Read on Twitter
Our new @JBiomech paper on sample size estimation for biomechanical waveforms is now online at https://doi.org/10.1016/j.jbiomech.2021.110451">https://doi.org/10.1016/j... . Coauthors @ScienceJos & Todd Pataky. Here is what it is about.
Sample size estimation is relatively infrequent in biomechanics but we were not sure how infrequent. From all 2018, 2019 JoB papers only 4% conducted a-priori power analysis. That& #39;s pretty low. Those that did (n=29) were not well reported and many examined waveforms.
We wanted to show that sample size estimation for biomechanics waveforms is possible using the open-source power1d package https://spm1d.org/power1d/index.html#">https://spm1d.org/power1d/i... using numerical simulation, but not in typical software e.g. JASP, G*Power, SPSS, etc.
You first need to think of the 1D predicted effect for your experiment. There are lots reported in the biomechanics literature.
A 1D effect can have many shapes so it is important that we can define these flexibly.
We also need to specify some 1D noise, but once you have a hypothesised experiment, you can use numerical simulation to calculate the sample size required to meet your target power.
So we did this for many example effects in the literature, also comparing between 0D (discrete) power results [yellow] and 1D (waveform) power results [blue].
![So we did this for many example effects in the literature, also comparing between 0D (discrete) power results [yellow] and 1D (waveform) power results [blue].](https://pbs.twimg.com/media/Ez5zbGdX0AI9yKE.jpg "So we did this for many example effects in the literature, also comparing between 0D (discrete) power results [yellow] and 1D (waveform) power results [blue].")
1D sample sizes were always larger than 0D but ranged from 0D+1 to 0D+20. This means if 0D sample sizes are used to test a 1D hypothesis then the experiment will likely be underpowered.
In the paper we rationalise current practice, provide some recommendations for unambiguous reporting of power analyses and help to show how sample sizes for 1D analysis can be justified. *end of thread*
