News organizations have always argued that space does not permit the sophisticated statistical detail that would fully explain the polls they conduct or report upon. The Web has changed all of that. Now news organizations can refer readers to their websites for more detail. Here's how polls should be conducted and reported ethically:

• Confidence levels and error limits should be fully reported, in detail, for subsamples as well as for the overall sample. At the very least, news organizations should provide a calculator so that readers can do the math themselves.
• News organizations should also provide an estimate of how accurately the sample being polled was selected. This "sampling error" should be added to the random errors described above.
• When reporting on polls of others, news organizations should either provide full disclosure of error limits (as above) or set and disclose their own standards about what is reliable enough to report. For example, a news organization might decide that no poll with a sample smaller than 500 will be reported. This approach is particularly useful for broadcasters who do not have enough airtime to go into details.
• News organizations should report possible sources of bias (circumstances that can affect poll results)--for instance, breaking news, high refusal rates or a multiple-day or multiple-week sampling period.
• News organizations should publish the full text of the polling script used by telephone operators.
• Wherever possible, polls should ask questions in multiple ways about issues that are hazy in the public's mind (the economy, for example; see main story).
• News organizations should publish their complete data sets in a generally usable format (Excel, CSV, HTML, XML), so interested parties can do their own analysis and so multiple surveys can be more easily combined.