Instead of Average
Sep-2022: The book Time to Listen is available, and Chapter 1 includes the popular story about how design for “the average pilot dimensions” was actually killing test pilots. Find out how designing multiple solutions is economically powerful.
newsletter #22 | 21-Mar-2017
Most organizations have several different products aimed at different market sectors. However each product or service is usually designed with one main user in mind. This product or service ends up only supporting a portion of the people it is aimed at. Real life scenarios are ignored because they’re deemed not important enough by the organization, or too complex, or not even recognized. (e.g. A memorable example of the latter is the 2014 release of Apple’s Health Tracker that omitted period tracking.)
The prevalence of these kinds of offerings reminds me of the episode from the 99% Invisible podcast: On Average. In that episode they point out that in WWII, pilots flying in cockpits designed for the “average man” experienced trouble controlling the planes. There were deaths, which lead to investigation and research, and in 1950 they eventually hit upon the solution of adjustable cockpit controls.
So, instead of designing for an imagined average user, you could design adjustable experiences. This, of course, puts the onus on the user, both in terms of attention and cognitive load. My opinion is that several elegantly- and specifically-designed solutions for different peoples’ thinking styles is a better approach. For example, Healthwise came up with three different designs to support three different thinking styles of people trying to lose weight. And the example below, from Caroline Jarret, who is currently writing a book titled Surveys That Work. Pay close attention to how Caroline refers to “average.” It’s called the “mean” or “arithmetic average,” and provides very little help to our design of a solution.
Here’s Caroline:
Back in stats class, you probably learned that a ‘mode’ is the value that occurs most frequently, and the ‘range’ starts with the lowest value and finishes with the highest.
So let’s say that I am a restaurant owner, and I’ve done a little pilot survey with 100 responses to find out whether a recent ad campaign to attract families with children has succeeded in attracting them or not. I get these answers:
The average number of children per family is just over 2.
If you look at the range, you’ll see there are a couple of very large families, including one with 9 children — perhaps not that surprising when we think about life today with blended families. And perhaps even more surprisingly, some families are turning up with no children in tow at all. So maybe if you’d designed seating in a restaurant to allow for two adults and two children per family, you’d have a lot of empty seats at tables for four, but some much larger groups struggling to organise themselves.
If you look at the most frequent number of children, the mode, then you’ll see that the mode is 1 child, with the group of people who turn up with no children at all the next biggest So if you designed a ticket price that’s aimed at attracting families with 2 kids, you’d fail to cater for the two biggest groups in your audience (and not be all that helpful for the larger families either).
I didn’t learn a lot about modes and ranges in statistics class because they’re awkward concepts mathematically, whereas means have lots of very interesting mathematical properties that are very handy for statistical purposes. But for designers, my experience is the other way around: means can hide a lot of the details that we need for design, whereas modes and ranges can be much more informative.
These sorts of Zipf distributions are everywhere in UX, with the most familiar being the search terms. (this section written by Caroline Jarret)
Caroline’s restaurant example is a good way of demonstrating how eager our culture is to design for the mean, and how much of a hassle an “average” design turns out to be for most people. If you expect several different groups at the restaurant (and assuming this is a restaurant that anticipates lots of kids), you’d design multiple solutions: tables for pairs of adults, tables for adults with one kid, and configurable tables pre-arranged in anticipation of adults with 2–4 kids and with 5–9 kids. (Ignore for purposes of this example the fact that it’s rife with assumptions around how a kid behaves and how to support a kid versus an adult. Focus on the average versus mode part of the example.)
