Hacker News Reader

Comment by sneak | Hacker News Reader

It seems to be working sufficiently for Uber and their happy customers. Uber is a premium service. Drivers should be fired if they don't know their way around, have a dirty or smelly car, or otherwise request things of me as a customer beyond my destination address. I routinely give out two and three star ratings, knowing it could well get people fired. 5 stars is for a perfect trip that could not be improved.

View on HN

How do I rate an uber drive?

Say I pick up an uber, and everything goes exactly as planned. No weird side streets, no funky smell in the car, and an amicable driver. Not the best uber experience - I've had uber drivers free water bottles and nightlife recommendations at the ready in the past - but we got exactly where we wanted to go.

The service was satisfactory, but not above average. What do I rate this driver?

The answer to this question varies depending on who you are. The problem is that there is no standard for riders to objectively rate uber drivers. It's the equivalent of the easy-A teacher in high school versus the strict grading-on-a-curve teacher.

People have their own biases and experiences that affect the scale they rate on. The differences in my own personal grading syllabus versus another persons can get people on uber fired for the exact same ride experience.

If Uber has anyone statistically competent on staff, they'll normalize your scores and compute an individualized scale for you. I.e., let i = individual rider, j = individual driver. Then they'll use a model like:

    score[i,j] = a[i]z[j]+b[i] + noise

Then they store the z[j] for each driver.

So if I consistently rate drivers between 2 and 4, while you consistently rate drivers between 4 and 5, then a[me] = 3, b[me] = 2 while for you a[you]=4 and b[you]=5. If you rate a driver as 4 (i.e. z[j] = 0), then Uber will tell me the driver is rated a[me]0 + b[me] = 2.

Uber can then rate their drivers internally based on z[driver], not the individual ratings.

It is of course possible that Uber is completely statistically incompetent, but I doubt that.

Uber aren't incompetent, they just don't care, much like any other marketplace with a huge market share. Most of the time the variance in how customers rate drivers will average out. Even though they'll accidentally fire and ban a few good drivers due to random bad luck it doesn't affect Uber much since they still have plenty of other drivers. Same reason why companies like Amazon are willing to ban reliable Marketplace sellers on spurious grounds like moving into an address that someone once scammed them from years ago.

Why do you believe Uber (with hundreds of employees, and in vicious competition for drivers [1]) doesn't care enough have one employee devote a week to doing textbook linear regression?

[1] Uber is willing to pay Lyft for the privilege of having their recruiter sit in the car with the driver.

Exactly. I really don't understand all the hubbub that people raise over Uber's rating system somehow being broken because:

1. Many people, like the writer, only give our 5 stars. (Simple: their votes don't really count much.)

2. People have different scales. (These can easily be rescaled.)

Yeap. For example, Criticker (movie ratings site) simply ranks everyone on a 10-tier scale, so if you only rate movies up to 30%, a movie with a 29% rating with be on the 9th or 10th tier.

I wish the 5 stars system worked like that. I love UberX, but I don't like the negotiation at the end of the ride over ratings. It hasn't been as bad for me as in the Forbes article, but there's a definite read-between-the-lines of always giving 5 stars.

I'd much rather give my feedback/rating in private once I'm out of the cab, and for the driver to do the same with me. I really don't need/want to know how they rated me. The "I gave you 5 stars, now you rate me" conversation is always awkward.

Whoa, that's insane. I've never had a driver ask me to rate them, and I've taken a lot of ubers (though more recently, lyfts). (This is in SF.)

I have - dude asked for a five-star rating.

There is no criteria for a five-star rating. Did he get me where I needed to go at the expected price? Pretty much - that seems at best to rate average.

I don't even know what a five-star rating would be.

I gave him one because I really didn't feel the need to spend the time figuring out exactly what I should give him.

However, the problem I have, in retrospect, is this:

DON'T FRICKIN' ASK FOR A FIVE-STAR RATING WITHOUT A GOOD REASON.

They shouldn't ask, which I would explain; if they persisted, I would give a low rating for that.

In terms of stars, drivers get dropped if they are below around 4.5 on average, so 5 stars basically means "keep this driver" and anything less means "fire this driver".

I've had a driver tell me he would give me a $20 coupon for future rides, pull my phone, input the coupon code (which didn't work, since it was first riders only) and rate himself five stars.

Perth, Australia here. UberX has only been here a short time though.

What conversation? You don't have to rate the driver in their presence, and I've never once done so.

Occasionally I'll have a driver demand a 5-star rating, which of course earns them an automatic 1.

Are you saying drivers are making you rate them while you're there? I've literally never seen anyone rate a driver in the car.

Yep. This is in Perth, Australia. At least one of the UberX drivers specifically said "See, I've just given you 5 stars, now you rate me" before I left the car. UberX has only been here a month though, so maybe drivers (and riders?) aren't sure how the system is meant to work. I've had 4 rides on UberX now.

No one's stopped me from leaving the car, and the rides were all okay so I'm not complaining about any drivers. But the implication I got was of reciprocal 5 stars.

Reply on news.ycombinator.com