Selection bias occurs when data is kept or discarded in a way that makes the sample no longer representative of the whole population. When studying the relationship between green-bean moisture and thermal response during roasting, this type of bias can quietly creep into data collection and distort the final conclusion.

Selection bias can start before roasting even begins

Risk can appear right at the step of choosing which raw-material lots to include in a test. If the person conducting the test tends to pick lots with moisture that's easy to measure, easy to handle, or that previously produced a “nice-looking” batch, the resulting dataset will no longer fully reflect the actual range of moisture variation in the raw material.

Moisture needs a representative sample

For the relationship between moisture and thermal response to be statistically meaningful, the study sample needs to cover a wide enough range of the moisture levels the raw material can actually have, including lots at both ends of the distribution. If only lots near the average moisture value are included, the conclusions drawn will be hard to apply to lots that fall outside the range studied.

Designing the sampling method before looking at results

An effective way to reduce bias is defining the sampling rule in advance — for example, choosing randomly within each moisture band, or selecting on a fixed schedule regardless of previous roasting outcomes — before starting to collect data. If the sampling rule gets adjusted after some results have already been seen, the risk of unintentionally favoring the initial hypothesis increases.

Placing moisture alongside other input variables

Moisture isn't the only variable affecting a bean's thermal response. Bean size, density, and lot origin can all act at the same time. If these variables aren't recorded simultaneously, an observed relationship between moisture and the heat curve could actually be driven by another variable that happens to travel alongside moisture in the chosen sample.

How moisture relates to the heat curve

Research on heat and mass transfer kinetics in roasting shows initial moisture can affect how quickly a bean absorbs heat in the early stage of a roast. However, the specific magnitude of effect recorded in a given study only reflects the exact raw material and roasting system that study used.

The risk of extrapolating from a small sample

When the number of test batches is small, a few unusual observations can create the impression of a clear trend when it's really just random variation. Extrapolating a conclusion from a small sample to the entire range of raw material used daily needs to be done cautiously, especially when that small sample was chosen in a non-random way.

Ways to reduce selection bias in roasting data

Some practices can help reduce this risk: logging every roasted batch within a period instead of only selecting “notable” ones; defining in advance the criteria for including or excluding a batch from analysis; and keeping batches whose results don't fit the initial hypothesis rather than discarding them.

Practical conclusion

To reliably verify the relationship between green-bean moisture and thermal response, sample selection needs to be designed before data collection, cover a wide enough range of actual moisture, and account for other input variables that could create confounding. A conclusion drawn from a biased, selectively chosen sample shouldn't be treated as a general rule for every batch of raw material.

References

• Hernández, J.A. et al. (2007). Analysis of the heat and mass transfer during coffee batch roasting. Journal of Food Engineering, 78, 1141–1148.
• Heat and Mass Transfer Kinetics on the Chemical and Sensory Quality of Arabica Coffee Beans (2022). Agronomy, 12, 2880.
• Batch-Scale Simulation of Heat and Mass Transfer of Coffee Roasting in Spouted Bed Roasters (2025). Beverages, 11, 162.