Sounds like a good option
OP: Ticks should not count. Old school. JMHO. Is that what we are saying here?
We will never reach such high numbers of ticks on the website (in short term at least, and even if we are optimimistic)
| RateCount | Tick Count | Show Review Avg | Show Tick+Review Avg | Show Weighted Review Avg | Show Score | Show Style Score |
|---|---|---|---|---|---|---|
| 0 | 5 | - | x | - | - | - |
| 4 | 5 | - | x | - | - | - |
| 5 | 5 | x | - | - | - | - |
| 6 | 20 | x | - | - | - | - |
| 15 | 50 | - | - | x | x | x |
Also, we should put Ticks in a separate secondary tab than Full reviews.
Paired with a more secure profile creation process (email verification, obligatory location, having to update username from generic word1word2XX before being eligible to have public ratings), this should be ok .
viper666 mentions the lack of local ratings from Quebec members makes it harder to plan a beer trip, IMO it’s part of the problem. The other problem and we can do this here and not on UT is rating places and we are not push it enough and making it more usefull for viewers (like how accecible the place is). I’m for letting the ticks count, but more efford on place ratings would help also.
2 Likes
We also hope that our coming deployment of brewer ratings helps guide place explorers as well. Because beers turn around very fast, brewery-focused guides seem to make far more sense in today’s beer world.
How will the new Brewer ratings work for already retired brewers?
Just to keep things clear. Bayesian weighting works well, and is even more important when sample size is low.
Bayesian weighting has NOTHING to do with the 10 rating cutoff. There only reason that cutoff is there is to give some protection from rating fraud.
There are however better methods for doing that, but those may require more coding/computing effort.
But I see no reason why scores are not shown for beers with <10 ratings, even if cutoffs may be used for toplists.
2 Likes
if ticks will ever count, their average should be normalized to the ratings average because people tend to give more stars
1 Like
You are mixing a couple of concepts here. I agree that the Bayesian score is great for things like keep a beer with low ratings from suddenly getting into the Top 50, but as a user looking up a a beer its a problem.
If a beer has a low number of ratings it means I know the Bayesian weighting will lower the score. Which means to know if it’s any good I need to open the beer page and look at the individual ratings. If I just look at the score quickly and see a 3.4 score I may not bother with it, and miss a beer with a significantly higher average rating.
1 Like
Seriously, the Bayesian weighting should only matter for Awards/Top beers listings… For the rest, I would prefer the real average, with the number of ratings clearly shown associated to the score.
This way, new trending good beers and new good brewers can be recognised instantly.
2 Likes
Must have or not for me isn’t decided by average score alone. On the contrary, I read the descriptions and based on that I decide whether I must have that beer or not. A score alone doesn’t mean anything to me. That’s why ratings matter to me personally.
1 Like
Real average shown next to the weighted average is pretty cool I agree. It would also be cool to have averages of each attribute visible.
2 Likes
I do
I do that too when I have time, but if I am at a bar while traveling where I don’t recognize anything and it has 50 taps I don’t necessarily want to do that for every beer.
1 Like
Well, one of the problems with unweighted averages are that the beer that will float to the top are those where the ratings are gamed.
Until we have implemented a way to correct for that (which isn’t impossible) the raw average for beers with few ratings do more harm than good.
1 Like
Right now my solution is to search on Untappd and mentally subtract .25 to .5 to estimate what I think the average might be on Ratebeer if it had enough ratings to show me an average.
I guess the point is that for me, I pretty much never look at the “best of” lists anymore. I care much more about the beers on tap at the bar in front of me, and finding a way to prioritize which of those I want to try. Locally it doesn’t really matter. I know the breweries and can make my own assessment based on how much i like a brewer and the beer description.
When traveling I may not know the brewers so it’s more of a combination of description (does this sound like something I might personally like) and the score. Right now weighted averages mean Ratebeer weight averages may signal me to skip the great, rare, up and coming brew in favour of a just okay more widely available beer.
IMO In that scenario the weighting is more likely to hinder me than fake / inflated ratings.
1 Like
This. Plus our user best is so paranoid about fake ratings we seem to catch them pretty quick. At this point we spend more time worrying about them than actually finding them or suffering because of them.
2 Likes
Thank you to everyone who has contributed to this discussion. We’ve taken all the feedback on board and have come up with a proposal here.
https://community.ratebeer.com/t/proposal-5-ratings-and-reviews/13649
2 Likes
I recently thought about a possible solution to be used EXCLUSIVELY in beers under, say, 50 ratings.
For these beers where one would really prefer to know the “real” average, it would be possible not to weight down to the number of ratings for that beer but for two factors:
(1) How many beers the raters have rated already (avoids fraud)… and
(2) Whether the raters have tended to be overly generous or overly strict in the past (i.e. raters who always give 0.8 stars above average would not result in loads of 4.2-4.3 beers appearing everywhere)
In the case where the few raters haven’t rated too many beers and whose averages are very different to the overall scores, ONLY THEN would the beer score drift towards 2.75 as per usual. Otherwise, it would be much closer to the unweighted average, which is useful for any local beer.
The following cases would happen:
-
An “ideal” rater whose average is the RB average, with >5k beers rated, rates a beer 3.6. That beer would show a 3.6 rating. And to be honest, it’s what my gut feeling would suggest if I saw that.
-
A local beer has two ratings of 3.0 (user1: 600 ratings, -0.5 average difference) and 4.0 (user2: 45 ratings, +0.02 avg diff). This is equivalent to a predicted 3.5 (600 ratings) and 3.98 (45 ratings), resulting in a shown rating of 3.53. This could be weighted down VERY slightly proportional to the total ratings accumulated by user1+user2 (=645).
Shown rating: 3.50 (or whatever, it’s the concept that matters) -
If two raters give 5.0 stars and it’s the tenth rating for both of them, and they give on average +1.0 stars… it’s a 4.0 rating that is significantly weighted down by the tiny sum of their total ratings (=20) leading to, say, a 3.3 or less.
IMHO this really solves the issues for beers with fewer ratings.
What do you think?
1 Like
I like this idea. As in you rate the rater. High probability of fraud, aka high fraudability 
