The bourbon family forest?

On pages 72-73 of the book Guide to Urban Moonshining: How to Make and Drink Whiskey (written by Colin Spoelman and edited by David Haskell, 2013, published by Harry N. Abrams), there is an illustration of something called the "American whiskeys family tree". This is reproduced in in the article The Bourbon Family Tree for GQ magazine, from where I sourced the copy here.

The author describes it as follows:

This chart shows the major distilleries operating in Kentucky, Tennessee, and Indiana, grouped horizontally by corporate owner, then subdivided by distillery. Each tree shows the type of whiskey made, and the various expressions of each style of whiskey or mash bill, in the case of bourbons. For instance, Basil Hayden's is a longer-aged version of Old Grand-Dad, and both are made at the Jim Beam Distillery.

So, while the vertical axis is indeed a time scale, the trees are only marginally family trees in the genealogical sense. This is much more an attempt to illustrate the corporate ownership of American whiskey, which is made principally from corn (and thus is generically called bourbon, although in Tennessee they seem to rarely use this word). The main distinctions among the brands are (i) whether the non-corn part is made from rye, a little bit more rye, or wheat, and (ii) the length of time it is aged between distillation and sale.

The reticulations among the trees apparently refer to blends. The ghost lineages at the right are described thus:

Willett, formerly only a bottler as Kentucky Bourbon Distillers, has been distilling its own product for about a year; I include the brands that it bottles from other sources for reference.

Single-malt scotch whiskies — a network

In a previous post I presented a Network analysis of scotch whiskies. This analyzed 109 single-malt scotch whiskies based on the tasting notes of a single author, measuring 68 characteristics (nose, color, body, palate, finish). The analysis demonstrated that the conventional way of classifying Scotch malt whiskies by region does not relate to their taste.

An alternative approach to classifying the whiskies is, then, to actually try to group them by taste. This topic was tackled in the book by David Wishart (2002) Whisky Classified: Choosing Single Malts by Flavour, Pavilion Books, London. His objective was: "if you like a particular malt whisky, then we can tell you what other brands taste similar." This book was revised in 2006, and there was a 10th anniversary edition in 2012. There is also an associated web page (Whisky Classified).

Wishart proceeded as follows:

Tasting notes in 10 recently published books on malt whisky were analyzed for 86 readily available single malt whiskies: Arthur (1997), Broom (1998), Jackson (1995), Lerner (1997), MacLean (1997), Milroy (1995), Murray (1997), Nown (1997), Shaw (1997) and Tucek and Lamond (1997). Tasting notes published by the distilleries were also reviewed, where available.

A vocabulary of 500 aromatic and taste descriptors was compiled from the tasting notes in the 10 books. These were grouped into 12 broad aromatic features: Body (Light-Heavy), Sweetness (Dry-Sweet), Smoky (Peaty), Medicinal (Salty), Feinty (Sulphury), Honey (Vanilla), Spicy (Woody), Winey (Sherry), Nutty (Oaky-Creamy), Malty (Cerealy), Fruity (Estery) and Floral (Herbal). The 12 flavour categories are scored on a scale of 0-4 according to the intensity with which each feature is present in a whisky.

The 86 single malts were classified using ClustanGraphics. The cluster analysis groups malts into the same cluster when they have broadly the same taste characteristics across all 12 sensory variables. Technically, the method minimizes the variance within clusters and maximizes the variance between clusters. The result was ten clusters of single malt whiskies.

The order of the 10 clusters A-J maximizes the row-wise rank correlation of the underlying proximity matrix. Readers who are familiar with malt whiskies may recognise the two extremes of strongly sherried malts (cluster A) and the heavily peated, mainly Islay malts (cluster J). Adjacent to these polar benchmarks are the lightly sherried (clusters B and C) and lightly peated (clusters H and I) malts, with the light-bodied, floral and malty clusters, including four largely unpeated groups (clusters D-G) falling in the middle.

The classification is shown at the bottom of this post.

I have re-analyzed these data using the manhattan distance and a neighbor-net network. A copy of my data spreadsheet is available online. (Note that other online copies of the data [eg. here] contain errors.)

Whiskies that are closely connected in the network are similar to each other based on the 12 characteristics, and those that are further apart are progressively more different from each other. I have added colours to the network representing the ten alleged groups:

Cluster A Cluster B Cluster C Cluster D Cluster E

blue light blue light green green black

Cluster F Cluster G Cluster H Cluster I Cluster J

brown orange pink crimson red

This shows that the book's order of the groups proceeds roughly from the middle right of the network (blue) clockwise around to the top right (red).

This analysis provides only a vague justification for the book's classification. The red group does form a distinct cluster in the network, as does the blue group, which are the two extremes of the classification order. However, some of the middle groups, eg. brown and light green, do not form network clusters at all. The other groups more or less form neighbourhoods in the network, but they do not form clusters.

Therefore, trying to use this classification as the author intended, to identify whiskies that taste similar to each other, will be difficult. For example, the network shows that Ardmore is similar to Old Fettercairn, and Glen Deveron is similar to Tullibardine, but these pairs are nothing like each other — and yet all four of them are classified together (in Cluster F).

So, single-malt Scotch whiskies do not really form groups, except for the peaty flavoured ones (mostly from Islay), and to some extent the sherry flavoured ones. The rest form a continuous gradient between these two extremes. They all taste different, to one extent or another.

Finally, Wishart is not the only person to have tried clustering these data — Luba Gloukhov also tried, using k-means clustering, to no greater effect. Clustering techniques only work if there are groups in the data, and in this case the data show continuous variation between the two extremes.

Wishart's Classification of Single-Malt Whiskies

Cluster A
(Full-Bodied, Medium-Sweet, Pronounced Sherry with Fruity, Spicy, Malty Notes and Nutty, Smoky Hints): Balmenach, Dailuaine, Dalmore, Glendronach, Macallan, Mortlach, Royal Lochnagar;
Cluster B
(Medium-Bodied, Medium-Sweet, with Nutty, Malty, Floral, Honey and Fruity Notes): Aberfeldy, Aberlour, Ben Nevis, Benrinnes, Benromach, Blair Athol, Cragganmore, Edradour, Glenfarclas, Glenturret, Knockando, Longmorn, Scapa, Strathisla;
Cluster C
(Medium-Bodied, Medium-Sweet, with Fruity, Floral, Honey, Malty Notes and Spicy Hints): Balvenie, Benriach, Dalwhinnie, Glendullan, Glen Elgin, Glenlivet, Glen Ord, Linkwood, Royal Brackla;
Cluster D
(Light, Medium-Sweet, Low or No Peat, with Fruity, Floral, Malty Notes and Nutty Hints): An Cnoc, Auchentoshan, Aultmore, Cardhu, Glengoyne, Glen Grant, Mannochmore, Speyside, Tamdhu, Tobermory;
Cluster E
(Light, Medium-Sweet, Low Peat, with Floral, Malty Notes and Fruity, Spicy, Honey Hints): Bladnoch, Bunnahabhain, Glenallachie, Glenkinchie, Glenlossie, Glen Moray, Inchgower, Inchmurrin, Tomintoul;
Cluster F
(Medium-Bodied, Medium-Sweet, Low Peat, Malty Notes and Sherry, Honey, Spicy Hints): Ardmore, Auchroisk, Bushmills, Deanston, Glen Deveron, Glen Keith, Glenrothes, Old Fettercairn, Tomatin, Tormore, Tullibardine;
Cluster G
(Medium-Bodied, Sweet, Low Peat and Floral Notes): Arran, Dufftown, Glenfiddich, Glen Spey, Miltonduff, Speyburn;
Cluster H
(Medium-Bodied, Medium-Sweet, with Smoky, Fruity, Spicy Notes and Floral, Nutty Hints): Balblair, Craigellachie, Glen Garioch, Glenmorangie, Oban, Old Pulteney, Strathmill, Tamnavulin, Teaninch;
Cluster I
(Medium-Light, Dry, with Smoky, Spicy, Honey Notes and Nutty, Floral Hints): Bowmore, Bruichladdich, Glen Scotia, Highland Park, Isle of Jura, Springbank;
Cluster J
(Full-Bodied, Dry, Pungent, Peaty and Medicinal, with Spicy, Feinty Notes): Ardbeg, Caol Ila, Clynelish, Lagavulin, Laphroaig, Talisker.

Network analysis of scotch whiskies

When using non-scientific data to illustrate mathematical methods, I am told that the most common sources are baseball averages, movie grosses and political polls. I do not wish to resort to such clichéd examples, at least not yet. So, I am going to continue my series (which started with the Eurovision Song Contest) by analyzing a well-known dataset by François-Joseph Lapointe and Pierre Legendre (1994) A classification of pure malt scotch whiskies. Applied Statistics 43: 237-257. These same authors have also re-used these data: Pierre Legendre and François-Joseph Lapointe (2004) Assessing congruence among distance matrices: single-malt scotch whiskies revisited. Australian and New Zealand Journal of Statistics 46: 615-629.

The data consist of measurements of 68 characteristics (nose, color, body, palate, finish) for 109 single-malt scotch whiskies. The original authors analyzed these data using a similarity matrix and a tree. From this they produced a classification of the whiskies. They concluded that there is, indeed, a weak but detectable relationship between their classification and the geographical location of the various distilleries.

I have re-analyzed these data using a weighted bray-curtis similarity and a neighbor-net network. The bray-curtis similarity ignores "negative matches", as discussed in the previous post, so that only shared characteristics generate similarity (not shared lack of a characteristic). The weights were used to give the five types of characteristic equal influence (the 68 characteristics are not equally distributed among the five types).

Whiskies that are closely connected in the network are similar to each other based on the 68 characteristics, and those that are further apart are progressively more different from each other. I have added colors to the network along geographical lines within Scotland: light purple = lowland, red = east, yellow = midland, light green = north, dark green = west, light blue = islands, dark purple = Islay, black = Speyside. These groups are the same as those used by Lapointe and Legendre.

Click to enlarge.

There is very little to say about this diagram, except that it is not very tree-like, thus calling into question any classification scheme, and there is very little evidence of geographical patterns. There are 21 whiskies at the left of the diagram that share the biggest split, none of which come from the lowlands, midlands or Islay, but that is it.

I have been reliably told, by people with extensive experience of the matter, that each and every Scotch single malt is unique, and that therefore personal preference for one over another is entirely justified. I now have the feeling that this may actually be true.

Note: There is a follow-up post (Single-malt scotch whiskies — a network) as well as one on bourbon (The bourbon family forest?).