those graphs of schools, GREs, and GPAs

[untitled]

Last admissions season, I created some graphs and analyses of previous years' admissions. The site went down right after everyone's decisions were made. I had planned on making some substantial updates, but, as you can guess, never did. Even so, I've just put the site back up (updating the data as best I could), and hopefully everyone enjoys it. I always plan on cleaning up the code and publishing it, but never had any serious interest so never put in the minutes. However, its a pretty simple perl script, and let me know if you are interested in helping out.

I originally did these while I was finishing up my applications, and so I did not use them to help choose to which schools I applied, and I hope you are all very careful in how you use this source (as well as rankings, advice, etc); I try to be upfront about what I am doing, and I try to avoid any bias, etc. I think the site explains how it works, but am happy to explain if there are any questions (sorry to sound so didactic, but I really enjoyed doing this, and want to be as useful to the next class as possible). Please feel free to give me feedback; I've immensely enjoyed implementing others' suggestions (as always, previous applicants, please let me know if you want your information removed or detached from your user info or anything).

[thesparky]

Lot of effort!

Just briefly checking some random things.

It seems like you might have to convert international GPAs into the 4.0 scale, otherwise you have, for instance, somebody with 4.3/5.0 (Sonaar) artificially higher on the GPA scale.

Also, when there are a lot of candidates, they crowd out the graph and it looks really tiny (eg. Berkeley).

[dertliali]

perfect!
I like the graphs and analys, i guess it takes serious time but thank you

[untitled]

Originally Posted by thesparky

Lot of effort!

Just briefly checking some random things.

It seems like you might have to convert international GPAs into the 4.0 scale, otherwise you have, for instance, somebody with 4.3/5.0 (Sonaar) artificially higher on the GPA scale.

Also, when there are a lot of candidates, they crowd out the graph and it looks really tiny (eg. Berkeley).

International GPAs do not always convert, so you have to look at outliers and figure out where to put them yourself. Also, there are some US schools that allow greater than 4.0 on a 4.0 scale. Really, it is very hard to compare GPAs across schools (especially as we usually keep our UG institution private), so that dimension is really rough all around. In cases where a conversion was included, the script sometimes catches it. And in cases where the gpa recorded was outside of a certain range, the script just scored it as a 4.5 or 3.0 so at least the dot exists.

I'll look at the crowding issue, I could make the graphs bigger, but I don't think changing it a lot will look good, and I don't think changing it only a little will help much. I plan on creating seperate graphs for 2007, 2008, and 2009, which might help.

[GymShorts]

You deserve an award for this. Do you plan on turning this data into an academic paper?

[xbar]

Future app cycles are indebted.

Cheers

[KGkhan23]

I love that the title on the firefox tab is "Untitled" .

[untitled]

Originally Posted by xbar

Future app cycles are indebted.

Cheers

Thanks. I saw your post count was 3, and thought that there must have been a new xbar, as I've seen xbar so much in the analysis, and could hardly believe that you never posted except your profile. I really appreciate everyones contributions, even just as a data point or an audiance.

Originally Posted by KGkhan23

I love that the title on the firefox tab is "Untitled" .

I'm a bit of a megalomaniac, I love having my name all over my stuff ... except ... anonymous forums ... ::sigh::

[untitled]

There is quite a bit of analysis that we can do with all of this, some of which has been discussed before. I don't know if many are interested in discussing this, but GymShorts sent me some interesting points in a PM.

Originally Posted by GymShorts

There are two kinds of analysis that I think could be possible.

First, you could use the data and do some kind of regression to obtain estimates for how much different factors contribute to one's chances of being admitted. I think TMers would fully support this effort. Imagine if you estimated an equation with coefficient. Then when a future students comes, you could say "Given your profile, taking real analysis will increase the probability of you getting into a top ten school by X percent."

Second, you could do a ranking of how difficult a school is to get into. I believe you have already done something like this, using the idea that if someone gets into Chicago but is rejected from Harvard, then that is one vote that Chicago is easier to get into than Harvard. To get it publish, you would want to do a hypothesis test. You would probably want to do a population proportion test. For example, consider all applications to Harvard and Chicago who got into only one as the population. If they are equally hard, then the population proportion for each group should be 0.5. Use the data to get a p-value and reject or accept the null hypothesis. Again, I imagine that TMers would be completely behind this idea because they would get a ranking of how difficult a school is to get into. The ideal strategy for them would be to find the schools that they prefer the most that are the easiest to get into. A rigorous analysis of the data is only to their benefit.

You are correct that the lack of random sampling is a problem. You might be able to correct for this using some survey methods that are used to correct for certain groups not responding as much. Even if you cannot, worse econometric crimes have been published in the AER.

There is csv file* summarizing a lot of things that would go into a ranking (the file scores -99 for did not apply, -1 for waitlist, 0 for rejection, 1 for acceptance, everything else should be reletively clear). In fact, I use that file when I do my regressions, which I talked about here (Analysis on the jump in quality of accepted applicants). Right now, I think getting a few more years of data will be helpful in controlling the confidence intervals, and I'd love to learn how to do multiple equation models** and learn survey techniques to correct for biases.

I haven't added in a test for the taking of Real Analysis yet, but that is planned. After that, I don't know if I can pull much more information out of the profiles, as the formatting is non-uniform, and I'm not terribly excited about going through by hand. Also, I can call R from perl, so I'd like to add some regressions to the analysis page, we'll see.

* If you play with the data, feel free to post results or send me R code if you want to talk about any of it.

** Multiple equation modelling would allow me to regress admit to school X on whatever predictors I have and on results from schools Y, Z, etc. Doing this would make the ranking by difficulty especially easy (I could just rank the coefficient on the constants) and, to me, especially believable. I have tried to rank based on pairwise comparisons of difficulty, and ended up doing it by hand and putting it here.

[untitled]

I added a check to see if real anal is mentioned after math courses, the results are in the csv file. I can then to logit models of taking real analysis. I can also check to see if taking real analysis changes the chance of getting into a school. The answer depends on the school, and for very few schools is the answer significant. Does anyone know, is there a test for significance of significance - for instance, if 10% of the results are significant at the 10% level, then I think you can conclude you are breathing. Is there any way that I can present a bunch of results, and then say that with a certain probability I can say that some percentage of my significances are actually significant? If so, that would certainly make my day.

Seriously, though, I can present any of this, if anyone wants to see it or has an idea what would be good to look at, but I don't think there is yet anything to see.