The first session is on mediation and this is the agenda for today.

Well we start with an overview of mediation.

What is mediation?

Why do we need mediation?

And then we will talk about the traditional and current tests for mediation.

So, what is mediation?

Mediation, when we are theorizing about mediation models,

we look at the mechanism,

we look at the process

through which our independent variable influences our dependent variable.

As an example how job meaningfulness influences job performance.

What is the mechanism?

How job meaningfulness influences job performance.

That's why we need mediation models.

And usually when you see mediation models this is

the figure that you will see in the paper.

So you'll have the independent variable leading

to the dependent variable via the mediator.

Three different paths in the model, A,

B and C. The most traditional way

to test for mediation came from Baron and Kenny.

They published a paper in 1986 and it described how we should test for mediation.

That was a huge advancement for us because it gave

us a consistent way to test for mediation,

to test for mediating models.

According to them the,

first step is to have a significant relationship between our independent

variable and our dependent variable meaningfulness and job performance.

If you have that significant relationship,

you move to our step two and then you needed to find

a significant relationship between your independent variable and your mediator.

In our example job meaningfulness and job engagement.

If you have that if you have a significant relationship you move to step three.

And now you are looking for the relationship between your mediator

and your dependent variable engagement and job performance.

If that's significant that's great because you have

initial support for your mediating model.

What do you have to do next is to run

a model in which you'll have your independent variable,

your dependent variable and control for mediator.

What you are looking for is the difference in

that co-efficient between the independent variable and dependent variable.

If that relationship goes to zero you'll have full mediation.

If that relationship between the independent

variable and the dependent variable decreases,

you'll have partial mediation.

Well, this language for mediation and partial mediation has not been

used very often since the most current models,

the most current tests were proposed

but if you are reading papers that were published up to let's

say 2012 you may find

this language embedded in the papers so it's good to know what that really means.

Once you did that,

once you ran these four different steps,

you went through all these four different steps and ran four different models,

it's time to do a Sobel test according to Baron and Kenny.

The Sobel test was proposed to check if

its indirect effect was significant or not and this is another language indirect effects.

When we are running mediation models the other way to say that,

to describe that is to say that well the indirect effect of our independent

variable on our dependent variable via the mediator is significant or not.

So keep in mind those different languages, mediator's,

interactive facts and once you have an indirect effect model you'll do a Sobel test.

The problem with a Sobel test is that one of

the assumptions is that the indirect effect co-efficient has a normal distribution.

But we know now that based on

Monte-Carlo simulations that interactive facts they are not normally distributed.

So there is this violation of this assumption.

So I strongly recommend you to not adopt Sobel tests when you

are conducting mediation models or when you are looking for indirect facts.

There is a better way to do that.

And the best way or the better way is running a bootstrapping analysis.

With bootstrapping analysis.

You'll get a confidence interval.

This is a non-parametric test.

So there is no assumption about the distribution of the indirect effect.

Now we have this re-sampling process.

You can get to these 95 percent confidence interval and

look if a zero is in the middle or is in this confidence interval.

If zero is in the confidence interval of your bootstrapping,

well you don't have any indirect effect.

You don't have evidence for an indirect effect,

but if zero is not in

the confidence interval yes then you have evidence for the indirect effect.

But now what people are doing,

what we researchers are doing is actually adopting a current way,

our current test for mediation and this test has been proposed by Hayes.

According to Hayes, you do not have to have

a significant relationship between your independent variable and your dependent variable.

And the reason is well you may have competing

mediators that would cancel the effect,

the direct effect of our independent variable on our dependent variable.

Let me give you an example.

So it could be that job engagement is

one mediator and we know I mean job meaningfulness would have a positive effect on

job engagement which could have a positive relationship with task performance

but if the other path is let's say task complexity or perceptions of task complexity,

the more meaningful the job is,

the less complex that task could appear to you,

could be perceived by you.

And then there is a negative relationship there.

And we know that this complexity has a negative relationship with job performance.

So engagement and test complexity or perceptions of test complexity

have opposing effects on the performance, on the DV.

So we could not have a significant effect or main effect or

direct effect of our independent variable on our dependent variable.

So yeah according to Hayes we don't have to have this direct effect,

a significant direct effect of our independent variable on our dependent variable.

So then the second step in our traditional test becomes the first step for

Hayes process if you want to call that process developed by Hayes.

So now we have to have a significant effect

between our independent variable and our mediator,

meaningfulness and engagement and we also have to have a significant effect

between our mediator engagement and our dependent variable job performance.

So the fourth step in the traditional test, well,

we don't need that as well because we don't need to have a main effect,

and according to Hayes we should not to do the Sobel test and the reason,

I just explained to you.

When we conduct a Sobel test,

we're violating one of the assumptions which is

the indirect effect has a normal distribution.

We'd go straight to the bootstrapping and again bootstrapping is a non-parametric test.

You can get a confidence interval and you can more precisely it's

a more robust test for your indirect effect for your mediation models.

In this session we covered a few important things.

First, we described what mediation is.

We are looking at the mechanism of the relationship

between your independent variable and your dependent variable and

then we explained the traditional and the current tests for mediation models.