Serial Mediation

pass

I need your help! How can I calculate a serial correlation in Jasp, so that the mediators are calculated one after the other instead of both parallel? I am a total beginner and don't know how I can do this best!

Thanks in advance!

EJ

Can you be a little more precise? Maybe a concrete example will help, with links to existing methodology (say in R)

Cheers,

E.J.

pass

I have three predictors (P1, P2, P3) and two Mediators M1 (L1, L2, L3), M2 (T1, T2, T3), and an outcome variable W.

The model looks like this: P1, P2, P3 -> L1, L2, L3 -> T1, T2, T3 -> W

Ihnwhi

Hi pass (@pass),

Although you cannot run serial mediation in the 'Mediation Analysis' option in the SEM module, there is a trick to do that: you can use the 'Structural Equation Model' option. Because the JASP SEM module is based on the lavaan syntax, what you need to do is to program the lavaan code.

As a simple example, let's say we have one predictor (P), two mediators (L & T), and one outcome (W). Please see the following figure.

In this case, you can program lavaan code as below:

L = b1 * P

T = b2 * L + b4 * P

W = b3 * T + b5 * L + b6 * P

ind1 := b1 * b2 * b3

ind2 := b1 * b5

ind3 := b4 * b3

where ind1 is an indirect effect from P to W via two mediators L and T (I guess this is what you mean by serial mediation), ind2 is an indirect effect from P to W via L, and ind3 is an indirect effect from P to W via T. Then, you copy that syntax and paste it into the syntax in JASP.

In your situation, there are multiple predictors and mediators (L1 ~ L3 & T1 ~ T3 seem like mediators measured across different time points), so the code might become more complex. But the principle is that you think of each equation as a regression model.

Please let me know if you have further questions. Hope this helps.

Cheers,

Ihnwhi

pass

Hi Ihnwhi,

thanks a lot!! That has already helped me a lot! :)

I was a bit vague with the formulation.. It is not multiple time points measured, but I have for example the construct P and then consider 3 variables for that construct. Exactly so for L and T. So I would have to calculate basically a mediation with two mediators but the whole times 3 ! Do you understand what I mean? I try to show it more clearly on a picture here:

Thanks again and in advance!!😉

Ihnwhi

Hi pass (@pass),

Thanks for your figure. So, if my understanding is correct, your P is a latent variable with three indicators (Pa, Pb, and Pc)? Likewise, L is a latent variable with three indicators (La, Lb, Lc), and so is T (measured by three indicators Ta, Tb, and Tc)? W might be an observed variable. Please let me know if my understanding is correct.

Cheers,

Ihnwhi

pass

Hey Ihnwhi,

not exactly! P for example stands for personality and is measured by three forms of personality Pa, Pb, and Pc. These are latent variables as well and are measured by 10 items each. Same counts for the other two constructs.

Thank in advance!

Ihnwhi

Hi pass (@pass),

All right. I am sorry I am still a bit confused. So, are P, L, and T like second-order latent variables? Or, you do serial mediation three times: (1) Pa, La, Ta, W, (2) Pb, Lb, Tb, W, and (3) Pc, Lc, Tc, W, respectively? Do you mind elaborating on your design a bit more?

Best,

Ihnwhi

pass

Hey. Sorry for the confusion.

I also find it hard to understand. I think P can be disregarded. What matters are the variables (for example Pa, Pb, Pc). All three P variables relate to all three L variables and all three T variables. Maybe an example will help.

For example, I am examining mediation:

Pa-> La-> Ta-> W

Pa-> La-> Tb-> W

Pa-> La-> Tc-> W

Pa-> Lb-> Ta-> W

Pa-> Lb-> Tb-> W

Pa-> Lb-> Tc-> W

Pa-> Lc-> Ta-> W

Pa-> Lc-> Tb-> W

Pa-> Lc-> Tc-> W

Pb-> La-> Ta-> W....

Ihnwhi

Hi pass (@pass),

Thanks for your elaboration. It seems like you would like to test serial mediation multiple times with different combinations of Pa, Pb, Pc, La, Lb, Lc, Ta, Tb, Tc. You can follow the way I programmed above, and you only need to change variables of interest each time.

For example, let's say we test Pa->La->Ta->W. You can construct the formula as (based on the figure you provided):

# Following three lines define latent variables

# Pa_item1 ~ Ta_item10 corresponds to indicators you chose

Pa =~ Pa_item1 + Pa_item2 + ... + Pa_item10

La =~ La_item1 + La_item2 + ... + La_item10

Ta =~ Ta_item1 + Ta_item2 + ... + Ta_item10

# Define regressions

La = a1*Pa

Ta = d21*La + a2*Pa

W = b2*Ta + b1*La + c*Pa

# Define indirect effects

ind1 := a1*d21*b2

ind2 := a1*b1

ind3 := a2*b2

Likewise, you can replace Pa, La, and Ta with other variables (Pb, Bc, Lb, Lc, Tb, and Tc).

Please let me know if you have further questions.

Hope this helps!

Cheers,

Ihnwhi

pass

Wow Ihnwi!

This is really great! Thank you very much! Next week I have collected my data and will try it out! 🙏

I will let you know!

Ihnwhi

Sounds great. Good luck!

Cheers,

Ihnwhi