I feel like i should have a basic idea of why this is, but im not entierly sure so i figured id ask.

I have a problem im trying to solve, it involves figuring out the causal connection between a certain set of stats, and a score given to individuals, this score is a linear combination of the stats.

The individuals are devided into ~3 groups (sub groups exist but that doesnt matter for my purposes), so to get a "feel" for it, i ran a simple correlation between each stat vector and the score metric, both for each individual group, as well as for all of them together.

For the correlation of one of these stats i got:All together: 0.479, Group A: 0.786, Group B: 0.663, Group C: 0.542

Edit: It is worth noting however that this specific metric is not a factor in the final score, I.E the multiplier is 0, but i am looking to potentially use it, and therefore need to understand this, and either way curious in the more general sense.

What confuses me is that the correlation of all the groups to the stat is lower then each individual group, whereas i would expect it to be somewhere in between.

I feel like the answer is something along the lines of "each group has a different kind relationship with the stat, and therefor all together its less linear then each individually", but i could be pulling that out of my ass.

If anyone has an explanation i would very much appriciate the help, thanks!