Limits...
A cautionary note on the power of the test for the indirect effect in mediation analysis.

Loeys T, Moerkerke B, Vansteelandt S - Front Psychol (2015)

Bottom Line: Recent simulation studies have pointed to the higher power of the test for the mediated effect vs. the test for the total effect, even in the presence of a direct effect.This has motivated applied researchers to investigate mediation in settings where there is no evidence of a total effect.On the basis of the results, we recommend that when the primary interest lies in mediation only, a significant test for the total effect should not be used as a prerequisite for the test for the indirect effect.

View Article: PubMed Central - PubMed

Affiliation: Department of Data Analysis, Ghent University Ghent, Belgium.

ABSTRACT
Recent simulation studies have pointed to the higher power of the test for the mediated effect vs. the test for the total effect, even in the presence of a direct effect. This has motivated applied researchers to investigate mediation in settings where there is no evidence of a total effect. In this paper we provide analytical insight into the circumstances under which higher power of the test for the mediated effect vs. the test for the total effect can be expected in the absence of a direct effect. We argue that the acclaimed power gain is somewhat deceptive and comes with a big price. On the basis of the results, we recommend that when the primary interest lies in mediation only, a significant test for the total effect should not be used as a prerequisite for the test for the indirect effect. However, because the test for the indirect effect is vulnerable to bias when common causes of mediator and outcome are not measured or not accounted for, it should be evaluated in a sensitivity analysis.

No MeSH data available.


Related in: MedlinePlus

The power to detect the indirect effect (IE). Data are generated according to the right panel of Figure 1 with a = 0.4 and c' = 0.16 and a residual correlation rho between M and Y equal to 0.336. The different power curves represent varying assumptions on unmeasured confounding of the M-Y relationship in a sensitivity analysis.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4290592&req=5

Figure 4: The power to detect the indirect effect (IE). Data are generated according to the right panel of Figure 1 with a = 0.4 and c' = 0.16 and a residual correlation rho between M and Y equal to 0.336. The different power curves represent varying assumptions on unmeasured confounding of the M-Y relationship in a sensitivity analysis.

Mentions: We mimic the setting of Rucker et al. (2011) with the direct effect equal to 0.16, the effect of X on M fixed to 0.4, but now with a spurious correlation between M and Y induced by a standard normal distributed variable U (the right panel of Figure 1). Typically, factors other than X that affect M also affect Y in the direction that M affects Y (Bullock et al., 2010). We assume here that U has the same effect on M and Y, and results in a spurious 0.4 effect of M on Y. Hence, the true total and indirect effect equal 0.16 and 0, respectively; but the spurious indirect effect, ignoring the unmeasured U, also equals 0.16. We simulate such data in samples of size N (25, 50, 100, 200), and repeated each setting 5000 times. The total and indirect effect are estimated each time using models (1), (2), and (3). Figure 4 shows the power to detect the total effect, as well as the power to detect the spurious indirect effect (curve with ρ equal to zero, cfr. infra). As expected, one finds as before substantial power gain for the test of ab vs. c with increasing sample size. However, this is misguided in view of the absence of an indirect effect.


A cautionary note on the power of the test for the indirect effect in mediation analysis.

Loeys T, Moerkerke B, Vansteelandt S - Front Psychol (2015)

The power to detect the indirect effect (IE). Data are generated according to the right panel of Figure 1 with a = 0.4 and c' = 0.16 and a residual correlation rho between M and Y equal to 0.336. The different power curves represent varying assumptions on unmeasured confounding of the M-Y relationship in a sensitivity analysis.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4290592&req=5

Figure 4: The power to detect the indirect effect (IE). Data are generated according to the right panel of Figure 1 with a = 0.4 and c' = 0.16 and a residual correlation rho between M and Y equal to 0.336. The different power curves represent varying assumptions on unmeasured confounding of the M-Y relationship in a sensitivity analysis.
Mentions: We mimic the setting of Rucker et al. (2011) with the direct effect equal to 0.16, the effect of X on M fixed to 0.4, but now with a spurious correlation between M and Y induced by a standard normal distributed variable U (the right panel of Figure 1). Typically, factors other than X that affect M also affect Y in the direction that M affects Y (Bullock et al., 2010). We assume here that U has the same effect on M and Y, and results in a spurious 0.4 effect of M on Y. Hence, the true total and indirect effect equal 0.16 and 0, respectively; but the spurious indirect effect, ignoring the unmeasured U, also equals 0.16. We simulate such data in samples of size N (25, 50, 100, 200), and repeated each setting 5000 times. The total and indirect effect are estimated each time using models (1), (2), and (3). Figure 4 shows the power to detect the total effect, as well as the power to detect the spurious indirect effect (curve with ρ equal to zero, cfr. infra). As expected, one finds as before substantial power gain for the test of ab vs. c with increasing sample size. However, this is misguided in view of the absence of an indirect effect.

Bottom Line: Recent simulation studies have pointed to the higher power of the test for the mediated effect vs. the test for the total effect, even in the presence of a direct effect.This has motivated applied researchers to investigate mediation in settings where there is no evidence of a total effect.On the basis of the results, we recommend that when the primary interest lies in mediation only, a significant test for the total effect should not be used as a prerequisite for the test for the indirect effect.

View Article: PubMed Central - PubMed

Affiliation: Department of Data Analysis, Ghent University Ghent, Belgium.

ABSTRACT
Recent simulation studies have pointed to the higher power of the test for the mediated effect vs. the test for the total effect, even in the presence of a direct effect. This has motivated applied researchers to investigate mediation in settings where there is no evidence of a total effect. In this paper we provide analytical insight into the circumstances under which higher power of the test for the mediated effect vs. the test for the total effect can be expected in the absence of a direct effect. We argue that the acclaimed power gain is somewhat deceptive and comes with a big price. On the basis of the results, we recommend that when the primary interest lies in mediation only, a significant test for the total effect should not be used as a prerequisite for the test for the indirect effect. However, because the test for the indirect effect is vulnerable to bias when common causes of mediator and outcome are not measured or not accounted for, it should be evaluated in a sensitivity analysis.

No MeSH data available.


Related in: MedlinePlus