2
$\begingroup$

I analyze clinical research data, This is a comparison of two treatment groups. The outcome is a continuous variable. We decided to adjust on predictors of outcome in imbalace, in posthoc (We assume it even though not recommanded). These are just robustness analyses, if not exploratory. There are two adjustment variables, a qualitative one: varcat and a continuous one: varcont. My objective is to have the adjusted difference between the two groups with their confidence interval. I wonder if I should do a linear regression or an ANOVA or ANCOVA. In clinical research we talk more about ANOVA and ANCOVA than linear regression. Aren't these tests specific cases of linear regression? Don't we have the same conclusions as with linear regression?

with a linear regression i should compute like below:

lm (outcome~treatment+varcont+varcat,data=data)

How should I compute when doing analysis of variance or covariance?

$\endgroup$

1 Answer 1

1
$\begingroup$

Aren't these tests specific cases of linear regression?

Yes, ANOVA and ANCOVA are both special cases of linear regression.

How should I compute when doing analysis of variance or covariance.

ANOVA would be:

lm(outcome ~ treatment, data = mydata)

ANCOVA would be:

lm(outcome ~ treatment + varcat, data = mydata)

Don't we have the same conclusions as with linear regression?

No, because they are different models.

My objective is to have the adjusted difference between the two groups with their confidence interval. I wonder if I should do a linear regression or an ANOVA or ANCOVA.

With this objective, if you know that both varcat and varcont are potential confounders, then you should run your proposed linear regression adjusting for both variables. I won't comment further about posthoc baseline adjustments, since you already know that it is not advised by regulatory authorities.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.