Supervising undergraduates in research is of course different in many ways from supervising graduate students -- e.g., expectations, scope of project, amount and type of interaction between student and advisor -- but there are also some similarities.
For example, selecting undergraduates as research assistants or advisees based on their grades in classes does not automatically lead to a fulfilling research experience for all concerned. This is a hazard of the admissions process for graduate school, but it also afflicts the selection of students for undergraduate research experiences, even when we have taught these students in our very own classes.
This is (mostly) not a rant about supposedly smart students who can't or won't do research, or who can do research but are such incredibly high maintenance that the research experience becomes a serious burden for everyone within a 5 kilometer radius of their research project. This is instead a prose poem of praise for the hard-working B students who excel at research and with whom it can be very enjoyable to work.
Some of my undergraduate research assistants are chosen after a competitive application process. Summer interns, for example, are from a highly-selective pool comprised of the top students from the top universities in the country. In this case, "top student" is typically defined as the students with the highest GPA. Most of these students are in fact excellent interns, but some of them are mediocre and some of them are abysmal.
It is fascinating in a semi-disturbing kind of way that the same situation applies for selecting students from my own institution. These are students I have met in person before I hire them as undergraduate research assistants or sign on as their advisor for a thesis or research project. In some cases these are the students who got top grades in rigorous courses taught by me and others, yet some of them, however bright, are not so smart when it comes to research.
When choosing undergrad research students from my own university, I don't always select the A students. Don't worry -- the hard-working and talented A students are not going without research experiences. In fact, every student who wants such an experience can find an advisor. But sometimes I select a B student who seems to be motivated and smart, but who just doesn't do as well in some classes as some other students.
In my experience, the success or failure of these B students at undergraduate research projects is indistinguishable from that of A students -- I have had experiences ranging from outstanding to ghastly with both -- but there are some mutual benefits to working with B students.
For the student, a successful research experience as an undergraduate may in some cases offset a modest GPA in graduate admissions efforts.
For the advisor, the B student might be easier to work with in some ways. An unscientific hypothesis some of my colleagues and I have recently discussed is that many B students might have the advantage of being less high-maintenance than some A students because they tend not to be so anxious. I think we are perhaps not the first to propose this; for example, see the cartoon in the 9 January 2009 Chronicle Review by V Hixson. In the cartoon, one professor says to another "Actually, I like the B+ students best.. bright, but still humble."
And many B students do just as well as A students in research. Reasons for this include:
Doing well in a classes, even really difficult ones, does not mean that someone has the skills necessary to do research.
Of course we don't expect that students, however stratospheric their GPA, will automatically know how to do research -- as advisors we try to teach this -- but some students learn and thrive as a project evolves, and others do not, no matter how 'smart' they are. The same is true of graduate students.
Random example: An apparently top student with A's in difficult courses worked with me on a straightforward bit of research, but it quickly became clear that he had no ability to make connections between different observations or thoughts, could not visualize phenomena, and only understood basic concepts if they were repeated to him many times. He had a great attitude about the work and eked out some results (with lots of help), but he never really understood what he was doing and never went beyond a 'problem set' kind of approach to the research.
Doing research as a student typically means you have to be willing to interact with at least one other person (the advisor) and possibly others as well (other students, other faculty, postdocs, technicians). Some people can do this well and some people can't, even with experience, no matter how high their GPA.
Random example: One of my A-student interns needed to learn and use a not-complicated technique. Others in the group were experienced with this technique, and several of us were available to answer her questions. After a day or so, A-student came to me, clearly upset, and said "I had some questions and X (an undergrad) and Y (a grad student) helped me a lot, but it is obvious that they care more about their own research than they do about mine." I replied "Well, I should hope so", a comment that shocked her. Her voice quavered as she said "But what about me?".
Heroically resisting saying something sarcastic and/or insensitive, I gave her a gentle mini-lecture about being a part of a community of researchers driven by individual curiosity and working together but also independently etc. etc. She was not happy to learn that the major focus of all our efforts and interests was not her. Alas, this realization did not change her world view, but we somehow got through the rest of the summer and she returned in the fall to her home institution and the professors who wrote her rave letters of recommendation and later admitted to me that they didn't like working with her either.
It would save a lot of time, money, stress, and grief if we could predict in advance which students would do well in a research experience, but in the absence of a reliable method of prediction, we'll just have to continue with the classic trial-and-error approach. Most of the time this works out well, especially if the students are mostly sane.