Lunar effect: Does the full moon really make some of us crazy?

It’s perhaps a common belief that the full moon affects human and animal behaviours, causing sleepless nights, mental disorders, howling wolves and increasing crimes on the streets. The phenomenon has been scientifically studied over the years, and the results have been either negative or disputable.

With Utopia AI Moderator, our text analytics tool, we at Utopia Analytics have analysed a huge amount of data, containing over 200 million user-generated content (UGC) items. What the tool does is learning from human decisions on which content is appropriate and which isn’t.

We thought it would be interesting to see if we could find any correlation between the quality of user-generated content and lunar phases.

Our hypothesis was that if there is truly any lunar effect, then during the night of a full moon, more user-generated content will be rejected due to being improper.

Data and preparation for analysis

The analyzed data ranges from September 1, 2018 until end of March 2019, containing 7 full moons.

The analyzed content items are divided into the following categories:
– Very short chat messages,
– General topic discussion forum messages, and
– Restricted topic news articles comments.
Every message has been analyzed by Utopia AI Moderator, and has been marked either accepted or improper, based on the meaning of the wording in its context.

We have used Utopia AI Moderator results in this study, because they are more consistent and reliable than human decisions over time.

Normally, improper rates (i.e. the ratio of rejected content to the total number of content) don’t vary much over time. On the other hand, improper rates between different types of content may differ considerably. Therefore, in this study the improper rates for the analysis were normalized.

For each type of content, the lowest and highest improper percentage was noted, and then the improper rate was normalized between 0 % and 100% between the noted limits. This way we were able to make all the types of content comparable to each other, and also amplified the changes thus making small variations more visible.

Results

The average of all user content is shown in Figure 1. It is hard to tell a pattern from the picture. Someone may see that in 2019 there is some structure, but in 2018 there clearly is not. Is the result purely random, or is there something?

Figure 1. Improper rates compared to the lunar phases with the full data set.

Figure 2 shows results for chat messages and news comments. News commenting (red) improper rate varies over the 7 months period, but it is most likely caused by news events, local or global. Chat messages (purple) are stable, and there is only a little variance.

Figure 2. Improper rates of news comments and chat messages compared to the lunar phases.

In Figure 3 we looked at general topic forum comments. During 2019, a clear pattern following the full moons can be seen, in 2018 similar pattern is recognisable but not quite correctly aligned with the full moons.

Figure 3. Improper rates of forum comments compared to the lunar phases.

Final conclusions

This blog is not a serious attempt to scientifically study lunar effect. Rather, it shows that a lot of interesting information can be obtained using fairly simple data analysis techniques if there is suitable data available.

The original question, if user-generated content is affected by moon phases, is left to the reader to be answered.