Dear, I have a model with several labor market institutions and i want to show (in a separate table) all the interaction effects between them.
My doubt is about the methodology: is more correct to add interaction terms one at time (progressively) to my baseline equation or have I to estimate a single baseline equation for each interaction term?
Why do you want to include interaction terms among all your variables? In general, you need some kind of prior theory or specific hypothesis before you test an effect.
David L Morgan Because the related literature has stressed the importance of interaction between labor market institutions.
To do this, you need to produce an estimator that reflects unit-level differences of interacted variables whose moderators vary within units. Please see the attached document and the link: https://www.jstor.org/stable/1165181?seq=1
Theory is paramount. If you are doi nArticle Fixed and Random effects models: making an informed choice
You can use whatever terms you feel are relevant to your model to accurately curve fit your data. A, B, C, D, AB,AC,AD,BC, BD even ABC and ABCD. When you add them you should check how well they improve the fit if they significantly improve the fit add them if they don't do not add them. This type of model is actually a curve fit that will usually do a good job interpolating between the known data points, a limited amount of extrapolation will probably also be acceptable but you should plot up the relationship to make sure it behaves reasonably.
Curve fits are often quite useful but please do not confuse them with rigorous kinetic models that may reflect a more reasonable understanding of the underlying kinetics.
Most "rigorous kinetic models" are actually just glorified curve fits. This can easily proved by observing that there are often a large number of different "rigorous kinetic models" for any significant reaction that has been studied and at best only one can be correct.
Since you are an academic I suggest that you emphasize kinetic models that have the appearance of modelling the interaction (chemical reactions etc,) that you believe are taken place. But keep an open mind. Quite often what we believe to be true is not correct. Believe the data more than the theory.
In your case, for example, if you want to determine if the effect of an extra year of job experience has a different effect on productivity (for instance) when someone working in the public institution versus working in the private sector. You should construct the model in order to consider whether or not this is the case. To run this form of model with interaction term you should also construct a new variable (e.g., job experience*public sector employment) and then to determine if the effect of an extra year of job experience has a different effect on productivity (measured through TFP, for example) when someone is working in a public sector versus working in the private sector. Your conclusion would be that the effect of the job experience does or does not differ depending on whether one works in a public sector or not. Constructing the new variable with interaction term is a technical matter depending what type of software you use. Hope it helps. Good luck!
My point (which was not at all clear, sorry) was to point you to the flexibility of the within-between random effects model for such analysis as then you can handle interactions with time invariant predictors. In addition to the paper already cited , you could look at this one
Article Explaining Fixed Effects: Random Effects Modeling of Time-Se...
- Badges
- Science method
- Similar topics
- Methodology