10
$\begingroup$

I am a biostatistician working in an applied field and I am responsible for writing the statistics methods section for the papers I collaborate on. In reading a lot of academic papers I've come across plenty of examples of badly written stats sections (mostly they are boring, uninformative, and lack precision, detail, and understanding of the methodology used).

Regardless of the subject matter and sophistication of the statistical methods used, what are some good examples of well written statistics sections in applied research articles?

How to define "well written" is subjective, but I would describe a statistics section as well written if it is lucid, gives (or seems to give) a full picture of how the analysis was conducted, addresses the assumptions made during the analysis, and incorporates the statistical process into the flow of the paper.

Here are some examples of papers that I think have good statistics sections:

BCG Vaccination Reduces Risk of Tuberculosis Infection in Vaccinated Badgers and Unvaccinated Badger Cubs

A Model for Predicting Mortality in Acute ST-Segment Elevation Myocardial Infarction Treated With Primary Percutaneous Coronary Intervention : Results From the Assessment of Pexelizumab in Acute Myocardial Infarction Trial

Others? Thoughts about what a "good" statistics section should include are also very welcome.

Improve this question
$\endgroup$
5
  • 1
    $\begingroup$ I understand your question's motivation. However, I see, at least, three problems with this question. The first is that it's too general (broad) and, thus, difficult/impossible to answer comprehensively. The second is that questions, containing "good", "better", "best" and similar attributes, IMHO don't make sense, if they lack clear specification of criteria for comparison. The third is that, while relative terms require criteria, absolute terms, like "good", call for highly opinionated answers, which don't fit well the StackExchange's model and, especially, Cross Validated. $\endgroup$
    – Aleksandr Blekh
    Commented Apr 21, 2015 at 16:49
  • 2
    $\begingroup$ @AleksandrBlekh You are correct about our site standards. However, this question does provide criteria; it is fairly narrowly focused IMHO; and if answers include objective support, they will not be opinionated. (I would hope that any answers that are purely opinions will be summarily deleted by high-rep community members.) Therefore I have elected not to cast a close vote. Given that there will not be one objectively best answer, though, I have also made the thread CW. Given its potential interest and importance, I have also upvoted it. $\endgroup$
    – whuber ♦
    Commented Apr 21, 2015 at 17:37
  • 2
    $\begingroup$ @whuber: I agree with your comment. Actually, mine was not aiming closing the question, but encouraging the OP in reformulating it with a minimum subjectivity focus (but, I guess, it's not easy, if at all possible). Also, I re-read the question and noticed some criteria I've missed initially. I upvoted it as well. $\endgroup$
    – Aleksandr Blekh
    Commented Apr 21, 2015 at 18:07
  • 2
    $\begingroup$ Thank you both for your comments and votes and I agree. It is a VERY broad question and I am not at all looking for a definitive "answer;" I'm interested in gathering examples of statistics sections that other applied statisticians consider thorough, informative, logical, and - ideally - elegantly written. A short piece in Nature in January pointed to (a blog post that pointed to) "well written" academic scientific writing and I am curious to see what other practitioners consider beautiful statistical writing. $\endgroup$
    – ccl
    Commented Apr 21, 2015 at 18:50
  • 1
    $\begingroup$ @ccl uninformative and lacking are certainly valid critiques of analysis sections, but boring? That's almost desirable in my book. Much of science is boring: science is not discovery, science is verification. $\endgroup$
    – AdamO
    Commented Dec 27, 2017 at 15:32

2 Answers 2

Reset to default
9
$\begingroup$

In mid 2000s, a group of medical statisticians put their heads together and issued STROBE statement (http://www.strobe-statement.org): STrengthening the Reporting of OBservational studies in Epidemiology. It was published in the same form in Lancet, PLoS Medicine, Journal of Clinical Epidemiology, and several others, which to me seems like the most amazing part of the whole exercise: putting heads together is not nearly as difficult as convincing a diverse group of editors to publish anything as is. There are various checklists based on the STROBE statement that help you define what a "well written" statistical part is.

In an unrelated area, the U.S. Institute of Education has been accumulating evidence on performance of the various educational programs in their What Works Clearinghouse. Their Procedures and Standards Handbook delineates what constitutes a solid study (by the education community standards; biostatisticians with clinical trials background find them falling quite short of what FDA requires). Spoiler alert: of the 10,000 study reports in the WWC data base, only 500 "meet WWC standards without reservations"... so when you hear somebody say about an educational product that it is "research-based", there's exactly 95% chance it's actually bogus, with research conducted by the publishers of that product without the control group.

Improve this answer
$\endgroup$
0
$\begingroup$

The following is a favorite article of mine: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2650104/

Here a very well controlled clinical trial was conducted to verify what was a commonly held belief that suppression of herpes outbreaks could reduce the transmission of HIV. It is an example of a null result. They also don't discredit their evidence because it is an enormous and well controlled trial. The design is immense, all aspects of possible confounding or bias were considered.

What I appreciate about the statistics section is its brevity, its focus on pre-specified analyses, clearly delineating primary vs. secondary hypotheses, and disclaimer for conflict of interest, describing intent-to-treat and per protocol analyses, and explaining the possible source(s) of bias.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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