topic Re: Nocollapse in Statistical Procedures

Nocollapse

sasnewbie12 — Thu, 05 Oct 2017 16:20:26 GMT

I am new to SAS, and running analysis on survey data, SAS 9.4.

When I run a survey regression procedure, it automatically collapses the strata that contain a single sampling unit into a pooled stratum, and gives me a blue message in the log; it also suggests using the NOCOLLAPSE option to avoid collapsing.

I tried the proc with and without the Nocollapse option and received different p-values.

Is it more statistically accurate to leave the default sas method of collapsing, or should I use the Nocollapse option?

Thank you

Re: Nocollapse

Reeza — Thu, 05 Oct 2017 16:24:35 GMT

@sasnewbie12 wrote:

I am new to SAS, and running analysis on survey data, SAS 9.4.

When I run a survey regression procedure, it automatically collapses the strata that contain a single sampling unit into a pooled stratum, and gives me a blue message in the log; it also suggests using the NOCOLLAPSE option to avoid collapsing.

I tried the proc with and without the Nocollapse option and received different p-values.

Is it more statistically accurate to leave the default sas method of collapsing, or should I use the Nocollapse option?

Thank you

It's usually important to consider the context of the strata's that are being combined.

PS. I'm moving this to the Stats forum.

Re: Nocollapse

ballardw — Thu, 05 Oct 2017 16:53:54 GMT

You will likely get different p-values as the option is going to affect variance calculations. The collapsing is in effect "borrowing" some data variance from another strata.

Your data and study design should tell whether or not the nocollapse option is appropriate. The planning should indicate why a strata has a single sampling unit and any concerns about that strata for the analysis.

Re: Nocollapse

sasnewbie12 — Sun, 29 Oct 2017 17:52:29 GMT

Ok this pointed me in the right direction. I found that collapsing was used while the data was being made, so therefore I would use the nocollapse option. I had someone else look at the data to help me on this.