Skip to main content

How Might Roasting Affect Drip Extraction: A Detailed Analysis

Royal Coffee Director of Roasting Jen Apodaca. All images courtesy of Royal Coffee, Inc.

In many cases, a roaster will make choices about heat application based on final, intended use. An espresso roast will differ significantly from a drip roast based on known effects on the final flavor profile.

Not long ago at the Crown office, Jen, Evan, and I were speculating about what choices might be made specifically during the roast that affect a coffee’s solubility. We regularly brew and take measurements of extraction percentages of our Crown Jewel selections, and occasionally theorize about why one particular roast seems to extract differently than a second roast of the same coffee. Could roasting styles be quantified in terms of time, temperature, extraction percentage, and TDS?

One of my working theories has been related to Maillard development — my assumptions usually correlate more time spent in this phase of the roast before first crack to higher solubility. I wanted to see if my coworkers had similar assumptions, so I asked Jen and Evan what they thought. Evan and Jen both focused on darker roasts, indicating that, to a point, they might also be more soluble. Jen bet that after 2nd crack “extraction percentages go down because of the physical change to the coffee.” Evan additionally speculated that longer sugar-browning, to a point, might lead to higher solubility after first crack.

I’d be remiss at this point if I didn’t mention that all these numbers really don’t get at the ultimate question about coffee: “How does it taste?” Extraction percentage will never fully illuminate flavor profile, but it can tell us at least a little about the way a coffee brews, and what steps might be needed to access its full potential.

The Variables

I quickly surveyed our records and was able to isolate a lot of data — 3,750 data points related to 135 brews of 111 roasts of 59 unique coffees.

For nearly every one of the coffees, two roasts were performed. This enabled me to quickly compare and contrast extraction data, and then see if there were major differences in roast styles.

For a single given brewed coffee there could be as many as nine roasting variables, three green coffee variables, and 10 extraction variables. This unfortunately makes for a wide playing field.

The Assumptions

Of all the roasts performed, only three entered second crack. For all coffees, total roast time varied between 7:30 and 11:59, ground ColorTrack readings fell between 54.45 and 62.68, except for the three dark roasts which measured between 70 and 71.2. Without exception, every roast was performed on our 1-kilo Probatino, though roast batch sizes varied. Time stamp observations are contingent on manual entry and observation.

Every extraction was brewed at the Royal Offices in Emeryville, Calif., at sea level. Any extraction data we collected was based on manual measurements. Manual brewing styles, of course, are highly variable based on the brewer. I think it’s unlikely, but our water composition may have slightly affected extractions, and I know that over the course of the past year the mineral content of the water in the office here has shifted dramatically despite high-quality catalytic carbon block filtration.

Not all roast pairs provided great comparisons, so below, I’ve made a distinction for times when I’ve narrowed the field to cut out noise of irrelevant data. In some cases I chose to filter results by brewing device, roast time range, or simply whether or not there was complete enough data to draw conclusions. Lastly, even though the data set is fairly robust, there are far, far too many variables and outliers to draw serious conclusions. In the future I hope to design a few experiments to narrow the list of variables, but for the time being, I think we can glean a little information from the analysis below.

The Analysis

Looking for overarching patterns across every brew of every roast, many factors appeared insignificant. Perhaps I’d bitten off more than I could chew… roasted weight loss, percentage of time spent in post-crack development, green coffee specs, and the difference between internal and external ColorTrack readings all seemed pretty random, with minimal relationship to brewing parameters.

The most significant factor appeared to be total roast time, which seemed to inversely correlate to extraction percentage, to the tune of about 1 percent for every 133 seconds of roasting time, with plenty of exceptions to the general trend.

There was also a slight trend of lighter coffees being the most soluble. The three darkest coffees did not follow the trend, but there is a 7.5 point gap between the lightest dark roast and the next darkest light roast, so it’s possible there is some curvature between medium and dark where extraction dips and rises.

Identical Brew Specs

Ultimately, though, I felt I needed to start filtering these results to make sure I was being fair. My first step was to eliminate any roast pairs that weren’t brewed identically. Since I wanted to look at significant factors in extraction variables, I narrowed my hunt to identical brews of different roasts of a single coffee that varied by more than 0.5% percent in extraction. I was left with 50 roasts of 25 coffees.

Of these 50 roasts, the only significant metric I could identify was the ratio of total roast time to post-crack sugar browning (or post crack development, hereafter abbreviated PCD). However, the data supports the opposite conclusion from what Evan expected! The graph below illustrates the trends in percentage of PCD versus extraction percentage in the 50 roasted coffees.

There appears to be a slight inverse relationship between the two. When comparing each pair, the roast that brewed with a higher extraction percentage also spent less time in PCD in 18 out of 25 cases. That’s statistically significant at 72 percent, but not exactly conclusive.

It would be unfair to forgo examination of the other brew pairs with slight or nearly no differences in extraction. If the trend above holds true, the opposite should also be true: coffees with minimal differences in extraction percentage should also express minimal differences in PCD percentage. As it happens, there were the exact same number of coffees that fell into this sample set: 50 roasts of 25 coffees. These pairs split almost down the middle into two sub-categories. Twelve pairs had nearly no difference in extraction (less than 0.2 percent), the other 13 had minimal differences (between 0.2 and 0.49 percent).

Of the 12 pairs with that expressed almost no difference at all in extraction percentage, the PCD percentage differed significantly (by more than 0.5 percent) in all but one of the pairs. And in the 13 pairs with a slight difference (between 0.2 percent and 0.49 percent) eight of the pairs had a lower PCD percentage. Arguably, this disproves the trend. The correlation between low PCD percentage and high extraction percentage is tenuous at best, and possibly irrelevant.

Similar Total Roast Times

My next step was to reorganize and narrow my dataset by total roast time. If I could concentrate on pairs where the same coffee was roasted twice at less than 30 seconds difference, perhaps I could find a common factor in Maillard reactions or post crack development that affected the brew.

I found 21 coffees (42 roasts) that met the 30 second or less roast differential prerequisite. Of these 21 pairs, 2 expressed no difference in extraction percentage, despite differences in roast style. Both of these were wet hulled Sumatran coffees brewed on Bonavita brewers.

Most of the other coffees failed to beat random odds. In fact, for two of the 21 pairs, we doubled down on brewing, and in both instances the roast that had a higher extraction percentage in the first brew-off had a lower extraction percentage in the second — a frustrating inconsistency, to be sure.

If I narrowed my results to coffees with large differences (0.5 percent or more) in extraction, only nine pairs remained. I looked for trends in the most soluble brew in each of the pairs.

Nothing unanimous or definitive though the majority of coffees with higher solubility trend towards shorter roasts with larger percentage of time spent during the initial drying stages, less time in Maillard reactions and PCD, and lighter ColorTrack readings.

With such a small sample set, the evidence here doesn’t seem to support definitive conclusions about coffee brewing extraction differences resulting from stylistic changes in roasting that affected coffees roasted to similar end times.

Automatic Brew Cycles Only

Frustrated by a lack of conclusive results, I decided to reorganize data again. It seemed plausible to me that manual brewing could introduce unaccounted-for variables that might be muddling results. If I could eliminate the human element from brewing differences, perhaps we’d see better trends. Bonavita’s automatic brew cycle ensured consistent brewing parameters given a consistent dose, grind, and water weight, so I narrowed all the roasts down to just those brewed on Bonavitas.

We did a fair amount of brewing on these devices, analyzing 53 roasts of 27 coffees, and I was finally able to see what I thought were pretty clear trends when looking at all the brews in broad strokes. I was able to reconfirm a slight inverse relationship between PCD ratio and extraction percentage… or so I thought:

I was also able to see again an overall slight trend of the inverse relationship of Maillard time to extraction percentage, roughly to 1 percent lower extraction for every 50 seconds spend in Maillard reactions, with one significant outlier: a Brazil that spent 5:35 in Maillard reactions and yielded a 23 percent extraction (not depicted below).

I was also able to reconfirm overarching trends indicating a pretty strong relationship between shorter roasts and higher extraction percentages.

There is also trendline evidence to suggest that lighter roast color results in higher solubility.

However, problems in these trends arose quickly when I compared the roast pairs. I honed in on 21 coffees that were each roasted twice and each roast brewed once. Of these 21, two were those pesky trend-bucking Sumatran coffees with identical extraction numbers but different roast styles, which are included in the statistics below:

Maillard reaction time, lower percentage of PCD, and ColorTrack readings all failed to beat random chance. So really, roast length seems to be one of the only major factors I can say with confidence makes a difference in these coffees’ solubilities. Again, however, there is an exception.

One coffee I excluded from the group above was roasted twice, and each roast brewed twice. The only difference in brew method was the grind setting, and the roast that had a higher extraction at a coarser grind setting (+0.76 percent extraction and had the lower (-0.45 percent) extraction the second time. So frustrating! Why did this happen?

The two roasts differed by 20 seconds in roast length, 25 seconds in Maillard time (the roast with longer total time had a shorter Maillard duration), 4.3 percent in PCD (quite significant), and 0.87 in ColorTrack reading. It’s hard for me to believe that a simple 0.5 adjustment in grind on the Mahlkonig EK43 could cause such a dramatic shift in extraction, which leads me to conclude that, at least in this instance, operator error may not have been eliminated as a factor simply by looking at auto-drip brew cycles.

The Conclusions

I don’t believe the evidence from our analyses can support decisive conclusions that apply to all brewing parameters for all roasts of all coffees. But if we hone our focus, I think there is good information here.

  • Bonavita brews seemed to produce a few associations, at least from an overall trend perspective, between roast styles and extraction. However, manual brew styles did not always uphold these trends, so it seems reasonable to conclude that the small variations attributable to the human influence on manual brewing are a more important factor in extraction than even the most dramatic differences in roasting.
  • Overall, roasting style may be a contributing factor to a coffee’s solubility in some cases, but not in all conditions nor universally across all coffees. Associations can sometimes be drawn between high extraction percentage and the following roasting metrics:
    • Lower PCD percentage
    • Shorter Maillard reaction time
    • Shorter total roast length
    • Lighter color roasts
  • I think it’s reasonable to surmise — without having done sufficient testing to decisively conclude — that much of the same stuff in coffee that is soluble in hot water is also being roasted away with longer times, higher temperatures, and darker colors. The peak solubility of coffee is likely far earlier in the roast — perhaps as early as the very beginning of first crack —  than I had expected. Solubility definitely seems to be affected strongly by length of roast and not just temperature or color.

After running all these numbers, the bottom line is that I don’t think roasting for extraction percentage is really the best way to roast coffee. Better to cup and use your palate than to try and roast by number to get the best tasting brews.

(Editor’s note: Royal Coffee is a current Daily Coffee News advertiser, though this piece was conceived and published for the benefit of coffee professionals everywhere.)



Warren Machanik

A great and interesting article. I would love to have seen a cupping result. While extraction percentage is a great and educative measure, would have loved to see how the mouth feel and flavour complexity changes. I am blown away to hear that there appears to be an inverse relationship between PCD and extraction percentage. I had always assumed a longer dev time would increase extraction percentage. Very good article.

Nick Kohout

Chris, this is pretty cool. Any chance we could get p-values and R-squared for the regressions? Confidence intervals would be cool, too, if you are anti-p.

Chris Kornman

Hey Nick, as you probably surmised, they mostly hover in the low-to-moderate correlation realm. BonaVita ext% vs. total roast time = -0.453 Pearson, for example. With many complicated factors, I don’t think we’ll see anything in this realm approaching a perfect correlation, but that’s also kinda the point of the research. No single variable in roasting will give you a surefire method of affecting extraction, though applied together you might have a better chance… (e.g., Bonavita ext% vs. short plus light = 0.559 Pearson). Email me, I can share some more details if you like.

Jacques Van Wyngaard

Loved it! Thank you for taking the time to be so objective. Good job guys!

Comments are closed.