3
$\begingroup$

In a paper that I was reading Bayesian stats, they are talking about a "tight" prior.

We control the “tightness” of the Minnesota prior by adjusting the values of parameter $b1$. A tight version of the Minnesota prior is de ned by $b1 = 0.2^2$, and a loose version sets $b1 = 0.9^2$.

Is a tight prior the same thing as an "informative prior"?

Is it saying that the variance of a distribution, if we take the example of a normal distribution, is smaller?

$\endgroup$
  • 1
    $\begingroup$ They say directly in the article: "We control the “tightness” of the Minnesota prior by adjusting the values of parameter b1. A tight version of the Minnesota prior is defined by b1 = 0.2^2, and a loose version sets b1 = 0.9^2. Here the words “tight” or “loose” are used in relative terms. One can certainly argue that b1 = 0.9^2 represents a tight prior compared to the case b1 = 10^2." You can say one is more "informative" than the other, but it's equally loose and informal. $\endgroup$ – AdamO Oct 17 '17 at 18:03
7
$\begingroup$

A tight prior is indeed a concentrated prior, so a Normal(0, 1) prior for a parameter is more tight than a Normal(0, 10) prior. This is a relative statement, it depends on the scale of your parameter.

A tight prior is also more informative I would say, although informative has a connotation of being helpful, while a tight prior has less of an opinion, it's just tight.

But sure, if you have a more informed prior opinion of the variable, you would express that with a tighter prior.

| cite | improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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