It’s not surprising that measuring economic growth in an economy the size and scope of the United States is a daunting, data-heavy task that requires adjustments as new information becomes available. But a new analysis of the government’s official U.S. quarterly gross domestic product estimates by CNBC shows the margin of error between advance and final numbers can be so broad as to render early estimates virtually useless.

The analysis shows “large and persistent errors that should give everyone — from investors to business executives to policymakers — pause in relying on this early data for key decisions,” said Steve Liesman, CNBC’s senior economics reporter, who conducted the research.

On Friday, the U.S. Bureau of Economic Analysis (BEA) will released its final estimate of GDP growth for the fourth quarter of 2015. Economists surveyed by the Wall Street Journal expect no changes to last month’s measure, which tracks the value of the goods and services produced by the U.S. economy in a given time period. It was revised upward from the preliminary reading in January, to 1.0 percent from 0.7 percent.

Liesman looked at preliminary and revised U.S. official quarterly GDP estimates going back to 1990 and found an average error rate of 1.3 percentage points. For example, if the government’s advance figure were to say the U.S. economy grew by 2 percent, later revisions on average change that from anywhere between 0.7 percent and 3.3 percent.

Something similar happened in the GDP figures for the first quarter of 2014 when the initial reading said the U.S. economy shrank by only 0.1 percent, but two months later the contraction was broadened significantly, to 2.9 percent.

The analysis also found that nearly a third of the time the government gets the direction of growth wrong, which means initial readings that the U.S. economy expanded are incorrect 30 percent of the time. The BEA itself acknowledges the issue, though it says preliminary data on acceleration or deceleration is accurate “between 75 and 80 percent of the time.”

“Revisions are not errors,” Brent Moulton, the BEA’s associate director for national economic accounts, told CNBC. “They represent improvements to the GDP estimate as more information becomes available.”

The sheer number of times GDP data are widely adjusted can lead to claims the U.S. government manipulates the data. Accusations of conspiracy-minded efforts to mislead on GDP data flare up from time to time, often when one side or the other of the U.S. political divide wants to accuse the other side of hiding bad economic news from the public.

But this and other analyses show the margins of error between advance and final GDP estimates are more likely caused by less insidious factors. Mainly, the government’s inability to measure economic growth or contraction of an $18 trillion U.S. economy so quickly after the end of each quarter. More accurate data takes time to collect. 

The main takeaway of the CNBC study is a warning to business executives and policymakers: Don’t make hasty decisions based on the government’s initial estimates of economic growth or contraction.