topic How do I contrast data using proc mixed for correlated data in Statistical Procedures

How do I contrast data using proc mixed for correlated data

tbn1 — Fri, 24 Aug 2018 16:26:25 GMT

I am trying to use proc mixed to analyze correlated data. Specifically, I am trying to assess whether mean number of pills prescribed changed over a 5 year time period, by provider.

Questions:

1. I get a p value for the model but I am trying to contrast the different years and the contasts keep coming up with empty fields.

2. Additionally, I am getting a message that says 'Invalid or missing data' and I am unsure what this refers to.

I appreciate any assistance. Thanks!

proc mixed data=gs;

class surgeon year(ref=first);

model pillnum = year/ s;

repeated / subject=surgeon type=cs r rcorr;

contrast 'year 2 vs 3-5' year 2 -2;

run;

Re: How do I contrast data using proc mixed for correlated data

sld — Fri, 24 Aug 2018 17:24:14 GMT

Your CONTRAST statement does not do what you want. See

CONTRAST and ESTIMATE Statements Made Easy: The LSMESTIMATE Statement

and

Examples of Writing CONTRAST and ESTIMATE Statements

Then follow up if you still have trouble.

We would have to see your data to know why you are getting "Invalid or missing data". Or speculate wildly.

I recommend adding <repeated-effect> to the REPEATED statement. See the documentation for MIXED | REPEATED It might help resolve your error message, more or less.

I hope this helps.

Re: How do I contrast data using proc mixed for correlated data

ballardw — Fri, 24 Aug 2018 22:56:33 GMT

@tbn1 wrote:

2. Additionally, I am getting a message that says 'Invalid or missing data' and I am unsure what this refers to.

When getting error or warning messages it is best to copy the code and messages from the log. Paste into a code box opened using the forum's {I} icon. Many messages have diagnostics but the message windows on this forum will reformat text making any of the diagnostic characters less useful. The code box does a much better job maintaining the log. Including the actually submitted code from the log clears up any potential questions about possible typos or options chosen.