«Paper initially prepared for presentation at the 41st Annual Conference of the Research and Planning Group, Track 1 – Student Learning Outcomes and ...»
The Impact of Class Size on Student Success:
The Importance of Controlling for Instructor and Course Characteristics
Paper initially prepared for presentation at the 41st Annual Conference of the Research and Planning
Group, Track 1 – Student Learning Outcomes and Success, Santa Barbara, CA, April 30-May 2, 2003
Recent policy debates have focused on class size as an explanatory variable that may, or may not, influence student achievement. This research paper reviews the literature on class size, and examines the impact of class size on successful course completion when controlling for instructor and class specific variables, the gender and ethnicity of a student, the reading abilities of students, the time a course is offered, and the length of time an instructor has been teaching.
Results indicate that across the sections taught by the same instructor, class size does not have a statistically significant impact on successful course completion. Students are as likely to earn a successful grade in a 20-seat section as an 80-seat section. The most important explanatory variables for student success are whether the student appeared to be “transfer directed” (a fulltime student), and whether the student had sufficient reading skills. White students performed slightly better in the courses, raising questions about the possibility of latent ethnic biases on the part of the instructor, but further research is needed. The findings suggest that for a narrow subset of classes in one social science discipline (taught by one instructor), class size has little meaning for explanations of why some students are more successful than others.
Matt Wetstein, Interim Dean Francisco Mora Planning, Research, and Grants Development Research Analyst San Joaquin Delta College San Joaquin Delta College 5151 Pacific Avenue 5151 Pacific Avenue Stockton, CA 95207 Stockton, CA 95207 209-954-5472 209-954-5472 firstname.lastname@example.org email@example.com
INTRODUCTIONPublic policy debates over class size reduction in the K-12 educational system have stimulated a growing body of research on the connection between class size and student achievement. That scholarship presents a mixed bag of findings. In this study, we seek to examine the inconclusive findings in the literature, and we analyze the impact of class size on student achievement in an isolated community college context. Our research controls for instructor specific effects and finds almost no evidence of class size effects on a simple measure of student achievement. In the pages that follow, we summarize the literature, set out our research model, and present new results that stress the importance of controlling for instructor specific variables when discussing class size effects.
While much has been written about class size and student achievement, there is no scholarly consensus on the issue. However, more scholars than not suggest a direct connection between the two variables. In a meta-analysis of research up to the early 1980s, Glass et al.
(1982) maintained that the evidence suggested small classes were associated with higher levels of achievement across all grades at the primary and secondary level. The connection between improved learning and class size became particularly important when class sizes were reduced below 20 students (see Glass et al. 1982; Pritchard 1999).
Some of the most compelling evidence on the connection between class size and student achievement has come from Tennessee’s experiment with class size reduction, and the systematic tracking of student performance after the initiation of that program in 1985 (Finn and Achilles 1999; Pritchard 1999). The Tennessee study involved 79 schools, more than 7,000 students, and a random assignment process to control for school level and curricular effects. The
counterparts in large classes on standardized exams; that the impact was even larger for minority students in early stages of the experiment; and that the impact of small classes in early primary grades had a lasting impact that persisted beyond five years (Pritchard 1999, 4; Nye, Hedges, and Konstantopolos 1999, 127). Krueger and Whitmore (2001) discovered that the impact of the Tennessee program could even be traced to ACT and SAT college entrance test taking patterns and student scores, with more noticeable increases for black students.
California’s experiment with reduced class sizes began in 1997, with an investment of $1 billion targeted at reducing class sizes to 20 or fewer at the first grade level, and in subsequent grades in later years (CSR 2002). The California initiative has generated some data suggesting a modest positive relationship between smaller primary classes and standardized test scores, however implementation issues like the inability to hire enough qualified teachers may threaten greater success (CSR 2000, 2-3; CSR 2002). Recent budget problems in the state make it unlikely that the California experiment with class size reduction will continue at the same level.
While studies have documented success stories, other research has concluded that smaller classes have only a small effect on student learning, or no proven influence at all (Pritchard 1999, 2). For example, Eric Hanushek (1998) has argued that the evidence drawing connections between class size and student achievement is “meager and unconvincing” and that analysis of 25 years of aggregate data on pupil-teacher ratios suggests that while student teacher ratios declined dramatically between 1970 and the 1990s, there were no corollary increases in achievement scores on the National Assessment of Education Progress (NAEP, see Hanushek 1998, 1-9). Hanushek (1998, 1) believes that policymakers have rushed to embrace supportive findings in this field because class size reduction is a simple fix for public school problems. He maintains that micro-level variables like good teachers in specific class settings with specific
little variance in student achievement is explained by class size (Hanushek 1998, 35).
Despite the growing body of scholarship suggesting a link between K-12 class sizes and student performance, the literature on universities and community colleges is quite sparse. A search of the Educational Resources Information Center (ERIC) database for articles and reports listing “class size” in the title and “college” as a key word turned up only 15 documents between 1995 and the present. Within that group, one study notes that recent findings point to factors other than class size as being important to student learning at the college level (Gilbert 1995).
Factors like teaching effectiveness, instructor practices, and course organization tend to be more important than class size for higher rates of learning. A similar conclusion emerges from a study focused on community college students and their opinions about instruction (Lesser and Ferrand 2000). These researchers discovered that class size, grades given, and college and field rankings had no significant correlations with student opinions about the quality of instruction. In an unpublished study at Fullerton College (2001), Institutional Researcher Ken Meehan found no correlation between class size and rates of student retention and success.
Two recent studies have examined the influence of class size on student achievement in college level introductory courses (see Kennedy and Siegfried 1997; Borden and Burton 1999).
Kennedy and Siegfried (1997) examined whether class size had an impact on achievement levels in introductory economics courses. Using a standardized national test from 1988-89, they found that class size had no significant association with achievement scores in their economics courses, even when controlling for student characteristics like SAT scores. In the Borden and Burton study (1999), five years of data from introductory Math and Sociology courses were used to examine whether course completion and course grades were influenced by class size. They found only a “small, overall negative impact of increasing class size on student grades and
providing a micro-level analysis that controls for some course specific factors in a community college setting. Most notably, the study controls for possible teacher impacts by isolating the analysis on one specific instructor teaching a limited set of courses over a five-year period. By using one instructor’s courses, we can control for variables that Hanushek (1998) and others have identified as significant: variation in teaching quality, presuming that all things being equal, an individual brings relatively constant levels of “quality” to the various section she/he teaches.
Thus, an analysis of one instructor can hold constant this potential variable. Additionally, the study aims to examine whether certain course characteristics like class size, time of the course offering, and a time trend variable might have an impact on student achievement.
The research proposes a multivariate logistic regression equation that explores the impact of various independent variables on successful course completion in introductory Political Science courses at a California community college. While class size is the variable of most interest to the study, other factors control for the demographic features of students (to explore potential ethnic and gender differences), the nature of a course (in terms of U.S. Government versus other Political Science topics), time of day the course was offered, and a time trend variable that attempts to assess whether the instructor has become “tougher” with age. Two control variables are also built into the study for student skill levels and transfer intention.
The study focuses on one professor’s courses over a five-year period. In this case one of the authors of this study served as a “guinea pig” for analysis. The period under review begins in 1996 with the start of the author’s teaching tenure at San Joaquin Delta College to the spring of 2001 (1996-2001). Complete data are available on 1,578 students who enrolled in the classes
instructor might become a “tougher grader” over time, a count variable was included in the study for each successive semester. Thus, courses in the first semester were scored “1,” with each successive semester increasing in value by one. This particular variable captures the consistency (or inconsistency) of the instructor’s grading method. If the coefficient is positive, it means that with time the instructor is becoming an easier grader. Conversely, if the coefficient is negative, it implies that the instructor is becoming a tougher grader over time.
Ethnicity is a variable intended to capture which ethnic groups do better in the author’s classes. It could also imply whether or not the instructor favors certain ethnic groups by giving students from that group higher grades. Ethnicity is measured by a variable identifying whether the student is white (a score of 1), or non-white/unknown race (0). If the coefficient comes out positive it means that the likelihood of successful completion is positively associated with being white. Conversely, if the coefficient is negative the likelihood of success decreases as a result of being non-white. Moreover, gender is used in the analysis to see if higher percentages of men or women are successful in completing a Political Science course, and to explore the possibility that the instructor may have certain sexist biases in one direction.
A variable measuring the time of day a course was taught is included in the model. This variable is designed to capture any impacts that might come from taking a class in the afternoon, as opposed to a morning section. The hypothesis behind this measure is that students taking a course later in the day might have a drop off in their scores compared against morning students, essentially because of “study fatigue” and a loss of alertness that comes from the passing of the day. In short, the expectation was a negative relationship between successful course completion and a variable measuring time of day. Scores on this indicator range between 7.50 (for a 7:30 am class) and 13.50 for a 1:30 pm class).
we call “transfer directed behavior.” We use reading assessment test scores as a crude proxy for academic skill levels, assuming that stronger readers have an obvious advantage of over weaker readers. Thus, we expect a strong positive relationship between reading scores and successful completion of a course. Similarly, we examine course-taking patterns of students during the semester of their enrollment in the Political Science courses. By tracking the number of units enrolled, we are able to determine whether a student is “transfer directed” and taking a full time load of 15 units or more, or whether the student is taking only a part time load. We hypothesize that transfer directed students would have greater odds of completing the Political Science courses, simply because their course taking patterns suggest a seriousness of purpose about completing the college’s general education requirements, and necessarily, a greater probability of transfer in subsequent years.
Of course the variable of most interest is the measure of class size and how it might affect successful course completion. Given the findings in some of the literature, the expectation was that larger class sizes would produce lower probabilities of student success. Thus, we expected to find a statistically significant relationship between class size and course completion.
The variables in the model are listed in Table 1, with their mean, minimum, and maximum values. The mean value for successful completion (found at the bottom of the table) was.51, suggesting that 51 percent of the students were able to complete the courses with a grade of A, B, C, or CR.
The statistical model used in this study is derived from a logistic regression analysis in the SPSS software package. Logistic regression was used because of the dichotomous nature of the dependent variable (see Aldrich and Nelson 1984).