Sample Size Estimation Problem

[Lakshmee]

I am trying to use a program called EZ-Quant to determine a sample size.

The program first asks you to choose one of the following options:
1. Judgmental assessment of universe error rate, or
2. Sample results from a similar audit universe.

I am selecting #1 because I don't have a similar audit universe.

After this, the program asks for a "Presumed Universe Error Rate."

The program explains this assessment is judgmentally derived from the percentage of the universe amount that is in error. It says the maximum is 50 percent.

I don't understand how I should come up with this number and need help.

Here is some background on what I'm trying to do.

I want to select a statistical sample from a list of transactions.
Once the sample of transactions is selected, each transaction in the sample will be examined for a certain type of error. I can go into more detail about what is being examined and what constitutes an error if need be but for now, I will just say, the transaction is either correct or in error.

The errors will then be projected onto the universe.

I initially looked at 4 transactions and found 2 of the 4 to be in error. Based on this, I am inclined to go with the 50% maximum error rate but I really don't know.

Any help would be most appreciated.

 

[rogojel]

hi,
this is just as you supposed, your original rough estimate of the error rate. Because the formula includes the term p(1-p) you get the largest sample size by assuming p=0.5.

regards

 

[Lakshmee]

rogojel, thanks for your reply. I would like to have a smaller sample. Therefore, I would like to input a smaller error rate but I would have to support this action and the number itself. Not sure how I could justify a smaller error rate.

 

[rogojel]

hi,
without data you cant really work on your error rate estimate - as an alternative you could input into the program different sample sizes and see what would be the precision of your measurement using that number of samples.

regards

 

[Lakshmee]

I have data and I have calculated the basics on it. Number of items, mean, mode, SD... I just need some guidance on coming up with a good error rate. I also need to figure out the desired max. precision amount which this programs says is equal to 1/2 the width of the sample projections confidence interval. Can someone put this is a way I can understand?