FIGURE 1.1 Pre‐test–post‐test design.
The weakness of this design is that there may have been a number of factors influencing the outcome of the study, but you may be aware of only some of them. An example is a study on the influence on growth of food supplements given to children. Was there bias in the way the children were recruited into the study? Did the supplement contain enough of the relevant energy or nutrients to have an impact on growth? Were the children at the right age for the supplement to have an impact? Was follow‐up long enough to observe the impact of the supplement on growth? Participation in the study may in itself benefit growth (more social interaction, more stimulation from adults and peers, more food sharing, changing dietary patterns at home as a result of parental involvement in the study, etc.). It is possible that the direct effect of the supplement may represent only one of several factors that can have in impact on growth.
One sample (Figure 1.2). Here, two or more treatments are compared in one group of subjects. Each subject acts as their own control. There may be a placebo being compared with an active treatment.
The danger here is that the order of administration may have an effect on the outcome. It is therefore desirable to randomize the order of administration of the treatment and placebo (making the design more like a cross‐over clinical trial – see below). Thus, for one subject, Intervention 1 will be the active treatment, and Intervention 2 the placebo; for another subject, Intervention 1 will be the placebo, and Intervention 2 the active treatment. A ‘wash out’ period may be required, so that any residual effect of the first treatment has time to disappear and not appear as an influence on the second treatment. This may not be possible where a psychological or attitudinal variable is being measured, because the views of a subject may be permanently influenced once they have started to participate in the study.
FIGURE 1.2 One‐sample design.
Two or more samples, unmatched (Figure 1.3). Similar experimental groups are given different treatments. One group acts as a control (or placebo) group.
This is a stronger experimental design than the pre‐test–post‐test or one‐sample designs. However, the groups may differ in some characteristic which is important to the outcome (for example they may differ in age). Even if the groups are not ‘matched’ exactly, some attempt should be made to ensure that the groups are similar regarding variables which may be related to the outcome. For example, in a study to see if the administration of oral zinc supplements improves taste sensitivity in cancer patients, the type, degree, and severity of disease would need to be similar in both the treatment and placebo groups.
Two or more samples, matched (Figure 1.4). The participants in the study groups are matched in pairs on a subject‐by‐subject basis for variables such as age and gender (so‐called ‘confounding’ variables). One group is then the control (or placebo) group, and the other group is given the active treatment(s).
FIGURE 1.3 Two‐sample (parallel) design.
FIGURE 1.4 Two‐sample (matched) design.
For example, if the aim was to test the effect of a nutrition education programme to persuade urban mothers to breast feed their infants (versus no education programme), it would be important to match on age of mother and parity (number of births). Matching could be carried out by first grouping mothers according to the number of children. Then, within each parity group, mothers could be ranked according to age. Starting with the two youngest mothers in the group with one child, one mother would be randomly assigned to the treatment group (the education programme), and the other would be assigned to the control group (no education programme).11 The next two mothers would be assigned randomly to treatment or control, and so on, by parity group and age, until all the subjects have been assigned. In this way, the possible effects of age and parity would be controlled for. If there were three groups (Treatment A, Treatment B, and a control group), mothers would be matched in triplets according to parity and age and randomly assigned to Treatment A, Treatment B, or the control group. Using this technique, all the groups should have very similar age and parity structures, and it could be argued that age and parity therefore have a similar influence in each group. Of course, other factors such as maternal education, income, or social class might also influence outcome. The difficulty with including too many factors as matching variables is that it becomes increasingly difficult to find adequate matches for everyone in the study. Use the two or three factors that you think will be the most powerful influence on the outcome as the matching variables. Then measure all the other factors that you think might be associated with the outcome so that they can be taken into account in the final analyses (see Chapters 10 and 11).
Clinical trials. These involve the assessment of the effects of clinical interventions such as drugs or feeding programmes. They are usually carried out in a controlled setting (that is, where the subject will be unable to obtain other supplies of the drug or where all aspects of diet are controlled). The intervention is compared with a placebo.
The design and analysis of clinical trials is a science in itself [7], and there are a great many variations in design which can be adopted. The so‐called Rolls Royce of clinical trials, the randomized double‐blind placebo‐controlled cross‐over clinical trial (Figure 1.5), requires careful thought in its planning, implementation, and analysis.
Randomized controlled trials should not be embarked upon lightly! It is very demanding of time, staff, and money. Moreover, there are important limitations.
In recent years, it has been recognized that while clinical trials may be appropriate for providing proof of causality in some circumstances (e.g. understanding the impact of a new drug on disease outcomes, or a specific nutritional intervention), there may not always be equivalent, controlled circumstances that exist in the real world (e.g. promotion of five‐a‐day consumption in a sample versus a control group). The generalizability of findings may therefore be limited when it comes to saying whether or not a particular intervention is likely to be of benefit to individuals or the population as a whole. To address these circumstances, alternate designs and analytical approaches have been developed