7
$\begingroup$

I'm a last year medical student, as we say "Intern doctor". In the future I want to do research on the issue that I want to get in.

Therefore I want to learn mathematics, statistics, R programming and so on, because I want to do best statistical analysis for my research. Proper statistical analysis is the most important part after the research issue as far as I'm concerned.

Consequently, for these things I watch on Youtube the Richard McElreath statistical rethinking videos and I do learning R programming on DataCamp courses, I bought annual premium plan. As far as I see the Youtube videos that I watch are somewhat hard for me to understand. Without statistical substructure I think to learn R programming on DataCamp would be on shaky ground.

What would you think about that? I want to learn the ideas, advices from more experienced people than me. What would you advice for me to learn these skills with very well?

$\endgroup$
6
  • 1
    $\begingroup$ Welcome to Cross Validated and the world of statistics! You might consider pursuing a master's in statistics or biostatistics. Less intense, you might consult with the head of the statistics or biostatistics department at your school. $\endgroup$
    – Dave
    Jan 12 at 21:04
  • 24
    $\begingroup$ Have you considered focusing on your medical skills and collaborating with a statistician for your research? Together, you and the statistician will have far more research and statistical capabilities than you could hope to achieve by yourself in any short or medium period of time. $\endgroup$
    – whuber
    Jan 12 at 21:07
  • $\begingroup$ I’d start with fun part, programming, data science etc. you can always catch up later with mundane theoretical layers $\endgroup$
    – Aksakal
    Jan 12 at 21:07
  • $\begingroup$ I have a master degree in biostatistics at Hasselt University in Belgium. I think a biostatistics program will help you as it teaches you statistics with applications mainly in medical/clinical context. For statistical software, R and SAS are used during the whole time and after two years, you can use them for your statistical tasks. $\endgroup$
    – TrungDung
    Jan 12 at 21:32
  • $\begingroup$ Would you mind including how much statistics you know at present? Did you take any classes in undergrad? How familiar with mathematics are you? $\endgroup$ Jan 12 at 23:26

5 Answers 5

14
$\begingroup$

What would you think about that?

This is a good question. I've spent a good amount of time during my Ph.D in Biostatistics consulting for academic physicians and their research. If you (and moderators) will allow for an opinion based answer then I'm happy to give it.

Medicine for some reason has created a culture in which the physician is intended to do everything themselves. Study design, data collection, analysis, writing, oh yea and on top of that their clinical duties and learning more about their specialty. These include responsibilities of an epidemiologist, data architect, statistician, just to name a few. Personally, I think that is a ridiculous onus to put on a researcher. This also might explain why medical research seems to be a copy-paste affair with bad statistics. Statistics is hard to learn, medicine is hard to learn, so learning both tends to mean taking shortcuts on one or the other or both (and understandably, it is the statistical rigour that is sacrificed).

Rather than succumb to these expectations it might be wiser to, as whuber notes, befriend a biostatsitician. Collaboration is a good way to learn, because you get consistent advice tailored to your specific situation as opposed to a mish mash of approaches from different courses with different learning goals. I'm not saying to defer all statistical work to a statistician, nor am I saying you should not learn about statistics independently, but I think rushing to learn all these things while also being a physician will lead to poorer work than if you were patient and collaborative.

The question is then "How do I meet/befriend a biostatsitician". Your medical school is likely attached to a university, in which there may or may not be an epidemiology department. Epidemiologists focus very carefully on how to do quality studies in a medical setting. THey should be well versed enough in statistics to help you out with design, data collection, and analysis. If you don't have an epidemiology department, there may be someone in a stats/math department, or in the sociology department (sociology is not exactly like biostatistics, but the difference between an epidemiologist and a sociologist grows smaller and smaller).

EDIT:

EdM makes a good point about the basis of fundamental probability and statistics. I'm not prepared to give a list of topics to learn and places to learn them. I think any undergraduate curriculum in science can give you enough to get started.

That being said, if pressed to offer one resource on a basis of prob and stats, I would recommend Introduction to Medical Statistics by Martin Bland. The book is geared towards medical students and in the introduction states

This book is intended as an introduction to some of the statistical ideas important to medicine. It does not tell you all you need to know to do medical research. Once you have understood the concepts discussed here, it is much easier to learn about the techniques of study design and statistical analysis required to answer any particular question.

The book however does not cover probability, and so you're free to pick up most introductory texts on the matter to cover that base. I agree with Bland that this book should serve as a good basis to read academic medical literature critically, and should serve as an excellent jumping off point to learn more about statistics in medicine.

$\endgroup$
11
  • 9
    $\begingroup$ (+1) Totally agree with this. Imagine a statistician trying to learn how to treat patients medically by taking online classes and watching YouTube videos. Of course there is no problem learning things outside your formal training, but to do the thing rigorously requires extensive training. It's not just statisticians gatekeeping statistics; the universe is gatekeeping statistics by making its analysis so challenging to do correctly. $\endgroup$
    – Noah
    Jan 12 at 21:56
  • $\begingroup$ Yes, excellent point @Noah. OP should browse this site to see just how much disagreement exists between people who make statistics their job! Confusion is abundant and having a close collaborator to help guide you through it is a much better option. $\endgroup$ Jan 12 at 21:57
  • 3
    $\begingroup$ I agree (+1), but the physician should learn enough basic probability and statistics to have productive conversations with the befriended biostatistician. A biostatistician might not be able to treat patients, but she presumably knows enough about basic biomedical issues to pick up on what's critical to statistical evaluation. What is the corresponding basis in fundamental probability and statistics that one might hope all academic physicians to have, and how might that be reasonably acquired? $\endgroup$
    – EdM
    Jan 12 at 22:45
  • 1
    $\begingroup$ @EdM That's a great point about the utility of physicians acquiring sufficient background to collaborate. Good medical schools are now trying to include that much instruction in their required programs, but their success rate might be spotty and other medical schools still need to catch up. $\endgroup$
    – whuber
    Jan 13 at 0:02
  • 1
    $\begingroup$ @NickCox: I see your point, but I would rather see this as a job more akin to lab technicians and what is in the German academic system called Mittelbau - people who do an honest job, are of course not at the top of the academic dunghill, but don't have to spend their weekends writing grants. I would assume there would be people interested in this. But I agree that the organization of universities is not, ah, conducive to this kind of setup. Which I would argue is a loss all around, but that's the way it is. $\endgroup$ Jan 13 at 13:47
6
$\begingroup$

I spent the last 14 years trying to get better at statistics including earning an applied master's in it (which in honesty was disappointing). I think the answer is that there is no simple way to do this. This is made worse by the fact that much of the academic literature that uses statistics is not done by true statisticians IMHO (which of course is true of me) and they do things (lots of things) that violate various rules because they are not true experts. Simple examples include applying a method wrong, ignoring threats to external validity, using stepwise regression, failing to address omitted variable bias, dropping variables from the model because they are not statistically significant and rerunning the model with the same data and on and on. One frightening article years ago looking at articles in the New England Journal of Medicine and Lancet found a wide range of statistical errors in articles applying logistic regression, and they are of course elite journals. And this ignores the fact that statisticians often disagree themselves.

So if you are going to learn statistics, its a lot of work. You might look at the UCLA (Idre) site or for time series the Duke University site as a starter.

$\endgroup$
5
$\begingroup$

The discussions above make fantastic points. I recommend a parallel approach of finding a biostatistician collaborator and learning about biostatistics. On the latter, concentrate on learning things that the biostatistician is unlikely to already know, to spur them on to a better understanding. I've tried to cover lots of things in BBR that are in this category---things that biostatisticians and clinical researchers have to unlearn in order to make progress. As just one example, it is commonly accepted that computing change from baseline is OK in analyzing patient outcomes. BBR goes to great lengths to show why you should not compute change from baseline. You'll see lots of discussions about learning medical statistics, and collaborating, at datamethods.org.

$\endgroup$
4
$\begingroup$

You're Getting The Order Wrong.

It's as though you've said "I want to learn medicine, therefor I bought a scalpel." R is a very useful tool. But it will be much harder to learn statistics from using R than it would to learn to use R after already knowing statistics.

If I were learning statistics from scratch, I would not start with Baysian. I would start really simple. Use physical objects to understand how possible outcomes work first (like dice and coins). If you flip a coin, how many outcomes are there? Can you list them out? How about two coins, etc. Getting a solid understanding of the set of possible outcomes is the best basis for thinking about how "likely" something is to fall into a given subset of those outcomes.

Conveniently, this is the way that most introductory statistics classes are taught. As such, I really recommend that you start learning statistics with an intro class.

Following that, probably the next most important skill is formal logic. That's likely to be filed under philosophy at your school or online. You will essentially never be able to actually "prove" anything with research in a strict, formal sense. Example: with the logic chain A -> B -> C, maybe you test B->C a million times, and get the desired result every time. That demonstration doesn't "prove" that B->C. But practically, we would probably accept research that asserted B->C. What you need there is the grounding and discipline to understand and clearly state your full logic chain and all your assumptions. Doing so will make the analysis easier, and will make your results more robust.

Once you have that, you can look at Baysian approaches much more usefully. I find it's easiest to think of Bayes as dealing with some epistemological problems with frequentist statistics, which means it wont make much sense unless you already understand the basics of statistics and of the logic chains. (Some people have claimed success with learning Baysian from the get-go. I have trouble even imagining how one could usefully do that. YMMV.)

Unfortunately, while you can get a solid working knowledge of applied statistics that way, it will be exceedingly difficult to get a true understanding without calculus, and fairly difficult calculus at that. So if you really want to dive deep, this is about the point where you would want to brush up on that. Then look for a calculus-based statistics. It might be called "Statistics for scientists and engineers" or similar.

$\endgroup$
0
2
$\begingroup$

It sounds like that you just have started with statistics and you do not have experience with it. That is totally OK. But your goal is the other extrem that you want to be a statistic pro. That is impossible. Relax your self and do not push you to much here.

Keep one very important thing in mind. A professional (or however you want to name that) scientist never work alone. Research is a team thing. The workload itself can be done by one person. But the thinking, developing and dropping of ideas and questions and arguing with research colleagues and other experts is always a team thing.

And that will bring you back to statistics. Even well experience research do not make there statistics alone. They make a plan how to collect and how to analyze the data. But when they finalized that plan they do not start to collect data but they contact a statistic person. They do statistic consulting (translated from German).

Because of that process you learn how to do things especially where are your borders of knowledge and expertise and when to ask other persons.

In short: You do not have to work alone. Your supervisor and your university should be able to offer you something here.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.