The first stage was, therefore, to compile a more comprehensive vocabulary of words used to
describe the flavour of single malt whiskies. To maintain a level standard, the principal
malt whisky produced at each operational distillery in Scotland was selected for profiling,
and tasting notes were collected from a panel of eight writers. This provided a whisky
flavour vocabulary of around 450 words, which has since been enlarged. The vocabulary was
next categorised by MacLean’s cardinal groups of aromas and flavours. In this exercise, some
categories were over-represented whereas others were under-represented. This can be
explained by the fact that off -flavours such as “sulphury”, “stale”, “sour” and “feinty”
are seldom evident in proprietary malts because faulty casks exhibiting off -flavours are
rejected before bottling. The whisky flavour wheel was further developed into a spider chart
and a flavour map11. The spider chart is compact, attractive in use, and does contain actual
data; its only defect is that a point on a flavour spoke can fail to show if the adjacent
scores are both zero, but this seldom happens. The flavour map appears to have been
constructed intuitively, having values for “smoky”, “rich”, “delicate” and “light”. The use
of orthogonal axes labelled “light” and “delicate” appears contradictory, and some of the
malts seem oddly placed; for example not many whisky enthusiasts would think of Ardbeg and
Springbank as “light”, while Lagavulin and Talisker are contrastingly shown as “rich” on the
x-axis. All of these diagrammatic aides seek to clarify the vocabulary of flavour, and,
where applied to individual malt whiskies, to augment the tasting notes. There is no attempt
at further analysis. Cardinal flavours We have, however, undertaken further analysis to
group malt whiskies into clusters such that the malts within each cluster taste broadly
similar. In MacLean’s cardinal groups the “peaty” category includes both “smoky” and
“medicinal” descriptors, the latter being more associated wi Thislay where the peat is
decomposed forest, as compared with heather peat elsewhere. By contrast, whiskies from
Orkney, Mull and Speyside can be equally smoky but display less medicinal character. It was
therefore decided to adopt two flavour groups, “smoky” and “medicinal”, with the latter
including “salty” and “bitter” flavour descriptors. The “woody” group is not easily
expressed in the vocabulary, which more frequently relies on terms such as “spicy” for
active casks, “vanilla” relating to American bourbon casks used in maturation and “honey”
for maturation in European oak casks. The “body” of the whisky, while not a flavour as such,
occurs frequently despite not featuring in MacLean’s descriptors; similarly “sweet” and
“dry” are often used. The outcome of the vocabulary review was to settle on 12 cardinal
flavours as enumerated in the flavour dictionary in Wishart12. They are “body”, “sweetness”,
“smoky”, “medicinal”, “ tobacco”, “honey”, “spicy”, “winey”, “nutty”, “malty”, “fruity “and
“floral”. The principal malts from all operational distilleries were next rated for these 12
cardinal flavours on a 5-point scale, according to the intensity of each flavour, as
recorded by the panel of whisky writers: “not present”, “low hint”, “medium note”, “defi
nite note” and “pronounced flavour”. The resulting flavour profile for Highland Park 12
years old. The malts were then classified by cluster analysis into 10 clusters of whiskies
that taste similar, as reported by Wishart13 (see box). Such an analysis is, however, only
as reliable as the cardinal flavour intensity ratings, which had been subjectively
determined from tasting notes. The results were then circulated to the industry and many
helpful comments were received, pointing out malts that appeared to be wrongly assigned to
the clusters. These errors could be traced back to the data, where, for instance, the
absence of smokiness in Auchentoshan and Glenkinchie whiskies had been incorrectly graded.
The data were reviewed, a second classification and then a third were circulated prior to
publication in Wishart12. Th e publisher wished to include an authoritative industry review
and logo, and this was agreed by the Scotch Whisky Heritage Centre provided that all its
members, which comprise the Scotch whisky producers, approved. Their written approval was
thankfully forthcoming, and it can, therefore, be claimed that the classification has the
full support of the Scotch Whisky Industry. The objective of the cluster analysis is to
group whiskies that have similar flavour and to find nearest neighbours based on their
flavour profiles. However, all perceptions are subjective in this field, and it is indeed
quite surprising how whisky experts differ in rating the “quality” of the same malt, let
alone how its flavour is described (see Whisky Magazine). Flavour chart Having obtained a
classifi cation that was fully supported by the industry, and which was later extended to
around 200 malts, the next step was to construct a flavour map from the data. Principal
component analysis was applied to 185 single malts rated on 12 cardinal flavours, to project
a 12-dimensional scatter on to its principal components. To display the scatter in two
dimensions clearly constitutes an oversimplification of the variability in the data.
However, 46% of the variance was explained by the first two components. This can be
described as the best two-dimensional view of the flavour profiles for 185 malts. The first
component is easily interpreted as ranging from “smoky” and “medicinal” to
“fruity”, “sweet”, “honeyed” and “winey”. The second component differentiates “light”/
“delicate” from “full-bodied”/“rich”. The third component, accounting for 10% of the
variance, had high weights for “tobacco”, “malty” and “herbal”, and might be interpreted as
differentiating young whiskies that display cereal/malty notes, feints and aldehydes yet to
be absorbed or modified during cask maturation. The fourth component, accounting for 9% of
the variation, scored high on “spicy” and low on “sweet”, perhaps relating to the type of
cask used and its freshness and activity. Summary Whisky Classified is an example of a
product segmentation used in marketing, where the goal is to define niche product segments
that each appeal to different consumer groups. Th ose that have found a malt whisky they
like can experiment with products in the same or neighbouring clusters. They do not need to
taste the whole range to discover their preferences. A common marketing tool is the
suggestion “If you like that, then try this …”. The cluster analysis identifies nearest
neighbours that flag up malts of similar taste. Malt whisky novices often start near the
lower centre of the flavour chart, with delicate light and fresh malts such as Glenfiddich,
Auchentoshan and Glengoyne. Seasoned malt enthusiasts, by contrast, tend to migrate to the
two extremes, either to the rich, full-bodied intensely winey malts to the upper left, or to
the heavily peaty/smoky malts to the right. This classification is also helpful when
choosing a representative range of malts, for example in planning a whisky tasting. A
selection of, say, pin malts can be made from the 6-cluster partition of the flavour
spectrum so that a good contrast of flavour styles is obtained. I always use this approach
when planning a malt whisky tasting for a small group of people. Readers of Significance
will have the opportunity to test my classification at this year’s Royal Statistical Society
Conference in Edinburgh. It will include a technical session on the statistical material,
followed by a practical tasting of around 30 malts that have kindly been contributed by the
whisky producers. It will also be presented to the industry at this year’s World Whisky
Conference in London. Scotch whisky has become Scotland’s largest export, and, despite the
present worldwide recession, it continues to expand and flourish. My classification by
flavour is a small contribution to helping new malt whisky consumers navigate their way
around this delicious and compelling subject.
