The monthly reporting of the unemployment statistics by the federal government provides important information for businesses, labor, and government agencies to plan and conduct their respective activities. This vast monthly effort by the United States Census Bureau involves the polling of about 60,000 households in 3137 county and county-equivalent cities to represent the entire population of the United States. The effort to make the sample and resulting statistics accurate is a good model of statistical sampling.
Many of the unemployment statistics and much of the raw data is released to the public each month. The most popular and the most widely reported of these statistics is the unemployment rate for the entire United States. That statistic represents the portion of the U.S. population that is unemployed as narrowly defined by the Census Bureau's employment and unemployment criteria. The following table shows some unemployment statistics and media comments from the year 2000:
Some Unemployment Rates and Media Comments
for the year 2000Month Unemployment Comments from the media Rate Change January, 2000 4.0% down 0.1% "Unemployment Rate Fell" (headline) February, 2000 4.1% up 0.1% "Job Growth in February Way Off the Recent Pace" (headline) " . . . sudden weakening in job growth" (article)
May, 2000 4.1% up 0.1% "Unemployment Rate Rose To 4.1 Percent Last Month In the strongest signal yet that the booming economy is finally slowing . . . " (lead paragraph) August, 2000 4.1% up 0.1% "In another clear signal that the economy is slowing . . . " (article) September, 2000 3.9% down 0.2 "Jobless Rate Fell to 3.9% in September" (headline) October, 2000 3.9% no change Every day brings new evidence that the once-booming national economy is slowing. (article) December, 2000 4.0% no change ". . . added jobs at a surprisingly robust rate in December." (article)
These reports come out every month, usually with attention-catching headlines and pressing comments. Every change is reported with the same urgency and importance as every other change, and little is said about the long term trends, although the trends are part of the Census Bureau's report. On a monthly basis, the changes that these reports denote may seem important and worthy of note, but in fact much of it is noise, not really a signal of any real change.
Let's look at unemployment rate information over a period of time so we can clearly see the signal and the noise. The following is graph of the unemployment rate for the last 5 years, or 60 months, of data published by the Bureau of Labor Statistics:
Observe during the year 2000 how stable the unemployment rate was. In that full year the unemployment rate only occurred at three different values: 3.9% (2 times), 4.0% (7 times) and 4.1% (3 times). Seeing the data in this form helps to understand that those month to month changes were really just noise. All those reports attributing cause and effect involving the unemployment rate were over-responding to the 'rocking of the boat,' and the variation was nothing more than 'dog bites man.' But it is hard for the news media to report, nothing really happened-nothing changed. But they should, because no real change was the real story of the unemployment rate in 2000. The changes like those throughout the year 2000 were only noise and were insignificant.
Better explanation would have been made and better understanding would have been gained, if the news media had discussed the relative stability of the unemployment rate in the year 2000.
A good graph, like the one above, will 'answer questions you do not even have to ask.' For instance, note how well the chart informs that there was a long trend downward from 1996 to the year 2000. This is truly noteworthy and the real story of the unemployment rate. Also, note that in the year 2001, there is the start of a trend upward. Rules on trends for data say that the data-points must increase or decrease for seven data-points in a row (some standards require eight) for it to be a valid trend.
Control limits calculated for the period above give the total width of the control limits as 0.497% (2 times the 0.248%). Since all of our changes above are 0.1% and 0.2% (less than the 0.248% in either direction), then that indicates that the entire year 2000 is only noise from month to month. However, the long trend from 1996 at above 5% to 2000 at 4% is a trend, a true signal and worthy of note. If you wish to see the control chart for the period July, 1996, to June, 2001, click this link for 5-Year Unemployment Rate Control Chart.
The lesson in the study is that interpretations of small amounts of data like only one month to another month can be misleading. 'Some news is only noise.' Although many people in and out of the news media have a hard time with that idea. A better way to understand data is to look at a graph and in many cases add control limits for complete understanding of signal and noise.
Would you like to know more or have McDonald & Associates work with you? Then Contact Us with your specific training and service needs. Questions or comments on this case are always welcome.
Return to Applications and Case Studies List
About Us Contact Us Classes and Services Home
Copyright © 2001, Michael L. McDonald