¦ 2017 Vol. 13 no.

Simple and parallel mediation:

A tutorial exploring anxiety sensitivity, sensation seeking,
and gender

Leanne Kanea and Andrea R. Ashbaugha, B

a
School of Psychology, University of Ottawa

Abstract In this tutorial, we demonstrate how to conduct simple and parallel mediation analyses Acting Editor De-
using the PROCESS macro for SPSS (Hayes, 2013). We begin by describing the principles of medi- nis Cousineau (Uni-
versité d’Ottawa)
ation. We then present a step-by-step tutorial describing how to test statistical assumptions and
conduct a simple and a parallel mediation using data from a project exploring whether anxiety sen- Reviewers
sitivity mediates the relationship between gender and sensation seeking in a sample of 295 under- One anonymous re-
viewer
graduate students. Results of these analyses showed that anxiety sensitivity, and more specifically
the belief that bodily sensations are dangerous, explains part of the relationship between gender
and sensation seeking. Finally, we interpret these results as if we are presenting the findings in
a research article. The tutorial serves as a concrete and need-to-know introduction to simple and
parallel mediation.
Keywords mediation, gender differences, anxiety sensitivity, sensation seeking. Tools SPSS,
PROCESS macro.

B [email protected]

LK: 0000-0002-9054-2596; ARA: 0000-0001-8590-0136

10.20982/tqmp.13.3.p148

late that the relationship between two variables is more

Introduction
complex than meets the eye. To illustrate this, we will use
In psychology, researchers are often interested in exam- the example of gender and sensation seeking. Sensation
ining whether variables are related to one another. To seeking refers to a personality trait implicating the search
achieve this, they can conduct regression-based analyses. for novel, intense, and complex sensations through vari-
Although causality cannot be inferred from this type of ous experiences (Zuckerman, 1994). It has also been con-
analysis, researchers anchor themselves in a theoretical ceptualized as a motivating need for novelty and intensity
framework to determine which variable is said to be the (Roth & Hammelstein, 2012). In a meta-analysis, Cross,
outcome and which is said to be the cause of this outcome. Copping, and Campbell, 2011 determined that men scored
For instance, one might wonder if gender is related to var- significantly higher on measures of sensation seeking than
ious personality characteristics like impulsivity or sensa- women (d = 0.41). However, various factors may account
tion seeking. Conceptually, it would make more sense to for the relationship between gender and sensation seek-
test whether gender predicts these characteristics than the ing. Indeed, the underlying processes are certainly any-
other way around. However, it would also be possible to thing but simple.
examine if these characteristics predict gender. A theo- In cognitive-behavioural models of psychological phe-
retical framework will help determine which option is the nomena, behaviours, thoughts, and emotions mutually in-
most probable and useful way of conceptualising the rela- fluence each other (Hofmann, Asmundson, & Beck, 2013).
tionship. A person may therefore choose to engage, or to not engage,
Moreover, theory may also guide researchers to postu- in activities that produce new and exciting sensations for

The Quantitative Methods for Psychology

2 148
 ¦ 2017 Vol. 13 no. 3

Figure 1 Simple mediation using the mediating effect of anxiety sensitivity on the relationship between gender and
sensation seeking. Notes: a is effect of gender on anxiety sensitivity; b is effect of anxiety sensitivity on sensation seeking;
c0 is direct effect of gender on sensation seeking.

many reasons. Indeed, if a person interprets sensations, effects associated with this pathway (Hayes, 2013). In ad-
such as rapid heart beat, sweating, and racing thoughts, in dition, there is the direct effect (c0 ), which is the effect of
a negative way (i.e., they do not like those feelings), then gender on sensation seeking while keeping levels of anxi-
they will likely seek less of these sensations. Many indi- ety sensitivity constant (Rucker et al., 2011). When combin-
viduals fear sensations like these because of their possi- ing the indirect and the direct effects, you obtain the total
ble negative consequences, a construct know as anxiety effect (c), which is in fact the result you would get by sim-
sensitivity (Taylor, Zvolensky, Cox, Deacon, & Heimberg, ply regressing sensation seeking on gender (Hayes, 2013;
2007). Anxiety sensitivity has three dimensions: Physi- Rucker et al., 2011). The coefficients associated with the
cal Concerns (the beliefs that bodily sensations are life- various pathways (i.e., a, b, ab, c0 , and c) are essentially un-
threatening), Cognitive Concerns (the belief that difficulties standardized regression coefficients. We will see more on
concentrating are dangerous), and Social Concerns (the be- their interpretations later in the tutorial.
lief that others will reject or laugh at observable anxiety The strength of the indirect and the direct effects will
symptoms; Taylor et al., 2007). Women report higher anxi- determine the result of the mediation analysis (MacKinnon
ety sensitivity than men (Stewart, Taylor, & Baker, 1997). et al., 2007). If the indirect effect is significant, then it is
Consequently, women may seek exciting sensations less considered to be successful mediation (MacKinnon et al.,
frequently than men. It would be interesting therefore 2007). When this occurs, the direct effect may disappear
to see whether gender affects sensation seeking through or remain significant. If it disappears, then there is com-
its effect on anxiety sensitivity. More specifically, perhaps plete mediation (i.e., the effect of X on Y is entirely due to
gender affects levels of anxiety sensitivity, which in turn M ), whereas if it remains, then there is partial mediation
may influence the tendency to seek novel and intense sen- (i.e., M does account for part of the relationship between
sations. X and Y , but, X still predicts Y even when taking into ac-
This kind of hypothesis can be tested using mediation count M ; MacKinnon et al., 2007).
analysis (e.g., Hayes, 2013; MacKinnon, Fairchild, & Fritz, This conceptualisation of partial versus complete me-
2007; Preacher & Hayes, 2004; Rucker, Preacher, Tormala, diation is very wide-spread in the scientific community
& Petty, 2011), wherein a mediator (M ; anxiety sensitivity) (Hayes, 2013). However, it does have its critics. Hayes
is proposed to explain the relationship between an inde- (2013) presents various arguments against the use of this
pendent variable (X ; gender) and an outcome variable (Y ; nomenclature. For instance, he posits that partial media-
sensation seeking; see Figure 1). This model is a simple me- tion is bound to occur because something will mediate all
diation because there is only one mediator. effects; it is simply a question of finding it (Hayes, 2013).
In this example, gender is proposed to influence anx- The process behind variable associations is too complex
iety sensitivity (a), which in turn would affect sensation and is bound to be mediated by some factor or another.
seeking (b). This is called the indirect effect (ab) of gen- Moreover, discovering that a variable completely medi-
der on sensation seeking through anxiety sensitivity. This ates the relationship between two others does not exclude
indirect effect is obtained by multiplying a and b, the two the possibility that other constructs, not assessed in the

The Quantitative Methods for Psychology

2 149
 ¦ 2017 Vol. 13 no. 3

Figure 2 Parallel mediation using the mediating effect of three anxiety sensitivity dimensions in the relationship be-
tween gender and sensation seeking. Notes: an is effect of gender on anxiety sensitivity dimensions, women are coded as
0 and men as 1; bn is effect of anxiety sensitivity dimensions on sensation seeking; c0 is direct effect of gender on sensation
seeking.

study, may also play a role in the relationship (Hayes, 2013; iety symptoms (Social Concerns; Taylor et al., 2007). These
Rucker et al., 2011). In essence, just because a researcher dimensions may play different roles in the relationship be-
finds evidence for one mediator does not mean that it is the tween gender and sensation seeking. Using parallel medi-
whole story. Therefore, readers should be cautioned if they ation, we can test the mediating effects of all three anxiety
choose to report their analyses using the partial-complete sensitivity dimensions (see Figure 2). We would then have
view of mediation. However, given that it is widely used in three possible indirect effects: one through each anxiety
scientific articles, it is necessary to comprehend its defini- sensitivity dimension.
tions to better identify its pitfalls. Well written and comprehensive books regarding me-
Furthermore, simple mediation is the simplest of medi- diation analysis are available for the interested reader
ation models. More complex models, such as parallel me- (e.g., Hayes, 2013). This paper will more so act as a concrete
diation, can include more than one mediator (Hayes, 2013). and need-to-know tutorial on simple and parallel media-
In parallel mediation, two or more variables (M1 , M2 , etc.) tion by extending the gender, anxiety sensitivity, and sen-
are proposed to mediate the relationship between X and sation seeking example introduced above. After introduc-
Y (see Figure 2). These mediators are allowed to correlate ing the research question and briefly discussing the data
with one another, but not to influence each other in causal- and the methods used to collect them, we will describe and
ity (Hayes, 2013). In parallel mediation, there are as many test the statistical assumptions of mediation. We will then
indirect effects as there are mediators. With three medi- present a step-by-step on how to conduct the analyses. Fi-
ators, there are the a1 b1 , a2 b2 , and a3 b3 pathways using nally, we will interpret the results as if we are presenting
M1 , M2 , and M3 respectively. This model is useful since the findings in a research article.
it allows for a more complex assessment of the processes
Example Mediation Analyses
through which X affects Y . For example, anxiety sensitiv-
ity is a multidimensional construct assessing three related To illustrate a mediation analysis, we will be using data
but distinct beliefs: the belief that bodily sensations are from a study examining anxiety sensitivity. Briefly, 297 un-
life-threatening (Physical Concerns), the belief that difficul- dergraduate students from the University of Ottawa partic-
ties concentrating are dangerous (Cognitive Concerns), and ipated in an online study for course credit. One participant
the belief that others will reject or laugh at observable anx- did not identify a clear gender and was therefore elimi-

The Quantitative Methods for Psychology

2 150
 ¦ 2017 Vol. 13 no. 3

nated considering our interest in gender differences. One selected options to ensure that you can redo your analyses
case with missing data was also deleted listwise, creating at a later time.
a final sample of 295 participants consisting of 148 women
Simple Mediation
and 147 men. Since the statistical analyses themselves are
of current focus, further sociodemographic characteristics Before running the analysis, we must first examine our
will not be discussed. All study procedures were in accor- variables to determine if mediation is appropriate. You
dance with the Research Ethics Board of the University of may have noticed that gender, our independent variable,
Ottawa and with the 1964 Helsinki declaration and its later is dichotomous. Thankfully, it is completely correct to con-
amendments. duct a mediation analysis with a dichotomous independent
Following informed consent, participants completed variable (Hayes, 2013). In fact, the interpretation that we
the UPPS-P (negative Urgency, Premeditation, Persever- will be able to make of the coefficients will be even more
ance, Sensation seeking, Positive urgency), a questionnaire informative, as we will see later on when we examine our
assessing impulsive behaviour (Lynam, Smith, Whiteside, results. Our other variables, anxiety sensitivity and sensa-
& Cyders, 2006). The UPPS-P has five subscales; Sensation tion seeking, are on a Likert-type scale, which should lend
Seeking, Negative Urgency, Positive Urgency, (lack of) Pre- itself well to regression.
meditation, and (lack of) Perseverance. For the current
Assumptions.
study, we were interested in the Sensation Seeking sub-
scale (SSs), which conceptualizes sensation seeking as a We first need to consider statistical assumptions. We will
disposition to impulsive behaviour (Cyders, 2013). provide examples of assumption testing using our data (for
Participants also completed the Anxiety Sensitivity more details, see Field, 2013; Hayes, 2013).
Index-3 (ASI-3; Taylor et al., 2007). The ASI-3 has three sub-
scales measuring to Physical Concerns (i.e., the belief that Linearity. In regression analysis, the relationship be-
bodily sensations are life-threatening; ASI.PHY), Cogni- tween X and Y should be linear to minimize error (Hayes,
tive Concerns (i.e., the belief that difficulties concentrating 2013). To our knowledge, no guidelines suggest a way of
are dangerous; ASI.COG), and Social Concerns (i.e., the assessing overall model linearity in mediation. However,
belief that others will reject or laugh at observable anxi- a mediation can be broken down into simple and multiple
ety symptoms; ASI.SOC; Taylor et al., 2007). Therefore, regressions, which each need to fulfil the assumption. The
the supplemental dataset includes the following variables: indirect effect should also be linear, which means that its
gender, ASI.TOT (ASI-3 total scale score), ASI.PHY, constituting effects (a and b) need to be linear. To exam-
ASI.COG, ASI.SOC, and SSs.1 ine these criteria for a simple mediation, you need to plot
In the introduction, we wondered if gender influences residuals against predicted values in four regressions: X
levels of sensation seeking through its effect on anxiety predicting Y (c), X predicting M (a), M predicting Y (b),
sensitivity. Considering that we have a total scale and three and X and M predicting Y (combined linearity of b and
subscale scores for anxiety sensitivity, we could go about c0 ). Should a or b be nonlinear, Hayes and Preacher (2010)
testing the mediating effect of anxiety sensitivity in many outline procedures for determining and testing indirect ef-
different ways. The option that appears the most appropri- fects.
ate is to begin by conducting a simple mediation using the We will follow this procedure by running a series of re-
ASI-3 total scale as our mediator. If this analysis provides gressions (Analyze > Regression > Linear). For
support for our hypothesis, we could then do a parallel me- the sake of time and space, we will only present results re-
diation with the three subscale scores to determine which lating to the X and M predicting Y regression (note: all
dimension of anxiety sensitivity is driving the mediation. other relationships respected the assumptions). First, en-
To conduct our analyses, we will be using version 23 ter gender and ASI.TOT as the independent variables
of SPSS (IBM Corp., 2015) and the PROCESS macro for SPSS and SSs as the outcome variable. In the Save tab, select
(Hayes, 2013), which uses a regression-based approach to the standardized residuals. Then, in the Plots tab, select the
mediation. This macro can be downloaded at the follow- standardized regression residuals (ZRESID) for the Y axis
ing address: http://processmacro.org/download.html. As a and standardized predicted values (ZPRED) for the X axis.
side-note, it is important to know that the PASTE option in Optionally, after you have run the analysis, you may add
SPSS (which allows you to save your syntax, or method, a Loess curve to the scatterplot by double-clicking on the
for later) cannot be used in the PROCESS macro. There- plot, and then going under Elements > Fit Line at
fore, we suggest writing down all the steps taken and the Total > Loess > Apply. Loess fits a non-parametric
curve that represents the relationship between variables
1
In the discussion, we will briefly discuss serial mediation. Therefore, we also included the Negative Urgency subscale (NUs) of the UPPS-P in the

2
dataset to provide readers with the possibility of conducting a serial mediation.

The Quantitative Methods for Psychology 151

 ¦ 2017 Vol. 13 no. 3

Figure 3 Checking the linearity and homoscedasticity assumptions using the multiple regression standardized predicted
and residual values: The influence of gender and anxiety sensitivity on sensation seeking.

(Jacoby, 2000). As can be seen in Figure 3, the regression from the error of all other cases (Hayes, 2013). This is es-
appears fairly linear since the Loess curve centers close to pecially relevant for studies using cluster sampling proce-
zero along the entire X axis. dures or dyadic data research (Hayes, 2013), wherein cases
Homoscedasticity. Estimation error should be relatively may be related to each other on the outcome because they
equal across all predicted Y values. If it varies, then we share some characteristics or context. In the presence of
have heteroscedasticity, which affects the standard error nonindependence of observations, the standard error of
of the regression coefficients (Hayes, 2013). To check ho- the regression coefficients could be either smaller or larger
moscedasticity, return to the same plot that we created to than what it should be (Hayes, 2013). Only knowledge of
examine linearity, but this time look for consistency in ver- one’s own data collection method will allow a person to
tical range across the X axis. In other words, see if the data determine if their data meets the independence assump-
spreads on the Y axis consistently and equally through- tion. Given that we sampled our participants from an un-
out the plot, resembling a rectangle. Our data in Figure dergraduate participant pool at the same university, it is
3 shows a relatively constant vertical range. unlikely that we have underlying common characteristics
Normality of estimation error. Estimation error should that would compromise the independence of our estima-
be normally distributed (Hayes, 2013). To examine this tion error.
assumption, we can create a Q-Q plot with the residuals
The analysis.
we saved from the regression by going under Analyze >
Descriptive Statistics > Q-Q Plots. The re- Now that we have determined that our data respects mul-
sulting plot for the X (gender) and M (anxiety sensitivity) tiple regression assumptions relatively well, we can con-
predicting Y (sensation seeking) multiple regression can duct our simple mediation. Once the PROCESS macro has
be seen in Figure 4. Our data fit well with the diagonal line, been installed on SPSS, you may access it by going un-
indicating normality. When there are minor violations to der Analyze > Regression > PROCESS. You will
this assumption, the results of the analysis should not be be greeted with the window shown in Figure 5. You will
affected unless the sample size is very small (Hayes, 2013). need to use the center arrows to guide your variables to the
Independence of observations. The error associated appropriate boxes. We want gender as the independent

2
with each data point (i.e., one case) should be independent variable, ASI.TOT as the M Variable, and SSs as the out-

The Quantitative Methods for Psychology 152

 ¦ 2017 Vol. 13 no. 3

Figure 4 Checking the normality of estimation error assumption using the multiple regression standardized residuals:
The influence of gender and anxiety sensitivity on sensation seeking.

come variable. On the left-hand side, you will see various of Ottawa. The computer will take a random bootstrap
options. First, Model Number refers to the type of analy- sample of observations (of size n) within this mini popula-
sis we want to run. The default setting is 4, which is the tion with replacement. This means that some observations
one we want since it is for simple and parallel mediation may be selected multiple times or not at all within each
(for the use of other model numbers, the interested reader bootstrap sample. The computer will repeat this process
can refer to Hayes, 2013). Second, we must decide which thousands of time, each bootstrap sample being slightly
type of bootstrapping to use for the indirect effects. This different. Hayes (2013) recommends and uses 10,000 boot-
technique merits explanation before we continue with the strap samples. The analyses are then run on all these boot-
analysis. strap samples to obtain the desired statistic for every one of
Bootstrapping is an alternative way to perform null hy- them. Hence, if there are 10,000 bootstrap samples, there
pothesis testing that can be applied to the test of the in- will be 10,000 computed indirect effects. These effects are
direct effect (ab) to determine if it is different from zero then placed in ascending order to determine the lower and
(Hayes, 2013). When using null hypothesis testing for an upper bounds of the confidence interval (CI), usually a 95%
indirect effect, one assumption is that ab is normally dis- CI. An example 95% CI may resemble −0.023 to 0.265,
tributed (i.e., if we were to redo the study multiple times, −0.416 to −0.146, or 0.024 to 0.137. In the first case, the CI
determining ab for each, the distribution of ab should be includes zero, which would indicate that the indirect effect
normal). Bootstrapping does not assume that ab is normal is not significant because zero is in the realm of possible
and therefore is preferable since we cannot really know values for the effect. In the latter two cases, the CIs do not
the shape of the indirect effect’s distribution in the popula- include zero and are entirely below or above zero. We can
tion. Bootstrapping is a resampling method (Hayes, 2013). therefore say with 95% confidence that the indirect effect
The goal is to construct a confidence interval around the is negative in the second case and positive in the third case.
examined effect (in our case, around the indirect effect For our analysis, we will use 10,000 bootstrap samples
ab). To achieve this, the current sample (of size n) is used and select the Bias Corrected method of bootstrapping. Ex-
as a mini population. We have 295 participants, who we amining the pros of bias-corrected bootstrap CIs over other
believe adequately represent the population from which methods is beyond the scope of this paper; the interested
they came, the undergraduate students of the University reader may refer to Hayes (2013). Lastly, we will select a

The Quantitative Methods for Psychology

2 153
 ¦ 2017 Vol. 13 no. 3

Figure 5 Screen capture of the completed opening PROCESS procedure window in SPSS (version 23).

95% confidence level. If we had covariates, we could also puter must do to compute the bootstrap CIs. The output
select which variables we would like them to influence (M from PROCESS should resemble Listing 1 given at the end
and/or Y ). Since we do not have any covariates, it does not of the article.
matter which option is selected. The screen should resem- Several key points should be noted in this output. Lines
ble the window shown in Figure 5. 8 to 14 remind the researcher of the model (number 4),
Next, under the Options tab, we will select Ordinary the variables, the sample size, the number of bootstrap CIs,
Least Square/Maximum Likelihood (OLS/ML) confidence and the level of confidence. Lines 16 to 39 show the regres-
intervals to obtain CIs for effects other than the indirect ef- sions associated with the mediator and the outcome vari-
fect which uses bootstrapping (optional if you do not wish able. It is these coefficients we will be using to construct a
to present these), Effect size, and Total effect model (see diagram to represent our mediation. We will return to this
Figure 6). Additionally, PROCESS by default only allows for in the interpretation of our results. Also, all coefficients in
variables that are eight characters or less to be entered in this output are unstandardized ones. If you would like to
the analysis. If your variable names are longer than eight obtain standardized coefficients, you may transform your
characters, you may go in the Long Name tab and allow variables into Z scores before entering them in the medi-
long variable names. ation. Considering that our independent variable is gen-
We are now ready to click OK on the main window. It der, it would have been inappropriate to standardize as it
may take some time before results appear in the output would have produced coefficients without any real mean-
window. This is normal considering the work the com- ing.

The Quantitative Methods for Psychology

2 154
 ¦ 2017 Vol. 13 no. 3

Figure 6 Screen capture of the Options menu of the PROCESS macro in SPSS (version 23).

Continuing, Lines 42 to 51 present the total effect model we can say here that men scored 0.029 points higher than
that we requested in the options. As previously explained, women on sensation seeking as a result of the indirect ef-
the total effect (c) is the result you would obtain by simply fect through anxiety sensitivity. However, what does 0.029
regressing sensation seeking on gender without considera- mean? Is it a small or a large increase in sensation seek-
tion of the mediator. You may notice that these values are ing? The bootstrap CI is completely above 0, so we know
identical to those in the next section of the output. The in- with 95% confidence that the indirect effect is positive, but
teresting information in the Total Effect Model section is we still do not know if the increase in sensation seeking is
the R-squared, which tells us how much variance in sen- notable. In addition, scores on the SSs vary from 1 to 4,
sation seeking our model can explain. In our case, we are but these are arbitrary values that carry no real meaning.
explaining 4% with our model, which indicates that many Partially standardized indirect effect. It is possible to
other factors are influencing sensation seeking. change part or all of the metrics used to describe the indi-
Direct effect. Finally, Lines 52 to 85 give information on rect effect to add to its interpretation. The partially stan-
the direct and indirect effects. Recall that, in our dataset, dardized indirect effect expresses ab using the original
our X is coded by a unit difference (women coded as 0; metric of X , but the standard deviation of Y (Hayes, 2013).
men coded as 1), which means that the effects can be in- In our case, men scored 0.048 standard deviations higher
terpreted as mean differences between women and men on sensation seeking than women as the result of the indi-
(Hayes, 2013). It appears that men scored 0.201 points rect mechanism of anxiety sensitivity.
higher on sensation seeking than did women when levels Completely standardized indirect effect. It is possible to
of anxiety sensitivity were kept constant. However, if we even further standardize the indirect effect by expressing
had coded men as 0 and women as 1, we would then say ab using the standard deviations of both X and Y (Hayes,
that women scored 0.201 points lower on sensation seek- 2013). As such, we could say that an increase by one stan-
ing than did men when levels of anxiety sensitivity were dard deviation on gender produced an increase of 0.024
kept constant. standard deviations on sensation seeking through the indi-
Indirect effect. You will notice that the output presents rect effect of anxiety sensitivity. However, considering that
three versions of the indirect effect: the indirect effect, the our X variable is dichotomous, this makes little sense. In-
partially standardized indirect effect, and the completely deed, the values we associated with men and with women
standardized indirect effect. The indirect effect, similar to were arbitrary. An increase in one standard deviation
the direct effect, is presented using the metrics of X and in gender therefore is uninterpretable. Completely stan-
Y , such that for every increase in one unit on X , there dardized indirect effect can be used with other types of

2
is a change of ab units on Y (Hayes, 2013). For example, variables, but it is not appropriate for dichotomous ones

The Quantitative Methods for Psychology 155

 ¦ 2017 Vol. 13 no. 3

Figure 7 The mediating effect of anxiety sensitivity in the relationship between gender and sensation seeking. Notes:
*p < .05, **p < .01, # p < .001; All presented effects are unstandardized; a is effect of gender on anxiety sensitivity,
women are coded as 0 and men as 1; b is effect of anxiety sensitivity on sensation seeking; c0 is direct effect of gender on
sensation seeking; c is total effect of gender on sensation seeking.

(Hayes, 2013). Thus, for our analyses we may use either the these dimensions drive the mediation more than the oth-
unstandardized indirect effect or the partially standard- ers, or if all three contribute to it. We can answer this
ized indirect effect. The important thing will be to specify question using parallel mediation. Recall that mediators
which one we choose in the text or in a figure note. in a parallel mediation are allowed to correlate, but not to
causally influence each other. Considering that all three
Interpretation of results.
dimensions are assessed in one questionnaire and that we
To finish the section on simple mediation, we will present have no theoretical reason to believe that one dimension
the results of our analysis as we would in a manuscript. All would lead to another, parallel mediation is appropriate.
information comes from the PROCESS output. On the other hand, if we believed that one dimension led
to another then serial mediation (see the conclusion for
Results from a simple mediation analysis indi- a more in-depth description of serial mediation) would
cated that gender is indirectly related to sensa- be the appropriate choice. With parallel mediation, we
tion seeking through its relationship with anx- can test each proposed mediator while accounting for the
iety sensitivity. First, as can be seen in Fig- shared variance between them (Hayes, 2013). However,
ure 7, men reported less anxiety sensitivity mediators that are too highly correlated may create mul-
than women (a = −3.584, p = .042), and ticollinearity, which affects the estimation of their partial
lower reported anxiety sensitivity was subse- relationships with the outcome variable (Hayes, 2013).
quently related to more sensation seeking (b = Assumptions. We verified the assumptions using the
−0.008, p =< .001). A 95% bias-corrected same methods we used for simple mediation, only this time
confidence interval based on 10,000 bootstrap we conducted seven additional regressions (i.e., X [gen-
samples indicated that the indirect effect (ab = der] predicting each mediator [the three anxiety sensitivity
0.029) was entirely above zero (0.003 to 0.074). dimensions]; each mediator predicting Y [sensation seek-
Moreover, men reported greater sensation seek- ing]; X and all three mediators predicting Y ; note: we al-
ing even after taking into account gender’s in- ready had regressed Y on X for the simple mediation). We
direct effect through anxiety sensitivity (c0 = noted no major assumption violations.
0.201, p = .003).
The analysis. To conduct our parallel mediation, we will
go to the same PROCESS window as before, which should
Parallel mediation
resemble Figure 8. The order in which you place the medi-
The results of our simple mediation suggest that anxiety ators will not affect the results of the parallel mediation.
sensitivity mediates the relationship between gender and Next, under the Options tab, we will select OLS/ML con-
sensation seeking. However, anxiety sensitivity has three fidence intervals, Effect size, Total effect model, and Com-
dimensions: Physical, Cognitive, and Social Concerns (Tay- pare indirect effects. Once again, if your variable names

2
lor et al., 2007). It would be interesting to know if any of are longer than eight characters, you may also go in the

The Quantitative Methods for Psychology 156

 ¦ 2017 Vol. 13 no. 3

Figure 8 Screen capture of the completed opening PROCESS procedure window in SPSS (version 23).

Long Name tab and allow long variable names. Click OK sation seeking than did women when levels of the three
on the main PROCESS window to obtain the output in List- dimensions of anxiety sensitivity were kept constant. In
ing 2. addition, if you examine the indirect effect section of the
This output is very similar to the simple mediation output, you will see the total indirect effect, which is the
analysis, with some key differences. First, there is a sec- sum of all indirect effects. This statistic is often not of inter-
tion that presents the unstandardized coefficients for each est because we usually want to look at the specific indirect
“outcome”, that is for each mediator and the final outcome effects (Hayes, 2013). Indeed, it appears that the only indi-
variable (sensation seeking). Then, we have once again rect effect that is different from zero with 95% confidence
the total effect model, which is the same as for the sim- is the Physical Concerns subscale of the ASI-3, as evidenced
ple mediation we ran earlier since, if you recall, the total by the bootstrap CI for Physical Concerns that is completely
effect is the result obtained by simply regressing sensation above zero. Thus, we can say here that men scored 0.058
seeking on gender. Next, we can see information on the points higher than women on sensation seeking as a result
direct effect of gender on sensation seeking. The results of the indirect effect through the physical concerns of anx-
are slightly different than they were for the simple medi- iety sensitivity, holding all other mediators constant.
ation because we have different mediators entered in the Next, you will find pairwise comparisons between the
model (i.e., three separate mediators versus one combined specific indirect effects (denoted by C1 , C2 , and C3 ). The
mediator) that account for a different proportion of the to- legend for these contrasts is at the end of the output. It
tal effect. Here, men scored 0.172 points higher on sen- should be noted, however, that these comparisons do not

The Quantitative Methods for Psychology

2 157
 ¦ 2017 Vol. 13 no. 3

Figure 9 The mediating effect of three anxiety sensitivity dimensions in the relationship between gender and sensation
seeking. Notes: *p < .05, **p < .01, #p < .001; All presented effects are unstandardized; an is effect of gender on anxiety
sensitivity dimensions, women are coded as 0 and men as 1; bn is effect of anxiety sensitivity dimensions on sensation
seeking; c0 is direct effect of gender on sensation seeking; c is total effect of gender on sensation seeking.

allow to test if one indirect effect is larger than another; 9, men reported less fear of physiological sen-
they simply tell us if the effects are different (Hayes, 2013). sations than women (a1 = −2.021, p = .004),
For example, if one indirect effect is .20 and another is and lower reported fear of physiological sen-
−.20, the contrast would probably identify these effects as sations was subsequently related to more sen-
being different, which they are because one is positive and sation seeking (b1 = −0.029, p < .001). A
the other is negative even if they are of the same strength. 95% bias-corrected confidence interval based on
However, if both indirect effects are of the same sign, then 10,000 bootstrap samples indicated that the in-
a significant contrast may be interpreted as one effect be- direct effect through fear of physiological sen-
ing larger than the other (Hayes, 2013). Here, we can see sations (a1 b1 = 0.058), holding all other media-
that C1 and C2 are significant. With the help of the legend, tors constant, was entirely above zero (0.017 to
we can determine that the indirect effect through the Phys- 0.132). In contrast, the indirect effects through
ical Concerns subscale is larger than the two other indirect both the Social and the Cognitive Concerns sub-
effects. For other analyses wherein more than one effect scales of anxiety sensitivity were not differ-
is significant, the contrasts may also provide additional in- ent than zero (−0.004 to 0.038 and −0.047
teresting information. to 0.005, respectively; see Figure 9 for the ef-
Interpretation of results. Once again, we will present fects associated with these pathways). More-
the results of our parallel mediation as we would in a over, men reported greater sensation seeking
manuscript. even when taking into account gender’s indirect
effect through all three dimensions of anxiety
Results from a parallel mediation analysis in- sensitivity (c0 = 0.172, p = .011).
dicated that gender is indirectly related to sen-
sation seeking through its relationship with the
Discussion and Conclusion
Physical Concerns subscale of anxiety sensitiv-
ity. This dimension pertains to the fear of phys- Mediation analysis enables researchers to examine the
iological sensations because of the belief that processes through which one variable affects another. Al-
they may have negative consequences and are though complex, modern computers and software make

2
life-threatening. First, as can be seen in Figure conducting mediations much more approachable. Using

The Quantitative Methods for Psychology 158

 ¦ 2017 Vol. 13 no. 3

Figure 10 The serial mediating effect of anxiety sensitivity and negative urgency in the relationship between gender and
sensation seeking. Notes: *p < .05, **p < .01, #p < .001; All presented effects are unstandardized; an is effect of gender
on mediators, women are coded as 0 and men as 1; bn is effect of mediators on sensation seeking; c0 is direct effect of
gender on sensation seeking; c is total effect of gender on sensation seeking; d is effect of anxiety sensitivity on negative
urgency.

the PROCESS macro for SPSS (Hayes, 2013), we were able sensation seeking mediates the relationship between gen-
to determine that anxiety sensitivity explains part of the der and anxiety sensitivity. We chose the former because
relationship between gender and sensation seeking. More we based ourselves in a cognitive-behavioural framework,
specifically, men reported less anxiety sensitivity than which suggests that beliefs (anxiety sensitivity) influence
women, and less anxiety sensitivity was associated with behaviour (sensation seeking).
more sensation seeking. We did not stop there, however. But what if we want to see if this sequence of elements
We then wondered which dimension of anxiety sensitiv- truly operates in this causal order? We could try running
ity was driving the mediation. Using parallel mediation, the mediation analysis with the mediator and the outcome
we found that the Physical Concerns dimension of anxi- in both possible configurations and see if the indirect effect
ety sensitivity was the only significant mediator of the re- holds in each model. We could also conduct a longitudinal
lationship. Men reported less fear of physiological sensa- study where we would measure anxiety sensitivity some
tions than women, which in turn led to more sensation time before sensation seeking to establish evidence of tem-
seeking. Considering the three dimensions of anxiety sen- poral precedence. However, to really determine if anxiety
sitivity (Physical, Social, and Cognitive Concerns), it makes sensitivity causally influences sensation seeking, we would
sense that it is the Physical Concerns that played the great- need to conduct an experiment where we would manipu-
est role. Sensation seeking produces many physiological late anxiety sensitivity (by either inducing or inhibiting it
sensations. A person who fears these sensations would by, say, telling people that anxiety-related sensations are
likely score high on the Physical Concerns subscale and either dangerous or harmless) and see how this manipu-
also engage in less activities that produce strong physio- lation affects level of sensation seeking on a standardized
logical sensations. task. Trait levels of sensation seeking would also need to
In the introduction, we alluded to the fact that be statistically controlled for in this experiment.
regression-based analyses (including mediation) imply Furthermore, simple and parallel mediation are not the
causality, but that it is in fact a false sense of causality. Re- only possible mediation models. In parallel mediation, we
searchers choose which variable is the outcome and which saw that the mediators are allowed to correlate but not to
is the predictor of this outcome, but the analyses could causally influence each other. If it is believed that one me-
also be run the other way around. In our simple medi- diator leads to another, then a serial mediation would be
ation example, we decided that anxiety sensitivity would the preferred model (see Figure 10). Briefly, in serial medi-
be the mediator in the relationship between gender and ation the a or b pathway is mediated by a second mediator.
sensation seeking. However, we could have also tested if There are then indirect effects through each of the medi-

The Quantitative Methods for Psychology

2 159
 ¦ 2017 Vol. 13 no. 3

ators (a1 b1 and a2 b2 ) and an indirect effect through both Hayes, A. F. & Preacher, K. J. (2010). Quantifying and
mediators (a1 db2 ). Essentially, it is like conducting a par- testing indirect effects in simple mediation models
allel mediation, but with an added pathway that tests the when the constituent paths are nonlinear. Multivari-
causal relationship between the mediators. For instance, ate Behavioral Research, 45(4), 627–660. doi:10.1080/
maybe anxiety sensitivity affects a person’s tendency to 00273171.2010.498290
act rashly when faced with negative emotions, a construct Hofmann, S. G., Asmundson, G. J. G., & Beck, A. T. (2013).
known as negative urgency (Cyders & Smith, 2007). Essen- The science of cognitive therapy. Behavior Therapy,
tially, a person who is afraid of the sensations associated 44(2), 199–212. doi:10.1016/j.beth.2009.01.007
with negative emotions may act impulsively in order to IBM Corp. (2015). IBM SPSS Statistics for Windows (version
escape these sensations. Negative urgency could then af- 23.0). Armonk, NY: IBM Corp.
fect levels of sensation seeking. Figure 10 shows this ex- Jacoby, W. G. (2000). Loess: a nonparametric, graphical tool
ample of serial mediation.2 The material we covered in for depicting relationships between variables. Elec-
this tutorial can easily be applied to serial mediation and toral Studies, 19(4), 577–613. doi:10 . 1016 / S0261 -
you may practice this technique using the supplemental 3794(99)00028-1
dataset (note: in PROCESS, the model number for serial Lynam, D. R., Smith, G. T., Whiteside, S. P., & Cyders, M. A.
mediation is Model 6). The interested reader can refer to (2006). The UPPS-P: assessing five personality path-
Hayes (2013) to learn about serial mediation in more detail. ways to impulsive behavior. Technical report. West
This paper presents only the tip of the iceberg regard- Lafayette, IN: Purdue University.
ing mediation. It is based in our previous experiences with MacKinnon, D. P., Fairchild, A. J., & Fritz, M. S. (2007). Medi-
this type of analysis, complemented with information from ation analysis. Annual Review of Psychology, 58, 593–
much more comprehensive resources (e.g., Hayes, 2013). 614. doi:10.1146/annurev.psych.58.110405.085542
We hope it will prove useful to researchers who are new Preacher, K. J. & Hayes, A. F. (2004). SPSS and SAS proce-
to the field or who require a quick refresher. Mediation is dures for estimating indirect effects in simple media-
a great tool to have in a statistical toolbox, whether it be tion models. Behavior Research Methods, Instruments,
to conduct it yourself or to better understand the literature & Computers, 36(4), 717–731. doi:10.3758/BF03206553
that uses the technique. Roth, M. & Hammelstein, P. (2012). The need inventory of
sensation seeking (NISS). European Journal of Psycho-
Authors’ note
logical Assessment, 28(1), 11–18. doi:10 . 1027 / 1015 -
The authors thank Veronika Huta, and Simon G. Beaudry, 5759/a000085
for their comments on assumption testing in mediation. Rucker, D. D., Preacher, K. J., Tormala, Z. L., & Petty, R. E.
(2011). Mediation analysis in social psychology: cur-
References
rent practices and new recommendations. Social &
Cross, C. P., Copping, L. T., & Campbell, A. (2011). Sex dif- Personality Psychology Compass, 5(6), 359–371. doi:10.
ferences in impulsivity: a meta-analysis. Psychologi- 1111/j.1751-9004.2011.00355.x
cal Bulletin, 137(1), 97–130. doi:10.1037/a0021591 Stewart, S. H., Taylor, S., & Baker, J. M. (1997). Gender
Cyders, M. A. (2013). Impulsivity and the sexes: measure- differences in dimensions of anxiety sensitivity. Jour-
ment and structural invariance of the UPPS-P impul- nal of Anxiety Disorders, 11(2), 179–200. doi:10.1016/
sive behavior scale. Assessment, 20(1), 86–97. doi:10. S0887-6185(97)00005-4
1177/1073191111428762 Taylor, S., Zvolensky, M. J., Cox, B. J., Deacon, B., & He-
Cyders, M. A. & Smith, G. T. (2007). Mood-based rash action imberg, R. G. (2007). Robust dimensions of anxiety
and its components: positive and negative urgency. sensitivity: development and initial validation of the
Personality and Individual Differences, 43(4), 839–850. anxiety sensitivity index-3. Psychological Assessment,
doi:10.1016/j.paid.2007.02.008 19(2), 176–188. doi:10.1037/1040-3590.19.2.176
Field, A. (2013). Discovering statistics using IBM SPSS statis- Zuckerman, M. (1994). Behavioral expressions and biosocial
tics (fourth edition). London, England: SAGE. bases of sensation seeking. New York, NY: Cambridge
Hayes, A. F. (2013). Introduction to mediation, moderation, University Press.
and conditional process analysis. A regression-based
approach. New York, NY: The Guilford Press.

2
95% bias-corrected confidence intervals based on 10,000 bootstrap samples indicated that the indirect effects through fear of physiological sensa-
tions alone and including negative urgency (a1 b1 = 0.042 and a1 db2 = −0.013) were entirely above or below zero (0.004 to 0.098 and −0.032 to
−0.002, respectively). However, the indirect effect through only negative urgency (a2 b2 ) was not different than zero (−0.008 to 0.058), suggesting
that gender influences levels of negative urgency only through its effect on anxiety sensitivity.

The Quantitative Methods for Psychology

2 160
 ¦ 2017 Vol. 13 no. 3

Listing 1: Output from the PROCESS procedure in SPSS (version 23) for the gender, anxiety sensitivity, and sensa-
tion seeking simple mediation analysis.

1 Run MATRIX procedure:

2 ************* PROCESS Procedure for SPSS Release 2.16.1 ******************
3
4 Written by Andrew F. Hayes, Ph.D. www.afhayes.com
5 Documentation available in Hayes (2013). www.guilford.com/p/hayes3
6
7 **************************************************************************
8 Model = 4
9 Y = SSs
10 X = Gender
11 M = ASI.TOT
12
13 Sample size
14 295
15 **************************************************************************
16 Outcome: ASI.TOT
17
18 Model Summary
19 R Rfsq MSE F df1 df2 p
20 ,1188 ,0141 225,9784 4,1923 1,0000 293,0000 ,0415
21
22 Model
23 coeff se t p LLCI ULCI
24 constant 26,1351 1,2357 21,1506 ,0000 23,7032 28,5671
25 Gender -3,5841 1,7505 -2,0475 ,0415 -7,0292 -,1390
26
27 **************************************************************************
28 Outcome: SSs
29
30 Model Summary
31 R R-sq MSE F df1 df2 p
32 ,2822 ,0796 ,3244 12,6292 2,0000 292,0000 ,0000
33
34 Model
35 coeff se t p LLCI ULCI
36 constant 2,8641 ,0744 38,4869 ,0000 2,7176 3,0105
37 ASI.TOT -,0080 ,0022 -3,6364 ,0003 -,0124 -,0037
38 Gender ,2012 ,0668 3,0127 ,0028 ,0698 ,3327
39
40 ************************** TOTAL EFFECT MODEL ****************************
41 Outcome: SSs
42
43 Model Summary
44 R R-sq MSE F df1 df2 p
45 ,1948 ,0379 ,3379 11,5530 1,0000 293,0000 ,0008
46
47 Model
48 coeff se t p LLCI ULCI
49 constant 2,6537 ,0478 55,5384 ,0000 2,5597 2,7478

The Quantitative Methods for Psychology

2 161
 ¦ 2017 Vol. 13 no. 3

50 Gender ,2301 ,0677 3,3990 ,0008 ,0969 ,3633

51
52 ***************** TOTAL, DIRECT, AND INDIRECT EFFECTS ********************
53
54 Total effect of X on Y
55 Effect SE t p LLCI ULCI
56 ,2301 ,0677 3,3990 ,0008 ,0969 ,3633
57
58 Direct effect of X on Y
59 Effect SE t p LLCI ULCI
60 ,2012 ,0668 3,0127 ,0028 ,0698 ,3327
61
62 Indirect effect of X on Y
63 Effect Boot SE BootLLCI BootULCI
64 ASI.TOT ,0288 ,0177 ,0027 ,0738
65
66 Partially standardized indirect effect of X on Y
67 Effect Boot SE BootLLCI BootULCI
68 ASI.TOT ,0488 ,0296 ,0043 ,1226
69
70 Completely standardized indirect effect of X on Y
71 Effect Boot SE BootLLCI BootULCI
72 ASI.TOT ,0244 ,0148 ,0022 ,0614
73
74 Ratio of indirect to total effect of X on Y
75 Effect Boot SE BootLLCI BootULCI
76 ASI.TOT ,1254 ,1295 ,0085 ,4062
77
78 Ratio of indirect to direct effect of X on Y
79 Effect Boot SE BootLLCI BootULCI
80 ASI.TOT ,1434 ,5763 ,0083 ,6730
81
82 R-squared mediation effect size (R-sq_med)
83 Effect Boot SE BootLLCI BootULCI
84 ASI.TOT ,0093 ,0066 ,0010 ,0294
85
86 ******************** ANALYSIS NOTES AND WARNINGS *************************
87
88 Number of bootstrap samples for bias corrected bootstrap confidence intervals:
89 10000
90
91 Level of confidence for all confidence intervals in output:
92 95,00
93
94 NOTE: Kappa-squared is disabled from output as of version 2.16.
95
96 ------ END MATRIX -----

Listing 2: Output from the PROCESS procedure in SPSS (version 23) for the gender, anxiety sensitivity, and sensa-
tion seeking parallel mediation analysis.

1 Run MATRIX procedure:

The Quantitative Methods for Psychology

2 162
 ¦ 2017 Vol. 13 no. 3

2 ************* PROCESS Procedure for SPSS Release 2.16.1 ******************

3
4 Written by Andrew F. Hayes, Ph.D. www.afhayes.com
5 Documentation available in Hayes (2013). www.guilford.com/p/hayes3
6
7 **************************************************************************
8 Model = 4
9 Y = SSs
10 X = Gender
11 M1 = ASI.PHY
12 M2 = ASI.SOC
13 M3 = ASI.COG
14
15 Sample size
16 295
17 **************************************************************************
18 Outcome: ASI.PHY
19
20 Model Summary
21 R R-sq MSE F df1 df2 p
22 ,1657 ,0274 36,4396 8,2676 1,0000 293,0000 ,0043
23
24 Model
25 coeff se t p LLCI ULCI
26 constant 7,8919 ,4962 15,9047 ,0000 6,9153 8,8685
27 Gender -2,0211 ,7029 -2,8753 ,0043 -3,4046 -,6377
28
29 **************************************************************************
30 Outcome: ASI.SOC
31
32 Model Summary
33 R R-sq MSE F df1 df2 p
34 ,0867 ,0075 29,6116 2,2199 1,0000 293,0000 ,1373
35
36 Model
37 coeff se t p LLCI ULCI
38 constant 11,2162 ,4473 25,0753 ,0000 10,3359 12,0965
39 Gender -,9441 ,6337 -1,4899 ,1373 -2,1912 ,3030
40
41 **************************************************************************
42 Outcome: ASI.COG
43
44 Model Summary
45 R R-sq MSE F df1 df2 p
46 ,0524 ,0027 34,9672 ,8078 1,0000 293,0000 ,3695
47
48 Model
49 coeff se t p LLCI ULCI
50 constant 7,0270 ,4861 14,4568 ,0000 6,0704 7,9837
51 Gender -,6189 ,6886 -,8988 ,3695 -1,9740 ,7363
52
53 **************************************************************************

The Quantitative Methods for Psychology

2
163
 ¦ 2017 Vol. 13 no. 3

54 Outcome: SSs
55
56 Model Summary
57 R R-sq MSE F df1 df2 p
58 ,3187 ,1016 ,3188 8,1959 4,0000 290,0000 ,0000
59
60 Model
61 coeff se t p LLCI ULCI
62 constant 2,8771 ,0829 34,7125 ,0000 2,7140 3,0402
63 ASI.PHY -,0289 ,0084 -3,4265 ,0007 -,0455 -,0123
64 ASI.SOC -,0074 ,0075 -,9909 ,3225 -,0222 ,0073
65 ASI.COG ,0125 ,0088 1,4223 ,1560 -,0048 ,0299
66 Gender ,1724 ,0671 2,5693 ,0107 ,0403 ,3045
67
68 ************************** TOTAL EFFECT MODEL ****************************
69 Outcome: SSs
70
71 Model Summary
72 R R-sq MSE F df1 df2 p
73 ,1948 ,0379 ,3379 11,5530 1,0000 293,0000 ,0008
74
75 Model
76 coeff se t p LLCI ULCI
77 constant 2,6537 ,0478 55,5384 ,0000 2,5597 2,7478
78 Gender ,2301 ,0677 3,3990 ,0008 ,0969 ,3633
79
80 ***************** TOTAL, DIRECT, AND INDIRECT EFFECTS ********************
81
82 Total effect of X on Y
83 Effect SE t p LLCI ULCI
84 ,2301 ,0677 3,3990 ,0008 ,0969 ,3633
85
86 Direct effect of X on Y
87 Effect SE t p LLCI ULCI
88 ,1724 ,0671 2,5693 ,0107 ,0403 ,3045
89
90 Indirect effect of X on Y
91 Effect Boot SE BootLLCI BootULCI
92 TOTAL ,0577 ,0235 ,0199 ,1152
93 ASI.PHY ,0584 ,0278 ,0168 ,1320
94 ASI.SOC ,0070 ,0096 -,0044 ,0384
95 ASI.COG -,0078 ,0114 -,0466 ,0053
96 (C1) ,0514 ,0288 ,0090 ,1268
97 (C2) ,0662 ,0362 ,0149 ,1665
98 (C3) ,0148 ,0169 -,0069 ,0665
99
100 Partially standardized indirect effect of X on Y
101 Effect Boot SE BootLLCI BootULCI
102 TOTAL ,0975 ,0386 ,0341 ,1897
103 ASI.PHY ,0987 ,0460 ,0287 ,2168
104 ASI.SOC ,0119 ,0161 -,0074 ,0651
105 ASI.COG -,0131 ,0191 -,0772 ,0092

The Quantitative Methods for Psychology

2
164
 ¦ 2017 Vol. 13 no. 3

106
107 Completely standardized indirect effect of X on Y
108 Effect Boot SE BootLLCI BootULCI
109 TOTAL ,0488 ,0193 ,0172 ,0952
110 ASI.PHY ,0495 ,0230 ,0144 ,1086
111 ASI.SOC ,0059 ,0080 -,0037 ,0325
112 ASI.COG -,0066 ,0096 -,0386 ,0046
113
114 Ratio of indirect to total effect of X on Y
115 Effect Boot SE BootLLCI BootULCI
116 TOTAL ,2507 ,5783 ,0811 ,6446
117 ASI.PHY ,2539 ,9142 ,0679 ,7241
118 ASI.SOC ,0305 ,0607 -,0230 ,1913
119 ASI.COG -,0337 ,3114 -,2309 ,0288
120
121 Ratio of indirect to direct effect of X on Y
122 Effect Boot SE BootLLCI BootULCI
123 TOTAL ,3346 7,9496 ,0838 1,5885
124 ASI.PHY ,3388 9,9316 ,0708 1,8026
125 ASI.SOC ,0407 1,6297 -,0330 ,4503
126 ASI.COG -,0450 1,4064 -,6041 ,0410
127
128 Specific indirect effect contrast definitions
129 (C1) ASI.PHY minus ASI.SOC
130 (C2) ASI.PHY minus ASI.COG
131 (C3) ASI.SOC minus ASI.COG
132
133 ******************** ANALYSIS NOTES AND WARNINGS *************************
134
135 Number of bootstrap samples for bias corrected bootstrap confidence intervals:
136 10000
137
138 Level of confidence for all confidence intervals in output:
139 95,00
140
141 ------ END MATRIX -----

Open practices

The Open Data badge was earned because the data of the experiment(s) are available on the journal’s web site.

Citation

Kane, L. & Ashbaugh, A. R. (2017). Simple and parallel mediation: A tutorial exploring anxiety sensitivity, sensation seek-
ing, and gender. The Quantitative Methods for Psychology, 13(3), 148–165. doi:10.20982/tqmp.13.3.p148

Copyright © 2017, Kane and Ashbaugh. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC
BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original
publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not
comply with these terms.

Received: 29/06/2017 ∼ Accepted: 20/07/2017

The Quantitative Methods for Psychology

2 165