Wine Spectator Ratings: Why the Kink at 90 Points in the Distribution?

Leo wrote on Sat, 14 Mar 2009 07:40:02 -0400:

It looks like the ratings might follow a normal distribution tho' they probably don't publish results of ratings like 10 or so (even as you say under 79) to avoid being sued or lose advertizing. I'm not sure the dip is significant since I don't know the statistics, including actual numbers of wines rated. Does the dip occur in other years?

Leo Bueno wrote in news: snipped-for-privacy@4ax.com:

Depending on how you see it, you can think that there are less than expected wines rated 89 points or more than expected wines rated 90 points.

Even a bit of both effects.

The fact could be that, being the rating a subjective measure, a taster would prefer to rate 90 points a wine that perhaps he is not sure if it should be 89 or 90, since 90 marks the entrance to the "great wines" category.

If you have a file with the raw data you could ask someone with an statistic software to do a K-S test to see if the data follow a normal distribution pattern. If you do not know anyone, I can do it with SPSS.

Best,

s.

I get a chuckle every time I read some referring to SPSS. 32 years ago, I got my first job in college helping to administer an old IBM 360 computer at college. The majority of users were social scientists at a neighboring college running SPSS (Statistics Package for the Social Sciences). It is amazing to me that all these long years later, researchers are still using this software package. Multivariate analysis hasn't changed much in the intervening years, though, I suppose.

Mark Lipton (A user instead of BMDP)

Hi Mark,

fortunately enough, no need to use it with the command line anymore

ANACOR(TABLE)=VAR1;VAR2;

etc.

Never used BMDP, but did use EQS many times to perfom SEM (Structural Equations Modelling)

s.

Unless I am misreading the graph, the actual numbers are on the vertical axis:

They are large numbers and I do not need a statistical test to tell me that these results are extremely unlikely to follow a normal distribution - or any other smooth distribution for that matter.

I would wager that is the correct explanation.

For the avoidance of doubt, I am saying: Yes the kink is significant - VERY significant.

(Apologies for all the follow-ups to my own posts today - I am far too used to Web forums these days - where you can go back an edit your own posts.)

Steve wrote on Sun, 15 Mar 2009 12:22:39 +0000:

I'd like to have seen the raw data as I mentioned but you confirm that the dip is probably significant. The other question in my mind remains, is this a one-off phenomenon and would the previous year's data show the same thing? Apart from the kink, I think a normal distribution would describe the published graph apart from the fact that the lower numbers are not given.

I posted the Excel data file. Hope you get chance to crunch the numbers.

Here it is. http://miamiw>If you have a file with the raw data you could ask someone with an

But it also points out the flaws in assigning such precise numerical values to an essentially subjective experience.

Also note the reluctance to give 85pts - perhaps because it sounds like a round number without making the wine much more desirable than

1. Was that noticeable in previous years?

Leo Bueno wrote in news: snipped-for-privacy@4ax.com:

Still had no time to do it. The data need some transforming which I performed at home with Excel, but I am too lazy to install the VPN- connection and the SPSS in my laptop at home. And, then, when I arrive at my office, work abduces me and I forgot. Perhaps tomorrow.

DaleW wrote in news: snipped-for-privacy@g38g2000yqd.googlegroups.com:

Or if it is the same patterns in other guides or wine writers.

You probably think of tasting for a wine magazine from a consumer point of view: "Is this really that good?". Pro tasters probably think that a 90 increases the chance of the winery to advertise in the magazine.

Oh, I also forgave to do the test today. Again. Let's see tomorrow.

s.

Finally I had the time to do some basic analysis. You can download a pdf file here:

I did some transformation to the data because I do not think SPSS can work with a frequency table so I had to build a 19001 cases table with just one column.

Just in case the problem was because of the 89-90 strange pattern, I built another set of data, exchanging the number of cases of 89 and 90.

First page is just some descriptives.

Second page shows that the data do not follow a normal distribution since the Null Hypothesis is that the variable is normal. However, if we look at the second data set, in which there is not the indent at 89 (see page 7 for a histogram), even without the dent at 89, the data does not follow a normal distribution.

Pages 5 and 9 show a Q-Q graphic, in which a point in the horizontal line means that the data goes as expected in a normal distribution. If the dot is above the line, there are more wines rated with a particular score than expected for a normal distribution.

All the best,

s.