How much of an impact does GPA, specifically math GPA, have on getting admitted

[davidolohowski]

I have a friend who is applying for stats PhD right now and he wants to do probability theory. But his mindset on GPA really got me interested in the impact of GPA on admission.

What he did was during his Master degree, he took all the graduate level pure math courses our math department has to offer. He was literally taking 9 or 10 courses each semester. He still managed to get good grades in all of them. But as far as I know, he does not have enough time to work on research during Master degree simply because he was busy doing all those courses. He did have a paper published in applied math during undergrad though. I am also guessing the reason he wants to do research in probability theory is because he enjoys doing proofs.

The other day, we got into an argument on how exactly does GPA and course work affect your chance of getting admitted. My stance is that GPA and math GPA is just served as a cutoff mechanism to filter out people who are clearly not gonna survive during a stats PhD. It essentially is just an indicator that you have the basic foundational ability in taking in new materials and understanding literature. The most important thing should be your reserach potential which is demonstrated through your SOP and reference letters. But he is arguing that the fact he excelled in all the pure math courses sets himself apart from many others and gives him an definitive advantage over others. Then he went on and argue that the most important thing for PhD admission is coursework while SOP and letters don't matter that much simply because everybody is likely doing different research and there is no universal assessment on reserach ability when people are doing completely different research.

So my question is, what is the actual impact of GPA/coursework on admission? Whose opinion is the right one? Also, just out of curiosity, given my friend's scenario, what is his chances of getting into a top stats PhD program?

Edited November 10, 2019 by davidolohowski

[bayessays]

Your friend is mostly right. In a field like computer science, research is super important and the grades are mostly to see that you're a somewhat competent student. Almost nobody does real statistics research prior to a PhD, so GPA and GRE scores are all you have to go off of. Also, most programs have intense qualifying exams that require you to be good at math to pass so they want the students who will do the best. Letters matter, so research is important in that it will get you positive letters, but math ability is really the most important thing.  If you go to a good school, and your friend has As in a lot of graduate math classes, he will do very well in admissions. 

[davidolohowski]

Thank you very much for clearing that. I always had the impression the most important part about stats admission is research potentials. But then I have another question which is sort of a comparison between me and him. I understand it's not really appropriate or maybe blatantly wrong in doing comparison with others but because both of us were trying to get into the same school, there's direct competition in this scenario, so I am indeed wondering. I am kind of in the opposite direction of him where I only took the generally required real analysis and grad level probability theory and did very well in them but those are the only pure math/proof based courses I took. I spent most of my time focusing my research. But I guess I was lucky enough that my research is currently prepared for publications for two papers, one applied and the other in statistical computing. My main interests are also different from his in that I am more applied heavy and interested in methodological development for applied problems. I also did my damnedest in terms of SOP and I got 2 very strong letters. So in this scenario, who has a better chance then? 

Edited November 10, 2019 by davidolohowski

[Casorati]

I kinda have the similar question. I am from a Canadian school and we do percentage grading. So 85-89 is an A and anything >=90 is an A+. Do admissions committees really distinguish 85 and 99 or they don't care that much above when grades are above a certain threshold?

Edited November 10, 2019 by Casorati

[bayessays]

@Casorati I don't think you need higher than an A, but when the number is sitting there for you to see, higher is better. 

@davidolohowski I think which of you will do better relies on so many factors - what school you go to (and the professors there, which will affect both how your grades and letters are interpreted), your exact classes, what types of journals your research is in, GRE scores, etc.  In general, math classes and grades in them are important.  But really good research and letters can mean a lot - but most people don't have letters or research that make a huge impact.  If you did well through real analysis, your friend's extra math won't be a huge deal and you could possibly be in better shape - taking absurd amounts of math isn't a free ticket into a PhD program.

My biggest point was more your question about the GPA cutoff.  In many fields, the difference between a 3.6 and a 4.0 isn't big because the research is the differentiator.  That is not the case in statistics. 

[davidolohowski]

2 hours ago, bayessays said:

I think which of you will do better relies on so many factors - what school you go to (and the professors there, which will affect both how your grades and letters are interpreted), your exact classes, what types of journals your research is in, GRE scores, etc.  In general, math classes and grades in them are important.  But really good research and letters can mean a lot - but most people don't have letters or research that make a huge impact.  If you did well through real analysis, your friend's extra math won't be a huge deal and you could possibly be in better shape - taking absurd amounts of math isn't a free ticket into a PhD program.

My biggest point was more your question about the GPA cutoff.  In many fields, the difference between a 3.6 and a 4.0 isn't big because the research is the differentiator.  That is not the case in statistics. 

Thanks for the reply. Since you mentioned letters and reserach making impact, it made me wonder how much of an impact my current situation in research have on admission. I'm also very interested in what makes the research component really meaningful. 

Here is a little specifics about my research experience and letters: The stats department at my school is about top 150 in the world. I am in astrostatistics. So I have two supervisors, one in astro and one in stats. Both of my profs are top in their fields with the stats prof being probably the best researcher in my department and is currently professor emeritus. In terms of paper, my astro prof told me I should definitely get the applied part of my work published in astronomy journals, hence the applied paper. My stats prof told me I should get the methodological work published as well, hence the computing paper. I was told to aim for journal of computational and graphical statistics. It is mainly on enabling Hamiltonian Monte Carlo for a certain form of intractable likelihood distribution. I will be first author for both papers. In terms of letters, my astro prof told me he wrote a very strong one and I should be a strong candidate at any school. But I would take that with a grain of salt because he came from a different field. For my stats prof, he straight up showed me his letter and he said in the letter that I was among the top 5 students he had during his entire career and my Masters thesis is on par of a PhD thesis, at least for the standard of my department. 

Based on above, how much does my research and letter component contribute to my application? I am aiming for schools around top 20 - 50 in the world.

Edited November 10, 2019 by davidolohowski

[cyberwulf]

Since very few have an established research track record before starting a PhD program in statistics/biostatistics, admissions committees in these disciplines are basically trying to find the smartest and most talented people they can. The reality is that prior research experience and a well-written SOP aren't a strong marker of intellectual ability (at least within the pool of people being considered for PhD admission at good programs). On the other hand, excelling at a bunch of challenging math courses taken with a strong peer group does provide some indication that an applicant has abilities that exceed most in that group, making them a good bet for PhD program success. Recommendation letters can also help make this point; the most helpful variety of these from an admissions perspective say something like "I think that Student X is as good or better than Student Y, who recently received his Ph.D. from Prestigious U, and is taking a faculty position at Top University."

Another reason that research experience is downweighted: there's a persistent myth that a large proportion of students applying to PhD programs with good grades are just "school smart" and won't have the creativity and motivation necessary to do dissertation research. In my experience, this profile is rare; for the most part, "school smart" students are also "research smart". Indeed, it's far more common to see students who appear "research smart" based on previous research experiences to struggle with the technical rigor required in a good PhD program.

[davidolohowski]

1 hour ago, cyberwulf said:

Since very few have an established research track record before starting a PhD program in statistics/biostatistics, admissions committees in these disciplines are basically trying to find the smartest and most talented people they can. The reality is that prior research experience and a well-written SOP aren't a strong marker of intellectual ability (at least within the pool of people being considered for PhD admission at good programs). On the other hand, excelling at a bunch of challenging math courses taken with a strong peer group does provide some indication that an applicant has abilities that exceed most in that group, making them a good bet for PhD program success. Recommendation letters can also help make this point; the most helpful variety of these from an admissions perspective say something like "I think that Student X is as good or better than Student Y, who recently received his Ph.D. from Prestigious U, and is taking a faculty position at Top University."

Another reason that research experience is downweighted: there's a persistent myth that a large proportion of students applying to PhD programs with good grades are just "school smart" and won't have the creativity and motivation necessary to do dissertation research. In my experience, this profile is rare; for the most part, "school smart" students are also "research smart". Indeed, it's far more common to see students who appear "research smart" based on previous research experiences to struggle with the technical rigor required in a good PhD program.

Thanks for the reply. I see your point. Then to what point is it enough to demonstrate ability in technical rigor? Is it like what my friend did, the more the merrier? Or does getting good grades in real analysis is simply enough? 

[Casorati]

I have a friend who applied two years ago. He took many advanced math courses such as measure theory, functional analysis, graduate probability theory and did very well in them, but he had no statistics research experience. He applied to ~6 schools ranging from Chicago to UNC and ended up getting into nowhere. 

Edited November 10, 2019 by Casorati

[icantdoalgebra]

To add onto this thread, I had asked one of the professors at my institution (UC Berkeley) what sort of is the most important things in considering an admitting an applicant and he stated (in order of decreasing importance) 

1. High GPA while taking many tough technical classes (math being the most prominent, but I got the vibe that courses from physics, computer science and other "difficult" majors are considered fine). He had also emphasized that most of the grades should be A or A-, but a B here and there would be ok.

2. Strong letter of recommendations from faculty members, but having one from industry should be fine (although I think he meant if its a research position that publishes papers). A point made by another professor echoes what cyberwulf has said, that perhaps the strongest reasonable letter they can write (apart from a Nash-like letter saying "this person is a genius") is comparison to either past students that went to strong schools or to current stats PhD students (although this might only apply to schools with already existent strong statistics programs)

3. A good statement of purpose that highlights why you would be a good fit for that specific program and statistics in general. Another professor advises students against writing too flowerly or over-the-top since the people reading them are busy professor members so they would rather see you get straight to the point.(Additionally if you write about how you began to love statistics as a little kid they would not believe it)

After interacting with a decent number of PhD students at Berkeley, I've also noticed that its not very common for the PhD students to have prior statistics research experiences (and a professor has confirmed for me that its usually rare to see students with prior theoretical statistics research experience), but they will have had research experience in other fields, and they usually will have a publication by the first year of PhD in their undergrad research (although this is likely factored into the letter of recommendations). 

A different professor had told me, in regards to GRE scores, that the quant section really should be easy for anybody looking to enter Berkeley (and I can confirm that the first year core classes are very mathematically intense here; the second semester theoretical statistics course makes Casella and Berger look like a walk in the park), but the verbal section does matter and they will still look at it.

Edited November 11, 2019 by icantdoalgebra