Subject: blowoff/trub, Student's t, experiments
Date: 1992-02-25 19:40:10 GMT
Subject: Blowoff/Trub, Student's t, Experiments
Mike and Al have been discussing blowoff versus trub removal (HBD 829, 830)
and have proposed experiments to identify what effect (if any) various
combinations of blowoff/trub removal might have. Al's proposed experiment
seems especially appealling. But before you begin boiling wort, take a
moment and consider how to run and evaluate the experiment.
I've read many accounts of "A vs B" experiments in the digest. Some brewers
make an experiment out of every batch, and have done so for 20 years!! I fear
much of this effort is wasted by failing to recognize and allow for the varia-
tion inherent in the brewing process. Will the changes you taste and smell
be due to the amount of trub left over, the amount of beer blown off, or
purely chance? And how can you find out without repeating the experiment
many, many times?
This was the type of problem faced by W.S. Gosset (a.k.a. Student) when
he was working for Guinness. He was trying to correlate the behavior of beer
produced at his experimental brewery with such things as the analysis of
malt and hops, and brewing and storage temperatures. These were the
days (1900s) when unsuccessful brews had to be drained into the Liffey.
The analysis of experiments was so important to Guinness that Gosset
spent a year in London studying under the great statistician Karl Pearson.
Here Gosset invented Student's t statistic (Guinness refused to let him
publish under his own name).
Student's t led to Gosset corresponding with R.A. Fisher, who, inspired
by Gosset (he also helped Fisher get his job at the Rothhamsted Experimental
Station) went on to invent the design and analysis of experiments.
Gosset was an influence on Fisher in ways most digest subscribers can
understand: Fisher asked for guidance from Gosset about home brewing and
what computing machine to buy. Gosset's home brewing advice is a
little disappointing to me though: "less trouble to buy Guinness and let
us do the work for you".
Although it would be desirable to run the blowoff/trub experiments with one
batch of wort, there are ways (invented by Fisher) of using two or more
batches and getting a meaningful result.
Any influence the trub or blowoff has on the beer will much easier to
evaluate if your brewing process is in statistical control. Large
uncontrolled factors in the process may overpower changes due to blowoff/trub.
How much variation is there from one batch to another? How much variation
is there within a batch? I seem to get bottle to bottle variation in
my beer (and the differences aren't always bad). If you know how much
variation occurs naturally in your brewing process, you'll be able to judge
if blowoff/trub removal makes a significant difference in your beer.
Al's experiment is a good one since it takes into account not only the
influence of blowoff and trub removal independently but the interaction
of both blowoff and trub removal. This is one of the advantages a designed
experiment has over "one factor at a time" experiments. Another major
advantage of designed experiments is that MANY more factors can be tested
with a modest number of runs. For example, the main effects of seven
factors can be evaluated with only eight experimental runs.
Mike, no matter what the experiment shows, the worst that can happen is
you'll end up with lots of beer. Contact me if you need someone to
help drink it.
The posts that comprise the Homebrew Digest Searchable Archive remain the
property of their authors.
This search system is copyright © 2008 Scott Alfter; all rights reserved.