The PROCESS macro provides several advantages over the traditional causal steps (Baron & Kenny, 1986) and products of coefficients (i.e., Sobel testing; Baron & Kenny, 1986; Sobel, 1982) approaches to testing mediation. First, PROCESS relies on a single statistic to specifically test the indirect effect of an independent variable (X) on the dependent variable (Y) through some mediator (M). Single test procedures are more powerful than the causal steps approach that requires a minimum of three tests; the failing to meet significance on any one of which will lead to rejection of the presence of indirect effects. While Sobel testing overcomes this limitation through the use of a single test for indirect effects, Sobel testing assumes a normally-shaped sampling distribution of the indirect effect over repeated sampling of the population. This assumption is often difficult to meet. However, PROCESS requires no such normality assumption. By employing bias-corrected bootstrap confidence intervals within a single test of indirect effects, PROCESS provides a method for testing of indirect effects that is more powerful than causal steps and that minimizes bias in results that can arise from non-normal sampling distributions (Hayes, 2013).

Baron, R. M., & Kenny, D. A. (1986). The moderator-mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal of personality and social psychology, 51, 1173-1182. doi:10.1037/0022-3514.51.6.1173

Hayes, A. F. (2013). Introduction to mediation, moderation, and conditional process analysis: A regression-based approach. New York: Guilford Press.

Sobel, M. E. (1982). Asymptotic intervals for indirect effects in structural equations models. In S. Leinhart (Ed.), Sociological methodology (pp. 290-312). San Francisco: Jossey-Bass.

Hi Shelly,

the "causal steps" approach has become known to have some serious problems, which is why experts on mediation recommend against using it. There are better approaches available (in SPSS), such as the PROCESS macro: http://processmacro.org/index.html

For an overview, see:

MacKinnon, D. P., Coxe, S., & Baraldi, A. N. (2012). Guidelines for the investigation of mediating variables in business research. Journal of Business and Psychology, 27(1), 1-14.

The PROCESS macro provides several advantages over the traditional causal steps (Baron & Kenny, 1986) and products of coefficients (i.e., Sobel testing; Baron & Kenny, 1986; Sobel, 1982) approaches to testing mediation. First, PROCESS relies on a single statistic to specifically test the indirect effect of an independent variable (X) on the dependent variable (Y) through some mediator (M). Single test procedures are more powerful than the causal steps approach that requires a minimum of three tests; the failing to meet significance on any one of which will lead to rejection of the presence of indirect effects. While Sobel testing overcomes this limitation through the use of a single test for indirect effects, Sobel testing assumes a normally-shaped sampling distribution of the indirect effect over repeated sampling of the population. This assumption is often difficult to meet. However, PROCESS requires no such normality assumption. By employing bias-corrected bootstrap confidence intervals within a single test of indirect effects, PROCESS provides a method for testing of indirect effects that is more powerful than causal steps and that minimizes bias in results that can arise from non-normal sampling distributions (Hayes, 2013).

Baron, R. M., & Kenny, D. A. (1986). The moderator-mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal of personality and social psychology, 51, 1173-1182. doi:10.1037/0022-3514.51.6.1173

Hayes, A. F. (2013). Introduction to mediation, moderation, and conditional process analysis: A regression-based approach. New York: Guilford Press.

Sobel, M. E. (1982). Asymptotic intervals for indirect effects in structural equations models. In S. Leinhart (Ed.), Sociological methodology (pp. 290-312). San Francisco: Jossey-Bass.

Dear Shelly,

Baron and Kenny’s (1986) procedure for testing Mediation effects  is so well known that it is widely used by researchers.However, more recently, some of its conditions and assumptions have been criticized, and other procedures have been suggested for testing Mediation effects, such as the product of coefficients method proposed by  MacKinnon and colleagues (2002).

Baron, R.M., & Kenny, D.A. (1986). The moderator-mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology, 51, 1173-1182.

MacKinnon, D.P., Lockwood, C.M., Hoffman, J.M., West, S.G., & Sheets, V. (2002). A comparison of methods to test mediation and otherintervening variable effects. Psychological Methods, 7, 83-104.

4 steps in establishing mediation

(Baron & Kenny, 1986; Judd &Kenny, 1981; James & Brett, 1984)

•          Step 1:  Show that the causal variable is correlated with the outcome.  Use Y as the criterion variable in a regression equation and X as a predictor (estimate and test path c in the above figure). This step establishes that there is an effect that may be mediated.

•          Step 2: Show that the causal variable is correlated with the mediator.  Use M as the criterion variable in the regression equation and X as a predictor (estimate and test path a).  This step essentially involves treating the mediator as if it were an outcome variable.

•          Step 3:  Show that the mediator affects the outcome variable.  Use Y as the criterion variable in a regression equation and X and M as predictors (estimate and test path b).  It is not sufficient just to correlate the mediator with the outcome because the mediator and the outcome may be correlated because they are both caused by the causal variable X.  Thus, the causal variable must be controlled in establishing the effect of the mediator on the outcome.

•          Step 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.   The effects in both Steps 3 and 4 are estimated in the same equation.

•          If all four of these steps are met, then the data are consistent with the hypothesis that variable M completely mediates the X-Y relationship

•          If the first three steps are met but the Step 4 is not, then partial mediation is indicated. 

•          The amount of mediation is called the indirect effect (total effect = direct effect + indirect effect)

Baron and Kenny’sprocedure for testing mediation

CONCEPTUALPROBLEMS: Some of its conditions and assumptions have been criticized, especially the X-Y requirement(first step)

•LeBreton,J.M.,Wu,J.,&Bing,M.N.(2009).Thetruth(s)ontestingformediationinthesocialandorganizationalsciences.InC.E.Lance&R.J.Vandenberg(Eds.),Statisticalandmethodologicalmythsandurbanlegends(pp.109-144).NewYork:Routledge.

•Zhao,X.,Lynch,J.G.Jr.,&Chen,Q.(2010).ReconsideringBaronandKenny:MythsandTruthsaboutMediationAnalysis.JournalofConsumerResearch,37,197-206.

Many research projects have been terminated early in a research program because the data did not conform first Baron and Kenny’s criteria, impeding theoretical development. The procedure tells to start by establishing a basic effect: “Once you have the effect, then you can look for mediation.” But if the effect is not found, the project is abandoned. So that, misapplication of the Baron-Kenny procedure was causing authors to drop projects that might be promising.

The product of coefficients method proposed by MacKinnon et al. (2002).

MacKinnon,D.P.,Lockwood,C.M.,Hoffman,J.M.,West,S.G.,&Sheets,V.(2002).Acomparisonofmethodstotestmediationandotherinterveningvariableeffects.PsychologicalMethods,7,83-104.

Mediation involves a causal relationship whereby an independent variable (X) impacts on a mediator (M),

which in turn impacts on a dependent variable (Y)

1stRegression equation: DV=M, IV=X (X(Path a)To show that there is a significant relationship between M and X)

2ndRegression equation: DV=Y, IV=X, M (Path b To show if there is a significant relationship between M and Y,controlling for the effect of X on Y) and To estimate the non-mediated or direct effect. The product a is the mediated or indirect effect

We could say that a relationship is mediated if:

1)X is significantly related to M(a statistically significant)

2) M is significantly related to Y after controlling for X (b statistically significant)

3) The mediated effect is statistically significant (the product ab is significant)

Using this model you ban use standardized regression coefficients in the SPSS output of multiple regression analysis

Good luck with your research!

I do not see where the questions was answered as to why the Causal Steps approach uses unstandardized rather than standardized coefficients. Moreover, I believe the same question can be asked of the Hayes Process Macro bootstrapping method for evaluating the significance of the mediational pathway. I am just going through the book now and it looks like it uses unstandardized rather than standardized coefficients. If anyone knows I would also be interested in the answer (not that I have any reason to think it should be based on standardized coefficients, and I can think of reasons why maybe it should not be, just curious).

As far as I know, tests are always conducted on unstandardized parameter estimates (i.e. using the original metric of the data). Standardized parameter estimates are computed for interpretation, which may or may not be helpful. For simple models, it is relatively easy to do this by hand.

I think the reason that the use of raw beta weights caught my attention is that I was reviewing Baron and Kenny in the context of path analysis and structural equation modeling which are alternate approaches for analyzing mediation. I think in both cases the convention is to focus on standardized coefficients. I realize they are not the same as B&K which is all OLS regression but it is still not clear to me why they use standardized coefficients and B&K uses unstandardized.

Richard J Contrada The regression/path coefficients that PROCESS produces are in unstandardized form.  Please see the link: http://www.processmacro.org/faq.html

Most of the answers here well point out to the superiority of OLS based mediation analysis using bootstrapping over the "causal steps" approach described in Baron and Kenny (1986). Adding on these, I would like to briefly discuss why reporting unstandardized, instead of standardized, coefficients and effect sizes in general, might be preferred in a mediation framework. My answer largely follows the line of thought by Hayes (2017).

First, Hayes strongly discourages reporting (only) standardized coefficients/effect sizes for cases where X is a dichotomous variable (which is very common in studies with a single treatment condition) (page 519). He argues that for such cases, std.-ed coefficients (hence all other direct, indirect, and total effects) are influenced by the distribution of cases between the two groups as well as differences between group means. Also, the standardized coefficients do not allow a comparison of group means and relative indirect effects straightforwardly. Instead, Hayes suggests to report either unstd.-ed or both unstd.-ed and partially std.-ed coefficients (reported by PROCESS).

For other cases, Hayes points out that if the key variables in an empirical model are based on similar measurement scales (e.g., a 7-point Likert scale), then reporting unstandardized coefficients gives a straightforward interpretation to all coefficients and effect sizes and there is no advantage in reporting standardized coefficients. Moreover, on page 519, Hayes has a good discussion (that I also agree with) on why the belief held by many social scientists that standardization coefficients should be reported is not quite warranted.

Two papers (more on the technical side) on the related issue that are recommended by Hayes are Kim and Mueller (1976) and Baguley (2009).

Hayes, A. F. (2017). Introduction to mediation, moderation, and conditional process analysis: A regression-based approach. Guilford Publications.

Kim, J. O., & Mueller, C. W. (1976). Standardized and unstandardized coefficients in causal analysis: An expository note. Sociological Methods & Research, 4(4), 423-438.

Baguley, T. (2009). Standardized or simple effect size: What should be reported?. British Journal of Psychology, 100: 603-617.