What is considered early? Forum

[elterrible78]

As someone who has no idea what these numbers mean, a brief explanation would be greatly appreciated.

Heh, no problem!

There are two different factors here: earlier application submission (Early) and binding-early decision application (ED) - you probably got that, though. I do want to point out that the "early" application means an earlier month, not a month early. So we're broadly comparing October v. November applicants, for example, no matter when in the month they applied. That needs to be clarified.

The coefficient for each variable is that substantive effect. In my analysis, the dependent variable is either 1 (applicant was admitted) or 0 (applicant was rejected). The coefficient shows you the "effect" of an independent variable (here, either "early" or "ed").

The coefficient for early = the % increase in your likelihood of being admitted for each 1-point increase in the value of the dependent variable (for "early", a one-point increase is equal to an earlier month the application was sent...September is a-point increase over October).

The coefficient for ED is the same, only here the independent variable is also binary, so it's either a 0 or a 1. In other words, the ED coefficient shows you the % increase in your likelihood of being admitted if you apply ED, period.

The P-values for each is basically the chance that the result is just...well, by chance, and not because of any actual causal effect. The lower the P-value, the higher the probability that there is actually something going on there rather than just random chance. In statistics, there are different "levels of confidence" that researchers use, but the basic cutoff that is used is .10 (beyond that, nobody really is comfortable saying that this is a statistically significant result). Some studies use .05, and those who want to be really strict use .01.

So, in this example, let's take Chicago: Applicants are 126.5% more likely to be admitted for each earlier month they apply, and there is a 0.000% likelihood that this is just random chance rather than an actual effect. Applicants are also 536% more likely to be admitted applying ED vs. regular decision, and there is a 2% chance that this result is just random chance, rather than an actual effect.

Again, I want to throw in that this is all controlling for LSAT, GPA, URM status, nontraditional status, and gender. In other words, don't take your 158 & 2.95 combo and try to convince yourself that if you apply ED, you're 5 times more likely to be admitted than a 175 / 3.8 regular-decision applicant.

Hope that helped![/quote]

edit: scooped by a much more succinct cotiger.

[elterrible78]

Actually, one last request. Could we get the chart for ED splitters, too? Lots of potential ED people here are big-time splitters.

I can definitely do this, although the dearth of data makes it necessary to go back to 2003 to get enough observations to make it meaningful, and there's definitely the caveat that things have very likely changed since 2010. I'll try to take care of it when I get back from class today if you're still interested.

[MoMettaMonk]

Thanks Cotiger and Elterrible.

[cotiger]

Actually, one last request. Could we get the chart for ED splitters, too? Lots of potential ED people here are big-time splitters.

I can definitely do this, although the dearth of data makes it necessary to go back to 2003 to get enough observations to make it meaningful, and there's definitely the caveat that things have very likely changed since 2010. I'll try to take care of it when I get back from class today if you're still interested.

Shweet, thanks bruh. It'll be nice to at least have it for reference.

[elterrible78]

Shweet, thanks bruh. It'll be nice to at least have it for reference.

Here you go:



For the schools in which the regression dropped the variable:

4 ED splitters applied to Columbia, all got in. Their means were 177.5 / 3.43.
Compare this to RD splitters who got into Columbia - their means were 176.7 / 3.45

Only 1 ED splitter applied to Duke (177, 3.2), got rejected.

7 ED splitters got into Northwestern. Means 174.3 / 2.95
RD splitters who got into Northwestern had means of 174.8 / 3.07

Just eyeballing that, it looks like ED didn't probably make a huge difference at Columbia, whereas it seems more important at Northwestern.

[koalacity]

Just popping in to admire the impressive math skills ITT and, more importantly, to thank elterrible for his amazing and extremely useful blog. On behalf of all current and future 0Ls, thank you!

[elterrible78]

Just popping in to admire the impressive math skills ITT and, more importantly, to thank elterrible for his amazing and extremely useful blog. On behalf of all current and future 0Ls, thank you!

No problem...wish I had more time to update or whatever. Glad people find it useful, and I love that av.

[vuthy]

This is really interesting. But one question: isn't there a bit of a selection bias here? For example, I would think that if you had two applicants (A and B) with identical numbers, but A applied in October and B applied in January, we could make some (admittedly big) assumptions about them that might indicate why A was a stronger applicant. For example, A is more likely to have hit his LSAT number on the first try, which could make a marginal difference in close cases (though I know it's highest score only, technically). Also, maybe A got his LORs together quickly b/c he had profs who were eager to write, whereas B was later because he had to ask multiple profs, who then wrote weaker letters. Or A was able to get his app out early bc he didn't struggle with writing his PS, whereas B isn't as strong of a writer, or struggled to find things to say.

I know these are all pretty far-fetched generalizations, but I do think that on balance, early appliers are at least marginally more likely to have their shit together in categories the regressions can't control. And that might explain some of this effect.

But mainly I think this is really interesting.

[drevo]

This is really interesting. But one question: isn't there a bit of a selection bias here? For example, I would think that if you had two applicants (A and B) with identical numbers, but A applied in October and B applied in January, we could make some (admittedly big) assumptions about them that might indicate why A was a stronger applicant. For example, A is more likely to have hit his LSAT number on the first try, which could make a marginal difference in close cases (though I know it's highest score only, technically). Also, maybe A got his LORs together quickly b/c he had profs who were eager to write, whereas B was later because he had to ask multiple profs, who then wrote weaker letters. Or A was able to get his app out early bc he didn't struggle with writing his PS, whereas B isn't as strong of a writer, or struggled to find things to say.

I know these are all pretty far-fetched generalizations, but I do think that on balance, early appliers are at least marginally more likely to have their shit together in categories the regressions can't control. And that might explain some of this effect.

But mainly I think this is really interesting.

This is what I had thought of too especially in respect to places like Yale. I would personally assume that absolute studs that get into Yale probably didn't just decide to apply to law school on a whim and have been preparing for the app for quite some time. Obviously this isn't everyone but I'd imagine it describes some.

Regardless, this is an awesome use of the available data and it is greatly appreciated.

[elterrible78]

This is really interesting. But one question: isn't there a bit of a selection bias here? For example, I would think that if you had two applicants (A and B) with identical numbers, but A applied in October and B applied in January, we could make some (admittedly big) assumptions about them that might indicate why A was a stronger applicant. For example, A is more likely to have hit his LSAT number on the first try, which could make a marginal difference in close cases (though I know it's highest score only, technically). Also, maybe A got his LORs together quickly b/c he had profs who were eager to write, whereas B was later because he had to ask multiple profs, who then wrote weaker letters. Or A was able to get his app out early bc he didn't struggle with writing his PS, whereas B isn't as strong of a writer, or struggled to find things to say.

I know these are all pretty far-fetched generalizations, but I do think that on balance, early appliers are at least marginally more likely to have their shit together in categories the regressions can't control. And that might explain some of this effect.

But mainly I think this is really interesting.

I think those are really good points. How likely they are to be true, and how much those things matter, is kind of left to question, but it's worth thinking about. There's no way to measure that kind of stuff, so we're left with what we have, though.

[elterrible78]

This is really interesting. But one question: isn't there a bit of a selection bias here? For example, I would think that if you had two applicants (A and B) with identical numbers, but A applied in October and B applied in January, we could make some (admittedly big) assumptions about them that might indicate why A was a stronger applicant. For example, A is more likely to have hit his LSAT number on the first try, which could make a marginal difference in close cases (though I know it's highest score only, technically). Also, maybe A got his LORs together quickly b/c he had profs who were eager to write, whereas B was later because he had to ask multiple profs, who then wrote weaker letters. Or A was able to get his app out early bc he didn't struggle with writing his PS, whereas B isn't as strong of a writer, or struggled to find things to say.

I know these are all pretty far-fetched generalizations, but I do think that on balance, early appliers are at least marginally more likely to have their shit together in categories the regressions can't control. And that might explain some of this effect.

But mainly I think this is really interesting.

This is what I had thought of too especially in respect to places like Yale. I would personally assume that absolute studs that get into Yale probably didn't just decide to apply to law school on a whim and have been preparing for the app for quite some time. Obviously this isn't everyone but I'd imagine it describes some.

Regardless, this is an awesome use of the available data and it is greatly appreciated.

The overall point is a really good one, but Yale is probably about the worst possible example to use, since the effect for applying early really just isn't there (based both on what their admissions folks say, and what the numbers bear out).

[cotiger]

The overall point is a really good one, but Yale is probably about the worst possible example to use, since the effect for applying early really just isn't there (based both on what their admissions folks say, and what the numbers bear out).

I think you're being too strict with your NSS determinations. Yes, if we're trying to prove that an effect for sure exists, we might demand the standard 95% confidence. But in this case (especially for ED), we're merely looking for guidance. There's an 84% likelihood that there's some boost at Yale for applying earlier. That's a high enough probability for me to say that it probably does exist. To be sure, if it does exist, it's one of the smallest, but there's a significant difference between the results we get for Yale and the results we get for Columbia/Duke, which have ~90% likelihoods of no boost.

[elterrible78]

The overall point is a really good one, but Yale is probably about the worst possible example to use, since the effect for applying early really just isn't there (based both on what their admissions folks say, and what the numbers bear out).

I think you're being too strict with your NSS determinations. Yes, if we're trying to prove that an effect for sure exists, we might demand the standard 95% confidence. But in this case (especially for ED), we're merely looking for guidance. There's an 84% likelihood that there's some boost at Yale for applying earlier. That's a high enough probability for me to say that it probably does exist. To be sure, if it does exist, it's one of the smallest, but there's a significant difference between the results we get for Yale and the results we get for Columbia/Duke, which have ~90% likelihoods of no boost.

Fair point.