# Using discrete ordinal predictors

**URL:** <https://discourse.pymc.io/t/using-discrete-ordinal-predictors/4524>\
**Category:** Questions\
**Created:** [February 18, 2020, 6:27pm UTC](https://discourse.pymc.io/t/using-discrete-ordinal-predictors/4524 "2020-02-18T18:27:11Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![JaredStufft](https://yyz2.discourse-cdn.com/flex036/user_avatar/discourse.pymc.io/jaredstufft/32/2481_2.png) [@JaredStufft](https://discourse.pymc.io/u/JaredStufft)\
**Post date:** [February 18, 2020, 6:27pm UTC](https://discourse.pymc.io/t/using-discrete-ordinal-predictors/4524/1 "2020-02-18T18:27:11Z")

</div>

I have some survey data that includes a Likert score as a response. I’d like to relate this likert score to some continuous outcome of interest. What’s the best way to include discrete ordinal features as predictors?

I understand using them as outcomes via `OrderedLogistic()` likelihood functions. I also know I can include them as dummy variables and just use a standard linear regression, but this doesn’t take into account the ordinality.

TL;DR what’s the best way to use a likert/discrete ordinal feature as a predictor of a continuous outcome in PyMC3?

EDIT:  
Thought about this a little more. Thinking about basically doing 2 models - one will model the Likert value itself as an outcome variable using the Ordered Logistic likelihood and update a set of coefficients for the other covariates. The goal of this is to get a model to predict the latent distribution behind the Likert. Then, use the predicted latent distribution as the predictor for my original question rather than the actual category itself. Any thoughts?