Why an opinion is measured on a scale, and not by yes or no

On a divisive question, a person may be firmly in agreement, broadly in agreement, torn depending on the case, or in disagreement for a precise reason. A binary format crushes these situations into two answers. The result is not only less precise: it is falsely clear-cut, since it attributes a firm position to someone who did not have one.

This crushing is not symmetrical, and that is what makes it misleading. The most nuanced people are forced to pick a side, while those who were already firm answer with no loss of information. A binary questionnaire therefore mechanically over-represents firm positions, and makes society look more polarised than it is.

The same mechanism operates in the reading of results. A percentage of “yes” suggests a homogeneous bloc, whereas it adds up firm support and lukewarm agreement that would tip at the first counter-argument. Yet it is the latter that decides the fate of a measure when it comes up for discussion.

The multi-degree scale, popularised by the psychologist Rensis Likert in the early 1930s, allows a person to say not only which way they lean, but with what force. This information about intensity is decisive: it is what distinguishes a light, revisable preference from a commitment that structures all the other answers.

It has a second, more technical advantage. By adding several graded items bearing on the same dimension, occasional hesitations cancel out instead of accumulating, and the estimated position becomes more stable. A single question, however well written, remains sensitive to mood, to recent news and to the context in which it is read.

It finally makes possible something a binary format forbids: measuring the distance between two people. Two opposite answers on a five-point scale may be close together or very far apart, and that difference is precisely what matters when one wants to understand a disagreement rather than merely record it.

The number of degrees is a trade-off. Too few, and nuance is lost; too many, and people stop distinguishing the rungs and answer at random within a zone. Five- or seven-point scales are the most common because they remain legible while retaining information.

The presence of a central point is more debated. Removing it forces everyone to take a position, which increases the apparent contrast but manufactures opinion where there was none. Keeping it leaves a comfortable way out, and attracts the answers of those who want to go quickly.

The choice depends on what one is trying to measure. When a real share of the public has no view, or rejects the alternative offered, removing the middle amounts to falsifying the measurement. When the question bears on a value everyone is necessarily confronted with, forcing the choice is easier to defend.

A scale does not prevent the tendency to choose the middle out of comfort, nor the opposite tendency to choose the extremes as an assertion of self. Some people never use the end points, others use only those, and these response styles mix with opinions in the raw result.

It obviously does not correct a badly written question. Grading a loaded wording produces a precise measurement of the question's slant, not of the person's opinion. The fineness of the scale never redeems the quality of the wording, and that is the most frequent error of improvised questionnaires.

That is why grading always comes with other precautions: items worded in both directions on the same dimension, descriptive rather than evaluative vocabulary, and concrete situations in preference to grand principles.

Knowing that intensity is being measured changes how one answers, and for the better. Using the whole scale, including its end points when one really is there, produces a more faithful profile than a series of cautious answers around the centre.

The reverse also holds: forcing oneself to take a side on a question where one genuinely hesitates does not improve the measurement, it degrades it. The middle, when it corresponds to a considered trade-off, is a full answer and not an absence of one.