Contact Info

Upcoming Talks

  • No upcoming talks
Past Schedule

Project Links

Should You Email a Professor During the Ph.D. Admissions Season?


  • If a professor advertises that they are looking for new Ph.D. students, does that increase the number of emails that they receive?
  • If a person emails me to let them know that they want to be my student, does that improve their chances to getting admitted?

Updated March 2017: Given the current uncertainity of the US government's policy for granting visas for international students, academics and researchers, I have removed my tally of emails by country. I do not want people to (incorrectly) think that my writings are tacit support for the supposed "travel ban" based on someone's country of origin.

Another Ph.D. admissions season has past. I reviewed probably about 150 applications this time around. Some applicants heeded my warning about not starting off with an anecdote that talks about their precocious childhood. Others did not. CMU's CS department PhD program acceptance rate this year was somewhere around 11%. Note that this does not include the other departments in CMU's School of Computer Science — I know that the Machine Learning Department's acceptance rate is even lower.

One aspect of admissions that is not often discussed are emails from the prospective students to faculty members. These emails are from people who are applying to the Ph.D. program and hope to curry favor or make sure that the professor looks at their application materials. The formula of these emails is almost always the same. The first paragraph says who they are, the second says what kind of research they are working on now, and then third paragraph says something about why my research matches up with their interests and why they think they are the ideal candidate. They also typically attach their CV. Occasionally they attach a photo.

In my conversations with other professors about these emails, the consensus seems to be that most just ignore them. Christos takes the extreme measure of using an auto-response message for all emails, including ones with addresses. I have a filter to automatically delete these messages from him. But he does put his phone number in that email and encourages people that really want to get in touch with him to text him. Nobody ever does.

Maybe it's because I am still young and haven't gotten anybody pregnant yet, but I actually try to respond to these types of emails if it looks like they really tried to target towards me. Unfortunately most of them are from people that are misguided and are clearly not qualified to be my student. Many of them have zero database systems research experience. My responses are almost always bad news for them. I tell them that they are not going to be my student. Most people do not respond. Some come back and plead for me to reconsider. It's never pretty. I always tell them that they should still apply if they really want to come to CMU because there may be another professor that is interested.

I also seem to get a lot of emails from people that work on graph processing projects and somehow think that's what I want to do too. One kid emailed me with a long proposal about how he wants to build a graph database system for his PhD. I don't find this area all that interesting and Jignesh Patel's research has shown that using a relational DBMS for graph processing outperforms specialized systems.

New Students Advertisement A/B Testing

I have been running a simple A/B experiment for the last two years on my homepage in regard to whether I was taking new students. For 2014, I had the following advertisement at the top my homepage to say that I was looking for new students. Then for 2015, I removed this tagline and didn't say anything about whether I was looking for new students on Twitter. I then measured whether having this advertisement affected the number of emails that I received.

Note that in 2014 that I was looking for applicants with a very specific skill. More on that below.

I realize that a true A/B test is supposed to show people different alternatives at the same time during a shorter period of time and measure whether the user takes a specific action. But in this case I have no idea of knowing whether somebody saw my website before they emailed me.

I then kept track of the all emails that I received during these two years. I wanted to determine whether advertising for new students made a difference in terms of the number of applications I received and the quality of the potential student.

For tracking the number of emails, I set my date boundaries from April to February. CMU's acceptance notifications go out at the end of January. The rejection emails are sent later, so I still got some emails from people that didn't know that they had been rejected yet. As you can see from the graph below, I got a total of 35 emails when I had the advertisement and only 16 when I removed it. I got the most emails in November for both years; this makes sense because that's when people normally submit their applications.

We can also compare the number of emails with the number of people that applied to CMU that specifically listed me as a potential advisor in their application materials. I also include the count from the first year that I joined CMU as another point of reference:

  • 2013: 21
  • 2014: 35
  • 2015: 40

So this means that for 2015 I got 54% fewer emails than in 2014 when I removed the advertisement. But I got 14% more people applying to be my student. I attribute this to two possible reasons. The first is that I am publishing more and my dank DB research skills are becoming more well known throughout the land. The second is that with fewer people emailing me and getting a response to why they are not a good fit for my lab, they didn't already know not to select me as potential advisor.

Should You Email a Professor?

If you plan on applying to CMU's Ph.D. program you may be wondering whether you should email me or another faculty member to improve your chances. I can't speak for other professors but the answer for me is "no". Of the 50 students that have emailed me over the last two years, only two were admitted. One was in 2014 and one was in 2015. For the one in 2015, I was not interested in them at all but they were admitted by other faculty members. This is why I always tell students that they should still apply.

Now for the admitted student that emailed me in 2014, they were already going to be accepted when they contacted me. They were writing to let me know that they did turntables and databases! This is an amazing combination of skills and that is why I currently co-advise Prashanth Menon (aka "DJ Two-Phase Locking") with Todd Mowry. Prashanth is trill as fuck. You need to be as good as him if you want to be my future student.

People don't realize what they are doing when they email a professor asking to be their student. You are essentially asking for somebody to give you $500,000. A Ph.D. in computer science takes roughly 5-6 years. If you become my student then that means I need to cover your tuition, stipend, health care, and studio time. Each student costs me $90k a year (before overhead). Yes, this is a lot. Yes, this makes it hard to do the kinds of things that I want to do. I've heard that CMU is tied with MIT for the most expensive Ph.D. students in the US.

So now think about this when you email me asking to be my student but you have no publications in a top-tier database research conference (SIGMOD, VLDB), no relevant industry experience, have never taken a database course, and are not majoring in computer science. Do you sound like you are worth $500,000 from my perspective? I don't care that you think you are a fast learner. I don't care that you are going to take a MOOC on databases over the summer before you start grad school to fill in your background gaps. I don't care that you want to come to CMU and build a graph DBMS. I don't care that you are "passionate" about "big data".

I'm sorry if I sound heartless, but it's reality. Just like not everyone can run a trap well, not everyone is cut out to do a Ph.D in databases.