Lecture
Ensuring internal validity is the main aim of planning an experiment when developing an intra-individual comparison of changes in the DV. At least two trials are required, corresponding to the levels of the IV. Identity is impossible here, since there is a single subject.
Factors of time, sequence, and task appear – the main threats of confounding (of these confounding variables) with the influence of the experimental factor.
An intra-individual experiment involves comparing changes in the DV as a function of changes in the levels of the IV, which are presented to one and the same person. That is, each subject serves as a control relative to himself.
The ultimate aim is to obtain an experimental fact (and not an artefact).
Types of schemes:
A1B1A2B2…AnBn (A and B are the levels of the IV).
In each new trial there is a successive change of conditions. The number of trials is set in advance by the experimenter.
The experimental effect here (or the main result of the action) can be represented as the difference of the averaged values of the DV in the experimental and control conditions. In addition, the significance of the results should be assessed.
This scheme is used in cases where an experiment with conditions A and B may be terminated at any moment for reasons beyond the experimenter's control (e.g. an experiment in a production setting).
Such a scheme controls changes in the DV over time.
The sequence effect here may manifest itself in that some trial gains by contrast, which distorts the magnitude of the effect.
Another threat arises from the confounding variable «the subject's knowledge», or their bias. Having grasped that the levels alternate, the subject may choose their response in advance in each trial.
Random distribution of the two levels of the experimental factor across a common sequence of trials, limited by the number n.
The control of confounding variables associated with the time factor here is based on the principle of equally probable assignment of different levels of a confounding variable to the experimental and control conditions.
The choice of the required number of trials is dictated by various reasons. For example, this is the duration of each trial. In the case of short trials, a large number of them can be used. Increasing the number of trials ensures control over the unreliability of the data.
The choice of n is also affected by the presumed magnitude of the experimental difference and the possibility of admissible confounds with the time factor.
The old way of organizing a random sequence is the use of a table of random numbers. Today this is done by a computer.
Quasi-random sequence: here there is an additional condition – the equal probability of the IV levels being represented in different parts of the overall sequence.
The total number of trials is divided into equal micro-sequences, in each of which all levels of the experimental factor are represented at random.
The forward and reverse order of the levels: ABBA and BAAB.
- variation and control of the position of the IV levels within their overall sequence, presented to a single subject.
This control can also be applied to sequences of tasks. The task factor is controlled. The time factor is controlled if the changes over time are linear in nature.
Confounds: systematic, non-systematic.
A non-systematic confound occurs when any of the confounding variables or their combinations irregularly intrude into the dependency under study. The source of a confounding variable caused by the time factor may be both internal causes (changes in the subject's states) and external ones (a chance distraction by noise, a telephone call, etc.).
One consequence of the irregular influences of confounding variables is the unreliability of the data, i.e. their variability under repeated trials. To control such confounds, one needs to bring the experiment closer to the infinite one – to increase the number of trials.
Systematic confounds may be introduced as early as the stage of developing the experimental tasks.
The sequence factor.
A symmetric sequence effect is such an influence of a preceding condition on a subsequent one in which the effect of this derived confounding variable does not change when the direction of transitions between the levels of the experimental factor changes – from A to B and from B to A.
(For example, a subject may always perform the second trial better)
An asymmetric effect is one in which the influence of the sequence of IV levels changes the direction or type of the effect (for example, the inequality of the conditions of adaptation to darkness and to light).
The regular alternation scheme is justified when it helps to control threats from non-systematic variability. A drawback of the scheme is the emergence of additional sequence effects (e.g. contrast).
The positional balancing scheme is usually used when there are few trials and the changes associated with the time factor are linear in nature. It controls symmetric sequence effects well.
There are sequence effects that are not controlled by any of these schemes – heterogeneous sequence effects.
Questions of external validity are connected not with the scheme, but with the representativeness of the methods and of the subject.
The representativeness of the subject is the extent to which they are a typical representative of the population.
Comments