Measuring GDP and its components is tricky business. GDP is supposed to measure all economic activity of a country, but of course not all activity is well monitored. So one has to work with proxies and estimates based on various indicators and surveys. Those efforts take time, and GDP estimates must be corrected after they’re first released. Everyone knows the initial GDP measure is imprecise and that revisions can be rather dramatic, yet the initial release receives the lion’s share of press coverage. Later revisions receive little attention, yet they matter quite a bit.

The latest GDP release included revisions going all the way back to 2012. ALFRED, a sibling of our FRED database, allows comparison of different historical “vintages” of data for many series included in FRED. The graph above shows the quarter-to-quarter growth rates for the two latest releases of U.S. real GDP. One can easily see that there have been some stark changes: In particular, the measure for the first quarter of 2015 was initially a much-discussed negative growth rate but was then revised to reveal a positive growth rate.

If these revisions average out to zero, we simply have an imprecise measure of GDP and that’s that. But if these revisions lean in one particular direction, then we may have some systematic bias in the initial estimates. One way to look for a significant bias is to examine the levels of GDP across data vintages. (The levels are the cumulation of the growth rates.) The graph below shows these levels, and one can see a difference between the two series all the way to the last data point.

How these graphs were created: For the first graph, search “real gross domestic product” on ALFRED. Choose “Series” instead of “Site” on the drop down menu (which is to the left of the search field). Click on the first choice and then on “% Chg” under “Units” below the graph. The second graph is simply the default version of the first graph, which has units in billions of dollars.

Suggested by Christian Zimmermann