# Questions about the second day 3 video

**URL:** https://isgw-forum.colorado.edu/t/questions-about-the-second-day-3-video/772
**Category:** Day 3
**Tags:** sem, openmx
**Created:** [June 3, 2026, 3:29am UTC](https://isgw-forum.colorado.edu/t/questions-about-the-second-day-3-video/772 "2026-06-03T03:29:00Z")
**Posts on this page:** 1
**Page:** 1

<div class="post-metadata">

### Author: ![kristen](https://isgw-forum.colorado.edu/letter_avatar_proxy/v4/letter/k/c4cdca/32.png) [@kristen](https://isgw-forum.colorado.edu/u/kristen)
#### Post date: [June 3, 2026, 3:29am UTC](https://isgw-forum.colorado.edu/t/questions-about-the-second-day-3-video/772/1 "2026-06-03T03:29:00Z")

</div>

I have some questions about the second Day 3 video, “Estimate heritability using data from relatives/twins”:

1. On slide 28 (roughly 19:34 in the video), it compares saturated and constrained models. How can a constrained model (m2) have a better fit (lower AIC) than a saturated model, wouldn’t a saturated model always have the best fit?
  - I realize the difference isn’t significant according to the LRT, but I’m curious how it’s possible for the constrained model to have even a slightly better fit.

2. On slide 29 (21:49 in the video) it talks about calculating maximum likelihood correlations based on the variances and covariances from fitting the model. Is there ever a reason why you would want to do this (vs. just calculating the correlations directly)? Are there times when the maximum likelihood correlations might differ from the directly-calculated correlations in a way that tells you something important about the data?
3. Question 3 about model violations has been [moved to its own thread](https://isgw-forum.colorado.edu/t/consequences-of-violating-twin-model-assumptions-gxe-assortment-etc/795)
4. Question 4 about dominance has been [moved to its own thread](https://isgw-forum.colorado.edu/t/what-does-dominance-mean-for-highly-polygenic-traits/)

Thanks for all the great videos, and sorry to ask so many questions!!