Harvard, C/O 2017, Applicants Thread (2013-2014) Forum

[gta]

But in all seriousness, the problem with this is it (I assume) ignores 25/50/75 percentiles, which we know are important. Also 5 years of data might actually be a bad thing. We are in a vastly different world with admissions than we were in the 2008-2009 cycle.

This was my first though, too, so I just did a quick Google search and found the following 25/50/75 info for Harvard entering class in fall 2009 (link):

171---173---176
3.76---3.89---3.96

And here are the most recent stats:

170---173---175
3.77---3.88---3.95

It looks like Harvard has been pretty well insulated from the collapse, as many people have said. These numbers are pretty darn close. The medians have stayed virtually identical. The LSAT 25th and 75th have each slipped, but the GPAs have barely moved (the 25th even improved.)

So I think the numbers are probably still good to use, which is too bad, as the calculator gives me a 13% shot at admission.

[midwest17]

I don't know what regression you're running, but to be useful you probably need to include indicator variables for passing the various percentiles, as has been mentioned. And even then you're going to run into problems with the inability to account for GPA/LSAT floors.

ETA: The upshot is that such regressions can be fun, but people should still use myLSN rather than relying on the regression.

[lawschool22]

But in all seriousness, the problem with this is it (I assume) ignores 25/50/75 percentiles, which we know are important. Also 5 years of data might actually be a bad thing. We are in a vastly different world with admissions than we were in the 2008-2009 cycle.

This was my first though, too, so I just did a quick Google search and found the following 25/50/75 info for Harvard entering class in fall 2009:

171---173---176
3.76---3.89---3.96

And here are the most recent stats:

170---173---175
3.77---3.88---3.95

It looks like Harvard has been pretty well insulated from the collapse, as many people have said. These numbers are pretty darn close. The medians have stayed virtually identical. The LSAT 25th and 75th have each slipped, but the GPAs have barely moved (the 25th even improved.)

So I think the numbers are probably still good to use, which is too bad, as the calculator gives me a 13% shot at admission.

That's not the problem though. The problem is that applicants have dropped. So if you had 10,000 applicants in 2008-2009 and you have 8,000 this cycle, then even if medians hold, your chances of getting in this year have gone up because you're competing against fewer applicants.

Of course I haven't looked at a HLS specific data to see what years applicants increased/decreased, so this could be completely false, but it was just a thought.

But you will see this effect with myLSN. As you increase the number of cycles, your chances generally decrease to a point.

Does anyone know what how many HLS applicants there were five years ago?

[fivestarfolds]

.

[Nooblarzlarz]

But in all seriousness, the problem with this is it (I assume) ignores 25/50/75 percentiles, which we know are important. Also 5 years of data might actually be a bad thing. We are in a vastly different world with admissions than we were in the 2008-2009 cycle.

This was my first though, too, so I just did a quick Google search and found the following 25/50/75 info for Harvard entering class in fall 2009:

171---173---176
3.76---3.89---3.96

And here are the most recent stats:

170---173---175
3.77---3.88---3.95

It looks like Harvard has been pretty well insulated from the collapse, as many people have said. These numbers are pretty darn close. The medians have stayed virtually identical. The LSAT 25th and 75th have each slipped, but the GPAs have barely moved (the 25th even improved.)

So I think the numbers are probably still good to use, which is too bad, as the calculator gives me a 13% shot at admission.

That's not the problem though. The problem is that applicants have dropped. So if you had 10,000 applicants in 2008-2009 and you have 8,000 this cycle, then even if medians hold, your chances of getting in this year have gone up because you're competing against fewer applicants.

Of course I haven't looked at a HLS specific data to see what years applicants increased/decreased, so this could be completely false, but it was just a thought.

But you will see this effect with myLSN. As you increase the number of cycles, your chances generally decrease to a point.

Does anyone know what how many HLS applicants there were five years ago?

Here's a handy spreadsheet for #s of all the t14 going back to 2007:
https://docs.google.com/spreadsheet/pub ... SlE&gid=30

[lawschool22]

But in all seriousness, the problem with this is it (I assume) ignores 25/50/75 percentiles, which we know are important. Also 5 years of data might actually be a bad thing. We are in a vastly different world with admissions than we were in the 2008-2009 cycle.

This was my first though, too, so I just did a quick Google search and found the following 25/50/75 info for Harvard entering class in fall 2009:

171---173---176
3.76---3.89---3.96

And here are the most recent stats:

170---173---175
3.77---3.88---3.95

It looks like Harvard has been pretty well insulated from the collapse, as many people have said. These numbers are pretty darn close. The medians have stayed virtually identical. The LSAT 25th and 75th have each slipped, but the GPAs have barely moved (the 25th even improved.)

So I think the numbers are probably still good to use, which is too bad, as the calculator gives me a 13% shot at admission.

That's not the problem though. The problem is that applicants have dropped. So if you had 10,000 applicants in 2008-2009 and you have 8,000 this cycle, then even if medians hold, your chances of getting in this year have gone up because you're competing against fewer applicants.

Of course I haven't looked at a HLS specific data to see what years applicants increased/decreased, so this could be completely false, but it was just a thought.

But you will see this effect with myLSN. As you increase the number of cycles, your chances generally decrease to a point.

Does anyone know what how many HLS applicants there were five years ago?

Here's a handy spreadsheet for #s of all the t14 going back to 2007:
URL

Yeah exactly, look at the big drop during c/o 2015 and 2016. So even if medians held, you're competing against fewer applicants - a lot fewer actually (decreases of 16% and 14% each year). This would likely increase your chances as compared to the earlier cycles, meaning they probably are not representative of your chances this cycle. Now we do know that c/o 2016 actually increased by about 1%, but this still puts it in the same ballpark as c/o 2014 and 2015.

If I were making this thing, I would not go past the c/o 2014 for my data.

[gta]

Yeah exactly, look at the big drop during c/o 2015 and 2016. So even if medians held, you're competing against fewer applicants - a lot fewer actually (decreases of 16% and 14% each year). This would likely increase your chances as compared to the earlier cycles, meaning they probably are not representative of your chances this cycle. Now we do know that c/o 2016 actually increased by about 1%, but this still puts it in the same ballpark as c/o 2014 and 2015.

If I were making this thing, I would not go past the c/o 2014 for my data.

11% of applicants were admitted for c/o 2013, compared to 16% for c/o 2015. That looks better, but we don't know the composition of the group that stopped applying. In theory, they could all have been well below both 25th's, and having them drop out wouldn't improve odds for any applicants with better numbers.

[Gary]

See JS.... If you don't release decisions then our attention goes to breaking the admissions secrets

[lawschool22]

Yeah exactly, look at the big drop during c/o 2015 and 2016. So even if medians held, you're competing against fewer applicants - a lot fewer actually (decreases of 16% and 14% each year). This would likely increase your chances as compared to the earlier cycles, meaning they probably are not representative of your chances this cycle. Now we do know that c/o 2016 actually increased by about 1%, but this still puts it in the same ballpark as c/o 2014 and 2015.

If I were making this thing, I would not go past the c/o 2014 for my data.

11% of applicants were admitted for c/o 2013, compared to 16% for c/o 2015. That looks better, but we don't know the composition of the group that stopped applying. In theory, they could all have been 3.4/166 Hail Mary's, and having them drop out wouldn't improve odds for any serious applicants.

Right, but either way it's more likely that the more recent cycles will be most representative of this cycle. We still get an idea of the composition of the people accepted by looking at the LSN data. All I'm saying is we should use the data that's more likely to be representative of this cycle. 2009-2010 was a vastly different applicant pool than this cycle.

Take a look at this:



Compare to this:



The only thing I changed were the cycles I looked at. Your chances drop by 21% if you extend the data back five years.

[patfeeney]

Yeah exactly, look at the big drop during c/o 2015 and 2016. So even if medians held, you're competing against fewer applicants - a lot fewer actually (decreases of 16% and 14% each year). This would likely increase your chances as compared to the earlier cycles, meaning they probably are not representative of your chances this cycle. Now we do know that c/o 2016 actually increased by about 1%, but this still puts it in the same ballpark as c/o 2014 and 2015.

If I were making this thing, I would not go past the c/o 2014 for my data.

11% of applicants were admitted for c/o 2013, compared to 16% for c/o 2015. That looks better, but we don't know the composition of the group that stopped applying. In theory, they could all have been 3.4/166 Hail Mary's, and having them drop out wouldn't improve odds for any serious applicants.

Right, but either way it's more likely that the more recent cycles will be most representative of this cycle. We still get an idea of the composition of the people accepted by looking at the LSN data. All I'm saying is we should use the data that's more likely to be representative of this cycle. 2009-2010 was a vastly different applicant pool than this cycle.

Take a look at this:



Compare to this:



The only thing I changed were the cycles I looked at. Your chances drop by 21% if you extend the data back five years.

ls22 your stats knowledge allows me to sleep at night. Thank you.

[gta]

The only thing I changed were the cycles I looked at. Your chances drop by 21% if you extend the data back five years.

Nice find, that certainly makes me feel better!

[lawschool22]

The only thing I changed were the cycles I looked at. Your chances drop by 21% if you extend the data back five years.

Nice find, that certainly makes me feel better!

No problem. Obviously with the sample size we have it's hard to say anything definitive, but it's the best we have.

[lawschool22]

Hey, to the guy/girl who made this regression tool for HLS, can you try doing it with just the past two cycles of data (i.e. c/o 2015 and c/o 2016)? That would be interesting a probably more accurate.

Also, is there some way to incorporate 25/50/75 percentile cutoffs?

It's a neat tool and interface, so with some tweaks it may end up being useful.

[pseu]

Hey all, I've been lurking this thread for some time and decided to finally make an actual post. Your intensity about tracking and speculating is both impressive and entertaining, and it has also kept me sane through the waiting, so thanks for that!

I had my JS1 early this week, so I am hoping that a round of JS2s come soon so I can at least know that the process is moving along. Are you all thinking that tomorrow or Monday may be a possibility? Again, complete speculation, but it is keeping me from going crazy to try to make some sense of the process haha

Good luck to everyone else waiting here!

btw,
App Complete 12/5
Interview Scheduled 12/19

[lawschool22]

Hey all, I've been lurking this thread for some time and decided to finally make an actual post. Your intensity about tracking and speculating is both impressive and entertaining, and it has also kept me sane through the waiting, so thanks for that!

I had my JS1 early this week, so I am hoping that a round of JS2s come soon so I can at least know that the process is moving along. Are you all thinking that tomorrow or Monday may be a possibility? Again, complete speculation, but it is keeping me from going crazy to try to make some sense of the process haha

Good luck to everyone else waiting here!

btw,
App Complete 12/5
Interview Scheduled 12/19

Yeah the speculating is what keeps us sane too lol. I think there are several people who think we will see some movement tomorrow and/or Monday.

[drawstring]

I'm thinking there will be some bloodshed tomorrow.

[scoopDeeDoo]

..

[scoopDeeDoo]

..

[Jessasaurus]

I'm thinking there will be some bloodshed tomorrow.

+1

[pseu]

Hey all, I've been lurking this thread for some time and decided to finally make an actual post. Your intensity about tracking and speculating is both impressive and entertaining, and it has also kept me sane through the waiting, so thanks for that!

I had my JS1 early this week, so I am hoping that a round of JS2s come soon so I can at least know that the process is moving along. Are you all thinking that tomorrow or Monday may be a possibility? Again, complete speculation, but it is keeping me from going crazy to try to make some sense of the process haha

Good luck to everyone else waiting here!

btw,
App Complete 12/5
Interview Scheduled 12/19

Yeah the speculating is what keeps us sane too lol. I think there are several people who think we will see some movement tomorrow and/or Monday.

Seems like that would make sense - looks like this time last year was hot, and another round seems due. As drawstring mentioned, the dings are probably gonna start rolling out as well, but if that happens at least there is activity. Beats waiting by so much haha

[patfeeney]

I'm thinking there will be some bloodshed tomorrow.

I know I'm pretty early on in the cycle/ late to the game, but I'm expecting to be lined up for the guillotine come sunrise.

[kaseyb002]

Hey, to the guy/girl who made this regression tool for HLS, can you try doing it with just the past two cycles of data (i.e. c/o 2015 and c/o 2016)? That would be interesting a probably more accurate.

Also, is there some way to incorporate 25/50/75 percentile cutoffs?

It's a neat tool and interface, so with some tweaks it may end up being useful.

Thanks for the feedback

So first off, I used a logistic regression (used to predict stuff between 0 and 1 e.g., admit vs. no-admit). A logistic regression will force the data to look like the S-curve you see on the graph, therefore, it will not be splitter friendly. If you're a splitter I would ignore the results.

For my methodology, I did almost nothing. I just pulled down the data from the past five years (didn't take time to scrub any of it) and created two regression models (one for data points at or above the median LSAT, and one below). I suppose I could cluster even further and create 4 different models: (1)above median lsat above median gpa, (2)above lsat below gpa, (3)below lsat above gpa, and (4)below lsat below gpa. However, I'm not sure how many observations that would leave me for each model. I strongly doubt I could do 25/50/75 cutoffs; there simply isn't enough data.

And in all honesty, no matter how good the model gets, the actual % probability number that shows up probably won't be as useful relative to the raw results you get on mylsn. The graph though, I think is useful for seeing how dramatic a difference a few points on the LSAT or GPA will make.

If anyone has some stats background and would like to make a better model, let me know. All I need are the coefficients and then I can plug 'em in right away. See the source code (all in javascript) if you want more info.

Oh and lastly, if people really care, I can partition the data and test for accuracy. On a previous data set I worked on I got like 86% or something.

As a side note...
I did a quick search, and there does seem to be a significant difference for splitters when you look between

2009-2011
http://myLSN.info/82syq7_1-14.jpg

versus

2011-2014
http://myLSN.info/yjwzhh_1-14.jpg

So maybe leaving that older data out could improve the model significantly.

[nothingtosee]

..

I love this site.

[kaseyb002]

But in all seriousness, the problem with this is it (I assume) ignores 25/50/75 percentiles, which we know are important. Also 5 years of data might actually be a bad thing. We are in a vastly different world with admissions than we were in the 2008-2009 cycle.

Finally, I'm curious about your regression. What confidence level are you at, and what was your correlation coefficient?

Here's the STATA printout. Again this was just a straight up, no tweaks, no data scrubbing "logit admit lsat gpa":

Model For Above Median LSAT Data Points

Logistic regression Number of obs = 690
LR chi2(2) = 369.27
Prob > chi2 = 0.0000
Log likelihood = -287.48419 Pseudo R2 = 0.3911

------------------------------------------------------------------------------
admit | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
gpa | 11.05645 .8674196 12.75 0.000 9.356334 12.75656
lsat | .3172556 .0563768 5.63 0.000 .206759 .4277521
_cons | -97.13444 11.28062 -8.61 0.000 -119.244 -75.02483
------------------------------------------------------------------------------


Model for Below Median LSAT Data Points

Logistic regression Number of obs = 817
LR chi2(2) = 236.28
Prob > chi2 = 0.0000
Log likelihood = -257.6438 Pseudo R2 = 0.3144

------------------------------------------------------------------------------
admit | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
gpa | 10.25789 1.111142 9.23 0.000 8.080089 12.43568
lsat | .54094 .0758397 7.13 0.000 .3922969 .6895831
_cons | -133.1748 14.55945 -9.15 0.000 -161.7108 -104.6388
------------------------------------------------------------------------------

And in case you're not familiar with that format, p-value for everything is 0 or in other words 99.999% confidence.

[chneyo]

...