Happy new year!
Context: I am conducting moderation analysis with binary outcome variable (DV). [Medication use(time1/dichotomous) moderating the effect of X(time1/continuous) on Y(time2/dichotomous)], but I am using a little-known statistical method to reduce problems associated with binary data.
I would like help to add some control variables ("gender", "conduct problems") to Hess et al.'s* test of moderation in logistic regression (code below), but I am very new to SAS, coming from SPSS. I could run it as is with no controls (two analyses of male and female), but I would like to try it with gender and conduct problems included in the model if possible.
- Could somebody please help me to code in the control variables (say "gender" and "conduct problems") without changing the moderating nature of the code below?
- Less important question: I am unsure what "
values of the parameters to be estimated" means. How would I go about deciding if I should change these values or not?
*Source of Hess et al.'s code and description: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2393725
Sorry if my terminology is poor, I have background in psychology not statistics.
For ease of access to code without annotations:
data test;
input n k X1 X2;
datalines;
30 7 0 0
33 4 0 1
33 23 1 0
28 6 1 1
run;
proc nlmixed data=test DF=120;
parms b0=0.1 b1=0.1 b2=0.1 b12=0.1;
u = b0 + b1*X1 + b2*X2 + b12*X1*X2;
p = exp(u)/(1+exp(u));
model k ~ binomial(n,p);
estimate 'phi12' exp(b0+b1+b2+b12)/(1+exp(b0+b1+b2+b12)) + exp(b0)/(1+exp(b0)) - exp(b0+b1)/(1+exp(b0+b1)) - exp(b0+b2)/(1+exp(b0+b2));
run;
Annotated SAS Code for Testing Moderation Effects in a 2x2 Experiment with a Binary Dependent Variable
/*---------------------------------------------------------------------------
SAS code for two-way interaction effects in binary data using logistic regression. The following code will run “as is” for any 2x2 experiment, changing only the data and DF value.
---------------------------------------------------------------------------*/
/*--------------------------------------------------------------------------
this is the test dataset (named “test”)
n is the number of subjects in a treatment condition,
k is the number of subjects choosing “1” in that treatment condition,
X1 and X2 are the dummy variables for the two treatment conditions
--------------------------------------------------------------------------*/
data test;
input n k X1 X2;
datalines;
30 7 0 0
33 4 0 1
33 23 1 0
28 6 1 1
run;
/* call proc nlmixed, specifying the degree of freedom (DF) to be n-k
where n is the number of observations and k is the number of parameters */
proc nlmixed data=test DF=120;
/* parms specifies the initial values of the parameters to be estimated
for a 2x2 experiment; these can be set at any value, here are set at 0.1
b0 is the intercept,
b1 and b2 are the coefficients for main effects,
b12 is the interaction coefficient */
parms b0=0.1 b1=0.1 b2=0.1 b12=0.1;
/* specifying the basic utility model for a 2x2 experiment */
u = b0 + b1*X1 + b2*X2 + b12*X1*X2;
/* the next 2 lines formulate a logit model; these can be used as is
even if the research design is modified from 2x2 */
p = exp(u)/(1+exp(u));
model k ~ binomial(n,p);
/* estimating the moderation effect between X1 and X2 for a
2x2 experiment */
estimate 'phi12' exp(b0+b1+b2+b12)/(1+exp(b0+b1+b2+b12)) + exp(b0)/(1+exp(b0)) - exp(b0+b1)/(1+exp(b0+b1)) - exp(b0+b2)/(1+exp(b0+b2));
run;
/* end of code */
