Chapter 8 Practice: A Four-Step Check and a Pre-Registered Experiment

Part One: checking a batch of plant raw material

The chemistry of this book is useful only if it changes what happens when a sack, a carton or a drum arrives at the door. This part sets out a four-step check that a buyer, a processor or a small brand can run on each batch, with equipment that costs far less than a laboratory and a reference laboratory for the questions that need one.

Step	What to do	What it catches
1. Paperwork and identity	Match the lot number on the goods to the COA. Check that the COA names the laboratory, the method, the limits and the result for each test (Chapter 6). Look at and smell the goods against a retained reference sample.	Wrong or mixed material; certificates that do not belong to the batch; tests against the wrong limits
2. Water	Measure water activity at receipt with a calibrated meter, on several units from different parts of the lot.	Material dried poorly or exposed to humidity in transit; mould risk ( $a_w$ above about 0.6) and faster quality loss
3. Quick chemistry	Use a simple screening test suited to the product: a refractometer (°Brix) for juices and concentrates; a colour test for total polysaccharide or PAC run against your own standard curve (Chapters 2 and 4). Send retained samples to an accredited laboratory on a schedule and whenever a screen looks wrong.	Drift between batches; dilution; a supplier’s numbers that cannot be reproduced
4. Blind tasting	Run a triangle test against the retained reference sample.	Changes in flavour or pungency that the other tests miss, such as over-drying or storage taint

The triangle test is the oldest and simplest blind test of difference (International Organization for Standardization 2021). Each assessor receives three coded samples, two the same and one different, in a balanced random order, and is asked which is the odd one out. If there is no detectable difference, an assessor will pick correctly one time in three by chance, so the number of correct answers follows a binomial distribution with $p = 1/3$ .

Calculation

The number of correct answers needed to conclude that the samples differ (one-sided, α = 0.05), the power to detect a difference when 30 % or 50 % of assessors can genuinely tell, and the p-value for one invented result. If a share $p_d$ of assessors can tell, the chance of a correct answer is $1/3 + 2/3\,p_d$ .

from math import comb

def tail(n, x, p):
    """P(at least x correct out of n) when each is correct with probability p."""
    return sum(comb(n, k) * p ** k * (1 - p) ** (n - k) for k in range(x, n + 1))

def critical(n, alpha=0.05):
    """Fewest correct answers that show a difference (guessing rate 1/3)."""
    return next(x for x in range(n + 1) if tail(n, x, 1 / 3) <= alpha)

print(" n   need   power if 30 % can tell   power if 50 % can tell")
for n in (12, 24, 36, 48, 60):
    x = critical(n)
    powers = [tail(n, x, 1 / 3 + 2 / 3 * pd) for pd in (0.3, 0.5)]
    print(f"{n:2d}   {x:4d}   {100 * powers[0]:22.0f} %   {100 * powers[1]:22.0f} %")

# One panel's result (invented): 24 assessors, 14 picked the odd sample.
n, correct = 24, 14
print(f"{correct}/{n} correct: one-sided p = {tail(n, correct, 1 / 3):.4f}")
#  n   need   power if 30 % can tell   power if 50 % can tell
# 12      8                       26 %                       63 %
# 24     13                       55 %                       93 %
# 36     18                       72 %                       99 %
# 48     22                       88 %                      100 %
# 60     27                       92 %                      100 %
# 14/24 correct: one-sided p = 0.0103

With 12 assessors the test misses most real but modest differences. With 36 it detects a difference that half the assessors can perceive almost every time. For routine checks of incoming goods, 24 assessors is a reasonable compromise; for a decision that matters, use more.

Note

A triangle test answers only “are these different?”, not “which is better?” or “how are they different?”. If the samples differ in colour or texture, present them in a way that hides those cues (coloured lights, ground samples, the same serving vessel), or the test will measure appearance rather than flavour.

Part Two: a pre-registered experiment

The question

Does freeze-drying keep more of ginger’s 6-gingerol, relative to 6-shogaol, than hot-air drying at 60 °C, when both start from the same fresh ginger and end at the same water activity? And is the difference large enough that people can taste it? Chapters 3 and 5 predict yes on the chemistry. The answer matters commercially, because freeze-drying costs much more, and that is exactly why the comparison should be designed so that the answer cannot be bent by whoever runs it.

Design

Element	Specification
Material	Bentong ginger, mature rhizomes, from lots representing different farms and harvest weeks
Unit	One lot: about 4 kg of fresh rhizome, washed and sliced to 3 mm on one machine
Allocation	Within each lot, slices are pooled and divided into two halves by computer-generated random allocation; one half freeze-dried, one half hot-air dried. The allocation list is prepared by someone who does not process or analyse samples
Freeze-drying	Frozen at −30 °C; primary drying at 10 Pa with the shelf controlled so the product stays below −20 °C; secondary drying at 25 °C
Hot-air drying	60 °C, air velocity and load fixed in advance
Common end point	Both halves dried to water activity 0.25 to 0.30, measured, then milled and packed in the same barrier pouches
Primary outcome	ln (6-gingerol / 6-shogaol), by a validated HPLC method with reference standards, dry basis
Secondary outcomes	6-gingerol and 6-shogaol contents (mg/g dry basis); water activity; colour; rehydration ratio of slices; triangle test
Blinding	Samples are coded before they reach the laboratory and the tasting panel; analysts and assessors do not know the drying method
Pre-registration	The protocol, analysis code and decision rule are published on an open registry before the first lot is processed

Number of lots

The outcome is compared within each lot, which removes differences between farms. The number of lots needed depends on how much the within-lot difference varies from lot to lot, which is not yet known.

Calculation

import math
from statistics import NormalDist

z = NormalDist().inv_cdf

def lots_needed(sd_log, fold, alpha=0.05, power=0.90):
    """Paired design: each fresh lot is split, one half freeze-dried and one
    half hot-air dried. Outcome: ln(6-gingerol / 6-shogaol). Detect a
    difference of `fold` between methods, two-sided, with normal approximation
    plus the usual small-sample correction z²/2."""
    delta = math.log(fold)
    za, zb = z(1 - alpha / 2), z(power)
    return math.ceil(((za + zb) * sd_log / delta) ** 2 + za ** 2 / 2)

print("sd of the within-lot difference (log scale) and the smallest fold change")
for sd in (0.4, 0.6):
    for fold in (1.5, 2.0):
        print(f"  sd {sd:.1f}, fold {fold:.1f}: {lots_needed(sd, fold):3d} lots")
# sd of the within-lot difference (log scale) and the smallest fold change
#   sd 0.4, fold 1.5:  13 lots
#   sd 0.4, fold 2.0:   6 lots
#   sd 0.6, fold 1.5:  25 lots
#   sd 0.6, fold 2.0:  10 lots

The protocol takes the cautious case (standard deviation 0.6 on the log scale, smallest worthwhile difference 1.5-fold) and plans for 25 lots, plus three more to allow for lots lost to equipment failure: 28 lots in all. A pilot of four lots will be used only to check the procedures, not to change the sample size.

Sensory test

Powder from six lots, chosen at random after processing, is brewed in a fixed amount of hot water and presented in triangle tests, freeze-dried against hot-air dried, to 36 adult assessors per lot, with the balanced presentation orders set out in the standard (International Organization for Standardization 2021).

Analysis and decision rule

The primary analysis estimates the mean within-lot difference in ln (6-gingerol / 6-shogaol) with a 95 % confidence interval, and reports it as a ratio. The result will be read by a rule fixed before any data exist, in the spirit of equivalence testing (Lakens 2017):

Calculation

import math

FOLD = 1.5                      # smallest difference worth acting on

def verdict(ratio, lo, hi):
    """ratio: freeze-dried ÷ hot-air dried, for gingerol:shogaol,
    with its 95 % confidence interval (lo, hi). Written before any data."""
    if lo >= FOLD:
        return "freeze-drying keeps clearly more gingerol (difference ≥ 1.5-fold)"
    if lo > 1:
        return "freeze-drying keeps more gingerol; size not shown to reach 1.5-fold"
    if hi < 1:
        return "hot-air drying keeps more gingerol"
    if lo >= 1 / FOLD and hi <= FOLD:
        return "no difference worth acting on (within 1.5-fold either way)"
    return "unclear: the interval is too wide"

# Invented results, to show how each kind of outcome would be reported.
for ratio, lo, hi in ((3.1, 2.2, 4.4), (1.6, 1.2, 2.1), (1.05, 0.82, 1.34), (1.3, 0.7, 2.4)):
    print(f"ratio {ratio:.2f} ({lo:.2f}–{hi:.2f}): {verdict(ratio, lo, hi)}")
# ratio 3.10 (2.20–4.40): freeze-drying keeps clearly more gingerol (difference ≥ 1.5-fold)
# ratio 1.60 (1.20–2.10): freeze-drying keeps more gingerol; size not shown to reach 1.5-fold
# ratio 1.05 (0.82–1.34): no difference worth acting on (within 1.5-fold either way)
# ratio 1.30 (0.70–2.40): unclear: the interval is too wide

The examples are invented, to show in advance how each kind of result will be reported. An interval inside 1.5-fold in both directions will be reported as “no difference worth acting on”, not as “no effect”; an interval that crosses the boundaries will be reported as unclear, not as “a trend”. The sensory results will be reported lot by lot and pooled, with the number of correct answers and the p-value.

Ethics, safety and conflicts of interest

The sensory panel consists of adult volunteers who give informed consent, can withdraw at any time, and are screened for ginger allergy; samples are prepared in food-grade conditions. The panel protocol will be reviewed by an independent ethics committee before recruitment. The protocol follows the principles of pre-registration (Nosek et al. 2018).

Disclosure

The author and his group operate websites on Bentong ginger and on freeze-drying technology (bentonggingers.com, freezedriedexpert.com) and have commercial relationships with food-processing supply chains in Malaysia, so they have a direct commercial interest in the outcome of this experiment. To limit that interest: the protocol, analysis code and decision rule will be published before the first lot is processed; the HPLC analysis will be done by an accredited laboratory independent of the author’s businesses; and the results, the raw data and the code will be published whatever they show.

What this experiment cannot show

It compares two drying methods for one cultivar, at one slice thickness, at one hot-air temperature. It says nothing about other crops, about other dryers, or about storage beyond packing. It measures molecules and taste, not any effect on health: a higher gingerol-to-shogaol ratio is a difference in composition, not a benefit to anyone’s body, and this book makes no such claim.

References

International Organization for Standardization. 2021. ISO 4120:2021 Sensory Analysis — Methodology — Triangle Test.

Lakens, Daniël. 2017. “Equivalence Tests: A Practical Primer for t Tests, Correlations, and Meta-Analyses.” Social Psychological and Personality Science 8 (4): 355–62.

Nosek, Brian A., Charles R. Ebersole, Alexander C. DeHaven, and David T. Mellor. 2018. “The Preregistration Revolution.” Proceedings of the National Academy of Sciences 115 (11): 2600–2606.

Fan, Y. (2026). Molecules and Terroir: Polysaccharides of Dendrobium officinale, Gingerols of Zingiber officinale, A-Type Proanthocyanidins and the Mass Transfer of Freeze-Drying [本草的分子与风土] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.23693057 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).