Thursday, September 23, 2010

San Francisco school choice goes in-house

Those of you who have been following school choice developments here know that, for the past year,  Atila Abdulkadiroglu, Clayton Featherstone, Muriel Niederle, Parag Pathak and I have been helping the San Francisco Unified School District design a new school choice system, which was adopted by the SF School Board last March.

The original plan was that we would continue to offer our services free of charge to implement the software, and then help monitor the effects of the new choice system.

Last week we heard from SFUSD staff that, because of concerns about sharing confidential data for monitoring the effects of the new system, they have decided to do further development in-house, and so will develop software to implement the new design on their own.

The SFUSD staff  have  been left with a sufficiently detailed description of the "assignment with transfers" design the Board  approved to move ahead with it if they wish. But it will take a good deal of care in implementing the new algorithm in software if its desirable properties--strategic simplicity and non wastefulness--are to be realized. (Both of these features were lacking in the old SFUSD assignment system, the one to be replaced.)

Below are links to some of the key developments before last week.

Here is a post with a link to the video of Muriel Niederle presenting the new design that the Board ultimately voted to adopt: SF School Board Meeting, Feb 17: new choice system.
And here is a link to the slides she presented, giving a description (with examples) of the new choice algorithm: Assignment in the SFUSD, and discussions of the features that make it strategically simple, non wasteful, and flexible.

In March 2010 the San Francisco Board of Education unanimously approved the new system. In their March 2010 press release (now here), the SFUSD reported (emphasis added):

"The choice algorithm was designed with the help of a volunteer team of market design experts who have previously been involved in designing choice algorithms for school choice in Boston and New York City. Volunteers from four prominent universities contributed to the effort, including Clayton Featherstone and Muriel Niederle of Stanford University, Atila Abdulkadiroglu of Duke University, Parag Pathak of MIT, and Alvin Roth of Harvard.
We are pleased that the district has decided to adopt a choice architecture that makes it safe for parents to concentrate their effort on determining which schools they prefer, with confidence that they won’t hurt their chances by listing their preferences truthfully,” said Niederle and Featherstone, the Stanford research team."

A very simple description of the basic assignment with transfer algorithm is given on page 370 of this paper (where it is called "top trading cycles"). Much of the SFUSD school Board debate has focused on what priorities to implement. What is described on page 370 is a simple version of the underlying choice engine into which the priorities go (with a somewhat different description than in Muriel's slides). The priorities can depend on any characteristics of the students (e.g. previous schools, or siblings, or home zip code) or of the school (e.g. neighborhood or historical student composition). But to keep the process strategically simple--to make it safe for families to rank schools according to their true preferences--the priority of a student at a school cannot depend on how that student ranked that school. (If you happen to be the programmer asked to implement this, drop me a line if you run into trouble:)

Some related developments can be followed on the blog of SF Board of Education member Rachel Norton, including this September 15 post on delay in the implementation of the middle school assignment plan: Recap: Assignment committee recommends delay

General background on the theory and practice of designing school choice algorithms can be found here, and my earlier posts on San Francisco schools are here.

Update, 9/30/10: many comments followed the link to this post at The SF K Files
Update 10/2/10: Rachel Norton, the SF Board member/blogger gets emails asking if the new system will be strategy proof, and she says it will be: Reader mail: questions on student assignment


Stan Goldberg said...
This comment has been removed by the author.
Stan Goldberg said...

will any of the parent programers or designers on this project who were excluded by the action of the school district contact me off list? I wish to follow up on the issue.

svg @

KWillets said...

A few questions (I will likely have more) -- When two kids want the same Top Trading Cycle, is the lottery number used to rank them?

At what point do the parents have to rank all the schools (I saw a mention of later rounds)?

KWillets said...

I've looked this over in more detail, and it appears that there is a huge amount of regional bias in the TTC mechanism. When all other factors are equal, a student in a popular school is more likely to get a TTC trade than a student in an unpopular school. So students living near popular schools not only have priority there, but also at other schools.

Tung Wai Yip said...

I'm a San Francisco parent going to enroll my kid next year. I've follow the design process, in particular to the algorithm proposed. But I am still horribly confused. WHAT IS THE NEW ASSIGNMENT ALGORITHM?

I've garnered these pieces:

1. SFUSD's priority

2a. Muriel's powerpoint (slide 20)

2b. Top trading cycles as in the Boston Public School Match paper

2a and 2b look rather different to me. In additional 2a's trading take both school and student's preference into account. But Muriel's transfer process (slide 17-19) seems to take only students preference into account. Another question is on Muriel's slide 20 point #1, it says "highest preference". But the term seems to be ambiguous. Is it highest preference to the student or to the school?

The press and the public seems to mostly concern about the priority list of 1. No one seems to care or understand what the assignment (e.g. 2a or 2b) process is really like. Can you tell me what exactly is the assignment process?

The SFUSD's FAQ is also confusing.

Point 6: If you do not get one
of your choices, you will be
offered your attendance area
school if it has openings.
Otherwise you will be offered
the school closest to your
home with openings.

Note: If these tie-breakers
do not resolve ties, then ties
will be resolved by random

It implies the priority matching mechanism in the Boston paper that you say should not be used. I think SFUSD people are as least as confused as I'm.

@KWillets, good observation !