Ways of approaching research : Quantitative Designs

Quantitative Designs

Qualitative Designs

Research is Focused

Data Collection Methods

My Home Page

The first session of the R&RA theme gave an overview of the research process. It also highlighted the fact that there are principally two forms of research design: quantitative and qualitative. The second session focuses on quantitative approaches to research.

Objectives

By the end of the session the students will be:

1. Able to define quantitative research

2. Able to identify features commonly associated with quantitative research rather than with qualitative research

3. Able to give an outline of the most prominent forms of quantitative research

4. Aware of key terms associated with quantitative research

 Definition

Quantitative research is:

"a formal, objective, systematic process in which numerical data are utilised to obtain information about the world"
(Burns and Grove cited by Cormack 1991 p 140).

Features associated with the quantitative approach

A brief consideration of the major distinctions between quantitative research and qualitative research can help put quantitative research into context.
 
 

Table 1 : Comparison of features of Quantitative and Qualitative approaches to research

Both designs, quantitative and qualitative are said to be systematic. In fact having a system or following a process is a defining principle of research.
Broadly speaking, quantitative research is thought to be objective whereas qualitative research often involves a subjective element.
It is thought that in gaining, analysing and interpreting quantitative data, the researcher can remain detached and objective.
Often this is not possible with qualitative research where the researcher may actually be involved in the situation of the research.
Consider a study being undertaken into waiting times in the Accident & Emergency (A&E) Department of a hospital.
A quantitative study, measuring how long people wait, can be purely objective.
However if the researcher was wanting to discover how patients felt about their waiting time, they would have to come into contact with the patients and make judgements about the way they answered their questions.
If the researcher asked the patient "how are you feeling having waited an hour to be seen by the doctor?" they would almost certainly register the patients non-verbal behaviour as well as document the response; in this way the researcher is adding a subjective element to the study.

Quantitative research is inclined to be deductive.
In other words it tests theory.
This is in contrast to most qualitative research which tends to be inductive.
In other words it generates theory.

Using the A&E waiting time example again, the quantitative approach might test the hypothesis that " Patients attending this A&E department do not wait for more than one hour to be seen by a doctor".
A qualitative approach which explores the feelings of patients who wait an excessive time to be seen by the doctor might generate the theory that "patients who experience an excessive wait to be seen by the doctor experience an enlargement of the symptoms that brought them to the department".

Quantitative designs of research tend to produce results that can be generalised.
Using our A&E example, we should find that, at least for the department under consideration, the results of the quantitative study tend to hold true. Providing, of course, that the research was conducted in an appropriate manner using appropriate sampling techniques.
However, qualitative studies tend to produce results that are less easy to generalise.
This has to do with the problem of the sample used at the time. We all know, for example, that our feelings about waiting can change dependent on our particular set of circumstances. Even if the researcher encountered the same group of clients on another day, they may find different results. Generally, it is difficult to generalise with qualitative results.

Lastly here, the most obvious difference between quantitative research and qualitative research is that quantitative research uses data that are structured in the form of numbers or that can be immediately transported into numbers.

If the data can not be structured in the form of numbers, they are considered qualitative.
(Note that qualitative data can sometimes be handled in such a way as to produce quantitative data. e.g. the researcher exploring feelings of patients can analyse the responses in clusters that are negative or positive so as to produce a figure/percentage of negative patient and positive patient feelings).

Therefore, objectivity, deductiveness, generalisability and numbers are features often associated with quantitative research.

When a researcher selects their approach to a study it should be a reflection of which approach is most suitable for the topic under consideration. However it is also reasonable to suggest that it also reflects the bias of the researcher. The majority of medical research is quantitative (and considered to produce "hard", generalisable results) whilst much of nursing research is qualitative (and considered to produce "soft" results).

There has been a lot of debate about whether or not nurses should aim to do more quantitative research which is generalisable.
Often nurses argue that nursing is about people in specific circumstances and, because people are individuals, findings about them cannot be generalised.
We can nevertheless create theories about the ways people will act or react in specific conditions.
Some argue that nurses lack the statistical skills to undertake quantitative research and that they "cop out" by doing qualitative research instead.
There is probably an element of truth in both sides of the argument.
What we should be concerned about is that the appropriate method is used for studying particular problems or aspects of life. Sometimes quantitative research is going to be the most appropriate approach in studying a particular subject.

Major types of Quantitative Research

Types of quantitative research include:
 

Some authors (see Burns & Grove 2nd Edn. 1993) include a fourth type of design known as a correlational (Ex post facto) study although others would dispute whether this is a quantitative approach to research.

Basically, experimental and quasi-experimental studies are designed to examine cause and effect. These studies are usually conducted to examine the differences in dependent variables thought to be caused by independent variables (treatments).
Descriptive and correlational studies examine variables in their natural environments and do not include researcher imposed treatments.

There is no universal standard for categorising research designs and different authors may change names of designs in their discussions of them.
Thus what is shown here is intended more to be informative than exhaustive.
This lack of universalism also causes problems when critiquing research as many published studies do not identify the design used.
Selecting an appropriate design for a study involves following a logical thought process. A calculating mind is required to explore all possible consequences of using a particular design in a study.
 

Experimental design.

The paradigm for scientific method in research is the true experiment or randomised control trial (RCT). Typical examples of RCT's include drug trials.
Experimental designs are set up to allow the greatest amount of control possible so that causality may be examined closely.
The three essential elements (OHP) of experimental design are:

Manipulation : The researcher does something to at least some of the participants in the research

Control : The experimenter introduces one or more controls over the experimental situation.

Randomisation : The experimenter assigns participants to different groups on a random basis.

(adapted from Polit & Hungler 4th Edn.1997)

The classic example is the before-after design or pre-test post-test design.
This is perhaps the most commonly used experimental design.
Comparison of pre-test scores allows the researcher to evaluate how effective the randomisation of the sample is in providing equivalent groups.
The treatment is fully under control of the researcher. The dependent variable is measured twice during the study (before and after the manipulation of the independent variable).
Example : Drug trials

In some studies the dependent variable cannot be measured before the treatment. For example we cannot effectively measure the response to interventions designed to control nausea from chemotherapy prior to the beginning of treatment.
Here we would us an approach known as the post-test only design
We may also wish to use this approach where pre-test sensitisation may occur. Subjects post-test response may be partly due to learning from, or as a reaction to, the pre-test.
In these instances the pre-test phase can be eliminated, however doing so removes the possibility of applying some very powerful statistical analyses.

A development of the pre-test post-test design is known as the Solomon 4-group design. Although somewhat cumbersome this design effectively measures the influence pre-testing has on post-test scores. It is a stronger tool than the pre-test post-test design but requires more complicated statistical analysis of the data obtained.

Factorial designs are a further development of the experimental technique which allow for two or more different characteristics, treatments, or events to be independently varied within a single study. This is a logical approach to examining multiple causality.
 

Quasi-experimental design.

Quasi-experimental designs were developed to provide alternate means for examining causality in situations which were not conducive to experimental control.
The designs have been developed to control as many threats to validity as possible in situations where at least one of the three elements of true experimental research is lacking (i.e. manipulation, randomisation, control group).
There are too many types of Quasi-experimental design (see Burns & Grove 1993 pp.305-316) to go into in great depth here.. Most are adaptations of experimental designs where one of the three elements is missing.
An example could be where the researcher uses groups (control and treatment) which have evolved naturally in some way rather than being randomly selected.
This is a quasi-experimental approach using non-equivalent control groups.
 

Descriptive design.

Descriptive designs are designed to gain more information about a particular characteristic within a particular field of study. A descriptive study may be used to, develop theory, identify problems with current practice, justify current practice, make judgements or identify what others in similar situations may be doing.
There is no manipulation of variables and no attempt to establish causality.

Correlational studies as already mentioned are not universally accepted as a form of quantitative research.
However they do crop up in the literature so we will briefly discuss them here.
As already noted they are also know as ex post facto studies. This literally means "from after the fact".
The term is used to identify that the research in question has been conducted after the variations in the independent variable has occurred naturally.
The basic purpose of this form of study is to determine the relationship between variables. However the significant difference from experimental and quasi-experimental design is that causality cannot be established due to lack of manipulation of independent variables.
"Correlation does not prove Causation"
Examples include many studies of lung cancer. The researcher begins with a sample of those who have already developed the disease and a sample of those who have not. The researcher then looks for differences between the two groups in antecedents, behaviours or conditions such as smoking habits.
 

Glossary of terms used in session 2

Mean

(full title = arithmetic mean) the value each item would have if all the values were shared out equally among all the items e.g., the mean of

2 + 2 + 2 + 2 + 4 + 4 + 5 + 6 + 6 + 6 + 8 = 47/11 = 4.27

the mean can be corrupted by "freak" values thus if the values above referred to the age of children on a children's ward, the age would be 4.27 but if we add a 15 year old to the sequence the average age = 5.16
 

Median:

the value of the middle item of a distribution e.g., in the sequence above there are eleven values and the value occurring in the middle position is the value "4" thus 4 is the median average for this sequence.

Mode:

(modal value) the point of maximum frequency e.g., in the sequence of numbers above, the value "2" occurs most frequently and is thus the modal value.

Normal distribution: this is illustrated by a normal (Gaussian) curve. see Graph 1

Normal distribution can be tested for by:

a) plotting the data on a graph and visually inspecting the line

b) using a probability graph in which if it is possible to draw a straight line through the values distribution is normal

c) applying the Chi-Square Test for normal distribution.

d) checking whether the mean, mode and median coincide (which they do if distribution is normal)

or

e) testing that:

Concepts that relate broadly to both quantitative and qualitative research

Association:

Sometimes there is a relationship between two variables but the relationship may not be causal i.e., neither variable is dependent upon the other. It may be seen that short men are more assertive than taller men but it does not follow that being short causes men to be assertive and we can not state that being assertive make a man short; there is no causal relationship.
 

Bias:

Distortion of the findings resulting from an undesirable influence.
 

Causality / Causal relationship:

A relationship in which one action brings about (causes) a particular consequence. More correctly, (since research can only hope to disprove a theory rather than prove it), a relationship in which failure to do ‘x’ means that ‘y’ will not follow. E.G., We can be less certain that bathing in the sun for two hours around midday causes skin to burn, than we can know that keeping out of the sun for the two hours around midday is unlikely to cause skin to burn.
 

Hawthorn Effect:

A psychological response in which subjects alter their behaviour because they are aware of their participation in a study.
 

Piloting:

A small-scale trial of the research method to ensure that the design is feasible. Although only a small number of subjects may be used, a variety of practical questions may be determined. E.G., can the subjects understand the questions they are being asked?
 

Population:

Literally means "all the people" and in research the term is most commonly used to refer to a specific group of people. However, in a research context, population refers to all the members or objects of any defined group which might be taken or about which information might be given.
A research population refers to the entire group to which the research results apply e.g., a relevant age group, or equipment group such as syringes.
 

Reliability:

Is concerned with the accuracy (consistency, stability and repeatability) of a measure in representing the true score of the subject being assessed on a particular dimension. The same results must be achieved, as far as possible, regardless of whom is doing the measuring. E.G., several nurses weighing the same patient on the same set of scales, in quick succession, should gain the same results. Reliability of measurement reduces influence or bias on the part of the person(s) doing the measurement, to a minimum.
 

Representative:

Refers to the extent to which a sample reflects the "truth" for the whole population in the study. The sampling technique should aim to ensure that the views of the population are reflected by the sample.
 

Validity:

Refers to whether a particular instrument actually measures the construct it is designed to assess. E.G., a cardiac monitor is not a valid tool for measuring the peripheral pulse. A cardiac monitor is a valid tool for measuring the electrical activity of the heart.
 

Internal validity:

The extent to which the effects detected in a study are a true reflection of what is real, e.g., if the detected effect is that better nutrition leads to greater height gain in infants, internal validity exists if the height gain can not be attributed to another factor. (NB this other factor may be referred to as a confounding or extraneous variable).
 

External validity:

"Tthe extent to which study findings can be generalised beyond the sample used in the study" (Burns and Grove 1993). E.g., One study may find that better nutrition leads to increased height gain in infants but external validity exists only if this finding is found with other samples.

NB the concept "Variable" appeared on the glossary of the previous session

Sampling:

Sample; Sampling; method of selecting a certain number of units from a total population.

(Macleod Clark J and Hockey L. 1981)

The way a sample is selected should be clearly demonstrated in a research report. The aim of a sample is that it should be as unbiased a cross section of the "parent" population as possible, i.e., a sample of subjects needs to be as representative as possible of the population under study.

To obtain a cross section we need to devise a sampling frame to define the boundaries (limits) within the context of the study and to reflect the organisation within which the sampling is taking place.

The larger the size of the sample, the lower is the likelihood of it failing to represent the population under study. However, the law of diminishing returns tells us that there is, for each study, a desirable sample size under which their may fail to be accuracy yet above which there is no better a reflection of the parent population.

Sampling may be

a) random

b) systematic

c) refinements of random and systematic

a) random-

b) systematic e.g., every 5th/xth

c) Refinements;

1. Stratified sample

Where there is heterogeneity in the population this can be reflected in the strata, i.e., each stratum can be weighed to reflect the heterogeneity. In this way a proportional representation of the whole population can be gained.

2. Cluster sample

Best used where there is a wide geographical spread. Clusters may be chosen subjectively to be representative of the whole. The clusters can be further stratified. E.G., if we want to know about all A&E patients in the country we need to take a sample from a variety of A/E’s. Each department can bring a number of patients into the sample according to whether they meet the stratification criteria.
 
 

Concepts strongly associated with quantitative research

Survey:

Involves the study of a large number of subjects drawn from a defined population.
 

Randomisation:

A method for controlling possible extraneous variables involving assigning objects (subjects, treatments etc.,) to a group or condition in such a way that every object has an equal probability of being assigned to any particular condition.
Randomisation can also be applied in other settings for research e.g., within a questionnaire there may be a sequence of questions, which, if reordered randomly, may evoke different responses because the previous question does not then influence the current question.

Control:

In order to increase the probability that findings accurately reflect the reality of the situation being studied, the study needs to be designed in such a way as to maximise the amount of control over the research situation and variables. Through control the influence of extraneous variables, variables which are not being studied but which could influence the results of the study by interfering with the action of the ones being studied, is reduced.
 

Manipulation:

Refers to the fact that we can create artificial divisions and circumstances in order that we can test a particular hypothesis. In experimental research the "causative" variable must be amenable to manipulation by the investigator; i.e., the researcher "does something" to subjects in the experimental condition. Subjects in the control group are not "manipulated" in the way that subjects in the "experimental group" are manipulated.
 

Measurement (levels)

Nominal:

Subjects of research are differentiated by possessing or not possessing a given characteristic, e.g., pass/fail, single/married, and divided into a number of categories but the difference between the categories is not measurable in any real sense. This is the least sophisticated level of measurement.
 

Ordinal:

Subjects are ranked in order from greatest to least or best to worst. Again there is no precisely measurable difference between the ranks.

Interval:

Genuinely quantitative measurement such as that of temperature is measured at the interval level of measurement. Here the difference between 10 and 11 degrees centigrade is the same as the difference between 11 and 12 degrees centigrade.
 

Ratio:

In a scale of measurement where the difference between points on the scale is precise (as in the measurement of height and weight,) and the scale starts at zero the level of measurement is referred to as ratio. Height and weight start at zero. You can not weigh less than 0.00kg and cannot be less than 0.00mm in length/height; these are ratio scales. You can however record temperatures of the weather in terms of minus x degrees centigrade and this is why the scale is interval and not ratio.
 

Page last updated Saturday April 10, 1999 22:55 +0100

Please e-mail any comments or queries to johnross@cwcom.net

[ Up ] [ Next ]