Interpreting interaction terms in multiple regression

[K Wilson]

Hi everybody,

I'm a master's student who's working on the last analysis for my thesis project. I'm studying possible predictors of satisfaction in women with midwife and obstetrician birth attendants. What I'm wondering is how to interpret significant interactions in multiple regression. I have attached my SPSS output as a Word document. Please look at it before reading any further.

BA is a dichotomous variable (0=MW group, 1=OB group). I know that because the interaction terms (BA*MAST and BA*NATU) are significant, that the relationship between MAST and TOTSAT depends on level of BA, and the relationship between NATU and TOTSAT depends on level of BA. I have been doing lots of reading about interaction terms in MR (e.g., Aiken & West, 1991) and am unsure about how to properly graph the interaction terms, or if this is even necessary to make an interpretive statement.

The way I've interpreted them (that I am unsure of) is by substituting in values of the continuous predictor variable at one SD above and below the mean in the regression equation (containing only the terms that make up the interaction and the interaction term) and then plotting the slopes for each category of the dichotomous variable (MW or OB). Doing this resulted in the following calculations for the BA*NATU interaction:

When BA = 1 and NATU is low:
TOTSAT = constant - .46(BA) + .36(NAT) - .37(BA*NAT)
= 2.902 - .46(1) + .36(3.81) - .37(3.81)
= 2.402

When BA = 1 and NATU is high:
TOTSAT = 2.902 - .46(1) + .36(5.59) - .37(5.59)
= 2.382

When BA = 0 and NATU is low:
TOTSAT = 2.902 - .46(0) + .36(3.81) -.37(0)
= 4.27

When BA = 0 and NATU is high:
TOTSAT = 2.902 - .46(0) + .36(5.59) - .37(0)
= 4.91

I graphed these values and interpreted the differences to mean that while a high score on NAT slightly decreases TOTSAT for the OB group, a high score on NAT increases TOTSAT for the MW group. Can someone please tell me if this is correct or if I am completely offbase here???


Another related question I have is how should I interpret the significant (and non significant) variables that comprise the interaction terms (BA and NATU are significant, MAST is not)? I read these are referred to as the main effects, but what do they mean in the presence of a significant interaction?

Any help would be greatly appreciated!! Thanks.

K

[JohnM]

I graphed these values and interpreted the differences to mean that while a high score on NAT slightly decreases TOTSAT for the OB group, a high score on NAT increases TOTSAT for the MW group. Can someone please tell me if this is correct or if I am completely offbase here???


Another related question I have is how should I interpret the significant (and non significant) variables that comprise the interaction terms (BA and NATU are significant, MAST is not)? I read these are referred to as the main effects, but what do they mean in the presence of a significant interaction?

Any help would be greatly appreciated!! Thanks.

K

I didn't look at your attachment, but your interpretation is correct. An interaction occurs when the effect of one variable is significantly modified by another variable.

If there are significant interactions in your model, you really can't say anything about the main effects of factors in the interaction, since they (main effects) lose meaning in the context of an interaction.