MCAT Estimator Spreadsheet (w/ Poll: How accurate was it for you?)

JSU said:

After putting in my GPA and last MCAT, I got this


(50% confidence score will occur within this range & Upper VR13 PS16 BS16 39

I would really like a 16 in the sciences but its probably a bit of a reach...
Predicted score was 4 points higher than what I actually got (based on GPA). Though that was expected...

Click to expand...

hahaha... yeah.... I should probably make the upper bound limit itself to a 15.

You must have an awesome GPA and great previous MCAT. The GPA alone won't calculate an upper/lower bound value, btw (as it would be n=1 and so a confidence interval would be undefined using the algorithm it uses). Even a perfect GPA is limited to a 13 on the MCAT, so your last MCAT score must have been great (35ish?).

JSU said:

I didnt put in SAT/ACT scores. Though I did put in actual MCAT (12,9,12). My sGPA is 4.0 and my cGPA is 3.94. I think the actual MCAT score threw it off.

Click to expand...

Yeah, I realized that after responding and edited my response. 4 points off probably isn't really that much in reality (it's about 1/2 of a standard deviation). I probably should have give larger options on the poll, haha... but hopefully no one is scoring 10-15 lower than "should" be...that would just suck!

Your GPA is awesome, so that would have made it expect the 39 MCAT. Your 12s on the sciences on an actual MCAT confirmed that you score well. That confirmation probably made it give you a high (read: impossible) upper bound because you consistently do well in the sciences (awesome sGPA + GREAT PS & BS scores both).

JSU said:

Thanks for the compliments. Im sure I would be using this tool every other day if I hadnt already taken the MCAT. I also anticipate that a lot of neurotic premeds will be sending you angry emails when their scores dont work out.

Click to expand...

hahaha.... hence the repeated warnings on it that it's nothing more than a tool. Part of why I put up the poll and all was to get a better feel for how well the tool actually predicts scores and further calibrate it. If people put in the effort to study and all, it should give results that are at least close to what they could get. I did go ahead and fix it a bit though to give slightly lower predictions.

fizzgig said:

i will pray and pray that these estimates pan out (well, not so much the gpa one)... unfortunately this test seems to be an entirely different beast from other standardized tests...

Click to expand...

haha, do pray. good luck!

keep in mind the confidence intervals. 50% confidence means that you have a 50% chance of getting a score in the ranges shown and a 75% chance of scoring above the bottom of that range. To get more conservative perspectives, use the 1 SD option instead. That will set your chance of scoring above the lower bound at about 85%. Using 2 SD (put a 2 in the SD box) further adjusts the values so that you have a 92% chance of scoring above the lower bound -- of course this also means the lower bound will likely be pretty low. For me, a 2 SD lower bound places me at a 92% chance of scoring better than 24... which I suppose is good in that it pretty much guarantees a score in the top 50%...hahaha....

And yes, the MCAT is unique; however, it still bears quite a bit of correlative relationship with other standardized exams.

latrala2300 said:

also, apumic, what's up with the med school selector being taboo? there are versions of it floating around on obscure torrents every now and then that get taken down every few days. wtf?

Click to expand...

The AAMC felt it violated their copyright. (Don't bother asking why it suddenly became an issue when it wasn't an issue in years past, because I have no idea.) It was removed from SDN, but apparently people are still BT'ing it.

Don Draper said:

Sources? For the exams you've listed in the sheet (SAT, GRE, ACT).

Also sources for the MCAT/GPA correlation?

Thanks.

Click to expand...

There are plenty out there... I don't have on me the papers I used at the moment (they were meta studies), but here are some sources:

GPA & MCAT:

http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2625543/?page=9

http://journals.lww.com/academicmed...nship_between_MCAT_science_scores_and.15.aspx


MCAT & other tests:

http://journals.lww.com/academicmed...year_profile_of_students__SAT_scores,.21.aspx

http://journals.lww.com/academicmed..._between_SAT_scores_and_MCAT_scores_of.2.aspx

Don Draper said:

These are 20-30 years old. And one article is about "Black Americans MCAT scores".

Even the 12 year study was only 199 students at a single university. No doubt if you focus on only one school you could find correlations. Some of the other ones were 63 students for the data set.

These are hardly conclusive. GPA at one school is completely different from GPA at another school.

It is also funny that half the studies you posted were for "Black" students (63 or 130 students). This is a very small subset of the application pool. Very small.

Click to expand...

As I said, those were examples I just found in a few moments of searching. I don't have the previously cited stuff right here on this computer. That being said, this tool is not meant to be conclusive. It is meant to give some general idea of what one might expect. You are treating it as though it were a published paper in and of itself. It's not. It's a statistically-based model to estimate an MCAT score prediction. The correlational values are really only relevant insofar as 1) they are positive, 2) in every study available, it was found P<0.05 in all cases, and 3) relevative r values are useful in terms of weighting different prior performance measures. Beyond that, it is largely a matter of evaluating population demographics, adjusting percentile ranking to fit different populations, and calculating standard error, confidence intervals, etc. The reality is that due to the many confounding factors, anything along the lines of an "exact prediction" or "perfect model" would be impossible. This is meant to give a rough estimation using multiple methods of calculation.

I should also mention that assumptions were intentionally made in the program. That was out of necessity. It is necessarily empirically-derived in every case? No.... Unfortunately, there are limits to how much data is out there, but I largely made it out of my own curiosity. I'm offering it to the SDN community as such. If others have something to add or ways to revise it, I am certainly open to that.

eldoctor said:

13 12 12 36
10 9 9 31

Those were my 50% scores. Don't add up

Click to expand...

They shouldn't add up. They are confidence intervals and independent of one another. You have a 50% chance of landing between each of those. A 13-12-12 is outside the interval but a 13, 12, or 12 is not (and a 12x3 would be within the confidence interval). Likewise, the lower limits are 10, 9, and 9 respectively but the lower limit for the total score is independent of the subscores and is 31, so 10-10-11 or 12-9-10 is fairly likely but a 10-9-9 is relatively unlikely (i.e., it would be a very bad day for you based upon previous performance).

Do you guys think it would help to have it include MCAT practice tests in the main scoring algorithm since a site apparently did an informal analysis of how Kaplan, PR & AAMC exams correlate with the actual exam? It obviously wouldn't be as "scientific" but would it be useful to you?

SoFreakedOut said:

Cool tool but it's definitely hard to get an accurate relationship (like the disclaimer you wrote in the spreadsheet) with GPA. Some people screwed up, were lazy, etc. during undergrad and so their GPAs are not reflective of their actual ability or the knowledge they can amass by simply studying content for the MCAT. On the flip side, some people go to easier schools or schools with grade inflation and the estimator will greatly inflate their scores. Nothing you haven't thought about though, I'm sure. The idea of including practice MCAT exams, someone else brought it up, is a good one.

Click to expand...

Right.

Maybe I will add the processing of practice MCATs to it all. It already does use practice MCAT scores in the graph part of the program, which I actually find more interesting than the rest of it, to be honest... graphical representations, pretty pictures, and all that I guess! lol

riddler said:

Odd. When I put in any SAT and ACT scores only, I had a range of 38-45. As soon as I put in any GPA, however, it dropped dramatically (with GPAs 3.8 and 3.6 inserted, it ranges 28-37). I'm hesitant to believe both score ranges. The 38-45 just seems too high to be accurate, and my GPA isn't very accurate as of now, as I only have one semester counting for it.

Any insight on what I can take from this? Neat program, though!

Click to expand...

The GPA is the better estimate. 28-37 (which is a very wide range) indicates that a strong predictor (your GPA) is incongruent with your SAT and ACT scores in terms of percentile. Keep in mind, however, that with only one semester, your GPA doesn't really mean much. Try entering your GPA 2-3 semesters from now and your GPA actually will actually have measurable predictive value. The program assumes you are entering your current GPA shortly before you take the MCAT (since this is how the studies were performed). As a result, a 3.8/3.6 just before the exam says your VR is likely to be moderately high (11-12) but your science scores will likely be much more moderate (i.e., 9-10).

In other words, you shouldn't take much of anything from this as a freshmen. Wait until you've been through most of your college career before trying a program like this.

It takes the math and verbal subscores. Each subtest is scaled to 800. As a result, the new out of 2400 method (which adds a 3rd test) shouldn't really make (much of) a difference (i.e., we simply ignore the new subtest).

Good to hear you beat the odds! I tried to program it so that, if anything, it would UNDER and not over shoot your score. I used this thread to do some calibration work on it, so now people seem to be either hitting it just about right or scoring a bit higher than the spreadsheet predicts. That's right where I want it. What I wouldn't want is for many people to have their MCAT scores predicted higher than what is likely attainable for them. Usually people prefer GOOD surprises to bad ones....

hiphopapotamous said:

It said I'd get a 34 and I got 39. cGPA/sGPA predicted I would get a 33 - not a very powerful regressor since GPA's meaning changes drastically depending on the institution. If I had gone to Brown, it would probably have predicted more accurately that I'd get a 37. A 39 is a common score among the few people with my GPA at my college.

Click to expand...

GPA is actually a very powerful coorelate of MCAT scores; however, as you mentioned, schools do vary on it. Due to this, I would suggest relying more on the other test scores if they are more consistent.

Being off by 5 points in the downward direction is probably acceptable considering that a 39 is unlikely for anyone. (But congrats on pulling it off!!!) I calibrated it to estimate on the low side, if anything. I'd much rather someone walk in expecting a 34 and get a 39 (to their surprise) than walk in thinking they're all that because it says they're going to get a 39 and then get a 34 (which is still an awesome score but if you're expecting a 39, a 34 might actually be disappointing -- imagine that).

phltz said:

This was off by 12 points.

Click to expand...

Impressive...details?

NYR56 said:

I agree with this, although I don't know what statistics have to say about it. My GPA prediction was a 24. I scored a 39. My ACT/SAT predictions were much closer though - ACT said 42, SAT said 36, averaging out to my actual 39.

Click to expand...

The stats used indicate a stronger correlation of GPA (esp. sGPA for the sciences & cGPA for VR) and GRE scores with the MCAT than SAT and ACT. This points to possibilities of changes in performance over time (i.e., GREs and GPAs would be measured at a closer time to when the MCAT was taken than would SATs and ACTs).

Rabolisk said:

The "MCAT Estimator" is a useless tool. Measuring correlation between GPA and other test scores and the MCAT is useful, but to use this specific tool to "estimate" your MCAT score is a waste of time. For example, the 90% confidence range for me was between 44 and 20. There is simply too much variance in the meaning of the variables (GPA especially) for this to be meaningful. After all, all of us here have taken at least one practice MCAT before, and that's we would estimate our score based on that.

Click to expand...

You setting the tool to a 90% confidence interval is ridiculous. A basic understanding of factors involved in test scores should tell you that such a high level of confidence for one score is essentially impossible. Furthermore, if you are getting such a huge range, it means your scores are all over the place or not enough data are available. For me, a 90% confidence interval would be 23-42, but what makes more sense is to use the 1 SD setting, which sets it to an 85% chance above a given value 69% within the given range). For me, this yields 27-38, which gives a reasonable estimate that I would score somewhere in the average to good matriculant range for MD schools. Even the very bottom of that range is acceptable at lower-tier schools. Using a 90% range basically means you're asking it to make an estimate that includes all possibilities (and it's set to allow for significant amounts of sampling error, which is something using your MCAT test scores as a predictor does NOT do).

phltz said:

My spreadsheet estimation of 30 was severely dragged down by a lousy GPA. Actual score was a 42.

Click to expand...

Ah, ok. That makes sense. A very low GPA could obviously do that to the stats. I suppose I could have it drop certain outliers from calculation; however, w/ only 3 sources of data for most people (GPA, ACT, SAT), it would likely be difficult to know for sure which is the outlier w/o further information (e.g., you flunked 24 credits' worth of courses freshmen year b/c you weren't trying but then got all As and A-s for the rest of your college career, resulting in an overall GPA of 2.89).

Another issue here is probably the "cap" this tool has. Even with perfect scores, an ACT can never predict higher than 43. A perfect SAT gives a 39 MCAT. A perfect GRE gives a 42 MCAT. A percect (4.0 cGPA & sGPA) GPA only gives 37. This is because the percentiles associated with those scores, when adjusted for the premedical population, simply do not give a range equal to that of the MCAT -- in other words, a 1600 on the SAT (old version/equiv. 2400 on new version) is really about as rare as a 39+ on the MCAT (by fraction of population taking the exam adjusted for the difference in populations).

For instance:
1600 SAT (2010): 1305/1,518,859 or top 0.08592% (Z=+3.135)
39+ MCAT: top 0.80% (Z=+2.41)
Since the Z-scores are for populations known to be separated by ~1 SD (HS seniors vs. college seniors), we would expect the difference between these Z-scores to be ~1 (it is, in fact, a 0.72 difference)