The Old GPA v MCAT Debate

[jult24er]

---

Members do not see this ad.

For the record, if you linearly regress AMCAS' data
http://www.aamc.org/data/facts/2008/mcatgpa-grid-3yrs-app-accpt.htm

the answer you get is:

GPA * 10 * 2.62
+
MCAT * 2.13

- 107.9

Please feel free to call this the "too-much-time-on-his-hands" score and refer to it as such on SDN to forever.

 

[mmmcdowe]

For the record, if you linearly regress AMCAS' data
http://www.aamc.org/data/facts/2008/mcatgpa-grid-3yrs-app-accpt.htm

the answer you get is:

GPA * 10 * 2.62
+
MCAT * 2.13

- 107.9

Please feel free to call this the "too-much-time-on-his-hands" score and refer to it as such on SDN to forever.

Your data is flawed because of the lumped groups of GPA and MCAT, but if you look at polling data the MCAT ranks first most often of all things (its less that 50% of the time though).

 

[jult24er]

Your data is flawed because of the lumped groups of GPA and MCAT, but if you look at polling data the MCAT ranks first most often of all things (its less that 50% of the time though).

Feel free to fix that ...

 

[235009]

useless.

rounding up my gpa and calculating using your crappy formula i get 56.6% but looking at the table and rounding down my GPA I get 73.3%. Even if i go down to the lower MCAT category i would be at 63.2% from the table.

Way to waste your time buddy

 

[jult24er]

useless.

rounding up my gpa and calculating using your crappy formula i get 56.6% but looking at the table and rounding down my GPA I get 73.3%. Even if i go down to the lower MCAT category i would be at 63.2% from the table.

Way to waste your time buddy

I guess linear is not so accurate across the whole spectrum.

For people between 3.0 and 4.0 and 27 and 45 it's:

2.43 MCAT + 5.86 GPA*10 - 222.21

And for people between 3.4 and 3.8 and 30 - 35 it's:

2.83 MCAT + 6.00 GPA*10 - 233.83

That predict you better?

 

[mmmcdowe]

Feel free to fix that ...

I can't, I would need every persons MCAT and GPA individually to do so.

 

[The Poet Sings]

perhaps i should have my boyfriend the stats major explain this to me, but until i can get a hold of him, how does any of this make sense?

 

[Pinkertinkle]

They both matter.

 

[BluntSurgeon]

They both matter.

quoting this

 

[Beta Cell]

perhaps i should have my boyfriend the stats major explain this to me, but until i can get a hold of him, how does any of this make sense?

All he did was was plot all of the MCAT/GPA groups and do a linear least-squares regression on the percentages given in the AMCAS table to find out your chances of admission to a medical school based solely on your MCAT and GPA. MCAT and GPA are the independent variables and % admitted with those MCAT and GPA is the dependent variable.

 

[235009]

I guess linear is not so accurate across the whole spectrum.

For people between 3.0 and 4.0 and 27 and 45 it's:

2.43 MCAT + 5.86 GPA*10 - 222.21

And for people between 3.4 and 3.8 and 30 - 35 it's:

2.83 MCAT + 6.00 GPA*10 - 233.83

That predict you better?

those two ranges definitely overlap so ur post doesn't make much sense (i.e. my stats fall in both ranges so which model should i be using?)

did you check the residuals on your models 😛 i'm assuming they're gonna look pretty horrible. also look up the term "ecological fallacy" because you are falling victim to it!

 

[jult24er]

perhaps i should have my boyfriend the stats major explain this to me, but until i can get a hold of him, how does any of this make sense?

Sorry, its a formula that calculates your percentage chance of getting in to at least one med school, based on your gpa and mcat.

For people between a 3.0 and 4.0 and a 27 and 45 that is:

2.43 MCAT + 5.86 GPA*10 - 222.21

So, GPA seems to be more correlated with getting accepted then mcat relative to how they are weighted in the LizzyM score (1.00 MCAT + 1.00 GPA*10) ...

Unfortunately, you can't say whether actually having the higher GPA gets people in, or if people with higher GPAs just tend also to have better ECs / interviews, etc. then people with higher mcats.

 

[jult24er]

those two ranges definitely overlap so ur post doesn't make much sense (i.e. my stats fall in both ranges so which model should i be using?)

did you check the residuals on your models 😛 i'm assuming they're gonna look pretty horrible. also look up the term "ecological fallacy" because you are falling victim to it!

I'd use the one more specificaly targeted to your range.

I never made any claims about any individuals ...

 

[235009]

I'd use the one more specificaly targeted to your range.

I never made any claims about any individuals ...

suggesting that people plug their individual scores in a linear regression model created using aggregate data = ecological fallacy

not because excel will do it for you doesn't mean the theory behind it is correct

...this:

I would need every persons MCAT and GPA individually to do so.

would be the correct way to do what you want to do

 

[jult24er]

suggesting that people plug their individual scores in a linear regression model created using aggregate data = ecological fallacy

not because excel will do it for you doesn't mean the theory behind it is correct

...this:



would be the correct way to do what you want to do

I honestly don't see why the "aggregate" thing is a big deal, i dont think the regression would change much if you had the indivdual data instead of those 3 point aggregated-blocks. I dont think thats what ecological fallacy means ...

 

[Excelsius]

For the record, if you linearly regress AMCAS' data
http://www.aamc.org/data/facts/2008/mcatgpa-grid-3yrs-app-accpt.htm

the answer you get is:

GPA * 10 * 2.62
+
MCAT * 2.13

- 107.9

Please feel free to call this the "too-much-time-on-his-hands" score and refer to it as such on SDN to forever.

Look at the graphs here: AMCAS Statistics Refinement: MCAT vs GPA vs Acceptance

Unless you are looking at the average ranges for stats, I don't think linear regression is the right model. You would have to break up the data at least into two parts to make linear regression useful, but even then I am not sure if that would be a good model. I am not a stats major so I don't know why you would even want to use regression here, but I'd think that in this case you'd need a nonlinear regression, rather than linear.

 

[docdad]

You can't use least squares regression for data like this, because of how GPA and MCAT scores are correlated. Linear regression assumes the two variables are independent and normally distributed.

There are other ways to handle data like this, but they're beyond the scope of this discussion.

 

[jult24er]

You can't use least squares regression for data like this, because of how GPA and MCAT scores are correlated. Linear regression assumes the two variables are independent and normally distributed.

There are other ways to handle data like this, but they're beyond the scope of this discussion.

So you cant have a regression model with, say, income and education level as independent variables, becasue those are correlated?

 

[MDman87]

suggesting that people plug their individual scores in a linear regression model created using aggregate data = ecological fallacy

not because excel will do it for you doesn't mean the theory behind it is correct

...this:



would be the correct way to do what you want to do

Good grief, man. I think the OP was just trying to help - by providing what he thought to be a useful formula. Clearly, it's not as useful as he thought. So what?! No need to tell him how stupid you think he is. You don't like it? Don't use it

 

[Longshanks]

I guess linear is not so accurate across the whole spectrum.

For people between 3.0 and 4.0 and 27 and 45 it's:

2.43 MCAT + 5.86 GPA*10 - 222.21

And for people between 3.4 and 3.8 and 30 - 35 it's:

2.83 MCAT + 6.00 GPA*10 - 233.83

That predict you better?

OK... what we got here is failure to communicate...

we have two overlapping groups with two different equations. Which one for which is it? And I guess both the MCAT and GPA matter less if you have poor numbers like me or if you're at your the top of the game, but if you're in that strong middle.. by damn your screwed with that gpa....🙄

 

[235009]

You can't use least squares regression for data like this, because of how GPA and MCAT scores are correlated. Linear regression assumes the two variables are independent and normally distributed.

There are other ways to handle data like this, but they're beyond the scope of this discussion.

So you cant have a regression model with, say, income and education level as independent variables, becasue those are correlated?

you COULD have both in your model if they are both significant (which would mean one is explaining a part of the variance that the other is not) however OP hasn't given us confidence intervals or p-values ....i wouldn't be surprised if neither variable is significant in this model.

 

[235009]

Good grief, man. I think the OP was just trying to help - by providing what he thought to be a useful formula. Clearly, it's not as useful as he thought. So what?! No need to tell him how stupid you think he is. You don't like it? Don't use it

the point is i don't think ANYONE should use this because it is wrong..... plus i always enjoy a good stats discussion... lighten up

 

[jult24er]

OK... what we got here is failure to communicate...

we have two overlapping groups with two different equations. Which one for which is it? And I guess both the MCAT and GPA matter less if you have poor numbers like me or if you're at your the top of the game, but if you're in that strong middle.. by damn your screwed with that gpa....🙄

Exactly, the correlation of acceptance with improving gpa or mcat is different depending upon how high your gpa and mcat are to begin with. So, when you try to use the entire range of data to derive one forumla that correlates gpa and mcat with acceptance rate, its not that accurate. The narower the range of data used, the better a fit the resulting forumla would be to an applicant's specific situation.

I think with the data AMCAS gives, you could come up with around 200 linear regressions, each for a different spot on the MCAT/GPA chart. But i thought the "3.1 to 3.9 and 28 to 40" would cover most people (although maybe with less accuray, since thats a large range) and the "3.5 to 3.7 and 31 to 34" would cover a good chunk of people, and with more accuracy.

 

[jult24er]

Good grief, man. I think the OP was just trying to help - by providing what he thought to be a useful formula. Clearly, it's not as useful as he thought. So what?! No need to tell him how stupid you think he is. You don't like it? Don't use it

Thanks!

 

[jult24er]

the point is i don't think ANYONE should use this because it is wrong..... plus i always enjoy a good stats discussion... lighten up

I thought it might not be perfect, but its more informative then nothing ... im not that good at stats, please, take the spreadsheet i posted and get us p-values and residuals and all that good stuff.

 

[jult24er]

Look at the graphs here: AMCAS Statistics Refinement: MCAT vs GPA vs Acceptance

Unless you are looking at the average ranges for stats, I don't think linear regression is the right model. You would have to break up the data at least into two parts to make linear regression useful, but even then I am not sure if that would be a good model. I am not a stats major so I don't know why you would even want to use regression here, but I'd think that in this case you'd need a nonlinear regression, rather than linear.

Just cause linear is simpler, for me to do and people to use. Maybe something else would fit better ... but, yeah, good point, i did break it up into smaller sections of the data as well. And note the LizzyM score suffers from the same linearity problem ...

Man, your graphs in the above look complicated ...

 

[235009]

I thought it might not be perfect, but its more informative then nothing ... im not that good at stats, please, take the spreadsheet i posted and get us p-values and residuals and all that good stuff.

nothing would be better in this case, especially with this neurotic crowd

i could get p-values and residuals (not in excel though) but i dont agree with your overall method so i won't

 

[JeetKuneDo]

I have no clue what's going on in this thread 😕

 

[Excelsius]

Just cause linear is simpler, for me to do and people to use. Maybe something else would fit better ... but, yeah, good point, i did break it up into smaller sections of the data as well. And note the LizzyM score suffers from the same linearity problem ...

Man, your graphs in the above look complicated ...

I didn't think they were complicated when I put them up, but the lack of response has proven me wrong. I have to go through them and redo everything and perhaps retain the raw data as an appendix.

As for LizzyM score, it is restricted only to average applicants - you can see from the graphs that average applicants have a more or less linear curve. When I go back through my graphs, I will be using curve fitting to come up with a realistic equation.

 

[jult24er]

I have no clue what's going on in this thread 😕

1) Your avatar is clever/cute on at least 4 levels I see.

2) This thread can be condescend into 2 points:

a. MDman87 graciosly stood up for me against a barrage of
scurilous attacks
b. I am proving, statisticaly, beyond any shadow of doubt, that:

For applicants with averagish stats (3.2 - 3.8, 27 - 38) the chance of getting at least one acceptance is:

(3.11 * MCAT) + (6.93 * GPA*10) - 280.39

which means that grades are more correlated with success than one might infer from a simple LizzyM score.

*and a more robust answer, accounting for the fact that gpa and mcat points have different value depending upon where they are (i.e. 31 to 32 vs 37 to 38) may be forthcoming from Excelsius.

 

[The Poet Sings]

thanks to those who explained it. now i wish i didn't use these formulas to run my numbers 🙁. so i'm just going to tell myself none of it is valid as is being argued