Rotation is when the items or options of a question are presented to each participant in a different (usually random) order – or even when the order of questions is varied randomly.
A rotation is often used to counter row effects. Very well known are the primacy and recency effect as well as contrast and consistency effects: Rotation is therefore used specifically when the order of questions or items is likely to significantly distort the response behavior in the overall sample. The sequence actually used in the individual case is generally not included in the evaluation during rotation.
In distinction to rotation, there is randomization, in which participants are randomly assigned to experimental groups. There, too, the order of questions can be varied – but then this order is usually highly relevant for the evaluation.
Tip: To understand the more complex code examples (below) read the captures Introduction in PHP and Introduction Arrays.
Attention: Rotation does not always make sense. Rotated items can also cause problematic measurement errors.
Useful rotation is when you want to extrapolate the aggregate to the population. So if you want to compare, for example, which option gets the most crosses in a multiple choice or which item is rated highest in a scale. Then, statistically, biases due to the series placement effects in the aggregate are likely to cancel out.
Not useful is the rotation if you want to compare the participants among themselves. This is the case with almost all types of correlation, e.g., with the question whether the satisfaction with a television genre (surveyed in a questionnaire/scale together with other items) is indicative of the duration of television viewing. If we rotate the items of the scale here, some participants seem to be more satisfied with the television program only due to series placement effects. This means that the rotation introduces a measurement error.
The measurement error due to rotation decreases with the number of items used to measure a construct. In the BFI-S personality scale with 3 items per personality factor, the measurement error is smaller than in the above example with one item. If you want to calculate correlations as well as compare the items with each other, you have to weigh the advantages and disadvantages. Before using the rotation, you should only pause for a moment and think about your goal.
Depending on what you want to rotate, the procedure for creating a rotation in the questionnaire looks slightly different:
Note: IIn distinction to rotation there is also the Randomization. Here, the order of the stimuli is not randomly determined, but only some are randomly selected from a number of stimuli and presented.