Admit Standardized Score and School List Builder

[HappyRabbit]

Hey everyone, as some of you might have seen, I’ve been working on a machine learning model over the last few months that can both score an application based on every data point used in the admissions process as well as build a school list specific to an applicant’s holistic profile. The goal of the created school list is to maximize the probability of at least one acceptance as well as the chance of the highest-ranked acceptance.

It accounts for every exception and every decision, such as when to apply DO/MD (even with high stats), how X-factors change your ability to apply to reaches, MCAT retakes, GPA upward trends, in-state vs out-of-state schools, research, extracurriculars, etc. The best way to see it in action is to just try it with any combination of stats and ECs that you can think of and see how the school list and score changes.

Applicants with the same score, for example, can have drastically different school lists depending on the builder’s holistic evaluation of the application (research-focused, service-focused, state of residence, etc).

The reason I made the school list builder was because I saw so many applicants missing out on their full potential simply because they weren't making the right lists - either by applying to the wrong schools and having to reapply, or by not applying to schools they were competitive for. As a low SES applicant, I know how much harder the process can be when you don’t have the same resources others do, whether that’s family in medicine or access to consultants/advisors.

This should hopefully make it much easier for everyone to know where to apply and what schools you are competitive for. The builder also allows you to customize your list by suggesting schools that you are competitive for that can replace the recommended ones.

That’s essentially the TL;DR of the post, you can try it out below:

Link: https://admit.org/

If you want to learn more about how it was developed, you can read this doc here.

Note: I still need to include post-bacc to the calculator, which is coming soon. However, you can still get an idea of where to apply if you did a post-bacc by increasing your GPA score (this isn’t exactly correct because only some schools reward reinvention, but post-bacc is being added soon).

---

Members don't see this ad.

 

[Angler]

@HappyRabbit I admit that when I first saw your initial iterations of admit.org, I was skeptical of your ambitions. I tried it out and read your google doc. I'm impressed at the amount of work that went into this and the results (from several scenarios that I pulled out of my back pocket).

Just a couple of quick thoughts since I only skimmed your methodology.

Do you account for imbalanced MCAT scores? Like a 126/122/130/132?

Are you thinking of a way to measure the quality of results? Maybe turn this into an academic research project? This might be difficult...

Anyway, keep working on this

 

[wysdoc]

Hey everyone, as some of you might have seen, I’ve been working on a machine learning model over the last few months that can both score an application based on every data point used in the admissions process as well as build a school list specific to an applicant’s holistic profile. The goal of the created school list is to maximize the probability of at least one acceptance as well as the chance of the highest-ranked acceptance.

It accounts for every exception and every decision, such as when to apply DO/MD (even with high stats), how X-factors change your ability to apply to reaches, MCAT retakes, GPA upward trends, in-state vs out-of-state schools, research, extracurriculars, etc. The best way to see it in action is to just try it with any combination of stats and ECs that you can think of and see how the school list and score changes.

Applicants with the same score, for example, can have drastically different school lists depending on the builder’s holistic evaluation of the application (research-focused, service-focused, state of residence, etc).

The reason I made the school list builder was because I saw so many applicants missing out on their full potential simply because they weren't making the right lists - either by applying to the wrong schools and having to reapply, or by not applying to schools they were competitive for. As a low SES applicant, I know how much harder the process can be when you don’t have the same resources others do, whether that’s family in medicine or access to consultants/advisors.

This should hopefully make it much easier for everyone to know where to apply and what schools you are competitive for. The builder also allows you to customize your list by suggesting schools that you are competitive for that can replace the recommended ones.

That’s essentially the TL;DR of the post, you can try it out below:

Link: https://admit.org/

If you want to learn more about how it was developed, you can read this doc here.

Note: I still need to include post-bacc to the calculator, which is coming soon. However, you can still get an idea of where to apply if you did a post-bacc by increasing your GPA score (this isn’t exactly correct because only some schools reward reinvention, but post-bacc is being added soon).

I think your suggested school lists are much too long to be of any practical use and include too many schools not in realistic range for an applicant’s numbers and geographical location.
Human judgement and experience is still of value in creating lists

 

[HappyRabbit]

@HappyRabbit I admit that when I first saw your initial iterations of admit.org, I was skeptical of your ambitions. I tried it out and read your google doc. I'm impressed at the amount of work that went into this and the results (from several scenarios that I pulled out of my back pocket).

Just a couple of quick thoughts since I only skimmed your methodology.

Do you account for imbalanced MCAT scores? Like a 126/122/130/132?

Are you thinking of a way to measure the quality of results? Maybe turn this into an academic research project? This might be difficult...

Anyway, keep working on this

Thanks! For now, I don't handle subsections along with post-bacc and graduate GPA - I plan on suggesting the applicant to apply to more baseline scores but I didn't want to add either until I was 100% confident it was correct. It's really hard to know what school screens and there are always exceptions to screening that are difficult to quantify, but I think the straightforward solution is to add more of a base. Let me know what you think about that.

 

[HappyRabbit]

I think your suggested school lists are much too long to be of any practical use and include too many schools not in realistic range for an applicant’s numbers and geographical location.
Human judgement and experience is still of value in creating lists

Yeah of course - there are limitations to every automatic school list builder and there's only so much a computer can really do. This should never replace the human eye, especially in very complex applications, but it should help applicants make a base list to build off of.

I currently build schools to be 35 in length and customizability is built in for applicants to drag schools in and out based on their personal preferences (location, etc). Schools not included in the main list but that are still in range for the student show up on the sidebar.

 

[HappyRabbit]

This is really well done. Only suggestion I have is to maybe decrease the required number of clinical hours for Harvard/Stanford, especially when an applicant has high stats, lots of research, etc. 300 is a somewhat arbitrary cutoff, and these two schools were more forgiving than Hopkins, UCSF, etc. of my low clinical hours given my strong spikes elsewhere.

Agreed, I'll have to put some thought into combining stats and research for these two particular schools to make a sliding scale that reduces the threshold for clinical and nonclinical as the former increase until a lower baseline is hit. Thanks for the feedback!

 

[Angler]

I doubt that any thoughtful applicant is going to use such a list as their end product. That is, until we get robots applying to med school. I think 35 is a reasonable number as a universe to research and choose from. Better than a list of, say, 15 schools. Maybe a list of also-rans would be good, and would be more useful than your right hand list of all schools.

Faha frequently has lists that are pretty substantial, and he/she/it provides a valuable service.

What's going to happen when you start med school? You obviously can't put this amount of work into this concurrently.

 

[HappyRabbit]

Yeah definitely, Faha's lists are really good and everyone should get their advice.

As far as med school, tomorrow's problem, still got 6 months left 😛

 

[BattalionX]

I like it. More than half of my interviews were from the target school portion and the other half in safeties. If I had this tool I may have been more encouraged to apply to more competitive schools like Rochester, Colorodo, UCSF, Emory. I think I undervalued my application odds just because I hyperfixated on my 512 and only applied to schools with MCAT averages at/under 516.

Obviously not perfect, but it's a cool tool.

 

[HappyRabbit]

I like it. More than half of my interviews were from the target school portion and the other half in safeties. If I had this tool I may have been more encouraged to apply to more competitive schools like Rochester, Colorodo, UCSF, Emory. I think I undervalued my application odds just because I hyperfixated on my 512 and only applied to schools with MCAT averages at/under 516.

Obviously not perfect, but it's a cool tool.

Yeah that's why I built it - lots of applicants either undervalue or overvalue their applications simply because they don't know better. Even if the builder is 90% accurate and there are some issues, I'm hoping that it can at least cause applicants to learn more about why the schools were suggested, find out they are more competitive than they think they are, and add more reaches (or in the latter case, add more safeties if they are overvaluing themselves). WAMC threads are going to still be more accurate but the generated list can be used as a beginning list rather than starting from scratch.

 

[FernThePlant]

Really cool to see the generated list vs my current cycle! 8 out of my 9 IIs were on the pre-made list -- looks like a great starting place for applicants to whittle down from based on mission fits and preferences

 

[HappyRabbit]

6 of my 7 interviews were on the list admit.org recommended, with the exception of Tulane which has a note that says "we do not recommend this school". Just curious why this note exists for Tulane?

Edit: Never mind, just realized this note exists for all schools that the algorithm does not recommend for a particular applicant, versus being a blacklisted school.

Are your stats higher than their medians?

 

[deleted982921]

So far based on my interviews and acceptances this seems spot on! I would make a note when entering MCAT scores because I entered my most recent score first and then my previous score, and it really impacted my list. When I fixed it the list was much more accurate

 

[HappyRabbit]

So far based on my interviews and acceptances this seems spot on! I would make a note when entering MCAT scores because I entered my most recent score first and then my previous score, and it really impacted my list. When I fixed it the list was much more accurate

oops, I'll try to see if I can add some headers to make that more clear. Thanks!

 

[HappyRabbit]

It seems the only large issue with the builder is applicants who, stat-wise, are at the border between MD and DO (think like 507/3.7) who have thousands of clinical hours but sub 50 nonclinical hours. The calculator applies a significant penalty because of the lack of nonclinical, recommending a DO list, but some applicants are still able to get 1 MD interview that turns into an acceptance.

Should I adjust the logic to support applying to MD in this case?

 

[BattalionX]

I will say I wish the list had additional recommended schools or the option to filter out MD/DO's in the list. I don't know why TCU (MD) is ranked lower than LECOM or DMU either lol. I have a friend who is applying with a 3.1 GPA and 510 MCAT (Hispanic, nontrad 10+ years, X factors, CEO of a company, low GPA cause of failed English 5+ times in the past), and while I understand no calculator/program should be set to accomodate for these statistical outliers, it would be cool to see which schools the program would recommend outside of the 3 MD's listed (Rush, FSU, TCU). Instead, it's just a ton of DO recommendations, and no further option to explore similar MD's to Rush/FSU/TCU. But just a thought from me.

 

[HappyRabbit]

I will say I wish the list had additional recommended schools or the option to filter out MD/DO's in the list. I don't know why TCU (MD) is ranked lower than LECOM or DMU either lol. I have a friend who is applying with a 3.1 GPA and 510 MCAT (Hispanic, nontrad 10+ years, X factors, CEO of a company, low GPA cause of failed English 5+ times in the past), and while I understand no calculator/program should be set to accomodate for these statistical outliers, it would be cool to see which schools the program would recommend outside of the 3 MD's listed (Rush, FSU, TCU). Instead, it's just a ton of DO recommendations, and no further option to explore similar MD's to Rush/FSU/TCU. But just a thought from me.

The ranking is independent of the builder and is taken from PD rank lists - I felt it would be weird to just automatically rank all MD schools over DO and rather use something more objective.

And yeah definitely, right now I'm working on only marking "not recommended" schools to those that the applicant should definitely not apply to rather than those the builder does not recommend for one reason or another. In the near future, not recommended schools will be exclusive to things that are extreme reaches stats wise or IS heavy schools.

 

[burneracc]

Seems like a great tool for people to work on school lists. As a data point, I applied this current cycle and got 4/12 interviews from reach schools, 1/12 from target (lol), and 2/6 from baseline (not sure why it gave me so few of them).

Would have loved something like this to start making a list back when I was making mine! Im sure your efforts will be appreciated by applicants for years into the future.

 

[HappyRabbit]

Seems like a great tool for people to work on school lists. As a data point, I applied this current cycle and got 4/12 interviews from reach schools, 1/12 from target (lol), and 2/6 from baseline (not sure why it gave me so few of them).

Would have loved something like this to start making a list back when I was making mine! Im sure your efforts will be appreciated by applicants for years into the future.

Thanks! Usually when few baselines are given, it's because the builder shifts the focus on getting into target/reach schools with state schools as the fallback. I'm hoping to make the options for schools more clear now, so rather than showing every not recommended school in red, it will differ between actual do not apply schools (in-state heavy, super out of reach, etc) and schools that are just a little above or below range but still eligible.

 

[Mr.Smile12]

Most of our STEMM-trained students love being overwhelmed with data. At least getting a sense of the landscape is helpful. It is something that I'm sure the commercial consultant companies are working on as well.

I also will say it helps us advisors with focusing students on next steps to refine the list and strategies networking or unwritten rules/expectations. The early warning about overreliance on bots applies, as does the fact that it can help one to be more efficient.

 

[deleted1124957]

Whats the word on the street, how much is LGBTQ status adding to the app?

Similar to URM?

 

[HappyRabbit]

Whats the word on the street, how much is LGBTQ status adding to the app?

Similar to URM?

Nah it’s quite different; it changes the reach ability at certain schools that favor LGBTQ applicants

 

[HappyRabbit]

First changes being added now are:

1) Fixing Stanford/Harvard edge case where it was being removed from lists if an applicant had low nonclinical hours but high research

2) Applicants with an Admit score at the border between MCAT and DO, who have low nonclinical but thousands of clinical hours, will not be penalized for lack of NC and told to apply to certain MD schools.

3) Reason behind "not recommended" for each school is being added (for example, IS heavy school or an HBCU, or if an applicant's stats are out of range)

4) Adding a yellow icon for schools that are slightly out of reach for applicants (or too much of a safety) so applicants can still add them to the list if they don't mind taking the risk.

Thanks for all the feedback so far!

 

[MrPuff]

Hello HappyRabbit,

Kudus for putting in the time and energy on the admit.org site. When I plug in my data I get a list that looks similar to the one I've already built, so the algorithm appears to function decently. One thing I noticed is that, for me, when I only change my MCAT from 516 to 518 I get pretty drastically different lists - 4 T20s v 12 T20s. Do you think that the difference (92nd to 95th percentile) in MCAT score should give that large of difference in the list output? Perhaps so, but it looks greater that I would have expected. Again, great job on building your school selection tool.

 

[HappyRabbit]

Hello HappyRabbit,

Kudus for putting in the time and energy on the admit.org site. When I plug in my data I get a list that looks similar to the one I've already built, so the algorithm appears to function decently. One thing I noticed is that, for me, when I only change my MCAT from 516 to 518 I get pretty drastically different lists - 4 T20s v 12 T20s. Do you think that the difference (92nd to 95th percentile) in MCAT score should give that large of difference in the list output? Perhaps so, but it looks greater that I would have expected. Again, great job on building your school selection tool.

Yeah it should, 516 is too low for most T20 schools without X-factors with the exception of schools like Pittsburgh which should be on your list. To quantify it, a 516 applicant really has max visibility at schools with 519 and not much more than that. 516 would require a combo of low SES and publications with great ECs to even have a small shot.

Most T10 schools screen below 518 for ORM non-tied/X factor applicants (unfortunate but is what it is)

 

[MrPuff]

Thanks for the reply! When I run the program I get Duke, Cornell and Northwestern - all 520 median accepts - and Icahn (518) for reaches. Pitt (518) is listed as a target. I have all five of these schools on my personal list as reaches. So again, the algo gets similar results to my list. I also have Duke (520) and Mayo (521) on my list. I don't have T10s on my list with the exception of Duke which I will likely remove. Thanks again.

 

[HappyRabbit]

Thanks for the reply! When I run the program I get Duke, Cornell and Northwestern - all 520 median accepts - and Icahn (518) for reaches. Pitt (518) is listed as a target. I have all five of these schools on my personal list as reaches. So again, the algo gets similar results to my list. I also have Duke (520) and Mayo (521) on my list. I don't have T10s on my list with the exception of Duke which I will likely remove. Thanks again.

You should try Duke if that's what's being recommended unless you don't want to for personal reasons. Good luck with applying!

 

[lesbianherbs]

Very cool! Comparing its suggestions to my cycle so far (as a high stat applicant):

I also applied to 4 other schools that were not suggested for me, and have not heard back from any of them, lol

 

[Emh_omega-323]

Interesting... I feel like the list made for me would have been overambitious for me as a CA ORM without some X factor-light features of my app not captured in this formula. As it is I have an A at a base tier, 2 pending decisions in target, and a pending decision in the reach tier. So not a bad estimate.

Also, I still think that "mission fit" is the school list making factor no one wants to deal with, but the most important one. Every school I have an II with has a special interest in a population I have unique volunteer work with.

 

[HappyRabbit]

Interesting... I feel like the list made for me would have been overambitious for me as a CA ORM without some X factor-light features of my app not captured in this formula. As it is I have an A at a base tier, 2 pending decisions in target, and a pending decision in the reach tier. So not a bad estimate.

Also, I still think that "mission fit" is the school list making factor no one wants to deal with, but the most important one. Every school I have an II with has a special interest in a population I have unique volunteer work with.

Over time I'll expand the questions asked and the way schools are suggested/ranked to account for mission fit (or at least as much as I can). It's very difficult but should be doable with time as I take each piece at a time.

 

[Mr.Smile12]

Mission fit is a difficult item for such an algorithm. Even AAMC helps schools document how they live up to their mission (Mission Management Tool, internal use within the Curriculum Dean's office). You know... for accreditation... all of that is confidential, with new things developed all the time. Networking can help you find out more. It also can change with leadership or organizational changes (like grant funding).

Again, I preach. Mission fit, mission fit, mission fit.

Mission Fit: Key to Your Success - SDN

The Student Doctor Network provides free tools, resources, and advising services to help students become health professionals.

www.studentdoctor.net

 

[alwaysrunning]

Wow! Well done!

I'm a parent of a premed who has been following the admissions process closely, including talking to a few admissions deans, in order to help my kid. Here are a few thoughts:

1 I love that you included a LOR from a lab director and LGBTQ. Both of those seem to have a bigger impact than applicants realize.

2 I think it would help to add poster presentation and conference presentation to the published prompt. They're not the same as published but they do have an impact.

3 I think you need a way to capture time spent working with medically underserved communities. Perhaps ask how many of the clinical and volunteer hours are with these groups. I've noticed this aspect has a big impact during admissions but I'm not sure exactly how

Thanks for putting this together. It's definitely going to help a lot of applicants!

 

[HappyRabbit]

Wow! Well done!

I'm a parent of a premed who has been following the admissions process closely, including talking to a few admissions deans, in order to help my kid. Here are a few thoughts:

1 I love that you included a LOR from a lab director and LGBTQ. Both of those seem to have a bigger impact than applicants realize.

2 I think it would help to add poster presentation and conference presentation to the published prompt. They're not the same as published but they do have an impact.

3 I think you need a way to capture time spent working with medically underserved communities. Perhaps ask how many of the clinical and volunteer hours are with these groups. I've noticed this aspect has a big impact during admissions but I'm not sure exactly how

Thanks for putting this together. It's definitely going to help a lot of applicants!

Yeah absolutely, over time I’ll expand the depth of questions to include things like abstracts/posters, leadership, and specifics around the clinical/nonclinical hours. Thanks for the feedback!

 

[MercifulCamper]

I'd say this is pretty good! I will say that last cycle I got into quite a few of my "reach" schools but was ghosted by a good amount of the "target"/"baseline" schools but this an n=1 and it gave me a similar list to what I applied to last year. My application was a high research hours/clinical hours application. Good job.

 

[HappyRabbit]

I'd say this is pretty good! I will say that last cycle I got into quite a few of my "reach" schools but was ghosted by a good amount of the "target"/"baseline" schools but this an n=1 and it gave me a similar list to what I applied to last year. My application was a high research hours/clinical hours application. Good job.

Yeah this can happen either because of yield protection or lack of mission fit (particularly in high research applicants applying to service schools)

 

[MercifulCamper]

I entered my info and the list is quite similar to the list I generated on my own. I applied to most of the top 25ish plus local schools in my home state and the state I went to college (the latter of which helped me absolutely zero!).

The targets and reaches generated by this algorithm for me were essentially the top 25ish schools while the baseline was filled with the local schools mostly. Using my list with the categories generated by this algorithm, I applied to 13 reaches, 9 targets, and 8 baselines (though when I was applying I considered it 22 reaches and 8 targets because the top 25ish seemed to be reaches for everyone and I didn't consider any school a baseline).

Looking back, the baseline schools were largely wasted applications. They "basically" ignored me. Honestly, I did even worse with my targets (generally T15-25ish). I got 3 times as many II's from reaches than baselines and 6 times more than targets.

In hindsight I now wonder if my time would have been much better spent applying to fewer schools by cutting the targets by say 50 percent and baselines by say 75 percent. Perhaps that would have generated stronger applications for the reaches that I ended up not receiving II's from (7 of 13).

I'm certainly not complaining. To the contrary, I'm tremendously grateful. It looks like barring major differences in the cost of attendance that I'll be attending my first choice medical school. I just question whether the traditional wisdom of applying to a broad set of schools makes sense for everyone. I can't help but think that applying to 10 fewer schools (20 instead of 30) would have generated a net increase in II's from schools that would have seriously considered me and therefore possibly more A's.

But, of course, it might not have. So who knows.

Anyway, if it's helpful for anyone, here is how my data breaks down based on this algorithm:

Reaches
Applied: 13
II's: 6
A's: 1 (3 decisions pending)
WL: 1 (3 decisions pending)
Post II: R 1 (3 decisions pending)
Pre II R: 7

Targets
Applied: 9
II's: 1
A's: 1
Pre II R: 8

Baselines
Applied: 8
II's: 2
A: 1
WL: 1
Pre II R: 6

I mean at the end of the day you got 9 interview invites - you are not the norm. If you were to just even get 3 interview invites from your targets and baseline schools in total your cycle would be a huge success. I think on SDN we need to give ourselves a reality check that it is hard to even get an interview sometimes when you apply to 20+ schools

 

[HappyRabbit]

I entered my info and the list is quite similar to the list I generated on my own. I applied to most of the top 25ish plus local schools in my home state and the state I went to college (the latter of which helped me absolutely zero!).

The targets and reaches generated by this algorithm for me were essentially the top 25ish schools while the baseline was filled with the local schools mostly. Using my list with the categories generated by this algorithm, I applied to 13 reaches, 9 targets, and 8 baselines (though when I was applying I considered it 22 reaches and 8 targets because the top 25ish seemed to be reaches for everyone and I didn't consider any school a baseline).

Looking back, the baseline schools were largely wasted applications. They "basically" ignored me. Honestly, I did even worse with my targets (generally T15-25ish). I got 3 times as many II's from reaches than baselines and 6 times more than targets.

In hindsight I now wonder if my time would have been much better spent applying to fewer schools by cutting the targets by say 50 percent and baselines by say 75 percent. Perhaps that would have generated stronger applications for the reaches that I ended up not receiving II's from (7 of 13).

I'm certainly not complaining. To the contrary, I'm tremendously grateful. It looks like barring major differences in the cost of attendance that I'll be attending my first choice medical school. I just question whether the traditional wisdom of applying to a broad set of schools makes sense for everyone. I can't help but think that applying to 10 fewer schools (20 instead of 30) would have generated a net increase in II's from schools that would have seriously considered me and therefore possibly more A's.

But, of course, it might not have. So who knows.

Anyway, if it's helpful for anyone, here is how my data breaks down based on this algorithm:

Reaches
Applied: 13
II's: 6
A's: 1 (3 decisions pending)
WL: 1 (3 decisions pending)
Post II: R 1 (3 decisions pending)
Pre II R: 7

Targets
Applied: 9
II's: 1
A's: 1
Pre II R: 8

Baselines
Applied: 8
II's: 2
A: 1
WL: 1
Pre II R: 6

The way the lists are created is to use baseline schools as a safety net, especially with high stat applicants. In this case there are two outcomes:

1) The applicant is competitive for the recommended reach and target schools in which case rejections from baseline schools due to yield protection don’t matter (you got interviews at reaches!)

2) The applicant is not competitive for reaches despite being recommended them due to unpredictable reasons (poor writing, etc) which baseline schools may notice and therefore send interviews out.

One of the components of the builder that I spent the most time on is determining how long to make the school lists for every scenario - you will notice that extremely competitive applicants have shorter school lists with most of the length weighted towards the top, whereas applicants who are less competitive are the complete opposite.

 

[Opuntia]

As an applicant with a 515, I anticipated only a few Top 20s (if any) to pop up on my recommended list. I have an X-factor though, so I applied to many Top 20s with a good number of IIs and acceptances (27 secondaries, 14 IIs, 7 As, 1 WL so far). Your list was SPOT ON with predicting where I would be successful! Love to see it. Thanks for this amazing tool. Great that it's free and without ads too. You should be super proud of this!!

 

[Mr.Smile12]

I just question whether the traditional wisdom of applying to a broad set of schools makes sense for everyone. I can't help but think that applying to 10 fewer schools (20 instead of 30) would have generated a net increase in II's from schools that would have seriously considered me and therefore possibly more A's.

Congratulations on your success.

You have to remember that when you receive advice in general, we don't know you, so most of the advice will be probabilistic or drawn from personal experience (applicant or advisor). The rule to apply to 20 schools is an overall average for all applicants (that's what we see in the AMCAS data). Final decisions are up to the individual who I would hope has a better sense of their fit with target/reach schools. We just give you the probabilities based on publicly accessible but averaged data.

In the end, most of us want to see you get into one school because that's also what the AMCAS data tell us (most only get 1 A). To have multiple bites at the A(pple) is something we would hope for. As with all tools and resources, they are there to help you, not make decisions for you. You can get the interview invitations, but you have to close.

 

[FlowRate]

Very cool project!

What's the underlying data for your ML model? What sort of algorithm are you using? Sounds like you're making manual adjustments on top of that?

Also seems like maybe hug of death? I can't access the website.

 

[HappyRabbit]

Very cool project!

What's the underlying data for your ML model? What sort of algorithm are you using? Sounds like you're making manual adjustments on top of that?

Also seems like maybe hug of death? I can't access the website.

I manually labeled thousands of applications with schools that credible SDN posters gave to applicants + verified the recommendations with my own input. You can do this to find the weights for every input into the builder as well as the more difficult part which is seeing how each category impacts each other (for example, weight of being URM alone is different than when combined with XYZ extracurriculars, and is different at every MCAT/GPA combination). This was the biggest flaw with the WARS calculator which treats each activity category as a fixed, independent boost and simply adds sections scores together to make one final score. Every aspect of the Admit school list builder is interlinked to make a more holistic evaluation of an application.

After that it’s just a matter of fine tuning, finding the weaknesses of the model, and adding to the training set + manual adjustments when I see something wrong.

As far as website not working, some niche ISP’s are blocking it but it works for 90% of people (I tried submitting forms to I think Optimum to remove the block but they said I need to be a paying customer to make one 😀)

 

[FlowRate]

I manually labeled thousands of applications with schools that credible SDN posters gave to applicants + verified the recommendations with my own input. You can do this to find the weights for every input into the builder as well as the more difficult part which is seeing how each category impacts each other (for example, weight of being URM alone is different than when combined with XYZ extracurriculars, and is different at every MCAT/GPA combination). This was the biggest flaw with the WARS calculator which treats each activity category as a fixed, independent boost and simply adds sections scores together to make one final score. Every aspect of the Admit school list builder is interlinked to make a more holistic evaluation of an application.

After that it’s just a matter of fine tuning, finding the weaknesses of the model, and adding to the training set + manual adjustments when I see something wrong.

As far as website not working, some niche ISP’s are blocking it but it works for 90% of people (I tried submitting forms to I think Optimum to remove the block but they said I need to be a paying customer to make one 😀)

Interesting, so your data set is school recommendations, not II/Acceptances? Too bad that old site where everyone posted their stats+schools went defunct.

 

[HappyRabbit]

Interesting, so your data set is school recommendations, not II/Acceptances? Too bad that old site where everyone posted their stats+schools went defunct.

Eventually as I build out other features the latter will be possible for students who want to contribute their data

 

[NotADoctor1]

Eventually as I build out other features the latter will be possible for students who want to contribute their data

Maybe you could try using some of the Sankey diagrams people post on r/premed. Some people provide a lot of detail about their application

 

[Mr.Smile12]

Maybe you could try using some of the Sankey diagrams people post on r/premed. Some people provide a lot of detail about their application

Yeah, but most people don't, especially nowadays where people are fearful of getting doxxed.

It just makes the paid consultant companies happier because they can keep all their data proprietary.

 

[HappyRabbit]

Yeah, but most people don't, especially nowadays where people are fearful of getting doxxed.

It just makes the paid consultant companies happier because they can keep all their data proprietary.

Working on a solution to this that should help admissions become more democratized, will share more soon.

 

[Mr.Smile12]

Working on a solution to this that should help admissions become more democratized, will share more soon.

A lot more people lurk and keep to themselves. Many are posting for "friends."

 

[HappyRabbit]

A lot more people lurk and keep to themselves. Many are posting for "friends."

Yeah there needs to be an incentive structure that motivates applicants to share their info. At the moment there is no upside aside from WAMC advice and the perceived infinite downside from being doxxed.

 

[HappyRabbit]

As an applicant with a 515, I anticipated only a few Top 20s (if any) to pop up on my recommended list. I have an X-factor though, so I applied to many Top 20s with a good number of IIs and acceptances (27 secondaries, 14 IIs, 7 As, 1 WL so far). Your list was SPOT ON with predicting where I would be successful! Love to see it. Thanks for this amazing tool. Great that it's free and without ads too. You should be super proud of this!!

Missed this, that's awesome! Congrats on your success this cycle 🙂

 

[FlowRate]

Worked when I was not on my work VPN. Would be interesting if the tool said why it was or was not recommending a certain school.