# Epsilon Shrinkage NaN

**URL:** <https://discourse.pumas.ai/t/epsilon-shrinkage-nan/925>\
**Category:** Estimation\
**Created:** [September 7, 2023, 4:33pm UTC](https://discourse.pumas.ai/t/epsilon-shrinkage-nan/925 "2023-09-07T16:33:59Z")\
**Posts on this page:** 6\
**Page:** 1

<div class="post-metadata">

**Author:** ![adunn34](https://yyz2.discourse-cdn.com/flex030/user_avatar/discourse.pumas.ai/adunn34/32/204_2.png) [@adunn34](https://discourse.pumas.ai/u/adunn34)\
**Post date:** [September 7, 2023, 4:33pm UTC](https://discourse.pumas.ai/t/epsilon-shrinkage-nan/925/1 "2023-09-07T16:33:59Z")

</div>

Hello! I’m getting an epsilon shrinkage of NaN and I’m not sure how to best rectify this. The model is a simple one compartment with first order elimination & absorption and the data I’m fitting includes includes 3 robust Ph1 studies as well as some sparse Ph2 data. Of note, when I fit only the robust data I do get an actual value for epsilon shrinkage. Any ideas would be greatly appreciated.

---

<div class="post-metadata">

**Author:** ![andreasnoack](https://yyz2.discourse-cdn.com/flex030/user_avatar/discourse.pumas.ai/andreasnoack/32/22_2.png) [@andreasnoack](https://discourse.pumas.ai/u/andreasnoack)\
**Post date:** [September 8, 2023, 12:59pm UTC](https://discourse.pumas.ai/t/epsilon-shrinkage-nan/925/2 "2023-09-08T12:59:11Z")

</div>

Do you have any subjects with just a single observation?

---

<div class="post-metadata">

**Author:** ![adunn34](https://yyz2.discourse-cdn.com/flex030/user_avatar/discourse.pumas.ai/adunn34/32/204_2.png) [@adunn34](https://discourse.pumas.ai/u/adunn34)\
**Post date:** [September 8, 2023, 4:18pm UTC](https://discourse.pumas.ai/t/epsilon-shrinkage-nan/925/3 "2023-09-08T16:18:56Z")

</div>

Yes I have 3 subjects who only have 1 observation. I just tried fitting with those 3 subjects removed and it does still result in an epsilon shrinkage of NaN. I will say the only reason I noticed was because I can’t use the GOF plot functions when epsilon shrinkage is NaN. It doesn’t seem to impact anything else (that I can tell).

---

<div class="post-metadata">

**Author:** ![adunn34](https://yyz2.discourse-cdn.com/flex030/user_avatar/discourse.pumas.ai/adunn34/32/204_2.png) [@adunn34](https://discourse.pumas.ai/u/adunn34)\
**Post date:** [September 8, 2023, 4:23pm UTC](https://discourse.pumas.ai/t/epsilon-shrinkage-nan/925/4 "2023-09-08T16:23:30Z")

</div>

I’m not sure if this is helpful - but when I remove the study with only steady state data (where I have to use the II,SS,ADDL inputs) the problem resolves.

---

<div class="post-metadata">

**Author:** ![andreasnoack](https://yyz2.discourse-cdn.com/flex030/user_avatar/discourse.pumas.ai/andreasnoack/32/22_2.png) [@andreasnoack](https://discourse.pumas.ai/u/andreasnoack)\
**Post date:** [September 10, 2023, 7:37pm UTC](https://discourse.pumas.ai/t/epsilon-shrinkage-nan/925/5 "2023-09-10T19:37:18Z")

</div>

In the original fit, were any of the individual (weighted) residuals `NaN` or `Inf`? If not, could you please try `wresiduals(yourfit, LaplaceI())` and see if there are any `NaN`s or `Inf`s?

---

<div class="post-metadata">

**Author:** ![adunn34](https://yyz2.discourse-cdn.com/flex030/user_avatar/discourse.pumas.ai/adunn34/32/204_2.png) [@adunn34](https://discourse.pumas.ai/u/adunn34)\
**Post date:** [September 11, 2023, 2:06pm UTC](https://discourse.pumas.ai/t/epsilon-shrinkage-nan/925/6 "2023-09-11T14:06:04Z")

</div>

Just checked for both the original fit and the above suggested. The only instances of NaN occur at dosing records (EVID==1) for both scenarios. Otherwise, there are no instances of NaN or Inf .