Sunday, March 17, 2013

Links


  1. Veenu Sandhu, Indulekha Aravind and Ranjita Ganesan in The Business Standard: Charity bazaar. "Indian businessmen and philanthropy have never walked hand in hand. But change is in the air."

  2. Peter Whoriskey in The Washington Post: Doubts about Johns Hopkins research have gone unanswered, scientist says. A grim story involving a paper in Nature, in which one researcher gets fired after raising questions about research in his own group, another commits suicide, and a correction is being issued.

  3. Christopher Drew and Jad Mouawad in NYTimes: Initial Tests of Battery by Boeing Fell Short.

  4. Two cool cartoons by Dan Piraro (whose comics are posted at the Bizarro Blog).

2 Comments:

  1. Rahul Siddharthan said...

    About the JHU case: quoting what I wrote elsewhere --

    ---

    The only thing that leaps out at me in the statistical analysis is the z-score cutoff of 1.5. Given that they are looking at 2-sided variations, this corresponds to a p-value of more than 0.1. It is in any case a dubious tactic to do things based on p-values or z-scores, but this is an unusually large choice -- more than 10% of your data would be "positive" just by chance. I assume this is what the WP article means by "One of the key problems, Yuan wrote to the Nature editors, was that the numerical threshold the investigators used for determining when a hit had arisen was too low. This meant they would report far more hits than there actually were."

    But then they say they verified all these using a cell viability assay. I have no idea how reliable that assay was -- but if it was reliable, I'd have no problem with the p-value, which is just an pointer then and not a prediction tool.

    ---

    Then someone pointed me to
    this reddit thread, where similar points are made.

  2. Rahul Siddharthan said...

    ps - even more relevant is the reddit AMA by the reporter, where it becomes very clear (at least in my reading) that he rushed to publication without making a serious effort to get a full picture. It is possible that Yuan is right and Boeke was pushing half-baked results without adequate proof. It is also possible that Yuan was protesting too much and had his own axe to grind. I really can't tell from the article, but it appears that the paper reports the data honestly (why else would they have a z-cutoff of 1.5, low enough to attract the attention of even casual readers?) If there was no fraud here, the WaPo have much to answer for.