Monday, May 30, 2016

MOOCs for statistics/research methods instructors

MOOCs aren't just for current students. I think they can serve as professional development for faculty members as well. I don't have time for a MOOC during the school year, but I am committing to doing one this summer.

I think that instructors can approach MOOCs in two ways: 1) professional development, and 2) a search for improved pedagogy.

As professional development, learn a new statistical skill or freshen up a dormant one. Learn R. Learn Python. Freshen up on your non-parametric skillz. Take a course on data mining or using statistics in order to gain business insights.

Unofficial documentation of your course progress is typically offered just by taking the course. Official documentation/grade reports are usually available for a reasonable fee (my husband has taken a few such philosophy courses and paid around $50 for the official documentation).

Another way to use these courses: Don't take them to learn new skills, take them to learn new ways to teach your old content. Steal a good discussion board prompt. Find a new text book. Discover a useful interactive website that explains effect size or confidence intervals.

Some universities advertise their own MOOC lists, and I welcome you to Google around and find different classes. Here are the data science/statistics courses for two of the big MOOC organizations, Coursera and EdX.

Hackathorn, Ashdown, & Rife's "Statistics that Stick: Embedding Humor in Statistics Related Teaching Materials"

Hackathorn, Ashdown, & Rife just shared some great resources for using humor to teach statistics. 
In their own words, "This resource consists of a 21-page word document that reviews literature on the use of humor in teaching, describes an instrument for assessing the use of classroom humor, and offers tips on using two additional resource features specific to teaching statistics: (a) 42 visual jokes and cartoons, organized by 12 statistical topics, and (b) 12 slide presentations."

You guys. It is a collection of hilarious jokes and memes to use when teaching. As well as some scholarly work about using humor to teach. Here is a link that will download the .zip file to your computer. Here is a link to STP's Office of Teaching Resources in Psychology's Teaching Resources. Scroll on down to the Statistics, Research, and Teaching header to find this resource.

A few samples of the cartoons they included:

Monday, May 23, 2016

John Oliver's "Scientific Studies" with discussion quesions

This hilarious video is making the rounds on the Interwebz. Kudos to John Oliver and his writing team for so succinctly and hilariously summarizing many different research problems...why replication is important but not rewarded, how research is presented to the public, how researchers over-reach about their own findings, etc.  I Tweeted about this, but am making it cannon by sharing as a blog post.

Note: This video has some off-color humor (multiple references to bear fellatio) so it is best suited to college aged students.

I will use this in my Online and Honors classes as discussion prompts. Here are some of the prompts I came up with:

1) In your own words, why aren't replications published? How do you think the scientific community could correct this problem? 
2) In your own words, explain just ONE of the ways in which a RESEARCHER can manipulate their own data and/or research findings. It should be one of the methods of manipulation described within the video. Also, don't just name the method of manipulation, explain it like you would explain it to a friend so that they could become aware of the issue AND know how to spot the problem. 
3) Given what you have learned in this video AND your own experiences, who/what do you think is the most to blame for spreading bad science? 
4) Given your response to item 3, describe one way to correct for this problem of misinterpreted data being shared inappropriately. 

5) Why are replications important?

6) What major shortcoming of the "champagne" study was glossed over by the media? What major shortcoming of the "chocolate/pregnancy" study was glossed over? What is the difference between how study authors handle limitations of their work versus how the media handles shortcomings in their work? 

7) What were the red flags from the "hydration" study. Which do you consider to be the most damning and why?

BONUS POINT: Come up with a catchy pick-up line using the spotty Oxytocin research described in the clip.

Additionally, here is another one of my blog post (with links to other posts) related to the topic of scientific reporting.

Monday, May 16, 2016

Cheng's "Okcupid Scraper – Who is pickier? Who is lying? Men or Women?"

People don't always tell the whole truth on dating websites, embellishing the truth to make themselves more desirable. This example of how OK Cupid users lie about their heights is a good example for conceptually explaining null hypothesis testing, t-tests, and normal distributions.

So, Cheng, article author and data enthusiast, looked through OK Cupid data. In this article, she describes a few different findings, but I'm going to focus on just one of them: She looked at users' reported heights. And she found a funny trend. Both men and women seem to report that they are taller than they actually are. How do we know this? Well, the CDC collects information on human heights so we have a pretty good idea of what average heights are for men and women in the US. And then the author compared the normal curve representing human height to the reported height data from OK Cupid Users. See below...

From, by Fangzhou Cheng 

From, by Fangzhou Cheng

I can think of a number of ways to use this example:

-Null hypothesis testing/effect sizes, in general: Do you control and experimental groups overlap? By how much? Essentially, we are more likely to find significance/large effects the less they overlap. These two figures demonstrate this idea pretty nicely.

-A conceptual example of one-sample t-test. The CDC can provide us with a given number representing average male or female height, which is our known mean/mu. We could then test that number against all of the male or female heights reported by OK Cupid Users. Well, not really test, as we don't have the raw data, but it conveys the idea conceptually.

-This might even make a good example for Social or Evolutionary Psychology.

-Higher level statistics classes could also learn from the code he the author generously shared.

-I remember learning in graduate school that men typically round up when researchers ask them their number of sexual partners, and women typically round down. We can add height to the list of things that people fib about, especially within the context of seeking out a dating partner.

More of Cheng's work can be viewed here.

Wednesday, May 11, 2016

Electronic Conference on the Teaching of Statistics

I hope you are all finishing your grading and enjoying the first bit of your summer vacations (those of you who teach at colleges that run on semesters, at least!). I know that the last thing you want to think about right now is professional development. Especially professional development related to the teaching of statistics. That being said, I wanted to remind everyone that the Electronic Conference on the Teaching of Statistics (eCOTS) is happening next week (May 16-20).

As a mom and generally exhausted person, I really appreciate the chance to attend a conference in my pajamas. Here is the program. It only costs $25, includes workshops and poster sessions, and recordings from the conference. So, even if you aren't mentally ready to tackle something like this right now, you could view the materials at a later date.

Monday, May 9, 2016

NPR series on Neonatal Abstinence Syndrome

My son, Artie, resting in the NICU
When my second son was born via emergency c-section, he spent a week in the NICU out of an abundance of caution. It wasn't fun, but Artie pulled through just fine. He is a fat, happy four month old now.

While we were there, I found out that many of the other NICU babies there were suffering from neonatal abstinence syndrome (NAS). They were born addicted to drugs. And those poor babies howled for hours as they were being weaned off of drugs and helped by the staff.

NPR's All Things Considered recently did a series about national efforts to help end NAS. Two of the segments from this series are potential leaning moments for statistics and RM classes. One discusses efforts to use proper research methodology to create the ideal treatment recommendations for NAS babies. The second discusses governmental efforts to use systematic data collection to better track NAS babies and get to the root of the problem.

1. Using clinical research to better treat NAS babies

NAS babies are born addicted to drugs and need to be weaned off of those drugs after birth. Detox is awful. To aid these babies, they are typically given smaller and smaller dosages of morphine or methadone in order to ease the detox process. However, there are no established guidelines for dosages. This NPR story briefly describes the research design for a proposed study that hopes to a) figure out which is better for babies: Methadone or morphine via b) longitudinal research. As I psychologist, I think it is also worth noting that this research has been proposed by developmental psychologists.

As such, I think this can be used in a statistics or RM methods class to demonstrate a real world problem and how psychologists are conducting research to better understand this problem. This particular example would also be suited to a developmental psychology class or a psychology class dedicated to drug abuse and addiction. I think this example also illustrates larger issues related to why this hasn't been studied yet: Research on infants already has many restrictions, rightly so. Imagine what it is like to write the IRB proposal related to dosing infants with morphine and methadone, even when it is a medical necessity.

2. Using systematic data collection to better understand NAS

As a Pennsylvanian, I'm especially pleased to hear this story from NPR about my state's efforts to better track data regarding children born with NAS. It seems like it should be very straight forward to track these babies, but there hasn't been a unified, organized way to do so. However, they are going to add this condition to the list of diseases, like TB and whooping cough, that the Department of Health specifically tracks with the help of physicians.

Why should we collect such data? To better understand the problem. A tracking program in Tennessee uncovered the fact that many of the mothers of NAS babies are receiving their drugs via perfectly legal prescriptions. This insight, provided by data, may lead to new avenues for preventing NAS by better educating doctors on pain management for pregnant women.

I think this can be used in class as an example of applied statistics within medicine. It also demonstrates that just because it might seem intuitive to collect certain data does not mean that such data is being collected. It also shows how such data can inform policy and problem solving (like the approach of educating doctors about precautions to take when prescribing opiates to pregnant women).

Monday, May 2, 2016

Ben Schmidt's Gendered Language in Teacher Reviews

Tis the season for end of semester teaching evaluations. And Ben Schmidt has created an interactive tool that demonstrates gender differences in these evaluations. Enter in a word, and Schmidt's tool returns to you how frequently the word is used in Rate Your Professor evaluations, divided up by gender and academic discipline.
Spoiler alert: Men get higher ratings for most positive attributes!
...while women get higher ratings for negative attributes. 

Out of class, you can use this example to feel sad, especially if you are a female professor and up for tenure.

In class, this leads to obvious discussions about gender and perception and interpersonal judgments. You can also use it to discuss why the x- and y-axes were chosen. You can discuss the archival data analysis used to generate these charts. You can discuss data mining. You can discuss content analysis. You can also discuss between group differences (gender) versus within group differences (academic area).

You could also use this to generate some data for classroom analysis. If you cursor an data point in a chart, you can see the exact number of instances of that word per millions of word of text. You could make your students enter all that data and run a t-test for a specific word, or by an academic area (say, a bunch of positive words just for male and female  psychology professors). Or, you could collect data for multiple words AND academic areas and make an ANOVA out of it.