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
- /
- Analytics
- /
- Stat Procs
- /
- Sample Size Estimation for Percentages - Differenc...

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

12-13-2013 10:23 AM

Hi all--

I work in direct marketing analytics and I have been using a free app called "Plan-alyser" to determine if the required sample sizes needed to ensure a pre-specified difference between two test response rates can be considered significant. Going forward I would like to use SAS to determine this.

Plan-alyser asks the user to input the required information:

1. Estimated response rate for the control panel.

2. Estimated response rate for the test panel

3. The Condidence level

How would I replicate this in SAS so I get an output which would determine a value for:

1. The control and test panel sample sizes required for the conditions listed above.

2. The percentage point difference in response that is considered significant

Is there a "Proc" or something else in SAS that will give me this desired output?

any assistance is appreciated. Thanks!!

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

12-13-2013 11:38 AM

The PROC is either Power or GLMPower.

Also you may find a Wizard in the Sas Programs folder (at least in Windows) titled SAS Power and Sample Size.

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

12-14-2013 11:38 PM

Hi,

You can use PROC SURVEYSELECT to do sample size partition in SAS.

Here is sample code.

Proc SurveySelect Data = datasetname Out = outdatasetname

Method = SRS /*SRS means Simple random sample*/

Seed = 07271987 /*this is a random & optional number, but in future if you want to generate the same sample with raw data then you need to re mention this seed number*/

Sampsize = 150; /*you can use samprate = .5 (50%)*/

Strata Variablename /*Doing Stratified sample*/

run;

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

12-16-2013 04:03 PM

This is good to know but what I really need to know is if the sample size between the test and control based on two proportions is statistically significant or not.

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

12-15-2013 09:32 AM

I think this might help:

proc power;

TwoSampleFreq Test = Fisher Dist = Exact_Cond Method=Walters

Alpha = 0.05

Sides = 2

GroupProportions = (p1 p2)

Power = 0.8

NTotal = .

;

where p1 and p2 are the two proportions

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

12-16-2013 10:27 AM

Great! I will try this out.

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

12-16-2013 04:04 PM

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

12-16-2013 06:14 PM

And quick question is Power = the same as entering the confidence interval?

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

12-17-2013 01:29 PM

There seems to be a lot of confusion here. Why would you want to know if the sample size is significantly different? I would think that you want to know if the proportions are significantly different, given the specified sample size.

Sometimes Wikipedia is your friend. The definition of statistical power there is in big bold letters--it is the probability that we conclude that the null hypothesis is false given that the null hypothesis is truly false. Given that, power= in PROC POWER is not the same as entering a confidence interval.

Steve Denham