Spotlight on: Beta distribution

แชร์
ฝัง
  • เผยแพร่เมื่อ 12 พ.ย. 2024

ความคิดเห็น • 7

  • @mohamadhs4564
    @mohamadhs4564 3 ปีที่แล้ว +3

    Thanks so much. Please produce some videos about discrete event simulation.
    Also, I'm not sure how to use the hazard ratio in a Markov model or how to convert it to transition probability.

    • @harshotsai
      @harshotsai 3 ปีที่แล้ว +1

      Yeah, use of Hazard ratios in MM would be great to know. Thanks!!!

    • @TMSnowsill
      @TMSnowsill  3 ปีที่แล้ว +1

      Great suggestions. With Discrete Event Simulation were you interested in using them as an alternative to Markov models (as they usually are in cost effectiveness analyses) or did you want to include things like interactions and competition for resources (operational research-type studies)?

    • @mohamadhs4564
      @mohamadhs4564 3 ปีที่แล้ว +1

      @@TMSnowsill Actually, I want to use it for cost-effectiveness.
      As you know the Markov model has many limitations and DES can overcome many of the Markovian limitations.
      Thank you for your attention.

    • @TMSnowsill
      @TMSnowsill  3 ปีที่แล้ว +2

      Fantastic. Yes I do hope to do a video series on DES as an alternative to Markov modelling.

  • @KaikiTsuki49
    @KaikiTsuki49 2 ปีที่แล้ว

    Thank you so much! Can you elaborate a little on the "c" value? When would you use c=0 versus the Jeffrey's prior, and what function does the prior serve?

    • @TMSnowsill
      @TMSnowsill  2 ปีที่แล้ว

      Hi there. There is perhaps some helpful detail available at en.wikipedia.org/wiki/Beta_distribution#Bayesian_inference
      If you have a good number of successes and failures in your data it usually doesn't make a big difference whether you use c=0 or c=1/2. In that case I tend to use c=0 because it doesn't make a difference and there's less requirement to explain what a Jeffreys prior is.
      If you have a very limited number of successes and/or failures then it can make a big difference - in the extreme case you can end up with an improper distribution which does not reflect uncertainty appropriately. For example, if you had observed 4 successes and 0 failures, you might choose Beta(4, 0), i.e., use c=0, but this has mean 1 and variance 0, i.e., no possibility of the parameter being anything other than 1.
      Most textbooks on Bayesian inference will give you some information about prior distributions, but you can also check out en.wikipedia.org/wiki/Prior_probability for a quick summary.