>>67215271The top right plots the CCV at a particular time to the CCV 1 minute after, so we are essentially trying to find the relation between CCV and CCV 1 minute after. As it turns out, for some streams with smooth-enough graphs which eventually plateaus, the relationship is approximately linear. That is, to predict the CCV 1 minute from now, you multiply the CCV today by a certain factor then add a constant. This "certain factor" can be interpreted as the average retention rate per minute of the streamer, while the "constant" can be interpreted as the average added viewership. We do this process of multiplying and adding numbers repeatedly to the first recorded CCV in the data to get the fitted model on the top left plot. The bottom plots are essentially analyses of the error between observed values and the regression line of the top right plot.
>>67215313I understand that YT CCV does not accurately record passive tabs, but it's the best and only data that we've got right now. The small spikes can be modelled as small fluctuations in the model and are (hopefully) captured as random errors in the model I made. Though, I do think that I should have been more cautious with my conclusions.
Regarding the video you suggested, when I took a look at it in VStats it seems that the graph doesn't seem that "pure" (see picrel). Up until the 48 minute mark everything seems relatively normal but there was a drop afterwards. I'll just try to use the "normal-looking" CCV time intervals in that stream, thanks for the suggesting the video anon.
>>67215503Yet despite knowing how shitty the site is, here we are still arguing about the numbers after 2+ years.
