LSAC UGPA/LSAT Search Forum

[MAPP]

Anyone use LSAC's law school predictor? Seems pretty legit I just haven't seen anyone mention it on TLS.

https://officialguide.lsac.org/Release/ ... fault.aspx

[Clemenceau]

People have mentioned it, just not in a positive context.

Use mylsn.info

Edit: to elaborate, its just not very accurate. I'm not sure what simple formula it uses, but I just plugged in my numbers and gave me less than 50% at a bunch of schools where I was admitted.

[MAPP]

People have mentioned it, just not in a positive context.

Use mylsn.info

Edit: to elaborate, its just not very accurate. I'm not sure what simple formula it uses, but I just plugged in my numbers and gave me less than 50% at a bunch of schools where I was admitted.

So maybe we could say that it underestimates your potential to get in but could still be used reliably? LSAC says they use a "logistic regression model," I don't know if we have any stats majors or econ grad students that could comment on this?? I'm an econ major so I know some basics of multiple-linear regression, and if LSAC is using all application and admission data numbers (i.e. they have population not sample data), then this predictor should be highly reliable.

[Clemenceau]

People have mentioned it, just not in a positive context.

Use mylsn.info

Edit: to elaborate, its just not very accurate. I'm not sure what simple formula it uses, but I just plugged in my numbers and gave me less than 50% at a bunch of schools where I was admitted.

So maybe we could say that it underestimates your potential to get in but could still be used reliably? LSAC says they use a "logistic regression model," I don't know if we have any stats majors or econ grad students that could comment on this?? I'm an econ major so I know some basics of multiple-linear regression, and if LSAC is using all application and admission data numbers (i.e. they have population not sample data), then this predictor should be highly reliable.

How can we call it highly reliable if it grossly underestimates/miscaclulates ones chances? I just now threw in a 3.7/169. Gives about a 25% chance at Duke and Columbia. However, experience(lsn) tells us that a 3.7/169 has very, very little chance at Columbia and is nearly an auto-admit at Duke. Please explain the reliability.

[MAPP]

People have mentioned it, just not in a positive context.

Use mylsn.info

Edit: to elaborate, its just not very accurate. I'm not sure what simple formula it uses, but I just plugged in my numbers and gave me less than 50% at a bunch of schools where I was admitted.

So maybe we could say that it underestimates your potential to get in but could still be used reliably? LSAC says they use a "logistic regression model," I don't know if we have any stats majors or econ grad students that could comment on this?? I'm an econ major so I know some basics of multiple-linear regression, and if LSAC is using all application and admission data numbers (i.e. they have population not sample data), then this predictor should be highly reliable.

How can we call it highly reliable if it grossly underestimates/miscaclulates ones chances? I just now threw in a 3.7/169. Gives about a 25% chance at Duke and Columbia. However, experience(lsn) tells us that a 3.7/169 has very, very little chance at Columbia and is nearly an auto-admit at Duke. Please explain the reliability.

Like I said, it would be nice to have someone comment on the methodologies of the logistic regression model. I can only speak for linear and multiple-linear regression. That being said, mylsn is only sample data (in most cases, it is such a small sample data set it could not be used to provide a statistically significant confidence interval). LSAC is using population data, and therefore they can provide a confidence interval that is statically significant.

In layman's terms, you can't really argue that a few data points on mylsn is superior in comparison to all data from all admissions and all enrollment at an institution that LSAC is using in their calculation.

[WhyYaCryin]

People have mentioned it, just not in a positive context.

Use mylsn.info

Edit: to elaborate, its just not very accurate. I'm not sure what simple formula it uses, but I just plugged in my numbers and gave me less than 50% at a bunch of schools where I was admitted.

So maybe we could say that it underestimates your potential to get in but could still be used reliably? LSAC says they use a "logistic regression model," I don't know if we have any stats majors or econ grad students that could comment on this?? I'm an econ major so I know some basics of multiple-linear regression, and if LSAC is using all application and admission data numbers (i.e. they have population not sample data), then this predictor should be highly reliable.

How can we call it highly reliable if it grossly underestimates/miscaclulates ones chances? I just now threw in a 3.7/169. Gives about a 25% chance at Duke and Columbia. However, experience(lsn) tells us that a 3.7/169 has very, very little chance at Columbia and is nearly an auto-admit at Duke. Please explain the reliability.

Like I said, it would be nice to have someone comment on the methodologies of the logistic regression model. I can only speak for linear and multiple-linear regression. That being said, mylsn is only sample data (in most cases, it is such a small sample data set it could not be used to provide a statistically significant confidence interval). LSAC is using population data, and therefore they can provide a confidence interval that is statically significant.

In layman's terms, you can't really argue that a few data points on mylsn is superior in comparison to all data from all admissions and all enrollment at an institution that LSAC is using in their calculation.

I'm almost positive that the LSAC search doesn't get updated year to year. There are HUGE differences in admission chances from one year to the next especially outside the T14 where some schools have decreased LSAT medians by several points. At Maryland, for example, the Fall 2009 class had an LSAT median of 163. This year's class is 157 I believe. However when you plug in any GPA/LSAT combination, they'll give you the same chances today that they did this time 2 years ago. And there has been more volatility in admissions numbers than Maryland. I took econometrics as well so I want to believe that regression works, but that search is a joke. As Clemenceau said mylsn.info is far superior. It gives you real numbers from real people year to year (assuming that the lsn profiles are not fake).

[oil]

People have mentioned it, just not in a positive context.

Use mylsn.info

Edit: to elaborate, its just not very accurate. I'm not sure what simple formula it uses, but I just plugged in my numbers and gave me less than 50% at a bunch of schools where I was admitted.

So maybe we could say that it underestimates your potential to get in but could still be used reliably? LSAC says they use a "logistic regression model," I don't know if we have any stats majors or econ grad students that could comment on this?? I'm an econ major so I know some basics of multiple-linear regression, and if LSAC is using all application and admission data numbers (i.e. they have population not sample data), then this predictor should be highly reliable.

How can we call it highly reliable if it grossly underestimates/miscaclulates ones chances? I just now threw in a 3.7/169. Gives about a 25% chance at Duke and Columbia. However, experience(lsn) tells us that a 3.7/169 has very, very little chance at Columbia and is nearly an auto-admit at Duke. Please explain the reliability.

Like I said, it would be nice to have someone comment on the methodologies of the logistic regression model. I can only speak for linear and multiple-linear regression. That being said, mylsn is only sample data (in most cases, it is such a small sample data set it could not be used to provide a statistically significant confidence interval). LSAC is using population data, and therefore they can provide a confidence interval that is statically significant.

In layman's terms, you can't really argue that a few data points on mylsn is superior in comparison to all data from all admissions and all enrollment at an institution that LSAC is using in their calculation.

I have a stat degree from one of schools mentioned in this thread, hopefully that qualifies me to speak on this matter.
While normally you could argue that the small sample size of mylsn would necessarily make it a worse metric than the LSAC logistic regression model, in my opinion the combination mylsn and human intuition (even untrained) actually function better. The logistic regression model has no ability to enforce any sort of GPA floor or any nuance in dealing with medians and quartiles, whereas even with a small and skewed sample size that mylsn provides we are able to roughly discern these. Maybe if LSAC used a basic machine learning algorithm I could see their tool being a better metric, but not a simple log regression model.

Additionally, as with any other regression, as you approach the extremes the regression becomes less accurate. These extremes could refer to either your quantitative scores, the admission chances, or the school caliber. So a 3.5/180 at T14 or a 4.0/178 application to HYS are likely not as trustworthy on the LSAC tool as a 3.6/160 to lower T1s. You can see this in the tool's unwillingness to denote apps as 100% even when they clearly would be auto-admits

[MAPP]

People have mentioned it, just not in a positive context.

Use mylsn.info

Edit: to elaborate, its just not very accurate. I'm not sure what simple formula it uses, but I just plugged in my numbers and gave me less than 50% at a bunch of schools where I was admitted.

So maybe we could say that it underestimates your potential to get in but could still be used reliably? LSAC says they use a "logistic regression model," I don't know if we have any stats majors or econ grad students that could comment on this?? I'm an econ major so I know some basics of multiple-linear regression, and if LSAC is using all application and admission data numbers (i.e. they have population not sample data), then this predictor should be highly reliable.

How can we call it highly reliable if it grossly underestimates/miscaclulates ones chances? I just now threw in a 3.7/169. Gives about a 25% chance at Duke and Columbia. However, experience(lsn) tells us that a 3.7/169 has very, very little chance at Columbia and is nearly an auto-admit at Duke. Please explain the reliability.

Like I said, it would be nice to have someone comment on the methodologies of the logistic regression model. I can only speak for linear and multiple-linear regression. That being said, mylsn is only sample data (in most cases, it is such a small sample data set it could not be used to provide a statistically significant confidence interval). LSAC is using population data, and therefore they can provide a confidence interval that is statically significant.

In layman's terms, you can't really argue that a few data points on mylsn is superior in comparison to all data from all admissions and all enrollment at an institution that LSAC is using in their calculation.

I have a stat degree from one of schools mentioned in this thread, hopefully that qualifies me to speak on this matter.
While normally you could argue that the small sample size of mylsn would necessarily make it a worse metric than the LSAC logistic regression model, in my opinion the combination mylsn and human intuition (even untrained) actually function better. The logistic regression model has no ability to enforce any sort of GPA floor or any nuance in dealing with medians and quartiles, whereas even with a small and skewed sample size that mylsn provides we are able to roughly discern these. Maybe if LSAC used a basic machine learning algorithm I could see their tool being a better metric, but not a simple log regression model.

Additionally, as with any other regression, as you approach the extremes the regression becomes less accurate. These extremes could refer to either your quantitative scores, the admission chances, or the school caliber. So a 3.5/180 at T14 or a 4.0/178 application to HYS are likely not as trustworthy on the LSAC tool as a 3.6/160 to lower T1s. You can see this in the tool's unwillingness to denote apps as 100% even when they clearly would be auto-admits

Ahhh finally someone with some qualifications to speak on this. Thanks for your input! So even though LSAC is using population data, you would consider mylsn to be superior simply because we add human intuition into the mix?

[oil]

Ahhh finally someone with some qualifications to speak on this. Thanks for your input! So even though LSAC is using population data, you would consider mylsn to be superior simply because we add human intuition into the mix?

There are many issues with the LSAC tool. For example, there's not even input for URM status. Despite it using the population instead of a sample, it is not using the data in a way that is particularly useful to anyone. mylsn.info may not be perfect, but I'd expect that it does at least slot you into a couple categories quite well. Sure thing (90%+), likely admit (65%-90%), coin-flip (35%-65%), small chance (10%-35%), no chance ( <10%).

There's no arguing that the LSAC could produce an outstanding admit prediction calc (using log regression even!) because of their access to pop data, but they have no actual interest in definitively telling kids they have no shot at top universities.

[rpfote]

Maybe if LSAC used a basic machine learning algorithm I could see their tool being a better metric, but not a simple log regression model.

Technically, logistic regression is considered a basic machine learning algorithm, but I take your point. Also, if I were to try and build a ML model I would probably consider a support vector machine or naive bayes before looking at logit.

I do agree that with the data that LSAC has, they could potentially create a better model to predict admissions, but speaking from the machine learning side of things (as opposed to statistics), the feature set is the most important part. myLSN seems to work by finding applicants that are similar to the numbers entered and creating a visualization of their results (and then extrapolating that into probabilities), depending on how LSAC uses their data, what features they select for their model, it would not be surprising that myLSN would provide a better prediction. The other advantage of myLSN is that it allows the user to look at the similar examples, and then the user can extrapolate.