topic Re: Testing if change in % is signifcant from one year to the next in SAS Procedures

Testing if change in % is signifcant from one year to the next

deleted_user — Wed, 07 Jan 2009 21:47:04 GMT

Can anyone tell me the appropriate statistical test in SAS for seeing if a change in percent or rate is significant?

I am looking at two different things:
1. mortality rates from 1 year to the next as well as over a ten year period to see if there was a statistically significant increase or decrease.

2. Percent of a county's births that are low birth weight from 1 year to the next and over a ten year period looking to see if change was significant.

Thanks

Re: Testing if change in % is signifcant from one year to the next

Doc_Duke — Wed, 07 Jan 2009 22:10:17 GMT

In just comparing two years, you could do it with a simple chi-squared test (PROC FREQ).

In looking at a longer time period, you might best do that with a logistic regression model treating year as a predictor variable (that would look for a linear trend in the change). (The MEASURES option in PROC FREQ will do ordinal association, to look at a monotonic change.)

Doc Muhlbaier
Duke

Re: Testing if change in % is signifcant from one year to the next

deleted_user — Thu, 08 Jan 2009 16:57:44 GMT

Thanks. When looking at % low birth weight between two years I just compared the overall average: 2005 at 7.7% and 2006 at 7.9%. When I plug this into proc freq/chisq it thinks I only have two observations, which I can understand why sas thinks that based on my number of observations but in reality LBW is calculated from thousands of records. How can I overcome this without putting every observation in sas?

Thanks

Re: Testing if change in % is signifcant from one year to the next

Doc_Duke — Thu, 08 Jan 2009 17:20:06 GMT

Here is an example of how to do a frequency table on counts

http://support.sas.com/documentation/cdl/en/procstat/59629/HTML/default/procstat_freq_sect003.htm