Moderated Mediation: Annotated SAS Output proc contents; run; proc means data=protest; run; The SAS System The CONTENTS Procedure Data Set Name WORK.PROTEST2 Observations 129 Alphabetic List of Variables and Attributes # Variable Type Len Format Label 6 LikingZ Num 8 7 RespapprZ Num 8 5 SexismZ Num 8 2 liking Num 8 F8.2 4 protest Num 8 PROTEST. PROTEST: experimental condition (0 = no protest, 1 = protest) 3 respappr Num 8 F8.2 RESPAPPR: appropriateness of response 1 sexism Num 8 F8.2 SEXISM: perceived pervasiveness of sex discrimination LIKING: liking of the target The MEANS Procedure Variable Label sexism SEXISM: perceived pervasiveness of sex discrimination liking respappr protest LIKING: liking of the target RESPAPPR: appropriateness of response PROTEST: experimental condition (0 = no protest, 1 = protest) data protest2; set protest; SexismZ=(sexism-5.1169767)/0.7837616; LikingZ=(liking-5.6367442)/1.0496973; RespapprZ=(respappr-4.8662791)/1.3481203; run; Proc Means; run; N Mean Std Dev Minimum Maximum 129 5.1169767 0.7837616 2.8700000 7.0000000 129 5.6367442 1.0496973 1.0000000 7.0000000 129 4.8662791 1.3481203 1.5000000 7.0000000 129 0.6821705 1.0000000 0.4674481 0 The MEANS Procedure Variable Label sexism SEXISM: perceived pervasiveness of sex discrimination liking respappr protest SexismZ LikingZ RespapprZ LIKING: liking of the target RESPAPPR: appropriateness of response PROTEST: experimental condition (0 = no protest, 1 = protest) N Mean Std Dev Minimum Maximum 129 5.1169767 0.7837616 2.8700000 7.0000000 129 5.6367442 1.0496973 1.0000000 7.0000000 129 4.8662791 1.3481203 1.5000000 7.0000000 129 0.6821705 0.4674481 0 1.0000000 129 5.6376896E-8 1.0000000 -2.8669135 2.4025460 129 -1.329287E-8 1.0000000 -4.4172203 1.2987133 129 -2.242571E-8 1.0000000 -2.4970168 1.5827378 Simple Mediation Analysis %process (data=protest2, vars=protest RespapprZ LikingZ,y=LikingZ,x=protest,m=RespapprZ,boot=10000,model=4); Model and Variables Model = 4 Y= LIKINGZ X= PROTEST M= RESPAPPRZ Sample size: 129 Outcome: RESPAPPRZ Model Summary R R-sq F df1 df2 p 0.4992 0.2492 42.1550 1.0000 127.0000 0.0000 Model coeff se t p LLCI ULCI Constant -0.7285 0.1359 -5.3625 0.0000 -0.9973 -0.4597 PROTEST 1.0679 0.1645 6.4927 0.0000 0.7425 1.3934 If you were to conduct an independent samples t-test comparing the two groups on perceived response appropriateness, you would get the t and p reported here. Mean standardized perceived appropriateness was 1.068 higher when the attorney protested than when she did not. Outcome: LIKINGZ Model Summary R R-sq F df1 df2 p 0.4959 0.2459 20.5483 2.0000 126.0000 0.0000 Model coeff constant se t p LLCI ULCI 0.0655 0.1514 0.4324 0.6662 -0.2341 0.3650 RESPAPPRZ 0.5168 0.0893 5.7884 0.0000 0.3401 0.6935 PROTEST -0.0959 0.1910 -0.5023 0.6163 -0.4739 0.2820 This is an ANCOV. If you were to conduct an independent samples t-test comparing the groups on how much the respondent liked the attorney, you would find a significant difference, with them liking her better when she protested, d = .465. Here we have held constant the effect of response appropriateness, and the adjusted means no longer differ significantly. ****************************** DIRECT AND INDIRECT EFFECTS ******************************* Direct effect of X on Y Effect SE t p LLCI ULCI -0.0959 0.1910 -0.5023 0.6163 -0.4739 0.2820 Indirect effect of X on Y Effect Boot SE BootLLCI BootULCI RESPAPPRZ 0.5519 0.1436 0.3115 0.8775 The indirect effect here is the product of the coefficients, 1.0679(.5168) = .5519. Simple Moderation Analysis %process (data=protest2, vars=protest RespapprZ SexismZ,y=RespapprZ,x=protest,m=SexismZ,quantile=1,model=1); Model and Variables Model = 1 Y= RESPAPPRZ X= PROTEST M= SEXISMZ Outcome: RESPAPPRZ Model Summary R R-sq F df1 df2 p 0.5442 0.2962 17.5340 3.0000 125.0000 0.0000 Model coeff constant se t p LLCI ULCI -0.7466 0.1328 -5.6206 0.0000 -1.0095 -0.4837 SEXISMZ -0.3075 0.1371 -2.2426 0.0267 -0.5790 -0.0361 PROTEST 1.0815 0.1607 6.7282 0.0000 0.7634 1.3997 INT_1 0.4709 0.1639 2.8732 0.0048 0.1465 0.7953 Back to the future. Interactions: INT_1 PROTEST X SEXISMZ R-square increase due to interaction(s): R2-chng INT_1 F df1 df2 p 0.0465 8.2551 1.0000 125.0000 0.0048 Conditional effect of X on Y at values of the moderator(s) SEXISMZ Effect se t p LLCI ULCI -1.2720 0.4825 0.2590 1.8633 0.0648 -0.0300 0.9950 -0.7872 0.7108 0.2027 3.5065 0.0006 0.3096 1.1120 0.0039 1.0833 0.1608 6.7386 0.0000 0.7652 1.4015 0.6418 1.3838 0.1950 7.0950 0.0000 0.9978 1.7698 1.2798 1.6842 0.2685 6.2724 0.0000 1.1528 2.2156 Values for quantitative moderators are the 10th, 25th, 50th, 75th, and 90th percentiles. Values for dichotomous moderators are the two values of the moderator. Perceived prevalence of sexism significantly moderates the effect of protesting on perceived response appropriateness. As perceived prevalence of sexism increases, the difference between the two groups gets larger. Moderation Analysis With Covariate %process (data=protest2, vars=protest RespapprZ SexismZ LikingZ,y=LikingZ,x=protest,m=SexismZ,quantile=1,model=1); Here we have identified Response Appropriateness as a covariate. Model = 1 Y= LIKINGZ X= PROTEST M= SEXISMZ Statistical controls: RESPAPPRZ Sample size: 129 Outcome: LIKINGZ Model Summary R R-sq F df1 df2 p 0.5323 0.2833 12.2551 4.0000 124.0000 0.0000 Model coeff se t p LLCI ULCI constant 0.0127 0.1506 0.0844 0.9329 -0.2854 0.3108 SEXISMZ -0.2109 0.1417 -1.4882 0.1392 -0.4914 0.0696 PROTEST -0.0297 0.1901 -0.1563 0.8760 -0.4059 0.3465 INT_1 0.4051 0.1714 2.3628 0.0197 0.0658 0.7444 RESPAPPRZ 0.4614 0.0906 5.0916 0.0000 0.2820 0.6408 Back to the future. Interactions: INT_1 PROTEST X SEXISMZ R-square increase due to interaction(s): R2-chng INT_1 F df1 df2 p 0.0323 5.5830 1.0000 124.0000 0.0197 ***************************************************************************************** Conditional effect of X on Y at values of the moderator(s) SEXISMZ Effect se t p LLCI ULCI -1.2720 -0.5450 0.2660 -2.0491 0.0426 -1.0715 -0.0186 -0.7872 -0.3486 0.2152 -1.6196 0.1079 -0.7746 0.0774 0.0039 -0.0282 0.1902 -0.1480 0.8826 -0.4046 0.3483 0.6418 0.2303 0.2340 0.9840 0.3270 -0.2329 0.6935 1.2798 0.4887 0.3119 1.5668 0.1197 -0.1287 1.1061 Back to the future. Values for quantitative moderators are the 10th, 25th, 50th, 75th, and 90th percentiles. Values for dichotomous moderators are the two values of the moderator. Since we have held constant the effect of the mediator, these effects are the conditional direct effects of protesting on liking. Those low in perceived pervasiveness of sexism actually like the attorney better when she did not protest. Moderated Mediation Analysis %process (data=protest2, vars=protest RespapprZ SexismZ LikingZ,y=LikingZ,x=protest,w=SexismZ, m=RespapprZ,quantile=1,model=8, boot=10000); Model and Variables Model = 8 Y= LIKINGZ X= PROTEST M= RESPAPPRZ Model and Variables W= SEXISMZ Sample size: 129 Outcome: RESPAPPRZ Model Summary R R-sq F df1 df2 p 0.5442 0.2962 17.5340 3.0000 125.0000 0.0000 Model coeff se t p LLCI ULCI Constant -0.7466 0.1328 -5.6206 0.0000 -1.0095 -0.4837 PROTEST 1.0815 0.1607 6.7282 0.0000 0.7634 1.3997 SEXISMZ -0.3075 0.1371 -2.2426 0.0267 -0.5790 -0.0361 INT_1 0.4709 0.1639 2.8732 0.0048 0.1465 0.7953 Interactions: INT_1 PROTEST X SEXISMZ We have seen this interaction earlier Outcome: LIKINGZ Model Summary R R-sq F df1 df2 p 0.5323 0.2833 12.2551 4.0000 124.0000 0.0000 Model coeff constant se t p LLCI ULCI 0.0127 0.1506 0.0844 0.9329 -0.2854 0.3108 RESPAPPRZ 0.4614 0.0906 5.0916 0.0000 0.2820 0.6408 PROTEST -0.0297 0.1901 -0.1563 0.8760 -0.4059 0.3465 SEXISMZ -0.2109 0.1417 -1.4882 0.1392 -0.4914 0.0696 INT_2 0.4051 0.1714 2.3628 0.0197 0.0658 0.7444 Interactions: Interactions: INT_2 PROTEST X SEXISMZ We have seen this interaction earlier. DIRECT AND INDIRECT EFFECTS We have seen these conditional direct effects earlier. Conditional direct effect(s) of X on Y at values of the moderator(s) SEXISMZ Effect SE t p LLCI ULCI -1.2720 -0.5450 0.2660 -2.0491 0.0426 -1.0715 -0.0186 -0.7872 -0.3486 0.2152 -1.6196 0.1079 -0.7746 0.0774 0.0039 -0.0282 0.1902 -0.1480 0.8826 -0.4046 0.3483 0.6418 0.2303 0.2340 0.9840 0.3270 -0.2329 0.6935 1.2798 0.4887 0.3119 1.5668 0.1197 -0.1287 1.1061 Conditional indirect effect(s) of X on Y at values of the moderator(s) SEXISMZ Effect Boot SE BootLLCI BootULCI RESPAPPRZ -1.2720 0.2226 0.1402 -0.0134 0.5549 RESPAPPRZ -0.7872 0.3280 0.1186 0.1397 0.6199 RESPAPPRZ 0.0039 0.4998 0.1258 0.2915 0.7829 RESPAPPRZ 0.6418 0.6385 0.1646 0.3542 1.0008 RESPAPPRZ 1.2798 0.7771 0.2172 0.3971 1.2584 As the value of perceived pervasiveness of sexism increases, the magnitude of the indirect effect of protest on liking through response appropriateness increases. Values for quantitative moderators are the 10th, 25th, 50th, 75th, and 90th percentiles. Indirect effect of highest order interaction Effect Boot SE BootLLCI BootULCI RESPAPPRZ 0.2173 0.1040 0.0323 0.4462 (Protest x SexismResponse Appropriateness)(Response AppropriatenessLiking) .4709(.4614) = .2173. proc sgplot; series x=SexismZ y=direct/curvelabel = 'Direct Effect' lineattrs=(color=blue pattern=ShortDash); series x=SexismZ y=indirect/curvelabel = 'Indirect Effect' lineattrs=(color=red pattern=Solid); xaxis label = 'Perceived Pervasiveness of Sexism'; yaxis label = 'Conditional Effect of Protest'; refline 0/axis=y lineattrs=(color=gray pattern=ShortDash); run;