Interesting note on methodology noted by a forum user (Licorice):

Each month, questionnaires are mailed to a probability sample of approximately 5,000 employer firms selected from the larger Monthly Retail Trade Survey (MRTS). Firms responding to MARTS account for approximately 65% of the total national sales estimate. Advance sales estimates are computed using a link relative estimator. The change in sales from the previous month is estimated using only units that have reported data for both the current and previous month. There is no imputation or adjustment for nonrespondents in MARTS.

Got it?

The number is cooked. Only same-store sale changes count and those stores that closed or were newly opened are ignored.

This implies that the report will may slightly understate results in a rapidly-expanding environment for the first month (although that understatement will be corrected via revisions in the second month) but is almost certain to grossly overstate results in a weakening consumer retail environment - and those overstatements will not be corrected.

Here's why.

A new store (that has no "last month" history) will typically have low sales numbers for its first month of operation, simply because nobody knows it's there. While most stores have some sort of "grand opening" or other promotion linked to their inception, the usual "buzz" associated with that tends to be fleeting (days.) Traffic then tends to build with familiarity, assuming that the store is successful. Since in a flat market this traffic comes from other competitors, the "new store" impact would tend to be neutral or slightly positive beyond actual sales results. That is, the traffic taken by the new store (from existing retailers) will be counted, because the retailer that loses to the new store is counted and the new store is also counted. The numbers balance; the under-reporting is limited to the initial "grand opening", which is normally a one-off until traffic and familiarity builds, and the new store is reported both for prior and current months as soon as the first month passes. This causes a revision (upward) in the second month of operation to the prior month's results.

But a closed store is ignored in the month it fails, and the traffic that shifts FROM it to other stores pumps their comps .vs. the previous month. Therefore, as stores close it looks like retail activity actually increased when in fact at best it was flat.

Examples will make this clear.

We start with one store in the world that has net sales of "100".

Store #2 opens with sales of 10. Half of that is new activity, half comes from Store #1. First month shows a sales report of "95", a decrease. But in the next month Store #2's numbers come online, the "95" is revised to the (true) 105, and Store #2s numbers (which have climbed to 60, while Store #1 has lost share and now also has an amount of 60) are all reportable. Net activity is now accurate at 120 and the previous month is revised to the (true) net 105.

Store #1 and #2 both are operating with sales of 60. Store #2 fails, and half of its business goes to Store #1. In the month it fails Store #1 shows an increase and Store #2's numbers are DROPPED ENTIRELY, since it did not report. This is not revised. We now report a "50% increase" in retail activity, which is total crap - we really had a 25% net decrease for the current month. But the revision to the previous month does get posted, and depresses the previous month's numbers.

Did this just happen?

The August to September 2009 percent change was revised from -1.5 percent (±0.5%) to -2.3 percent (±0.3%).

Oh, it did! Now we know where the revision to the previous month came from - stores closed in the present month and their sales loss was intentionally dropped from the current month.

Cute folks, cute. This intentional omission of stores that fail to report for the current month but did for the previous, instead of counting a closed store as the zero that it is, mean that so-called "improvements" by competitors who pick up the previous store's traffic are not balanced by the loss of the failed store.

As such the report intentionally overstates results and you cannot obtain an accurate magnitude for the distortion. You can, however, detect that it happened by the negative revision to the previous month's data - and in this case, we got a big one.

If you were wondering how we can possibly have "improving" retail sales data when sales tax information from the states refuses to reflect this alleged "improvement" in retail sales, along with how states can post double-digit sales tax declines while "retail sales" are down by a much smaller percentage, you now understand. The Census Bureau intentionally lies by omitting the "zero" for a store in the month it closes - that loss of sales is never reported - not even retrospectively in a revision the next month.

The Market Ticker once again dissects The Goebbels Information Bureau residing in our Government.