Burger Wellington Data Analysis

It’s not big data, it’s burger data

Simon Carryer
7 min readJul 15, 2018

Every year, an increasing proportion Wellington’s best restaurants take part in a celebration of meat in a bun, known as “Burger Wellington”. The festival sees a staggering variety of interpretations of the simple burger, and putting aside the tedious semantics of what constitutes a burger, it seems every year the boundaries are pushed further in an effort to impress judges and tempt crowds.

I wanted to understand more about the changing fashions of Burger Wellington — what are the hot new ingredients, what trends have died out, and what makes a burger worth winning a spot on the winner’s podium. With several years of burgers now available for analysis, it’s time for some serious burger data analysis.

Summary statistics

The dataset I used was my own stitching-together of several different spreadsheets I found around the web, from 2014 to 2018. The formats changed a little over the years, and I had to do a bit of wrangling to get it all together. You can see the results of that work here.

In total, my dataset had 555 burgers exactly, though I suspect I’m missing some burgers from 2014.

A huge increase in the number of participating restaurants!

I didn’t see any major trends in terms of price — the average burger for each year has been just under $21, and finalist burgers have actually been fractionally cheaper on average, just over $20.

Slightly more interesting was the trend in patty types:

Beef is cementing its role as the King of Patties

Overall, it looks like there’s been a movement away from more unusual patties —including seafood, lamb, and pork — and towards the traditional beef and chicken burgers. Beef patties accounted for 29% of burgers in 2014, and over 37% in 2018. Chicken burgers have nearly doubled in popularity, from 10% of burgers in 2014 to 18% in 2018, overtaking both pork and “other”, to take second place in the patty-rankings.

One of the clearest trends I saw was the rise of the “milk bun” over all other bun types.

Milk buns forever!

The milk bun, a soft, slightly sweet bread from Japan, has apparently taken the burger world by storm, from being completely unknown in 2014 to comprising 20% of burgers in 2018. Brioche, a richer, eggy bread, exploded onto the bun scene in 2015, but has been losing ground to milk buns since 2017.

Savvy readers will notice that the values in the above chart don’t sum to 100%. Many burger descriptions don’t mention the kind of bun they use, but it’s become more common to do so. In 2014, only around 70% of burger descriptions mentioned a bun. In 2018, over 80% do. The bun has become a more important aspect of the burger experience. Many burger descriptions decline to mention the kind of bun but instead indicate the bakery — usually Pandoro or Arobake.

Another interesting trend is the battle of the mayonnaises. Mayonnaise is a burger staple, but its more exotic cousin, aioli, has been making inroads into burger stardom.

Just over 5% of burgers used aioli in 2014, and that has increased to around 15% in 2018. It doesn’t look like mayonnaise is going anywhere though — it has been used in around 20% of burgers every year.

Typical burgers for each year

To illustrate the changing trends in burgers, I tried to find the “typical” burger for each year. For each year, I found the ingredients that were most on-trend for that year (used more in that year than in any other year), and picked the top bun, patty, sauce and other ingredients to create a hypothetical “typical” burger. This was not very scientific, but gives some idea of the changing trends. Here are the burgers, by year:

2014: Pork burger, with Zany Zeus feta, coleslaw, and chipotle mayo, in a Pandoro bun

2015: Spicy chicken thigh burger, with jalapeños, grated carrot, lemon and pineapple, in a Zaida’s bun.

2016: Seared Angus beef patty, with Zany Zeus halloumi and fennel slaw, in a Pandoro brioche bun.

2017: Venison patty, with cos lettuce and a chorizo and kūmara hash, in an Arobake milk bun.

2018: A beef and bacon patty with kimchi and swiss cheese, in a Zaida’s bakery milk bun.

Picking winners

What I’m really interested in is the characteristics of the five or so burgers each year that are chosen as “finalists”. In particular, I wanted to know if the successful burgers were succeeding in part due to more creative choices of ingredients.

What defines an unusual ingredient? By extracting noun phrases from the burger descriptions, I could get a rough list of the ingredients for each burger. I used a somewhat imperfect technique for this. The descriptions are not a complete list of burger ingredients, which means it’s likely I’m under-reporting the incidence of ingredients. It’s also possible that more unusual ingredients are more likely to show up in the burger’s description — if your burger has kimchi in it, it’s likely you’ll mention that in the description. You might be less inclined to mention if your burger has, say, lettuce in it.

With this list of burger ingredients, I could count the occurrences of each ingredient. The average number of of all a burger’s ingredients gives an overall score for the burger. A burger with a higher score has more ingredients that have been used in other burgers. This is an inexact approach, but it seems to work. One of the most unique burgers, by this metric, was a 2018 offering from Beach Road Deli, described as:

“Seki Spit roast pork, crackling, rukau (cooked taro leaves), coconut cream and mainaise pickle in a yellow bun, with taro, maniota, and kūmara 'Island' fries.”

The Matterhorn’s “Game Changer” burger from 2014 was another example of a burger with a high uniqueness index:

“Riverbank Market venison burger with house-cured duck bacon, confit duck neck, smoked Ōtaki beetroot ketchup and Milton’s sour cherry compote on a pine-nut bun”

Almost none of the ingredients in the above burgers were seen in another burger, in any year.

By contrast, one of the most traditional burgers was Franziska’s “Che Burger” from 2018, with the following description:

“Beef patty with fried egg, cheese and chimichurri mayo in a housemade bun, with fries”

Concrete Bar’s “Lock Stock” from 2017 was similarly faithful to burger mainstays:

“Smoked cheese and Rocket Fuel-stuffed Angus beef burger with coffee rub, candied bacon and tangy ranch slaw in a Pandoro bun, with fries.”

On average, each of the ingredients in the above burgers has been seen in over thirty other burgers since 2014.

This approach to measuring burger “interestingness” is of course flawed, but while we might argue about the scores of individual burgers, hopefully as applied to the whole dataset, it will give us some useful information.

For example, here’s a chart of the average number of ingredient uses by year:

In 2017, the average burger’s ingredients had each been seen in 12 other burgers

The trend is reasonably flat, though it looks like 2017 was a nadir in burger uniqueness, and 2018 has returned to more interesting offerings.

So do finalist burgers use more unique ingredients? Overall, it appears the answer is yes. For all years, the average non-finalist burger’s ingredients were used in nine or ten other burgers. By contrast, finalist burgers’ ingredients were typically seen in fewer than six other burgers.

But! We should be careful about making this kind of claim. There are only 22 finalist burgers in our dataset. If we grabbed a random set of 22 burgers, we might find by chance that they happened to be more unique than the others. We don’t know whether the uniqueness of finalist burgers is a real effect, or whether it’s the product of chance selection. Luckily, there are statistical techniques we can use to rule this out.

If we take a random set of 22 burgers, and compare them to the others in terms of uniqueness, we might expect to find little or no difference between them. Most of the time, we’d be right — our random set looks much the same as the rest. But sometimes, by chance, we’d happen to get a very unusual set of burgers — they’re much more or less unique than the others. If we perform this experiment over and over again, taking a random set of burgers and comparing them to the others many hundreds of times, we can get a sense of the usual range of variation here. What we find is that over half the time, when we take a random set, the difference in uniqueness score between the groups is less than one point. Over 99% of the time, the difference between the two random groups is less than three points. The difference we saw between our finalists and the rest was over four points. That means either we’re witnessing a one-in-several-hundred fluke, or there’s a real difference between finalists and other burgers.

Conclusions

There’s a lot more I’d like to get into here. Finalist burgers use more unusual ingredients, but surely there’s some limit to how exotic you can go? The finalist burgers are usually not the most unique burgers for their years. Is there some optimal level of uniqueness? It turns out there’s really not enough data to robustly answer these questions, especially with such an inexact approach to labelling ingredients.

I’m looking forward to updating my dataset next year and expanding my research. In the meantime, you can see all of my working in my github, here.

--

--