Interaction interpretation 2 continious variables

[Wessel de Kroo]

Hello,

I’m having a problem that may be more of a statistics issue than a Stata issue.

Q1: how do I go about interpretation of the -0.025 coefficient for the interaction term here:

I have the following interaction regression:
Interaction term regressions:

	(1)	(2)
VARIABLES	MsciUsa_TotReturnIndex	MsciUsa_TotReturnIndex
exhibit_force_event	0.028	2.101**
	(0.758)	(1.009)
reduce_relations_event	1.564***	2.586***
	(0.220)	(0.399)
c.exhibit_force_event#c.reduce_relations_event		-0.005***
		(0.002)
Constant	2,931.348***	2,738.647***
	(84.614)	(104.429)

2. when we add an interaction effect. then the lower order term are not the main effects. exhibit_force_events is only the effect when the other term in the interaction is equal to 0?

3. a negative value for the interaction term will imply that the higher reduction of relations, the greater(more negative) the effect of exhibiting force will be on msciUsa_TotReturnIndex. However how do I interpret the -0.005 coefficient for c.exhibit_force_event#c.reduce_relations_event. unsure about this one.

4. here specifically. after including the interaction term the independent variable: exhibit_force changes from insignificant to significant , accompanied with a large change in coefficient.

does this mean that the independent variable is significant confitional on the inclusion of the interaction variable in the model?

Any advice is welcome

I already checked the following book:
https://scholar.google.nl/scholar?hl...le+regression%

perhaps my institution does not have full licensing , since I can sadly enough emrely check half. some relevant information is not visible for me.

[Clyde Schechter]

Q1: how do I go about interpretation of the -0.025 coefficient for the interaction term here:

You don't show any coefficient with that value. The interaction coefficient is -0.005.

2. when we add an interaction effect. then the lower order term are not the main effects. exhibit_force_events is only the effect when the other term in the interaction is equal to 0?

Correct!

3. a negative value for the interaction term will imply that the higher reduction of relations, the greater(more negative) the effect of exhibiting force will be on msciUsa_TotReturnIndex. However how do I interpret the -0.005 coefficient for c.exhibit_force_event#c.reduce_relations_event. unsure about this one.

I suppose you could put it this way. But just to be clear, the higher reduction of relations, the more negative the effect of exhibiting force will be. I don't know why you would calling that "greater." Algebraically, it is "smaller" whether that means less positive or more negative. Every unit increase in exhibit_force_event is associated with an expected decrease of 0.005 units in the effect of reduce_relations_event.

4. here specifically. after including the interaction term the independent variable: exhibit_force changes from insignificant to significant , accompanied with a large change in coefficient.

No! No! No! exhibit_force means two different things in the two models (as you, yourself, noted in question 2.) So nothing changed from insignificant to significant. You're looking at two different things, one of which is "significant" and the other is not. But nothing changed its own significance.

Consequently, the answer to

does this mean that the independent variable is significant confitional on the inclusion of the interaction variable in the model?

No, no, thousand times no!

Continuous by continuous interactions are difficult to understand. I think a picture is worth thousands of words here. Try this. Select a set of interesting or important values of exhibit_force_event, and a set of interesting or important values of reduce_relations_event. Just for the sake of illustrating the code, let's assume that those values are (1, 3, 5, 7, 9) and (20, 40, 60, 80, 100), respectively. Then after your interaction regression, run:

margins, at(exhibit_force_event = (1 3 5 7 9) reduce_relations_event = (20 40 60 80 100))
marginsplot

to see how the expected outcome looks as a function of these two variables. You can also see how the marginal effects of each variable look by running:

margins, dydx(exhibit_force_event) at(reduce_relations_event = (20 40 60 80 100))
marginsplot

and

margins, dydx(reduce_relations_event) at(exhibit_force_event = (1 3 5 7 9))
marginsplot

I think you will understand your model much better after examining these graphs. I think that you should consider that any audiences you present your findings to, unless they are professional statisticians, will find these things similarly difficult, and you might want to include these graphs in your presentation.

[Wessel de Kroo]

q1: coefficient -0.005 I meant, yes clear.

2.ok

3.clear

4.clear , nothing changes its own significance.

consequent answer also very clear.

- maginsplots-:

Very helpfull for me presentation in a bit.

I had an initial question on the use of marginsplot but the use and application seems crystal clear now.

I really like the marginal effect plots! they too can give an will be an insightful addition.

very clear advice , thank you !









I see my initial post had a decent amount of typo`s. Sorry about that.