[Bob-san]

General statistics says a "large sample" is 30. However, 30 participants wouldn't accurately show bankruptcy rates. Expanding that to 300 would be more accurate. If the sample could be 3000, you'd probably have a pretty darn accurate sample though it'd take much longer. Even so, if 3000 participants are available, "go for it". A sample of 200 showing bankruptcy rates would probably be reasonably accurate but wouldn't be high-confidence. Sampling 200 of 118,308 would probably be accurate so long as the participants are accurately surveyed. Quantifying 10% of the reasons or even 1% of the reasons (or a fraction of a percent; 1/2%) would be even more accurate for such a large sample.

For example, if I were in a bank and we had 1,000 customers with a specific type of account. I am given a task to review those thousand customer accounts and report on any sales we'd lose by discontinuing service for that type of account. If I were to review 100 of the customers and present my findings, my boss would probably be happy that my sample is accurate and that high confidence can be placed. It's also a lot easier to sample 100 instead of 1000.

[General Battuta]

General statistics says a "large sample" is 30. However, 30 participants wouldn't accurately show bankruptcy rates. Expanding that to 300 would be more accurate. If the sample could be 3000, you'd probably have a pretty darn accurate sample though it'd take much longer. Even so, if 3000 participants are available, "go for it". A sample of 200 showing bankruptcy rates would probably be reasonably accurate but wouldn't be high-confidence. Sampling 200 of 118,308 would probably be accurate so long as the participants are accurately surveyed. Quantifying 10% of the reasons or even 1% of the reasons (or a fraction of a percent; 1/2%) would be even more accurate for such a large sample.

For example, if I were in a bank and we had 1,000 customers with a specific type of account. I am given a task to review those thousand customer accounts and report on any sales we'd lose by discontinuing service for that type of account. If I were to review 100 of the customers and present my findings, my boss would probably be happy that my sample is accurate and that high confidence can be placed. It's also a lot easier to sample 100 instead of 1000.

While this is nice, I hope you actually read the methodology used in this study, which is significantly better than anything you outlined here.

Also, I hope you're not relying on your boss being 'happy' or this kind of ballpark estimate of confidence. You can calculate very precise confidence, so your best bet is to choose a reasonable confidence interval and then select a sample size to get that confidence. Not vice versa.

[Liberator]

Heh, I'm happy enough to make you think a bit more about what you posted, all I wanted was more information and you provided it, thank you.

That said, I don't understand why we don't have a simple bar chart available somewhere that is the totality of the subject.  You said the total number of people filing bankruptcy in the sample time frame was 118,000 and change.  So why not take 6 months and make a bunch of calls to whatever agency is responsible for recording all the various information that would be useful in this issue.  And don't say it's too much work for no reason.  The reason is to provide a basis of fact to proceed from.  I'm tired of the world running on guesses.  No matter how educated a guess is, it's always eclipsed by the fact that you are guessing about.  We need facts to make world shattering decisions.  As far as the work goes, it wouldn't be too difficult to write a program to coalate, organize and tabulate the data in question and while the contents of the filings are certainly private, the nature of court filings are not except in certain circumstances.  I can pick up a trashy tabloid and know what Tom Cruise had for supper 3 nights ago, why is it so hard to compose an easy to read and understand document that doesn't take a PH'D in Statistics to understand?

[iamzack]

Probably because it's a waste of time and effort to get information that can be found using a less costly and time-consuming method.

[Liberator]

I was actually roundabout suggesting that there be an indexed database of legal filings.

[iamzack]

Probably because people would object to it for privacy reasons.

[Liberator]

Privacy reasons?  I'm not talking about indexing the contents of the cases themselves, just the filings, the public stuff.  I could give a rats ass if someone doesn't want me to know they're filing bankruptcy because they have $300k in credit card debt from buying a hole in the wat, er a boat, or because they have failed in a monumental way at business and have no recourse to get on with they're lives.  I don't care about that.  What I am talking about is a paid access database of legal filings updated automatically every 2 weeks or 2 days or whatever.  The only real problem is getting the infrastructure in place to handle it.  Besides, for most people, there is no reasonable level of privacy outside your own home anymore anyway, between cameraphones, traffic-cams, MySpace, Facebook, Twitter, etc where we record (almost) every second of our lives to expect that someone somewhere can't find out anything about you they want is the very epitome of burying your head in the sand.