Match Algorithm

[mmazurek]

In an effort to gain insight into the best strategies to employ when assigning programs to a rank-order list, I developed a simple simulation that uses the match algorithm as described on the CaRMS website with a live demonstration using the sample data provided. I'm including a link to the project below in case anyone is interested in playing with the code to see how things play out when the variables are tweaked. From my experience—and perhaps, quite intuitively—it seems the best strategy is to rank ALL programs you would even remotely consider matching to, in order of descending preference, regardless of how strong of an applicant you are or how well you felt your interview went.

 

http://jsfiddle.net/hyVFP/

 

Cheers,

 

M

[Alastriss]

In an effort to gain insight into the best strategies to employ when assigning programs to a rank-order list, I developed a simple simulation that uses the match algorithm as described on the CaRMS website with a live demonstration using the sample data provided. I'm including a link to the project below in case anyone is interested in playing with the code to see how things play out when the variables are tweaked. From my experience—and perhaps, quite intuitively—it seems the best strategy is to rank ALL programs you would even remotely consider matching to, in order of descending preference, regardless of how strong of an applicant you are or how well you felt your interview went.

 

http://jsfiddle.net/hyVFP/

 

Cheers,

 

M

 

Thanks, but we all know already that the best strategy is to rank anywhere you want to go in descending preference. The big question you are considering not ranking a program is that you are saying you would rather go unmatched than go to said place in question.

[rmorelan]

Thanks, but we all know already that the best strategy is to rank anywhere you want to go in descending preference. The big question you are considering not ranking a program is that you are saying you would rather go unmatched than go to said place in question.

 

Yeah basically - they aren't lying in the manuals when they say that is the best approach. I for fun analyzed it a bit as well - from a Economics and Computer Science point of view it is optimally configured in the interests of the applicants. The only way to potentially do better than the strategy mentioned is to have access to more information than you actually have access to. Any approach to "game the system" actually results in sub optimal performance.

[medaholic]

Little trivia - the match algorithm used in CaRMS was developed by Lloyd Shapley and Alvin Roth who won the Nobel Prize in Economics last year

 

http://www.nobelprize.org/nobel_prizes/economics/laureates/2012/press.html

 

http://www.slate.com/articles/business/moneybox/2012/10/nobel_prize_in_economics_lloyd_shapley_and_alvin_roth_win_for_deferred_acceptance.html

 

Interesting that an algorithm that was figured out almost half a century ago is still solving today's problems.

[dan0105]

 

http://jsfiddle.net/hyVFP/

 

M

 

Did you program that yourself just for the match? Well done, I thought I was the only nerd who would do something like that...

 

On a different note, rank everything!!! Even if you don't want it very much, it is far more likely that those positions would be better than anything left open in the second iteration for you (because why didn't you apply to it the first time then?). And don't take chances with hoping some schools will have second iteration spots because last year they did, blah blah, most of the schools try very hard to make sure they don't repeat the same mistakes twice.

[anticipate]

In an effort to gain insight into the best strategies to employ when assigning programs to a rank-order list, I developed a simple simulation that uses the match algorithm as described on the CaRMS website with a live demonstration using the sample data provided. I'm including a link to the project below in case anyone is interested in playing with the code to see how things play out when the variables are tweaked. From my experience—and perhaps, quite intuitively—it seems the best strategy is to rank ALL programs you would even remotely consider matching to, in order of descending preference, regardless of how strong of an applicant you are or how well you felt your interview went.

 

http://jsfiddle.net/hyVFP/

 

Cheers,

 

M

Cool! I learnt something.

[Bawse]

Now here's a question for those who have a more intricate understanding of the match algorithm:

 

From the program's point of view, is it better to rank higher the BEST candidate or a lower candidate who will rank them first? IE: Do programs employ the same ranking strategies as med students??

[medaholic]

Now here's a question for those who have a more intricate understanding of the match algorithm:

 

From the program's point of view, is it better to rank higher the BEST candidate or a lower candidate who will rank them first? IE: Do programs employ the same ranking strategies as med students??

 

Yes, the algorithm tries to find a stable "match" so both sides get what they want, with a slight preference for applicants.

[rmorelan]

Now here's a question for those who have a more intricate understanding of the match algorithm:

 

From the program's point of view, is it better to rank higher the BEST candidate or a lower candidate who will rank them first? IE: Do programs employ the same ranking strategies as med students??

 

As far as I understand it yes they follow the same approach. First they have no true idea if someone will rank them first.

 

If you have more information it is possible to make better choices - but the trouble is you don't have access to that information e.g., if you knew for sure someone was ranking you first you could rank them first and get a certain spot - but and this is key you will never know that no matter what they say. People get burned when they try to game the overall algorithm.

[Alastriss]

As far as I understand it yes they follow the same approach. First they have no true idea if someone will rank them first.

 

If you have more information it is possible to make better choices - but the trouble is you don't have access to that information e.g., if you knew for sure someone was ranking you first you could rank them first and get a certain spot - but and this is key you will never know that no matter what they say. People get burned when they try to game the overall algorithm.

 

But they still look at likelihoods (Ie you have trained in city x, you have family and spouse in city x, you lived in city x, etc)

 

What I don't get is why they keep saying that CaRMs ranking works in favour of the student.

[dan0105]

I think that is from the concept that if a student could potentially go into TWO different programs, the student's list will determine which program they are placed in.

 

Ie, if program A ranks the student #6 and program B ranks the student #1,

and the student's preference was A,B, as long as there is space, its the student's list that decides, and the student goes to A.

[dan0105]

What I don't understand is how knowing what a program ranked you would help?

 

Say program X wants you, and even illegally contacts you to tell you that you are their #1 applicant. How does this change anything? If you would rather go somewhere else I don't see how your rank list would change... you would still just put X wherever in the order of preference, would you not?

[opsoclonus]

If you understand how the algorithm works then what the program says shouldn't influence your decision. Some people don't understand it though and think that if they change their list to rank that program first, they will have a better chance of matching. I don't know the reasoning behind that thought but it happens all the time in the U.S.

[dan0105]

http://www.dcs.gla.ac.uk/~rwi/SPA.html

 

A good article on the exact specifics of the match criteria. Participant vs Program favoured is discussed around paragraph 5. It might be a bit too nerdy with the graph theory for most.... be warned.

[thebouque]

There is no Da Vinci code. Both students and programs should rank in order of preference.