Mediation Analysis - JuliA Renberg`s Portfolio
Transcription
Mediation Analysis - JuliA Renberg`s Portfolio
Results #3 EDRS 821 Julia and Luis Determine if perceived teacher competence is a mediator between effort and overall amount learned. First, bivariate correlation was conducted to check if there is significant correlation between a) IV and M; b) M and DV. In this case, there is significant correlation for a) effort and teacher competency (r = .40, p < 0.05) and b) teacher competency and overall amount learned (r = .64, p < 0.05). Additionally, although not required, there was significant correlation between IV (effort) and DV (overall amount learned) with r = .37, p < 0.05. Effort 1s Effort 1s Pearson Correlation Sig. (2-tailed) N Pearson Correlation Sig. (2-tailed) N Pearson Correlation Sig. (2-tailed) N Overall Amount Learned 1s Teacher Competency 1s Overall Amount Learned 1s .366** .000 430 1 1 431 .366** .000 430 .399** .000 430 430 .642** .000 430 Teacher Competency 1s .399** .000 430 .642** .000 430 1 430 **. Correlation is significant at the 0.01 level (2-tailed). To establish mediation effect, Baron & Kenny’s causal step approach was followed. Step 1: We ran regression model to see if there was an association between effort (IV) and mediator - teacher competency (DV). Model Summary Model 1 R .399a R Square .159 Adjusted R Square Std. Error of the Estimate .158 .67040 a. Predictors: (Constant), Effort 1s Model 1 Standardized Coefficients Beta Unstandardized Coefficients Std. Error B (Constant) 1.579 Effort 1s .426 a. Dependent Variable: Teacher Competency 1s .151 .047 t Sig. 10.483 9.012 .399 .000 .000 Step 2: We checked whether the association of independent and dependent variable reduced significantly (partial mediation) or disappear (full mediation) when the mediator was added. Model Summary Model R R Square Adjusted R Square 1 .366a .134 .132 b 2 .653 .426 .424 a. Predictors: (Constant), Effort 1s b. Predictors: (Constant), Effort 1s, Teacher Competency 1s Model 1 (Constant) Effort 1s 2 (Constant) Effort 1s Teacher Competency 1s Std. Error of the Estimate .74882 .61016 Unstandardized Coefficients B Std. Error 1.279 .168 .430 .053 .254 .154 .153 .047 .649 .044 Standardized Coefficients Beta Repeated work via Processor: MATRIX procedure: ***************** PROCESS Procedure for SPSS Release 2.13 *************** Written by Andrew F. Hayes, Ph.D. www.afhayes.com Documentation available in Hayes (2013). www.guilford.com/p/hayes3 ************************************************************************** Model = 4 Y = AmtLe_1s X = Effrt_1s M = TComp_1s t .366 .130 .590 7.601 8.129 1.653 3.255 14.753 Sig. .000 .000 .099 .001 .000 Sample size 430 ************************************************************************** Outcome: TComp_1s Model Summary R .3993 R-sq .1595 MSE .4494 F 81.2085 df1 1.0000 df2 428.0000 p .0000 Model constant Effrt_1s coeff 1.5793 .4263 se .1507 .0473 t 10.4829 9.0116 p .0000 .0000 LLCI 1.2832 .3333 ULCI 1.8754 .5193 ************************************************************************** Outcome: AmtLe_1s Model Summary R .6528 R-sq .4262 MSE .3723 F 158.5865 df1 2.0000 df2 427.0000 p .0000 Model coeff .2541 .6490 .1529 constant TComp_1s Effrt_1s se .1537 .0440 .0470 t 1.6528 14.7526 3.2552 p .0991 .0000 .0012 LLCI -.0481 .5625 .0606 ULCI .5562 .7355 .2452 ************************** TOTAL EFFECT MODEL **************************** Outcome: AmtLe_1s Model Summary R .3657 R-sq .1338 MSE .5607 F 66.0844 df1 1.0000 df2 428.0000 p .0000 Model constant Effrt_1s coeff 1.2790 .4296 se .1683 .0528 t 7.6008 8.1292 p .0000 .0000 LLCI .9483 .3257 ULCI 1.6098 .5334 ***************** TOTAL, DIRECT, AND INDIRECT EFFECTS ******************** Total effect of X on Y Effect SE .4296 .0528 t 8.1292 p .0000 LLCI .3257 ULCI .5334 Direct effect of X on Y Effect SE .1529 .0470 t 3.2552 p .0012 LLCI .0606 ULCI .2452 Indirect effect of X on Y Effect Boot SE TComp_1s .2767 .0355 BootLLCI .2173 BootULCI .3593 Normal theory tests for indirect effect Effect se Z p .2767 .0360 7.6775 .0000 ******************** ANALYSIS NOTES AND WARNINGS ************************* Number of bootstrap samples for bias corrected bootstrap confidence intervals: 1000 Level of confidence for all confidence intervals in output: 95.00 NOTE: Some cases were deleted due to missing data. The number of such cases was: Questions Please give complete but concise responses to the following questions regarding the analyses you have conducted. 1 Explain any differences between the B&K approach and the PROCESS output. The Baron and Kenny approach does not require testing of the indirect effect of the mediator. Their four step analysis (although many reduce them to two steps, as we did) to establish mediation requires that: 1) The causal variable is correlated with the outcome; 2) The causal variable is correlated with the mediator; 3) The mediator affects the outcome variable; 4) To establish that M completely mediates the X-Y relationship, the effect of X on Y controlling for M (path c') should be zero for full mediation (or reduced to a very small amount for partial mediation). SPSS program provides us with necessary information to test for significance of indirect effect, but does not actually performs any tests. PROCESS output does essentially B&K steps, but also runs the Sobel test and a more superior bootstrapped test. Similarly, ROCESS calculates indirect effect of the mediator (in this case, .2767), where as SPSS program only provides us with calculated coefficients for paths a and b and we have to manually calculate the combined effect as ab (.426 x .649). 2 Write a short (1-2 sentence) summary of your model finding (write in terms of constructs). What does your model suggest? The present model suggests that teacher competency is a mediator, because when entered into the model it partially “takes over” or “wipes out” the direct effect of effort on the overall amount learned (still significant), reducing the unstandardized coefficient from .43 to .15. Therefore, effort affects teacher competency, and teacher competency affects the overall amount learned. 3 What is the bootstrapped CI for the indirect effect? Is it significant? Indirect effect of X on Y Effect Boot SE TComp_1s .2767 .0355 BootLLCI .2173 BootULCI .3593 According to the PROCESS output, the bootstrapped CI for the indirect effect (see above) ranged from .22 to .36. This result indicates that the indirect effect of the mediator was significant because there is no zero included in this range. 4 Conduct a Sobel test using your SPSS output—is it significant? To test for significance of indirect effect (in other words, whether or not the total effect of X on Y is significantly reduced by the addition of a mediator to the equation), the Sobel test was performed (entering unstandardized coefficients and standard errors from the SPSS output for a path (IV to M) and b path (M to DV) into the Sobel test calculator on Kristopher Preacher’s website). The result was significant with the t value exceeding 1.96 - t (430)=7.69, p<0.05. These results are the same for the PROCESSOR output: Normal theory tests for indirect effect Effect se Z p .2767 .0360 7.6775 .0000 5 Add a second mediator (perceived teacher caring) to your model using PROCESS. Draw a picture of this model filling in path coefficients. Write 3-5 sentences explaining the model. Run MATRIX procedure: ***************** PROCESS Procedure for SPSS Release 2.13 *************** Written by Andrew F. Hayes, Ph.D. www.afhayes.com Documentation available in Hayes (2013). www.guilford.com/p/hayes3 ************************************************************************** Model = 4 Y = AmtLe_1s X = Effrt_1s M1 = TCare_1s M2 = TComp_1s Sample size 430 ************************************************************************** Outcome: TCare_1s Model Summary R .3776 R-sq .1425 MSE .7029 F 71.1519 df1 1.0000 df2 428.0000 p .0000 Model constant Effrt_1s coeff .9714 .4990 se .1884 .0592 t 5.1558 8.4352 p .0000 .0000 LLCI .6011 .3828 ULCI 1.3417 .6153 ************************************************************************** Outcome: TComp_1s Model Summary R .3993 Model R-sq .1595 MSE .4494 F 81.2085 df1 1.0000 df2 428.0000 p .0000 constant Effrt_1s coeff 1.5793 .4263 se .1507 .0473 t 10.4829 9.0116 p .0000 .0000 LLCI 1.2832 .3333 ULCI 1.8754 .5193 ************************************************************************** Outcome: AmtLe_1s Model Summary R .6654 R-sq .4427 MSE .3624 F 112.8032 df1 3.0000 df2 426.0000 p .0000 Model constant TCare_1s TComp_1s Effrt_1s coeff .3221 .1711 .5007 .1307 se .1529 .0482 .0602 .0468 t 2.1072 3.5513 8.3123 2.7963 p .0357 .0004 .0000 .0054 LLCI .0216 .0764 .3823 .0388 ULCI .6226 .2657 .6191 .2226 ************************** TOTAL EFFECT MODEL **************************** Outcome: AmtLe_1s Model Summary R .3657 R-sq .1338 MSE .5607 F 66.0844 df1 1.0000 df2 428.0000 p .0000 Model constant Effrt_1s coeff 1.2790 .4296 se .1683 .0528 t 7.6008 8.1292 p .0000 .0000 LLCI .9483 .3257 ULCI 1.6098 .5334 ***************** TOTAL, DIRECT, AND INDIRECT EFFECTS ******************** Total effect of X on Y Effect SE .4296 .0528 t 8.1292 p .0000 LLCI .3257 ULCI .5334 Direct effect of X on Y Effect SE .1307 .0468 t 2.7963 p .0054 LLCI .0388 ULCI .2226 Indirect effect of X on Y Effect Boot SE TOTAL .2988 .0363 TCare_1s .0854 .0267 TComp_1s .2135 .0340 (C1) -.1281 .0492 BootLLCI .2339 .0382 .1528 -.2312 BootULCI .3779 .1421 .2891 -.0353 Normal theory tests for specific indirect effects Effect se Z p TCare_1s .0854 .0262 3.2537 .0011 TComp_1s .2135 .0351 6.0897 .0000 Specific indirect effect contrast definitions (C1) TCare_1s minus TComp_1s ******************** ANALYSIS NOTES AND WARNINGS ************************* Number of bootstrap samples for bias corrected bootstrap confidence intervals: 1000 Level of confidence for all confidence intervals in output: 95.00 NOTE: Some cases were deleted due to missing data. 16 The number of such cases was: ------ END MATRIX ----- Teacher Caring .50 .17 Overall Amount Learned .13 Effort 0. .43 Teacher Competency .50 The presented above regression model is statistically significant and explains about 44% of variance of our criterion “overall amount learned” (R2 adj = .44, F (3, 426)= 112.80, p<0.05). The model shows that the direct effect of our causal variable (effort) on the dependent variable (overall amount learned) is replaced by two mediating variables (teacher caring and teacher competency). The cumulative mediating effect of these variables is only partial, as the effect of IV on DV is reduced from (unstandarized coefficient changed from .43 to .13), but remained significant. The Sobel and the Bootstrapping tests indicate the indirect effect for both, teacher competency (t (430)=6.09, p<0.05;[.15, .28] and teacher caring (t (430)=3.25, p<0.05;[.04, .14] were significant. The indirect effect of teacher competency (.21) was stronger that of teacher caring (.09). Write up Note: Write up your results using the Baron & Kenny method describing what steps you took and the results of your analysis (APA format). Include a diagram of the proposed model. You should include the following: Hypotheses that correspond to the above question. Results: Divide this into two sections. o Descriptive statistics: Table of the correlations among all variables in the study. Correlations Overall Amount Learned 1s .366** .000 431 430 ** .366 1 .000 430 430 .378** .571** .000 .000 430 430 ** .399 .642** .000 .000 430 430 Effort 1s 1 Effort 1s Pearson Correlation Sig. (2-tailed) N Overall Amount Learned 1s Pearson Correlation Sig. (2-tailed) N Teacher Caring 1s Pearson Correlation Sig. (2-tailed) N Teacher Competency 1s Pearson Correlation Sig. (2-tailed) N **. Correlation is significant at the 0.01 level (2-tailed). Teacher Caring 1s .378** .000 430 .571** .000 430 1 430 .739** .000 430 Teacher Competency 1s .399** .000 430 .642** .000 430 .739** .000 430 1 430 o Hypothesis testing: results of the hypotheses tests. Discussion: Include a brief discussion of your results. Be sure to explain clear the mediational relationship.