H1: Do positive (but not negative) emotions predict increases in ego-resilience and life satisfaction?
Positive and negative emotion scores were calculated for each day and averaged across the month (positive: M = 1.95, SD = 0.51; negative: M = 0.63, SD = 0.38). The two scores were not correlated on any day (median r=−.05, median p=.51) or in aggregate (r=.17, p=.12).
Correlations between composite emotion scores and the outcome variables are displayed in . Positive emotions predict ego resilience and life satisfaction, and also predict increases over the course of the month, over and above any change predicted by T1 values. Effect sizes are in the medium-to-large range.
Correlations Between Emotion Indices and Resilience and Life Satisfaction
Negative emotions predict T2 resilience only (r=−.22, p=.05). This is smaller than the relationship between T2 ego-resilience and PE, but the difference is only marginally significant (z=.168, p=.09). Negative emotions do not predict any life satisfaction scores (ps > .13). All estimated effect sizes for negative emotions are small.
Ego-resilience and life satisfaction were not correlated at either timepoint (rT1
=.54), but change in ego-resilience predicted change in life satisfaction (β=.32, p
H2: Do positive emotions (but not life satisfaction) partially mediate the relation between initial and final ego-resilience scores?
We constructed a mediation model in which T1
ego-resilience predicted T2
ego-resilience, with daily positive emotions as a mediator. We used the Sobel test for mediation with bootstrap estimation for coefficients (Preacher & Hayes, 2004
). Positive emotions continued to predict T2
ego-resilience (β=.22, p
=.001), as did T1
ego-resilience (β = .74, p
< .001). The indirect pathway – T1
through positive emotions – was significant (β = .09, p
< .01), indicating that positive emotions partially mediate the relationship between T1
We assessed life satisfaction as a mediator in place of positive emotions, using life satisfaction at T1
, and their average. In each case the indirect path was nonsignificant and near zero (
< .01, p
>.15; change in direct path ≤ .01).
H3: Are increases in ego-resilience responsible for the relation between positive emotions and increased life satisfaction?
Positive emotions predicted change in life satisfaction (β=.15, p=.03). To determine whether this is consistent with contributing to life satisfaction specifically by building resources, we tested change in ego-resilience as a mediator. T2 life satisfaction was treated as the outcome, with T1 life satisfaction as a predictor. Thus, semipartial correlations of other variables with T2 life satisfaction predict variance above and beyond that predicted by T1 levels. Positive emotions were an additional predictor, and change in resilience was a mediator.
When the paths through all predictors are taken into account, change in resilience remains significant (β=.15, p=.02). The direct path from positive emotions to change in life satisfaction becomes nonsignificant (β=.08, p=.22), but the indirect path through change in resilience is significant (β=.05, p<.01). This pattern indicates that the relationship of positive emotions to change in life satisfaction is fully mediated by change in resilience.
The relations described here can be illustrated by viewing change in ego-resilience across positive emotions quartiles () and change in life satisfaction across ego-resilience change quartiles ().
Raw Change in Ego-Resilience Scores by Positive Emotion Quartiles.
H4: Do negative emotions reduce the effects of positive emotions?
Although negative emotions were not a significant predictor of change in ego-resilience or life satisfaction, they may reduce the predictive value of positive emotions. We tested a model in which the relationship between T1 and T2 ego-resilience was simultaneously mediated by positive emotions, negative emotions, and their interaction. Positive emotions remained a significant predictor of change in resilience (β=.27, p<.001). Negative emotions became a significant negative predictor (β=−.17, p=.01), but there was no significant interaction (β=.10, p=.10). The indirect path through positive emotions accounted for a significant portion of the relationship between T1 and T2 ego-resilience (β=.09, p<.01), but the paths through negative emotions (β=.03) and through the interaction (β=−.03) did not (p≥.15). This model was a significant improvement over the model with positive emotions as the sole predictor (ΔR2 =.03; ΔF=4.1, p=.02).
The significant coefficient for negative emotions indicates how strongly they predict low change in ego-resilience when PE is held constant at zero (because variables used here are standardized, zero represents the mean score). The interaction term indicates that NE will relate to ego-resilience change differently depending on the level of PE. Specifically, when both NE and PE are above zero, the positive interaction term will partly cancel out the negative NE term. Using the “pick a point” strategy (Rogosa, 1980
) we determined that the “balance point” is when PE is ≥.45 SD above the sample mean. At that level of PE, an increase in NE no longer significantly predicts a decrease in ego resilience change. In contrast, PE becomes a stronger predictor of ego resilience change at high levels of NE (PE becomes nonsignificant when NE is >1 SD below
the mean). This indicates that high levels of PE reduce the impact of any increase in NE, but high NE does not reduce the impact of PE.
We also tested positive emotions, negative emotions, and their interaction as predictors of increases in life satisfaction. Positive emotions remained significant (β=.17, p=.01), negative emotions remained nonsignificant (β=−.10, p=.15), and the interaction was nonsignificant (β=.09, p=.19). This model did not improve over using positive emotions alone (ΔR2 =.01; ΔF=1.5, p=.23).
H5: Are rising levels of positive emotions necessary?
We tested whether increases in ego-resilience and life satisfaction required not just the presence of positive emotions, but an increase in positive emotions over baseline. We created a change variable from the Week 4 positive emotions score with Week 1 positive emotions partialed out, and repeated the tests described above, using the change variable in place of the aggregate positive emotions variable. Change in positive emotions did not significantly predict baseline, one-month, or change scores for ego-resilience or life satisfaction (r<.19, p≥.09). It also did not have a significant indirect effect in the relationship between T1 and T2 ego-resilience (β= .10, p = .12). We replicated these tests again with aggregate positive emotions and change in positive emotions entered simultaneously (despite overlap in predictors, all tolerance scores were >.78). The regression coefficients for positive emotions change declined, while aggregate positive emotions remained significant. Thus, it appears that absolute levels of positive emotions matter more than positive emotions relative to baseline.