As I noted earlier, the Census Bureau is releasing their poverty and household income data for 2013 tomorrow at 10am. But what you may not know is that the sample size for these data is smaller this year and that means it will take a slightly larger change in the poverty rate to achieve statistical significance.
Typically in recent years, there’s been about 100,000 addresses sampled for the poverty/income survey. But this year, because the Bureau is testing some quite different questions and procedures regarding how they collect income data, they’ve split the sample. About 30,000 households are getting the new questions, and thus the usual data upon which we’ll focus tomorrow will come from a sample almost a third smaller than last year.
This means that the confidence interval (CI) measuring the statistical certainty of the 2013 results will be wider than those for earlier years. In recent years, the 90% CI for the poverty rate has been between 0.2 and 0.3 percentage points, meaning that there’s a 90% chance that the true estimate is plus or minus this amount from the Census estimate. (Technically, if you took many re-samples of the same population under the same conditions and same design, then 90% of the time the CI would bracket the true parameter.)
This year, because of the smaller sample size, that CI will be a bit larger. Furthermore, the math behind all this results in needing a slightly larger value than the CI discussed above (the CI of the level) to generate a statistically significant change.
Punchline: I expect the rate to tick down a tenth or two or even three. But it will take a larger change than that to achieve statistical significance, i.e., a change that we can be reasonably confident is distinguishable from no change at all. Thus, though one never knows, it’s unlikely we’ll see a significant change in the poverty rate from 2012-13.