2013 was a good year for Sizeadvisors: The number of measurements increased by about 44 percent, the number of ratings increased by about 57 percent, and the software was running nearly without interruption after a database upgrade. On the first Christmas day, we reached 2200 ratings – thanks to all contributors for the nice present! – and we've already passed 2500 ratings.
Over the year, I was busy trying to find ways to make Sizeadvisors better: Talking to members of the site and custom shoemakers, finding and testing other algorithms, fixing smaller problems, and so on.
Meanwhile, I kept thinking whether there's a way to integrate all that feedback? It took me a while, but I think I've found it.
So, here's what will change in the coming months. Don't worry. The interface will mostly remain as it is.
One list to rule them all
Currently, there are two different ways to fitting recommendations: 'Similar fittings' based on your ratings, and 'Personal fittings' based on your measurements. Each has its unique advantages, and disadvantages.
For instance, 'Similar fittings' is rather conservative, and less risky, in general. But you'd need to have at least one well-fitting pair of shoes to get interesting results. Also, the rationale for a fitting recommendation – a last/width/size combination in Sizeadvisors' terminology – was hardly intuitive. Finally, there's no way to recommend EU-sized shoes if you could only rate UK or US sized shoes, and visa versa.
On the other hand, 'Personal fittings' is more progressive, and takes a broader view. It's also more general, since (nearly) everybody can submit measurements. But its results are also more risky. There's also what might be called the density problem – some areas in the multi-dimensional space described by the various measurements are sparsely populated, and there's no way to estimate the fit and generate recommendations if nobody is "around" you.
As a result, it makes sense to merge both lists, so that each method can reduce the disadvantages of the other without loosing too much of its own advantages.
Following the crowd
I think the new method will be much more intuitive than the current ones. The fit estimations will be based on the ratings of members you follow.
In other words, it'll be like subscribing to a newspaper. Only you don't subscribe to news, but to another member's (list of) ratings. In fact, you should be familiar with the basic model, if you use Twitter, Facebook, Pinterest, tumblr., Instagram, or similar sites.
It's not quite the same, however. There are some differences:
- Opt-out: Usually, you'd actively select a newspaper you'd like to read. This is easy, because you can judge for yourself whether you're interested. In contrast, it's not always easy to see whether you should follow other people's fit ratings. Also, the available data at Sizeadvisors grows on a regular basis, so it would be quite a bit of work to remain up-to-date. Thus, the new method will thus automatically decide whom you follow, based on signals from your existing ratings and measurements. You will get the option to stop following others, though.
- Offsets: If you subscribe to a newspaper, you read the stories as they are. In contrast, Sizeadvisors' new method will allow you to follow someone with an offset. For instance, if you follow somebody with an offset of -2 UK units, all the sizes of his UK or US-sized ratings will be scaled down 2 units.
- Aggregation: With a newspaper, you have to go through all individual stories. Ratings, however, will be aggregated according to last/width combinations and sizes. Just like it's been done now. The idea is the more evidence for a good fit, the higher it should be in your list of results.
How it works
Just imaging think of a shoe that fits you very comfortable. So, you probably rated it with 4 stars. If it's a popular last/width combination, there are other members who likewise rated it with four stars.
Since you and all these other members have something in common, there's a good chance you want to follow them – possibly with an offset if they rated a different size than you did. This makes it possible to take all the last/width combinations they ever rated, scale the sizes according to the offset, and provide you with a combined list of fittings, ordered from best to worst according to the available evidence.
Another way to put this is to think of clubs. Being able to say "A shoe on Loake's Capital last in F fits me very comfortably" gets you the membership in a club in which all members who can say the same thing. While you're there, everyone puts his experience on the table, so to speak, compares them, and so everyone can walk away knowing more about the fit of other last, widths, or sizes. As a result, everyone knows a quite a bit more.
The more shoes and lasts you can rate as "comfortable", the more clubs you'll join. Thus, you – or rather the software – will be able to pool all the knowledge from all the different clubs, and this is the evidence on which your recommendations will be build.
The same principle applies if you provided measurements. Only the club is defined by people who have similar measurements in common. In fact, the new method is very much like the 'Personal fittings'. As a result, you get results very similar to 'Similar fittings' but the hard-to-understand tables will be gone. Instead, you get a list of individual ratings and annotations if you want to check why a fitting has been recommended.
Aside from being much more intuitive, in my opinion, there are additional advantages:
- Suggestions for finding better fitting lasts. If your first try in buying a quality shoe turns out to be less than perfect, the current methods leave you in a kind of limbo. There are no recommendations for last/width combinations that might fit better. The new method will change this, if you rated at least one shoe with 3 stars, if it's build on a somewhat popular last/width combination.
- More individual control. You'll be able to stop following another member, if you have the feeling you don't want or should follow him. Why would you want to do this? The previous scenario is one example. Say, you can only rate one pair of shoes with three stars – a fairly comfortable fit. While there are probably lots of other members who did the same, their reasons for doing so might differ from yours. For instance, another member may have found them too wide, while you may have found them too narrow. Right now, the software can't detect these differences. But you can. And your feedback will one day make it possible for the software to detect these differences, too.
- More results for EU-sized shoes. There will be some cases where you can follow another member with an offset of zero. If there is no offset, his ratings for EU-sized shoes don't need to be scaled, so they can be used as they are. This also works the other way around: If your only experience is with EU-sized shoes, there's a better chance, you can follow someone else with an offset of zero, so your recommendation will also include UK- or US-sized fittings, and shoes.
As far as I can tell, there are only two small disadvantages. The first is that you'll possibly have to adjust the "size filter" setting. It'll only be available if you submitted measurements, but you may not have noted it if you mostly used the 'Similar fittings' method.
Its default will be set to +/- 0.5 units from your reference size(s). This means, if your reference size is a UK 8.0, recommendations will only cover last/width combinations that would fit in sizes between UK 7.5 to 8.5. For most people, this is a sensible default, but if you have (very) wide or narrow feet, you may want to relax the setting.
The second is a small decrease in privacy. To be able to say whether you want to continue following another member, you have to have some way to see what he rated and how. Obviously, others need the same opportunity, so they will be able to see what you've rated and how.
To keep this to a necessary minimum, only members who follow you will be able to see your ratings. Additionally, they will only see your ratings for lasts they also rated. Likewise, you will also only be able to see the ratings of those you follow, and only for lasts you've rated, too.
If you haven't provided measurements, yet, people weren't able to see any of your ratings. Obviously, this will be different, then. In contrast, if you provided measurements, this makes hardly any difference, except some of your ratings may now be shown together in one place.
If you, for some reason, disagree with this, you have to cancel your account. A simple e-mail will be sufficient. No questions asked.
I've spend the last two months rewriting and amending the underlying software. It's quite a bit of work to do in one's free time.Almost everything is in place now. However, I've decided to rework some parts, again – namely the one based on measurements. I'll expect this to be finished in about two months or so.
To not let others wait any longer, I will nevertheless upload the current code. Unless something unexpected happens, it'll be available on March, the 11th. In my opinion, the advantages of the current changes justify the decision. An update of the documentation (help, etc.) will follow shortly.
If you have not provided any ratings, yet, there'll be a link to the old method of "Personal fittings". Or you can simple submit a rating for your best fitting shoes and see how it works.
As usual, I welcome feedback.