First visualization

John has an idea, but he wants to make sure that it will work. His idea is to hit one big home run to propel his channel to the top. He wants to make a big video where he will spend a large amount of money to gain many views in a short amount of time. To see if his idea would work, he first wants to analyze whether high delta_views picks will then bring more views in the future. In other words, he wants to detect outliers and see if they influence the behavior of the time series.

At the same time, he can detect if having bad weeks with fewer views can also have an influence. Using his result, he can then compare high and low outliers and see if there are any patterns to predict what will happen with those outliers' delta_views.

Now he sees a potential influence in some graphs, like the ones in the second row and fourth column. It was his idea to deal with this outlier and bring them back around the curve.

He stated unequivocally that there could be an influence. He intends to delve into the analysis of those view slopes, as planned.

Slopes Analyze

All channels delta_views

First, he wants to get an idea of the difference in slopes of distribution. How much do outliers influence channels? How much do they not have? 

The plot he sees shows a kind of exponential distribution with a large plateau at the beginning. His opinion is to say that few channels are affected by outliers. But what he really wants to know is if there is any difference between falling channels and rising channels.

Hmm.. The distribution is very chaotic, and there is not much difference between positive and negative slopes. Moreover, there is no particular pattern to extract from this.

Nonetheless, he only used delta_views from now on. He thinks: "What makes a big channel apart from the number of views it generates?" Subscriber! Let's do the same thing with delta_subs!

All channels delta_subs

Interestingly, the distribution looks the same. So, also concerning delta_subs, outliers seem not to influence the curve. Let's try again and separate the positive and negative slopes.

"Ah, ah!" he thinks. The distribution appears to be chaotic, but an intriguing fact is that channels appear to have more falling delta-subs curves than rising ones. This could be due to the fact that a subscriber can only subscribe once. In comparison, a single person can watch the video an infinite number of times. So after some time, the curve of the subscriber will arrive at a plateau since there is no infinite number of humans.

Unfortunatly, this observation doesn't help him in his task: Is thre any pattern that can be different between falling and rising channels.

He doesn't give up. What he wants is to have his channel in the "Howto & Style" category. And hence, wants to analyse outlier in this particular category.

Howto & Style delta_views

The distribution is very similar compare to the graph with all youtube channels. So channel delta_views's curve behavior in the Howto & Style categorie models well the overall behavior of all youtube channels according to this indicator(delta_view). But as John see the seperate graph, he change his mind.

He sees that Howto & Style categorie tends to have more falling channels than rising one in termns of delta_views. However, there still no particular pattern. The graph looks still chaotic and there is not so much difference between positive and negative slops difference distribution.

John try one last visualization using delta_subs in the Howto & Style category.

Howto & Style delta_subs

Those graphs he produced look basically the same as the previous ones. So he came to the conclusion, according to his analysis, that by doing so, there are no identifiable patterns between negative and positive slopes.


One thing he can note is that according to delta_views and delta_subs, outliers don't always make your channels rise or fall; it's only in rare cases that this happens. In that sense, spending a large amount of money for a video to generate buzz doesn't guarantee that his channels will grow. In contrast, it will be a better choice not to believe in outliers in order to grow his channel. However, it also works on the opposite side; a low outlier doesn't mean that his channel will fall and disappear.