0
$\begingroup$

I have a question regarding bayesian statistics. Is it possible to end up with a posterior probability of 1, that a slope is positive?

My likelihood data shows a greatly significant relationship, with a p-value of 0.000715. But does it make sense that I end up with a posterior probability of 1?

And do you know a smart way to calculate the posterior probability of a positive slope?

Improve this question
$\endgroup$
12
  • $\begingroup$ Exactly how did you compute this posterior probability? Was it based, perhaps, on a Monte-Carlo method or is the calculation an exact analytical one? $\endgroup$
    – whuber ♦
    Feb 27, 2016 at 17:59
  • $\begingroup$ And I suppose we are to assume that this p-value corresponds to a positive point estimate? $\endgroup$
    – conjugateprior
    Feb 27, 2016 at 18:01
  • 1
    $\begingroup$ That means you have an estimated posterior probability of $1$. As in all simulations, this value is uncertain. Unless the prior assigned zero probability to negative slopes, mathematically there must be some positive posterior probability of a negative slope--but your simulation just hasn't gone on long enough to detect that. $\endgroup$
    – whuber ♦
    Feb 27, 2016 at 18:13
  • 1
    $\begingroup$ You ended up with the result that, after taking several thousands of samples from the joint posterior, the number of times that particular coefficient was non-positive was 0? $\endgroup$
    – conjugateprior
    Feb 27, 2016 at 18:13
  • 4
    $\begingroup$ You may run 10⁶ or 10⁹ iterations and still find a probability of one and still be unable to decide whether or not this estimate equal to 1 corresponds to a posterior probability of exactly 1. Simulation cannot help you with this question. $\endgroup$
    – Xi'an
    Feb 27, 2016 at 21:01

1 Answer 1

Reset to default
1
$\begingroup$

(I'm assuming that you're doing some sort of linear regression.) The only way you could end up with a posterior probability of 1 that the slope is positive is if either

1) the prior is zero for non-positive slopes, or 2) the likelihood is zero for non-positive slopes.

That's a straightforward consequence of Bayes' Rule.

Improve this answer
$\endgroup$

Your Answer

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

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