Here's a kind of innumeracy I see everywhere.

"X percent of Y is Z." This kind of statistic is often quoted by people who are trying to prove something, but even if the statistic is true, it doesn't even begin to prove anything. To know whether that statistic has any significance, you also have to know "what percent of non-Y is Z?". Often this bit of information is missing from the argument entirely.

For instance (I'm making up these numbers):

"10 percent of everyone who drinks Coca-Cola dies of cancer within 10 years." Does that constitute evidence that drinking Coca-Cola causes cancer? Not at all. You also have to answer the question, "What about the people who do not drink Coca-Cola? How many of them die of cancer in any 10 year period?" If, say, the answer is 5 percent, then you have a plausible argument (which is not proof) that drinking Coca-Cola causes cancer. On the other hand, if the answer is 20 percent, then you have equally good (or bad) evidence that drinking Coca-Cola prevents cancer. But without that other number, you have nothing at all.