The downside of the DC school lottery

[Anonymous]

Anonymous wrote:Good god, this idea to basically replicate the old system again.

It’s Groundhog Day with these people.

The only benefit of the old system was that if you had a sucky lottery number it took longer to find out. With each school managing its waitlist there would still be movement through October, the notorious "October Shuffle." You had hope longer.

OP's proposal isn't about giving better outcomes to everyone, it's about making people who get bad draws not find out about it right away.

[Anonymous]

Anonymous wrote:The real problem is that the school lottery is a repeated game with highly correlated outcomes and it is being modelled as a single outcome. As pointed out by a previous poster, it also exaggerates the idea of the ordering of choices being meaningful when there are really tiers of preference. For example, if the preference between two choices is either zero or swamped by the noise of uncertainty, then this heavy emphasis on efficiency with respect to trading after assignment is silly. Ordinal ranking just isn't communicating all the information that parents have compiled.

Hmm.
Roth et al didn’t seem to write a lot about the DC lottery. They describe the N.Y. lottery, but in that algorithm, schools rank as well as students (similar to the medical match that they developed earlier.) So it’s hard to know for sure what the DC algorithm does. (If anyone has seen any description of the detailed algorithm of the DC lottery itself, please send.)

That said, the system seems to be a deferred-acceptance matching system that uses multiple rounds of matching before assigning all outcomes.
It does model multiple possible outcomes via the multiple rounds of matching.

But as OP said, the current algorithm uses one lottery number for each child each year.

You raise another issue with the current lottery— that there’s no way for students to communicate ties in preference.
In practice, I don’t think this influences the match much. I don’t see a way to use tie information to improve the quality of the matching. But you’re totally right that it gives parents a false sense of agency. It would probably be better for parent psychology if there was a different way to indicate beyond mere ordinal ranking.

[Anonymous]

Anyone who read the OP and thought “That’s the Old Lottery System” should probably bow out of discussion of the OP.

Because the post said, and now many replies have noted, that it is just a small modification of the current lottery.

[Anonymous]

Could we FOIA to see the actual code that executes the lottery? Has it been published?

[Anonymous]

Anonymous wrote:

Anonymous wrote:The real problem is that the school lottery is a repeated game with highly correlated outcomes and it is being modelled as a single outcome. As pointed out by a previous poster, it also exaggerates the idea of the ordering of choices being meaningful when there are really tiers of preference. For example, if the preference between two choices is either zero or swamped by the noise of uncertainty, then this heavy emphasis on efficiency with respect to trading after assignment is silly. Ordinal ranking just isn't communicating all the information that parents have compiled.

One legit complaint about the current lottery is that it has no way of accounting for intensity of preference. If I slightly prefer A to B but you strongly prefer A to B, then utility would be maximized by giving you A and me B, even if that's not a trade I would voluntarily agree with. The problem is there's no way of formalizing that and trying to do so opens all sorts of avenues for gaming.

The current system doesn't deal with ties well either. Imagine there are schools A and B, equivalent in all ways but some distance apart. I live equidistant between them so I am indifferent, but you live within walking distance of A and strongly prefer it. I list my choices as A then B, strictly because of alphabetical order, and I have a higher lottery number so I get A and you get B. We could trade and both be better off (I would have the psychic income of helping you out).

But these are edge cases.

These are interesting points. One way to address intensity would be to give each kid 100 points which they could distribute however they want among up to 12 schools. The school I assign the most points would be my #1, etc. That way there would be both a ranking and intensity would be shown by the varying number of points. I recognize this is probably not practical and too confusing, but an interesting thought. Also not a statistician so can't say how an algorithm would process the point numbers.

[Anonymous]

Actually, you could go the other way and get a single lottery number for your whole time in OSSE schools. That way you could decide whether to move right away rather than waiting through several lottery years to find out your outcome.

[Anonymous]

Anonymous wrote:Anyone who read the OP and thought “That’s the Old Lottery System” should probably bow out of discussion of the OP.

Because the post said, and now many replies have noted, that it is just a small modification of the current lottery.

OK, OP.

I am sincerely asking, how does this take into account the overall rankings of you choices? If it doesn't, then it's just the old system under one umbrella.

[Anonymous]

I do not understand why there are so many slots per entry. Does it increase or decrease your chances of getting in if you do not use all the slots? And why do people even put in slots that they don’t really want?

[Anonymous]

Anonymous wrote:The system the OP is describing is (I think) pretty close to what DC used to do when each school ran its own lottery. it was a mess. The biggest issue from an economics point of view is that it led to a situation where there could have been a lot of mutually beneficial trades -- which means it was inefficient at allocating a scarce resource. For example under the old system it was entirely possible for the following scenario to take place:

KidA gets into MV and has a bad waitlist number for IT, his parents prefer IT

KidB gets into IT and has a bad waitlist number for MV, his parents prefer MV

Under the new system, that won't happen because the parents will rank their choices and if KidA has a good number, he will rank IT first and get in there. KidB would get into MV with a good number.

Ding ding ding ding! This is the correct answer. All the rest of you are wrong. OP, you’d need to argue against this suboptimal outcome. You can’t. You lose.

All the rest of you are also wrong.

[Anonymous]

They could tell you your master lottery number before you submit so you could at least focus your mental energy on the tiers of schools that you are likely to get.

[Anonymous]

Anonymous wrote:I do not understand why there are so many slots per entry. Does it increase or decrease your chances of getting in if you do not use all the slots? And why do people even put in slots that they don’t really want?

It increases your chances of getting into a school if you can list more schools. It will not affect how good your number is. But we matched with our #12 choice. If we could only list 10, we would have had no match. We did really want it, just not as much as we wanted the 11 above it. But we enrolled and it has been good so far.

[Anonymous]

Anonymous wrote:

Anonymous wrote:The system the OP is describing is (I think) pretty close to what DC used to do when each school ran its own lottery. it was a mess. The biggest issue from an economics point of view is that it led to a situation where there could have been a lot of mutually beneficial trades -- which means it was inefficient at allocating a scarce resource. For example under the old system it was entirely possible for the following scenario to take place:

KidA gets into MV and has a bad waitlist number for IT, his parents prefer IT

KidB gets into IT and has a bad waitlist number for MV, his parents prefer MV

Under the new system, that won't happen because the parents will rank their choices and if KidA has a good number, he will rank IT first and get in there. KidB would get into MV with a good number.

Ding ding ding ding! This is the correct answer. All the rest of you are wrong. OP, you’d need to argue against this suboptimal outcome. You can’t. You lose.

All the rest of you are also wrong.

No, you are wrong. The DC Lottery implementation already requires gaming because only 12 options are listed rather than every option. Also, the "game" is repeated with high positive serial correlation whereas the DC Lottery implementation and associated proofs of efficiency assume only a single "game". And many folks have pointed out that defining efficiency with respect to a single set of ordinal preferences given by parents with limited time and bounded and uneven access to information means that this particular view of evaluating the process is dubious. But hey, abstract economic thought is always perfect, right?

[Anonymous]

Anonymous Also, the "game" is repeated with high positive serial correlation whereas the DC Lottery implementation and associated proofs of efficiency assume only a single "game". [/quote wrote:

Please expand on this thought.

[Anonymous]

Anonymous wrote:Also, the "game" is repeated with high positive serial correlation whereas the DC Lottery implementation and associated proofs of efficiency assume only a single "game".

Please expand on this thought. (Formatting corrected)

[Anonymous]

Anonymous wrote:

Anonymous wrote:

Anonymous wrote:The real problem is that the school lottery is a repeated game with highly correlated outcomes and it is being modelled as a single outcome. As pointed out by a previous poster, it also exaggerates the idea of the ordering of choices being meaningful when there are really tiers of preference. For example, if the preference between two choices is either zero or swamped by the noise of uncertainty, then this heavy emphasis on efficiency with respect to trading after assignment is silly. Ordinal ranking just isn't communicating all the information that parents have compiled.

One legit complaint about the current lottery is that it has no way of accounting for intensity of preference. If I slightly prefer A to B but you strongly prefer A to B, then utility would be maximized by giving you A and me B, even if that's not a trade I would voluntarily agree with. The problem is there's no way of formalizing that and trying to do so opens all sorts of avenues for gaming.

The current system doesn't deal with ties well either. Imagine there are schools A and B, equivalent in all ways but some distance apart. I live equidistant between them so I am indifferent, but you live within walking distance of A and strongly prefer it. I list my choices as A then B, strictly because of alphabetical order, and I have a higher lottery number so I get A and you get B. We could trade and both be better off (I would have the psychic income of helping you out).

But these are edge cases.

These are interesting points. One way to address intensity would be to give each kid 100 points which they could distribute however they want among up to 12 schools. The school I assign the most points would be my #1, etc. That way there would be both a ranking and intensity would be shown by the varying number of points. I recognize this is probably not practical and too confusing, but an interesting thought. Also not a statistician so can't say how an algorithm would process the point numbers.

Then you open the door to strategizing. Should I load my points into a hard-to-get-into school, or should I spread them around? Once you allow strategizing you open the door to suboptimal outcomes.