Introduction to Statistical Interactions
Back To: FAQ Home  Interpretation   Fundamentals   Polynomials   References

What is a statistical interaction in the context of OLS regression?

An interaction is a conditional relationship between an independent variable and the dependent variable. Generally, this type of conditional relationship refers to a case where the relationship between X1 and Y, varies according to the value of X2.

What is an interactive model?, How does one test for an interaction?, and When should one include a product term in a model?
Y= b 0 + b1 (X1) +b2 (X2) +b3 (X1 X2)

A) The equation above is an example of an interactive model. b3 represents the interaction between the X1 and X2, which is referred to here as the "product term".   In a traditional linear regression model (without a product term), the slope of Y on X1 has a constant value across all values of X2. The inclusion of a product term--b3(X1X2)--indicates the presence of a conditional multiplicative relationship between the two independent variables. In other words, the regression of Y on X1 is conditional upon the specific value of X2(Aiken and West 1991).

B) It is possible, using OLS regression, to test for relevant interactions by including a product term in the model estimation.  First, create the product term by multiplying the relevant variables and then estimate the model including the new "product term" variable in the regression equation.  The p-value provided for the t-test indicates the level of significance of the product term.  If the product term is significant one can conclude that the hypothesized interaction actually exists in the data.

C) The suspicion that an interaction may exist in a certain data set can arise from apriori information or from theorizing. When one suspects that an interaction may exist in the data, the interaction should be included. For an example of a statitical interaction click here: (Statistical Interaction Example).



Back To: FAQ Home   Interpretation   Fundamentals   Polynomials   References

Last updated: August 16, 1999