No Correlation Between Difficulty and Enjoyment

While gathering the results of my research, I noticed an apparent proportionality between difficulty and enjoyment in Iteration 1's experience graph:

Difficulty and enjoyment seem to rise and fall together through levels 1, 2, and 3.

If there is a correlation in Iteration 2, it seems to be less clear:

Intrigued by this, I tried creating scatter graphs of individual players' enjoyment rating vs difficulty rating. If there was a strong positive correlation between them, I'd see an upwards-rightwards trend between them.

However, no such trend is immediately apparent. Though a slight positive trend can be seen in Iteration 1's scatter graph, where an empty space exists around areas with high difficulty and low enjoyment or vice versa, this trend is still challenged by a few outliers (A rating of 1 difficulty but 3 enjoyment to the far left, or 4 difficulty but two enjoyment to the right). Iteration 2's graph is even harder to make sense of, with ratings (likely by chance) not landing in the 3,3 center of the scatter cluster, forming a ring shape. Here, it seems, it would be easy to draw a positive or negative trend between difficulty and enjoyment.

The experience curve values were based on averages, so I've also tried to create a scatter graph using those values.

Here, we can see a very clear correlation. Enjoyment increases with difficulty. Why does this trend appear when considering my players collectively, but less so individually? What does this mean? Is it even a valid observation?

Apparently, individual players show very little correlation between difficulty and enjoyment. A player may consider an easy game to be very fun, and a difficult game to be not so fun. But, when these results are compiled into an average, a clear correlation emerges, such that enjoyment tends to increase with difficulty on average, between all players.

I'm wary of this being due to an unforeseen statistical quirk - similar to how rolling two dice tends to produce a sum of 7 more than any other number. I don't know much about statistics, and wouldn't know how to identify this kind of mistake. Alternatively, maybe my sample of testers is just two small to draw any useful conclusions.

If the correlation is valid, however, then maybe the graph of Average Enjoyment vs Difficulty Ratings eliminates deviations to reveal a genuine underlying relationship. In this case, the trend would be showing what I suspected; that more difficult challenges are more thoroughly enjoyed by the player, and that too much hand-holding damages the game's experience.

But, unless I can confirm this concept, I'm not in a position to draw this conclusion.