So, how important is Halloween shopping to the economy? Is the expected 15% drop in sales a big deal or just a blip? To answer this question, you have to ask "Compared to what?" For instance, Halloween spending is paltry compared with Christmas shopping, which clocks in over 0B. Or, following columnist George Will's line of argument, it is approximately equal to all spending by campaigns for all federal offices in 2012 presidential election cycle. (Halloween has an overall edge in our consumption affections over politics given that spending in non-presidential election years is lower than in presidential cycles.) Alternatively, Halloween spending equals the total spent by all 17,450 K-12 school districts on energy.
So, how do you size up our nation's expenditure on Halloween?
Inequality is a hot topic in the news. Just search the internet for that word in news stories and you find story after story after story describing the current level of inequality or arguing over its consequences. Rarely do these stories carefully consider the way inequality is measured–measurement is just taken as a given.
It turns out, the choice of what to measure is critical; While inequality of pre-tax income has increased dramatically in the past 30 years, inequality of consumption has hardly budged.
I was recently looking at some press releases from the Bureau of Labor Statistics and saw this item titled "San Mateo, Calif., has largest county 1st quarter 2013 over-the-year wage gain at 14.8%." That's a curious headline because San Mateo is a smallish county among the 335 largest US counties which were the sample for the BLS's report. If you click through the headline to the report itself, in the opening paragraph you'll learn that Fort Bend, TX produced the highest employment growth. Fort Bend is similarly a smallish county.
But before we get all excited about the virtues of small counties, we also learn at the top of the report that Sangamon, IL was the slowest growing county in terms of employment. And, you guessed it, Sangamon is a small county. And Williamson, TX saw the slowest wage growth. I won't even mention Williamson's size.
This recent blogpost on measuring institutional drop-out rates at community colleges provides a brilliant example of the point behind the title of Madison & Steen's Calculation vs. Context. Matt Reed, a dean at a two-year college, shares an interaction he had with a fellow conference panelist on how to measure community college success. Reed notes that our official definition of "dropping out" arguably fits four-year schools better than their two-year peers: "We get penalized when students do a year at the community college and then transfer to a four-year school. Even if they go on to complete the bachelor's successfully, that student still shows up in our numbers as a dropout."
His fellow panelist wasn't ready to let two-year schools off the hook on transfers, though. She points out that while there may be good reasons to transfer, such "churn" often signals dissatisfaction with the institution.
Reed notes that such bean-counting quibbles would be of only modest concern (presumably schools can education prospective students about the meaning of their transfer data) if the President hadn't just announced a plan to tie federal loans to "school performance." Indeed, in a world where a government bureaucrat's definition of a word can spell the doom for tens or hundreds of institutions, we'd better have a citizenry (and political class) that understands the nuances and challenges of creating effective indices. Sadly, I fear we don't live in that world.
The Houston Astros are not having a very good season. And their worst-in-baseball 51-108 record is only the start. According to Nielsen, their Sunday game in Cleveland earned a 0.0 rating in the Houston metro area. That is, not one of the 581 Houston metro TV meters tuned into the game at any time.
The Houston Chronicle blog linked above isn't so sure about Nielsen's statistical approach:
Nielsen's figures are, however, subject to challenge on a number of fronts. For one thing, the company's business is based on the concept that Nielsen can measure what millions of television viewers are watching by monitoring the behavior of hundreds...Nielsen's 0.0 rating does not preclude the probability that some households without Nielsen meters watched the game.
The author seems to misunderstand two aspects of statistics. First, while we often report point estimates, much of statistical thinking is precisely about the error bar. We know it is incredibly unlikely that our point estimate is exactly right.
The second misunderstanding is deeper. While 581 TVs is only a small sample of the more than 2 million Houston residents, it turns out that a well designed sample of that size really can produce tight estimates of a population parameter. I remember the wonder I experienced when I first realized this in an intro statistics course. I hope I don't forget that sense of awe–even incredulity–when teaching new students who, like the Chronicle, find this fact hard to believe. And what an important fact it is, for if we were not able to make reasonably precise estimates based on samples of merely hundreds of observations we would not be able to provide insight into so many critical questions of public and personal life.