# Using 'strata' as a feature 

**URL:** <https://community.drivendata.org/t/using-strata-as-a-feature/11315>\
**Category:** Poverty Prediction Challenge\
**Created:** [January 20, 2026, 8:12pm UTC](https://community.drivendata.org/t/using-strata-as-a-feature/11315 "2026-01-20T20:12:47Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![adam.przychodni](https://avatars.discourse-cdn.com/v4/letter/a/8c91f0/32.png) [@adam.przychodni](https://community.drivendata.org/u/adam.przychodni)\
**Post date:** [January 20, 2026, 8:12pm UTC](https://community.drivendata.org/t/using-strata-as-a-feature/11315/1 "2026-01-20T20:12:47Z")

</div>

Hi 🙂

First of all, thank you so much for organizing this challenge! It’s a fantastic initiative, and I really appreciate the work you’re doing to advance poverty measurement tools.

I have a quick question regarding the **`strata`** variable. Given the focus on S2S robustness and the limited number of surveys in the test set, should we be using `strata` as a **predictive feature**?

My concern is whether the model should learn to generalise to new, unseen strata in the private set, or if it’s intended to capture local effects from the existing IDs.

Thanks again for all your hard work on this!

Best regards,

Adam