Interesting statistics equation

I've been going by this

Works out very well ;)

Cheers

I like the point in the explanation that says this.

CORRECT SOLUTION: Score = Lower bound of Wilson score confidence interval for a Bernoulli parameter

Say what: ...

What this math totally ignores is what the xkcd cartoon shows, that if an item or vendor has even a few negative reviews it likely has a problem and should be avoided. I've seen vendors with 98%+ rating that when you read the negative reviews clearly indicate a customer relation problem even if you only have a 1 in 50 chance of dealing with it. I prefer 99.8%+ ratings and clean comments.

Anybody who thinks reviews by humans obey axioms of randomness must logically conclude that most people can make up good random passwords. ;)

XKCD for my money.

Cheers

Phil Hobbs

Bad example. People are not objective entities; how motivated are folks to produce good reviews vs. bad ones? Is there an incentive to produce reviews of ANY kind?

I've a friend who reviews damn near everything she does. And, amusingly, always seems to have something positive to say! I.e., what *value* do her reviews have, to me? How reflective of the actual "quality" of the product are they?

Another friend (and spouse) RECOMMENDED a repair service when we were grumbling about the various things around the house that "needed attention". Not 30 minutes later, they were commenting about all the screwups they'd experienced -- with the repair service THEY RECOMMENDED! (I.e., what value do I assign to their recommendation?)

I think most customer service organizations weight written correspondence at 100:1 or 1000:1; if you get one complaint letter, they assume there are 99 (or 999?) other folks who are just as pissed but didn't take the time to write! The next question becomes: have they opted to tolerate their dissatisfaction? Or, are they taking their business elsewhere??

I don't think you have to assume anything about an individual person's motivation to use this method, as it's inferring entire population characteristics from a sample. If this person or that hadn't left a negative or positive review for whatever reason, someone else would have, for whatever reason. That's sort of the point - a straight average weights the fact that people who have extremely negative or positive experiences are more likely to leave reviews much too heavily.

I think you do have to assume that the reviews are in some sense objective as compared to how anyone else would feel given the same service, and that the reviewers aren't outright lying. But that's true of any customer-ranking system you care to mention.

The point of the math seems to be (and I think what is reflected in that xkcd comic) is that straight averages weight both positive and negative reviews too heavily, when the sample size is small.

shows that the estimator is designed to allow for sampling error, that is to say it doesn't matter if your sample population reviews are 68% drawn from pathological jerks who love to leave negative reviews, it can still return an accurate estimate of how the entire population would feel about it.

extremely positive and extremely negative I should say

This may be a problem only with certain ethnicities, but I've seen reviews that were clearly written by a competitor (or by a friend of the owners).

"We went to this restaurant and found it dirty and expensive and the food had no taste and three of us got sick afterward. We much prefer S*** Mah*** just down the street. The owner Baljinder makes sure everyone is happy. etc.

Rule of thumb- if they name someone that's who the review is meant to benefit.

For Newegg reviews it's usually best to sort by number of reviews as well as rating and compare. For example, EVERY hard drive or SSD has some kind of horror story if they've sold enough.

--sp

Not specifically Newegg or disk drives, but reading the reviews is important. Some will rate something "I wish I could have given it zero stars because the UPS guy left the package in the bushes".