It's cloudy today (why data points are not trends)

Yesterday, we had the kind of weather you expect at the end of an academic year. The high temperature was in the upper 70s and the sun was shining. It felt like we were ready for summer.

Today is a different story. While I was looking forward to sitting on the deck and grading papers (well, that might be a little exaggerated), the fact that it's cloudy with a light breeze and temperatures in the upper 50s means that all bets are off. How disappointing. I guess summer isn't here after all.

Of course, we had some summer back in February. It was way warmer than usual. Then things got sort of back to normal. Then they got below normal. A month ago today, the high temperature here was 42 degrees. You'd think I'd be happy with 56.

With all this change in weather patterns every day, you can almost feel sorry for folks who have questions about Global Warming. How could such a shift align with what we see as such volatility on a regular basis? Maybe we're just in February but it wil cool off.

This isn't really about weather, much to my son Matt's regret. It's about our annoying tendency, made worse by media hype, to see data points in isolation. What might today's sudden drop in temperature mean for our long-term planning? Why doesn't it match what the five day forecast said just 3 days ago? (Why do the NOAA people get to keep their jobs with records like this? And why isn't the House Oversight committee holding hearings?)

Yesterday, the April jobs report came out. It was below expectations, while still being in positive territory. It's below the amount necessary to handle labor market expansion due to population (although there's other data that explains we've been seeing increased rates of retirement for since the turn of the century). It may be true that the warm weather in February (see above) artificially shifted hiring patterns earlier, but it's too soon to tell.

Related to the jobs report and election uncertainty in Europe, the Dow fell yesterday ending what all media sources called "the worst week of the year". It dropped 1.6% compared to April 27th. Oh, and in the second week of March, the Dow had it's "best week of the year" gaining 2.1%. Tuesday, the Dow hit its highest point in four years! And then the market corrected downward.

Here's another crazy example: Gas prices. Two months ago, there was great panic about how gasoline prices had risen dramatically. If you project that trend line out, we could be looking at gas prices approaching $5 per gallon by the end of the year. Last month, gas prices started to fall, dropping about 25 cents a gallon from the previous $4 mark. I was waiting for the media stories about how this could usher in the potential for $3 gas, if the trend continues. But they didn't write that story because you can't evaluate gas prices in short term patterns. There are seasonal impacts, supply issues, and the growth of the Chinese middle class that all play a role. But gas prices are back up a bit. What might that mean for Obama's reelection prospects?

One more example: We've begun paying attention to the national polling about President Obama and Governor Romney. As I posted on Facebook recently, the number of firms doing polling and the frequency of the polling means that we have new data all the time. What do these pieces of information tell us? First, as somebody finally pointed out on NPR this morning, we don't vote nationally but have 50 state elections. At that level these polls tell us nothing. Second, individual polls are skewed by the pollster, impacted by blips in the media, are subject to the specifics of the sample frame, and the nature of the question wording. But if you look at the trends, particularly at the state level, you see something much more substantial.

I've complained about 24 hour news incessantly on this blog and I will continue to do so. If you have to fill that much airtime, reflecting on how the trend lines have shifted on a seasonal basis (which happens in the stock market, at the gas pump, with hiring, and with the weather) isn't a hot story. It doesn't give pundits much to speculate about.

My point is that we have to take the long view. Minute fluctuations are meaningless unless we hype them out of proportion and then the hype impacts trends more than the fluctuations did.

I'd write more, but with a 20 degree drop in temperature since yesterday I'd better go get my parka out. If this pattern holds, the temperature could be in the TEENS by Monday!

Required Classes for Civil Discourse: First Course -- Statistics

In the month since my last post, I've focused a lot of energy on job search issues. There's some reason for hope on that front but I continue the hard work of staying patient and letting things unfold at their own pace.

Of course, I've continued to follow all of the interesting political/cultural dynamics as well as issues in higher education. One of the interesting things to show up on the higher education front is a significant research project conducted by sociologists at New York University and the University of Virginia. They found that the gains we'd hoped to find in students over time aren't happening. Universities need to demand more rigor, they argue, and to expect students to demonstrate what they've learned along the way.

I couldn't agree more. In fact, when I consider some of the dynamics currently in play in American Society, I've decided that there are some courses that ALL CITIZENS should be required to master whether in high school or in college. Mastery means that they take away the learning with them and know how to apply it to everyday situations. Over my next few posts, I'm going to describe lessons people SHOULD have learned from my required classes and explore why those matter. Given my sociologist background, the first example shouldn't be particularly surprising.


Introduction to Statistics

There are three things I want everyone to remember out of a statistics class. Those who argue that you can make statistics mean whatever you want didn't pass their statistics class. The heart of research and measurement is the careful explanation of what is being considered in ways that everyone can share. We agree on the measurement strategy, tell others what we've done, and then they can see the same thing.

Too much of the debate over deficit reduction and spending cuts has used funny statistics. For example, Rep. Paul Ryan claimed that the president's budget raised spending when it simply maintained legislatively mandated cost of living increases. These statistics may mean that one can make the case, but you wouldn't pass my stats course by playing fast and loose with definitions. Cherry-picking the data is not acceptable by any party. If Rep. Ryan made clear that he was describing spending over a decade or President Obama explained that he was counting proposed Medicare savings as part of his calculations, we'd be better off. When they just quote statistics that support their policy solutions without explaining their data, they just aren't being fair. The struggles we face are too important to paint over them with distorted data. It's hard to imagine how the sides will come together without first accepting a set of statistics that can be adequately defined and measured. Only then can policy situations be worked out.

Here's a second lesson one should take away from statistics class: inferences stem from the data. One of the new Republican Representatives from a Southern state gave an interview claiming that "the people" had given her a mandate to make drastic cuts in federal spending. The story went on to show that she had won her election by 51% to 49% over her Democratic opponent. Here's why she didn't pass statistics:

1. Current estimates are that a congressional district contains about 700,000 people. About 25% of those are under voting age, which leaves 525,000. 
2. The census department reports than in 2008, 71% of the eligible population was registered to vote which leaves 372,750. 
3. The 2010 election had a national turnout of 41.5% of registered voters, which leaves 154,691.
4. The candidate had 51% of her votes. If her district is "normal" that means she got 78,892 votes. Her opponent got 75,798. 
5. She won by 3,094 votes. 

Not only is this NOT a mandate for her position but we need to think about her responsibility to the other 75,798 people who didn't vote for her. My example is about a Republican but the same misapplication of statistics occurs for Democrats as well. Statistics don't lie and we shouldn't make them out to be more than they are. This is why our fascination with various polls becomes quickly problematic. We interpret trends when the data is actually quite varied. The right way to read them is to look at the diversity of opinion and not simply the horserace characteristics.

The third thing everyone should know out of statistics involves how to understand the Normal Distribution.

Not everything fits into a normal distribution, but it's good to remember as a mechanism for understanding the varieties of positions folks take. The normal distribution illustrates that most people cluster around a centrist position. The farther to the extreme one goes, the fewer people actually hold that position.

This is good to learn because it's easy to get our impressions formed by the extreme positions (the blue bars of the chart above). Consider how we think about the major cable news organizations involved in discourse today. We have the sense that Fox News represents one side (maybe including the green bars on the right) while MSNBC represents the other side (with blue and green bars). Here's the point: from a probability standpoint, they always reflect a minority view. There is no voice for the red bars above (CNN has an unfortunate tendency to put two blue-bar representatives together in a "debate"). When we talked about The Silent Majority forty years ago, the red bars are who we were talking about. [NOTE: please don't confuse the colors of the bars above for our unfortunate shorthand for political positions; that's not what this means and if it were my chart I'd use different colors.]

Here's another application of the normal distribution to current events. I'm currently working my way through Robert Putnam and David Campbell's American Grace, a very long research piece on the role of religion in contemporary society. They do a wonderful job showing the changes in religious expression over time. One of the critical trends they explore has to do with the rise of those with no religious commitments whatsoever, especially among younger generations. They demonstrate that these shifts are in reaction to how religion is perceived in American society today, especially in reaction to the role of women, attitudes toward sexual orientation, opposition to evolution, and wariness of other religions. But they go on to demonstrate that those highly visible religious position are only held by 25% of the 40% who claim to be evangelicals. The upshot is that the modern Christian church is being perceived by one of the blue bars in the normal distribution. Critics as well as casual observers miss the ways in which faith is meaningful for a host of Americans. Thinking of Westboro Baptist protesting military funerals or the Florida pastor wanting to burn the Qu'ran, they say, "If that's what it means to be Christian, I don't any part of it." As a result of failing to see the red bars, they miss the open-minded Christians who are maintaining a consistent faith without denouncing science, social change, or diversity of thought.

The normal curve cautions us against extrapolating from an extreme position and to pay more attention to the probability that such positions are held by "normal" people. I've read some stuff recently about the number of people who felt abused by organized religion. Reporting on his book about those abused, Jack Watts quotes the Barna Group, a major evangelical religious research operation, suggesting that 37% of the unchurched have experienced some type of negative interaction with religion.  I would suggest that many of the New Atheists and the young people estranged from faith are treating the blue bars as if they represented all people of faith. Maybe if we all worked on seeing complexity rather than oversimplifying the extremes, the church would be a source of healing our civil discourse and not a contributing factor to our challenges.

So if everyone mastered a basic statistics class, they could agree upon the nature of our challenges, resist the temptation to read more into the data than is reasonable, and be careful to see the complexity of the situations we face. You might even find yourself liking the math!