by another_jim on Sat Sep 25, 2010 9:46 am With your help, and the efforts of new member Kendall Mills, we now have a chance to find out exactly how the particle distribution of coffee ground for espresso changes when the grinder is set finer or coarser. We can also find out if these changes are systematically different for different types of grinder.
We welcome samples from any grinder. Although only a limited number of samples will be analyzed, the more samples we get, the stronger the final sample will be, and the better the analysis. Therefore, even if your sample is not used, sending it in will improve the overall result. So please participate if you can.
The instructions for creating the samples:
If you decide to submit a sample, you must act between Monday and Wednesday, ordering a bag of Counter Culture's Aficionado from the batch roasted for this project. Please enter "TGPII" in the comment section of your order, on the final page of the checkout, so the correct roast batch is sent to you.
In addition, you will also need a grinder suitable for espresso, a 0.1 gram scale, a 1 second or better timer, a ridged La Marzocco double basket, either from the company or OEM, a pump espresso machine with a 58mm group, and a bottomless portafilter.
Please follow the preparation instructions as precisely as possible. Poorly prepped samples will show up as outliers, and generate extra work in discarding and replacing them. It may also force the use of robusticized statistical tests which would generate weaker conclusions. If you are uncertain about the quality of your samples; let me know, so we can discuss and vet them.
Prepare the samples on the 7th to 10th day post roast, Monday 4 October through Thursday 7 October.
Submit a pair of samples, one sample using a 15.0 gram dose, and the other an 18.0 gram dose. Please dose to 0.1 gram precision, so that you see 15.0 and 18.0 when weighing the samples.
You may submit multiple pairs of samples, with different pairs from different grinders or with one pair using single dose grinding and one using a filled hopper. Please makes sure all samples make up matched pairs of 15 and 18 gram doses, since unmatched samples cannot be used.
The shot from each dose must produce between 22.5 to 27.5 grams of espresso, by weight, at the allotted shot time (please try to be more precise and get as close to 25 grams as you can, but with coarsely adjusting grinders, this may be the limit of achievable precision). The shot time is computed by adding half the time before the first drop falls from the bottomless, plus all the time after the first drop falls. This combined time must equal 25 seconds. The alternative formula is easier: the shot time is 25 seconds plus half the dwell time.
Here is a suggested procedure for this rather tricky part. Get your grind level roughly right, and measure the dwell time before the first drop appears. Now compute your shot time. For instance, suppose the dwell is 5 seconds, then your shot time will be a 27.5 seconds (25 seconds plus half of the 5 second dwell). Now make the shot, cut it at the correct time, and weigh it. Adjust the grind until you get as close to 25 gram shots as you can. Please note that on some groups the dwell time will change depending on the fill level of the basket, so check it for both the 15 and 18 gram dose
Once you have the shot for each dosage level flowing correctly, you will grind and prep the submission sample exactly the same way as the shot sample. Submit the full 15.0 or 18.0 grams, and make it exactly the same way as for the pulled shot, using the same grind setting, hopper technique (single dose or filled hopper), doser thwacking, chute sweeping etc etc. It may be simplest to make the sample in the same basket you used for the shot, and then dump it into the sample bag. We want the sample to be as close as possible to what was in the basket when you pulled the money shot.
Here is the unavoidable red tape part. Place this grind sample in a sealed and labeled sandwich bag. DO NOT ASSUME YOUR BAGS WILL STAY TOGETHER; THEY WILL ALMOST CERTAINLY NOT. Tape a label with your name, the grinder brand, the machine brand, the dose, and the hopper info (single dose versus filled hopper) on EACH BAG. Anyone sending a sample that does not have a firmly taped on label, or that has incomplete data on that label, will have that sample and its matched pair discarded.
Send the samples UPS 3rd day, USPS priority mail, or equivalent to:
Notify me when you send the sample, and I will notify you when I receive it.
The final selection of samples for the project will be discussed after all the samples are received, and the final choices will be made by Kendall Mills.
The details of how the samples will be analysed:
Every sample's particle distribution will be measured by Kendall Mills on a laser diffraction counter.
The raw count data from the counter will be summarised into five parameters per sample, the proportion of coarse particles to the total, along with the mean and variance of the coarse and fines distribution. A mixture of two log-normal distributions will be used as the model to for these five parameters.
The estimated parameters from the individual samples will be combined into a data set and graphed in the parameter space. The graph will be visually explored for systematic features such as clustering (e.g. all the conicals' parameters in one clump and the flats' in another) or reduced dimensionality (e.g the data falling into a line or plane), and whether these systematic features correspond to the grinder types, size of burrs, cost of grinder etc.
The features and correlations found in this investigation of individual samples are for information only; they will not be tested statistically, since the data are biased by the effect of unmeasured confounding variables, making such testing unsound.
The parameters from each paired sample will be differenced (either using s18 - s15, or (s18 - s15)/s15 as the formula) and charted in the same way as the unpaired variables. But instead of merely visual exploration, the degree of clustering and linearity will be measured and tested for their strength and correlation to the grinders' specifications.
Due to the repeated measures form of these differenced data, the effect of the unknown confounding variables is removed, and doing statistical tests is warranted. In this type of design, if the correlations are strong enough to be readily visible on a graph, even a sample as small as 6 to 10 items can be used to reject the null hypothesis that there is no correlation of particle distribution changes to grinder type.
The conclusions we hope to reach:
Once we know exactly from the paired comparisons how different grinders work to adjust grind fineness, e.g. by mostly altering the proportions of fines to coarse particles or by mostly changing coarse particle sizes; and once we have a visual impression of how the grinders differ in their overall grind distributions, we will have enough information
-- to pose more refined questions and get more solid answers
-- to know whether we were sane or delusional when buying these expensive grinders.
Questions?
If you have specific questions about your submission, send me a private message using the button on this post. Also use a PM to notify me when you send out the sample.
If you have comments or reservations about this announcement, please post them in the TGPII discussion thread; this announcement is locked for clarity's sake. I will update the announcement myself if necessary.
Thank you
