Gifted Child Quarterly
57(3) 197–204
© 2013 National Association for
Gifted Children
Reprints and permissions:
sagepub.com/journalsPermissions.nav
DOI: 10.1177/0016986213490022
gcq.sagepub.com
Methodological Brief
In gifted education research, logistic regression analysis
(also known as logit analysis) has been used to analyze
binary outcomes.
1
For example, logistic regression has been
used to identify factors (e.g., family, school, environment,
and individual) associated with high-achieving and under-
achieving gifted students (Baker, Bridger, & Evans, 1998;
McCoach & Siegle, 2003), investigate variables related with
gifted students dropping out (Renzulli & Park, 2000), and
describe the phenomenon of bullying among gifted students
(Peterson & Ray, 2006). Binary outcome variables (also
known as Bernoulli variables) represent two categories indi-
cating that an event has occurred (e.g., 1 = student dropped
out, 0 = student did not drop out) or that a characteristic is
present (e.g., 1 = high achieving, 0 = underachieving).
Predicting giftedness (1 = yes, 0 = no), though we realize
that giftedness can be measured on a continuum of abilities,
has been modeled using logistic regression as well
(Konstantopoulos, Modi, & Hedges, 2001). However, unlike
linear regression results, interpreting and presenting logistic
regression coefficients may be more challenging for applied
researchers (Liberman, 2005; Long, 1997).
The issues with using ordinary least squares (OLS) regres-
sion with binary outcomes have been well researched (Long,
1997). While decades ago, using OLS to model the relation-
ship of predictors with binary outcome variables was accept-
able, it is considered unacceptable today by most researchers
(Allison, 1999). Using traditional linear regression to model
binary outcomes affects the model parameter estimates and
standard errors as a result of linear regression model viola-
tions (Long, 1997). Using traditional OLS regression may
also result in impossible or nonsensical predicted values that
are greater than one or less than zero. This Methodological
Brief presents an overview of logistic regression (LR), pro-
vides an example of interpreting logistic regression results,
and closes by discussing multilevel extensions.
Data Source and Measures
Our data came from the National Education Longitudinal
Study of 1988 (NELS:88; Institute of Education Sciences,
U.S. Department of Education, http://nces.ed.gov/surveys/
nels88/). We limited our analysis to public school 10th-grade
students (n = 4,832) with nonmissing variables of interest.
As our analysis is for pedagogical purposes only, we did not
use weights (see Hahs-Vaughn, 2005, for a primer on using
weights with national datasets) or account for the nested data
structure (i.e., students within schools). Our dependent vari-
able (DV) was if a student had ever been in an Advanced
Placement (AP) course (F1S34E; 1 = yes, 0 = no). Our inde-
pendent variables (IVs) of interest were the following: if a
student had been enrolled in a gifted and talented class in the
eighth grade (BYS68A), gender (1 = female, 0 = male), and
socioeconomic status (SES). SES is an NELS developed
composite (M = -0.09, SD = 0.75, min = -2.23, max = 1.91)
created using parents’ level of education, occupation, and
household income. The specific research question of interest
490022GCQ XX X 10.1177/0016986213490022Gifted Child QuarterlyHuang and Moon
research-article 2013
1
University of Virginia, Charlottesville, VA, USA
Corresponding Author:
Francis L. Huang, Curry School of Education, University of Virginia, PO
Box 800785, Charlottesville, VA 22908-8785, USA.
Email: flhuang2000@yahoo.com
What Are the Odds of That? A Primer on
Understanding Logistic Regression
Francis L. Huang
1
and Tonya R. Moon
1
Abstract
The purpose of this Methodological Brief is to present a brief primer on logistic regression, a commonly used technique when
modeling dichotomous outcomes. Using data from the National Education Longitudinal Study of 1988 (NELS:88), logistic
regression techniques were used to investigate student-level variables in eighth grade (i.e., enrolled in a gifted class, gender,
and socioeconomic status) that were associated with taking an Advanced Placement course in the 10th grade. Through the
use of the NELS:88 data, the authors provide an example of how to interpret both categorical and continuous independent
variables and illustrate how model fit can be assessed.
Keywords
quantitative methodologies, policy/policy analysis, assessment
by guest on April 21, 2016 gcq.sagepub.com Downloaded from