<p>csh123: you seem to think a math degree is all about turning coffee into theorems. That’s a really simplistic view that is not representative of even the pure math programs there are out there (though yes, some pure math students end up having no clue about the world outside of math, those are usually the kind who go on to do PhDs in pure math).</p>
<p>Biostatisticians don’t do the labs, they help plan the experiment and analyze the data. Same with bioinfo people. They don’t handle microarrays, they develop models, software and procedures to make sense of the data (often of the large p, small n kind, many parameters but few experimental units, something not covered in undergraduate statistics, or even most graduate statistics courses unless specialized in that particular area).</p>
<p>With a math or CS major that included a wide variety of applied courses, you can do any job that requires either data analysis or programming. </p>
<p>You want another example: the epidemiology department at my current place of employment created a graduate course specifically to train students with a skill that no job candidate seems to have: analysis of large databases (i.e. massive data sets linked through different registries and what not). It’s not quite a statistics nor a computer science course, it’s somewhere in between, and the examples come from the medical field, but there’s no real medicine or biology involved. It’s really about handling and managing massive databases. No statistics course covers that, because the theoretical ones work with the data in an abstract sense, and the applied ones involve example data sets that fit in a single Excel spreadsheet (and no matter how many rows and columns there are, that’s a “small” dataset in terms of dimensions). No computer science course covers that either, because the courses relating to databases are about creating the software structures and objects, connecting them together, and getting values in and out of them. They are not about analyzing their content in a meaningful way (because that would be statistics). </p>
<p>The matter of fact is that, statistics is at the core of science, and computer science is necessary to do any modern statistical analysis. You can’t do scientific research without doing data analysis at some point. In biology and medical courses, they “blackbox” the mathematical part of data analysis, they just concentrate on the conclusions based on the outputted numbers. But researchers come up with experiments and problems whose features do not correspond to your basic statistics course assumptions. If they know what they’re doing, they’ll get a statistician involved before implementing their experiment or study. If not, they’ll come crying to a statistical consultant after their paper was rejected. Or they’ll see their research invalidated by other scientists in follow-up studies.</p>
<p>One of my colleagues works with volcano data. She doesn’t have any background in geology or vulcanology. She’s done pure math as an undergrad and statistics in her graduate studies. What’s the issue with the volcano data: it’s seismic information, measured continually, you don’t get to “save” all the data. You want to know when the next eruption is coming based on the recent and historical information, so as to make a timely alert that’s not a false alarm. The volcano guys can explain everything about the processes that go on that create eruptions and earthquakes, and measuring the various volcanic activity indicators is not a problem. The issue is making sense of the measures with the clock-ticking. That part is all mathematics and statistics and computer science, you want efficient algorithms and programs that don’t need many computations, because if it takes an hour to crunch the numbers and spit out an alert, it will be too late already. They don’t need another vulcanologist to solve that problem, they need statisticians and mathematicians and computer science people. </p>
<p>One of my statistics profs at McGill was one of the big names in neuro-imaging, particularly the analysis of fMRI data. No background in biology or medicine, and getting the fMRI data was a “solved” problem (thanks to physicists and computer scientists working with doctors and biologists). What to do with it was another matter. Massive data sets (we’re talking about 3D maps of the temperature in the brain, over time), highly correlated in space and time, with few subjects. Even getting an “average brain” is difficult, because it’s a complex structure and nobody has exactly the same shape/size of the brain. Working on that involved advanced mathematics and collaboration with neurobiologists, computer scientists, mathematicians (said prof was a friend and former classmate of a fields medalist, it really helped with solving some geometry related problems), probabilists and statisticians. The applications of analyzing data from smooth random fields (the mathematical objects necessary to make sense of fMRI) include climate and astrophysics data, just to give a couple of examples (not to mention all those other fMRI datasets one can get, not necessarily the brain).</p>
<p>So let me put it another way: how many years of posdoc does your average PhD in biology ends up doing before getting a “real job”? How many years of postdoc do PhDs in statistics do before getting a real job? I don’t have the numbers at hand, but I know postdocs have been the norm in biology for quite some time. In statistics, they’re still optional.</p>