The effect in the data (unless it isn't random fluctuation. At the moment I'd go with the assumtion that it is a genuine correlation, as the effect is present over all subgroups), might (1) not be monocausal, i.e. a combination of contributing factors like development experience, age of IDEs/tooling, etc. might play a role as well as other aspects. (2) a cause might not be "explained away" because the control variable was considered. Let me elaborate

The years of experience variable might be taken to explain away the effect of experience, and then conclude that tabs vs. spaces must be due to another effect than years of experience. But chances are, that "years of experience" and "tabs vs. spaces" are just correlated to a common, causal property (like "programming proficiency" or however you want to call it). Both "years of experience" and "tabs vs. spaces" are then just incomplete reflections of the underlying cause, both rendering the effect of the underlying cause incompletely.

What I am trying to say is: Its complicated, probably you won't be able to find the one true cause for the effect in the data. If this were physics, one could come up with a predictive theory to put this to the test. In social studies, we just cannot control the parameters well enough.

If you are interested in reading more on this, "Causality" by Judea Pearl is a good (but exhausting read).