The Census Bureau has released a new dataset today: the 2005-2007 American Community Survey 3-Year Estimates. The estimates were created from the ACS sample survey for all geographies that have a population of 20,000 or more. Throughout this decade, the Census has released annual ACS estimates for areas with 65,000 or more people. They can't create annual numbers for areas smaller than 65k because the estimates are not statistically robust enough. But now that they have three years of data, they can create averages for that three year period that are statistically sound for areas with at least 20k.
From this point forward, they will continue to produce annual 65k+ numbers every year, and they will also release 20k+ numbers each year by adding the latest year and dropping the oldest one (so next year, they will release annual 65k+ numbers for 2008 and 3 year 20k+ estimates for 2006-2008).
All of the American Community Survey datasets are available through the American Factfinder.
The Census Bureau is also in the midst of preparations for the next decennial census, which is the 100% count of the population for all geographic areas. The questionnaires will be sent out on April 1, 2010.
Subscribe to:
Post Comments (Atom)
1 comment:
A little follow-up: an article published in the NY Times today that uses the new ACS data to examine changes in the city's population. Census Shows Growing Diversity in New York City
They have a few maps that show changes in income and rent in the city. They don't say so in the article, but the divisions they're mapping are called PUMAs. Since they have greater than 20k people, they're the smallest unit you can get (below the borough) to show change within the city.
It's worth pointing out (for the sake of information literacy) that in doing their analysis, they broke a number of rules. They said they were looking at data from 2007, but these are really averages between 2005 and 2007. They provide one exact figure for each data point, i.e. 81,000 people, when it reality the ACS reports data as an estimated range - i.e. btw 79k and 83k, with a confidence interval. Finally, they compared the new numbers to numbers from the 2000 Census, which is a completely different dataset (a 100% count and not a survey). So to some degree, the conclusions they make in their analysis may be distorted since the comparison is improper.
Post a Comment