Ratings: 3 stars vs 5 stars. Why 5?

Question

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?

Answer 1

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.

Answer 2

There are two problems here; you lose a lot of granularity for detailed ratings (with reviews/ect), 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 explaination 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.

Answer 3

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.

Answer 4

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.

Answer 5

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 ot 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.

Answer 6

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(Netflex also hosted an Challenge). If you have ratting 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.

Answer 7

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.

Answer 8

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.

Answer 9

Edit:

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.