CHAPTER

25
ELEMENTARY QUANTITATIVE METHODS
Cathy Lewin
Education and Social Research Institute, Manchester Metropolitan University, UK

Key concepts studied, whether a person or an object, each of

whom or which is often referred to as a case.
Statistical methods are a wide range of tools and Attributes can be real measurements or something
techniques that can be used to describe and interpret that can be counted or quantified (for example age,
data that are quantitative or can be measured height, income, test scores). Numbers can also be
numeri- cally. Numerical data can make a valuable used to ‘measure’ opinions and attitudes through
contribu- tion in both quantitative and qualitative ranked responses to data collection methods such as
research whether it be simple percentages or the survey questions or structured observations (for
results of complicated techniques. The use of mixed example educational level, socio-economic status,
methods (see Greene et al., in this volume) has rating of services such as banking). Variables can
become increasingly popular as a means to harness also be assigned specific values (0, 1, 2 and so
the strengths of both approaches, triangulate data forth) to represent categorical attributes or
and illuminate statistical findings with, for example, characteristics that cannot be measured numerically or
case studies and/or vignettes. ranked in any way such as eye colour or gender.
Quantitative researchers require knowledge of a
range of very precise methods and procedures, all of
which are associated with specific terminology and a How can statistics be used?
range of principles arising from probability theory. Statistics are particularly useful when asking questions
This chapter seeks to provide the foundations re- of large numerical datasets, enabling researchers to
quired to understand quantitative research. The first summarize and make comparisons. Descriptive statistics
section of the chapter provides the reader with an are used to describe and summarize data and include
introduction to statistics – what can be measured measures of central tendency (average) and dispersion
and how – and introduces the concepts of reliability (the spread of data or how close each case is to the
and validity. The second section covers sampling measure of central tendency). Descriptive statistics
stra- tegies: how to choose what will be included in have an important role to play, enabling data to be
quantitative studies in the social sciences. The third explored before any further analysis is undertaken but
section provides an introduction to questionnaire also as a primary means of describing how things are
design, a data collection instrument commonly used rather than seeking to explain why phenomena occur.
within quantitative paradigms to survey a large numb- Inferential statistics are used to identify differences
er of respondents. Finally, the last section explains between groups, look for relationships between at-
how statistics can be used to describe and explore tributes and create models in order to be able to make
numerical data. predictions. Inferential statistics are introduced and
discussed in Barnes and Lewin (in this volume) and
Introduction to statistics Jones (in Part VI, in this volume). Statistics can be
applied to a single variable (univariate analysis), two
What can be measured? variables (bivariate analysis) or more than two vari-
Statistics are applied to variables or measurements of ables (multivariate analysis). The kind of statistical
attributes or characteristics of whatever is being tool that can be used also depends on the type of data
involved and whether specific conditions have been
215
 PART VI SAMPLING, CLASSIFYING AND QUANTIFYING

met. The significance level of a statistical test is also selected at the

established, that is the likelihood that a difference or
relationship has been identified when it does not truly 216
exist. If the probability of this occurring is very
small it means it is less likely that the result has
occurred by chance and so the researcher can be
more confident about the findings.

What do we mean by reliability and validity?

Reliability refers to the stability or consistency of
measurements; that is whether or not the same results
would be achieved if the test or measure was
applied repeatedly. For example, a question may be
worded ambiguously and answered differently on
different occasions. Validity refers to whether or
not the
measurement collects the data required to answer
research question. A measure can be reliable (always
generate the same result) but not valid (not measure
the intended concept). However, if it is not reliable
then it cannot be valid. There are various aspects of
validity that should be considered when designing any
measurement (see, for example, de Vaus, 1995, for
more on this) and threats to validity can differ
according to the statistical approach undertaken (see
Jones, in Part VI in this volume, for a discussion of
threats to validity in relation to modelling).
Causality can be inferred if it can be
demonstrated that changing the value of one variable,
the independent variable, has an effect on the value
of another, the dependent variable. It is a means of
explaining a phenomenon through its likely causes.
Internal validity refers to the confidence that can be
placed in causal inferences. There may be other
(unaccounted for) variables at play. Some variables
will have a direct effect on others while others may
have an indirect effect. There are many threats to
validity in quantitat- ive research including history
(circumstances changing over time), testing (test
practice effects), mortality (attrition or being unable
to collect data from all original participants) and
maturation (developmental changes in participants).
Generalizability or external validity refers to the
possibility of expanding any claims of causality
from the group or sample being studied to the
population that the group represents – that is, that
the same effect will be found in another group
and/or in other contexts.

Quantitative designs
Quantitative research can employ a number of
differ- ent designs, one of which is usually
outset depending on the kind of research question
being investigated. Experimental design is the primary
approach in the positivist paradigm (see Garratt
and Li, in this volume). This involves the
manipulation of at least one independent variable to
see whether or not it has any impact on the
dependent variable. Tests can be conducted before
the experiment begins – pre-test – and after it has
been completed – post-test – or just at post-test.
These data are used to identify differences between
two or more groups on measure- ments of the
dependent variable. Laboratory experiments take place
in contrived settings but allow researchers to have
more control whereas field experiments are
conducted in naturalistic environments where it is
often easier to recruit participants. Many argue that
results achieved in laboratory settings are not general-
izable to naturalistic settings casting doubt on the
external validity of such experiments. Often in
social science research, quasi-experimental designs are
adopted when it is not possible to allocate individuals
random- ly to groups (see section on the principles of
sampling below). For example, in educational
research whole schools or whole classes are often
assigned to groups (rather than individuals being
randomly assigned to groups) because of practical
and logistical issues.
Randomized controlled trials or RCTs are one
form of experimental design in which participants are
allocated truly randomly to an experimental group
(for example, those exposed to the independent
variable such as a new drug) and a control group
(those not), enabling unmeasured or unknown vari-
ables to be taken into account and strengthening
claims for internal validity. These approaches are
expensive due to the large numbers of participants
required. Furthermore, random allocation can be hard
to achieve in social science research and there are
ethical considerations that necessitate constraints.
Nevertheless, there are often opportunities to set up
randomized experiments when an experimental and
control group occur naturally, for example when there
is a limited number of places on a course and
participants are selected randomly. RCTs are often
referred to as the ‘gold standard’ for quantitative
research although the value of such an approach is
not universally accepted by social scientists.
A cross-sectional design is often used in survey research
and involves the collection of quantitative data on
at least two variables at one point in time and from
a number of cases. These data are then used to look
for patterns of association or relationships either in
the group as a whole (all cases) or in subgroups
sharing
 25 ELEMENTARY QUANTITATIVE METHODS

characteristics or attributes (females or males for

Probability or random sampling strategies
example). It is problematic to establish causality in
simple statistical tests of relationships (see Barnes and Simple random sampling is the simplest strategy in which
Lewin, in this volume) but causal inferences can be each population member has an equal chance of
made using more sophisticated techniques such as selection through ‘pulling names from a hat’ or
regression analyses (see Jones, in part VI, in this assigning each member a unique number and using
volume). random number generators (tables of random numb-
A longitudinal design, often an extension of a cross- ers or a computer program that generates random
sectional design when a survey is administered repeat- numbers within a specified range). However, a
edly at regular time intervals over a number of com- plete list of the population is required and this
years, can be used to more easily establish causality is not always available. Systematic sampling is similar
but is expensive to conduct. but uses the sampling frame (a complete unordered list
of all members of the population) rather than
random numbers in the selection process. A
Principles of sampling
member of the population is selected at regular
Social science research can focus on a specific intervals from the sampling frame. The sampling
population or complete set of units being studied (for frame should not be ordered (names listed
example, all state secondary schools in one country or alphabetically for example) or there may be a bias in
all nurses working in a region) when time, costs and the selected sample.
accessibility often prohibit the collection of data from Stratified sampling involves ordering the sampling
every member or about every item. In these situations frame by one or more characteristics and then
it is necessary to select a representative sample of the selecting the same percentage of people or items from
population, one in which the same range of charac- each subgroup either using simple random or system-
teristics or attributes can be found in similar propor- atic sampling. This will ensure that characteristics of
tions. It is only with a truly representative sample that the population are represented proportionately (for
you can generalize the research findings to the whole examples males and females). The more characteristi-
population. So judgements have to be made to cs that are used, the more complex this procedure will
ensure that the sample is as representative as be. Only characteristics that are considered to be
possible adopting one of a number of different likely to affect the data analysis should be considered.
sampling strategies to go some way towards When the population is large and widely
overcoming poten- tial limitations. A census involves dispersed it may be more appropriate to initially
collecting data from all members of the population select sub- groups such as geographical areas rather
and is a true represen- tation. Sampling, however, than ran- domly select from the whole population.
results in an estimate of population characteristics This is known as cluster sampling. For example, a
because the sample selected may not be truly number of hospitals could be randomly selected
representative. Researchers should explain the from the list of all hospitals in a country and then
sampling strategies used in their research so that the sample identified through a random sampling
readers can make judgements about potential bias strategy (simple, systematic or stratified) applied to
that might be introduced or other limitations. In lists of nursing staff at those hospitals selected
probability sampling each member or item of the initially. An extension of cluster sampling is stage
population has an equal or known chance of being sampling in which more than one level of grouping is
selected. It is usually possible to generalize findings used to generate the sample such as selecting a
from analysis of data collected from such a sample region, then a school, then a class, then a number of
to the population overall. Non-probability sampling students within that class.
covers all other approaches.
There are many ethical considerations that need to
be addressed such as participant consent (see Piper Sampling error
and Simons, in this volume). Some samples will be Probability or random samples have less risk of bias
easier to access than others by the nature of the (selecting subgroups disproportionately, for example
population characteristics. For example, access to twice as many men as women) but will still be subject
employees in companies will be easier than self- to a degree of sampling error or the difference
employed people working from home. between attributes or characteristics of the sample
and the
217