### INTRODUCTION

### Aging and memory

### Possible mediators

### METHODS

### Participants

### Measures

### Executive function

### Memory

### Analysis

^{2}) test result. However, this largely depends on sample size, so we also looked for other indices of fit. Following the guidelines of Hu and Bentler,20 values of the comparative fit index (CFI) and Tucker-Lewis Index (TLI) close to 0.95 and a root mean square error of approximation (RMSEA) value below 0.06 are all indicative of good model fit. We employed traditional χ

^{2}comparisons between the nest and nested models to obtain a more parsimonious model. To test the differences between men and women, we conducted multi-group path analysis.

### RESULTS

### Model tests

^{2}test statistics and other fit indices of the two competing models are presented in Table 2. The first model is the hypothesized full mediation model (Figure 2). The χ

^{2}test values were not significant, and all fit indices satisfied the criteria for acceptability (χ

^{2}=19.60, df=17, p=0.295, TLI=0.994, CFI=0.997, RMSEA=0.026). Therefore, the model in which the effects of aging on memory decline are fully mediated by executive function is sufficient to explain our data.

^{2}test of difference. We created Model 2 by adding two additional pathways to Model 1; thus, Model 2 had two additional restrictions and fewer free parameters. Compared with Model 1, Model 2 lost two degrees of freedom, which resulted in the following values: Δχ

^{2}=7.14 (p<0.05)(χ

^{2}critical value=5.99 at the p<0.05 level). Δχ

^{2}was significant at the 0.05 level, indicating that Model 2 provided a better explanation of our data than Model 1. Moreover, all of the traditional fit indices were better for Model 2 (χ

^{2}=12.46, df=15, p=0.644, TLI=1.006, CFI=1.000, RMSEA=0.000) than for Model 1. Therefore, the partial mediation model was accepted as the final model for this study.

### Path analysis for the final model

### Multi-group analysis of gender differences

^{2}test values, and all fit indices satisfied the criteria for goodness of fit (χ

^{2}=33.202, df=34, p=0.507, TLI=1.002, CFI=1.000, RMSEA=0.000). The regression weights of each structural path are presented in Table 4. Since measurement invariance was satisfied, we imposed equality constraints on each structural path and all structural paths between men and women, and conducted a χ

^{2}difference test. As described in Table 5, there were no differences in structural paths between men and women, and when we constrain all the paths in the models, the overall goodness of model fit was not changed significantly (Δχ

^{2}=7.613, p=0.179, ΔTLI=0.005). In sum, the hypothesized model was cross-validated among men and women, and the research model can be applied to both middle-aged men and women.