According to a well-known anecdote, Thomas Edison was once asked by a French reporter how it feels to fail 999 times. Edison apparently replied: “Young man, I have not failed 999 times. I have simply found 999 ways how not to create a light bulb.”
Well, I haven't found our light bulb, yet. But I believe I'm an important step closer.
It's been a while since the last blog post. You may want to know what I was up to. I spend over a year trying to find a way to improve recommendations based on measurements. But nothing I tried seemed to work until recently.
I'll spare you the technical explanation. In the end, the change was rather simple. Finding it, however, was tricky. It's quite literally a different perspective.
So, here's a summery of what changed with the latest update:
- A new calibration for recommendations based on measurements improves results, on average.
- A new setting hides many results for your own protection, unless you change it.
- Recommendations based on many ratings should now be closer to the top of the list.
And here's why…
A new calibration
Firstly, and most importantly, the fundamental calibration how recommendations are generated using foot measurements has changed. For many members, this leads to recommendations closer to one's personal point of view.
The change is best explained using an example. For instance, if you study the details under 'Personal fittings', you may see a result like this:
|Foot volume||Foot length||Rated Size||Score|
|Member 1||about equal||UK 6.5||UK 8.0||4 stars|
|Member 2||about equal||UK 8.0||UK 8.0||4 stars|
|Member 3||about equal||UK 9.0||UK 8.0||4 stars|
In other words, members 1, 2, and 3 each rated a particular last-width combination in UK 8 with 4 stars. The odd thing is each has a different foot length (expressed in their respective UK reference size here).
I'll come back to this oddity shortly.
The major benefit is that many members now get many more other members considered as being 'close'. Their ratings too are contributing to the estimation of the fit. This means
- a larger variety of last/width combinations can be estimated, and
- more recommendations are being based on a higher number of ratings, making them more reliable, on average.
So, what about the oddity? Is a rating from, say, member 1 relevant to member 3? The answer is yes.
People's subjective choice of a shoe size, and its fit, is mostly a matter of foot volume, not foot length. The pattern expressed in the table above is widespread in the data:
- Member 1 has a high volume foot relative to his foot length. People like this often wear shoes in sizes larger than their reference size.
- Member 2 has a normal volume foot relative to his foot length. People like this often wear shoes in sizes close to their reference size.
- Member 3 has a low volume foot relative to his foot length. People like this often wear shoes in sizes smaller than their reference size.
However, as soon as you ignore the foot length, the common denominator becomes clear: In absolute terms, they all share a similar foot volume, and thus choose the same size for this particular last-width combination.
Even some custom shoe makers make lasts on foot volume, mostly. I was talking to an orthopedic custom shoe maker a while ago, and he revealed that some of his colleagues do not measure the foot length when making a custom last. They initially rely only on girth measurements.
Why is this so?
There are many reasons. Most importantly, people can feel if the leather of a shoe is sufficiently close to the skin. But they cannot feel toe allowance in most cases. It's just free space.
Even those who could theoretically feel the lack of a sufficient toe allowance – i.e. people like member 3 in the above table – probably formed an opinion about "their" personal shoe size early in life. Since their foot volume is rather low compared to the shoes they tried, they sized down and got used to a certain 'tight' feeling around their toes.
This is also the reason why many people are skeptical about measurement devises. None of them measures foot volume (and thus 'fit'). Consequently, their results differ from one's personal experience. All a measurement devise can do is to provide you with a shoe size that ensures sufficient toe allowance.
Overall, given a sufficient sample size of members being close to you, recommendations should now better reflect sizes you got used to wear.
However, there's just one problem, which required the introduction of a new setting…
Testing a new setting
Secondly, there's now a new setting (or option) available under 'Profile' that filters the results under 'Personal fittings'. By default, it's set to 'Strict'. This means only fittings close to your reference sizes (plus half a size) are shown to you. Wearing shoes in your reference size should ensure a toe allowance of 1.5 UK units, unless the last maker or manufacturer has totally different ideas about sizing than anyone else.
But why a setting?
With the new calibration, the sizes of the recommended fittings for someone who got used to wear a UK 8 shoe on a UK 9 foot are now roughly what he expects to get recommended. For instance, member 3 above rates a shoe in UK 8, so he probably expects to see recommendations somewhere around UK 8.
Unfortunately, from a medical point of view, this may not be the best possible recommendation. Technically, the toe allowance of a shoe in UK 8 is probably too high for Member 1, and too low for Member 3.
If the toe allowance is too low, your toes may be pressed together. Apparently, this increases the risk of developing foot deformations. If the toe allowance is too high, your feet have to overcome additional resistance during every step. Apparently, this increases the strain on the joints in your foot, knees, up to the lower back which could be painful. Therefore, you'll have to actively change the new setting if you want to see last/width combinations in sizes other than your reference sizes. Each option weakens the restrictions imposed by the filter.
I additionally added appropriate info and warning signs even when you changed the setting.
So, if someone wants to continue to wear UK 8 pair of shoes on UK 9 feet, it'll be his decision and responsibility now.
If you do that, please consult an expert about potential problems, or read the studies yourself. I can relate if you'd ignore those warnings, since I'm in the same position. Without an appropriate supply of shoes, there's just no other way.
However, you are now being warned which may be more than you get elsewhere.
Testing a new order
Thirdly, ratings are now aggregated by a method that considers the total amount of ratings for a particular fitting. Typically, the more ratings, the more reliable is a recommendation. The estimated fit (the number in the green boxes) is corrected accordingly. Consequently, other fittings appear on top of the list.
The disadvantage is that the numbers decreased across the board. The list still reflects the best lasts given the data, even if not a single one manages to be higher than, say, 3 stars.
Another disadvantage is that popular lasts are more likely to appear on the first page. For instance, don't be surprised if all the top results you see are from Loake or Allen Edmonds.
A new name
Fourthly, and finally, the name of the method based on measurements changed. It's now called "Personal fittings" I didn't like the old designation, anyway. But now, the old name really wouldn't have been appropriate, anymore.
I've cleaned up some last information:
- The last designation for Allen Edmonds have sort of changed. Many people use "5-65" now, instead of the old 5 (65), for the #5 last, for instance. I've turned everything to the new system.
- Almost nobody uses the correct width designations for Alden – ie. "C/E" for an "E" width. I've switched all to the common designations.
Information about shoe models (if available) are also out of date, and will be updated as soon as I find the time.
Even when one ignores the new setting, recommendations based on measurements now better reflect one's personal expectations concerning shoes sizes, on average. On a scale from -1 to 1 with 1 showing perfect correlation, a comparison of rated sizes to the top ten recommended sizes is 0.75 now.
Obviously, there are still few outliers (probably due to incorrect measurements), but the new perspective makes it easier to spot them and ignore their ratings in the future. It may also make it easier, I believe, to estimate patterns for well-known lasts, which in turn might help to estimate subjective fit preferences for those who provided ratings of these lasts.
The changes also have a negative effect: For some people, the list of recommended fittings became too long to be browsed conveniently. I'm considering a search function for brands and lasts (just like the search function for "Similar fittings" ), as well as a way to aggregate all results according to last/width combinations.
Which fitting (ie. last-width-size combination) floats to the top of one's personal list is still a matter of the available data, of course. In the past three years since Sizeadvisors went online, the growth of members, measurements, and ratings increased each year.
Hopefully, the new configuration will make the available data grow even a bit faster now.