Interaction between two continuous variables- how to estimate outcome variable based on different values of predictors.

Reply
N/A
Posts: 1

Interaction between two continuous variables- how to estimate outcome variable based on different values of predictors.

Hello,

I have an interesting and strong interaction between two continuous variables (age and a blood marker), and would like to be able to estimate the outcome variable based on three ages of interest and varying levels of this biomarker.

Questions:

1. I have tried to set up Estimate statements, but am not sure how to handle these without a categorical variable (below- question marks are the concern area). Any suggestions?

2. If I wanted to plot the blood markers versus the outcome variable at these fixed ages, what is the best way to do this in SAS? Including an estimate of slope and 95% confidence interval for both?

proc glm data=prac;

class gender hypertension ;

model outcome=age education gender hypertension bloodmarker age*bloodmarker/solution clparm ;

estimate intercept 1 education 12 bloodmarker .5 age 60 bloodmarker*age ????? /E;

estimate intercept 1 education 12 bloodmarker 1 age 60 bloodmarker*age ????? /E;

estimate intercept 1 education 12 bloodmarker 1.5 age 60 bloodmarker*age ????? /E;

estimate intercept 1 education 12 bloodmarker .5 age 70 bloodmarker*age ????? /E;

estimate intercept 1 education 12 bloodmarker 1 age 70 bloodmarker*age ????? /E;

estimate intercept 1 education 12 bloodmarker 1.5 age 70 bloodmarker*age ????? /E;

estimate intercept 1 education 12 bloodmarker .5 age 80 bloodmarker*age ????? /E;

estimate intercept 1 education 12 bloodmarker 1 age 80 bloodmarker*age ????? /E;

estimate intercept 1 education 12 bloodmarker 1.5 age 80 bloodmarker*age ????? /E;

run; quit;


Thank you!!!!

Bri

Respected Advisor
Posts: 2,655

Re: Interaction between two continuous variables- how to estimate outcome variable based on different values of predictors.

You may wish to go at this a different way, as these estimates average over gender and hypertension status, and I'm not sure if that is meaningful.  It could be, and I am misinterpreting what you are asking, but would using the AT option in an LSMEANS statement be a possibility?

For instance, the predicted means of gender at the various values could be obtained with:

lsmeans gender/at bloodmarker=0.5 age=60;

lsmeans gender/at bloodmarker=1 age=60;

and so forth.

Steve Denham

Ask a Question
Discussion stats
  • 1 reply
  • 245 views
  • 0 likes
  • 2 in conversation