Teaching Statistics

This class was a challenge because I had 70 students. That meant I had to figure out a way to effectively teach while maintaining a manageable workload for myself, and to do so in a way that still upheld the principles of my teaching philosophy (which, if enacted in a vacuum, would not involve 70-person classes, or at the very least would have a smaller lab-type component).

  • Statistics is a required course for Sociology majors (and also fulfills UNH students’ quantitative proficiency requirements). Because some of my students will be going on to take Research Methods in the Sociology Department, I had to be sure to equip them with introductory statistical skills for potential quantitative analysis they would then engage in. This included teaching them a statistical software program. In our department, some faculty use SPSS and others use STATA. I decided to use SPSS because I think the ability to use menus instead of code makes it more intuitive for beginners. I also found a textbook I liked that offered SPSS Demonstrations with each chapter, walking through SPSS procedures step-by-step, and that also had videos of these demonstrations on the textbook website. With 70 students, we would not fit in a campus computer lab, so having this textbook resources was very helpful for students.
  • One of my goals was for students to see the value of using statistics – how statistical analysis enables researchers to unveil patterns that are not always easily discernible when you have some or all raw data. This goal was reinforced throughout the semester as we explored social data. Another goal of mine, which exists with some tension with this first one, was for students to critically evaluate statistical analyses or claims that were presented to them. As my class was held during the election season, there were plenty of popular media articles that reported on sample polling data, with corresponding headlines and analysis, and then tucked away a margin of error in the last paragraph that challenged their claims. Another method I used to have students both understand the utility of statistics as a research method for sociological inquiry as well as to be able to critically evaluate claims was for students to do group reading research presentations. Students signed up on a Wiki for a date and an article. The articles used statistical methods we had already covered in class at that point in the semester. Students presented the article to the class, focusing on the included tables, which involved interpreting these tables as well as explaining what method the author used and why. These presentations also helped reinforce and concretize some of what we had learned for students. You can find the assignment here.
  • The textbook I selected was Social Statistics: Managing Data, Conducting Analyses, Presenting Results, by Thomas J. Linneman. I found the book approachable. It also included advanced techniques not often covered in introductory textbooks, which often end right when things are starting to get really interesting and enable the researcher to do evaluations that will lend themselves to causal claims. Our class definitely dove into the basics, but we also spent considerable time on bivariate and multivariate linear regression, and I also exposed students to and they engaged with hierarchical linear regression, reference group and dummy variables, beta coefficients, interactions, and students were exposed to but had the option (for extra credit) of exploring logistic regression.
  • Throughout the class, I used real data that I thought was interesting (e.g., considering the timing with Secretary Clinton’s bid for president, the General Social Survey question about whether women are suited for politics is an interested dependent variable, and it was revealing for students to see how even a substantive portion of women and liberals think women are not suited for politics) and/or that I thought students would find interesting (e.g. provocative ones like number of sex partners or student data that we surveyed on the spot from presidential preferences to number of pets).
  • While the physical layout of the class was clearly designed for a lecture-style class without much student movement, I tried to incorporate engaging activities when possible.
    • On the first day we did my usual Introductory Sociology first day in-class experiment in which students evaluate candidates for student government, only to learn afterwards that everyone has the same candidate statement except half the class’s statements were for Emily and half were for Greg, though with a larger class we could differentiate by both race and gender (I also added in Lakisha and Tyrone); after analyzing and discussing our data, we take a look at this similar study that was completed on a bigger scale. This was an introduction to both the ability to see patterns from grouping and aggregating data, and starting to talk about the stories data tells, its distribution, etc.
    • For learning about sampling distribution, I passed around dice for students to roll and record their number during our do-now. Students then came up and 1) put an X on the whiteboard to help us construct a frequency table of the rolls (which would trend towards being a flat amodal distribution), and 2) put a folded piece of paper with the number they rolled into a bowl. This bowl of die rolls constituted our target population. Students then came up and each picked out five papers from the bowl, recorded them, but the papers back, and then calculated the mean of the numbers they had randomly selected. Students then constructed a frequency table on the board using these sample means. In this way we were able to put together a sampling distribution, which as more students came up to record their mean value became more normal looking, and which also enabled us to talk about p-values.
    • While we immediately scaled back the next class to multivariate analysis with only two independent variables, our first transition was a more complicated exposure. I divided the class into 8? or 10? groups, and each group got a handout with an SPSS output of a bivariate relationship (two groups had duplicates, and everyone had the same dependent variable). Groups then came up and presented their answers and interpretations to questions about the output (which was up on the slides at the same time for the rest of the class to see). I then talked through the logic of performing a linear regression with more than one independent variable. We combined all the independent variables the groups had looked at into one model, and that was our first dive into multivariate linear regression.
    • To talk about sample size, distribution graphs, and normal distribution and skew, we started by students counting their heart beats over a specific period of time (we did 15 seconds), recording the number on a sticky note, and then bringing up their sticky notes row by row to construct the distribution graph.
    • We used Smarties for learning about odds and odds ratios.
    • Students completed four problem sets, available HERE and exemplar responses available HERE. Partially because this was such a big class, I borrowed an idea from a professor who sometimes teaching undergraduate statistics, and gave students the option of working alone on their problem set or working with one other person (but only turning in one problem set for the pair). As the semester progressed and students grew more and more comfortable with SPSS and our statistical analyses, the problem sets gave them more and more freedom to choose their own variables and test out their own research questions.
    • I administered three tests and a final exam. They were multiple choice, but rigorous. I personally do not tend to be a big fan of traditional tests for summative assessments (these came with dehumanizing bubble sheets, called “exam scanning sheets,” for which there is a place to right your ID number and user ID / IT ID, but not your name (maybe to keep your name/id correspondence separate)). However, it seemed the most practical pathway forward. That being said, part of teaching is meeting students halfway, or meeting them where they are and then pushing them a bit. Students are accustomed to tests, and I saw students putting in a lot of effort in advance of tests to genuinely review the material and concepts and prepare for the exam. This active learning is quite beneficial.
  • Other items of note:
    • One way I like to demonstrate a conceptual (as opposed to just formulaic) understanding of what mean or average is is to use some object that is in unequal piles and redistribute it / balance it out so that each pile is equal. I think this semester I used crackers.
    • Here’s a great TEDx talk on the difference between correlation and causation.
    • Check out this Guess the Correlation game!
    • I’m still left wondering why students don’t like box plots as much as I do!?!