# Pedictive posterior visualization for logistic(or binomial) regression

**URL:** <https://discourse.mc-stan.org/t/pedictive-posterior-visualization-for-logistic-or-binomial-regression/4350>\
**Category:** Modeling\
**Created:** [May 29, 2018, 2:08pm UTC](https://discourse.mc-stan.org/t/pedictive-posterior-visualization-for-logistic-or-binomial-regression/4350 "2018-05-29T14:08:53Z")\
**Posts on this page:** 6\
**Page:** 1

<div class="post-metadata">

**Author:** ![linas](https://yyz2.discourse-cdn.com/flex030/user_avatar/discourse.mc-stan.org/linas/32/5342_2.png) [@linas](https://discourse.mc-stan.org/u/linas)\
**Post date:** [May 29, 2018, 2:08pm UTC](https://discourse.mc-stan.org/t/pedictive-posterior-visualization-for-logistic-or-binomial-regression/4350/1 "2018-05-29T14:08:53Z")

</div>

Hi,

I am looking for a good way to visualize predictive posterior for binomial regression. In my case y[i]~binomial\_logit(N, alpha+beta_x[i]) - or in vectorized form y~binomial\_logit(N, alpha+x_beta) where x is a matrix and represents 10+ predictors for each observation, beta is a vector.

---

<div class="post-metadata">

**Author:** ![jonah](https://yyz2.discourse-cdn.com/flex030/user_avatar/discourse.mc-stan.org/jonah/32/200_2.png) [@jonah](https://discourse.mc-stan.org/u/jonah)\
**Post date:** [May 29, 2018, 4:30pm UTC](https://discourse.mc-stan.org/t/pedictive-posterior-visualization-for-logistic-or-binomial-regression/4350/2 "2018-05-29T16:30:45Z")

</div>

Is this a visualization for posterior predictive checking (i.e., comparing predictions to observed binomial counts) or for visualizing out-of-sample predictions? More specifically, what are you interested in showing in (learning from) this visualization? There’s tons of ways to visualize these things, so knowing more about your purpose would be helpful.

---

<div class="post-metadata">

**Author:** ![linas](https://yyz2.discourse-cdn.com/flex030/user_avatar/discourse.mc-stan.org/linas/32/5342_2.png) [@linas](https://discourse.mc-stan.org/u/linas)\
**Post date:** [May 29, 2018, 8:23pm UTC](https://discourse.mc-stan.org/t/pedictive-posterior-visualization-for-logistic-or-binomial-regression/4350/3 "2018-05-29T20:23:46Z")

</div>

Really i would like to learn both but immediate need is comparing predictions to observed binomial counts.

---

<div class="post-metadata">

**Author:** ![linas](https://yyz2.discourse-cdn.com/flex030/user_avatar/discourse.mc-stan.org/linas/32/5342_2.png) [@linas](https://discourse.mc-stan.org/u/linas)\
**Post date:** [May 30, 2018, 2:17pm UTC](https://discourse.mc-stan.org/t/pedictive-posterior-visualization-for-logistic-or-binomial-regression/4350/4 "2018-05-30T14:17:26Z")

</div>

Is it as simple as having in generated quantities y\_pred[i]=binomial\_logit\_rng(N, alpha+beta\*x[i]) and then taking 90% credible region of y\_pred and comparing with observed binomial counts?

---

<div class="post-metadata">

**Author:** ![jonah](https://yyz2.discourse-cdn.com/flex030/user_avatar/discourse.mc-stan.org/jonah/32/200_2.png) [@jonah](https://discourse.mc-stan.org/u/jonah)\
**Post date:** [May 30, 2018, 7:40pm UTC](https://discourse.mc-stan.org/t/pedictive-posterior-visualization-for-logistic-or-binomial-regression/4350/5 "2018-05-30T19:40:17Z")

</div>

Yeah that’s one option. The bayesplot function `ppc_intervals` makes that plot. Many of the other `ppc_` plots are also fine to use with binomial models.

You can also compare the probabilities rather than the observed counts (I think both are useful).

---

<div class="post-metadata">

**Author:** ![arthur-albuquerque](https://yyz2.discourse-cdn.com/flex030/user_avatar/discourse.mc-stan.org/arthur-albuquerque/32/11331_2.png) [@arthur-albuquerque](https://discourse.mc-stan.org/u/arthur-albuquerque)\
**Post date:** [February 25, 2022, 12:14am UTC](https://discourse.mc-stan.org/t/pedictive-posterior-visualization-for-logistic-or-binomial-regression/4350/6 "2022-02-25T00:14:06Z")

</div>

Hi Jonah, how could one compare the probabilities instead of observed counts?