Monday, June 27, 2016

The Economist's "Seven Brothers"

I use this story from The Economist as a conceptual explanation of the one-sample t-test. TL:DR: Sex ratio disparity data out of India serves as a conceptual introduction to the one-sample t-test.

So, at its most basic, one sample t-test uses some given, presumably true number/mu, and tests your sample against that number. This conceptual example illustrates this via the naturally occurring sex ratio in humans (your mu) versus 2006-8 sex ratio data from different states in India (your sample data). Why look at this data? Social pressure, like dowries, high rates of sexual violence against women in India, etc. make male offspring more attractive than female offspring to some families. And the data provides evidence that this is leading to disturbing demographic shifts. See the table below from The Economist:

If you would like, you could even eyeball and enter the data for the different states and calculate your t-test.

I like this example because it illustrates how a very simple statistical test could be used to provide evidence that something fishy is going on in India. A tiny example of forensic statistics in the spirit of Freakonomics.

The data is from the UN and demonstrates how data collection is used by enormous organizations to uncover concerns in given countries/regions.

This data also demonstrates a global issue. The article itself goes into greater detail about this complicated issue: Why it is occurring and possible outcomes (increased sex trafficking, violence against women) that may result from this sex imbalance.

Monday, June 20, 2016

Teaching your students about the de facto ban on federally funded gun research

Organizations have frequently tried to shut down/manipulate data for their own ends.

Big tobacco and lung cancer and addiction research. The National Football League and Chronic Traumatic Encephaly.

And for the last 20 years, the National Riffle Association has successfully blocked funding for research investigating public safety and gun ownership.

Essentially, the NRA has concentrated on eliminating funding at the CDC for research related to better understanding how guns hurt people. It started in 1996 with the Dickey Amendment and no one has been willing to fight to bring back funding.

The APA wrote a piece on this in 2013 that summarizes the issue.

In the wake of the shooting in Orlando, NPR did a story explaining how the American Medical Association is trying to change the rules governing gun research and the L.A. times published this column.

I think this precedence is unfortunate from both sides of the gun debate. I grew up in rural Pennsylvania. I've been to the shooting range. I like deer jerky. There were free gun safety classes offered at my junior high school and I know plenty of nice, safe people who own guns and some that even hunt so their families actually have enough food.

Is proper gun safety transmitted as a social norm within families? Are their differences in gun behaviors between people who hunt versus purchase guns purely for personal safety?

What about the mental health aspect of gun violence? Have certain municipalities succeed in keeping suicide by gun rates low? What about state-level screenings to keep guns out of the hands of the mentally ill? What works?

Maybe it really is the case that if guns are outlawed, only outlaws will have guns. But show us the data.

How to use in class?

-By stopping the money, the research has stopped. This may be useful as a way of teaching your advanced undergraduates and graduate students the importance of the flow soft money in research.

-Contextualize this research by adding in the stories of how the tobacco industry and the NFL have tried to manipulate/stop research to hit home to your students the importance of research as a tool in setting public policy.

-Research can be politicized. We should be weary of where the money is coming from, be that money to sponsor research or political donations to stop research.

-Research can change our government. Research can improve the world by providing more information when making very important governance decisions.

Also, while not directly related to the issue of funding manipulation, I like this video which explain the complexity of gun issues in the US. It uses A LOT of data to illustrate the problems we must solve. 

Monday, June 13, 2016

Carroll's "Sorry, There’s Nothing Magical About Breakfast"

I love research that is counterintuitive. It is interesting to me and makes a strong, memorable example for the classroom. That's why I'm recommending Carroll's piece from the NYT. It questions the conventional wisdom that breakfast is the most important meal of the day.

As Carroll details, there is a long standing and strong belief in nutrition research claiming that breakfast reduces obesity and leads to numerous healthy outcomes. But most nutrition research is correlational, not causal. AND there seems to be an echo-chamber effect, such that folks are miss-citing previous nutrition research to bring it in line with the breakfast research.

Reasons to use this article as a discussion piece in your statistics or research methods course:

-Highlights the difference between correlation and causation
-Provides an easy to understand example of publication bias ("no breakfast = obesity" is considered a fact, studies that found the opposite were less likely to be published)
-This pop NYT article includes links to all of the referenced research.
-Describes different research methods used to explore the issue. Controlled research in which breakfast eaters were forced to skip breakfast. Meta analysis. Diary studies. Adult research versus kid research.
-Conflict of interest: Should we trust pro-breakfast research conducted by companies that sell cereal?
-Conflation: Research does support the notion that children who eat breakfast do better in school than those who do not. This has been used to support the bigger notion that breakfast = good. However, this research has used poor kids who get free breakfast at school as their participants. So, under-nourished, rapidly growing kids who live in a condition of food uncertainty and poverty do better when they have at least one guaranteed meal a day, a guaranteed meal that involves spending more time at the physical school. Can we generalize this to, say, middle class adults, with different nutritional needs than growing children, who have greater control over their lives, and are financially secure?

Monday, June 6, 2016

Rich, Cox, and Bloch's "Money, Race and Success: How Your School District Compares"

If you are familiar with financial and racial disparities that exist in the US, you can probably guess where this article is going based on its title. Kids in wealthy school districts do better in school than poor kids. Within each school district, white kids do better than African American and Latino kids.

How did they get to this conclusion? For every school district in the US, the researchers used the Stanford Educational Data Archive to figure out 1) the median household income within each school district and 2) the grade level at which the students in each school district perform (based on federal test performance).

This piece also provides multiple examples for use within the statistics classroom. Highly sensitive examples, but good examples none the less.

-Most obviously, this data provides an easy-to-follow example of linear relationships and correlations. The SES:school performance relationship is fairly intuitive and easy to follow (see below)

From the New York Times: Positive linear relationship between parental SES and performance on standardized tests

-The data is provided in an interactive format. You can type in the name of a given school district so that you can see where that school district falls in the scatter plot. This makes this example interactive and more applicable to your students' lives and experiences. Below, I have highlighted the school district in the city where I teach.

-The data provides a good example for explaining between group and within group differences. As discussed, between school districts, high SES students outperform low SES. However, within school districts, white students out perform black and Hispanic students (see below: Here, the data is divided by school district as well as white, Hispanic, and black students within each district).

From the New York Times: SES x test performance x race

So, there is a lot to unpack here. A lot of sensitive stuff to unpack. is all illustrated with interactive scatter plots that beautifully illustrate correlation and linear relationships.

I think caution should be used with this example. You can also delve into issues of race. The data demonstrates, time and time again, that if you break up data by ethnicity, regardless of SES, white students perform better than Latino and African American students. There are many historical/SES issues related to underperformance among African-American and Latino students. If you are going to share the data related to these issues, I think that it is worth the time to address these so that racial stereotypes aren't used to explain this data (the authors of the NYTs piece do a good job of doing so).