mvnForum Homepage


Print at May 30, 2020 9:15:39 PM

Posted by vnork at May 9, 2009 8:57:46 PM
Re: Conference of the Monarch (Thoughts)
 
As I said earlier, I intended to sample about 100 pirates because you really don't need a much larger sample than that to get good results. It was well over 10% of the online population and big enough based on things like the law of large numbers and the central limit theorem. Yes, it surprises a lot of people that you can get accurate measurements from when you poll only a few thousand registered voters for a country the size of the USA, but it works.

This is where you're wrong. Your method of choosing 100 pirates is very inefficient, and is open to a lot of sampling error. Therefore, you need a sample of much larger than 100 to get accurate results. The polls you're referring to use much more efficient sampling methods than you used, so that comparison is invalid.

Illustrative example: Let's say Gallup decided to do a presidential poll by randomly selecting an American, then polling 5,000 people who live closest to that random American. This is a form of random sampling, but it's prone to error. You would have to repeat this many times to achieve the accuracy of individually selecting 5,000 people at random from across the nation. Not all forms of random sampling are made equal.

Now imagine a /w list of ships in an archipelago, expanded to a list of jobbers. Just as an example, let's say there's 200 jobbers on ships on Viridian. There's a SMH with 75 pirates that's screened to where 95% of jobbers are above able. The rest of the ocean is about randomly distributed with only 25% above able. You sample the first 100 jobbers on the list of ships. Keep in mind that the 75 jobbers on the SMH are clumped together on the list. Even though your sample is half the population, it is pretty much only representative of the entire population when the frame samples roughly 35 SMH jobbers and 65 other jobbers. Any other time, you will have large amounts of sampling error. Twenty percent of the time, a data point will report that 75% of pirates are able, when in fact only about 50% are. Another twenty percent of the time, it will report that 22% of pirates are able. To get accurate statistics, you would have to repeat the sampling process many times. This is not an efficient sampling method.

Instead, imagine having a list of jobbers taken from the /w list of ships. By randomly selecting every 4th jobber along the entire list, you can get a random sample with a much, much, much lower margin of error with only one sample of 50 jobbers. This is a much more efficient sampling method.

-Vnork

Puzzle Pirates™ © 2001-2016 Grey Havens, LLC All Rights Reserved.   Terms · Privacy · Affiliates