be well advised first to use observation to see if we can discover an acceptable explanation and then see if experimentation will help clarify the relationship between the phenomenon under scrutiny and the factors which appear to be associated with it. In the natural sciences where it is possible to control many of the variables, which may be influencing the interaction(s) in which we are particularly interested, experimentation will often provide an acceptable solution. However, in the social sciences, where the main focus of interest is usually human behaviour, it is rare that one is able to apply sufficient controls to make experimentation successful. Further, one is usually concerned to know why people behave in a particular way and inference from observed or experimental data is notoriously weak in doing this satisfactorily. Accordingly, having exhausted the possibilities of observation and experimentation most marketing researchers will wish to undertake some form of sample survey of the population in which they are interested. The third principle is that just as one proceeds from observation, to experimentation to sample survey so one should first undertake qualitative research before attempting to quantify the direction and extent of any hypothesised relationships. In doing so projective techniques will often be

396

Michael J. Baker

found to be especially useful in clarifying the underlying motivations, which influence actual behaviour. Finally, it is clear that one-off or cross-sectional studies may provide brief illumination of a problem it is much preferable if decision-makers and researchers keep track of issues of interest to them through longitudinal studies.

References and Recommendations for Further Reading

While many of the references given may appear dated they are original sources that underpin many of the current textbooks. Whether you wish to revert to the original sources or depend on current explanations of them is a matter of choice. Alreck, Pamela L. and Settle, Robert B. (1985) The Survey Research Handbook, Homewood, Ill.: Richard D. Irwin Baker, Michael J., (2000) Marketing Strategy and Management, 3rd. Edition, Basingstoke: Macmillan Blaikie, Norman, (2000) Designing Social Research, Cambridge: Polity Press Campbell, D. T. and Stanley, J. C. (1966) Experimental and Quasiexperimental Design for Research Chicago: Rand McNally Cohen, M. R. and Nagel, E. (1934) An Introduction to Logic and Scientific Method New York: Harcourt Cooper, P. and Branthwaite, A. (1977) Qualitative Technology: New Perspectives on Measurement and Meaning Through Qualitative Research Market Research Society Annual Conference, March, pp. 7992 Gordon, Wendy and Langmaid, Roy (1988) Qualitative Market Research: A Practitioners and Buyers Guide, Aldershot: Gower Gordon,G.,. Kimberley H.T and MacEachran, A. (1975) Some considerations in the design of effective research programs on the diffusion of medical technology, in W.J. Abernathy, A. Sheldon and C.K. Prahalad (eds.) The Management of Health Care, Cambridge; Ballinger. Kennedy, A.M. (1982) Longitudinal Research Methods: Applicability in Industrial Marketing, Working Paper, Department of Marketing, University of Strathclyde Levitt, Theodore, (1960), Marketing Myopia, Harvard Business Review Likert, R. (1967) The Human Organisation, McGraw-Hill, New York Mayer, Charles S.(1965) Marketing Research Sec 24 and Statistical and Mathematical Tools Sec 25 in Marketing Handbook, 2nd edn, ed Albert Wesley Frey, New York:Ronald Press Moser, C.A. and Kalton, G. (1971) Survey Methods in Social Investigation, 2nd Edn London: Heinemann Popper, K. R. (1968) The Logic of Scientific Discovery, rev edn. London: Hutchinson Seiler, R. E. (1965) Improving the Effectiveness of Research and

Selecting a Research Methodology

397

Development New York: McGraw-Hill Seltiz, C., Jahoda, M. Deutsch, M. and Cook, S.W. (1959) Research Methods in Social Relations, 2nd edn., London: Methuen Tull, D. S. and Albaum, G. S. (1973) Survey Research (Aylesbury: International Textbook Co. Ltd) Wancevich, J.M. and M.T. Matteson (1978) Longitudinal Organisational Research in Field Settings, Journal of Business pp. 181-291. Webb, John (2000), Questionnaires and their Design, The Marketing Review, 1, 197-218

About the Author

Michael J Baker is Emeritus Professor of Marketing at Strathclyde University. This paper was written while Distinguished Visiting Professor at Monash University, Australia. Currently, he is President of the Academy of Marketing and Dean of the Senate of the Chartered Institute of Marketing and holds Visiting Appointments at Glasgow Caledonian University, Nottingham University and Surrey University. He is also an Honorary Professor at the University of Wales, Aberystwyth.