Mediation Analysis
To evaluate the model, we performed mediation analyses
described by Baron and Kenny1 and following more
recent guidelines presented by Preacher and Hayes.35 In
traditional mediation, multiple regression is used to determine
the strength of the association between predictor
and outcome variables both before and after the
inclusion of a third (mediating) variable. Although Baron
and Kenny describe mediation as resulting when the effect
of the predictor on the outcome (ie, the total effect) is
reduced to zero after inclusion of the mediating variable,
Preacher and Hayes emphasize the importance of measuring
the size and statistical significance of the indirect effect
(ie, the effect of the predictor through the mediator) as
opposed to simple reduction of the total effect. The Sobel
statistic37 is a widely used test of the strength of the indirect
effect and was included in our analyses.
Separate mediation analyses were performed on the
outcome of exercise behavior (months of regular exercise)
for each of 2 predictor variables: Perceived importance
of exercise and exercise self-efficacy. Readiness to
change exercise behavior served as the mediator in each
model. Given that this plan called for the estimation of
coefficients in 2 separate mediational models, was adjusted
to P .025 (ie, .05/2) accordingly.
Consistent with the Baron and Kenny approach, we
first established the assumptions of mediation in each
model in 3 steps: (1) First, by regressing the outcome
variable on the predictor variable, (2) second, by regressing
the mediator on the predictor, and (3) finally, by
regressing the outcome on the mediator after controlling
for the predictor. After establishing the existence of
relationships among these variables, we tested mediation
via evaluation of the indirect path by using the Sobel
statistic.
It is important to note that the use of cross-sectional
data poses significant limitations for mediation analyses,
as one cannot make statements about causality in correlations
determined at a single time point. Although mediation
is perhaps the most appropriate test given the
model to be tested, the results need to be interpreted
with caution.
Mediation AnalysisTo evaluate the model, we performed mediation analysesdescribed by Baron and Kenny1 and following morerecent guidelines presented by Preacher and Hayes.35 Intraditional mediation, multiple regression is used to determinethe strength of the association between predictorand outcome variables both before and after theinclusion of a third (mediating) variable. Although Baronand Kenny describe mediation as resulting when the effectof the predictor on the outcome (ie, the total effect) isreduced to zero after inclusion of the mediating variable,Preacher and Hayes emphasize the importance of measuringthe size and statistical significance of the indirect effect(ie, the effect of the predictor through the mediator) asopposed to simple reduction of the total effect. The Sobelstatistic37 is a widely used test of the strength of the indirecteffect and was included in our analyses.Separate mediation analyses were performed on theoutcome of exercise behavior (months of regular exercise)for each of 2 predictor variables: Perceived importanceof exercise and exercise self-efficacy. Readiness tochange exercise behavior served as the mediator in eachmodel. Given that this plan called for the estimation ofcoefficients in 2 separate mediational models, was adjustedto P .025 (ie, .05/2) accordingly.Consistent with the Baron and Kenny approach, wefirst established the assumptions of mediation in eachmodel in 3 steps: (1) First, by regressing the outcome
variable on the predictor variable, (2) second, by regressing
the mediator on the predictor, and (3) finally, by
regressing the outcome on the mediator after controlling
for the predictor. After establishing the existence of
relationships among these variables, we tested mediation
via evaluation of the indirect path by using the Sobel
statistic.
It is important to note that the use of cross-sectional
data poses significant limitations for mediation analyses,
as one cannot make statements about causality in correlations
determined at a single time point. Although mediation
is perhaps the most appropriate test given the
model to be tested, the results need to be interpreted
with caution.
การแปล กรุณารอสักครู่..
![](//thimg.ilovetranslation.com/pic/loading_3.gif?v=b9814dd30c1d7c59_8619)