Wednesday 2 July 2008

Adulteration (part two)

I've decided to look more closely at the differences between the gravities recorded in brewery logs and those obtained from the analysis of pub samples. To make life easier for all of us, I've calculated the pub OG as a percentage of the log OG. To be fair, I've included all the samples I have for the periods covered.

There's a huge significant difference between the results for Whitbread Bitter and Mild. I've chosen pretty much the same years for both, 1931 to 1939 for Bitter, 1933 to 1940 for Mild. The Bitter pub samples had, on average, 96.47% of the OG in the brewing log:


Of the 19 samples, 18 were below the brewery OG, just one, every so slightly, above. Two samples were within 1% of the brewery OG, 4 between 1% and 2%.

It was a very different story for Whitbread Mild:


Of the 23 samples, 18 were above the brewery OG, 5 below. 3 were within 1% of the brewery OG, 8 between 1% and 2%. On average, the Mild samples had 102.49% of the brewery OG.

Maybe someone who understands statistics better can me if can say if that's significant.

5 comments:

Lars Marius Garshol said...

I'm not exactly a statistics expert, but I did do a stat course at university 15 years ago. So, let's see...

For the bitter samples I get a standard deviation of 1.97, and thus a 95% confidence interval of 92.6 to 100.3. So the 100 is within the interval, but only by a hair. It's extremely close to being significant.

For the mild I get an stddev of 3.6, and an interval of 95.5 to 109.5, so that's clearly not significant.

Anonymous said...

My intuition tells me it's significant, but it looks like the maths disagrees - I think the problem is the small sample size.

Also, maybe a comparison of the two sets of data would produce a more significant result - i.e. the likelihood of both the bitter and the mild figures occurring by chance must be lower than the likelihood of either one of them occuring by chance. Sadly I've no idea how to work this out... can you help, Lars?

Anonymous said...

Would they have been filtering back? Perhaps they filtered back stronger beer into the mild

Andy Holmes said...

It would also help to know something about the accuracy of the measurement and what tolerance any measuring instruments were calibrated to.

Ron Pattinson said...

Tom, you see exactly the same with the Barclay Perkins beers. Except that the Mild is even further out of whack. Not sure why that should be. Though there was a difference in the beer range of Whitbread aand Barclay Perkins. BP had a Best Mild at about 1043.

Anonymous. That's my assumption.

My next step is to harvest more details from the BP, Whitbread and Truman's brewing logs for the years covered. I have exact dates for the samples. How long between mashing and selling for Mild? A week? Two, tops. Tracking down a dozen brews that could have been the sample will be simple. If I do this for every one I have, I should get some pretty conclusive evidence.

Andy, good question about margin for error. I know that calculating OG from FG and alcohol content isn't 100% accurate. You can't expect them to nail the brewery OG perfectly every time. Though, when CAMRA first had beers analysed before the breweries made their OG's public, they were very accurate.

I've used Whitbread Double Brown, a bottled beer that can't be easily tampered with as a control. These come from the 1930-1939.

brewery 1055.2
pub 1054.9
pub 1053.6
pub 1052.2
pub 1054.37
pub 1054.2
brewery 1054.1