Personality & Individual Differences · Simplified
Four Ways to Answer the Same Questionnaire
Four people who genuinely stand in the same place, and four rather different sets of numbers.
Step 1 of 2
First, a prediction
Four fictional people have exactly the same true standing on a trait. One answers the twenty items straightforwardly. One tends to agree with whatever an item says. One prefers the ends of the scale. One stays near the middle.
Half the items are reverse worded, so that agreeing with an item and agreeing with its opposite pull the score in opposite directions. Whose score does that reverse wording rescue?
Step 2 of 2
The same person, four ways of answering
All four rows come from one underlying set of reactions to the twenty items. The only difference between the rows is how each person turns a reaction into a number between 1 and 5.
Key terms
- Acquiescence
- Tending to agree, more or less whatever the item says.
- Extreme and midpoint responding
- Preferring the ends of the scale, or preferring the middle of it, independently of what the items say.
- Reverse wording and balanced keying
- Writing some items so that agreement means less of the trait, and then recoding them before adding up. On a balanced scale, agreeing with everything no longer produces a high score.
- Recoded score
- The score after reverse-worded answers have been flipped. It is the number that would go on the record.
Each marker is one respondent's recoded score on the 1 to 5 scale. The upright line is the score their shared true standing should have produced.
What this shows
Reverse wording fixes one problem, not three
Key idea: Agreeing with everything adds the same amount to every answer. On a balanced questionnaire that addition lands on the positively worded items and on the reverse-worded ones alike, and the recoding turns the second half of it into a subtraction, so the two cancel and the score comes out close to where it should be. Take the reverse wording away and nothing cancels: the same person now scores more than a point higher than they should, which on a five-point scale is a great deal. That is the strongest practical reason to balance a scale. The other two styles are untouched by it. Preferring the ends stretches the score away from the middle whichever way the item is worded, and preferring the middle squashes it, and recoding cannot undo either, because both distortions survive the flip. So a balanced scale is worth building and it is not a general defence against response styles.
Response styles are routinely written about as though they were failures of character or attention. They are not. Agreeing readily and using the ends of a scale both vary systematically between cultures and with education, staying near the middle may be caution, ambivalence or items that genuinely do not apply, and an apparently careless pattern can be produced by fatigue, by pain, or by a badly built form met with a screen reader. What a grid like this can show is that a pattern is present. It cannot show why, and it cannot establish anything about a person's motives. On the numbers: four invented respondents whose answers come from a formula, one shared set of underlying reactions so that the rows differ only by style, and a fixed random seed so the same grids appear every time. Real respondents differ from each other for many reasons at once, and no real questionnaire separates them this cleanly.
The longer version adds three further styles including random and straight-line responding, split-half correlations, longest-run statistics and a diagnosis exercise in which you name a style from its evidence. It is at Response-Style Simulator in the main collection.