The Social Science Statistics Blog has a great entry on the government new home sales figures. The 90% Confidence Interval around this number indicates that these figures are not significantly different from zero. Thus, when the press discusses these numbers as if they are real, they are quite likely making a Type I error. Below is a portion of the blog entry:
The government released its report on new home sales for the month of February; here is how the story was reported by Reuters (as seen on the New York Times website):
WASHINGTON, March 26 (Reuters) — Sales of new homes unexpectedly fell in February, hitting their lowest level in nearly seven years, according to a report released on Monday. New-home sales slid 3.9 percent, to an annual rate of 848,000 units, the lowest since June 2000, from a downwardly revised pace of 882,000 in January, the Commerce Department said. Sales for November and December were revised down as well.
And here is the Census Bureau press release:
Sales of new one-family houses in February 2007 were at a seasonally adjusted annual rate of 848,000, according to estimates released jointly today by the U.S. Census Bureau and the Department of Housing and Urban Development. This is 3.9 percent (±17.4%)* below the revised January rate of 882,000 and is 18.3 percent (±12.2%) below the February 2006 estimate of 1,038,000.
There are several amazing things about this. First, with all of the resources of the federal government, we can’t get better than a 17.4% half-width for a 90% confidence interval?