This podcast is brought to you with the support of
The process for brewing coffee appears simple: One pours hot water over some coffee grounds, and then drinks the liquid that passes through a filter. This superficial perspective, however, belies a sequence of complicated physical and chemical processes that govern the quality of the resulting beverage.
In today’s lecture, Professor Bill Ristenpart discusses the origins of the “Coffee Brewing Control Chart” widely used to interpret the quality of drip brew coffee and how several implicit assumptions in the derivation of the chart yield questionable interpretations in current practice. Also discussed are several unanswered questions regarding drip coffee brewing that are the subject of ongoing sustained research efforts at the UC Davis Coffee Center.
Special Thanks to Softengine Coffee One, Powered by SAP
This episode of the Expo 2019 Lectures podcast is supported by Softengine Coffee One, Powered by SAP. Built upon SAP’s business-leading Enterprise Resource Planning solution, Softengine Coffee One is designed specifically to quickly and easily take your small-to-medium coffee company working at any point along the coffee chain to the next level of success. Learn more about Softengine Coffee One at softengine.com, with special pricing available for SCA Members. Softengine: the most intelligent way to grow your business.
Table of Contents
2:50 An overview of the academic work taking place at the UC Davis Coffee Centre and Bill’s personal journey in coffee
15:00 The results of Bill Ristenpart’s study of flat-bottomed baskets vs semi-conical baskets using discrimination testing
46:20 The results of the same study that used a sensory descriptive analysis framework and consumer preference testing
57:25 Audience questions
Full Episode Transcript
Heather Ward: Hello everyone! I’m Heather Ward, the SCA’s Senior Director of Content Strategy, and you’re listening to the SCA Podcast. Today’s episode is part of our Expo Lecture Series, dedicated to showcasing a curated selection of the extensive live lectures offered at our Specialty Coffee Expo. Check out the show notes for relevant links and a full transcript of today’s lecture.
This episode of the Expo 2019 Lectures podcast is supported by Softengine Coffee One, Powered by SAP. Built upon SAP’s business-leading Enterprise Resource Planning solution, Softengine Coffee One is designed to quickly and easily take your small-to-medium coffee company working at any point along the coffee chain to the next level of success. Learn more about Softengine Coffee One at softengine.com, with special pricing available for SCA Members. Softengine: the most intelligent way to grow your business.
The episode you’re about to hear was recorded live at the 2019 Specialty Coffee Expo in Boston. Don’t miss next year’s lecture series in Portland – find us on social media or sign up for our monthly newsletter to keep up-to-date with all our announcements, including ways to get involved in next year’s Expo and early-bird ticket release!
The process for brewing coffee appears simple: One pours hot water over some coffee grounds, and then drinks the liquid that passes through a filter. This superficial perspective, however, contradicts a sequence of complicated physical and chemical processes that govern the quality of the resulting beverage.
In today’s lecture, Professor Bill Ristenpart discusses the origins of the “Coffee Brewing Control Chart” widely used to interpret the quality of drip brew coffee and how several implicit assumptions in the origin of the chart yield questionable interpretations in current practice. Also discussed are several unanswered questions regarding drip coffee brewing that are the subject of ongoing sustained research efforts at the UC Davis Coffee Center.
Bill Ristenpart is a Professor of Chemical Engineering and the founding director of the Coffee Center at the University of California Davis. He received his Ph.D. from Princeton University and his postdoctoral research at Harvard University. In 2012, Prof. Ristenpart co-developed ECH 1, “The Design of Coffee,” which is now the most popular elective general education course on campus, taught to almost 2000 students per year.
Also, I will jump in occasionally to help you follow along.
2:50 An overview of the academic work taking place at the UC Davis Coffee Centre and Bill’s personal journey in coffee
Bill Ristenpart: Thank you, Peter and thank you. Thank you for showing up at 9am for this lecture. And so, what I will be talking about today is kind of encapsulated in this title here. I will be focusing on drip coffee brewing and really, specifically focusing and zooming in on one particular aspect of that which is how does the geometry of the brew basket affect matters? But first, because I’m from the coffee center, not many people are familiar with the coffee center. I thought first I’d break this talk into two parts. I’d spend a few minutes just talking about what is the coffee center. I’ll talk a little bit about what we’re trying to do, what we’re trying to accomplish at UC Davis with academic research folks on coffee and then, as promised, I’ll dive deeply into what happens when you brew coffee and what you change. Is this the geometry of the brew basket?
So, part of one, let’s talk a little bit about the coffee center. Okay, so first of all, if you’re not familiar, UC Davis is located in California. It’s about an hour’s drive from San Francisco. It’s very close to Napa Valley, where a lot of good wines come from. It’s pretty big campus, almost 30,000 undergraduates,100 different majors. This is what it looks like. Lots of bicycles around, many international students. There are many famous University California campuses, everybody’s heard of Berkeley, UCLA. UC Davis is the only one that really focuses on food science and on agriculture and so in that regard, it’s pretty famous for its programs, especially in wine.
So, here are some pictures of something called the Robert Mondavi Institute for Wine and Food Science and so it’s a brand new facility costs about US$100 million to build, based on the generosity of Mr. Mondavi, who is a pretty famous vintner in Napa Valley and so are there are vineyards, there’s a pilot winery, pilot brewery, lots of really cool infrastructure. I’d like to show one of these buildings right here is the pilot winery, and so it’s actually the world’s first LEED platinum-certified food production facility, and so that LEED Platinum is a very high standard for energy, sustainability and efficiency. There’s a winery. There’s the Anheuser Busch Brewery. There’s a California tomato food processing facility. The facilities here are used by hundreds of students, dozens of faculty. They teach it at the undergraduate/graduate level. Really busy place for wine and food science and what I’d like to point out for coffee audiences is that precisely 0% of this infrastructure was dedicated to coffee. That is none of it.
So, why is it how can we build something 10 years ago, not have any attention paid to coffee? The reasons for that are historical.as you guys I’m sure are aware. Coffee is not grown in the United States to any significant extent and so the Department of Viticulture and Enology dates back to the founding of the University of California back in the 1880s, and it’s literally written into the California State Constitution that there will be a university that will study things like wine and grapes. There was no Napa Valley of coffee. The coffee’s not grown there. There was no agricultural impetus, no congressional or government impetus and without that government support, there was no coffee academics, no coffee professors, no coffee science. And that’s not just UC, that’s basically everywhere in the United States. Historically, there is just no coffee academia. So, Peter mentioned I’m a chemical engineer and so a lot of people ask me what does chemical engineering have to do with coffee? Why are you talking about it? So, let me give just a couple minutes about how I got into it. My origin story in coffee.
So, a few years ago, my colleague, Professor Tonyaa Kuhl and I had an idea. We wanted to use coffee to teach chemical engineering.
Heather Ward: Bill has two pictures on screen. The left picture is a pile of green coffee beans and the right picture is cup of brewed coffee. An arrow points from the green beans to the cup of coffee and there’s a question mark in it.
Bill Ristenpart: So, what do I mean by that? So, this is actually a slide we used from our lectures. You guys all recognize this. We have green coffee beans and then you do something, you do something, and it turns into a cup of coffee and chemical engineers are in here. What do we do? We design ways to convert raw materials into some type of more valuable products and if you think about what do you need to know to do this stuff in that arrow in the question mark. These are the type of things we teach in our curriculum. We talk about transport phenomena which means how does heat move from here to there. How do fluids move from here to there. How do molecules like caffeine or other molecules move from a solid phase to a liquid phase. Thermodynamics, that’s heat and its relationship to energy and work. And then what makes chemical engineers distinct is the focus on chemical reactions. When you guys were roasting coffee or when you’re brewing coffee, you might not think about it, but you’re doing a tremendous amount of chemistry.
So, the key point here is that all these things are really crucial for understanding coffee and so Tonya and I developed a class. It is called the ECH 1, the Design of Coffee. It’s one hour of lecture per week, two hours of hands-on lab activities where the students get to roast coffee on little benchtop roasters. We get to brew coffee measure the pH versus time, teach core scientific and engineering principles using coffee as a working example. and so, we do like I mentioned a whole bunch of things. We talk about conservation of mass, measure pH, talk about chemical kinetics and we do all these experiments and analysis and then we switch to design where the students work in teams, and it’s a really fun tasting competition where with the students working groups to make a liter of the best-tasting coffee using the least amount of electrical energy. So, you guys know how difficult it is to make good tasting coffee. Imagine if you try to make good tasting in coffee while also minimizing your energy usage. So, it’s a classic optimization problem, but it’s one that’s fun.
Heather Ward: Bill is showing a chart titled “number of students per academic year.” The point to note is that the introductory beer course, the blue line, had over 1200 students in 2011 but attendance dropped by 200 students in 2015. Coffee on the other hand, the red bar, went from 18 in 2012 to over 1500 by 2015.
Bill Ristenpart: Just to say a little about the class, this is the number of academic students per year taking a couple of classes at UC Davis. The Blue bars here are the beer introductory course, and you can draw your own conclusion about the trend there. The red is coffee and so we started with 18 students in 2012 and now we’re with all the different varieties of the class. Almost 2000 students per year take the coffee class at UC Davis and this is my absolute favorite slide. Here’s the student newspaper.
Heather Ward: Bill is showing a clip out of the student newspaper at UC Davis.
Bill Ristenpart: Every year they have a vote, the students do a survey and a couple of years ago, the best general education course the students voted, number one was Design of Coffee, number two was Introduction to Human Sexuality, number three was Introduction to Beer Brewing. So, at UC Davis coffee is better than beer and sex.
So that’s how I got into coffee. So, it was really a teaching exercise, but very early on as we started teaching it became very quickly apparent that there was a lot of unmet needs in the coffee industry, both for research and for education focused on coffee which historically there hadn’t been so early on we started developing collaborations with the Specialty Coffee Association, the National Coffee Association, the thought leaders in the coffee industry and so this is a little timeline on what’s going on. That’s when I rolled out our freshman class. We started getting some faculty together to think about things. We renovated our undergraduate coffee lab in 2015. In 2016 Pete’s Coffee gave a very generous founding gift to found UC Davis Coffee Center and Since then, I’ve been spending a lot of time working with different thought leaders and companies in the coffee industry to build up the coffee center. So, what’s the excitement about? One of the most exciting things is that there’s a whole building that suddenly became available on campus. So, this building the previous occupants moved out and administration gave us the green light and said, hey, you can have this building if you raise some funds, raise some support from the coffee industry. And so that’s what we’ve been doing and the main idea for this building is to have everything you need for coffee science and advanced education under one roof.
So, there’s going to be an experimental green beans storage facility. There’s going to be a pilot blustery. There’s going to be a brewing espresso laboratory, a dedicated sensory descriptive laboratory, chemistry laboratory, innovation, space classroom space, office space, outside greenhouse space, everything you need for advanced research and education focused specifically on coffee. So, as part of this, I’ve been spending the past few months doing a tremendous amount of meeting with architects, and so we’re in the final stages of architecture. So, you can see here, here’s an example of the floor plan from above.
Heather Ward: Bill is showing the floorpan of the Coffee Centre, with different rooms with names such as ‘Advanced Coffee Analytical Laboratory’, ‘Pilot cold brew and advanced packaging facility’, ‘Sensory and cupping laboratory’, ‘Peet’s coffee pilot roastery’, and others.
Bill Ristenpart: Here’s some renderings, we’re putting a lot of work into making this a beautiful space where people from industry, you guys, you’re all invited to come visit Davis to come see this when it’s built. We want to have it be a place where it can hold not only education events and do research, but also hold receptions, social activities and things like that. So, here’s a rendering from outside looking in. You can see roastery here. The architects don’t have any stock photos of coffee so this is corn in there rendering but I promise we’ll have coffee. Here’s the sensory and descriptive lobby, a couple of traditional cupping tables and more importantly, well, some sensory isolation description booths, which I’ll talk about more in this talk. So, we’re really excited about this as we as we’re moving forward. One thing I should emphasize is that it’s not just me and one of the great things about UC Davis there’s more than 2000 faculty on campus. There’s a subset here’s about 40 or so different professors from all these different departments around campus who have interest in expertise that pertains to coffee.
So, you can see everything from plant genetics. Juan Medrano is one of the guys who helped sequence the genome of cafe arabica. Plant science, obviously microbiology, all the ways around to this being California, we have somebody in law school involved all the way down to sensory science. So, there’s quite a few faculty involved and there’s lots of cool things going on. So, I think many of you have seen this coffee tasters flavor wheel.
Heather Ward: Bill is showing the cover of the Journal of Food Science. The cover page shows the coffee taster’s flavor wheel.
Bill Ristenpart: I think less well known is that one of the main contributions to this was by a graduate student. That’s Molly Spencer, who was at UC Davis. She did the statistical analysis that gave rise to the precise positioning of all the different terms in the lexicon in the coffee taster’s flavor wheel. She published that on the cover of the Journal of Food Science, which, if you’re a graduate student in food science, that’s like the Holy Grail, that’s really awesome.
There’s lots of other cool things going on. One of my colleagues, Daniela Barile, is a food chemist. She just published some really nice work on oligosaccharides in coffee. That’s a fancy word for sugars. So, things that lead sweetness to coffee. I mentioned this work of the Coffee Genome Project is being led by Juan Medrano and his colleagues. Linda Harris is a food safety microbiologist, and so you guys all know cold brews, exploding lots of concerns about food safety in cold brew. If you have sitting on the shelf for a while, things like salmonella and listeria and e-coli can start growing. Understanding how storage conditions and treatment conditions affect that as well. And then here’s my name. There’s some really cool stuff going on right now that I’m doing a collaboration with Jean-Xavier Guinard on the Coffee Brewing Control Chart. We’ll see more about that in a second.
Oh, and something that’s very exciting right now is we’re actually hiring our very first staff position for the coffee center. We delayed the closing date for receiving applications until after this conference. So, if you know of somebody or if you’re interested in coming and being the head roaster at UC Davis and teaching students how to roast and being in charge of the pilot roastery, check it out, please apply and that’s supported in part by a generous philanthropic donation by Probat but it’s a UC Davis employee and we’re taking applications right now. So please let your friends know.
As part of this already mentioned some other cool talks besides what I’m talking about today. Dr. Scott Frost is a Post-Doc. He’s talking in detail about the Coffee Brewing Control Chart. Lots of fascinating data that’s later today, this morning 11:30. Mackenzie Batali is talking tomorrow morning. She’s fractionating coffee and doing detailed sensory descriptive analysis of it and relating it to the chemistry. That’s tomorrow morning. And then one of my colleagues in the sociology department, so, on the other side of campus, from engineering, David Kyle, he’s doing also another talk tomorrow morning with his students talking about one of the classes he’s developed at UC Davis focusing on cultural and historical and sociological aspects of coffee consumption. So, if you really into that stuff, I recommend you check that out.
15:00 The results of Bill Ristenpart’s study of flat bottomed baskets vs semi-conical baskets using discrimination testing
Bill Ristenpart: So, that was just a few minutes talking about the whirlwind introduction to UC Davis and the coffee center. Lots of cool, I think and exciting things going on. Let’s get into coffee brewing, coffee extraction and so very specifically, let’s talk about what happens during brewing when you change the geometry.
First, a little bit historical perspective. In the world of coffee science, this is a pretty important name. There’s a guy named Ernest Earl Lockhart, really fascinating story. This guy was one of the early explorers of Antarctica. So, there he is wearing his parka and whatnot and then he went to MIT. He was a biochemist, started studying coffee, and 20 years later he looked like this.
Heather Ward: Bill has two black and white photographs of Earnest Earl Lockhart side by side. The left is of Earnest with a long beard wearing a 1940s arctic snow jacket. The second of him, clean-shaven, wearing glasses and a suit and tie in the 1960s.
Bill Ristenpart: hopefully, if you study coffee for too long, that’s not what happened to me. I don’t know. So, what did he do? That’s a nice history. On the stage, there’s a nice article about it. What did he do? He was the director of something called the Coffee Brewing Institute, and this is back in the fifties, back when they’re very interested in understanding how to make instant coffee, soluble coffee, things like that and he wrote what’s really kind of a seminal paper. It’s called “the Soluble Solids in Beverage Coffee as an Index to Cup Quality” and he was trying to do something very challenging. He was trying to assign a single number, an index, a single number to assess or judge “the quality of a cup of coffee and highlighted here is a statement that is as true in 2019 as it was half a century ago. The quality or acceptability of coffee, beverage or any other food product is very difficult to describe or measure.” Very true. Very true. But he had a pretty cool idea. He wanted to link something about the soluble solids.
So, what does that mean? This in terms of the implications it’s now encaptured in the Coffee Brewing Handbook. So many of you might know Ted Lingle. There’s a picture of him in the 1990s holding up an early version of the Coffee Tasters Flavor Wheel. Ted went and compiled all the stuff that Lockhart and the Coffee Brewing Institute did and put it together in The Brewing Handbook, which my understanding is still the number one selling book that the SCA publishes. So, what’s in the Brewing Handbook? Well, the main feature of it is that something called the Coffee Brewing Control Chart.
Heather Ward: Right now, we recommend pausing this podcast and googling an image of the Coffee Brewing Control Chart for this next section before hitting play again.
Bill Ristenpart: So, if you have had the opportunity to take some of the training that Mercier and others offer, you’ve probably learned about this. But just very briefly, what does it show? On the vertical axis is the soluble concentration otherwise known as the strength or the total dissolved solids. It’s literally measuring how much stuff is in your cup of coffee. How strong is it? That’s the vertical axis and the horizontal axis is extraction, and that refers to how much mass, how much of the molecules did you rip out of the solid phase of the coffee grounds into the liquid phase. That’s the horizontal axis. The diagonal lines are the brew ratio. How much water to coffee grounds, what maceration did you use? And so, these two charts are the same.
This is the kind of like published version. This is exactly the same thing. My version will be easier to read, and the key thing here is that even though this has been kind of taught over the past half-century, it has several technical and sensory shortcomings. So, I won’t go into great detail about it. Dr. Frost is talking about it in much more detail at 11:30. Very briefly one of the problems is that this chart suggests that there’s a big difference between, for example, 17.9% and 18.1% extraction and so that’s not true.
It’s not true that you can basically suggest that there’s a huge difference at a critical extraction value, not necessarily true. Another problem is that it conflates sensory descriptive attributes. What does it taste like with hedonic or preference judgments, like ideal?
So, Lockhart and his colleagues put forth the idea that there’s this ideal range right here and that’s not necessarily true. Some consumers like it here, some don’t. In modern sensory science, you don’t conflate those two things. You don’t mix together sensory and hedonic judgments. And so and this is not an observation originated by us. I think a lot of the thought leaders in the coffee industry have realized this for a while and so we’re very delighted to partner with the Specialty Coffee Association starting a couple years ago to tackle this problem, to do some research and to update and expand the Coffee Brewing Control Chart. So, the Specialty Coffee Association provided the funding with underwriting from Breville Corporation, which makes these precision burrs. And so, there’s me, the people doing it. Here is Professor Jean-Xavier Guinard. So, I’m an engineer. He is a food scientist, and he has tremendous expertise and sensory and consumer science. And then the people actually who really do the work. Scott Frost. and here’s McKenzie Batali who’s now a Ph.D. student at Davis.
So, this is a very big project. We’re doing lots of work over the entire range of the Coffee Brewing Control Chart. Right here for the purpose of this talk, I want to focus on one aspect of what we did, which is again, this idea I’ve been teasing here: what’s the difference, what happens when you use a different shape geometry to brew your coffee.
Heather Ward: Bill’s slide is titled “what is “better” for drip brewing, flat bottom or conical?”. The slide has two diagrams – one of a flat bottom basket with square sides and one of a semi-conical, triangular basket.
Bill Ristenpart: So, you have a filter basket. The water comes in, it drips through and a very simple question asked which one’s better? What’s better? Should we use a flat or should we use a conical? So easy to ask, not so easy to answer. But first, I’m going to say, why does this matter? I mean of all the things you study, why this? Well, it matters a lot if you’re, for example, a manufacturer of a drip coffee brewer. A large fraction, something like half of coffee drinkers still, they don’t go to expensive cafes. They go and brew coffee at home and so what you’re looking at here, apologies is hard to read. This is a plot of the price of different dip brewers versus manufacturer and each of these points is an individual point.
I had a statistics undergraduate go and do some work, and she went and scraped the Internet for pricing information. She’s giving a poster outside and so if you want to see some details about this. Here, the prices range from about US$20 up to US$300 and a key thing here and what I really want to focus on here is looking at the different types of shapes and so each of these points here is a different model of Brewer, and we differentiated it based on the shape. So, some are pear cones, these are very few. The vast majority are flat bottom baskets, and there’s quite a few that are semi-conical, semi-conical brew baskets and the vast majority of the ones crowded around here, the more inexpensive ones are flat and there’s a pretty big variation in price between the ones that are semi-conical. So, why is this? How come half the brewers are flat? How come half are conical? What does it go and why is there such a preference for semi-conical at the very high end of the price range?
Well, presumably it has something to do with how the geometry affects the taste of your coffee. So, here’s the questions that I wanted to ask: very specifically, how does this geometry affect the extraction, and how does it affect the flavor profile? So, how are we going to answer that? Well, here’s the main idea, a very simple idea. We’re going to change only the basket shape and then see what happens. So, Breville not only providing underwriting, but they have this very convenient brewer. Convenient in the sense of it has a nice little brew basket that you can insert, a semi-conical brew basket. So, you can easily swap out with one brewer, whether it is flat or semi-conical and what we want to do with this work is hold a whole bunch of things constant. We’re going to have the water composition constant. Many of you guys know about the PPM and the specific water chemistry. We hold that constant, so we use the same water. We hold the same feed temperature. What’s the temperature of the water coming and have that constant. The brew ratio, how much water total to coffee grounds. The feed flow rate, the filter type, all those things. Just keep everything locked down. And what do we change? The independent variable is the basket geometry.
So, we just swap that out and then because we want to make sure that whatever results we find don’t pertain only to one certain type of coffee or certain grind size, we repeated the experiments with a few different grind sizes and coffees at different roast levels.
So, when we set things up like that, that’s the independent variable. What are the dependent variables? That’s what the consequence is. Well, it’s whatever you get. So, you change what total dissolved solids you have, the strength of your coffee. You’re going to change your effective extraction. The mass flow rate. How quickly it comes out is going to change. The brew temperature might change, and the sensory profile might change. So, those are all the things we’re going to measure as a consequence of changing the basket geometry.
Just to show what we mean by a different grind size distributions, we have a nice little Mahlkonig grinder.
Heather Ward: Bill is showing a photo of a Mahlkonig Kenia coffee grinder. Next to it are six grind size distribution curves. They show the grind size distribution curves for fine, medium and coarse grinds using two types of coffee: a light roast and a dark roast.
Bill Ristenpart: We used for the purposes of study three different grind sizes. If you guys have this model the numbers are three, four and five. For the purpose of this study, just internally we called it fine, medium and course. That’s with respect to each other. I’m not trying to imply any type of industry-standard there, but this is finer. So, here at this grind setting, the median size about 800 microns. So, what you’re looking at here are particle size distribution. So, here’s the count of particles versus size. So, this is 800 medians. Grind four is about 1000 microns. Grind five is about 1200 microns. So, these air not insignificant changes in grind size. This is a 25% increase in grind size. This is a 50% increase in grind size.
Then we did it for a couple different roasts. The light quality in here is not very good, but you can see this is kind of a little bit lighter and this is a little bit darker. So, when we say light and dark roast, I know that people have firm opinions about what a light roast is, maybe in your mind substitute lighter. This is a lighter roast, and this is a darker roast. So, that’s our raw materials and here is the type of data that we acquire. So, what are we looking at here? This is a plot of total dissolved solids as a function of time and so in these experiments, what we can do is fractionate the coffee. And so, we start the brew. We capture the first few seconds of drip brew that comes out and then we get a new tube and then capture the next few seconds and then the next few seconds, and then we go and independently measure each of those. So many of you brew coffee, you just measure at the end of the brew, you get one number and that’s the total time-integrated. That’s the total aggregate TDS. These are the total dissolved solids of each faction and right off the bat, you can see that wow, there’s wild differences between the flat geometry and the conical one. So, if looking at the flat, you see it comes out at a really high TDS. So, 4.0, that’s the first stuff comes out is really concentrated, and then it gets progressively weaker, and then it hits a plateau, and it kind of bounces around, somewhere around 0.6/0.7 for a very long time and then as you reach the end of your brew cycle, the last two factions, a smaller mass bump up a little bit.
So, that’s flat. It starts off, I think, and this is the way it’s intuitive for most people. The first stuff it comes out of strong and then gets weaker and weaker. What’s surprising to me? This is how long and flat this tale is here. That’s with the flat. And again change nothing so this is the same ground size, same roasts, same temperature, same everything else. Just change the geometry of the brew basket and look, the dynamics completely change. What happens? The first stuff coming out is actually much weaker total dissolved solids and then it goes up. Hits a peak sometime, like more than a minute into the brew and then it decays, and goes down to some lower level. So, the dynamics are dramatically different, and we have data like this. This is for a particular grind size and for a particular… I’ll be happy to take questions at the end. Thank you.
So, we have representative data. This is just one set of data. We have data like this for all the different grind sizes, all the different roast levels and it’s an example of this. We’re also measuring in detail the temperature. That’s one of the dependent variables that also is a consequence. You change the geometry, you change the heat flux out of the system, you change the brew temperature.
Heather Ward: Bill has a series of graphs that show the average brew temperature across the different variables – dark to light roast, fine grind to coarse grind. The flat bottomed brewer using coarsely ground light roasted coffee has the lowest brew temperature of 89.6 Celsius, whereas the conical brewer with coarsely ground dark roasted coffee has the highest brew temperature of 94.2 Celsius.
Bill Ristenpart: and so what you are looking at here are histograms of the observed brew temperature and there’s eight different conditions here. Here’s the dark roast, the flat bottom, the fine grind size. Here’s the light roast and you could see the temperatures. There are some pretty significant differences. That’s another consequence of changing the geometry.
You also change the temperature a little bit. That’s also going to affect the dynamics of what happens during the brew and so, we have lots of data like this and me as a chemical engineer I’m having a lot of fun trying to model this using kind of classic chemical engineering transport phenomenon methodology that involves lots of differential equations.
So, how many people here would like to hear a whole bunch of calculus right now? Well, two or three hands. If you want to see some of the stuff like solving this even in an approximate fashion involves what’s known as a series of coupled, first order, linear, ordinary differential equations and it’s pretty complicated stuff. I’m not going to focus on that today. I have a graduate student who’s coming on board right now, actually, and we’re spending a lot of time analyzing that. What I’d like to focus on here though is, I think what many people in this audience care about, which is what’s the final thing, what happens?
And so that was the dynamics. Here’s the final total dissolved solids of each of the brews that we did for some of the early stages of the work.
Heather Ward: Bill is presenting a graph which shows that the flat bottom brewer produces brews with lower TDS compared to semi-conical brews. The graph also shows that TDS increases as you make the coffee grinds finer.
Bill Ristenpart: So, what you’re looking at here – this is a plot of the total dissolved solids as a function of the different conditions. So, the left two are the flat bottom brewer, the right two are the semi-conical brew basket with the medium in the fine grind and so a few things jump out right here. This is the final TDS of the whole brew. So, right off the bat, just in general, the cone yielded much higher TDSs than the flat. So, if I had to have hypothesized ahead of time, I would’ve said well probably there’s a longer path length, maybe the resonance time is longer. It has more time in there to extract, etc.
Now what’s the benefit of this? We’re getting hard data about how much higher it is and you can see that intuitively, you know that if you make it a finer grind, you should get a higher TDS and that’s true. Here’s the medium, here is the fine and it goes up in both cases.
The second general trend is the basket shape yielded, actually, a larger TDS change than changing the grind sides by 25%. So, here’s the flat bottom medium, here’s the semi-conical medium. That’s a pretty big change. Just keeping the basket geometry the same and changing the grind size yielded a smaller TDS change. And so, in other words, just changing the basket geometry had a bigger impact on this dependent variable of the total brew strength than changing the grind size by a pretty significant amount.
The third thing that jumps out as well, a lot people like to say there’s a lot of variability from brew to brew. And if you’re not familiar with box plots, what do they show? The individual points here are the measurements and each conditioning here. We have 42 angles, 42 brews. So, there’s more than 160 brews up here on the graph. And there’s quite a bit of scatter and the box plots here represent; a box plot traditionally shows the 25th and 75th percentiles. So, in other words, the size of the box shows that that’s the range over which 50% of the data points are inside that. And so you could take that as kind of one measure of the spread and so roughly the size of those boxes is on order of plus or minus .05 TDS. So why is there that variability? Well, we put the coffee grounds in, did the brew and measured it. This is what we observed and so one of the hypotheses that we have for why this is happening is that water as it’s moving through coffee grounds it follows the path of least resistance and so if you have one brew versus another with more or less channeling.
So, if it channels it finds an easy path through the grounds, it goes through, it doesn’t pick up as much dissolved solids, and so the TDS would be lower. And another brew just because it doesn’t have that easy channel, the resonance time is longer. It spends more time in contact with the grounds and has a higher TDS.
So, these are the physical measurements. Here’s a key question, how did these changes matter for flavor? How do these matter for flavor? This is an interesting question and it turns out kind of a controversial question and so if you’re like me and don’t spend any time on social media, then you might have missed this. But I thought I’d just put it up here. There was a post a couple weeks ago by Mr. Scott Rao.
Heather Ward: Bill is showing an Instagram post by Scott Rao where he is talking about Bill’s paper. Bill has highlighted wording written in capital letters that says “anyone can tell the difference between 1.1% and 0.7% TDS.” Under this quote, Bill has a text box that reads ‘This is what’s known as a “testable hypothesis.”‘
Bill Ristenpart: I guess, and I think it’s fair to say that he had a fairly scathing analysis of the quality of the data and of the methodology of experimentation. And I won’t say much about it. I encourage you to go read it. It’s a very interesting post. It engendered several hundred comments. I had journalists contacting me asking me my opinion about it. I think the comments there say much more about Mr. Rao than they do about U. C. Davis but what I would like to focus on here, my initial inclination was to ignore this completely. But I think there was some confusion out there in the industry. And what I’d like to focus on here is one of the specific criticisms that they are lodging which is, like, you know, what’s the point of measuring or doing any type of sensory analysis on these. Anyone can tell the difference between 1.1% and 0.7% TDS.
So, this is like you did experiment wrong because these are big changes in extraction therefore, this is a waste of time. Well, he very helpfully put this in capital letters and what I like to do is to spend a little time thinking about this, because this is a claim. This is what’s known in science as a testable hypothesis. And I think one thing about the coffee industry is that many people have said this to me, that there are lots of bold claims made with very little data. Here we have some data and so now let’s test this hypothesis. So, how do we do that? How do we test this hypothesis? How do we assess whether or not this is true? Whether it’s true that anybody can tell the difference between 1.1% and 0.7% TDS. Well, the way you do that in sensory science is something called discrimination testing and I think more colloquially known as triangles and here’s the basic idea. I think many people here are familiar with this idea. You serve, for example, three cups of coffee. Two of them are exactly the same, and one is different, and you present it to a taster, and they have to tell you which one is the odd one out, which one’s different. And the result of this, the data that you acquire, is a simple yes or no. Yes, they got it right or no, they did not get it right. And so, that’s the basic idea. Anybody who’s gone through Q certification knows all about this.
A key thing here and what I’d like to use as a teachable moment here is that you have a 33% probability of getting it right just by chance. I mean, you could not even taste it and grab one of these and say this one’s different. So, you have to have a much higher than 33% success rate to conclude that there is a statistically significant difference. Just to really drive this home, you have to start using by nominal statistics and so you might have heard of the phrase binomial distribution. The classic thing that they teach in undergraduate statistics is flipping coins and so you guys don’t know the flipping a coin is 50/50 odds, 0.5 Here’s a simple question, if you flip four coins, how often are you going to get two heads and two tails? How often are you going to have two successes? Two times you get it right, two times you get it wrong. And a lot of people who haven’t thought carefully about statistics will say like oh about 50% of the time, I guess, right? I mean, you should get two heads. Well, that’s not quite right. So, the way you think about it is you start tabulating all of the different possible outcomes. So, let’s call heads a success, tails a failure and X is the number of successes, and if you think about possible outcomes, it’s possible you can get four tails. Okay, so F, F, F, F.
How many different combinations are there where you can get one success? Well, here they are in different orders.
Heather Ward: Bill has a table showing all the 16 possible combinations when flipping four coins. It shows that the odds of getting two heads when flipping four coins is 6 in 16.
Bill Ristenpart: And if you keep going through and adding it all up here, the proportions there’s only one of 16 possible outcomes where you get all failures. There’s only one out of 16 where you get all successes. There’s six out of 16 where you can have exactly two heads, and so that’s the table and if you graph that, it looks like this. So, here’s the probability versus number of successes, and it is true, absolutely true that you’re going to get the most likely individual outcome is two heads, but it’s only 38% of the time. So, if you want to have a good bar trick, what you do is you bet your friend I’m going to flip four coins. I bet that you’re going to get not two heads, right because, look, if you add up all those other things, that’s a 62% probability you’re going to get not two heads and that’s good odds if you bet even money. But that’s four coin flips.
We can generalize this and let’s think about Q Certification. To pass your Q you got to do six triangles and you got to get five of them right. And so, another very related question is if you do six triangles, where again the chance of you getting it right is 33% just by chance. It’s possible that somebody goes through Q Certification doesn’t train it all, they get all six right. What are the odds of that? Well, you can generalize the binomial distribution. If you’re mathematically inclined here are the formulas. Here, I’ll just focus on the actual probability. Here’s what it looks like.
Heather Ward: Bill has a picture of a cupping table with 6 groups of 3 bowls. In each set of 3 bowls, two are the bowls are the same while the last is the odd one out. A cupper has a 1/3 chance of correctly identifying the odd one out in each cluster through random chance. Next to this picture is a graph showing the chances of getting zero out of six right, all the way to correctly guessing all six.
Bill Ristenpart: So you can guess that if you have 1/3 chance, the most likely thing is you will get two right and that’s true but there is a finite, a small but finite probability that somebody, and that’s exactly 1.6% probability that somebody will get five out of six of the channels right just by chance. And to get all six right, it’s only about one in a thousand but the cut off for Q Certification is five out of six and so you can keep this in mind, of any Q Grader you meet there’s a 1.6% probability that maybe they just got lucky. So, and that actually it turns out that six, that’s the minimum number of triangles that one can do to have any type of statistical significance, where you have that type of small result.
What if we do many more triangles? We can keep going. So, here’s the probability distribution. If we do 45 triangles and so this is the probability versus the number successes and 1/3 of 45 is 15 and you can see it’s centered around 15. But here’s the key idea: if you’re doing the whole bunch of triangles with a large panel and if you observe that like, yeah, I got 15 out of 45 right, well that’s basically no better than random. If, however, you get a result way over here.
Heather Ward: Bill is pointing out to the far right of the graph, far away from the “hump” of the distribution graph, where the distribution tails off and becomes very small. From this point on, the graphs get too complicated to describe. To make it easier to follow along, we recommend listening while looking at Bill’s presentation slides. There’s a link in the episode description. We are on slide 45 at the moment.
Bill Ristenpart: these are not zero, they’re just infinitesimally small and what that means is that in the lingo of statistics, you can reject the null hypothesis that it’s represented by binomial distribution. So, in other words, there is, in the context of tasting, there is a perceptible, statistically significant perceptible difference.
So that’s a little detour into binomial statistics. We did a whole bunch of triangles. We did recruit 45 panelists to do things and so the way we do things at the Coffee Center is we have what’s known as a sensory descriptive laboratory and so we have space to brew the coffee etc. and we have descriptive isolation booths. So, this is what it looks like. You go in there, there’s a booth. You can’t see the people to your left or right. That minimizes context bias. You don’t want to be looking and seeing somebody grimacing or yum! You don’t want to have that influence from other people. It has red lights to minimize expectation bias because you look at a cup of coffee and you see how strong it looks, how dark it looks, that’ll affect how you perceive it. So, what we can do is we could slide the triangles through. Here is the three cups of coffee.
Their job again, very simple. Just taste all three and which one’s different. Which of those two is different, or which of those three is different. And so we did a two by two factorial design. So, we compared our flat bottom and our conical using two different grind sizes, the medium and the fine. So, when you do it like that, when you have those four possible scenarios, there’s actually six triangles one can do. You can compare in the same geometry two different grind sizes. You can compare the same grind size; two different geometries and you can do all of the combinations. So, six different triangles. So, we recruited 45 untrained consumers, not expert panelists just people who the only selection bias we have is that they were willing and interested to come taste some black coffee. So, a good proxy for a consumer in a café. So, we have 45 untrained consumers, they each taste the six experiment triangles plus a controlled triangle where we did what we thought it was a very obvious difference between a very dark roast and a light roast and when you add up all those numbers, that’s a total of 945 cups of coffee. 945 cups of coffee served by Dr. Frost.
Here’s what the raw data looks like and so what is this? These are the zeros and ones. It’s a one if you get it right and it’s a zero if they got it wrong and so each row is a different person. Each column is a different one of those triangles and so let’s go back to our binomial distribution. Let’s look at a few of the results. The first one is the dark roast versus light roast and when we did that, we had 39 successes and so, what I thought of as a very obvious difference, it turns out that actually six out of our 45 untrained consumers couldn’t even tell that difference. But clearly there’s a statistically significant difference because it’s way out here, far away from the curve over here.
So that was kind of internal control. Here is fine versus medium in the flat. So, we changed our grind size by 25%. Changed our grind size by 25% doing everything the same, everything the same. Same waters, same fluoride everything. Only 18 out of our 45 panelists were able to correctly identify which one was the different one. So, that’s no better than random. It’s out here, it’s in the middle of the cloud. It’s not a statistically perceptible difference. and then we had a bunch of results in between.
So, here’s flat versus cone. So here we use the same grind size medium and 25 out of 45 and so here, that’s this little bar right here and it turns out that that’s right on the edge of where you have statistical significant differences. You’re able to discriminate it.
So, if you keep putting things on here, but back to our factorial design, here is all the results and so the thing’s highlighted in red here are the success rates and so between flat bottom, fine and flat bottom medium only the 18 out of 45. Same thing in the semi-conical. Again that 25% change in grind size, only 15 out of 45. No better than random. What is better than random was basically everything else. So, here the flat bottom versus semi-conical at the same grind size of 25 out of 45. Over here flat bottom versus semi-conical of the fine grind side, same thing and then the diagonals were also significant as well. So, that’s results.
So, I think I should emphasize that these differences, these large differences in the extraction of what we observed by no means a slam dunk, at least not for untrained consumers. Another way of looking at the data is looking at the relative proportions of correct versus incorrect. So, green here is correct, red is incorrect. This is on a scale of 0 to 45, that’s total number panelists, plotted here versus the difference in TDS. So, we had the mean TDS for example, the flat bottom fine versus flat bottom medium. And so here’s the key thing, and you can see that there’s kind of a trend here. So, for statistical significance, it was somewhere around here, okay? And that is somewhere in the range between 0.24 and 0.4% TDS. That’s right on the right on the edge of statistical significance they/you can perceive it and look at this. So here the biggest difference we have, the biggest difference was between 1.25% TDS and 0.7% TDS and 33 of our panelists got it right. More than 1/4 of our panelists did not and I should emphasize we also did this with some pretty famous coffee industry people who also tasted this and also had a hard time judging some of these things.
So, back to our testable hypothesis. So, anyone could tell the difference between 1.1 and 0.7% TDS. Well, I think the moral of the story here is that just because something is written in all capital letters on social media doesn’t mean it’s true. And I think the bigger story is if you want to really do proper testing, if you want to understand whether some change you’re doing… there’s a key thing here. I mean, some people might not like to hear that such large changes in the total dissolved solids in the percent extraction are not going to be appreciated by a pretty sizeable fraction of your consumers but that’s what the data are indicating to us.
46:20 The results of the same study that used a sensory descriptive analysis framework and consumer preference testing
Bill Ristenpart: So that was discrimination testing and so for those of you who are not aware, there’s three pillars of sensory science. One is one that I just went through great detail, discrimination testing. Is there a difference? Is there a difference? That’s yes or no. The next question is, what does it taste like? What does it taste like and that that’s where you do sensory descriptive analysis. So, we also integrate TDEL and doing what’s known as descriptive analysis and how do we do that? This is different. We don’t use untrained consumers. What we basically did is we recruited the best performing tasters, invited them to come serve on an expert panel and what does the expert panel do? Well, they spend a tremendous amount of time calibrating the panel. And the idea is, you see words here, you define your product sets, they come in, they spend time collaborating, tasting some of the representative products with the panel of judges led by the sensory scientist. They start thinking about what attributes they’re going to measure and so we use the coffee tasters flavor wheel and the corresponding lexicon and then there’s lots of time spent on calibrating the panel using different reference standards. So, you can see a few of them here. They smell tobacco so that everyone in the panel agrees that, you know, this aroma that we’re smelling we’re going to call that tobacco or another references rubber, like whiskey, a whole bunch of different references.
And so, they spend a lot time calibrating and then they go into the booths and they taste one coffee at a time. So, here’s a list of the different flavor sensory attributes that were measured for this particular study and you can see a lot of things here that are good: floral, chamomile, berry, dried fruit, raisin. Things that are not so good: musty, dusty, burnt wood or rubber. And basically happens we again use isolation booths, the cup of coffee comes in, and then on the little iPad, they use an unlabeled bar here to indicate whether that particular attribute is low or high and they assess for each sample all of these different attributes. How much berry did I detect? How sour is it? How bitter is it?
So, for this part of it, we did a two by two by two factorial design. So, we wanted to do it for a couple different roasts. We did fine and coarse. We used coarse because we found from the discrimination testing that fine versus medium wasn’t even discriminable, so it made an even bigger difference, a 50% difference in grind size and then, we again brewed everything up in the different geometries. We had 12 judges. You always want to do trial replicates, so they tasted blind the same brewed coffee three times without knowing when it was. There’s a total of 26. We did all the full physical measurements. Another 288 cups of coffee served with 26 sensory attributes. That’s 7000, almost 7500 sensory data points. And so here’s what raw data looks like. So, remember the discrimination testing was all zeros and ones. Now their numbers ranks between zero and 100. So, if it’s super, super bitter a judge gives 100. If there’s no trace of berry then they give it a zero for berry and so again, each row is judged, each column is a different system attribute. A lot of data. A lot of data.
And so how do we make sense of it? How do we pull trends out? One thing we do is something called analysis of variance known as ANOVA. You look for statistically significant trends like which ones are reproducibly higher like this one has a lot more bitterness than something else. And it turned out that there were about 18 of the 26 attributes that were by ANOVA statistically significant.
And so, because it’s such a cloud of data, there’s lots of different ways to represent it. One of the most accepted ways is something called principle component analysis or PCA and so this here’s a couple of PCA plots. And so, what you’re looking at here, the axes here what’s known as the principal of components, and this is a way of taking multi-dimensional data. This is 26-dimensional data and trying to boil it down to two dimensions. And what you’re looking for in a PCA plot is how much separation is there between things.
So, here in red are all the treatments with the different geometry. So, red is conical, blue is the flat bottom and the first thing that kind of jumps out here, the DR stands for dark, LR stands for light and you can see that there’s tremendous horizontal separation based on roast, which is not surprising. We expect those things to taste differently. More interesting is the vertical separation based on the basket geometry, so you can see that there’s a tremendous amount of vertical separation just by changing the geometry of the brew basket. And one of the nice things about this is you can then analyze what are the drivers…what’s driving these differences in the PCA? And so, here are some of those 18 statistically significant flavor attributes and, as you can see the drivers up here, there is sourness and citrus. So, the light roast and the either fine or coarse in the conical is driven up this way.
So, in other words, using the conical geometry which remember yielded these higher TDSs was also associated by an extra panel with much higher sourness and citrus flavors. In contrast, the flat bottom was more associated with sweetness and dried fruit, kind of floral flavors. So, I think that’s worth emphasizing that again all we’re changing is the geometry of the basket and when we do that, we change the TDS, of course, but we also change the flavor profile of what’s coming out in a pretty significant way. And for the more scientifically inclined here is one of the figures in the paper.
This is showing a subset of the statistically significant interactions; the roast versus geometry, the roast versus grind, grind versus geometry and what you can look out here, for example, is that here just picking bitterness, for example, here is for the light roast in the cone versus flat. The different letters here represent different treatments that were statistically significantly different. Okay, so basically A, B, C, D, each of these were different but looking at geometry of the cone versus flat, these are statistically, significantly different, for example, in bitterness.
So, that’s our expert panel. We measured in great detail what does it taste like as a consequence of these different treatments.
The third pillar of sensory science is back to the original question. Which one’s better? What do people like the best? What’s their preference? So, to do that, we have to do what’s known as consumer preference testing and so we brought in another group of untrained consumers. Again, a cohort that’s supposed to represent… the only selection bias is that they’re willing to come taste some black coffee. So, it’s a good proxy for customers in a café. So, 85 untrained consumers each tasted just four cups of coffee, brewed either in the flat or the conical and either the light or the dark roast. Same grind size and everything and they ranked each on a nine-point hedonic scale, going from all the way from “I hated it” all the way to “I loved it,” and we also did something called CATA, Check All That Applies. We had a subset of the different sensory descriptive attributes and ask them do you taste berry? If you do check it.
So, here’s the hedonic outcome. So, here’s one of them. So, here’s our traditional nine-point scale from dislike extremely up to like extremely and the different colors here represent the different treatment. So, dark roast in the flat bottom, light roast in the flat bottom, etc. The conical chronicle and this is called a violin plot. A violin plot and the way we did, we separated these 85 consumers and their taste preferences into this violin plot using something called hierarchical agglomeration clustering. And so what we did is look for different clusters or you could think of a different market segments and how did they associate with each other? How did they like the coffee?
And the key take-home message here is that when we separated it here using the agglomeration clustering each cohort strongly disliked one of the treatments. And so, for example, here cluster one was characterized, they hated the dark roast conical. In contrast, cluster two hated the light roast conical so that was kind of a roast preference, difference. And then over here, you see these guys cluster four, for example didn’t like the light roast in the flat bottom. And so, when you asked which one’s best, what do people like best. I mean, this points to, well it depends. Different people like different things. And so, some strongly like this and some like this. So, as you guys, anybody who is working in a café knows that everybody has different preferences. We can also go a little bit deeper though using the CATA and so here are the different attributes that we asked and here’s the percentage of people who responded for each of the different treatments with that, and so you can see most of them said bitter. Coffee is typically bitter so most of the consumers clicked on the bitter and you can see all the way through here, down to raisin, not many clicked on raisin.
And the interesting thing here is the differences in the treatments, about how much, so here, for example, sour, a lot of people indicated sour for the purple and green which is the light roast in the flat bottom and the semi-conical. And so you can take data like this and do something called a Lift analysis and see how did these CATA scores, how did those affect the hedonic pleasure judgments? Here’s what that looks like. So, here’s overall opinion of the coffee. So, these are like the bonuses or penalties associated with checking that and so, in other words, if a person indicated as floral that was associated with a full one-point increase on the hedonic scale. So, in other words, they liked it. In contrast rubber, everybody hated rubber. So, if they tasted rubber in the coffee that was associated with almost a negative 1.5 penalty. Going back to our sensory descriptive, we can isolate what brew treatment yielded the most pronounced rubber flavor as judged by our expert panel, and if you want to avoid that, then that’s the brew technique you want to avoid.
So, I think I’m running out of time and I want to leave time for questions. So, just to conclude here right, the basket geometry does a whole bunch of things. There’s a whole bunch of things. It clearly strongly affects extraction dynamics in the final TDS. It does yield statistically significant perceptible differences. It does but they’re not obvious. They’re not slam dunks. Not everybody can tell the difference. Not everybody can tell the difference. Very significant differences in the flavor profile and certain combinations engender strong dislike amongst certain market segments in consumer populations. So, everything we’re talking about here, it’s currently in peer review at the Journal of Food Science. Scott Frost did all this work and what I would like to emphasize is this is a small part of a big project. So we served a couple 1000 cups of coffee and that’s a small part of the big project.
The bigger project Scott is talking about again, this is a reminder. It’s an hour and a half later this morning. He’s talking about the coffee brewing controlled charts. I strongly encourage you to go see that. Tomorrow morning Mackenzie Batali is going to be talking about some of our sensory descriptive work of the fractionation. And so just to acknowledge people, again, I’m a chemical engineer so if I said anything imprecise about sensory science, it’s my fault. I’ve learned a lot from Jean-Xavier Guinard. He’s a great college. I’ve had a lot of fun working with him. Scott did all this work. McKenzie’s been helping. Big thanks to our sponsors, the Specialty Coffee Association with underwriting from Breville. We also had a lot of help from other institutes donating coffee and stuff and we have army of undergraduates who are helping out.
So, with that, I’m very… Thank you so much for your attention. I’m very happy to take questions.
57:25 Audience questions
Bill Ristenpart: Or if there’s no questions, I can go into calculus.
Heather Ward: A member of the audience has asked if Bill can share the recipe he used when experimenting with the Breville brewers.
Bill Ristenpart: Oh, yeah. So, I guess I did mention that here, I think we used a constant brew ratio. So, I think for every single brew, I think we used 55 grams of ground coffee to 1000 grams of water. Is that correct? Yes. Yeah. 55 to 1000. The Breville doesn’t do pulsing, not that I’m familiar with. So, just a constant flow rate. At Scott’s talk later today, we use the Curtis Brewers where we did do stuff with different water pulsing duty cycles. That’s a different set of work. Here today we’re just talking what’s the difference of flat versus conical. Thank you. Yes.
Heather Ward: An audience member is asking why do the first few drops have a lower TDS compared to the drops immediately afterwards in the conical brewer?
Bill Ristenpart: Oh, yeah, yeah. Yeah. Oh, good question. Good question. Yeah, we’ve been fascinated by this and so, for those who don’t know there is a mathematical modeling out there. There’s a group in Ireland who have looked at the pretty complicated set of convective diffusion equations. The main model predictions kind of match on the people’s intuition that it’s going to start high and then decrease. and here we have data that’s very reproducible. I don’t think you can the error bars. So, we’ve done this many times. We’ve confirmed that in this geometry the conical basket with, I believe this is a fine grind. We get this over and over again. So, and the first time I saw it, I thought, no, this must be wrong. It’s supposed to come out strong and then become weaker. That’s not how it behaves. It looks like this. So, I can hypothesize. One of my main hypotheses is that what’s going on, here we have a very different surface area to volume ratio of the bed of grounds and so I think what happens is that the conical takes longer to heat up and then retains heat better and so I think early on because, don’t forget the first water comes in, it’s hot, but landing on room temperature coffee grounds. So, you will have loss of heat. So, the first brew is not happening at a high temperature. If it comes in early on and it can reach the thermal equilibrium more quickly, and we have temperature data exploring this, that’s my main hypothesis. That it’s a transient thermal effect. So, short answer that’s a long way of saying I’m not sure, but we’re working on it. But, I think we very clearly observe this over and over again.
Heather Ward: An audience member is noting that many baristas will agitate the grounds when using a conical pour-over brewer. They note that the Breville used in the experiment doesn’t agitate the grounds.
Bill Ristenpart: Oh, no, yeah absolutely. So, that’s a great question. Our goal here was not to be expert baristas, but our goal here was to see how the apparatus behaved the way a typical consumer would do it. So, most consumers at home would not be sitting there stirring the coffee. So, what we’re looking at is how did that basket geometry, just swapping that out. How did that affect the flavor profile?
Heather Ward: An audience member is asking what type of coffee was used in the experiments
Bill Ristenpart: What do we use? Yeah, we used a Colombian wet washed, but we kept it constant through the experiment, So, yeah, I believe so. Yes. I’m sure it’s in the methodology section of the paper, but I don’t recall, I’m sorry. Yes.
Heather Ward: An audience member is asking whether Bill has done consumer preference testing specifically targeted at specialty coffee consumers?
Bill Ristenpart: That’s a good question. So, for our consumer results, our hedonic stuff, you have to keep in mind like we did in Davis, California. So, it’s a very millennial trending, Northern California population, many students. Whether or not that’s representative of Boston and Houston, that’s a different question. So, doing consumer testing is tricky but it tells us something. It tells us at least how students at UC Davis or staff at UC Davis how they responded to it. To do what you’re asking for that’s a bigger project and so suggest it to Mr. Giuliano. Yeah, okay. Yes, sir.
Heather Ward: An audience member is asking Bill to give more information about the clusters of consumer preferences from slide 55.
Bill Ristenpart: Yeah, they’re different. Good question. They’re different groups of individuals who tasted the coffee and they’re not separated by ethnicity or gender or anything like that. They’re separated by how they responded on that nine-point scale. So, over here you can see cluster one. They all had roughly similar responses to the dark roast flat bottom. The light roast flat bottom and light roast conical and what made cluster one stand out is wow, they all really dislike this one. So, it’s internal to how they responded to the coffees. So, it’s not based on age or anything like that.
No, no I think there’s 85 total and I forget the numbers it’s probably 20 people here, 25 here, 25 here. Yeah, four separate groups. Differentiate it’s statistically based on how they responded to the coffee and there is a bit of arbitrariness here. You can specify I only want to groups or I want 20 groups but if you specify too many groups then it loses meaning. Too few groups… Yeah.
They’re all doing the same test but it’s how you lump them together based on the response. Yes. Yeah, very good question. Thank you. Yes.
Heather Ward: An audience is asking whether the results would be similar if the experimenters changed the bed depth or the brew ratio?
Bill Ristenpart: Yes, I think so. So, I think physically what you do is you would get a higher TDS, so you’d bump everything up to higher TDSs.
I mean, that would be my first hypothesis. I mean, I think we’re seeing that for the conical. I think if you make it even bigger in terms of the bed, it would be weird if it went away. But, as a scientist, I’d say we’d have to do the experiment and check that hypothesis. Yeah. We have at least for this brew ratio we’ve done many, many experiments, so that trend is robust, and we’re trying to sort it out.
Heather Ward: An audience member is asking whether it would make economic sense to choose a conical basket brewer with less coffee compared to a flat bottom brewer because conical brewers delivers a higher TDS.
Bill Ristenpart: Yep. I mean, so, as a professor and a scientist I don’t usually make economic advice but I think one of the implications is that, like a significant fraction of consumers wouldn’t be able to tell if you water down your coffee but, whether or not that’s a good idea to do, there’s another segment of consumers who might be very sensitive to that change and would be very displeased by that. So, it’s up to you. Yeah. Very good questions. Yes, sir.
Heather Ward: An audience member is asking whether Bill has considered making the varieties of coffee one of the variables in these experiments.
Bill Ristenpart: So, we’re ramping up some projects now and we’re specifically looking at commodity coffees versus specialty coffees, not for the geometry business but for other things as well and so, that’s one of the goals for the Coffee Center is like, we want to be a place where we can do research to answer basically any question that pertains to coffee. So, I have a very, very long list of unanswered scientific questions. It would have been great to all this with a rare Yirgacheffe and see how the blueberry pops, but, we haven’t done that yet. If you’re interested, come talk to us. Yeah.
Heather Ward: An audience member is asking whether Bill and his team will post the results of their research on social media.
Bill Ristenpart: So, I don’t do social. So, the way scientists do things, we publish peer-reviewed scientific things and so one of the things I’m really excited about for the Coffee Center is we do have a strong collaboration with the Specialty Coffee Association and so this stuff, for example is a good example. We have not only this paper that’s in peer review right now at a science-y journal written in very scienc-y language. We also partner with them to make a kind of plain English version available in the 25 Magazine. So, there is a summary of the first part of this research in the current edition of 25 and so it’s easier for me to understand. Hopefully, it’s easier for you guys as well, but we want to have kind of that dual publication strategy. I don’t spend time on Instagram fighting with people, so that’s not what I do. Okay, good question.
Thank you so much for your attention.
Heather Ward: That was Professor Bill Ristenpart at Specialty Coffee Expo in April 2019. Remember to check our show notes for a full episode transcript of this lecture and a link to coffeeexpo.org for more information about this year’s event.
This has been an episode of the SCA Podcast’s Expo Lecture Series, brought to you by the members of the Specialty Coffee Association, and supported by SAP’s Softengine Coffee One. Thanks for listening!