“… any statis…

“… any statistical procedure should possess the following desirable features:
(1) It should have a reasonably good (optimal or nearly optimal) efficiency at the assumed model
(2) It should be robust in the sense that small deviations from the model as- sumptions should impair the performance only slightly….
(3) Somewhat larger deviations from the model should not cause a catastrophe.”

A quote from Huber as quoted in Casella & Berger.

I love the use of catastrophe in this quote.

Huber, P. J. (1981). Robust Statistics. New York: Wiley.

Advertisements

“This is the us…

“This is the usual difference between a probability definition and a statistics definition. The probability definition deals with one probability structure, but the statistics definition deals with an entire family.” – Casella & Berger p. 468, section 10.1 #C&B

Casella, George, and Roger L. Berger. Statistical inference. Vol. 70. Belmont, CA: Duxbury Press, 1990.

Comp Ravings Part 3

Comps, you are a cruel mistress.  In some ways it’s easy to take the hours of studying, the shared misery/complaining of how “unfair” comps are and such with other students.  It helps prepare you to go through something you really don’t want to do but you need to do for a higher goal.  I see this as something in academia that you will need to do in the future (i.e. grants).  But one thing that just blows all that out of the water? Really nice weather and holiday weekends.  

Today is Memorial Day.  Am I spending it remembering how my grandfathers and other veterans fought in wars?  No, but I cannot earnestly say I’d be doing that if I weren’t studying.  As some who know me, I love the beach.  The sand, the spray, friends from home, the sun – it doesn’t get much better.  So getting messages about all those lovely things while being shackled and caged at a desk is insufferable at times. 

So what can you do?  

1) Request comps to be in the winter.  This is not really a feasible option as most programs start in September and you know almost nothing at that time and if you waited 6 months after your classes it’d be 10x worse when it came to studying.  But I figured I’d suggest it because hey you’re already miserable during the sunless weather calamity depression that is winter. 

2) Study outside.  Most places have great places to eat for lunch, and you can set up shop there.  If you’re not doing computer work, that’s more the better.  I’m usually very pro-electronic only, but if it gets me outside, I’m going to that printer and taking down some trees.  Things to look for – places with outside seating that isn’t very windy, outlets if you need juice, mobile (read outside) blackboards – believe me, they exist.  But be forewarned, you may look like you’re John Nash in A Beautiful Mind.  

3) Find a view.  Some places (top floors of buildings are good).  Trust me – when it’s sunny – this  is where I want to be.Image

 

4)  Make Food Fun!  No – I’m not talking about Pinterest meals.  I’m saying you have to eat, so try to meet up with the friends/significant others/pets you’ve been neglecting lately.

5) Apparently writing blog posts gets out some thoughts and is a short pleasant distraction.

Biostatistics Comprehensive Exams (Comps)

Have a limited amount of time to cram in a whole year’s (or two’s) of information into your brain before being tested on a large field of topics?  If yes, you may be taking comps.

These should apply to both Master’s and PhD level exams.

Quick tips:

1) Small working groups.  Over 5 is way too many.  Sometimes 1 (just you) is just fine.  

2) Planning helps.  If you plan to working with people – try to plan as much as you can to not waste time.

3) Past exams – may departments have previous exams that you can either access (from past students) or from the department itself.   Also, other departments are kind enough to release theirs online (http://www.vanderbilt.edu/biostatistics/graduate/comprehensive-exams/comprehensive-examinations/)  (http://www.biostat.washington.edu/node/1285) (http://www.math.unm.edu/stat/exams/index.php) are some examples (Thanks Vanderbilt website).

4) Sooner rather than later.  This is the most “duh” tip and the most that is usually overlooked/not possible.  Post spring break (meeting ~ 1 / week) is a good starter to get yourself in review mode.  Finals cause problems with having everything due and comp study suffers.  It happens to everyone, just punch it when you have the freedom to do so.

5) Do at least one timed session.  It’s like you’ll run out of time or misappropriate time on the exam if you don’t

6) Drop anything else.  What’s worse than trying to cram 1 year into a short time?  Doing that and not passing.

Comprehensive Exams

So we embark on another year where finals are over and comprehensive exams are coming online.  Casella & Berger, Billingsley, and Chung will be names I mention as much as any friend in the next 2 weeks.

Here are some pointers that I hope serve me well:

1) Communication: Most important person to discuss your responsibilities while studying/prepping is your advisor.  They need to know not to expect much from you other than frazzled emails that have minimal to no progress on projects.  Objective #1 in your first year is passing comprehensive exams.  #2 is research, but you should be able to sacrifice 2 weeks of no work for harder/more productive work 2 weeks after the exams are over.

So communicate this to collaborators as well.   If they don’t understand that, you may have to weigh some other options because if they are unrelenting during this time, re-evaluate connections.

Tell family and friends that you’re going to be crazy/boring/stressed/a hermit.  You can still be totally social during this time.  But just minimally.

2) Productive distractions: As from 1) above, research is at an all time low.  Try to make sure if you’re wasting time it has 1 of 3 properties 1) they advance you in another way (blog, read a paper, read a blog) 2) are well time-defined so that there are clear cut endpoints (1 episode of BLANK TV show is at least over in 20min-1hr) 3) Scheduled – they become very good motivators when you’re cruising.

3) Figure out which other students can spitball the same way you do.  Not all students work together the same way (or not at all).

4) Treat this as a short-term project.  Do what you normally do for projects: define goals, define timelines, have some re-evaluation midway to see if the process is working, iterate.

5) Find some good tunes.  They help.  Daft Punk just came out with “Random Access Memories”.  Fitting name for this post as well.

fMRI Researchers need to read this article (if they haven’t already)

I have recently read “Guidelines for reporting an fMRI study” (http://www.sciencedirect.com/science/article/pii/S1053811907011020), from NeuroImage 2008 by Russell A. Poldrack, Paul C. Fletcher, Richard N. Henson, Keith J. Worsley, Matthew Brett, and Thomas E. Nichols.  and I have to say the article is very comprehensive.

Many of the discussion points they bring up are becoming much more commonplace in fMRI but not fast enough.  I was also surprised to see that it was cited only 97 times (as of May 7, 2013 on Google Scholar).  Maybe there are many other guideline papers out there, but it seems as though this one would be a good practical one to cite.  I thought that maybe it would be because people didn’t know where to put it (and as such leave it out).  Something like a “We completed this manuscript using the defined checklist of PAPER” for example to cite it may be a bit awkward, but I think it could be used somewhere.

What I’m more surprised about is how this hasn’t been adopted by editors as a first pass look at a methods section or something that NeuroImage makes standard.  The authors do grant that adoption of these methods will lengthen manuscripts (and somewhat sizably depending on inclusion of information), but it is not unheard of having large supplements (for some papers) already.  And let’s face it, webpages are cheap to store information, especially given the fact some neuroimaging papers attach some form of data, even in the form of highly detailed figures (which immediately negates large supplements of texts, tables, and flowcharts of process).

Again – I don’t know how much it’s been adopted, just that I think it should.  Maybe put a nice badge on the paper saying “Hey – it’s as much info as you can get” about the process the authors used, but at least that’d be saying something other than “the reviewers thought you could maybe attempt (and fail) at trying to reproduce this”.   It’s not that authors don’t want people to reproduce their work (they don’t want people NOT reproducing it), but they’re not forced to have that level of scrutiny in the neuroimaging field, and I like the fact that this paper wants that to change.

Given the publication, more people have been releasing their data, with the helps of the 1000 connectomes project, NITRC, ABIDE, ADHD-200, and people like Michael Milham pushing people to share.  One thing of note though, most the data sharing has been from resting-state fMRI and not as much task-based (as the article is particularly describing), which I think should (and is likely to) change in the near future.

I guess all I want is more “zombie” editors in the sense they demand “give me your brains”!