topic Multicolinearity in Statistical Procedures

Multicolinearity

saza — Thu, 17 Mar 2022 18:14:17 GMT

To study the effect of the multicollinearity on response prediction, fill in the table below. For the last two columns, predict BODY FAT of an adult male with WEIGHT= 180 lbs, ABDOMEN = 90 cm and CHEST = 100 cm.

The following code gives me a value of zero for everything.

proc reg data=bodyfat.bodyfat;
model pctFatSiri = weight abdomen chest;
where weight = 180;
where abdomen = 90;
where chest = 100;
run;

Re: Multicolinearity

PaigeMiller — Thu, 17 Mar 2022 18:23:35 GMT

The output indicates you have only one observation in your data. You can't fit a model to that, and there is 0 degrees of freedom, all regression coefficients are missing, there is no multicollinearity (or in the trival sense, with one observation, everything is perfectly correlated with everything else)

Fix the input data so that there are enough observations to fit the model.

Re: Multicolinearity

saza — Thu, 17 Mar 2022 18:25:44 GMT

proc reg data=bodyft;
model pctfatsiri = weight;
model pctfatsiri = abdomen;
model pctfatsiri = chest;
model pctfatsiri = weight abdomen;
model pctfatsiri= weight chest;
model pctfatsiri = abdomen chest;
model pctfatsiri = weight abdomen chest;
where weight = 180 Abdomen = 90 Chest = 100;
run;

like this?

Re: Multicolinearity

PaigeMiller — Thu, 17 Mar 2022 18:45:00 GMT

There are so many problems here that I don't even know where to begin, most of them relating to confusion on your part about (seemingly) the entire problem. I don't have the time right now to go into all of this.

Re: Multicolinearity

saza — Thu, 17 Mar 2022 18:42:10 GMT

That was the table that was given with that problem set

Re: Multicolinearity

SAS_Rob — Thu, 17 Mar 2022 18:43:00 GMT

What you are trying to do is very similar to this example in the documentation.

SAS Help Center: Example 102.3 Predicting Weight by Height and Age

You would not want to use a WHERE statement to limit the observations in the input data set. Instead it seems you are to take the parameter estimates from each of the models you fit and make predictions and get prediction intervals based on those values for the independent variable. You could use the OUTPUT statement to get predictions/intervals for all the observations and then limit that data set using a WHERE statement to the values that you are interested in.