Understanding changes in LOG OUTPUT in small samples

[deengyn]

Hi all, 

I'm doing a school project on complications occurring within 30-days post-op, it was defined as a binary variable. 

The sample is small (n = 139) and we have decided on 5 variables:

• Age (continuous)
• Comorbidities (binary)
• Diagnosis type (3 class category)
• Treatment type (3 class category)
• Sex (binary)

Now, I've done collinearity, univariate logistic analyses, and then eventually onto the multivariate logistic. It was stable, everything looked great. 

My supervisor suggested looking at 2 additional numeric variables which were also part of our dataset; in clinical terms, it would pose a great question and it has to do with treatment.

So I did some exploratory work. I looked into the variables, performed the same steps of looking into scatter plots, univariate, etc etc.

I tried the full model with 7 variables and again, stable. The output wasn't far from the initial with 5. 

However, should I change the code and add

units age = 5;

SAS would not output it. I'm sure it has to do with sample size, but could someone help me understand the background of this. 

[Reeza]

---

@deengyn wrote:

However, should I change the code and add

units age = 5;

SAS would not output it. I'm sure it has to do with sample size, but could someone help me understand the background of this. 

---

What did the log or error message show? Please post the log/error. 

AFAIK UNITS would only affect the calculation of the odds ratio, everything else should remain the same. 

---

@deengyn wrote:

Hi all, 

I'm doing a school project on complications occurring within 30-days post-op, it was defined as a binary variable. 

The sample is small (n = 139) and we have decided on 5 variables:

• Age (continuous)
• Comorbidities (binary)
• Diagnosis type (3 class category)
• Treatment type (3 class category)
• Sex (binary)

Now, I've done collinearity, univariate logistic analyses, and then eventually onto the multivariate logistic. It was stable, everything looked great. 

My supervisor suggested looking at 2 additional numeric variables which were also part of our dataset; in clinical terms, it would pose a great question and it has to do with treatment.

So I did some exploratory work. I looked into the variables, performed the same steps of looking into scatter plots, univariate, etc etc.

I tried the full model with 7 variables and again, stable. The output wasn't far from the initial with 5. 

However, should I change the code and add

units age = 5;

SAS would not output it. I'm sure it has to do with sample size, but could someone help me understand the background of this. 

---