NBS BIF #15 Hints and Hauls: Was it Over When the Canadians Bombed BIF #14? HELL NO!

NO I HAVE NOT!

..... you weren't talking to me ...



::quietly backs out of the room::

:: ::electronically, somehow, on a forum:: ::

 

This is awesome! I don't aim for 0.0% rDev, but I do consider it a point of accuracy when I'm not significantly "off" - by style, by ratings, and by consensus.

Granted, there's another factor at play here: many of us are frequently the only reviewer for a beer, as we often drink local. Therefore: "free" 0.0% rDev. Perhaps that balances out over time, as only a second reviewer has the potential to dramatically skew that rDev immediately (in a way that (s)he could not if reviewing a beer like Heady or Bud Light).

But just as a sample point, on my last 50 beers alone, I have 10 clocking in at 0.0% rDev. Of these, a whopping eight have been neither reviewed nor rated by anyone else, so the 0.0% is automatic. (I have another 8 out of those 50 with an rDev of less than 1% (+/-), so that's 10 beers with reviews where I'm under 1% - but that means that 45% of my <1% rDev reviews, out of my most recent 50 reviews, may or may not have anything to do with perceived accuracy.)

Also worth noting: Ian tends to review very thoughtfully, including styles that many others simply do not like. A damn good malt liquor is probably going to get a very high rating from him, because it's brewed well to style, even if the BA community at large is going to trash it. Hence: very high +rDev. Just something I've noticed anecdotally (primarily via NBS/NBW and WBAYDN?).

Since this is summed, it's not quite as informative as if it showed the individual trend over time - i.e., displaying whether a reviewer is getting more negative or positive over time, instead of the summed effect. You could interpret the derivative to make that assumption (if the trend is linear, then your rDev is roughly constant) ... if and only if the rate of reviews is constant (say, ~15 beers a month every month), or close enough to make that assumption.*

Without knowing that second data point, though, the rDev slope (BIF #16: Graphing the rDev Slope?) could be influenced simply by reviewing more beers during a particular month. Still, over time, the slope should ease, as the influence of one review on the whole should be less, so I think that means that you can interpret, say, Terry's slope as to mean he scores below average, and he's scoring farther below average as time goes by (to the surprise of no one, he's getting more negative as he ages ). And Ian, most likely, is rating beers roughly the same above average as he was a couple of years ago - he found his sweet spot, numerically - and hence only a small rise since then.

So, now I want to see a chart plotting every single beer rDev over time. It's only, what, 50,203 unique data points.

*Even then, it would tend to tail off and "curve" back toward the mean; it wouldn't technically be linear, but decelerating in the same direction (negative with positive acceleration, or vice versa) - unless of course your reviews over time are indeed skewing more negative or positive (farther from the mean). BUT THEN you have to take into account that rDev by definition requires the input of more users than just us, so the mean, itself, incorporates a constantly growing subset of data, against which your own personal review is balanced - so while your number of reviews grows over time, impacting your rDev slope, so do the total number of reviews from other BAs! BUT THEN BUT THEN, as I conjectured above, we in this group are often one of a small handful of a beer's reviews, which would tend to insulate our rDev slopes somewhat from the rest of BA (reviewing Bell's will impact your rDev slope much less over time than some obscure brewery in Kiribati, which will impact your rDev slope more).

More work is necessary on this thought-provoking chart. I may need a beer.

 

Oh what a tangled web we weave
When first we practice to do math