It's the most common interface used to display votes is the “star rating system,” in which a particular number of points (often expressed as stars) is assigned to an item by each reviewer. We find this model on many sites, from Amazon to Yelp. Stars seem like a pretty straightforward mechanism, both for your users to consume (5-star rating systems seem to be everywhere, so users aren't unfamiliar with them) and for us plan.

But 3 stars, for making a decision, I think give better insight to the consumers vs 5 stars. But what is the physcology behind the number 5 in ratings? Why not 3?

  • 2
    Well, why do you think that 3 stars give better insight? Commented Jun 28, 2012 at 19:14
  • 5
    Because sometimes you have the need for something between good(3) and excellent(5). Sometimes you need darn good(4). Commented Jun 28, 2012 at 19:17
  • 2
    no data, but a hunch is that 5 stars tend to be the 'just right' amount of granularity for most situations.
    – DA01
    Commented Jun 28, 2012 at 22:45
  • 3
    Youtube case study on why 5 star ratings don't work
    – JoJo
    Commented Jun 29, 2012 at 7:08
  • 3
    @JoJo - For youtube yes, much on youtube is love or hate, but for products it is a very different story. Commented Jun 29, 2012 at 13:14

9 Answers 9


There was a fantastic case put forward a while ago (if I find it I'll edit this answer) that the ideal number is actually 4 stars.

The idea is that people naturally gravitate towards the 3 in a 5-star system (or the 2 in a 3-star system) because it's easy. Go ahead and look at your iTunes library; if you're anything like me you have squillions of 3s.

By only providing positive or negative options and removing "neutral", you apparently get more meaningful feedback from the users.

I can't vouch for any of that mind you, having never needed to apply it, and I'm not sure if rating things out of 4 would make users feel disoriented or uncomfortable, but it's an interesting thought.

Edit: Still can't find the actual article I was referring to, but Zeldman presents a similar case here. The comments present some lively discussion both supporting and opposing the need to abolish the "neutral"/"maybe" option, including some that dispute the "most people gravitate to 3" claim he makes.

One other really terrific source for this discussion is the blog Life With Alacrity, which has an article explicitly debating the 5-star scale and another about comparative assessment of rating systems more generally

Further edit: In the interest of clarification, I'll summarise my opinions/understanding thusly: Research has been done into the distribution of optional ratings on a conventional five-star scale, and that research seems to suggest people gravitate toward higher ratings (3s, 4s and 5s), especially 4. An optional 5-star system actually permits 6 ratings; 0–5. That means in an even distribution you would expect each rating to have 17% of the whole, and in a normal distribution you'd expect the most popular ratings to be 2 and 3.

It is probably unreasonable to suggest that users will follow a normal distribution of ratings, especially if they own the content or have otherwise self-selected to like it (as in an iTunes library). It's also worth pointing out that users rarely deliberately choose "no rating"; being the default rating (and thus making other ratings optional) it's likely to be overrepresented in the set.

There is no general solution here, then, unless you force people to select a rating (i.e. exclude "zero stars" as an option), and if users are equally likely to encounter items they strongly dislike as those they strongly like (e.g. 1-star items are as likely as 5-star items).

Since that describes very few systems (one notable exception being actual newspaper movie critics, which I understand are the originators of the system), you may benefit from experimenting with non-standard systems, and especially of systems that exclude a "neutral" option entirely.

  • 15
    If you're going to do a survey without a neutral option, please make sure you have not applicable as a choice. This is especially an issue with customer satisfaction type surveys since they almost always include a number of questions that won't be relevant to all customers. Commented Jun 29, 2012 at 12:48
  • 1
    Hm, I find this pretty hard to believe honestly. In detail ratings systems people tend more toward the middle of the road, but even then I almost always see 5 used to mean "okay or better" and 1 used as "sucks". I just don't see many neutral views unless a long, thought out response is attached them them (at which point your "easy" argument stops making sense)
    – Ben Brocka
    Commented Aug 22, 2012 at 13:00
  • 2
    Kit: The sources you linked don't seem to support the idea that we see 3 as the most common rating. In fact, they suggest that we see bimodal distributions in real life, not normal ones. I think @Ben Brocka is correct on this - in my experience neutral views just aren't that common, probably due to the self-selection involved in leaving feedback.
    – kastark
    Commented Aug 22, 2012 at 13:21
  • 1
    @dhmholley yes, I agree that those sources dispute the claim I originally made. I thought I had made it clear in the edit I made but I'll look at revising it. I have left the original assertion in place because I believe there are systems where neutral ratings are effectively worthless and I found the idea interesting. As I said: I have no evidence to support that claim, and I've provided the evidence that may dispute it in the spirit of finding the truth.
    – Kit Grose
    Commented Aug 22, 2012 at 22:28

There are two problems here; you lose a lot of granularity for detailed ratings (with reviews/etc), and users of undetailed ratings (just a rating, no review) don't really use neutral votes.

Detailed ratings like Amazon's reviews really need extra granularity. By detailed I mean users are doing more than rating; they're explaining the why, they're giving a full review to accompany their rating. See Collective Choice: Rating Systems which specifically shows why 3 point systems (eBay is their example) aren't that helpful for in-depth reviews; notable they aren't granular, aren't statistically helpful, and they aren't distinct.

In contrast, 5 star rating systems are more granular (there's the "a little better than average, a little worse than average" levels), they're more distinct thanks to that extra granularity, and they more often follow a normal distribution when used with detailed ratings, as explained in more detail by Using 5-Star Rating Systems.

For the second point, just see Five Stars Dominate Ratings, a Youtube Blog post sharing their stats on rating usage. Youtube is a "shallow ratings" or undetailed ratings situation. Undetailed means there's no feedback other than the rating; no explanation of why, just a rating. Generally in this situation people either like it or they don't. Using 5 Star Ratings Systems specifically points out how undetailed ratings systems tend to follow a bimodal distribution clustered around two numbers (usually highest and lowest).

Presenting the ternary system is just asking for confusion; if your rating system really should have a shallow rating system of "I like it, it's meh, I don't like it" I'd strongly recommend ditching stars altogether and doing a thumbs up/down system that's easy to understand. Following a bimodal distribution with only 3 ratings means one of those ratings (almost always "neutral") is going to be ignored.


It's not just 3 stars vs 5 stars. It's the general grading principle: the more granulated a scale is the deeper the insight but it's also harder to decide.

The reason for the popularity of a 5-point grading system is in its simplicity. Five points is the fewest number that allows to capture the entire sentiment: very bad, bad, neutral, good, very good. Any larger scale just increases the granularity of good & bad rankings (not the extremes).

The problem with the 3-point scale is that it forces the person to vote either in favor or against something when they may not feel as radically and they aren't neutral about it. It can potentially frustrate people resulting in either incorrect votes or no votes at all.

If you want to simplify grading, then drop down a notch to a yes/no, like/dislike, upvote/downvote, etc system. This way neutrality will be signified by a hard zero when votes in favor are negated with votes against, but you'll still be faced with the problem of insufficient scaling.

  • 1
    Intuitively you'd be correct, but the page I linked above seems to contradict that in practice: user ratings aren't generally normally distributed as you might expect.
    – Kit Grose
    Commented Jun 29, 2012 at 6:05
  • I will say, too, that I think the main advantage of the 5-star system on the web is its familiarity. It's been used for much longer than the web has been around (e.g. by professional critics)
    – Kit Grose
    Commented Jun 29, 2012 at 6:06
  • @KitGrose: Yeah. I should've looked for studies first. =)
    – dnbrv
    Commented Jun 29, 2012 at 13:12

While I can't cite any direct studies I believe the breakdown of 5 stars over 3 stars is to allow for a fine grain detail when it comes to reviewing products. In a 3 star rating system user either: like/love it, think its average/ok, or hate/dislike whatever they are reviewing. With a 5 star rating system users have the ability to say something like:

"(Rating of 4) - It's a good product but I dislike this one thing about it so I don't completely love it"

These two additional levels of flexibility allow users to be more honest with their responses. With the 3 tier system, this simply can't be done unless the user wants to enter a complete review of the product. From a UX stand point providing the lowest barrier possible for users feedback is important. With a simple one click system most users are more willing to give the feedback if they believe the system allows then to honestly convey their experience with the product.

  • 1
    If one asks everyone who views a movie to give it a rating, and declining to give a rating is no easier than giving one, then finer-grained ratings may be more useful than coarse-grained ones. On the other hand, if people are encouraged to "like" all of the things they think are absolutely superb and somewhat arbitrarily also "like" some of the things they think are good even if not exceptional, then the percentage of users who bother to rate something positively will correlate with quality without individual users having to make fine distinctions.
    – supercat
    Commented Jun 10, 2014 at 17:02

The book I have on writing questionnaires (Bradburn N; Sudman S and Wansink B (2004) Asking Questions – a definitive guide to questionnaire Design. San Fransisco:Jossey-Bass) - which I have used as my definitive guide - argues strongly for 5 or 4 point scales, depending on what you are after. Although it applies to questionnaires specifically, the arguments are the same for a rating system, which is just a questionnaire with one question.

A 5 point scale has the advantage that people can express approval or disapproval, without being extreme. They can also express the extreme response if they wish to. But people tend to be reluctant to be extreme. If there were 3 stars, you would lose people who quite like it, as they might put "like" or "no feelings". As it is, you can combine the "like" an "love" into all people who have a positive response, and that should be more accurate.

This information should always, of course, be treated as extremely noisy. Making business decisions on the basis of it requires that some detailed analysis has been undertaken.

4 point scales have the same advantages, but remove the middle option, forcing people to make a positive or negative choice. This is not a good option on web rating, because a lot of people are fairly neutral about the products. Losing this means your results will only reflect people who have some strong feelings. You will lose an important balancing factor, which, as the data is noisy, is useful to have.


Users have a bias for neutrality when they can't decide. I think, if their opinion does not fit snuggly as a 1 or 3 star, they will default to 2 star

With 3 star ratings, you're capturing the absolute best, absolute worst, and the good, bad and average get thrown in together. Most apps benefit from 5 star ratings which compensate for this bias.

  • Thus even when a bimodal distribution is not a problem, on a 5-point scale the upward bias often results in only 2 or 3 meaningful data points. This is problematic because it minimizes differentiation. In many cases, a 5-star rating system where most of the ratings are either 3 or 4 is actually no better then just a thumbs-up/thumbs-down rating system. Source My problem is exactly that. The bias. The average grade for things online is about 4.3 stars out of five. How can that help me to make a better decision? Commented Jun 28, 2012 at 21:04
  • 2
    I see. I personally don't like star ratings at all. There are alternatives, depending. Ranking articles in relation to each other (selecting the superior article), when implemented properly, is much more accurate than starred ranking. Users make meaningful judgements, articles are ranked properly. Can eliminate clustering
    – Darragh
    Commented Jun 28, 2012 at 23:53
  • @Darragh: That sounds promising. Can you give an example of a public site which uses such a system? Commented Jun 29, 2012 at 20:09
  • @JonofAllTrades I've yet to see a working example. web.mit.edu/newsoffice/2011/… they apparently showcase an implementation, link is on that page
    – Darragh
    Commented Jun 29, 2012 at 22:18
  • @Darragh Isn't that just a more advanced version of up and down voting? When you up or down vote something on stackexchange or reddit as an example you are saying "this item is better or more relevant than those around it and deserves to appear higher in the list"
    – rsparis
    Commented Jul 5, 2012 at 1:23

IMHO star-system does not work good.

Example: lets assume that we have 2 "4-stared" votes for article. So it means that two users, who voted are satisfied and we have average rating - 4 stars. And after some time came new user which does not liked the article and he rates article with "1-star" vote. So now our average rating is (4+4+1)/3 = 3 stars. We have middle average rating for article. As we can see, one user can be more powerful than two others.

I think like-dislike (plus-minus) system works better. You may check it on YouTube. In our example we would have: 1 minus and 2 pluses = +1. So this means, that article liked for more than a half users. It is easy to use and very informative.

  • 1
    The like/dislike system is efficient for knowing if something is good. The star (or better yet, verbal score) system is good for knowing how good something is. Commented Jul 4, 2012 at 18:20


The like/dislike system is efficient for knowing if something is good. The star (or better yet, verbal score) system is good for knowing how good something is.

If IMDB had a like/dislike system, you could probably know if a movie is any good or not. However, you wouldn't know if a movie is great, good, average, bad or the worst thing you could every see.

Perhaps the best system would be "like/dislike" followed by a question "how much?".

For an overview on giving scores, read the following section...

If you use a scale of 1 choice then you are really giving no choice at all (even though you can count % of voters, I am not sure how effective that measure is). P.S. TV talent shows work this way.

If you use a small scale of 2 choices (like/dislike) then you don't give an option for neutral (e.g. "not good, not bad"), so users' closest option is not to vote. If users do vote for them as dislike, it causes stuff that isn't bad to look bad. For example: serial down voters on SE.

If you use a scale of 3 choices (e.g. bad, neutral, good) then people might not be sure how to rank things they liked a little - good or neutral (and the same for things they disliked a little).

If you use a scale of 4 choices, there is not neutral option for things you neither liked or disliked.

If you use a scale of 5 choices, you have bad, baddish, neutral, goodish, good, so you can differ things you liked vs things you liked only a bit (same with disliked).

It is hard for people to objectively give scores on a larger scale e.g. 7 choices or 11 choices (0...10), even though they are informative.

Take the 7 choice scale of exceptionally bad, bad, baddish, neutral, goodish, good, exceptionally good for instance, the scale is very informative, however, users may vote exceptionally good or exceptionally bad on a whim instead of contemplating on the meaning ("Is this one of the best (or worst) I have ever seen or not?"). Don't use numbers for above 5 choices, verbal choices help provide a common scale.

  • Can you expand on your evaluation of binary voting systems? Are you suggesting that people will downvote anything which is imperfect, thereby disguising the difference between "really suxors" and "needs work"? You mention SE, which I'd consider a very strong counterexample of a successful binary voting system. Commented Jun 29, 2012 at 20:08
  • @JonofAllTrades I am suggesting that in those cases users would either not vote or vote down. And yes, I have often seen good answers on SO voted down because some users didn't consider them good enough - usually because they didn't spoon feed the asker the entire code for solving, but instead gave guidelines and required that the asker actually learnt something. I think that a 5-star system could on one had provide better feedback and on the other hand reduce the amount of votes (perhaps for the better), since users would have to be a little less trigger happy (think a bit first). Commented Jun 29, 2012 at 20:30
  • @DannyVarod Check out the IMDB rating for film Dark Knight. There are 2,5% who voted with "1". How could 10-based rating solve such speculation?
    – webvitaly
    Commented Jul 4, 2012 at 18:37
  • @webvitaly (a) I did not suggest 10-based ratings, I suggested 5 or 7 based ratings. (b) I suggested using descriptions, not numbers. (c) When the number of voters is big enough, the percentage of false votes becomes a lot smaller. (d) 2.5% is neglectable in the final average - if you had an average of 7.0 and then you get false votes, so that the percentage of total votes for 1 is now 2.5%, the new average only goes down to 7*0.975+1*0.025 = 6.85 which rounded up to 6.9. Commented Jul 4, 2012 at 21:03
  • I like that approach; alternatively, I think it might be helpful to let people select a range of values: "I'd say this product deserves a 6 and at most a 9, but don't really know where it should fall between that".
    – supercat
    Commented Aug 20, 2014 at 21:24

I would like to add some technical aspects to answer your question as other answers have answered Psychological aspects.(and I know this is not what you expect, just for your knowledge)

Many websites like lets say Netflix have some thing called recommend engine, which would recommend users Movies which users would like to watch, now it also shows ratting of those movies! , but you would be amazed to know that the matrix of users who have watched and rated the movie is very sparse.

Now the toughest job is to guess the ratings (Netflix also hosted a Challenge). If you have rating scale down to 3 it becomes very difficult to guess those if you only allow ratings in integer values(or say at interval of 0.5 as most of them do). Hence they have chosen 5 for above answered technical reason, and for the psychological reasons in above answers.

  • 1
    Thanx for explaining technical aspect as well
    – Aman J
    Commented Oct 11, 2012 at 14:05

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.