Our agency issued 300,000 bills in 1999 for $360 million. Our accounting system doesn't allow us to fully track collections so we want to take a random sample of the 300,000 to measure our collection percentage.
I believe there are two sample size formulae that may be relevant: A "Variables Sampling" formula that includes: Population size Confidence level Sampling error Estimated Standard deviation or, an "Attribute Sampling" formula that includes: Population size Confidence level Percent of occurrence desired precision Some say we should use "Variables," because each bill can have a varying collection % (a bill can be 100% collected, 0% collected, 50% collected, etc.) Some say we should use "Attribute," because they believe Variables Sampling is geared toward finding information like the total amount billed--which we obviously already have. Also, "Attribute" requires an "estimated standard deviation." We can easily calculate the actual standard deviation of our population, so it doesn't seem that we'd be using "Variables" correctly. (I'm not sure if it's relevant to the question, but we also plan to stratify our sample, since a small number of bills represent a large amount of the total dollars.) Any idea which sample size formula is correct? We're a government agency, so it's important that we use the right method. Thanks in advance, Gerry . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
