Which of two predictors reliably correlates with the DV changes, depending on the analysis. Evidence that it's their shared variance that correlates?

I'm working on a manuscript addressing hypothesized links between each of two empathic tendencies (measured with newly created, reliable self-report scales)—to share pleasures [Resonant Pleasure], and to experience a 'Helper's High'—and self-reports of satisfaction with life.

My question is based on the fact that depending on the analysis, and on the sub-sample involved, which of these two traits correlates reliably with satisfaction, keeps changing.

I'm guessing that this is because the variance shared with satisfaction, is also shared between scores on the Resonant Pleasure and Helper's High scales. They correlate at r(421) = .38. I would appreciate any guidance on whether I'm right about that. Details follow.

Both Resonant Pleasure and the Helper's High show reliable, positive zero-order correlations with scores on the Satisfaction with Life Scale: at r = .19 and .21, respectively.

Some analyses were conducted with data from a subset of the sample, because only that subset had completed the Positive and Negative Affect Scales (PANAS).

In data from the full sample, univariate tests within a multivariate multiple regression analysis show no effect of Resonant Pleasure on Satisfaction (F(1,400) < 1). But they do show a reliable effect of Helper's High on Satisfaction (F(1,400) = 3.73, p = .05).

In contrast, a parallel analysis using only cases that provided PANAS data, shows the opposite. For this subset, univariate tests do show a reliable effect of Resonant Pleasure (F(1,239) = 4.75, p < .05); but they do not show an effect for Helper's High (F(1,239) < 1).

In a path analysis using data for the full sample, the path coefficient linking Resonant Pleasure with Satisfaction with Life (at .11) is reliable. The comparable path coefficient linking Helper's High with Satisfaction (at .06) is not. (We conducted no path analysis using data from the subset, with satisfaction as an outcome.)

As mentioned, the Resonant Pleasure and Helper's High scales overlap. Perhaps it's their shared basis in positive empathic sensitivity that correlates with Satisfaction with Life: such that analyses that look for unique contributions of IVs, in predicting satisfaction, will show the contribution alternating between the two, but never showing a contribution for both.

Does the variability (across analyses) in which scale correlates with satisfaction, constitute evidence that what's correlated with satisfaction is something the scales have in common?

Many thanks in advance for any advice.