We're insulting more readers!

First-person heds on news stories are never good. But some are dramatically more offensive and clueless than others:

Survey: We're losing our religion
Poll on faith in U.S. shows more claim no religion; percentage of Christians down overall.

A wide-ranging study on American religious life found that the Roman Catholic population has been shifting out of the Northeast to the Southwest, the percentage of Christians in the nation has declined and an increasing number say they have no religion at all.

It's kind of hard to avoid the conclusion that "our religion" is Christianity, isn't it? Which may well be what the paper meant (and is almost certainly true in the aggregate of the newsroom), but diversity-wise, isn't that about as smart as calling the pale-pink crayon "flesh" color?

That's enough of a reason to spike the hed and remind the desk that first-person heds are banned, but we aren't done yet. Where's the evidence that some notional "we" are losing our religion? Hiding in the second graf, you say?

Fifteen percent of respondents said they had no religion, up from 14.2 percent in 2001 and 8.2 percent in 1990, according to the American Religious Identification Survey.

And -- aw, you'd think they were paying attention or something! -- the story includes the confidence interval (not the confidence level, but progress is progress). The interval, aka "margin of sampling error," is given as half a percentage point.* So given the other potential sources of error that result from being in the field for nine months,** we can safely say that the apparent increase in the "no religion" category approaches significance at traditional levels of confidence.

Remember another of the lessons from last fall, though? The difference between statistical significance and practical significance? What we're almost sure of, assuming the sample is an accurate reflection of whatever "us" the hed writer had in mind,*** is that the proportion of heathens in the population has risen almost a full percentage point. In other words, if this was 1992 and we had a poll indicating that a nearly significant increase had brought Ross Perot's support up to 15 percent, the hed would read "We're Voting For Perot!"

What we seem to have, then, is the classic news-vs.-polls problem. The things that make a poll good usually aren't the things that get attention in the news meeting, and the things that get attention are usually the least valid, or least generalizable, aspects of the poll. That's a general fault of how journalism handles science. But we really need not to compound it with heds that panic some readers and gratuitously insult a portion of the rest.

Now go eat your beets.

* About time the NYT dropped its irrational fear of reporting any survey data to the first decimal point, isn't it?
** Yes, religious affiliation is almost certainly more stable than voting intent, but -- you pays your money and you takes your choice.
*** Probably a dumb assumption, given the paper's circulation area.