# ANCOVA that handles pseudoreplication?

**URL:** <https://discourse.datamethods.org/t/ancova-that-handles-pseudoreplication/6386>\
**Category:** modeling strategy\
**Tags:** statistical-model\
**Created:** [January 25, 2023, 4:17pm UTC](https://discourse.datamethods.org/t/ancova-that-handles-pseudoreplication/6386 "2023-01-25T16:17:30Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![cmagelssen](https://discourse.datamethods.org/user_avatar/discourse.datamethods.org/cmagelssen/32/3561_2.png) [@cmagelssen](https://discourse.datamethods.org/u/cmagelssen)\
**Post date:** [January 25, 2023, 4:17pm UTC](https://discourse.datamethods.org/t/ancova-that-handles-pseudoreplication/6386/1 "2023-01-25T16:17:30Z")

</div>

Hi,

This spring I will run a motor learning experiments  
with elite alpine ski racers. I plan to run a between-subjects experimental design with pre and post data. I am going to have pseudoreplication on pre and post, because each participant perform five trials. What is an appropriate statistical model for this data?

Here are my initial thoughts:  
I can aggregate the performance on pre and post, and run a simple ANCOVA. In this case my model will look something like this:

My model will look something like this:  
post\_i ~ constant + pre + group + error\_i.

But that approach will through away information. Is it possible to run a multilevel “ANCOVA” variant on these data, such that I can use all trials in the model without calculating an average score for each participant?

If not, I think I can run a multilevel growth model. With lme4 formula it will be something like this:

performance\_i ~ 1 + time + group + time:group (1 + time | participant)

This will help me include all trials in the model but will have slightly lower power compared with the ANCOVA model (my intuition). But if there is a way to run an ANCOVA with pseudoreplication, that would be very useful.

Christian