Yesterday I blogged about Megan McArdle after she used a statistic, and then claimed in an online chat that she was not using a statistic, but rather a "hypothetical." McArdle took a fair bit of criticism, and posted a response. Astoundingly, she says this:

The reference is to my off-the-cuff remark about slashing pharmaceutical profits by 80%.  I should note, to be fair, that there were two portions of the comment:  one in which I repeated an estimate I had heard from several people, that the US accounted for something in the range of 85% of pharma net profits after you accounted for various issues, which I then turned into 80-90% when typing--a fairly common way to give a range on an uncertain verbal statistic.

When typing? Also, "a fairly common way to give range on an uncertain verbal statistic" (that's a lot of qualifiers) is to not state it as a fact! Even worse:

I don't want my off-the-cuff comment, based on conversations with people who were not speaking on the record, to become the source of a fake statistic for the right.  80% may not be right, and I can't back it up with any hard numbers, because there are no hard numbers available.  But multiple corresponding sources suggest that the number is well over 50%.  60% is probably the floor of likely. 

When you are in a hole, stop digging.