A Guide to Education Research in the Era of NCLB

Download Report

Transcript A Guide to Education Research in the Era of NCLB

A Guide to
Education Research
in the Era of NCLB
Brian Jacob
University of Michigan
December 5, 2007
How has the environment for
education research changed?
•
•
•
•
NCLB: evidenced-based programs
Accountability
Tight state and local budgets
Heightened oversight by foundations
Outline
• What are the different types of
education research, and what are the
goals of each?
• What distinguishes good evaluation
research from bad research?
• What are some of the common
challenges to collaboration between
education researchers and
practitioners?
Types of Education
Research
• Basic/Descriptive
• What teacher characteristics do principals value most highly?
• How does the gender gap in student achievement change over
the course of schooling?
• Program Evaluation
– Formative (Process Analysis)
• Which aspects of a curricular intervention work well and which
aspects need to be refined?
– Summative (Impact Analysis)
• Does a new curriculum have a positive impact on student
achievement?
How to Tell a Good Impact
Analysis when You See It
• Good intervention
– Well-defined question
– Coherent and explicit theory of action
• Good research design
– Causal inference
• Good implementation
– Attrition, contamination
• Good analysis/interpretation
– Magnitude
– Generalizability
The Problem of Causal Inference
• What is the casual impact of a particular program or policy on
the outcome of interest?
– Do teacher induction/mentoring programs reduce teacher attrition?
– Does computer-aided curriculum improve student achievement?
• We need to know what would have happened in the absence of
the program (i.e., the “counterfactual”)
• We often start with a correlation
– Students who attend magnet schools score higher than students in
traditional schools.
• But this correlation may not reflect a “causal” impact
– Many potential “confounding” factors
– Students who attend magnet schools are more motivated and have
more supportive families than those who attend traditional schools.
Threats to Causal Inference
• Selection
– Students/teachers/schools who participate in a program are
systematically different than those who do not participate.
– Example: families that choose to send their children to
charter schools; teachers that engage in professional
development.
• Concurrent events
– Concurrent events may have been responsible for the effects
that are attributed to the program.
– Example: High school reforms in Chicago implemented at the
same time as accountability provisions.
• Maturation
– Naturally occurring changes over time may be confused with
a treatment effect
– Example: Schools that were improving because of new
professional development adopt a tutoring program.
Common Research Designs
• Example: Success for All (SFA) is implemented in 12
out of 32 elementary schools in a district in 2001-02.
• Matched Comparison Group
– Compare test scores in SFA schools with “similar” non-SFA
schools in years after 2001-02.
• Pre/Post Design
– Compare test scores in SFA schools after 2002 with test
scores in the same schools before 2002.
• Pre/Post with Matched Comparison Group
– Compare test score changes from, say, 1996 to 2005 in SFA
schools with changes over the same time period in the
“similar” non-SFA schools.
Randomized Control Trials (RCTs):
The Gold Standard
• Randomly assign some students/teachers/schools to receive the
treatment, and others to not receive the treatment.
• Randomization assures that the treatment and control groups
are equivalent in all ways - even in ways that one cannot
observe, such as “motivation,” life circumstance, etc.
– Avoids concurrent event/maturation concerns since both treatment
and control group should experience these effects
• Decision about the level of random assignment depends on the
nature of the treatment, and the possibility of “contamination” of
the control group.
– One-on-one tutoring: student level random assignment
– Curriculum or pedagogy: classroom or school random assignment
• Some policies/programs cannot be evaluated via RCTs.
– Example: the competition effects of school choice
Some Concerns with RCTs
• Ethical Concerns: Should we deny treatment to some
students/schools?
– Assumes that we know that the program is effective
– If there are limited resources, then randomization is arguably the
“fairest” method for allocating the treatment.
– Many research designs can ensure equitable distribution of the
program while, at the same time, maintaining random assignment.
• Group 1: 3rd grade classes get the curriculum in year 1, and then 5th
grade classes get the curriculum in year 3
• Group 2: 5th grade classes get the curriculum in year 1, and then 3rd
grade classes get the curriculum in year 3.
• Logistical Concerns: How disruptive would it be for
schools/districts to conduct random assignment?
– Depends on the context, but often not very disruptive
• Professional development program with existing student tests
– Requires evaluations to be planned at the same time that the
program is implemented, and not merely attempted after the fact.
Good Research Worries
about the Problem of Attrition
• Attrition occurs when participants (e.g., students,
teachers) leave the school/district prior to the
conclusion of the study.
– Difficult to collect outcome data for these individuals.
– Differential attrition is when members of the treatment group
leave the study at a different rate than members of the
control group.
• If those who “attrit” are different in important ways,
this can bias the results of the study - even in RCTs.
– Example: “lottery” studies of magnet/charter schools
– Many students who lose the lottery leave the public school
system. If the most “motivated” lottery losers leave, one
would overstate the benefits of attending a magnet/charter
school.
Good research also …
• Pays attention to the magnitude of the
effects, and not just the statistical
significance.
• Looks at how effects change over time.
• Addresses the generalizability of the results
(external validity)
• Explores variation across students, teacher
and/or school subgroups
• Discusses the limitations of the study
Common Challenges for any
Research Collaboration
• Planning for evaluation in advance rather than
conducting it after the fact
• Concern about denying treatment
• Obtaining consent from principals, teachers and
parents
– Financial incentives; give information back to schools
• Respecting the organization of schools
– Be aware of school schedules and the academic calendar
• Respecting the time constraints of school and district
staff
– Use existing tests when possible; limit survey length
Conclusions
• Good research is becoming increasingly
important in education
– And can be very useful for practitioners (case
studies)
• Good research requires advance planning
and collaboration between researchers and
practitioners
• Collaboration is possible, as shown in the
following case studies