> Part of the confusion stems from the fact that the rental economy — taking its cue from the Internet in general — sees everything as either horrible or great, with little room for nuance. Lyft nods to this when it tells passengers reviewing drivers that “anything lower than 5 indicates that you were somehow unhappy with the ride.” Drivers can be dropped from their services when they fall below 4.5, but it is unclear what it takes to get banned as a passenger.

I've never used Lyft, but as an Uber customer, it seems like someone at Uber graduated from college and, either as a result of, or in a sly commentary on grade inflation, created an ambiguous rating system that encourages rating-inflation, while penalizing drivers based at the whims of what riders consider "4-star" to be.

4-stars on every other rating service, is pretty solid...yet if a driver averages 50/50 on both 4s and 5s, they are sanctioned (from what I've heard from drivers, the rating-cutoff can shift depending on the area and excess of drivers...in Silicon Valley, it might be higher than in other locales, for instance)...I've never given a driver less than 5-stars because I've never received awful service (though, certainly, I've had drivers who were late, or cars that were messy)...it's bad enough that they're scraping by, I don't need to pile on by giving them a slight demerit that could end their driving. But since the star-rating is basically a binary system, why not just make the question, "Were you satisfied with your ride?" and solicit the user for comments if the answer is the rare "No"...such a system would garner real feedback and improvement, at a very small price in the sensory satisfaction of clicking on a star (this might also prevent the problem of fat-fingered/drunk users accidentally giving a low rating).

I agree completely. A binary yes/no would be a lot more useful in gauging customer satisfaction with drivers. When 5 stars is "yes" and 1-4 stars are "no," Uber forsakes the gradient system that, as far as I can tell, is the only merit of star rankings.

That said... Amazon has likewise mutated star rankings. What was the statistic, that 80% of Amazon's products have a 4-5 star ranking? For any given listing, 5 stars indicates satisfaction, 4 stars indicates likely problems, and 3 stars indicates serious defects. I've never seen 1 or 2 star reviews except on long-tail products with less than a handful of reviews. Despite turning star rankings into blunt instruments, Amazon still features probably the most valuable, extensive reviews on the internet.

I definitely agree with the binary yes/no

Also, instead of a percentage, list the good/bad ratio. 99% doesn't seem much different than 98%, but the first is 99 good experiences per bad one while the second is 49.

The great part about the Amazon reviews is that for any product you can read some of the 4 and 5 star reviews and some of the 1 and 2 start reviews. Seeing why people were satisfied and why people were dissatisfied is very valuable in determining which start ratings reflect what you care about when purchasing a product.

My Uber driver on Wednesday flat out told me he won't pick up certain racial groups, because they tend to leave lower-than-five-star ratings. It was offensive, but I understand that his > 4.5 star rating is essential to his well-being.

(This is the modern day equivalent of not being able to hail a cab if you are a person of color.)

Uber & friends will need to fix that, or their drivers are going to keep cherry-picking riders to maintain their ratings and giving a lot of customers bad experiences.

Perhaps they could simply throw out the grumps' ratings (e.g., this rider rated the driver a 2, but that's normal for this rider, so just ignore it).

What was the ethnic group he said he wouldn't pick up?

I had routinely given Uber drivers ratings between 3-5 stars (good, better, best), because service was always good enough that I didn't feel a need to dip lower. Then a driver mentioned that if a passenger gave him less than 5 stars, he would have been better off not taking the fare at all. That was pretty eye opening: that a $30+ fare wasn't worth being rated a 4, because of the future lost business that results from having a 4.8 average instead of a 4.9.

This was a few years ago, before Uber X launched. So even then, Uber had limited their scale (which provides 50 ratings points) such that only the top 5 ratings were acceptable. I think it has only tightened since then, I rarely see a driver below 4.8 at all.

Now I always give drivers a 5 unless something truly unacceptable happens, in which case I file a complaint with Uber. The ratings feedback loop has been completely broken for me.

It's ridiculous, and I don't see how they get the results they want when people have no idea what each rating number is supposed to mean. Some people start at 5 and remove points for issues. Some people start at 3 and only go up if you've gone beyond the call of duty. The only thing that makes sense is an up / down vote, when the scale can't be agreed upon by all participants.

I suppose that the driver -> passenger ratings might actually work as intended, since the drivers are presumably well aware of the reality of the ratings scale.

I don't think this is unique to Uber or car services by any stretch either. I have noticed that any service with too many ratings choices seems to suffer from poor ratings quality. Yelp is my go to example here. There should be no shame in a restaurant rated 3 stars, but in actuality it's seen as the kiss of death to a lot of places. Here again, some people are rating as if 3 stars is the baseline, some people as if anything more than 1 is earned at great cost. When you're pulling your responses from everyone in the world, I don't think you can really get consensus on a scale that goes beyond up / down.

Vast majority of humans can't comprehend how to rate any thing on more than good/bad scale. Even less people understand how you could rate some thing on 1-5 scale where whole range is in "good" category, so 1 is a somewhat good mark. One of the most obvious examples are schools. In my country before reform we had 1-5 marks. 1 and 2 were not considered valid mark but rather an absence of any score. So instead of five scale we had three scale. And since 3 was a placeholder for "passable" actual marks showing quality of the work were 4 and 5. And of course in time were invented unofficial corrections + and -, then ++ and -- etc.

So when adults with 20-30 years of supposed experience of grading quality of work can't use even a simple 1-5 system then it is obvious that general population will fail at grading even more spectacularly. And thus we have 4.5 out 5 as "awful" score :) .