Statistical analysis can be a powerful tool in understanding social, educational, psychological, and developmental processes. In cases where it is impossible or impractical to collect data on every individual, classroom, teacher, and school of interest, statistical analysis allows us to examine data on a sample of individuals (or classrooms, schools, etc.) in order to infer patterns in a larger population. For example, we might want to examine data on achievement test scores and per-pupil spending for a sample of schools to determine whether there is an association between spending and achievement patterns in the population. Or we might want to examine the association between race/ethnicity and achievement patterns. Moreover, if we find such an association, we might wish to ask additional questions, such as whether race/ethnic differences in achievement patterns can be accounted for by race/ethnic differences in family socioeconomic characteristics or in school quality.
In this course we learn to answer such questions using regression analysis—a statistical tool that allows us 1) to describe average patterns of association among multiple variables observed in a sample and 2) to make inferences about the patterns of association among these variables in a population. Regression analysis is a powerful statistical method with many variations. Our goal in this course is to develop an understanding of the basic methods, including their limitations, and to develop skill in using regression analysis to answer educational research questions. Finally, because an important part of any analysis is communicating the results to an audience, we will also place considerable emphasis on learning to present (in writing, tables, and figures) the results of regression analyses. By the end of the semester, students in this course should be sufficiently skilled in regression analyses that they can critically examine published research using regression and can carefully perform their own analyses.
EDUC 7456: Multilevel Models
Why study multi-level models? It turns out that most human behavior takes place in nested settings. Here are just a few examples: Children nested in families, students nested in teachers (and in turn in schools), households nested in counties; school districts nested in states; repeated observations over time nested within individuals, etc. The fundamental phenomenon of interest in much behavioral research involves individuals being affected by the groups or organizations in which they live or to which they belong. This course is certainly methodological, but it’s also conceptual in the sense that we will develop a lens for understanding how social organizations shape people’s lives.
Multilevel models (MLM), also known as hierarchical linear models (HLM), random-effects or random-coefficient models, variance component models, and (generalized) linear mixed models, are used when the units of observation (e.g., students) are grouped within clusters (e.g., schools). In such clustered data, observations for the same cluster cannot be assumed to be mutually independent for given covariate values as required for conventional regression models. Longitudinal or repeated measures data can also be thought of as clustered data with measurement occasions clustered within subjects; hierarchical models for longitudinal data are also known as growth curve models. This course will consider the statistical foundations of hierarchical linear models and focus on their application in behavioral and social research.
EDUC 7326: Quasi-Experimental Design for Causal Inference in Social Sciences
Assessing the causal effects of social and educational policies and practices is one important aim of educational and social science research. Educational researchers may want to know, for example, what effect a particular teaching practice has on student learning, what effects accountability policies have on teaching practices, or what effect early childhood education programs have on school readiness, and so on. Sociologists may want to know what the effect of certain neighborhood conditions are on child development, or what the effects of social networks are on individual behavior.
Historically, however, much educational and social science research has not been designed in such a way as to allow researchers to make credible causal inferences about the effects of educational and social practices and policies. In part, this is because many quantitative studies in education and the social sciences are essentially correlational in nature—they may show that there are statistical associations among sets of policy and practice variables and outcomes, but they do not provide convincing evidence of the causal linkages among these variables.
In recent decades, however, the so-called counterfactual or potential outcomes model (also called the “Rubin Causal Model”) and related developments have dramatically changed the way that social scientists have thought of causality. The new causal framework is not so much a set of technical models, but a precise logical framework for thinking about causality—and what constitutes evidence of causality—in the social sciences.
This course introduces students to a toolkit of quantitative methods to enable them to make valid causal inferences, particularly in the absence of a true randomized experiment. The methods covered in the course include 1) randomized experiments, 2) instrumental variables; 3) the use of natural and quasi-experiments; 4) longitudinal methods, including comparative interrupted time-series methods and difference-in-differences methods; 5) regression discontinuity; 6) matching estimators, including propensity score matching; 7) fixed effects estimators, and 8) value-added models. These methods offer considerable power to researchers interested in generating convincing and credible evidence of casual effects.
Miscellaneous Thoughts on Teaching
The core of my passion for teaching is simple: I like to take materials or concepts that seem impenetrable and translate them something transparent and conquerable. This motivation comes from my own—often painful—experiences with math classes growing up, which led me to believe I was “not a math person.” The self-misperception stayed with me through college and didn’t get corrected until I was required to take quantitative methods courses in graduate school. Once there, I discovered a deep passion for statistics and econometrics that drives me to this day. I feel lucky that I had a few great professors who gave me the chance to discover a field I love and overcome a counter-productive self-narrative. In this way, great teachers changed the course of my life, which also inspires my interest in the power of teachers. Unfortunately, not all students get a second chance at confronting their educational fears.
This kind of experience has been shown to be particularly salient for women and minority students. Indeed, women are less likely to take advanced science courses in high school and choose science- or math-related careers (Charles 2005). Previous research has shown that women’s avoidance of math and science careers may stem from their performance in mathematics and science in high school and even at the onset of formal schooling (Entwisle & Alexander 1990; Hyde, Fennema, & Lamon 1990; Busch 1995; Downey & Vogt-Yuan 2005). However, research suggests that this phenomenon is less related to actual skill and instead accounted for by underlying psychological experiences in the classroom. Stereotype threat occurs when a person who belongs to a group that has a negative stereotype attached to it (say, women in math classes) subconsciously conforms to the negative stereotype by performing a task to a lesser degree than they would otherwise. This may play a crucial role in women’s’ seeming underperformance in and aversion to historically “male” subjects (see, e.g., Kiefer & Sekaquaptewa 2007, Nosek, Banaji,& Greenwald, 2002).
It has become very common for schools of education to require all doctoral students to complete a year-long sequence in statistical training, so that they can become informed consumers of research regardless of methodological approach used. The CU-Boulder School of Education is one such program that has made a commitment to strong methodological training for all students, regardless of background or degree program. However when students return to doctoral programs in education, they often have spent a fair bit of time away from academia and sometimes have gone years without participating in math-intensive coursework. In addition, graduate students choose to pursue a doctorate in education for a very wide variety of purposes, many of which have little to do with quantitative research. These factors can lead to particularly high levels of anxiety surrounding statistical coursework embedded within education graduate program. In turn, that anxiety may produce a barrier to deep engagement with the material.
I adopt several strategies to try to provide many avenues of access for students of all levels in my introductory courses. One example strategy is the use of iClickers—remote-like devices—that students bring to class. The instructor can then design multiple choice questions and solicit anonymous responses to relevant to course content. The iClicker System allows for active participation by all students and provides immediate feedback to the instructor about any confusion or misunderstandings of the material being presented. The main potential benefits of iClickers are (a) to increase student engagement in class and (b) to provide an anonymous but direct line of communication between the instructor and students who feel hesitant or scared to reveal their struggles with the material.
I bring iClickers into my classes primarily to address the fact that I’ve often found that students are hesitant to reveal when they’re struggling. Students sometimes “suffer silently” and assume they are the only one not “getting it.” I begin my courses by encouraging my students to assume any confusion that will arise stems from the instruction, not the learner. Instructors can make mistakes, explain things poorly, or assume prior knowledge they shouldn’t. I tell them that I want every student to feel comfortable speaking up if something is unclear. That said, I understand that advanced statistics courses can involve a certain level of vulnerability, and some students may prefer to use the clicker to interact anonymously. I have found that, in my course evaluations, there are always a few students who tell me the iClickers helped them interact during class even when they didn’t necessarily feel comfortable raising their hand.