topic Re: Effect of a variable in Statistical Procedures

Effect of a variable

Erik3 — Wed, 15 Sep 2021 12:14:21 GMT

Good day

In my data set I have three grouping variables in total, age_group no_jabs and illness, and a dependent variable severity.

I want to determine whether a grouping variable, jabs, has an effect on my continuous variable, severity.

I first checked normality of severity with a histogram plot with kernel and normal overlay. CONCLUDED normality.

Thereafter I want to test homogeneity of my variances using Levene's test, code I used:

proc GLM data = health; *Variance/Levene's test;
CLASS age_group no_jabs comorb;
MODEL severity = age_group|no_jabs|comorb;
output out=ResData r=ResVar p=PredVar;
run;quit;

and I looked at the Pr > F column in the model row of the output, which is: <.0001: Does this mean I have homogeneity in my variances?

If so, how do I finally test the effect of the no_jabs variable on my severity variable?

Re: Effect of a variable

PaigeMiller — Wed, 15 Sep 2021 12:30:02 GMT

I first checked normality of severity with a histogram plot with kernel and normal overlay. CONCLUDED normality.

Completely unnecessary and not needed for PROC GLM. see https://blogs.sas.com/content/iml/2018/08/27/on-the-assumptions-and-misconceptions-of-linear-regression.html

and I looked at the Pr > F column in the model row of the output, which is: <.0001: Does this mean I have homogeneity in my variances?

Nothing in your code requests a homogeneity of variance test. So your output does not contain a homogeneity of variance test. To obtain a homogeneity of variance test, you would need to use the MEANS command with the HOVTEST option.

Re: Effect of a variable

Erik3 — Wed, 15 Sep 2021 14:23:10 GMT

Thank you PaigeMiller, I now use(effect of jabs on severity, use no_jabs as my GrpVar)

ods select hovftest;
proc GLM data = health;
CLASS no_jabs;
MODEL severity = no_jabs / SS3;
means no_jabs / hovtest;
run;quit;

instead of putting in ' age_group|no_jabs|comorb ' in my model statement. I got Levene's test output and assume I'm on a more correct path now.

If I now want to determine whether no_jabs has a significant effect on my severity I used:
proc GLM data = health;
CLASS no_jabs;
MODEL severity = no_jabs / SS3;
lsmeans no_jabs / adjust=tukey pdiff;
run;quit;

Does this seem OK?

Re: Effect of a variable

PaigeMiller — Wed, 15 Sep 2021 15:20:04 GMT

First, you used a model with all main effects, all two-way interactions and the three-way interaction. Now you are using a model with just one variable. I'm not sure why you switched, but if you believe that there are interactions, or even other main effects, I would use the larger model to determine what is and is not statistically significant.

Re: Effect of a variable

Erik3 — Wed, 15 Sep 2021 16:01:37 GMT

Ok, I'll make use of the larger model rather. I only want to see whether the one variable has an effect. But I understand your reasoning.
Thank you!!