Professional Documents
Culture Documents
C 5 T R P: E O R D: Instructional Goals
C 5 T R P: E O R D: Instructional Goals
Discussion Questions
1. What are the basic research design issues? Describe them in some detail.
Basic research design issues are primarily a function of the purpose of the study
(whether it is exploratory, descriptive, or hypothesis‐testing), and relate to such
aspects as the type of study to be done (causal or correlational), the setting in which
it will be done (natural or contrived), how much researcher control will have to be
exercised (very little in the case of field studies, to very much in the case of
experimental designs), how many times data will have to be collected (one shot
versus longitudinal), and the unit of analysis – i.e. the level at which data will be
aggregated. For most correlational studies, the field setting with minimal researcher
influence will be the choice. Most field studies are generally crosssectional, though
some could be longitudinal. Longitudinal studies, though better for understanding the
dynamics of the situation fully, also consume more time and resources. Thus, the
costs of a study also determine some of the design choices. The unit of analysis
depends on whether the research question focuses on individuals, dyads, groups, or
entire systems.
Answer
This would be a causal, hypothesis‐testing study. It will be a field experiment,
using one or two restaurants to manipulate incentive plans and see if this causes an
increase in profit margins in those restaurants. Because of the manipulation, there will
be some researcher interference with the natural flow of events. The time horizon for
the study would be longitudinal since data will be obtained on sales, profits, etc., both
before and after the manipulation. The unit of analysis would be the restaurant.
Scenario C
A manager is intrigued why some people seem to derive joy from work and get
energized by it, while others find it troublesome and frustrating.
This would be a correlational, hypothesis‐testing, field study with minimal
researcher interference. It will be a one‐shot study and the unit of analysis will be
individuals.
Exercises
Doing the exercises in the chapter will help the student to apply the basic
elements of research design to different scenarios. More in‐class exercises are offered at
the end of this chapter (with answers), that can be used either in the classroom or as
exam questions.
Answers to Exercises in the Book
© 2009 John Wiley & Sons Ltd.
www.wileyeurope.com/college/sekaran 48
Exercise 5.1
A foreman thinks that the low efficiency of the machine tool operators is directly
linked to the high level of fumes emitted in the workshop. He would like to prove this to
his supervisor through a research study.
1. Would this be a causal or a correlational study? Why?
2. Is this an exploratory, descriptive, or hypothesis‐testing (analytical or predictive) study?
Why?
3. What kind of a study would this be: field study, lab experiment, or field experiment?
Why?
4. What would be the unit of analysis? Why?
5. Would this be a cross‐sectional or a longitudinal study? Why?
1. This would be a causal study because the operator wants to prove to the supervisor that
the fumes are causing operators to be low in their efficiency. In other words, the
machine tool operator is trying to establish the fact that fumes cause low efficiency in
workers.
2. This is an analytical study because the machine tool operator wants to establish that
fumes cause low efficiency and convince his workshop supervisor through such analysis
(i.e. establish cause and effect relationship).
3. This would be a field experiment. Though the study would be set up in the natural
environment of the workers where the work is normally done, the amount of fumes will
have to be manipulated while other factors, such as atmospheric pressure, may have to
be controlled. Because of the location of the study, it will be a field experiment.
4. The unit of analysis would be the individual operators. The data will be collected with
respect to each operator and then the conclusion will be made as to whether the
operators are less efficient because of the fumes emitted in the workshop.
5. This would be a longitudinal study because data will be gathered at more than one
point in time. First, the efficiency of the operators would be assessed at a given rate of
fume emission. Then the fumes emitted would be manipulated to varying degrees, and
at each manipulation the efficiency of the workers would again be assessed to confirm
that the high rate of fume emission causes a drop in operators’ efficiency.
Exercise 5.2
Many were concerned about the operations of the infamous BCCI, the
international banking institution. If the Government had desired to probe into the
details, would this investigation have called for:
1. A causal or correlational study? Why?
2. An exploratory, descriptive, or hypothesis‐testing study or case analysis? Why?
© 2009 John Wiley & Sons Ltd.
www.wileyeurope.com/college/sekaran 49
3. A field study, lab experiment, or field experiment? Why?
4. A cross‐sectional or longitudinal study? Why?
1. This would have been a correlational study since the details desired are to be probed
into without trying to establish cause and effect relationships.
2. It would be a descriptive study describing the operations of BCCI.
3. This would be a field study since data will be collected in the natural environment.
4. This will be a cross‐sectional study since the data will be collected at only one point in
time.
Exercise 5.3
You want to examine how exposure to thin or heavy models in advertisements influences
a person’s self‐esteem. You believe that the effect of exposure to models in
advertisements depends on the extremity of the model's thinness or heaviness.
Discuss the design decisions that you as a researcher will make to investigate this issue,
giving reasons for your choices.
An experimental study (causal study) will be needed since the goal of the study is to test
a cause‐and‐effect relationship.
A pretest will help you to select advertisements containing female models for the
following four conditions: moderately thin, extremely thin, moderately heavy, and
extremely heavy;
- participants view an advertisement booklet containing ads with female models.
- participants rate each model in terms of size (‐5 =extremely overweight, +5 =
extremely thin) and attractiveness (‐5 = extremely unattractive, + 5 = extremely
attractive).
Based on these scores, you will be able to select advertising models in each condition.
Note that models should differ from each other in terms of size but not in terms of
attractiveness (you have to perform post hoc comparisons to test for this).
Exercise 5.4
You want to investigate the specific effects of specific emotions on customers'
behavioral responses to failed service encounters across industries.
Discuss the design decisions that you as a researcher will make to investigate this
issue, giving reasons for your choices.
Exercise 5.5
War on Cancer
Dr. Larry Norton of Memorial Sloan‐Kettering Cancer Center predicts that cancer
treatment will undergo major changes. Several drugs are being developed to battle
cancer without harming healthy tissues. It is a question of discovering which of
these drugs does the job best.
Design a study that would help find which drug would do the trick.
An experimental study (causal study) of the Solomon Four‐Group design will be
needed. A count of the healthy tissues in cancer patients at more or less the same stage
of cancer progress will first be taken, with about 20 patients in each group. Each group,
but the control group, will be treated with the different new drugs. The cancer cell
count will be taken both before and after treatment for a period of time (say, 3 to 6
months) to see which has the most beneficial effect. Thus it would be a longitudinal
study with some manipulation (drug variation) and control (matched or randomized
groups). The unit of analysis will be groups.
h & i: Unit of analysis will be individuals (stock brokers) since the data analysis will
be done on each individual’s responses without any aggregation across groups.
Discussion Questions
1. Describe different data sources, explaining their usefulness and disadvantages.
Secondary sources of data constitute all available sources from which existing
information can be extracted. This could be published or unpublished information in the
Library, in Government or Company records, in data tapes, on the internet, etc. Primary
sources of data can be any source which provides information elicited directly for the
first time. This could be organizations, individuals, trash cans, or whatever source from
which the desired information can be obtained.
Secondary data sources are useful because they offer a base on which to build
further research without having to reinvent the wheel. Past findings can spur new
research; past data can also offer opportunities to analyze them from a different
perspective or angle. Under certain circumstances, however, secondary data might be
misleading if one does not take the changed circumstances into consideration. For
example, if old archives of national data are utilized to examine and draw conclusions
about gender differences at the workplace, this might become an exercise in futility. The
workplace has changed much now compared to 20 years ago – there are more educated
Exercises
Exercise 8.1
A production manager wants to assess the reactions of the blue‐collar workers
in his department (including foremen) to the introduction of computerintegrated
manufacturing (CIM) systems. He is particularly interested to know how they would
perceive the effects of CIM on:
a. their future jobs
b. additional training that they will have to receive
c. future job advancement.
Design a questionnaire for the production manager.
Questionnaire
Cria Software Enterprises
4571 Southwood Ave
Cupertino, California 94024
□ Surveyor
□ Other
Exercise 8.2
Answers may vary.
Exercise 8.3
Answers may vary.
Exercise 8.4
Answers may vary. The questionnaire should include valid and reliable scales measuring service quality,
customer satisfaction, and customer loyalty.
Exercise 8.5
Design an interview schedule to assess the Intellectual capital, as perceived by
employees, in an organization after identifying the dimensions and elements.
The dimensions of intellectual capital would primarily comprise, among other
things, the following:
The knowledge‐base of the employees; Their
know‐how
The patents owned by the organization;
The secret formulas utilized by the company (special formulas / manufacturing process,
and the like).
Answers
a. Purposive Judgment Sampling.
b. A Systematic sampling design (using university listing of students).
c. A Stratified Random sampling with stratification along profession, gender, age, etc.
d. Simple random sampling because of the importance attached to high generalizability.
Problem
1(a).You want to estimate the production days that would be lost during the next
three months by sampling the vacation intentions of a few employees. You randomly
select 36 employees in the organization and find that the average number of days they
intend taking off is 16 during the coming three Summer months, with a standard
deviation of seven (7) days. Based on these sample statistics, you want to estimate at a
99 percent confidence level, the days that will be lost due to the entire population of
workers taking vacation time during the next three months, so that the plant manager
knows how much temporary help he should plan on hiring during the summer months in
order for work to proceed smoothly.
Calculation
μ X zSx
163.01
12.99to19.01
If there are 100 employees in the organization expected to take vacation during
Summer, then the most optimistic estimation of the days lost through vacation time
during the summer would be (13 100 =) 1,300 days and the most pessimistic would be
(19 100 =) 1,900 days. This would mean that temporary help would be needed
anywhere between 1,300 and 1,900 days worth of labor, for production to proceed
smoothly.
1(b). Now you might want to narrow the gap in the estimation since 1,300 and
1,900 days is a wide range, and you would like more precision using the same
information that is available to you. Well, it can be done only at the cost of taking more
chances that your estimation, though more precise, may not be stated as confidently as
before. You may, however, decide to lower your confidence level to gain more precision
in your estimation. At the 95 percent confidence level, the most pessimistic and the
most optimistic estimations would range from 1372 to 1828, a narrower range – we are
now gaining more precision, i.e., narrowing confidence interval – but only at the risk of
being wrong five percent (as opposed to one percent) of the time. We could seek a
further narrower interval range at the 90 percent confidence level (1408 to 1792), if
taking a higher risk in being wrong in the estimation is not going to be disastrous. You as
a manager will have to decide on the amount of risk you want to take by increasing or
decreasing the confidence level.
Problem
How large a sample do you need if you want to estimate the mean age of a
population of 100 residents and be within 3 years of the true mean with a 95 percent
level of confidence, when you know that the sample mean is 28 years and the standard
deviation is 17 years?
Sx s/ n
1.53 17/ n
1.53 n 17 n
17/1.53
n 123
Discussion Questions
1. Identify the relevant population for the following research foci, and suggest the
appropriate sampling design to investigate the issues, explaining why they are
appropriate. Wherever necessary, identify the population frame as well.
a). A company wants to investigate the initial reactions of heavy soft‐
drink users to a new “all natural” soft‐ drink.
b). A gun manufacturing firm would like to know the types of guns possessed by various age
groups in Washington, D. C.
When guns are sold, the dealers are expected to maintain records of the names and
ages of the buyers, and the types of guns sold. Thus, the population frame for gun
owners can be had from the records of gun dealers in the Washington, D.C. area. Once a
complete list is compiled, a stratified random sampling of appropriate age groups can be
done, and the relevant information obtained.
c). A hospital administrator wants to examine if single parents working in the hospital have
a higher rate of absenteeism than parents who are not single.
The marital status and parental status of the employees may be available from the
records of the hospital. If so, the population frame would be the hospital records from
which the information can be directly extracted. A stratified random sampling design can
be used and a t‐test done to examine if there are differences in absenteeism in the two
groups. If the marital or parental status of the employees is not available, then a
convenience sample of those who are known to be single parents and those who are
known to be not single parents can be used to check their records for absenteeism.
10. Why do you think the sampling design should feature in a research proposal?
Details of the sampling design, giving details of how many and what cadres of
personnel will be included in the survey, helps the sponsor to see if such people would be
available and could respond to the survey. It also helps determine the extent to which the
findings will be generalizable to the entire organization.
In published studies, it also helps the reader become aware of the settings to which
the findings can be generalized.
Exercises
For the situations below, what would be the relevant population and the most
appropriate sampling design? Make sure you discuss the reasons for your answers.
Exercise 10.1
The medical inspector desires to estimate the overall average monthly
occupancy rates of the cancer wards in 80 different hospitals which are evenly located
in the Northwestern, Southeastern, Central, and Southern suburbs of New York City.
All the 80 hospitals in the four parts of New York City form the population from
which the overall occupancy rates are to be established. A simple random sampling design
would be best suited because hospitals of different sizes, and those serving clientele from
different socio‐economic status, would then be included in the survey, and be most
representative of the population.
Exercise 10.3
The McArthur Co. produces special vacuum cleaners for conveniently cleaning
the inside of cars. About a thousand of these are produced every month with stamped
serial numbers and stored serially in a stock room. Once a month, an inspector does a
quality control check on fifty of these. When he certifies them as to quality, the units
are released from the stock room for sale. The production and sales managers,
however, are not satisfied with the quality control check, since quite often, many of
the units sold are returned by customers because of various types of defects. What
would be the most useful sampling plan to test the fifty units?
A systematic sampling design would be most useful since a probability sampling
design is called for, and also because it would be easy to locate the pieces to be sampled
since they have serial numbers and are stored according to the serial numbers.
Exercise 10.4
A consultant had administered a questionnaire to some 285 employees using a
simple random sampling procedure. As she looked at the responses, she suspected
that two questions might not have been clear to the respondents. She would like to
know if her suspicion is well founded.
A double sampling design will be used in this case. That is, a sub‐sample of the
original sample can be approached to see whether or not the two questions were clear
to them.
The population will comprise all the students of the University. A sampling frame
is available; hence probability sampling is a viable option. There are identifiable
subgroups (students of the five different faculties and graduate schools) which may be
expected to have different parameters to the variable of primary interest (attitude). For
that reason you may want to use a stratified sampling design. A minimum sample size of
30 for each stratum is necessary.
Exercise 10.6
T‐Mobile is a mobile network operator headquartered in Bonn, Germany. The
company has enlisted your help as a consultant to develop and test a model on the
determinants of subscriber churn in the German mobile telephone market. Develop a
sampling plan and pay specific attention to the following aspects.
Define the target population. Discuss, in as much detail as possible, the sampling frame
and the sampling design that you would use. Give reasons for your choice.
Sampling design: proportionate stratified sampling with loyal customers and customers
who have switched mobile phone providers in proportion to their original numbers in
the population.
CHAPTER 11
Instructional Goals
1. To make statistics come alive and become meaningful to students through examples that
they can relate to.
2. To get students to analyze data using any of the available PC software programs.
3. To impress on students that the data collected should first be edited, the blank responses
given an appropriate value, and then entered for analysis.
4. To stress the importance of getting a feel for the goodness of data after recoding.
5. To explain how the frequency distribution becomes a tool for explaining sample
characteristics.
6. To connect the application of descriptive statistics such as the means, standard deviations,
and variance to the dependent and independent variables of interest to any study in order to
get an idea of the central tendencies and to obtain a feel for the data.
7. To make sense out of Pearson Correlations, by interpreting significant correlations, and to
sensitize students to extremely high correlations (over .8) which might cause concerns
regarding the discriminant validity of the measures.
8. To help them interpret the results they obtain from data analyses.
Answer
The tabulation indicates the percentages on all the demographic data (as shown
in parentheses above). A profile of the respondents will be as follows.
Answer
The mean age of the respondents is 37.5 and since the mode is also 38, it can be
presumed that many of the respondents are about 38 years old. The average number of
years the respondents have been married is about 12, with some not married at all
(minimum = 0).
The amount of stress experienced by the subjects in this sample seems to be a little
over the average, but not high (3.7 on a 5‐point scale); the same can be said for their
performance level (3.6 on a 5‐point scale). The extent of their job involvement is close to a
4 (four) on a 5‐point scale and can be considered to be moderately high.
While there is more variance in stress and job involvement (1.79 and 1.63), the
variance in performance is rather low (.86). This would indicate that there is greater
dispersion in the experienced stress and extent of job involvement, but most of the people
have been performing at the same “just above the average” level.
We would thus describe the profile of the average worker in these three
organizations, assuming that the sample is representative of the population, as being about
38 years old, married for about 12 years, performing at just above the average level, fairly
involved in the job and experiencing a moderate amount of stress.
Exercise #3 on Correlations
From the same research done by Ms. Jones, the following inter‐correlation matrix
is shown. Interpret these results.
Table 3: Pearson Correlations
Age 1.0
# of yrs. married .86 1.0
Stress .43 .61 1.0
Job Involvement .53 .32 .58 1.0
Performance .09 .06 .49 .36 1.0
a. All correlations above .30 are significant, at least at the .05 level.
Answer
Eight correlations in the above matrix are significant at least at the .05 level, of
which four are significant at least at the .01 level. Only two of the ten correlations in the
matrix are not significant.
Age and the number of years married are highly correlated (r = .86) and this is not
surprising because older people are likely to have been married for a longer period of
time than younger persons.
Older persons also experience less stress (r = .43) possibly because they have either
learned the ropes well and things do not stress them, or they are better able to handle stress
and hence hardly experience them as stress, or they are given easy jobs.
Older people are also more job involved (r =.53), but age is not related to
performance. That is, though the older employees may be more involved in their jobs than
the younger ones, they do not necessarily perform better than the latter.
Those experiencing more stress are more involved in their jobs (r = .58) and also
perform better (r = .49). The more the job involvement, the better the performance (r
= .36). The above relationships do not indicate causality, merely correlations.
Those married for a longer period experience less stress, possibly because the
family demands are less on them since they would have settled down to a routine family
life which does not induce stress. The persons who are married for longer periods of time
are also more involved in their jobs. But the length of marriage is not related to
performance (r = .06).
If performance is the main variable of interest to this study we would say that
neither age nor the length of marriage has any relationship to performance, but that the
amount of stress and the level of job involvement do have significant bivariate
relationships with the performance of employees. The more job involved people are, the
better they perform; the more stress they experience, the better they perform. Note:
Based on the correlations among stress, job involvement, and performance, it would be
inappropriate to conclude that the high performer gets job involved and hence
experiences stress. Causal inferences cannot be drawn from correlational analyses.
Discussion Questions
1. What activities are involved in getting the data ready for analysis?
Data coding involves assigning a number to the participants’ responses so they can be entered
into a database.
3. Data editing deals with detecting and correcting illogical, inconsistent, or illegal data in
the information returned by the participants of the study. Explain the difference between
illogical, inconsistent, and illegal data.
Inconsistent responses are responses that are not in harmony with other information.
Illegal codes are values that are not specified in the coding instructions.
One way to handle a blank response is to ignore it when the analyses are done. An
alternative solution would be to look at the participant’s pattern of responses to other
questions and from these answers deduce a logical answer to the question for the
missing response. A second alternative solution would be to assign to the item the mean
value of the responses of all those who have responded to that particular item. In fact,
there are many ways of handling blank responses, each of them having its own
particular advantages and disadvantages.
Take for instance the perceived inequity measure of the Excelsior Enterprises case.
Perceived inequity is measured by five survey items: (1) ‘I invest more in my work than I
get out of it’; (2) ‘I exert myself too much considering what I get back in return’ ; (3) ‘For
the efforts I put into the organization, I get much in return’ (reversed); (4) ‘If I take into
account my dedication, the organization ought to give me a better practical training’;
6. There are three measures of central tendencies: the mean, the median, and the mode.
Measures of dispersion include the range, the standard deviation, the variance (where the
measure of central tendency is the mean), and the interquartile range (where the
measure of central tendency is the median). Describe these measures and explain which
of these measures you would use to provide an overview of a) nominal, b) ordinal and c)
interval data?
Scale Measures of Central Measures of Visual Summary Measure of Relation Visual Summary of
Tendency …for a Dispersion ‐‐‐‐‐ …between Variables Relation
Single Variable …for a Single ‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐‐…for …between
Variable a Single Variable Variables
Nominal Mode ‐ Bar chart, pie chart Contingency Table Stacked bars,
(Cross‐Tab) Clustered bars
Ordinal Median Semi‐interquartile Bar chart, pie chart Contingency Table Stacked bars,
range (Cross‐Tab) Clustered bars
Interval Arithmetic mean Minimum, maximum, Histogram, scatter Correlations Scatterplots
standard deviation, plots, box‐ and
variance, coefficient whisker plot
of variation
Ratio Arithmetic or Minimum, maximum, Histogram, scatter Correlations Scatterplots
geometric mean standard deviation, plots, box‐ and
variance, coefficient whisker plot
of variation
The Mean. The mean or the average is a measure of central tendency that offers a
general picture of the data without unnecessarily inundating one with each of the
observations in a data set. The mean or average of a set of say, 10 observations, is the
sum of the 10 individual observations divided by 10 (the total number of
observations).
The Median. The median is the central item in a group of observations when they are arrayed
in either an ascending or a descending order.
Measures of Dispersion
Variance. The variance is calculated by subtracting the mean from each of the
observations in the data set, taking the square of this difference, and dividing the total
of these by the number of observations.
Standard Deviation. The standard deviation, which is another measure of dispersion for
interval and ratio scaled data, offers an index of the spread of a distribution or the
variability in the data.
Other Measures of Dispersion. When the median is the measure of central tendency,
percentiles, deciles, and quartiles become meaningful. Just as the median divides the
total realm of observations into two equal halves, the quartile divides it into four equal
parts, the decile into 10, and the percentile to 100 equal parts. The percentile is useful
when huge masses of data, such as the GRE or GMAT scores, are handled. When the
area of observations is divided into 100 equal parts, there are 99 percentile points. Any
given score has a probability of .01 that it will fall in any one of those points. If John’s
score is in the 16th percentile, it indicates that 84% of those who took the exam scored
better than he did, while 15% did worse.
The measure of dispersion for the median, the interquartile range, consists of the
middle 50% of the observations (i.e., observations excluding the bottom and top 25%
quartiles). The interquartile range could be very useful when comparisons are to be
made among several groups. For instance, telephone companies can compare
longdistance charges of customers in several areas by taking samples of customer bills
from each of the cities to be compared.
What is the best way to provide an overview of the gender of the respondents?
8. Consider the following reliability analysis for the variable customer differentiation.
What could you conclude from it?
R E L I A B I L I T Y A N A L Y S I S ‐ S C A L E (A L P H A)
Item‐total Statistics
Reliability Coefficients
N of Cases = 111.0 N of Items = 3
Alpha = .5878
The scale is not reliable. We can delete item 1 to increase the Cronbach’s alpha, which will
increase to .7454.
Instructional Goals
1. To make students understand and apply the different types of analyses and tests to get
at the inferential information they need.
2. To help them interpret the results they obtain from data analyses.
Discussion Questions
1. When we collect data on the effects of treatment in experimental designs, which
statistical test would be most appropriate to test the treatment effects?
We would use a t‐test with paired samples to determine the treatment effects.
2. A tax consultant wonders whether he should be more selective about the class of
clients he serves so as to maximize his income. He usually deals with four categories of
clients: the very rich, rich, upper middle class, and middle class. He has information on
each and every client served, the taxes paid by them, and how much he has charged
them. Since many factors relating to the clients vary (number of dependents, business
deductibles, etc.), irrespective of the category they belong to, he would like an
appropriate analysis to be done to see which among the four categories of clientele he
should choose to serve in the future. What kind of analysis should be done in the
above case and why?
Undergraduate Students can be exposed to a simple ANOVA design.
An ANOVA will help the tax consultant to make the decision, since the differences
among the four groups on the dependent variable, income, can be obtained through
this analysis. A further test such as the Duncan Multiple Range test, will determine
which among the four groups provides the most income.
CHAPTER 13
QUALITATIVE DATA ANALYSIS
Instructional Goals
1. To make qualitative data analysis come alive and become meaningful to students
through examples that they can relate to.
2. To explain that qualitative data can be very rich and to stress the importance of
reducing qualitative data to key ideas and concepts.
3. To get students to analyze data using a structured approach.
4. To stress the importance of reliability and validity in qualitative data analysis.
Discussion Questions
1. What is qualitative data? How do qualitative data differ from quantitative data?
Qualitative data is data in the form of words. Quantitative data is data in the form of
numbers.
According to Miles and Huberman, there are generally three steps in qualitative data
analysis: data reduction, data display, and the drawing of conclusions.
109
(Kassarjian 1977).
Validity refers to the extent to which the research results 1) accurately represent the
collected data (internal validity) and 2) can be generalized or transferred to other
contexts or settings (external validity).
4. How can you assess the reliability and validity of qualitative research?
Two methods that have been developed to achieve validity in qualitative research are
discussed next.
‐ Supporting generalizations by counts of events. This can address common concerns
about the reporting of qualitative data; that anecdotes supporting the researcher’s
theory have been selected, or that too much attention has been paid to a small number
of events, at the expense of more common ones.
‐ Ensuring representativeness of cases and the inclusion of deviant cases (cases that
may contradict your theory). The selection of deviant cases provides a strong test of
your theory.
Narrative analysis is an approach that aims to elicit and scrutinize the stories we tell
about ourselves and their implications for our lives. Narrative data are often collected
via interviews. These interviews are designed to encourage the participant to describe a
certain incident in the context of his or her life history. Like this, narrative analysis
differs from other content analysis; it is focused on a process or temporal order, for
110