Hi folks, I am trying to get a handle on how you write hypothesis with Path Analysis. Specifically I am using SmartPLS for the data analysis and I am wondering how to formulate the hypothesis well. I have three independent variables (two of which are moderators) and each has 4-5 indicators for each latent variable and the dependent latent variable has 5 indicators.
So when I write the moderator hypothesis do I mention the indicators. It gets pretty confusing pretty quickly.... with 14 indicators over the three latent constructs in one moderating relationship. I could just stick with the latent variables however there are arguments from the literature review about some of the indicators. As you can see it is challenging going from a regression mindset to SEM.
Louise, it is necessary to formulate hypotheses for the measurement models and for the structural model.
For example, hypotheses for the measurement models could be:
X1 - x4 are indicators of F1, x5 - x10 are indicators of F2, and x11-x14 are indicators of F3.
These hypotheses need to be tested because the assumption of unidimensional constructs may not hold in your sample.
For the structural model, Han already provided valuable input. Hypotheses for the structural model could be:
F1 has an effect on F4 (the dependent variable), or: F1 has a positive (negative) effect on F4; the same applies to F2 and F3.
A moderator hypothesis could be:
The effect of F1 on F4 is dependent on the levels of F2, such that the effect on F4 is stronger (strengthened) with increasing values of F2.
Hope this helps, Karin
How to write hypothesis for SEM (PLS-SEM)?
Think how to write a hypotheses is dependent on your research objective(s) and research question(s) e.g. whether your research focus is on testing the correlationship, influences, differences between / among constructs or specifically testing the moderating effect etc. From you question, sounds like you want to develop a hypothesis to test the moderating effect.
Following is an example for your perusal - e.g. IV = HR Practices, DV = Firm Performance and Moderator = Leadership Style, the alternative hypothesis can be H1: the relationship between HR Practices and Firm Performance can be moderated by leaderships style such that the relationship is stronger with multiple leadership styles (or transformative leadership) and weaker with only single leadership style (or transactional leadership). The above hypothesis is clearer than H2: the relationship between HR Practices and Firm Performance can be moderated by leadership style.
If your moderator is latent variable with indicators (e.g. like your case above on the 2 moderators that each with 4-5 indicators), normally we don't need to use the indicators. We might use the indicator if the moderator is demographic variable e.g. age, gender etc. whereby in itself is an indicator.
Hypothesis development totally depends upon what you want to accomplish through your research. Normally, indicators of a construct is not used for hypothesis development.
Louise, it is necessary to formulate hypotheses for the measurement models and for the structural model.
For example, hypotheses for the measurement models could be:
X1 - x4 are indicators of F1, x5 - x10 are indicators of F2, and x11-x14 are indicators of F3.
These hypotheses need to be tested because the assumption of unidimensional constructs may not hold in your sample.
For the structural model, Han already provided valuable input. Hypotheses for the structural model could be:
F1 has an effect on F4 (the dependent variable), or: F1 has a positive (negative) effect on F4; the same applies to F2 and F3.
A moderator hypothesis could be:
The effect of F1 on F4 is dependent on the levels of F2, such that the effect on F4 is stronger (strengthened) with increasing values of F2.
Hope this helps, Karin
There are tow types of hypotheses, The research hypotheses should be written in the If- Then structure as shown in the constructed hypothetical causal model based on theory and previous research and the statistical hypotheses in the Null and Alternative forms should be derive from the research hypotheses. and since you are using path analysis which is based on regression then:
Ho: All beta weights are zero. i.e., Ho: B1 =B2 =B3====Bk =0
H1: At least one of the beta weights is not zero
There is no rules to mention your indicators while writing hypho....
You must know the nature of your hypho whether they are; Directional hypho, non-directional hypo, or it-then statement hypho...
For your clarification kindly find my recently article with the application of PLS-SEM
Article How Important Is Customer Satisfaction? Quantitative Evidenc...
I am following the answers, informative and important subject to be followed. Thank you
Hello. I think that in SEM context we must formulate prediction or mechanism instead series of the hypothesis. Initially, SEM is based on a mix of constructs (derived from different theories), and the researcher's goal is to explain the complexity involved in the model even if the case concerns a moderated model.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling