Government Data With a Smiley Face?
Harry J. Clawar Ph.D.
Did you ever sit there and look at an object feeling a gnawing needling sensation that something was wrong. Some statisticians make useful discoveries that way while perusing data. Many of you are familiar with my findings that the spot gold markets displayed, for several years, the aberrant characteristic of increasing prices overseas then abruptly selling off from the London am fix to the NY close. This theory and demonstration started one day with this kind of uneasy feeling while observing the POG price curves. Several others have since extended the validity of this finding, through various methodologies, back to the mid-nineties.
Several months ago I got this very same sensation when examining various government statistics. Many of these statistics are presented weekly, and then revised the following week. It suddenly seemed to me that, in a great many cases, the initial reports compared to the revisions tended to be in the direction of presenting a more favorable face on the economic situation. To validate my impression, I started to collect data on weekly Initial Claims for unemployment, and their revisions, as presented by briefing.com on the Economic Calendar of statistics page. Unfortunately, they do not maintain a history of the initial numbers after they are revised. I therefore had to patiently collect the data each week.
As with my original gold price anomaly findings, I'm releasing preliminary Initial Claims irregularity outcomes after only 2 months of data collection. This is a short period of time to base a trend upon. I will continue to collect data to see if the trend holds as it did for the POG findings.
In the table presented below you can observe that for 7 of 8 weeks Initial Claims reported (Prior Week's First Report) were revised upwards (Actual Prior Week's Claims) the following week. If reporting errors are more or less random, you would expect about 4 overestimates and 4 underestimates. The probability of 7 or more underestimates out of 8 occurring merely by chance is about .035.
There are 3 possible reasons for this outcome of low probability:
- It is in fact one of those rare outcomes where 7 or more randomly fall in the same direction.
- The outcome represents an unintentional consistent bias in the data reporting.
- The outcome is intentionally muted during its initial presentation.
The first possibility will be evaluated by the collection and analysis of subsequent reports. The second possibility would mean that the government reporting methodologies are error prone in a consistent direction and they have done nothing which has mitigated those directional errors.
If either possibility 2 or 3 are correct, what would be the reason for either ignoring errors or introducing them into the reports? It seems to me that when economic statistics are reported, the great emphasis is given to the current number. Media reporting of revisions appear almost as an afterthought. The psychological impact comes from the new (initial) number not so much the revision. If the "good face on the initial numbers" is verified by subsequent data collection and analysis, then someone might just be playing a psychological game with those who patiently wait weekly for the new numbers before buying or selling in the markets.
Harry J. Clawar Ph.D.
Hjc@angelfire.com
September 22, 2003
Email this Article to a Friend 
Printer Friendly Version
