Supplementary PROCESS Documentation: Decimal Place Precision in Output

Supplementary PROCESS
Documentation
This document is an addendum to Appendix A of Introduction to Mediation,
Moderation, and Conditional Process Analysis that describes options and output added to PROCESS since the printing of the book in May 2013. Whenever a new feature is added to PROCESS the version number is changed.
This supplementary documentation is cumulative. All features described
as additions to earlier releases can be found in later releases as well. This
document was produced and uploaded to www.afhayes.com on February 27,
2015.
Decimal Place Precision in Output

(Version 2.10)
PROCESS generates numerical output to four decimals places of resolution. This can be changed with the dec argument when using the decimals
option. This argument is set to F10.4 by default, meaning numbers in the
output will contain up to ten characters, with four of these to the right of
the decimal. In this argument, Fa.b sets the number of characters allocated
to numbers to a and the number of decimal places to display to the right
of the decimal point to b. For example, decimals=F12.6 species twelve
characters with six to the right of the decimal place. In the dec argument, a
should be larger than b.
In the SAS version of PROCESS, the F should be left o the dec
argument. For example, to set 12 characters for numbers with six after the
decimal, use decimals=12.6.
Additional Johnson-Neyman Output

(Version 2.10)
When the Johnson-Neyman technique is used, the PROCESS output in
this release includes a table at the top that provides information about the
percent of cases in the data with values of the moderator above (% above)
or below (% below) the points of transition in signicance identied using
the JN method. For example, output such as
Moderator value(s) defining Johnson-Neyman significance region(s)
Value
% below
% above
3.5087
1.5504
98.4496
4.9753
44.1860
55.8140
reects that in the data, only 1.55% of the cases have a value of moderator
less than 3.5087, whereas 98.45% of the cases have a value of the moderator
larger than 3.5087. Similarly, 44.19% of the cases have a value of moderator
less than 4.9753 and 55.81% of the cases have a value of the moderator
larger than 4.9753.
Models 75 and 76
(Version 2.10)
Two new conditional process models are added to PROCESS that specify two moderators of the X M, M Y, and/or X Y paths. See
the conceptual and statistical diagrams in the model templates document
available through the web page for Introduction to Mediation, Moderation,
and Conditional Process Analysis at www.afhayes.com.
Index of Moderated Mediation

(Version 2.10)
For some models, the indirect eect of X on Y through mediator Mi can
be expressed as a linear function of the single moderator in the model. The
slope of the line relating the indirect eect to the moderator is the index
of moderated mediation described in Hayes (2015). The table below formalizes how this index is constructed in terms of the model coecients
from the statistical model as diagrammed in the model templates section
of this Appendix. An inference as to whether this index of moderated mediation is statistically dierent from zero is a formal test of moderation of
the indirect eect by the moderator in the model. PROCESS automatically
produces this index of moderated mediation through each mediator in the
model as well as bootstrap condence interval for inference. The index of
moderated mediation and the bootstrap condence interval produced by
PROCESS is an automatic implementation of the formal test of moderated
mediation described in sections 12.3 and 12.4.
PROCESS Model
7 and 8
14
15
58 (dichotomous W only)
59 (dichotomous W only)
74
Index of Moderated Mediation

(a3i bi )
(ai b3i )
(ai b2i )
[a1i b3i + a3i b1i + a3i b3i (2Wlow + )]
[a1i b2i + a3i b1i + a3i b2i (2Wlow + )]
ai c2i
Unless the moderator is a dichotomous variable, is set to one, yielding

an index of moderated mediation that is the slope of the line relating the
size of the conditional indirect eect of X on Y through Mi to the moderator.
When the moderator is a dichotomous variable, is set to the dierence
between the two values of the moderator coding the two groups, so as to
produce an index that is equal to the dierence between the two conditional indirect eects. This mathematical transformation does not aect the
resulting inference using a bootstrap condence interval.
The index of moderated mediation is provided for models 58 and 59 only
if the moderator is dichotomous. When the moderator is not a dichotomous
variable in these models, the function relating the indirect eect to the
moderator is not a line and the method described in Hayes (2015) cannot
be used. In the expression for the index of moderated mediation of these
two models, Wlow is the smaller of the two values used to code groups.
Because there is only one indirect eect in Model 74 when the moderator
is dichotomous, the index of moderated mediation is produced in Model
74 output only when the moderator is not dichotomous.
Covariance Matrix of Regression Coefcients

(Version 2.11)
PROCESS will display the variance-covariance matrices for the regression coecients in each part of the model by specifying covcoeff=1 in
the PROCESS command line. By default, the variance-covariance is not
produced in the output.
Comparing Conditional Effects in Moderation

Analysis
(Version 2.12)
In models with only a moderation component (models 1, 2, and 3), two
conditional eects of X on Y can be formally compared with a statistical
test. In model 1, evidence of moderation of Xs eect on Y by M leads
to the corresponding claim that any two conditional eects of X on Y for
dierent values of M are dierent from each other. But in models 2 and
3, one can choose dierent values of the two moderators and ask whether
the conditional eect of X on Y diers between two groups dened by
dierent values of the moderators. This requires the use of the contrast
option, setting its argument to 1 (i.e., contrast=1), combined with the
mmodval and/or wmodval options. For instructions, see Hayes (2014).
Long Variable Names in PROCESS for SPSS

(Version 2.13)
Prior to version 2.13, PROCESS for SPSS allowed long variable names
but did not recognize characters after the eighth. This could produce
inaccurate output when the user includes two variables in a model that are
not unique in the rst eight characters. With the release of version 2.13,
the SPSS version of PROCESS no longer accepts variable names longer
than eight characters. If any of the variables in the model have names that
are longer than eight characters, PROCESS will terminate and request the
oending variable name(s) be shortened.
Effect Size in Mediation-Only Models with

Covariates
(Version 2.13)
With the release of version 2.13, PROCESS will produce point and bootstrap interval estimates of several measures of eect size for indirect eects
in models 4, 5 and 6 when covariates are included in the models of both
M and Y (the default covmy=0 option). In a model with one or more covariates, the denitions of several of the eect size measures for indirect
eects discussed in section 6.3 are modied. Dene SDX+ and SDY+ as the
the standard error of estimate (i.e., the square root of the mean squared
residual) when X and Y, respectively, are estimated from only from the
covariates (and, if the cluster option is used, from dummy variables coding
496
Supplementary PROCESS Documentation
the clustering variable). The partially standardized indirect eect is then

dened as
ab
abps =
SDY+
and the completely standardized indirect eect is dened as
abcs =
SDX+ (ab)
= SDX+ (abps )
SDY+
Thus, these two measures gauge the indirect eect relative to variation in
Y and (for the completely standardized indirect eect) X not accounted for
by the covariate(s).
When one or more covariates are included in the model of Y, the total
eect of X is no longer the regression weight for X in a model of Y that
includes only X (denoted c throughout Chapters 4 and 5). As a result, the
ratio of the indirect to the total eect, PM , is no longer ab/c but, rather, the
more general
ab
PM =
c + ab
The denominator of PM above is the regression weight for X in a model
estimating Y from X and the covariate(s) (and cluster variable dummies
when the cluster option is used) but not the mediator(s).
The eect size measures described by Fairchild et al. (2009) (variance in
Y explained by the indirect eect) as well as 2 (Preacher & Kelley, 2011)
have not been generalized to models with covariates and so PROCESS
suppresses their printing in this case. No eect size measures are generated
in models in which the covariates are restricted to the models of M only
(covmy=1) or Y only (covmy=2).
References
Hayes, A. F. (2015). An index and test of linear moderated mediation.
Multivariate Behavioral Research, 50, 1-22.
Hayes, A. F. (2014). Comparing conditional eects in moderated multiple regression: Implementation using PROCESS for SPSS and SAS. White paper
downloadable from http://www.afhayes.com/public/comparingslopes.pdf

Supplementary PROCESS Documentation: Decimal Place Precision in Output

Uploaded by

Copyright:

Available Formats

Supplementary PROCESS Documentation: Decimal Place Precision in Output

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Supplementary PROCESS Documentation: Decimal Place Precision in Output

Uploaded by

Copyright:

Available Formats

Supplementary PROCESS

Decimal Place Precision in Output

Additional Johnson-Neyman Output

Index of Moderated Mediation

Index of Moderated Mediation

Unless the moderator is a dichotomous variable, is set to one, yielding

Covariance Matrix of Regression Coefcients

Comparing Conditional Effects in Moderation

Long Variable Names in PROCESS for SPSS

Effect Size in Mediation-Only Models with

Supplementary PROCESS Documentation

the clustering variable). The partially standardized indirect eect is then

You might also like