Hi Meletios,

in my humble opinion, 30 hypotheses are way to much. Your paper has to focus on a central story and i wonder what this story might be. In addition, as you mention yourself, you simple have no room to develop 30 hypotheses. Think of a reader/reviewer who has to dig through this all.

With regard to the model: The term "robust" is a bit misused here (the point is data fit which is albeit just a vehicle to test the inherent causal assumptions). If I were the reviewer, I would like to know the test of the model in terms of chi-square and the structure/df. A fitting model is nice but not of so much worth, when there are many paths contained in it (i.e. has a high saturation).

Further: What is the sample size? The fact that a fitting model will not fit anymore, if you reduce its complexity, is a warning sign as (as far as I remember correctly), the RMSEA is biased in large complex models.

My approach would be to...

a) focus on a essential part of the model

b) build a auxiliary model around that path that 1) contains theory-based relevant control variables (addressing backdoor path adjustment) and 2) potentially instruments that create a sparse and testable structure.

I attached a presentation (short workshop) on graphical approaches. If the slides are too hard to follow: the last slide gives some essential introductions.

HTH

Holger

Hi Meletios,

in my humble opinion, 30 hypotheses are way to much. Your paper has to focus on a central story and i wonder what this story might be. In addition, as you mention yourself, you simple have no room to develop 30 hypotheses. Think of a reader/reviewer who has to dig through this all.

With regard to the model: The term "robust" is a bit misused here (the point is data fit which is albeit just a vehicle to test the inherent causal assumptions). If I were the reviewer, I would like to know the test of the model in terms of chi-square and the structure/df. A fitting model is nice but not of so much worth, when there are many paths contained in it (i.e. has a high saturation).

Further: What is the sample size? The fact that a fitting model will not fit anymore, if you reduce its complexity, is a warning sign as (as far as I remember correctly), the RMSEA is biased in large complex models.

My approach would be to...

a) focus on a essential part of the model

b) build a auxiliary model around that path that 1) contains theory-based relevant control variables (addressing backdoor path adjustment) and 2) potentially instruments that create a sparse and testable structure.

I attached a presentation (short workshop) on graphical approaches. If the slides are too hard to follow: the last slide gives some essential introductions.

HTH

Holger

Hi Meletios,

I agree with Holger, you have too many hypotheses in a paper (as a rule of thumb, no more than 5 hypos in a paper). Pay attention to novel hypos (grounded by foundational theories), eliminate replicated hypos, and/or divide the model into 2 or 3 models (for 2 or 3 different papers).

Try to reduce the number of hypotheses focusing on the main relationships. Present a single paper with the whole model with better results.

Holger Steinmetz thanks a lot both for your comments and your file. As a matter of fact yes I should have mentioned model fit. Indeed, model fit in the complex model is satisfactory, but after your suggestions I will try to create a more compact combining good model fit figures. Hence, I will try to reduce hypotheses and produce 2 models as Nguyen Tho suggested. If I will fail to create a model of good fit I shall follow Alejandro Ros-Gálvezes point of view with the total model focusing in the main relationships.

Thanks a lot guys. Have a nice and joyfull weekend.