Ask a room of researchers whether a rating scale should have four points, five, or seven, and you will get a surprisingly heated argument. It looks like a technical squabble about numbers. It is actually a disagreement about human nature — specifically, about whether "I don't really have a view" is a fact worth recording or a cop-out worth designing away.
Let us clear the mechanical part first. Five and seven points are the workhorses, and they share a feature: an odd number of options gives you a middle. Four and six are the forced-choice formats, even-numbered so there is no centre to sit on. More points buy finer discrimination and, up to a limit, better reliability — but the returns flatten out around seven, beyond which people cannot reliably tell an eight from a nine. Fewer points are faster, kinder on a phone screen, and easier to label clearly. None of that settles anything, because the real fight is not about how many points. It is about that middle one.
The case for taking the middle away
The forced-choice camp has a strong argument, and it has a name behind it: Jon Krosnick's work on satisficing. Answering a survey honestly is work, and many respondents do the least work that will get them through. A midpoint is the path of least resistance — the option you pick when you cannot be bothered to weigh the question. Remove it, the argument goes, and you force people to actually think, to lean one way or the other, and you get sharper, more decision-useful data. In a lot of customer and product work, that is exactly what you want: not a shrug, but a direction.
The case for keeping it
The other camp has an equally serious name behind it: Philip Converse and the idea of non-attitudes. A large share of people, on a large share of topics, simply do not have an opinion — and no amount of scale design creates one. If you take away the middle, those people do not suddenly develop a considered view. They flip a mental coin. You have not measured an attitude; you have manufactured one, and then you average it in with the real attitudes as if it were the same substance. A false lean is worse than an honest "neither", because the false lean looks exactly like signal.
The middle is doing two jobs
Here is the part both camps tend to skate over, and it is the heart of the matter. The midpoint is quietly doing two completely different jobs, and we pretend it is doing one. "I feel genuinely balanced about this" is a real, informative attitude — the considered centre. "I don't know, I haven't thought about it, I don't care" is the absence of an attitude. Both land on the same middle box. They mean opposite things, and your scale cannot tell them apart.
So the honest description is not that the middle is neutral. The middle is ambiguous. And that reframes the whole debate. Forcing a choice does not solve the non-attitude problem; it hides it. It converts "I don't know" into a coin-flip lean and launders noise into a clean-looking number. You bought discrimination and paid for it in truth. Keeping an odd scale does not solve the problem either, but at least it does not fabricate a direction — it just leaves the ambiguity sitting there, visible, where a careful analyst can be suspicious of it.
So what should you actually do
Stop treating this as a default and start treating it as a decision that follows the question. If your goal is genuinely to force a prioritisation — pick a side, choose between two options, commit — an even scale is defensible, because indecision is not the answer you need. If your goal is to measure how an attitude is really distributed, keep the midpoint, and seriously consider giving people a separate, off-scale way to say "no opinion" or "not applicable" so the middle of the scale can go back to meaning balanced rather than absent.
Better still, when a rating only makes sense for people who actually have experience of the thing — a product they may not have used, a colleague they may not work with, a service they may never have contacted — screen for it first. Ask a short gate question ("Have you had enough contact with X to rate it?") and show the scale only to those who say yes. Now the non-attitudes never reach the rating at all, instead of being fished out afterward, and you lose nothing: a score from someone with no basis to give it was never information in the first place. It also shrinks the survey for the people it does not apply to, which is its own small mercy. That family of moves — separating a real view from the absence of one, before or during the scale — does more for your data than any argument about four versus five.
And whatever you choose, choose it on purpose. Label the points so the middle is not left to interpretation. Keep the format consistent across a study so answers stay comparable. Remember that response styles differ — some people, and some cultures, gravitate to the middle or the extremes for reasons that have nothing to do with your question. The reflex to reach for a five-point scale because it is what everyone uses is the actual mistake here, more than any specific number.
The number of points is a small decision wearing a big decision's clothes. The big decision is what you believe a neutral answer is, and whether your instrument lets a person tell you "I honestly don't have a view" instead of guessing on their behalf. A survey that cannot tell a fence-sitter from a shrug is not capturing nuance. It is averaging two different kinds of silence and calling the result an opinion.