Journalists like numbers. That’s not to say they like math; most reporters would be doing something far more financially rewarding if they liked math. But they do like numbers because they are good anchors for articles. For example “GDP rises 3.1 percent” simply makes for a stronger headline than “GDP seems to be a bit better than it had been.”
The federal government likes numbers, too, and no agency likes them more than the U.S. Census Bureau. Indeed, the Census Bureau’s whole reason for being is to count stuff — people, houses built and sold, retail and restaurant sales, factory orders, construction spending and on and on. You usually see these numbers reported as percentages, often with one digit to the right of the decimal. Nothing says precision like a decimal point.
For example, the Census Bureau announced that the rate of housing starts was 12.1 percent higher in December than in November. That sounds like a strong confirmation of what apparently is a sustained recovery in residential construction. Most outlets would report, understandably, that housing starts rose 12.1 percent.
But Census Bureau data is based on surveys and is, therefore, subject to a margin of error. In this particular case, the margin of error at a 90 percent confidence interval is 13.4 percent. In other words, it’s possible that housing starts actually soared by as much as 25.5 percent in December. Or they might have fallen by as much as 1.3 percent. And since the Census Bureau can’t say with 90 percent confidence — a standard not quite a strict as a typical public opinion poll — that housing starts either rose or fell, the data is not statistically significant.
The key issue isn’t the size of the margin of error but rather the size of the margin of error relative to the reported change. For example, December housing starts were estimated to be 36.9 percent higher than in December 2011. The Census Bureau says the margin of error in this case is 22.0 percent. That’s very high, but the data is still statistically significant because even in the worst-case scenario December 2012 was better than December 2011.
Economic data that is not statistically significant is fairly common. Over the past month, for example, the changes the Census Bureau reported for housing starts, new-home sales, retail and food service sales and construction spending were not statistically significant. That doesn’t mean those data releases were meaningless. It just means you shouldn’t get too worked up over good or bad news from a single month. It’s the trend that matters.