Business research methods alan bryman emma bell 3rd edition

Locke, K. Remenyi, D. Strauss, A. Corbin Basics of qualitative research: Techniques and procedures for developing grounded theory Thousand Oaks, CA. Discussions of ethics tend to sound worthy, sometimes border on the philosophical, and occasionally stray right off the point. Why should this be? Ethics relate to moral choices affecting decisions and standards and behaviour. So it is quite hard to lay down a set of clear rules, which cover all possible moral choices. Especially in research, where the practical aspects of a study e.

Sometimes it can be quite a shock, when you have been used to getting pretty clear ideas about how to do something, to find you have to make your own decisions about how things will be done. Ethical choices we have never imagined can just creep up and hit us. An obvious example would be when, as a very honest student, we start to collect some data together and realize that one source of data is completely out of step with the rest.

As a professional researcher, that is an interesting challenge, which will create its own new pattern of research and investigation. But as a business student with a fast approaching hand-in deadline, the temptation to lose the odd piece of data can be great. We are not suggesting that we have to be great moral advocates here, perhaps that is a matter for our own consciences, but we must anticipate as much as we can the moral choices and dilemmas, which the practice of research will bring, and try to find appropriate ethical ways of dealing with them.

And how the data you collect will be used? And whose data is it, if they spoke or wrote it? Reality is messy — do we want to smooth the mess and create simple answers, or do we want to understand messy reality in order to change or anticipate it? Can data be recreated from your notes? Do we pretend it worked? What if an interviewee starts to see things in a new light and uncovers painful memories or ideas? Latter can also happen in focus groups- conflict, personal animosity could develop — how can this be handled?

What effect does this have on your data? Does it affect validity of results? Whose is it? And how exactly do you transcribe? Do you include repeated phrases or words? Do you attempt to record body language which may affect the meaning of what is said? Remember that provided the process was justified and conducted ethically and professionally, then a not very exciting outcome does not really matter.

We cannot all discover gravity or relativity, but we can all design sound research plans and carry them out professionally. How could you get ethical approval for this? However just how untrue may be surprising. Visit www. There is also some useful discussion of ethical research issues in an article by Jane Richardson and Barry Godfrey which focuses on ownership and authority to use interview transcripts which may be in the public domain.

See references below for details. Can you identify any more stakeholders? Some will be specific to the kind of research study undertaken, for example a study of recruitment practices could affect potential employees. Once you know who might be affected by your research study, you could design a simple risk analysis — for each stakeholder identify the type of risk from your research, its potential impact low, medium or high and the probability that it will happen unlikely, possible, probable.

Entering this into a grid, will give you a clear idea of priorities in designing an ethical study, and should lead you to think about strategies to reduce undesirable impacts. So what does participant anonymity involve? It is not usually just a case of not putting their names in the final report. It will be important to decide whether you need to devise a code for each participant so you know who they are but they cannot be named by others , or whether this is not needed by the study so no-one will have a code or a name.

Can you refer to their title, role, function, department, site etc? All these, in conjunction with your results, may reveal identity. Can you stop yourself referring to someone, in your study, to others in their company, who might try to identify them? If you have, for good reason, collected personal details, have you checked whether you comply with the requirements of any data protection legislation in your country? Does it really affect the research outcomes and thus will be important data to collect?

Or could you redesign your study so that this kind of data was not important and need not be collected? Informed consent requires you to prepare for all research participants some documentation which shows them what you are doing and why, what their role in the research is, what will happen to the data you collect from them and what they are agreeing to do.

It will also usually set out how you will keep and dispose of the data and how the required confidentiality will be ensured. This is very detailed and seems like a lot of work, but in fact a short text can often achieve all the requirements of informed consent. This, or a brief statement referring to this documentation, must then be signed by your participants. Remember that no undue pressure should be brought to bear on any participant or gatekeeper, since this, however well-intentioned, will influence their involvement in your research and will prove not only unethical, but may also invalidate results.

The first issue is the way data are collected and recorded. You may be using a specially designed relational database in which to record observations and related information, or we may be talking about a highlighter pen and notes in the margin of an interview transcript, or a clipboard and pencil. Whatever method is used to collect, and transfer data to a retrievable record, then it must be designed for purpose, systematic and capable of capturing all relevant details.

Take for example a semi-structured interview method: what kind of system could be used to record the interview? Video recorder? Digital recorder? Tape recorder? Notepad and pen? Pro-forma with main questions and spaces to record answers? Reflect for a moment on what kind of issues could arise which might affect research objectivity depending on choice of system. If so, what would you do? How could you ensure continuing objectivity?

The second issue is when a research study is under way and something unexpected happens to cause a problem with your data. Or a failed tape recording. Or a key participant withdrawing from the study, as they have a right to do. At this stage of the research, however honest we are, there will be a temptation to fix the problem.

So we should anticipate this temptation and understand, before it happens, that that is the road to failure in research.

Academic and professional audiences will not be fooled, because they will understand and look for such issues. The moral responsibility of the researcher is considerable and when researchers are found to have transgressed, they are likely to be held to account in the media.

Sadly, there are many examples to be found, but at least these will have been held to public account. If you are researching an organization of which you are part, then you already have an understood role or status within this organization. However, an internal researcher may be in a position to conduct a kind of research, which may be impossible from an external perspective.

Can you think of an example in business research? Could you possibly find more useful and reliable data covertly than openly declaring your intention and gaining official agreement for access? In a few cases, the answer may be yes, but if so, there must be approval from any research ethics committee relating to your studies or research or professional body ethics approval e. Assurances must then be given about the use to which the research data will be put and to what extent it will be anonymised.

Spying is not research! When should we think about ethics in a research study? What elements would you include in a consent form for interview based research? In what circumstances might covert research be justified? How would you deal ethically with this? What practical activities can you suggest to anticipate and prevent unethical research practice?

Godfrey To find out things about people we need to ask research them. So we ask some of them. We sample the population. Problem 2: we wanted to find out things about people, so we researched a sample of them. To what extent do our results relate to all people, and to what extent do they only relate to our sample? Problems 1 and 2 put sampling in a nutshell. Sampling is a practical way of studying people and their activities, thoughts, attitudes, abilities, relationships etc in relation to business.

That would mean that our findings can be generalised to the whole group. To make this happen, we have to learn about a number of issues and technical words and phrases in sampling. In the next section there is a brief glossary based on Box 4. To learn more about each technique, read the textbook and web search further, or ask questions about these techniques in livechat. This sounds underhand but is often used, at least in pilot studies or short term projects where there is insufficient time to construct a probability sample.

Therefore, where this is used, the results cannot be generalised to the population though many newspapers would like you to believe otherwise! Generalisability: being able to use sample results as if they applied to the whole population — this must be based on sound sampling processes Multi-stage cluster sampling: When drawing a sample from a geographically dispersed population, the logistics suggest that cluster sampling can help.

The sampling frame is first broken into clusters eg geographic areas , and a random or systematic sample taken. Then the population of each cluster is sampled randomly to provide random sampling which is logistically feasible.

This can of course introduce bias, but using both cluster and systematic sampling can usually produce effective samples. Non-probability sample: Random selection was not used so some units in the population may have had a higher chance of being selected e. Probability samples keep sampling error low and usually offer a sample which can be seen to be representative Quota sampling: Regularly used in market research and opinion polling. Like a stratified sample, this sample is chosen to include a certain proportion of particular variables e.

There is no sampling frame here, so it is not random, but sometimes it is difficult to pre-define the population eg staff in a company who contribute creative ideas. This technique is often used in qualitative approaches. Purposive sampling: Using your own judgement to select a sample. Often used with very small samples and populations within qualitative research, particularly case studies or grounded theory.

This approach cannot yield any statistical inferences about the population. Cases may be selected for being unusual or special or particularly related to your research question. Stratified sampling specificies any characteristics, which you wish to be equally distributed amongst the sample, eg gender or work department. Provided the sampling frame can be easily identified by these characteristics, then strata for each characteristic are identified and within each group, random sampling or systematic sampling can proceed.

Sample is chosen directly from the sampling frame which ideally should not be in any specific order except alphabetical.

Once you know the sample proportion required eg 1 in 20, start with a random number generated item in the list, then choose every 20th name until the sample is complete. Random sampling: also called probability sampling — see explanation above. Define the population. Define the sampling frame F this may be the same or it may exclude certain groups or individuals as not relevant to the study.

Decide the sample size Z. Using a table or computer programme to generate random numbers, collect Z amount of different random numbers within the range 1-N. Apply the chosen random numbers to the sampling frame to identify your random sample. Used in random sampling. Use whatever digits in the random numbers apply within your sampling frame total and ignore duplicates.

You may find it is simpler to use Excel spreadsheet function to generate random numbers. Sampling fraction: Number required for sample divided by number in total sampling frame expressed as a fraction or percentage.

These techniques offer varying levels of generalisability but always less than a random sampling method. Think about these three techniques and decide how justified you think each is for conducting business research. In the definitions of random sampling above, we have ignored this question so it is now time to tackle it.

Unfortunately there is no right answer to sample size. You cannot just apply a consistent proportion to the total sample frame. Instead the following issues need consideration: www.

If the population total is ,, then your sample size is 10, — yes this would probably be a good sample size but see the next problem on this list. We can see that this unit or person could be quite unrepresentative of the total population by itself. So relative sample size is not important. Absolute size is.

The bigger the sample size, the more the sample is likely to represent the population and the lower is likely to be the sampling error. Referred to as the Law of Large Numbers. If you have not done any work on statistics before, do some quick web-searching or look at the index of the textbook to find out. If you wish to conduct a statistical analysis on your data, the minimum size of sample for any one category of data should be 30, as this is most likely to offer a reasonable chance of normal distribution.

If your sample frame is 30 or less, then it would be wise to include the whole frame, rather than sampling.

Of course, the population you are researching may be way below in total, and it may in any case be very costly or time-consuming to use a large sample size. Practical considerations are important in research studies. Just bear in mind that if you choose a sample size which is small in absolute terms, then you must justify this action and take into account the fall in generalisability and representativeness which may result.

Inevitably your respondents are less likely to be as motivated as you, the researcher, about your research, so some — and sometimes a majority — will not respond, ie refuse to take part. All this is taken into consideration when a choosing your sample size and b calculating the actual response rate. So if all questionnaire respondents are chosen from one company or organisation, the best to hope for is that our results can be generalised to the whole workforce of that company or organisation.

We cannot assume that these results will in fact describe other workforces, as very different conditions and variables may apply in other organisations. However, we cannot then apply these conclusions to other countries without further research, nor can we apply these conclusions over time to the same country, as major variables could have changed over time. Think back here to what we discussed earlier about epistemology — what we can really know.

For practical time and cost reasons, media production teams often take quota sampling research or research done by more dubious methods and suggest its applicability to everyone watching or listening to a programme.

Look out for examples and try to find out what kind of sampling was applied to their research. If you are worried about the representativeness of your sample, in some cases it may be possible to check this by using a test of statistical significant difference to compare the profile of characteristics in your sample with that of another data list eg a census or company database. Clearly if there is no statistically significant difference between your sample and the full population data list, you have added more authority to the representativeness of your sample.

If you are using a non-probability sampling technique then even the flimsy size rules associated with probability sampling fall away. Your sample size for purposive or snowball sampling will really depend on your research questions and objectives. In qualitative research, the focus will not be on trying to estimate things about a population, but in trying to understand or relate the data to theory or ideas.

How many people do you need to talk to, to understand their perception of something for example? It could be just one.

Or it could be several or many. The question is here, what are you trying to find out and what sample size would give me confidence that my results had validity? We will go further into this when we discuss different qualitative methods, but often a good lead can be taken from research studies in peer-reviewed academic journals, where information has been given about sample size in relation to research question.

Find one that is close to your area of study which you would want to do anyway in your literature review and check the sample size studied in this type of enquiry. Why are random numbers useful for sampling? How do you calculate a response rate? What kind of minimum size would you need in a sample used for statistical inference? What level of certainty is needed for statistical sampling in academic research?

What reasons would you give for not exceeding a sample size of ? Quant it at ive research m et hods: collect ing and analysing quant it at ive dat a Suggested reading: Research Methods for Business Students, Saunders, M, Lewis, P et al.

Suppose for example that you want to know the three most useful management textbooks that a large group of managers have found effective. The question might look like this: Q1 What are the three management books which have been most useful to you so far in your management career?

You might leave three lines of space so that the respondents can write in their answers. Think about how this might be coded as a question response for analysis. Since most managers will not choose the same three, you will have a wide range of different answers. We cannot code each book separately with a sample size of and potential books in the answer range. So can we make any useful data out of this question? However it is a form of question which is quite common eg what five competencies are needed by successful salespeople?

What three benefits do you feel you have gained from mentoring? Etc etc. It is possible to turn the question into a list of possible answers from which respondents have to tick three which apply to them. This means you can give each possible answer a unique code in advance and then count the frequency with which each code is used.

If we take the first 50 responses and make notes on the characteristics, which define the responses, it becomes possible to group the responses. Once grouped, a code can be assigned to each group and you can then go back and code each answer according to this pre-defined group.

We now have 5 unique codes and can go through all the responses collecting numerical data for each code. Now we have a data set for analysis. Think about analysing this data. A set of data results is going to look pretty boring, and how much is it going to tell you about your research question?

In the next chapter we will investigate questionnaires further, but for now, we need to think about the data which will result from our questions, how useful it might be, and how we might analyse it. Always aiming for higher ground.

Different variables will require different kinds of analysis, so it is important to identify what you are asking for in your research. There is a fixed space interval between each variable and this is a consistent space. We could also include answers involving age, income, number of staff, revenue etc. There is an even more precise form of this variable which is sometimes called a ratio variable.

This last category changes the entire set into ordinal rather than interval variables, and this will constrain what can be done with the data, although it is still useful. So why put such potential interval data into groups in a survey? There are good reasons. It is usually best to treat these variables as a special kind of nominal variable. This can be done in Excel or another spreadsheet first, or put directly into a statistical package such as SPSS for Windows.

To make the transition from, say, questionnaire to data matrix, answers will need coding. For example, nominal variables will be text names and will need to be given a unique number to allow entry into a statistical package. This is vital for two reasons. The first is that codes are often worked out on scraps of paper quite quickly; if the paper is lost and you have a break between entering your data and coming to make sense of it, it is possible you will have a hard time remembering exactly what the results mean.

The second is that it is important not to lose sight of the question when analysing the results of quantitative data. Unusual patterns in the data must be scrutinised and going back to exact coding and possible different interpretations of the question wording, which may have caused the response, will be vital. So keep a retrievable, clear and accurate record of coding as the link between respondent and data. SPSS for Windows is the most commonly used tool to produce all statistical tests and analysis outlined in the sections below.

Using the package is very straightforward, provided you have access to it on a computer. Screen tabs allow you to switch between these two views. Data View is the screen through which you enter your data like a spreadsheet.

You must enter your data so that each column represents a variable, and each row represents a case. For example, if you have information on the age, salary and qualifications of employees, you enter the variable data for each employee along a row, with column headings of age, salary, qualifications.

To describe your variables, you go to Variable View. Text variable names can be a maximum of 8 characters with no spaces. This means it is helpful to make a rough plan of how you will enter data into SPSS — in which order you will show the variables and what variable names you will use. It is also possible to enter labels for Values all except interval values , so for example you may have a variable labelled Gender, which has values labelled Male and Female, though you have coded Male as 1 and Female as 2 in the Data view.

When you perform an analysis with SPSS by clicking Analyse and entering any relevant information about what you want done it is held as Output in an Output viewer screen which only appears after an analysis has been done.

This is simple to do and researchers do this from time to time, but it does impose constraints on how statistical inferences can be drawn, since cases in the lower response stratum are treated as if there were more of them than there are ie higher weighting in the dataset. Best avoided if possible unless you are really confident in statistics. The textbook recommends a useful summary of ways of looking at data on p We usually begin by attempting to describe particular values, their range, their central tendency, their dispersion around the mean.

We can look at the data trends over time, and look for proportions in the data. This is called univariate analysis because we are looking usually at one variable at a time. Once we have a clear picture of how the individual variables are behaving, we can start looking for relationships between variables — bivariate analysis.

A range of methods is shown below for these two kinds of analysis. Tables show a list of categories types of response and the numbers of people responding to each.

Sometimes just as a number, sometimes a percentage of the total choosing this response. When building a frequency table for interval variables, categories will usually be grouped if not the table would probably be too long. Make sure your groups of categories are exclusive eg for ages , 40 etc not , as this leads to difficulties of coding for age If using an interval variable, then a histogram would be used rather than a bar chart.

Note that pie charts should not show more than six segments — more than this will be very difficult to read, so either use a bar chart, or group the data before producing the pie chart. The measure is a single figure so is not representable in a chart, however, a series of means, medians etc can be charted or shown in a table. Mean is calculated only for interval variables. Median is calculated for interval or ordinal variables.

Mode can be calculated for any variable. The standard deviation is the average amount of variation around the mean calculated by taking the difference between each value and the mean, totalling these differences and dividing the total by the number of values. A higher standard deviation therefore means greater variation around the mean. The box plot shows where the median of the data lies and how the data clusters around that median or middle value.

This will also affect your later statistical analysis. When you are putting the last minute touches to a report before a deadline at study and at work it is easy to imagine that everyone will know what this graph shows. This leads to a big problem if we leave it at that. You must check each graph to ensure it has a clear title, the units of measurement involved are shown, any data source is shown, the sample size is shown where relevant, the axes are labelled, the variables read in a comparable way if more than one chart uses the same axes and variables eg left to right or top to bottom and there is a key or legend which is readable importing from Excel often leads to very tiny illegible legends — they must be reformatted.

It can also be helpful to introduce a chart in the text with an idea for the reader of what it will show, then after the chart in the text, explain what you think it showed. Of course, readers will want to make up their own minds, but it is helpful to let them know what you think they should look for in the chart.

This is always a good first step. Then if you wish to look at a trend over time for a single variable, the most common method is the use of index numbers — such as the FTSE index of share movements over time based in London. The base period is usually represented by the number or as in FTSE. Then each value is converted to an index number by dividing the data value for the case by the data value for the base period and multiplying by Why bother converting each value to an index number?

Generally because it makes comparison across time or numbers much simpler — can be done at a glance. Try to find an example from the web or media of a trend using index numbers. Suppose we want to take the trend further and estimate where it will go after the actual data we have to hand?

Here we are into forecasting and we will be covering this in our last but one chapter. Though this can sometimes seem obvious — eg if the two variables include something like age or gender which can influence the other variable but not be influenced by other variables.

Presumably the amount you eat could be influenced by your age, but your age could not be influenced by the amount you eat! If one variable is suspected of being the independent variable, this is shown as a column variable not a row variable.

Such tables are used to look for patterns of association in the data. This produces a percentage, which describes the proportion of variation in one dependent variable accounted for by the other independent variable. A similar analysis where more than one independent variables are involved is called multiple regression analysis. It does not assume a linear relationship.

Such a test can also estimate the chances of no relationship in fact existing between two variables, when bivariate analysis suggests that there is. Then decide the level of statistical significance we find acceptable, ie the level of risk that we would reject the null hypothesis ie say the variables are related when in fact they were not related.

It is usual to say that the maximum level of 0. We can choose a more stringent level of certainty e. We should bear in mind that the likelihood of a statistically significant result will increase with sample size — for the obvious reason that the bigger the sample in relation to the population, the less likely that any analysis on the sample will differ from the population by chance.

So if we think there is likely to be low statistical significance, we should increase sample size if possible, to make the analysis more sensitive to statistical significance. Very small samples, below 30, are more likely to show an unacceptable p level ie above 0. This test looks at each cell in a contingency table and calculates the expected value if there was no relationship but the value was a product of chance, works out the difference between each expected value and the given value and sums the differences.

This produces a single chi square value for the table, which is not important in itself, but is produced with a statistical significance level p. This is the number we are looking for, to check against our desired level of certainty. This helps us to be sure that the correlation we expect from the sample, really does exist in the population. The test produces a D statistic, which is used to calculate whether the sample distribution differs from the full population distribution by chance only.

The lower the t statistic, the more likelihood of any difference in the groups being caused by chance. Similarly a paired t-test can be used to measure pairs of variables, e. A high F statistic and a significance p level of below 0.

There are some data requirements for ANOVA, but broadly this can be used provided there are at least 30 values in each group and each value is independent of others. This chapter has been very factual and is not easy to take in, unless you are already familiar with statistical analysis and find it easy to follow.

It is intended just to give some revision pointers based on earlier reading or teaching you may have experienced. Then what kind of values will be produced if people respond to this question? Why is it important to think through the data likely to be produced from your research at an early stage? Why do you need to know the difference between interval, ordinal, nominal and dichotomous variables? What is bivariate analysis? What is the minimum number of cases you need to make a sample useful for statistical analysis?

What is the purpose of using index numbers and an example from the web or media? Surely it is quite straightforward to write them? In fact, designing questionnaires is particularly difficult. What do we need to think about? The format and design of the questionnaire — not too offputting, not too long, not too difficult to read, easy to know what you have to do to complete it 2.

How much general information do you need to have about the respondent? If you need biographical data such as age and gender etc, why is that? What extra value will it add to your research question? Should you start with easy questions like gender or end with them? What proportion of open and closed questions should be in the questionnaire? Closed questions start with a verb Do you come here often? They are easy to code but limiting in detail.

Open questions give much richer information but are widely variable across responses and therefore harder to code and analyse. What kind of questions can we ask? Straight questions with a clear answer? Questions about which people must reflect? Tickbox questions or written answer questions? Likert scale questions?

Such scales may or may not have middle points which allow a neutral response. How do we lay out the questions? For example if they are scale responses — do we lay them out horizontally or vertically?

How much space on the page do you give someone to write an answer to an open question? Should you include check questions, for example asking the same thing two different ways to ensure you are getting consistent answers?

How much information do you give the respondent about why you are asking the questions? Technical research detail? Just enough to know who you are and how data will be used? How do they get it back to you? If email — what does that do to anonymity? Should you include stamped addressed envelopes, drop boxes? Should you communicate with potential respondents before the survey itself? And after delivery to encourage completion? How many times could you prompt for a reply?

Should you use post, fax, email or online surveys? What happens if the response rate is too small to be useful? Now in its fourth edition, the text featuresnew chapters on the nature of business research and on sampling in qualitative research. Oxford University Press is a department of the University of Oxford. It furthers the University's objective of excellence in research, scholarship, and education by publishing worldwide.

Search Start Search. Go directly to our online catalogue. Buy Print Edition. Features About the Author s Features Each chapter is filled with examples of 'real world' research, placing the theories and concepts being discussed in context. A wide range of examples from various business functions - including human resource management, marketing, and strategy - demonstrate the relevance of research to business and management students Coverage of all aspects of business research ensures that students are fully prepared for conducting their own research projects A 'telling it like it is' feature draws on interviews with real research students and their supervisors to give invaluable advice on potential pitfalls to avoid and successful strategies to emulate when undertaking a research project 'Key concepts' are drawn out throughout the text and explained with unrivalled clarity, ensuring that students gain a full understanding of business research terminology and processes.

Table of Contents Part One 1. The Nature and Process of Business Research 2. Business Research Strategies 3. Research Designs 4. Planning a ResearchProject and Formulating Research questions 5. Getting Started: Reviewing the Literature 6. More Like This. More Copies In Prospector. Loading Prospector Copies Table of Contents. Loading Table Of Contents Loading Excerpt LC Subjects. Business -- Research -- Methodology. More Details. Includes bibliographical references pages [] and indexes.

Similar Series From NoveList. Similar Titles From NoveList. Similar Authors From NoveList. Librarian Reviews.