We could add to this simplest form of repeated measures design almost infinitely by substituting further treatments consecutively. For example, we could substitute bowel retraining for bran. We could also add treatments, offering bran and bowel retraining together, once again to the same group of participants (see Figure 15.1). We could also introduce further periods of no treatment (control phases) between each active intervention.

Interrupted time series analysis

When we add several data collection points between different levels of the independent variable (treatment or control), we are using an interrupted time-series design in which statistical comparisons are made between different time points in a treatment-control sequence. Between the introduction of each new intervention, there are several time points at which measurements are taken.

Each new phase represents a further level of the independent variable, and a degree of randomisation can be included, in terms of the order in which each level is offered, greatly strengthening the design. Statistical comparisons are made between each consecutive set of measures, in much the same way as we might monitor patients’ responses to a series of different interventions in a clinical context.

There are numerous advantages to the repeated measures design. Notice that, in the above example, we do not need to add extra participants to add extra experimental conditions. This can be particularly advantageous where participants are hard to find, as in a complaint which is comparatively rare. Also, there is typically no subject variability in repeated measures designs, because the same participants take part in each experimental condition. Thus, systematic subject bias between the groups is entirely avoided. Even where subject variability is large, repeated measures designs are still helpful, because the overall number of subjects required to reduce this to an acceptable level will always be considerably less than in designs where different subjects receive different levels of the independent variable. Finally, where further measurement points are added (as in time series analysis) there may be increased control of maturation effects (uncontrolled variation over time) because of the ability to examine trends in the data and use statistical procedures to account for these. In summary, repeated measures designs are very good at accounting for subject variability because participants effectively act as their own controls.

However, control of environmental sources of bias is less good. Even though subjects receive all levels of the independent variable, they are still open to the effects of confounding differences between the different levels, and similar attempts at control must be made as in true experiments. One particular problem with repeated measures designs is the presence of practice effects and other carry-over effects between the conditions. In our constipation example, we might easily find that offering bran exerted such a powerful influence that subsequent offering of bowel retraining contributed little to the group’s bowel motion frequency. In such a case, a return to baseline might offer a ‘wash-out’ period during which the effects of the first intervention would have a chance to wear off before the next intervention was offered. However, had we offered bowel retraining first, it might prove very difficult to offer such a washout period, because it is extremely difficult to ‘unlearn’ things one has been taught. These sorts of order effect can be compelling drawbacks to repeated measures designs in other contexts, too, such as studies of perception or attitude, in which offering one experimental condition primes to participant by facilitating their performance during subsequent conditions or alerting them to the objectives of the research in ways which interfere with its purpose.

Typically, some sort of counter-balancing across the conditions is an adequate solution to many order effects, and in health research such counterbalancing is usually referred to as a crossover study. In this approach, subjects receive all levels of the independent variable, but in different orders (see Figure 15.2). Thus, any bias introduced by order is eliminated. We might further refine this by allocating subjects randomly to one or other of the counterbalanced conditions. Statistical comparisons are made between the pooled results of participants according to condition, but regardless of order. In some designs, a specific statistical adjustment may be made to investigate the impact of order effects on the pooled results. As they lack subject variability across conditions, repeated measures designs typically are more sensitive to effects of the independent variable than is the other major contender in quasi-experimental research, the independent groups (or divided groups) design.

Figure 15.2 Counterbalancing in repeated measures designs.

Independent groups designs

The true experiment is actually an independent groups design experiment. The only difference between the true experiment and other independent groups designs is that randomisation and/or control are missing. The advantage of the independent groups design over repeated measure designs is that order effects are not an issue, because each participant receives only one level of the independent variable. On the debit side, all the difficulties of subject variability in the true experiment are also present, but accentuated because of the lack of randomisation or other control. However, as we noted earlier, randomisation may be difficult to achieve in many settings. In these settings, where repeated measures studies are impractical (e.g. because issues of order effects cannot be overcome), non-randomised independent groups studies are usually the only practical alternative. Effectively, these are non-randomised control group studies, where we cannot be certain that the experimental and control groups are similar in nature.

Control group time series designs

As we saw in the description of repeated measures designs, sometimes multiple data collection points are possible between the introduction of different levels of the independent variable. Once again, the gain for researchers is the diminution of the risk of chance findings (because there are more data points, which tend to reduce variability in the data), and the ability examine and account for uncontrolled variability.

Stay updated, free articles. Join our Telegram channel

Join