Visualising C-BARQ Scores

/ 2014/01/03

Good behavior, health and longevity play no role in breed popularity; these were the findings of a recently published study. In addition to being open access there are 3 more articles discussing the details in the “related links” section so I won’t reiterate what has been said. Instead I was more intrigued by the way the data was presented. It was unimaginative, confusing and uninformative.

The order of C-BARQ traits had no consistency

The order of C-BARQ traits had no consistency

The first problem is the lack of consistent order in C-BARQ variables. Table 1, Figure 1 and Figure 2 all have different order. There is no reason why for this. If we want to refer to a description or compare tables and figures there is no easy way to do it.

Figures 2 and 3 could also benefit from a splash of color or even just an informative glyph. Given that most of the results are not significant (statistically speaking), something to highlight the significant results would have made for a better figure.

“The representation of numbers, as physically measured on the surface of the graphic itself, should be directly proportional to the quantities represented.”  – Edward Tufte

A rose diagram  (petal /polar area / radial bar / sunburst) – most notably associated with Florence Nightingale – is a poor choice to visualise CBARQ data.

CIRCOS - 00

Smaller values are lost. Large ones exaggerated. 
Behavior and temperament measures for the 10 most popular dog breeds.DOI: 10.1371/journal.pone.0074770

The layout is deceptive; by dividing the circle into 14 equal segments, the authors seem to imply that all traits (at their maximum) are quantitatively equivalent.  I can only conclude from the above that beagles and dachshunds were dead – they have zero ENERGY – and are dobes so much more energetic than goldens? TRAIN a dachshund? Not likely, they score zero for trainability.  The graph paints a distorted picture of breed traits, large values are exaggerated and small values are rendered invisible.

The use of areas to represent linear values is ready-made for misinterpretation because we are not very good at comparing areas and using a pie graph would not help matters either because we are equally bad at comparing angles. It’s also counter-intuitive to think a 7.07 x 7.07 cm square is half the size of a 10 x 10 square and the same is for rose diagrams with data being proportional the root of the radius. [see Lie factor and Data-Ink ratio]

Half the radius (green) is 1/4 the area of yellow
Half the area is 70% the side length

There is a better way to show the data. Originally designed to visualize genomic data, the Canadian software package Circos has proven useful across a number of disciplines.  The circular format offers a number of advantages noted in the website including the reduction in memory load.

The figure below illustrates a very simple table and the Circos output; see here for an in-depth explanation of other options.

circos 720gig

Click on image to see animation.Credit: http://www.circos.ca

In the figure below, specific to fear/ aggression, we can see stranger aggression (STRAGG) is most prominent in Dachs and just by ribbon size alone we can see Golden, Poodle and Labrador score lowest in fear and aggression.

circos-table-aekvpwq

 

Here I take the same data in Figure 1 and visualize it using Circos. To ease comparison I kept the same order and color scheme. This format also allows us to compare several breeds at a glance.

And we can also visualise each individual breed without losing the smaller values or exaggerating larger ones.

I think Circos format offers an appealing way of displaying data; it has strong data-fidelity plus it allows us to compare a number of breeds in a single graph. It’s not the only way to visualize breed temperament but it’s a good one. One thing is certain, rose petal graphs should never be used again…. for anything![No 3D graphs either]

They say there are “lies, damn lies and statistics”, I like to add bad graphs to that list.

Related links

REFERENCE

Ghirlanda S, Acerbi A, Herzog H, Serpell JA (2013) Fashion vs. Function in Cultural Evolution: The Case of Dog Breed Popularity. PLoS ONE 8(9): e74770

Related posts

8 thoughts on “Visualising C-BARQ Scores”

  1. Gosh was I confused. I have an “order” in for a Samoyed puppy from a friend of mine. I have waited many years after losing my Akita and Samoyed and now it is time to get another dog. I have been batting around other breeds and looking at the stats above, I doubt I would pick any of them. My only reservation with the Samoyed is mine only lasted to nine. A brain tumor took her. Trainable… yes, but a lot of work. She was very devoted and I count that for high marks. The Akita was loyal beyond reproach and very obedient to any command. His problem was other dogs.

    • It’s not too far of a leap from an Akita to a Sammy and while I’ve enjoyed them both on visits they wouldn’t be the breed for me. One has to find what’s important and go with it.

  2. I’m a vet tech, and comment almost daily that wouldn’t it be an amazing thing if breeders selected for health? For traits such as healthy teeth, strong heart or clear skin? We see dogs and cats daily that are victims of irresponsible and cruel breeding. Why does an animal need a nose a certain length, the ear a specific shape or the eye a certain color?

    I like Border Collies and herding dogs. One of the great things about them is, to the best of my knowledge, they are not an AKC breed. They have been traditionally bred and selected for aptitude and temperament. What a novel idea. If the AKC gets their hands on them, the breed will became physically uniform and most likely lose the personality that makes them special.

    Border Collies aren’t for everyone. They require a lot of exercise and a job. If they aren’t given those things, they just become tortured souls. If only folks would look more at the emotional needs of the animals they acquire, and look less at the physical attributes they deem so important.

    And finally, the traits that are bred in dogs and some cats can really only be considered flaws. If we had a child with a face flattened like a pug or a persian cat, with difficulty breathing and malformed teeth, we would love it, but probably perform surgery to make it happier and healthier. Why do we intentionally breed characteristics that are health liabilities in our pets?

    I’ve newly discovered your blog and am enjoying it. Thanks.

  3. I’ve not heard that term, but it certainly is fitting. Thanks for the link.

  4. I like the idea … but the start and end segment widths as proportions of row and column totals doesn’t make much sense with the CBARQ data, I don’t think. It shouldn’t produce something that changes markedly for each breed whether you use three breeds or six or nine. Maybe taking offsets from the group mean might make most sense at the characteristic end … StrAgg etc. Not sure what the widths should be at the breed end either. Equal, possibly.

    • Thanks. I’ll revisit it when I have time. Mostly I wanted to introduce CIRCOS because I was writing (as yet unfinished) something on brain connectome and they use a circular approach to visualize the data sets. And I really hate rose petals.

Comments are closed.