A very brief update: we’ve successfully moved to a cluster of servers (instead of just one). We switched from MediaTemple to Linode and couldn’t be happier. We had some unexpected downtime this morning, so apologies if you were greeted by an overexcited white page stating, “It works!”

Now that we can focus on writing code instead of trying to fix slowdowns, expect more updates in the near future. College notification deadlines are approaching (ides of March, anyone?) – good luck to those of you waiting for your decisions!

We’ve been running the free essay editing and feedback system on MyChances.net for over a year, and over the holiday we decided it was time to make improvements based on lessons we have learned.

When will my essay be rated?

This is probably the #1 question that people would ask about the essay rating system. The problem was, it was really hard to answer. Nobody could choose which essay to read – we assigned it to them – which should have meant it was easy to know when an essay would get rated. If we had assigned essays in order of submission, the answer indeed would have been pretty obvious. But we didn’t do that. We wanted to reward people who contributed often to the site. Therefore, we assigned essays in the order of the # of credits that their author had.

Let’s say that the queue had one person’s essay, and this person was a pretty regular contributor, having 25 credits. His essay has been rated 4 times already, needing just one more to be completed. All of a sudden, I decide to submit an essay. Since I have 2,500 credits, my essay now takes precedence over his. So despite being 80% complete, he will have to wait for my essay to become 100% complete before he can get that last rating.

As a consequence, it became almost impossible to see how long it would take before one’s essay would get rated. It was dependent on stochastic processes.

Let the market decide

We wanted our authors to be able to get a sense of when their essay would be rated, while still rewarding those who contributed to the site. Therefore, we decided to allow the authors to offer as many or as few credits as they wanted, and to choose how many reviews they wanted to receive. The raters, in turn, would be able to see a list of all available essays, and how many credits were being offered for each one.

In this way, the authors could help determine how soon their essay would get rated. If they wanted a quick rating, for example, they could submit an essay for just 2 reviews at 15 credits a pop, jumping to the top of the list and getting done more quickly.

The title is the thing

Now that our raters could select one essay among many, things started to look pretty boring with names like “Untitled 1852″. Sure, perhaps you could select essays based on how many credits they pay (I sure do), or based on how many reviews are left to complete, or even based on some sort of credit-per-character metric. But in the end, I wanted to know if I would be evaluating an essay that looked interesting to me. So I created a new, publicly visible title field. Now I just might choose to evaluate “On leadership and lollipops” at 5 credits over “Untitled 2858″ at 10.

Note: ‘publicly visible title’ means that the title is world-viewable, including to non-members and search engines, until the essay has received its final rating. At that point, the title becomes completely hidden. The essay itself, of course, is always totally hidden from public view, which brings me to…

Privacy is still king

Every step of the way during the system redesign, I asked myself, “Would I use this if I were a college applicant?” For me, the only way that the answer would ever be “yes” is if I felt that my privacy were adequately protected. Many students are interested in getting feedback, and almost all are concerned about privacy. They don’t want their essays showing up in search results, or becoming easy fodder for copy-and-paste plagiarism.

Essays are still never shown to guests. And, while most members can now provide feedback on any active essay that they choose, they still must complete their rating of each essay they select before they can even view another. There are also limits in place on the number of essays that any particular member can view in one day.

Whither?

To where from here? We need to allow people to evaluate the quality of essay feedback that they receive. More details once that is implemented.

As always, if you have concerns, complaints, or suggestions for improvement, please post them here. Thanks to all of the members who *are* essay reviewers and essay authors. I am consistently impressed both with the essays that get submitted, and with the insightful, sincere feedback. Keep it up!

We’ve loaded up the 2008-2009 official college admissions data, which you can see on our college rankings page and each colleges’ pages. Certainly, other people were aware of this about a year ago, but the number of applicants to Tulane skyrocketed in 2008 (to over 35,000), and apparently hit nearly 40,000 in 2009. Talk about a turnaround after Katrina!

Have you tried to post on another person’s wall but find that your post never appeared? Have you received an email telling you that there is a new post waiting for you, but once you click through, you find nothing?

We ran out of integers to uniquely identify wall posts (oops!), but we bought some more, just in time for the holiday season.

2006 NYT College Preference Matchups

In 2006, New York Times columnist David Leonhardt wrote an article that was ostensibly about Harvard ending its early admissions program. The meat of the article, though, was far more interesting, filled with a discussion of college rankings*. A survey of 3,200 high school seniors from 500 high schools was performed in 1999. The students were asked about which schools admitted them, and which college they ended up attending. From this information, a model was created and used to calculate how often students would choose one college over another. See the NYT Chart for the results.

That data is now 10 years old, so over the past several months, MyChances developed tools to perform a similar analysis. For 2009, we surveyed 1,200 students at 950 high schools. How do our results stack up?

2009 MyChances College Preference Matchups

Let’s look at the same colleges as in the Leonhardt article, focusing on three examples: Harvard vs Stanford, Georgetown vs Brown,  and Duke vs Princeton.

73% of students would have chosen Harvard over Stanford in 1999. This is completely unchanged in 2009.

Among students admitted to Georgetown and Brown, only 22% were estimated to go to Georgetown in 1999. Now, our evidence suggests that Georgetown has pulled even with Brown, winning 53% of the matchups.

Finally, 91% of students admitted to Princeton and Duke previously chose Princeton. Now, we estimate that 35% would choose Duke – a far more narrow advantage for Princeton than before.

Though it is interesting to make some of the same comparisons as were made in the 2006 New York Times article, you’re not limited to those schools using our tools. We had enough data to rank over 200 colleges in 2009; hopefully we’ll be able to expand to 300 in 2010, ultimately providing estimates for all 1,700 colleges in our database in the near future.

* = Paper is “A Revealed Preference Ranking of U.S. Colleges and Universities” by Avery, Glickman, Hoxby, and Metrick

(Note: College Rankings #3, an exposition of our methods, has not yet been published.)

Tonight I added a couple of features to the scatterplots (scatter charts / scattergrams) that I introduced in my last post. The two new variables that you can now access are: acceptance status and prediction.

Let’s say you wanted to know how accurate our prediction engine is within one particular range of predictions. For example, you want to see what really happens when we claim someone has around 80-100% chances. Set the axes to Prediction, add a small amount of jitter, and you can now get a sense of how many miscategorizations we’ve made in that region. Take a look at Boston College to see an example.

You can now graph our predicted probability of acceptance on scatterplots.

Clearly we do a pretty good job at Boston College. For one, you can instantly see that the blue (accepted) applicants cluster on the right side with high predictions, while the red (rejected) cluster on the left. Additionally, look at the mean (average) lines. The mean prediction for accepted members was 83.3%, while that for rejected members was 35.6% – a very large difference.

Another way to visualize the data would be to set one axis to Accepted, and another to Prediction, so you get perfect separation between the blue (accepted) and red (rejected), and can perhaps more easily see how well our predictions separate the accepted from the rejected at any particular range of chances.

These updates were requested by Christian Romero; thanks, Christian.

We have rolled out our interactive flash scatterplots (also known as scattergrams), available on every college page under the ‘My Analysis’ tab.

These graphs display the accepted and rejected applicants scattered across a 2D canvas according to the variables that you choose. For example, you might look at Unweighted GPA & SAT, or Instate & Average AP Score. To get started with this new tool, see Cornell’s scatterplots.

For any given SAT score, valedictorians appear more likely to get into Cornell than non-valedictorians.

Because there are many, many overlaps, you can set a level of jitter, so each point floats near its true value. For example, if you look at Unweighted GPA and Valedictorian Status, everyone will clump on top of one another. (You either are a valedictorian, or you aren’t, so there are only 2 slots that you might possibly fit into – hence lots of clumping.) If you set a 20% jitter to Valedictorian Status, things will spread out nicely, so you can see what is really going on.

With your feedback and criticism (please post it here or in the forums), we’ll work on improving the tool. Enjoy!

In my previous post, I introduced the new college rankings system that we have implemented. In short, the system ranks colleges based on where their admitted students decide to attend. In this post, I will discuss some of the approaches that might be considered in creating a preference-based ranking. In the next post, I will discuss the preference ranking system that we have implemented.

2009 MyChances.net College Rankings

Yield isn’t enough

The goal of a preference-based ranking system is to capture people’s true preferences and represent them faithfully. To discover people’s preferences, a reasonable place to start might be a college’s yield. Yield is calculated as follows:

yield = (# of students attending) / (# of students accepted)

So how can we use yield to compare two schools? Suppose we match the University of Georgia (#70 on our list) against Pomona (#50 on our list). In this matchup, Georgia’s 55% yield actually beats out Pomona’s 39% yield. More of Georgia’s admitted students end up attending—so Georgia appears to be preferable to Pomona. But there is a problem here: we have no direct evidence that students, given the opportunity to attend either school, would choose Georgia. We simply don’t know what the students who were admitted to both schools would do.

In the abstract, there is another problem with this approach. Imagine that 100 students apply to both Georgia and Pomona. Suppose Pomona accepts 50 of them but Georgia accepts all 100. Now, suppose the 50 rejected from Pomona all decide to go to Georgia, giving it a 50% yield. Suppose, also, that 40 of Pomona’s accepted students also get into Harvard, Yale, or Princeton, and they all go off to those schools. This leaves Pomona with a 20% yield. Going by yield, it appears that Georgia is the preferred college by far—but in reality, all of the students admitted to both Pomona and Georgia who attended one of the two decide to go to Pomona. Yield, in this situation, gives us exactly the wrong answer about which school is preferred over the other!

Each student matters

What can we learn from the failure of yield as a measure of preference? Summary statistics simply don’t tell us enough. We need to drill down to the level of individual students. Only then can we build up a picture of their collective preferences. How can we do this? One approach might simply be to ask them what their preferences are. For example, we could survey a bunch of students applying to college, and ask them to order all of the schools they are considering, from most favorite to least favorite.

This is better; if those 100 people in our previous example honestly represented their preferences, we would probably see that the Pomona was preferred over Georgia. This is the intuitively correct result given our (fake) example. But even this approach isn’t perfect.

Talk is cheap; opinions, cheaper

One problem is that there is no cost associated with ranking a school #1 on your own personal list. Until you actually have to decide which college you are going to attend—and pay tuition to—for the next 4 years, your opinions have no teeth. Let’s say you rank UNC as your #1 school and Duke as your #8 (out of 8), because your Tar Heel family hates those Blue Devils. You apply, and get into both schools. Did I mention that you got a merit scholarship to Duke? All of a sudden, you find yourself attending your supposedly bottom-ranked school. You didn’t lie when you gave us your rankings, but you probably exaggerated how much you preferred UNC over Duke. Furthermore, you didn’t have all of the information that you used to make your decision—such as your merit scholarship—when you reported that Duke was your #8 school.

In general, asking people for their preferences leads to these additional problems:

• They may give feedback about colleges where their feedback is of questionable value. If someone with a 1.5 GPA says that they rank State U over Harvard, should that hurt Harvard—even though this person almost certainly wouldn’t be given the opportunity to attend there, anyways?
• They almost certainly give feedback that is based on imperfect information. At the moment where people are making their decision to attend one school out of several that they were admitted to, they have acquired as much information as they think they need to make this huge decision. Beforehand—and, in particular, before they have applied to and been admitted to colleges—their stated preferences may be much more labile.

Understanding these flaws helps flesh out a framework for a powerful-yet-simple preference-based college rankings system: one where students simply report where they were admitted and where they decided to attend. In my next post, I’ll get into some of the details of how to take this information and construct a ranked preference list. I’ll even demonstrate how this approach addresses a common criticism of the currently popular college rankings: that there is no way to truly distinguish between schools closely ranked (e.g., #3 vis-a-vis #5).

Presenting: our new college rankings.

The college admissions landscape is littered with college rankings. In 1983, US News first ranked American universities. Since then, rankings have been a fixture of the college world: they are produced by various businesses (US News, Princeton Review, Forbes, Atlantic Monthly), and heeded by students and colleges alike. To gain advantage, some universities have been alleged to manipulate their own rankings. And, while some of the factors used in the rankings are justifiable (alumni giving rate), some seem to be arbitrary (peer assessment surveys asking other colleges about your college’s ‘faculty dedication to teaching’). Each year, the methodology changes slightly, producing a slightly different list. In the end, the factors that are used to come up with the rankings seem arbitrary; the occasional change in the weighting of each factor, capricious. There is a need for a new approach.

Criteria for a ‘good’ college ranking system

1. The system should be difficult to game; any ‘gaming’ of the system should actually benefit students. In contrast, consider the allegations that some schools tried to manipulate the US News rankings by encouraging more students to apply in order to decrease their acceptance rate.
2. The factors measured should be relevant to students. In contrast, what Cornell’s dean thinks about the faculty dedication at the University of Texas may be irrelevant.
3. The overall procedure for generating rankings should be stable from year to year. In other words, any change in the rankings between 2008 and 2009 should be explained by a substantive change in the underlying factors, not by an arbitrary change in how those factors are weighted.

The MyChances College Rankings

We have implemented the MyChances College Rankings based on revealed student preference. In this system, the college admissions process is treated like a chess tournament. The colleges play matches (which occur when 2 colleges admit the same student). In each match, there is a winner (the college that the student ends up attending) and a loser. The winner gains points; the loser forfeits them. When a high-ranked school beats a low ranked school, the high-ranked school gains few points, and the low-ranked school loses few points. If a low-ranked school beats a high-ranked opponent, it gains more points than if it beat an equally-matched opponent. After playing many games, the colleges that students prefer rise naturally to the top of the rankings.

Does the method of revealed student preference meet the 3 criteria outlined above? I believe it does.

Consider point #1 (gaming the system). Imagine that MIT wanted to beat out Harvard by trying hard to avoid admitting any students that they thought would be admitted to Harvard. They would end up succeeding in a model based on acceptance rate and yield (since their yield would likely increase), but their actual student body would be less qualified. In the revealed preference model, however, they would be less successful. They would not compete head-to-head with Harvard, so would ‘win’ more. But they would be winning against weaker ‘opponents’, earning fewer points for each victory.

For point #2 (relevance), the idea of revealed preference is that it aggregates the sum total of what matters to students – whatever those factors might be. It is likely that students behave rationally (by attending the school that they find most desirable). So long as other students share similar values, then revealed preference rankings will work well in explaining, and even guiding, their decisions.

For point #3 (stability), the tournament style system is simple and straightforward. It is responsive to changes in student preference over time. It does not rely on aggregations of various statistical factors, or college faculty survey results; nor does it depend upon arbitrary weighting of those factors.

The details of the procedure that we use to generate the rankings, and our use of chess-style Elo points, will be explained in a later post. For an academic treatment of a similar college ranking system, I recommend the working paper, “A Revealed Preference Ranking of U.S. Colleges and Universities,” 2005, by Christopher Avery, Mark Glickman, Caroline Hoxby, and Andrew Metrick (free link).

Today we’re letting everyone in on a sneak-preview of our latest tool: the college cross-admit preference tool. We think it’s a simple but powerful way to see which colleges are most favored by admitted college students.

To use it is simple: type in the names of two colleges that you want to compare (perhaps Florida and Florida State?). You’ll then see which fraction of site members prefers which school. Preference is determined by the relative fraction of members admitted to both schools who end up attending one or the other. For example, if 25% of students admitted to both College A and College B ultimately go to College B, we say they prefer College B over College A. When the results are statistically significant at the 95% level, you’ll see the results lit up in bright colors.

For the hardcore college admissions followers out there, this will remind you of this graphic from a 2006 NY Times article. One difference is that our list isn’t limited to 17 schools; as the data continues to become available, we’ll display this information for all 1700 schools that we track.

Requests? Feedback? Suggestions? Let us know.