Our mission at the U.S. Bureau of Labor Statistics is to publish information about the labor market and economy. We always seek to improve our methods and provide the most accurate data in a cost-effective manner. All statistics, however, come with some uncertainty. Last year I wrote about how we deal with uncertainty in our measures. Today let’s talk about how we recently have improved our uncertainty measures in the Producer Price Index.
You may think it’s odd that an agency that tells the public what we know also works hard to explain what we don’t know. It may seem like we’re airing our dirty laundry, but that’s not how we see it. At BLS, one of our core values is to be transparent about our methods. Not only don’t we consider the laundry dirty, but we believe that airing it—that is, giving you more information about the strengths and the limitations of our data—is central to our mission. It’s part of our responsibility to give you information you can use to make better decisions.
The Producer Price Index (PPI) program measures the average change over time in the prices U.S. businesses receive for the goods they produce and the services they provide. BLS started publishing the PPI 126 years ago, making it one of our oldest measures. In 2014, the PPI expanded its coverage to provide a broader view of price change for goods, services, and construction. The PPI for final demand measures price change for goods and services sold for personal consumption, capital investment, government, and export. The PPI for intermediate demand tracks price change for goods, services, and construction products sold to businesses.
The PPI for final demand was unchanged in October 2016 and was up 0.8 percent over the last 12 months. But these figures are subject to sampling error. What’s that? It’s the uncertainty that results when we collect data from a sample of prices, rather than gathering prices from each of the millions of transactions that occur every day. For the PPI, we collect about 93,500 prices every month. A different sample of prices might give us different estimates of price change. Fortunately, we have tools to measure this sampling error. Most BLS programs collect data from sample surveys because it is far too expensive and would overburden businesses and workers to send all our surveys to everyone. Instead, we select samples carefully using scientific methods. These sampling methods work well, but they can’t avoid the possibility that the characteristics of a sample may differ from those of the population. We provide estimates of this sampling error by publishing variance estimates with the data. We recently released the first-ever variance estimates for the PPI.
If you aren’t into math, skip the next paragraph.
The measure of variance we use for the PPI is called a standard error. We use the standard error to calculate what statisticians call a confidence interval around the estimate. For example, the 1-month median absolute percent change in the PPI for final demand in 2015 was 0.30 percent. The standard error of that median was 0.11 percent. We can use these two numbers to calculate a confidence interval. In this example, we will use what we call a 95-percent confidence interval. To calculate that confidence interval, we take the estimated median price change of 0.30 percent, plus and minus two times the standard error of 0.11 percent. This gives us a confidence interval between 0.08 percent and 0.52 percent. We call this a 95-percent confidence interval because, if we were to choose 100 different samples of producer prices, the median price change would be between 0.08 percent and 0.52 percent in 95 of those samples.
OK, if you don’t like math, you can come back now. The chart above shows estimates of 1-month PPI changes (the red dots) each surrounded by its sampling uncertainty (the blue bars). If the blue bar crosses the 0.0 percent line, it means the change is not significantly different from zero.
Variance estimates are just one way BLS evaluates and explains the quality of our data and our methods. We have published information about our methods almost since our beginnings in 1884. Carroll Wright, the first BLS Commissioner, insisted on the “fearless publication of the facts.” We believe the fearless publication of the facts means not just explaining our measures and methods in highly technical terms. We want our measures and the uncertainty around them to be understood by a wide range of people, not just those who have advanced degrees in economics or statistics. We continue to seek clearer ways to explain uncertainty. One way is a new chart we are publishing on the monthly changes in nonfarm employment. In the future, we hope to publish more charts like this and simpler explanations of our methods. If you have ideas on how we can explain our data and methods more clearly, please share them with us below.
BLS data are the gold standard of economic statistics. But even gold bars have marks to indicate their impurities. Similarly, we at BLS don’t hide our impurities. We want you to understand the strengths and limitations of our data so you can use that knowledge to make good decisions.
