turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

Find a Community

- Home
- /
- SAS Programming
- /
- General Programming
- /
- Procedure for comparing means

Topic Options

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

01-15-2014 12:56 PM

I have a data-set that contains the # of sales("count") by time period ("period" - 1=first 6 months of year, 2 = last 6 mo's) by vendor ("group" - 4 vendors labeled 1,2,3, and 4).

What I'm trying to figure out is how different the variability is for each group between the first and last periods. I was wondering if there is a procedure that would take my data, create multiple samples from each group, and measure and record the means from each of the samples, by group and period, so I could then compare those sample means to say something like "Group 1 has a standard deviation of 1% between period 1 and 2, and Group 4 has a SD of 20%, which shows that Group 4's sales over 2 periods can fluctuate much more than Group 1's".

Any help is greatly appreciated!

Here's a basic data-set to give you an idea of how mine looks, I tried to design it so that Group 4 has a larger variation between periods 1 and 2 ...

data a;

input period group count;

datalines;

1 1 10

1 1 15

1 1 7

1 2 12

1 2 15

1 2 18

1 3 15

1 3 20

1 3 22

1 4 20

1 4 25

1 4 30

2 1 12

2 1 18

2 1 5

2 2 15

2 2 10

2 2 18

2 3 16

2 3 22

2 3 24

2 4 60

2 4 65

2 4 70

;

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

Posted in reply to triley

01-15-2014 02:05 PM

I may be shooting gnats with a 40 inch cannon, but I see this as a perfect example of testing for homogeneity of variance in PROC GLIMMIX. The dependent variable is a count, so it most likely follows a Poisson distribution, the measures on the vendors are repeated, and so would likely be correlated.

proc glimmix data=a;

class period group;

nloptions tech=nrridg;

model count=period/dist=poi ddfm=kr2;

random _residual_/group=group cl solution;

lsmeans period/ilink cl;

covtest homogeneity;

run;

The test of covariance parameters shows that at least one group had a value not equal to the other groups. The covariance parameter estimates are the variances of the groups, comparing period 1 to period 2. This can be generalized to more periods and groups if desired.

Steve Denham

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

Posted in reply to SteveDenham

01-16-2014 04:43 PM

Thanks for the quick response! I'm new to the GLIMMIX procedure, and just want to make sure I am reading the output correctly.

When I read the table below, is it fair to make the statement that Group 2 has the lowest 'count' variability over the 2 periods. And how would I express that using these #'s. For example, could I say that Group 4 is 40 times more variable than Group 3?

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

Posted in reply to triley

01-17-2014 02:36 PM

I assume you could say that. Ratios of variances are often reported. The big "BEWARE' here is that all of the estimates have fairly large standard errors, and if you constructed confidence intervals around the estimates (using chi-squared quantiles) you could have a really, really large range of possible ratio values. I would just say that for these data, Group 4 is much more variable than any of the other groups, and not try to quantify it, provided I reported the values in the table.

Steve Denhm