comp.soft-sys.sas - The SAS statistics package.
I'm analyzing survival with Cox regression as a function of a single predictor: experimental group membership. This is all toward the end of making a figure of hazard ratio coefficients as commonly seen in medical literature, like so: subgroup (favors intervention) 1 (favors usual treatment) ------------------------------------------------------------ Systolic HF <-----------o------> Diastolic HF <----------------o---------------> etc... To get the coefficients and confidence intervals for each subgroup, I ran the Cox model limiting the sample to patients in one subgroup, then repeated for the other. The coefficient estimates for the effect of the intervention from the subgroup analysis were identical to the coefficients obtained from an interaction model run on all the cases. When the analysis was run separately by subgroup, I observed that experimental group membership had a statistically significant effect for systolic heart failure (HF) patients, but not for diastolic HF patients. I made the figure, then went back to the data to check the interaction coefficients so I could put a p-value on each pair (systolic vs diastolic, men vs women, young vs old, etc). I ran the analysis on the entire dataset using a multiplicative interaction model using three RHS variables: 1. experimental (0/1 dummy for group membership) 2. systolic (0/1 dummy indicating type of heart failure) 3. experimental*systolic (multiplicative interaction of the 2) I was surprised in doing this that the interaction for experimental*systolic was only of marginal significance (p=0.07 in stata, 0.09 in SAS). I had expected that if one subgroup had a statistically significant effect of experimental group membership and the other subgroup didn't, then the interaction of experimental group membership by subgroup should be significant. Can anyone straighten out my logic here? Is the problem as simple as the fact that the interaction effect is too small to be statistically significant, or am I committing a larger logical or statistical error here? I have checked the construction of all the dummy and interaction variables and they carry the right counts. Thanks, brad smith
Hi all I need to run fixed effect for 2sls regresssions. Could you please advise me how to do it in SAS? Thanks a lot Best regards Thu Phuong
Hi I'm trying to understand how to interpret the results of a logistic regression when there are one or more Ordinal Independent variables. eg. y= b0+ b1x1 + b2x2+ b3x3 where y =1 or 0, x2= 1 or 2 or 3, x3=1 or 2 or 3 or 4 Can anyone also tell me how this basic sas code needs to be changed?I am modeling y=0 rsubmit; Proc logistic data= test; model y = x1 x2 x3; run;
Hi everyone, I have a simple problem that I am stuck on. I have a list of 27 predictor variables (let's assume they're all numeric for now) and I want to run 27 logistic regression models without having to call the macro I wrote (below) 27 times manually....I want to automate this for efficiency purposes. At this point, I am only interested in univariate analysis and will then move towards multivariate analysis. How can I do this? My macro so far: %let list=a b c d e f g; %macro lr (data=,riskfactor=, outcome=, outformat=); proc logistic descending data=&data; format &outcome &outformat; model &outcome=&riskfactor/link=glogit; run; %mend; %lr (data=anne2, riskfactor=site, outcome=mutation, outformat=mutationnew.) I'd like to loop through &list, using a do loop one by one, but i'm not sure exactly how to write this. Something like: %do i=1 %to dim(&list); should work, but i'm not totally sure, can anyone help me with the last couple of steps? TIA. NM.