Tuesday, February 19, 2008

On Numbers (2)

Erm. ICv2.com have updated their direct market sales estimates for January, but something smells fishy there. For some reason, everything seems to be about 5% higher than you'd expect, if you compare it to the December numbers, even in cases where there's no readily apparent reason for it.

Of course, it's not impossible that there's been a sudden major trend reversal in 2008, but it seems rather unlikely. In case some statistical error is involved - either on Diamond's part, which is where the raw data comes from, or on the part of ICv2.com's Milton Griepp - I'll hold off working on the new DC column for a few days.


Anonymous said...

Unlike December, there were 5 weeks in January. If these charts do in fact record a month's reorders, there you go.

- tone e

Anonymous said...

I highly doubt it. For one thing, a fifth week has never resulted in an across-the-board five percent increase before. For another, if reorders were the cause, books which came out in the last two weeks of January wouldn't be affected - but they are.

Given the sudden, inexplicable sales jump between COUNTDOWN #17 and #16, I'm more inclined to believe that there's a bend in Diamond's source data for the month.

Anonymous said...

Ah, that was me, by the way.


matches said...

Well, they do call them *estimates* for a reason. Do we even know how they're calculated?

Marc-Oliver said...

In terms of method, yes, we do. Using the Diamond index data, you only need a few actual numbers to get pretty good estimates of the rest. Presumably, those actual numbers come from retailers or small publishers.

That said, I should think the fact that you can tell when there's a five-percent discrepancy in the numbers from one month to the next speaks for the overall accuracy of the trends, rather than against it.

matches said...

Possibly - and I apologize if I'm just being dense - but even if that info is being obtained from small publishers (who I expect would be more reliable than any single retailer), isn't there a fairly significant potential for rounding error? 1.4450000001% and 1.454999999% are both going to show up on the Diamond charts as 1.45% - when we're talking about numbers that small isn't there a good bit of room for error?

Marc-Oliver said...

Sure, there's always a margin for error, statistically.

Having followed these numbers for close to five years now, though, I have to say it seems to be astonishingly small, as far as month-to-month fluctuation is concerned. To date, by all accounts, they've been very, very consistent - it would be a first if a simple margin of error was responsible for an apparent across-the-board jump like the one in the January numbers.

The only precedent I'm aware of for this sort of thing were the ICv2.com estimates for October 2003. They all seemed about 10% too high when they were first released, compared to earlier months. A few months later, ICv2.com corrected them, confirming that there had been a glitch in their initial calculations.

In this case, as I've said, I suspect it's something similar, but I'm more inclined to think the error occurred at Diamond this time around, due to the anomalous COUNTDOWN performance suggested by the index points.