How to read relevant similarities

When I look at the recommendations for the Similar Fittings, I see 2 tables. One is Your Ratings, which I understand is what I have entered in the system. But I can't understand the information in the Relevant Similarities table. Can you explain how to read the headings and how it is related to my entries.

Example, my rating reads:

1-511 AE in C, size 8.5, Score 4
Leydon Adlend in D, size 8.5, Score 2

Relevant Similarities:

1-511, size Diff: -1.0, Rating Diff: -1.0 # Comp 1
Leydon size Diff: -1.0,Rating Diff: -2.0,# Comp 1

Thanks.

8 comments:

Avatar

Claus Sept. 19, 2013

Yes, sure. No problem.

You probably noted these are the details for a particular last/width combination you didn't mention. Let's call this the target. The question is what can be said about the fit of the target for a particular size, given your ratings and the data from other members?

The answer thus uses your ratings (table 1), and comparisons extracted from the ratings of others (table 2). In other words, table 2 lists the comparisons from other members about AE's 1-511 last in C to the target (line 1), and from Alden's Leydon last in D to the target (line 2).

It's probably easier to understand when you imaging both tables merged into one. It's not always possible, but it works in your case. Then, it reads:

1. Based on your first rating and the first comparison, the target's fit in 7.5 (your rated size of 8.5 plus the size difference of -1) is estimated to be 3 stars (your 4 stars plus the rating difference of -1)

2. Based on your second rating and the second comparison, the target's fit in 7.5 (your rated size of 8.5 plus the size difference of -1) is estimated to be 0 stars (your 2 stars plus the rating difference of -2).

Since both estimations are for the same size, an weighted average is used to combine both. In your case, this is (3 + 0) / 2 = 1.5. In other words, the target in 7.5 (UK I presume) is unlikely to fit well according to the existing data.

To be fair, however, it should be noted that the method is likely to produce false negatives (lasts may fit well although they are estimated not to do so). The advantage is that false positives are relatively rare (lasts that do not fit well although they are estimated to do so).

If you have no other information, and you really want a shoe on the target last, you may want to ignore the second line. Then, the estimated fit for the 7.5 size would be 3 stars and this may be good enough for you.

Hopefully, this helps. If not, please let me know and I'll try once more.

Avatar

green_garden Sept. 19, 2013

Claus, thanks for the explanation. A little bit more clarification.

The weighted average, is this how the score for the target shoe gets calculated? Unfortunately, I can’t seem to locate the exact source of the data I sent over so I can’t verify if the 1.5 average shows up anywhere. So let’s assume I was looking at C&J 314 last in 7.5UK. The overall score for the C&J 314 should be 1.5 stars, right?

I’m still trying to understand how you can come-up with the recommended comparison. To me you need to have 3 data points:

1. My rating to a control shoe – 1-511 – additionally, you need the size I wear and my rating – 8.5C, 4 stars
2. Other’s rating to the control – 1-511 - did the person enter a rating for a 8.5C also?
3. Other’s rating to the target – C&J 314 - did the person enter 7.5K UK?

If you don’t mind, can you then use the above to illustrate how you generate the size Diff and Rating Diff data?

The size entry for #2 and #3 and how it relates to the size I entered for #1 is interesting to understand. Reason is, typically I can ask somebody to tell me how they rate a 10.5C AE 1-511 lasted shoe and then ask them what size they wear on a C&J 314 shoe and how they rate that shoe. So if they say they rate the 10.5C a 4 star and wear a 9.5 UK for C&J a 4 star, then I would be fairly confident that I can wear a 7.5UK with favorable results. I was wondering if the system does this or only looks at people who actually entered the same size as my control shoe.

I hope I’m not asking too many questions. Just want to understand the details because I don’t think I can totally rely on the weighted average. It’s better to understand shoe to shoe compare since I know which of my rated shoe fit the best.

Avatar

green_garden Sept. 19, 2013

Claus,

Please use the below as a replacement to the 2nd to the last paragraph. I edited to make my point clearer:

"The size entry for #2 and #3 and how it relates to the size I entered for #1 is interesting to understand. Reason is, typically I can ask somebody who wears a AE 1-511 shoe to tell me how they rate the fit. If he wears a 10.5C AE 1-511 lasted shoe, I could then ask them what size they wear on a C&J 314 shoe and how they rate that shoe. So if they say they rate the 10.5C a 4 star and wear a 9.5 UK for C&J a 4 star, then I would be fairly confident that I can wear a 7.5UK with favorable results. I was wondering if the system does this or only looks at people who actually entered the same size as my control shoe."

Thanks again.

Avatar

Claus Sept. 19, 2013

In principle, this is indeed how it works.

For instance, another member rated, say, a US 10.5 (with 4 stars) for the control (AE's 1-511 in C) and the target in UK 9.5 (C&J's 314 last in E, I presume) with 3 stars. This is where the size difference of -1, and the rating difference of -1 comes from in the second table, first line.

So yes, the other member would have told something like this if you'd asked him. He'd say: "If AE's 1-511 in C fits you perfectly fine, go down one size and the fit for C&J's 314 is 3 stars, in my experience."

In other words, a UK 7.5 for C&J's 314 last would probably fit you somewhere around 3 stars, at best, given our current data.

(Actually, come to think about it, using the maximum instead of a weighed average might be a good idea.)

But this is not generated from somebody who rated the same sizes as you did.

Avatar

green_garden Sept. 20, 2013

Claus,

Perfect. Thank you for explaining.

Based on this, I would actually look at the ratings where I'm confident of the fit. In my case, the weighted average went down because the system calculated my bad fitting shoe together with a good fitting shoe. I would have discarded the ratings for the bad fitting shoe anyway.

A few thoughts for improvement:

1. I know that you already ask people for their overall rating. But I think it would be good to know if they found the shoe too big or too small. Right now when I'm comparing a not so good fit for me (anything below 4) against another not so good rating, it is actually helpful to compare if the other person thought it was too big or small. I think that helps the overall compare. Because if my not so good shoe is too big, and the other person's rating is too small, then maybe the target shoe would be a match for me.

2. In the "Personal Fittings" section, I wish there is a way to query by Member #. I am finding that there are members who's measurement are very close to mine. If I can query all their ratings, it may help me with comparisons also. Right now, I have to go into each rating and check if the member is there.

Thanks again. I hope to take advantage of this great resource.

Avatar

Claus Sept. 22, 2013

green_garden,

thank you for your suggestions. The 'too small/too large' suggestion has already been raised by someone else. I'm afraid, changing the rating scores is impossible, for the old ones would have to be converted or discarded. We'd probably loose more ratings than the information we'd gain.

I've been skeptical about quering members in the past [1], but I've started testing it. The tables under 'Similar fittings' are obviously not helpful enough, so I'm currently looking for a better way to present the information.

Don't forget to enter a rating when your C&Js arrived! :)

[1] http://sizeadvisors.com/feedback/44/

Avatar

green_garden Sept. 23, 2013

Hi Claus,

I wasn't suggesting to re-do the whole scoring. You know how you have annotations for every score, I think you can just add another field to capture "too big or too small" and show it as a second annotation. Maybe I'm thinking of this too simplistically.

I've been wanting to get a C&J captoe. Based on the info on the site, I may take the jump on an Audley. Still considering it.

Avatar

Claus Sept. 25, 2013

Right after the start of Sizeadvisors someone suggested additional fields like "Too wide", "Too small", etc. and I added the respective fields. However, participation was low, ie. almost noone used them.

I thus replaced these fields with the 'annotations' field you see now. This works much better.

Unfortunately, I haven't found a way to display annotations for the 'Similar fittings' recommendation method. I think, I'll need a different design for the site to do that.

I'm working on it.

Add Your Comment

Only members may post comments. If you'd like to post a comment you must log-in or register.

Wanted ratings

  • Leiden by Allen Edmonds on an unknown last by Allen Edmonds in US 11.0
  • Chester (Rubber) by Loake on the 024 last by Loake in UK 9.0-F
  • Stow by Tricker's on the 4497s last by Tricker's in UK 7.5-5
  • Truman by Truman Boot Co. on an unknown last by Truman Boot Co. in US 7.0
  • Fifth Avenue by Allen Edmonds on the 65 last last by Allen Edmonds in US 8.0-E

More wanted shoe ratings »

Get shoe fit estimates out of

5,958 ratings.

Add yours to improve results. Click here to join. It's free.