ASA Connect

 View Only

  • 1.  Bayesian Bio Stats

    Posted 01-01-2017 18:20
    I'd be curious to hear from some Bio Stats guys on this article.

    http://www.medpagetoday.com/PrimaryCare/GeneralPrimaryCare/62291?xid=nl_mpt_DHE_2016-12-30&eun=g710605d0r&pos=0#

    Michael L. Mout, MS, Cstat, Csci
    MIKS & Assoc. - Senior Consultant/Owner
    4957 Gray Goose Ln, Ladson, SC 29456
    804-314-5147(Mbl), 843-871-3039 (Home)


  • 2.  RE: Bayesian Bio Stats

    Posted 01-02-2017 12:43
    The argument is perfectly valid, although the author just uses Bayes Theorum, which doesn't make the approach Bayesian per se.

    As usual with Bayes-based arguments, the critical issue becomes estimating the prior. Part of the reason why I support controlling for alpha level in multiple comparisons (which some have opposed) is that I believe when you do a shotgun approach with many questions, quite likely the prior probability of any specific alternative hypothesis being true is low. With well-chosen hypotheses, it's much less of an issue, and the ordinary 0.05 alpha is probably adequate.

    Ed

    Ed J. Gracely, PhD
    Associate Professor
    Family, Community, & Preventive Medicine
    College of Medicine

    Associate Professor
    Epidemiology and Biostatistics
    Dornsife School of Public Health

    Drexel University
    2900 W. Queen Lane,
    Philadelphia PA, 19129
    Tel: 215.991.8466      | Fax: 215.843.6028
    Cell: 609.707.6965
    Egracely@drexelmed.edu
    drexelmed.edu  |  drexel.edu/publichealth

    This email and any accompanying attachments are confidential. The information is intended solely for the use of the individual to whom it is addressed. Any review, disclosure, copying, distribution, or use of this email communication by others is strictly prohibited. If you are not the intended recipient, please notify the sender immediately and delete all copies. Thank you for your cooperation.




    Original Message


  • 3.  RE: Bayesian Bio Stats

    Posted 01-04-2017 14:37
    Although I generally agree with the article, I believe the use of its approach requires a certain amount of caution.

    People's prior assessments are based not just on evidence but on their values and views. For example, executives at a company developing a product may well have a rosier subjective assessment of the product's success chances than outside analysts. The article's approach depends on assuming that introducing priors will result in a more skeptical prior assessment and hence a higher, and in the author's view more realistic, hurdle needing to be met by the evidence observed in the trial.

    But the exact opposite could occur under the same theory. Rosier prior assessments tend to result in lower observational hurdles, collecting less evidence to reach conclusions. When we combine a tendency for business executives to be optimistic about the success of their own projects with a tendency to believe theories that result in reducing costs, a more subjective approach may have the unintended effect of increasing rather than decreasing the risks of failure.

    I think it's reasonable to use a Bayesian approach when internally planning the success of trials. Optimistic assumptions can be stated and perhaps subjected to challenge in such a setting. Moreover, a company is not likely to ever repeat a trial, so an estimation of its success requires subjective factors. A Bayesian approach is a more realistic model of a company's actual decision process, including its biases. At the same time, it's understandable why government, and particularly the FDA, has been wary of approaches which may be subject to manipulation or the influence of over-optimistic assumptions.

    Jonathan Siegel
    Associate Director Clinical Statistics

    Sent from my iPhone


    Original Message


  • 4.  RE: Bayesian Bio Stats

    Posted 01-05-2017 12:02

    The article recaps a well-known issue that can be mitigated by moving from .05 to .01 or less as the standard of evidence. Of course, maintaining a power level of 80% would require larger n's or more powerful designs. When searching for the Higgs boson, the CERN group used a 5-sigma rule which gives a Type I error rate of about 1 in 3.5 mil. An interesting topic in social psychology is why the medical and social sciences tolerate a 1 in 20 error rate - would you cross a street if there were a 1 in 20 chance of being run down (but only 90% of the time)? 

    ------------------------------
    Chauncey Dayton

    Original Message


  • 5.  RE: Bayesian Bio Stats

    Posted 01-06-2017 05:35

    "When searching for the Higgs boson, the CERN group used a 5-sigma rule which gives a Type I error rate of about 1 in 3.5 mil."

     

    I saw an interesting discussion about this a couple of years ago, where Tony O' Hagan wondered if the 5-sigma rule could be replaced by a more rigorous Bayesian rule.

     

    Some physicists explained the rationale behind the current approach. Basically they are looking at a probability distribution of energy values. There is a predicted distribution (from theory) and an empirical distribution that has some measurement error.

     

    If a new particle is present there will be an unexpected bump on the empirical curve and it will depart from the theoretical distribution at that point and only that point. The location and magnitude of the bump, if it's there, are not known in advance, though theory may suggest a general region of interest.

     

    The whole thing is a hot mess of multiple comparison issues and past experience with 3-sigma and 4-sigma criteria has shown they produce an unacceptable rate of things that, after a lot of subsequent effort, are shown to be false positives.

     

    The 5-sigma rule is a crude rule of thumb, not rigorously derived. The Type I error rate is unknown but likely to be well above 1 in 3.5 million.

     

     

    Blaise F Egan, CStat
    Lead Data Scientist
    Modelling and Analysis Team
    BT Innovate & Design
    ____________________________
    Office:  0331 664 5220
    Mobile: 0748 330 7421
    Email:  Blaise.Egan@BT.com

    www.benevolent.bt.com  Not a member? Please join the Fund today!

     

    This email contains BT information, which may be privileged or confidential. It's meant only for the individual(s) or entity named above. If you're not the intended recipient, note that disclosing, copying, distributing or using this information is prohibited. If you've received this email in error, please let me know immediately on the email address above. Thank you. We monitor our email system, and may record your emails.

    British Telecommunications plc
    Registered office: 81 Newgate Street London EC1A 7AJ
    Registered in England no: 1800000

     

     




    Original Message


Related Content