Hello everyone,

I am struggling at the moment to build the model for my research on value creation in academia-business relationships (so companies working with universities). I would like to use structural equation modelling for this.

On one hand, I have "relationship factors" (such as trust and commitment) which influence the "perceived value" of the researcher which again influences "satisfaction" and "intentions"

So, to put it simple:

Relationship factors -> perceived value -> Satisfaction / intentions

Now I am interested in understanding "perceived value" a bit better. So, what a researcher perceives as value of a relationship with businesses. Therefore, I would like to understand which specific value elements have the highest influence. By definition value is defined as the trade-off between benefits and sacrifices. As I found out in interviews, researchers value a lot the benefits for students, society and businesses, not that much their own benefits. Therefore, I would like to understand how much the benefits for the different stakeholder influence the overall value perception of the researcher.

E.g.: do benefits for students drive the perceived value more, or the benefits for the business?

How do I have to conceptualise this now? I guess it is a formative construct as benefits and sacrifices are an integrated part of the construct, isnt it?

What I think I have to do is the following

- Measure the benefits/sacrifices for students which the researcher perceives due to the relationship with a company (I can use multiple items for this)

- Measure the benefits/sacrifices for society ..

- Measure the benefits/sacrifices for the business ..

- Measure the benefits/sacrifices for the university ..

- Measure the benefits/sacrifices for the department ..

- Measure the researchers own benefits ...

Apart from this, I measure the overall perceived value of the relationship. My question is now: using this data, can I analyse the influence of the different stakeholder benefits on the researcher's perceived value (e.g. 23% of the researcher's perceived value of the relationship is explained by benefits to society)?

The next question is: can I then put this formative construct into my above shown model? So, can I do this:

Relationship factors -> various constructs to measure the benefits/sacrifices of the different stakeholders -> relationship value -> satisfaction / intentions

As I have neither worked with formative constructs nor with SEM before, I am very unsure how to deal with this.

Many thanks in advance! Any comments on this would be appreciated a lot!

Thorsten

I'm glad you're interested in this, and you're asking a lot of probing questions! Let's answer a few of those questions.

It's great that you're conceptualizing a model to test. SEM is a very flexible tool that allows considerable freedom in answering questions involving somewhat abstract constructs. Regarding your measurement questions, the way that you define a latent factor is to measure a set of observable variables (manifest variables) that represent or measure it. If you want to include these perceived value constructs, you'll have to find decent measures of them to start with. Then you'll define the latent perceived value factors as the underlying trait that explains why all of your value items are related to one another.

Once you have the variables and data you need, you should be able to fit any regressions between latent factors that you want. However, I'll add the caveat that it is often very difficult to determine which direction the causal arrows should flow. In the simplest case, x -> y and y -> x will have the exact same model fit, and are indistinguishable. You will be able to get variance accounted for, but you won't be able to prove that satisfaction/intentions don't cause relationship factors. You'll also have to figure out the variance accounted for on your own, but that's just a little algebra once you have the model results.

I'm glad you're interested in this, and you're asking a lot of probing questions! Let's answer a few of those questions.

It's great that you're conceptualizing a model to test. SEM is a very flexible tool that allows considerable freedom in answering questions involving somewhat abstract constructs. Regarding your measurement questions, the way that you define a latent factor is to measure a set of observable variables (manifest variables) that represent or measure it. If you want to include these perceived value constructs, you'll have to find decent measures of them to start with. Then you'll define the latent perceived value factors as the underlying trait that explains why all of your value items are related to one another.

Once you have the variables and data you need, you should be able to fit any regressions between latent factors that you want. However, I'll add the caveat that it is often very difficult to determine which direction the causal arrows should flow. In the simplest case, x -> y and y -> x will have the exact same model fit, and are indistinguishable. You will be able to get variance accounted for, but you won't be able to prove that satisfaction/intentions don't cause relationship factors. You'll also have to figure out the variance accounted for on your own, but that's just a little algebra once you have the model results.