2.2 Introduction

Broadly speaking, research designs can be divided into quantitative and qualitative. Quantitative research designs are based on measuring or counting things. They may be used to describe populations numerically; for example, to determine the average age of the population or the percentage who smoke. They are also designed to provide answers to questions, in other words to test hypotheses. Philosophically, quantitative research designs use a deductive approach; that is, they start with a statistical model and use the data collected to accept or reject the model. Quantitative research is designed to eliminate bias, or at least reduce it through the design itself or by statistical analysis. For this reason, quantitative researchers stand back from the research to ensure that they themselves do not introduce their own biases. Quantitative research is designed to be generalisable by use of a sufficiently large and representative sample.

Qualitative research is in many ways the opposite. It is based on natural enquiry, for example, observing or interviewing people. Rather than trying to answer a question, qualitative research is designed to explore the meaning of things. For example, a quantitative study might ask the question ‘does this treatment work?’, whereas a qualitative study might ask ‘why does this treatment work and in what circumstances?’ Qualitative research usually employs an inductive approach in that it moves from the data collected to a theory. Qualitative researchers often immerse themselves in the research, since elimination of bias is not a primary aim.

When designing a study, we usually start off with a research question followed by a series of aims, each addressing different aspects of the research question. It is the research question and study aims that determine whether a quantitative or qualitative approach should be used. For example, if the research question is ‘how do people feel when their lifetime partner is diagnosed with a terminal illness?’, then a qualitative approach is most suited to address this issue. Alternatively, if the research question is ‘do hip protectors reduce fracture rates in elderly people?’, then a quantitative approach is preferable. It is becoming increasingly common to see mixed approaches taken, where both qualitative and quantitative research are combined in a single study.

2.3 Quantitative research designs

2.3.1 Epidemiology

In many studies, the main aim is to determine whether exposure to a risk factor or health intervention is associated with a disease outcome. The scientific discipline that explores these relationships is called epidemiology. Epidemiology is the discipline underlying much of public health and clinical research, and is the foundation for much of the ‘evidence’ in evidence-based practice (EBP).

Every individual is unique; however, many individuals share common characteristics. For example, some drink or smoke, others may work in the same industry. Quite often the common characteristic is residing in the same city. It is the study of health in populations of people that differentiates epidemiology from clinical medicine, the latter being concerned with the study of health in individuals.

Last’s Dictionary of Epidemiology describes epidemiology as ‘the study of the distribution and determinants of health-related states or events in specified populations, and the application of this study to control of health problems’ (Last 2000:62).

The distribution part of the statement refers to who contracts or suffers from the disease. For example, which age or gender groups are most affected? When did they get the disease? In which place or location did the disease occur? Determinants refer to what caused the disease. The study of disease causation is also known as aetiology.

Last (2000) uses the expression ‘health-related states or events’ since epidemiology now covers a wide range of health or health-related outcomes. In the earliest days of epidemiology, the main concern was study of infectious diseases. As these were gradually brought under control, attention then moved to acute diseases such as heart disease and cancer. More recently, we have focused on chronic diseases such as asthma and diabetes. We also now give priority to mental health problems, such as depression, and measures of quality of life. Finally, we are also interested in events such as accidents or birth defects. Clearly our understanding of disease has become much broader over time and this has shaped research.

Whenever we measure disease, we always do so with respect to the population at risk. For example, we might make the statement ‘there were 1700 new cases of prostate cancer in Australian males in 2006’. We chose the male population since only males can get prostate cancer. The male population would be the estimated mid-year male resident population as at 30 June 2006.

In epidemiological studies, one of the primary goals is to measure the association between exposure to a risk factor (or health intervention) and disease—the stronger this association, the more likely the relationship to be causal. Two commonly used measures are the relative risk and the odds ratio. The relative risk (also known as the rate ratio) is calculated by dividing the percentage of people with the disease in the group exposed to the risk factor by the percentage of people with the disease in the group not exposed to the risk factor. Thus, a relative risk of 2 would be interpreted as those exposed to the risk factor would be twice as likely to get the disease as those not exposed to the risk factor. The odds ratio is a similar measure based on odds rather than on the percentage with the disease.

Finally, clinical epidemiology is the branch of epidemiology related to the use of epidemiological methods in patient populations. Importantly, this includes clinical trials of new drugs or health interventions. It is clinical epidemiology that underpins much of the high-level evidence for clinical interventions.

2.3.2 Sampling error

With quantitative research, we are nearly always working with samples of the population of interest rather than the whole population. We then make the major assumption that the conclusions drawn from our study are applicable to the whole population of interest, a process known as ‘inference’. Sampling error is error due to the use of a sample rather than the whole population. Imagine that you had a ward of 30 patients and were trying to determine their average age. If you chose two patients at random and asked them their age, and then took the average, it would likely not be an accurate reflection of the true average age of the whole ward. However, if you took a sample of 10 patients at random, then the average age would likely be much closer to the true value. Thus, sampling error can be reduced by increasing the sample size.

2.3.3 Bias

Bias is defined as a consistent deviation from the truth. More formally, it is defined as ‘deviation of results or inferences from the truth or processes leading to such deviation’ (Last 2000). There are dozens of different types of potential bias, including bias relating to the collection, analysis, interpretation, publication or review of data (Last 2000). With respect to quantitative studies, the main biases of interest include:

• Sampling bias: bias due to the use of a non-random sample.

• Ascertainment bias: bias due to a failure to represent equally all classes of persons supposed to be represented (e.g. the use of electronic white pages to recruit subjects omits those who do not have a landline telephone).

• Selection bias: error due to systematic differences in characteristics between those who take part in a study and those who do not (e.g. volunteers might differ from non-volunteers).

• Recall bias: systematic differences in the accuracy of memory (e.g. people with mesothelioma—a cancer of the lining of the lung—are probably more likely to remember any exposure to asbestos).

• Confounding bias: this will be described in Section 2.3.5.

Unlike sampling error, bias cannot be reduced by increasing the sample size. One feature of quantitative studies are type I and II errors. A type I error occurs when the researcher, based on the study results, declares a treatment to be effective when it really is not. This is most often due to some type of bias. On the other hand, a type II error occurs when, based on the study results, a researcher declares a treatment to not be effective when it really is. This most often occurs when the researcher has used too small a sample size. In this case, the study is said to be underpowered.

2.3.4 Validity

The validity of a study (the word validity comes from the Latin validus, meaning strong) is the degree to which inferences or conclusions drawn from the study are warranted (Last 2000). We distinguish between internal and external validity. Internal validity is where any differences in outcome between study groups (apart from sampling error) can only be attributed to the hypothesised effect under investigation; in other words, the study is unbiased. External validity refers to how well it is possible to generalise the conclusions drawn from a study to other populations.

2.3.5 Confounding

Confounding is a special type of bias in which a measure of the effect of an exposure on risk is distorted because of the association of exposure with other factor(s) that influence the outcome under study (Last 2000). For example, in 1971, a case-control study by Cole found a link between coffee consumption and bladder cancer. In other words, the more coffee you drank, the more likely you were to get bladder cancer. However, it was later shown that heavy coffee drinkers were also more likely to smoke cigarettes, and it was the cigarette smoking causing the bladder cancer rather than the coffee drinking. Here we say that cigarette smoking confounds the relationship between coffee drinking and bladder cancer.