Outliers and jitter

[andresriancho]

Outliers and jitter https://youtu.be/ykNt8pSQFZQ?t=1505

How to take advantage of this? Maybe...

• Group the samples in bins (like histogram)
• Get the lower X% (those are the ones with the lowest jitter)
• Assert we have more than Y samples
• Graph the filtered data
• Use only those samples to compare

X and Y are two important variables that I should "brute-force" to test how well they work. Intuition tells me:

• X should be around 2, 5, 10, 15, 20%
• Y should be around 50, 100, 150, 200, 250.

This might be very similar in nature to the box test... but it's worth a try.

Check this video to understand other potential filters I could implement. They were successful with a "low percentile filter" 2-25% (which is what I was explaining above). Note that they do eliminate 0-2%

[andresriancho]

I believe I could use this as a jitter filter and try the other statistical analysis methods with and without it.