Content uploaded by Lisa Heschong
Author content
All content in this area was uploaded by Lisa Heschong on Mar 28, 2019
Content may be subject to copyright.
CALIFORNIA ENERGY
COMMISSION
Daylighting In Schools:
Reanalysis Report
TECHNICAL REPORT
October 2003
500-03-082-A-3
Gray Davis, Governor
CALIFORNIA
ENERGY
COMMISSION
Prepared By:
Heschong Mahone Group
Lisa Heschong, Project Director
Fair Oaks, California
Managed By:
New Buildings Institute
Cathy Higgins, Program Director
White Salmon, Washington
CEC Contract No. 400-99-013
Prepared For:
Donald Aumann,
Contract Manager
Nancy Jenkins,
PIER Buildings Program Manager
Terry Surles,
PIER Program Director
Robert L. Therkelsen
Executive Director
DISCLAIMER
This report was prepared as the result of work sponsored by the
California Energy Commission. It does not necessarily represent
the views of the Energy Commission, its employees or the State
of California. The Energy Commission, the State of California, its
employees, contractors and subcontractors make no warrant,
express or implied, and assume no legal liability for the
information in this report; nor does any party represent that the
uses of this information will not infringe upon privately owned
rights. This report has not been approved or disapproved by the
California Energy Commission nor has the California Energy
Commission passed upon the accuracy or adequacy of the
information in this report.
ACKNOWLEDGEMENTS
The products and outcomes presented in this report are part of the Productivity and Interior
Environments research project. The reports are a result of funding provided by the California Energy
Commission’s Public Interest Energy Research (PIER) program on behalf of the citizens of
California. Heschong Mahone Group would like to acknowledge the support and contributions of the
individuals below:
Heschong Mahone Group, Inc.: Principal in Charge: Lisa Heschong. Project Director: Lisa
Heschong. Project staff: Puja Manglani and Rocelyn Dee.
Subcontractors: Jack A. Paddon and James L. Engler of Williams + Paddon Architects + Planners
Inc., Marshall Hemphill of Hemphill Interior technologies, and James Benya of Benya Lighting
Design.
Review and Advisory Committee: We are greatly appreciative of the following people who
contributed to the review of this report: William Beakes of Armstrong Industries, Jerry Blomberg of
Sunoptics, Pete Guisasola of City of Rocklin Building Department, Rob Samish of Lionakis
Beaumont Design Group, Michael White of Johnson Controls, Chuck McDonald of USG, John
Lawton of Velux, John Mors of Daylite Company, Joel Loveland of Lighting Design Lab, Anthony
Antonelli of Ecophon, Steve Fuller and Martin Powell of Albertsons, Jehad Rizkallah of Stop and
Shop, Paul McConocha of Federated Departments, JimVan Dame of My-Lite Daylighting Systems
and Products, Doug Gehring of Celotex, Ivan Johnson of TriStar Skylights, Robert Westfall of
Solatube International Inc., Leo Johnson of PJHM Architects, George Loisos of Loisos/Ubbelohde
Architects, Jim Kobs of Chicago Metallics, Steve Ritcher of Crystollite, Jackie Stevens of So-
Luminaire, Peter Turnbull of PG & E, Sean Flanigan of WASCO Products, Richard Schoen of
Southern California Roofing, Mike Toman and Jeff Guth of Ralphs and Food for Less, and Lori
Johnson of Target.
Project Management: Cathy Higgins, New Buildings Institute; Don Aumann, California Energy
Commission.
PREFACE
The Public Interest Energy Research (PIER) Program supports public interest energy research and
development that will help improve the quality of life in California by bringing environmentally safe,
affordable, and reliable energy services and products to the marketplace.
This document is one of 33 technical attachments to the final report of a larger research effort called
Integrated Energy Systems: Productivity and Building Science Program (Program) as part of the
PIER Program funded by the California Energy Commission (Commission) and managed by the New
Buildings Institute.
As the name suggests, it is not individual building components, equipment, or materials that optimize
energy efficiency. Instead, energy efficiency is improved through the integrated design, construction,
and operation of building systems. The Integrated Energy Systems: Productivity and Building Science
Program research addressed six areas:
Productivity and Interior Environments
Integrated Design of Large Commercial HVAC Systems
Integrated Design of Small Commercial HVAC Systems
Integrated Design of Commercial Building Ceiling Systems
Integrated Design of Residential Ducting & Air Flow Systems
Outdoor Lighting Baseline Assessment
The Program’s final report (Commission publication # P500-03-082) and its attachments are intended
to provide a complete record of the objectives, methods, findings and accomplishments of the
Integrated Energy Systems: Productivity and Building Science Program. The final report and
attachments are highly applicable to architects, designers, contractors, building owners and operators,
manufacturers, researchers, and the energy efficiency community.
This attachment (#A-3) provides supplemental information to the program’s final report within the
Productivity and Interior Environments research area. It includes the following report:
Daylighting in Schools: Reanalysis Report. This study expands and validates previous
research by Heschong Mahone Group that found a statistical correlation between the amount
of daylight in elementary school classrooms and the performance of students on standardized
math and reading tests.
The Buildings Program Area within the Public Interest Energy Research (PIER) Program produced
these documents as part of a multi-project programmatic contract (#400-99-413). The Buildings
Program includes new and existing buildings in both the residential and the non-residential sectors.
The program seeks to decrease building energy use through research that will develop or improve
energy efficient technologies, strategies, tools, and building performance evaluation methods.
This report is Attachment A-3 (Product 2.2.5) to the Final Report on Integrated Energy Systems:
Productivity and Building Science Program (Commission Publication #P500-03-082). For other
reports produced within this contract or to obtain more information on the PIER Program, please visit
www.energy.ca.gov/pier/buildings or contact the Commission’s Publications Unit at 916-654-5200.
All reports, guidelines and attachments are also publicly available at www.newbuildings.org/pier.
ABSTRACT
The “Daylighting in Schools: Reanalysis Report” is part of the Productivity and Interior
Environments research project, one of six research elements within the Integrated Energy Systems:
Productivity and Building Science Program. The Program was funded by the California Energy
Commission’s Public Interest Energy Research (PIER) Program.
This study expands and validates previous research by Heschong Mahone Group that found a
statistical correlation between the amount of daylight in elementary school classrooms and student
performance. The researchers reanalyzed student performance data from two school districts to
answer questions raised by the previous study. The reanalysis found that:
Elementary school students in classrooms with the most daylight showed a 21% improvement
in learning rates compared to students in classrooms with the least daylight.
There was no teacher assignment bias that might have skewed the original results; more
experienced or more educated teachers were not significantly more likely to be assigned to
classrooms with more daylighting.
The daylighting effect does not vary by grade.
Physical classroom characteristics (daylighting, operable windows, air conditioning, portable
classrooms) do not have an effect on student absenteeism. This seems to contradict claims
that have been made about the health effects of daylight or other environmental conditions, as
reflected in absenteeism rates of building occupants.
These results, which are consistent with the original findings, affirm that daylight has a positive and
highly significant association with improved student performance. These findings may have
important implications for the design of schools and other buildings.
Author: Lisa Heschong, Heschong Mahone Group
Keywords: Daylight, Productivity, Student Performance, Window, Skylight, Absenteeism,
Attendance, Health, Classroom Condition, School Design
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT TABLE OF CONTENTS
i
TABLE OF CONTENTS
EXECUTIVE SUMMARY ___________________________________________V
1. INTRODUCTION _______________________________________________ 1
1.1 Study Tasks________________________________________________ 2
1.2 Literature Review of Research on Teachers' Influence_______________ 2
1.2.1 Differences with Our Study _______________________________ 5
1.3 Summary of Previous Study ___________________________________ 5
2. TEACHER SURVEY ____________________________________________ 9
2.1 Methodology _______________________________________________ 9
2.1.1 Survey Structure ______________________________________ 10
2.2 Teacher Characteristics______________________________________ 11
2.2.1 Years of Experience____________________________________ 11
2.2.2 Education Level, Certificates and Honors ___________________ 12
2.2.3 Classroom Preferences _________________________________ 14
2.2.4 Criteria for Classroom Selection __________________________ 15
2.2.5 Permanent vs. Portable Classroom Preference_______________ 17
2.2.6 Classroom Energy Management __________________________ 18
2.3 Conclusions _______________________________________________ 20
3. TEACHER BIAS ANALYSIS _____________________________________ 23
3.1 Hypothesis________________________________________________ 23
3.2 Methodology ______________________________________________ 23
3.2.1 Teacher Credentials____________________________________ 23
3.2.2 Assignment Bias ______________________________________ 24
3.2.3 Decision to Focus on Daylight Code Only ___________________ 26
3.2.4 Daylight Code as a Dependant Variable ____________________ 26
3.2.5 Teacher Assignment Bias Models _________________________ 27
3.3 Findings __________________________________________________ 27
3.4 Conclusion________________________________________________ 29
3.5 Discussion ________________________________________________ 29
3.5.1 Findings of Different Study Population Models _______________ 30
3.5.2 Conclusions of Different Study Population Models ____________ 33
4. GRADE LEVEL ANALYSIS______________________________________ 35
4.1 Hypothesis________________________________________________ 35
4.2 Methodology ______________________________________________ 36
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT TABLE OF CONTENTS
ii
4.3 Findings __________________________________________________ 36
4.4 Conclusions _______________________________________________ 38
5. ABSENTEEISM ANALYSIS _____________________________________ 39
5.1 Hypothesis________________________________________________ 40
5.2 Methodology ______________________________________________ 41
5.3 Findings __________________________________________________ 42
5.3.1 Absenteeism Findings __________________________________ 42
5.3.2 Tardiness Findings_____________________________________ 43
5.4 Conclusions _______________________________________________ 43
5.5 Discussion ________________________________________________ 44
6. RE-ANALYSIS CONCLUSIONS __________________________________ 47
6.1 Grade Level Analysis________________________________________ 47
6.2 Absenteeism Analysis _______________________________________ 47
6.3 Teacher Survey ____________________________________________ 47
6.4 Bias Analysis ______________________________________________ 48
6.5 Re-Analysis Report _________________________________________ 49
7. APPENDICES ________________________________________________ 51
7.1 Statistical Terminology ______________________________________ 51
7.2 Teacher Survey ____________________________________________ 53
7.2.1 Three Most Important Criteria in Selection of Classroom _______ 55
7.2.2 Permanent v. Portable Classrooms ________________________ 63
7.2.3 Additional Comments___________________________________ 67
7.3 Bias Analysis Models________________________________________ 71
7.4 Grade Level Models ________________________________________ 81
7.5 Absenteeism Models ________________________________________ 89
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT TABLE OF CONTENTS
iii
TABLE OF FIGURES
Figure 1: Daylight Code Definitions ___________________________________ 6
Figure 2: Surveyed vs. Original Population Distribution by Daylight Code ____ 10
Figure 3: Number of Years Teaching for Survey Respondents _____________ 12
Figure 4: Teacher Education Level __________________________________ 13
Figure 5: Teacher Certificates and Honors (Recognition) _________________ 14
Figure 6: Most Preferred Attributes of Classrooms ______________________ 17
Figure 7: Permanent vs. Portable Classroom Preference _________________ 17
Figure 8: Teachers' Energy Management of Classrooms _________________ 19
Figure 9: Teachers' Lighting Management of Classrooms_________________ 20
Figure 10- Correlation of Teacher Variables to Daylight Variables, Student level
Analysis, Capistrano __________________________________________ 25
Figure 11 - Change in Capistrano Math Model with Addition of Teacher
Variables ___________________________________________________ 28
Figure 12 - Change in Capistrano Reading Model with Addition of Teacher
Variables ___________________________________________________ 28
Figure 13: Surveyed, Original, and Expanded Populations ________________ 30
Figure 14: Daylight Affect for Different Populations, with and without Teacher
Variables, on Reading Tests in Capistrano _________________________ 31
Figure 15: Daylight Affect for Different Populations, with and without Teacher
Variables, on Math Tests in Capistrano____________________________ 31
Figure 16: Teacher Variables and Daylight effect on Reading for the Three
Populations Compared ________________________________________ 32
Figure 17: Teacher Variables and Daylight effect on Math for the Three
Populations Compared ________________________________________ 33
Figure 18: Capistrano Grade Level Models with Interactive Variables
Summary ___________________________________________________ 37
Figure 19- Seattle Grade Level Models with Interactive Variables Summary __ 38
Figure 20- Distribution of Absences and Tardies ________________________ 41
Figure 21- Equation for natural log of attendance data ___________________ 42
APPENDICES:
Figure 22 - Capistrano Reading Models, Original Population, with and without
Teacher Variables ____________________________________________ 71
Figure 23 - Capistrano Math Models, Original Population, with and without
Teacher Variables ____________________________________________ 72
Figure 24 - Capistrano Reading Model, Teacher Survey Population, with and
without Teacher Variables ______________________________________ 73
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT TABLE OF CONTENTS
iv
Figure 25 - Capistrano Math Model, Teacher Survey Population, with and without
Teacher Variables ____________________________________________ 74
Figure 26 - Capistrano Reading Model, Expanded Population, with and without
Teacher Variables ____________________________________________ 75
Figure 27 - Capistrano Math Model, Expanded Population, with and without
Teacher Variables ____________________________________________ 76
Figure 28 - Descriptive Statistics, Capistrano Original Population ___________ 77
Figure 29 - Descriptive Statistics, Capistrano Teacher Survey Population ____ 78
Figure 30 - Descriptive Statistics, Capistrano Expanded Population _________ 79
Figure 31- Capistrano Grade Level Interaction, Reading Daylight___________ 81
Figure 32- Capistrano Grade Level Interaction, Math Daylight _____________ 82
Figure 33- Seattle Grade Level Interaction, Reading Daylight ______________ 83
Figure 34 - Seattle Grade Level Interaction, Math Daylight ________________ 84
Figure 35- Descriptive statistics, Capistrano Grade Level, Reading and Math _ 85
Figure 36- Descriptive statistics, Seattle Grade Level, Reading ____________ 86
Figure 37- Descriptive statistics, Seattle Grade Level, Math _______________ 87
Figure 38 - Capistrano Absenteeism Model ____________________________ 89
Figure 39 - Capistrano Tardiness Model ______________________________ 90
Figure 40 - Capistrano Absenteeism/Tardiness Descriptive Statistics________ 91
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT EXECUTIVE SUMMARY
v
EXECUTIVE SUMMARY
This report is a follow-on study to the Daylighting in Schools study1 that was completed
in 1999, which found a compelling statistical correlation between the amount of
daylighting in elementary school classrooms and the performance of students on
standardized math and reading tests. This re-analysis of the original study data was
intended to answer key questions raised by the peer review of the earlier study, and
expand our understanding of methodological choices for further work.
The original findings potentially have very important implications for the design of
schools and other buildings where people live, work and play. Daylight used to be
common and even required in schools, homes and offices, but fully daylit buildings
became increasingly rare as electric lighting became more the norm. This re-analysis
study helps to provide greater certainty for the original findings.
For this re-analysis study HMG conducted four tasks:
The Teacher Survey collected information from a sample of teachers in the Capistrano
school district about their education and experience levels, preferences for classroom
features and operation of those features. The primary purpose of the survey was to
provide input to a subsequent "assignment bias" analysis. In addition, we learned some
useful information about teacher preferences, attitudes and behaviors in response to
classrooms conditions.
While the teachers we surveyed generally had a preference for windows, daylight and
views in their classrooms, these preferences were not found to be driving classroom
preferences. Far more important was an almost universal desire for more space, a good
location, quiet, lots of storage and water in the classroom.
Environmental control was also found to be an important issue for teachers, especially
for those who did not have full control. Teachers seemed to hold a basic expectation that
they would be able to control light levels, sun penetration, acoustic conditions,
temperature and ventilation in their classrooms. They made passionate comments about
the need for improvement if one or more of these environmental conditions could not be
controlled in their classroom.
The Teacher Bias Analysis further examined information from the Teacher Survey. The
survey data was coded into variables and statistically analyzed in relation to both
assignment to daylit classrooms and the student performance models. The goal of the
Bias Analysis was to discover if the original study had over-inflated the effect of daylight
on student learning by not accounting for a potential "assignment bias" of better teachers
to more daylit classrooms.
We conclusively found that there was not an “assignment bias” influencing our results.
None of the individual teacher characteristics we identified were significant in explaining
assignment to a daylit classroom in the Capistrano District. Considering all teacher
characteristics together only explained 1% of the variation in assignment to daylit
classrooms. We did find that a few types of teachers, those with more experience or
1 Heschong Mahone Group (1999). Daylighting in Schools. An investigation into the relationship between
daylight and human performance. Detailed Report. Fair Oaks, CA.
(http://www.h-m-g.com/Daylighting/daylighting_and_productivity.htm)
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT EXECUTIVE SUMMARY
vi
honors, were slightly more likely (1%-5%) to be assigned to classrooms with more
windows or some types of skylights.
When we added the teacher characteristics to the original student performance models,
the daylight variables were not reduced in significance. Further analysis of other sub-
populations repeated these findings. Among twelve models considered, we identified a
central tendency of a 21% improvement in student learning rates from those in
classrooms with the least amount of daylight compared to those with the most.
In the Grade Level Analysis, we re-analyzed the original student test score data for
both Capistrano and Seattle by separate grade level, instead of aggregating the data
across the four grade levels (2-5). Our goal was to determine if this method would more
accurately explain the relationship of student performance to daylighting. We tested for
statistical significance and correlation, and we looked at any patterns discovered in the
analysis.
The data did not show any significant patterns between a daylight effect and the
separate grade levels, neither an increase or decrease in daylight effects by grade level.
Thus, we conclude that there do not seem to be progressive effects as children get
older, nor do younger children seem to be more sensitive to daylight than older children.
Allowing the results to vary by grade did not noticeably improve the accuracy of the
models. Therefore, we conclude that looking at data across grade levels is a sufficiently
accurate methodology.
In the Absenteeism Analysis, we used absenteeism and tardiness data in the original
Capistrano data set as dependent variables and evaluated them against the full set of
explanatory variables from the original study, plus the new information on teacher
characteristics. These models would allow us to assess whether daylighting or other
classroom physical attributes potentially impacted student health, as measured by
changes in student attendance.
Student attendance data is certainly not the best indicator of student health. Yet to the
extent that attendance data does reflect student health, our findings do not suggest an
obvious connection between physical classroom characteristics and student health.
Notably, daylighting conditions, operable windows, air conditioning and portable
classrooms were not found to be significant in predicting student absences.
Overall, the strength of the daylight variable in predicting student performance stands
out sharply across all of these re-analysis efforts.
This analysis also demonstrated that the findings of these models are more strongly
dependent upon the sample population then the subtleties of the explanatory variables.
Thus, we believe that it will be more informative to replicate this study with a different
population, to continue to try to refine the models with further detail in the explanatory
variables.
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT INTRODUCTION
1
1. INTRODUCTION
The Daylighting in Schools study1 completed in 1999 by the Heschong Mahone
Group on behalf of the California Board for Energy Efficiency found some a
compelling statistical correlation between the amount of daylighting in elementary
school classrooms and the performance of students on standardized math and
reading tests.
The study was reviewed by a panel of experts, recruited by Lawrence Berkeley
National Laboratory and involved a wide range of disciplines related to the study.
In general the review panel was satisfied with the soundness of the basic
methodology and the rigor of the statistical analysis. An additional “classroom
level analysis” (included in the Appendix of the detailed version) verified the
robustness of the initial results. The peer reviewers, however, expressed two
primary concerns2 that could only be addressed in follow-up studies. These are:
1. The results might be confounded by a potential bias whereby "better"
teachers might be more likely to be assigned to more daylit classrooms
2. The analysis might be more accurate if performed by grade level,
rather than aggregating data from four grade levels together
The study described in this report, supported through the California Energy
Commission's Public Interest Energy Research (PIER) program, was designed to
address these two concerns, while also expanding other areas of our knowledge
about the interaction of students, teachers and daylighting. The series of four
tasks described in this report were the necessary first steps in resolving
remaining questions about the Daylighting and Schools study. The results of
these initial re-analysis studies will also be used to inform the methodology and
data collection for the forthcoming PIER productivity studies in schools, retail,
manufacturing, and offices.
This report discusses the re-analysis of the 97-98 school year student
performance data on standardized math and reading tests from the Capistrano
Unified School District in Southern California and the Seattle Public School
District in Seattle Washington. The re-analysis of the original study data was
intended to answer key questions raised by the peer review of the earlier study,
and expand our understanding of methodological choices for further work.
1 Heschong Mahone Group (1999). Daylighting in Schools. An investigation into the relationship between
daylight and human performance. Detailed Report. Fair Oaks, CA.
2 Heschong Mahone Group (1999). Daylighting and Productivity. An investigation into the relationship
between daylight and human performance. Review Report. Fair Oaks, CA.
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT INTRODUCTION
2
1.1 Study Tasks
Four study tasks were defined, which are briefly summarized here, and described
fully later:
• Teacher Survey
• Teacher Bias Analysis
• Grade Level Analysis
• Absenteeism Analysis
The Teacher Survey surveyed a sample of teachers in the Capistrano school
district to determine their years of teaching experience, education level, and
other characteristics that might be associated with being a "better" teacher. While
we were conducting a survey, we decided to include a few additional questions to
learn more about the teacher's perspective on classroom assignments, their
preferences for the physical qualities of classrooms, and how they operated their
classrooms.
The survey fed into the second task Teacher Bias Analysis. The teacher
information from the survey was coded into variables that could be analyzed
statistically. First we looked at the assignment bias, to see if some types of
teachers were more likely to be assigned to more daylit classrooms in the
Capistrano District. Next, we added the information about the teachers to the
original Capistrano student test score models to see if accounting for teacher
characteristics would impact the significance or magnitude of the daylight
variables.
In addition to the tasks described above, we also re-analyzed the original data in
two other ways. The Grade Level Analysis looked at the original student test
score data for both Capistrano and Seattle by grade level to see if this was a
more accurate way to study the relationship of student performance to
daylighting.
The original Capistrano data set also included information on student
attendance--both absences and tardiness. This gave us the opportunity to see if
daylighting, or other physical characteristics of the classrooms in Capistrano,
were associated with changes in attendance. For the Absenteeism Analysis
task, we set student absenteeism and tardiness as dependent variables, and
used the full set of explanatory variables used in the original study, plus the new
information on teacher characteristics, to see if daylighting or other classroom
attributes were associated with student attendance.
1.2 Literature Review of Research on Teachers' Influence
We looked to research by educational researchers in our effort to understand
how teacher characteristics might be described and included in our models.
Various educational researchers have analyzed the relationship between teacher
performance and student achievement, and have identified a number of teacher
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT INTRODUCTION
3
characteristics that seem to fairly reliably predict student learning in the
classroom. Factors that have been found to be significant in previous studies
include a teacher's general intelligence, teaching experience, certain personality
traits, knowledge of the subject matter, knowledge of teaching strategies,
continuing education, and certification1. The following summary is based on an
extensive literature review by Prof. Linda Darling-Hammond of the Stanford
University School of Education (Darling-Hammond 2000) of the recent research
on the relationship between teacher performance and student achievement. The
reader is referred to her report for specific citations or further detail on studies.
This literature review helped inform the classification of teacher characteristic
variables for in this study. The discussion below includes both the approach of
other researchers to define variables of interest and a brief summary of some of
their findings.
General intelligence: General intelligence as measured by IQ test or college
grade point average shows the weakest performance as a predictor of
subsequent student performance. While early studies in the 40's positively
correlated teachers' intelligence and student achievement, these correlations are
generally statistically insignificant and have not held up over time. Two meta-
reviews of these studies performed in the 80's found little or no correlation.
Teaching experience: Researchers have usually measured teaching experience
by the number of years a teacher has spent in the profession. While various
studies have found a positive relationship between teachers' experience and
student learning, this relationship is not always significant or linear. Although
many studies conclude that inexperienced teachers generally perform less well
than those with more experience, the benefits of experience tend to level off after
approximately five years. This seems, however, to be dependent on the
organizational structure of the school district: in districts that emphasize the
importance of continuing education, long time teachers are more likely to improve
throughout their career.
Teacher personality traits: Studies have found scant correlation between
student learning and various teacher personality traits. One exception is a set of
personality traits variously defined as "flexibility," "creativity," or "adaptability."
This would seem to be consistent with a theory that a teacher's ability to
creatively adjust their teaching methods to fit the needs of the students and the
instructional goals would correlate positively with student learning. Some
researchers have found that "flexibility" is also closely correlated to variables
measuring a teacher's professional education, implying that teachers who have
studied formally are more likely to be able to adjust teaching strategies for
students' different learning styles.
Knowledge of subject matter: Knowledge of the subject matter to be taught, as
measured by number of college classes taken or by scores on a subject matter
1 Darling Hammond, L. (2000). Teacher Quality and Student Achievement: A review of state policy evidence.
Education Policy Analysis Archives, Vol. 8, number 1, available on-line, http://epaa.asu.edu/epaa/v8n1/
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT INTRODUCTION
4
test, has been found to be less important than might be expected. A variety of
studies have shown small, statistically insignificant relationships, both positive
and negative. One recent study found that teachers' coursework in the subject
field relates positively to student achievement in mathematics and science, but
that the number of courses show diminishing returns above a certain threshold
level (Monk, 1994). A teacher's knowledge of the subject was found to be more
important for higher-level classes and higher-achieving students (Hawk, Coble, &
Swanson, 1985). Thus, a certain level of subject matter knowledge appears
important, but above that point, other factors, such as the ability to effectively
convey this knowledge, become more important to student achievement.
Teaching strategies: Knowledge of teaching strategies has been measured by
number of education classes taken in teaching methods and level of college
degree (BA or MA). These variables generally capture variance in teacher
performance more effectively than the variables discussed above. Ferguson and
Womack (1993) studied 200 graduates of one teacher education program. They
concluded that the amount of education coursework was responsible for more
than four times the variance (16.5 %) in teacher performance than measures of
content knowledge, as determined by National Teacher Examination subject
matter test scores and GPA in the major (4.5 %).
Continuing education: It is also seems to be important that teachers continue to
refresh and update their knowledge through continuing education. Greater
student achievement has been linked to mathematics teachers' opportunities to
participate in sustained professional development courses. Similar results have
been suggested for literature-based instruction. Not only is the amount of
ongoing education important, but also how recent it is.
Certification: Standard certification usually requires a teacher to graduate from
an accredited teacher training program, have a major or minor in the field to be
taught, and pass a test on basic skills and teaching strategies. Therefore,
certification status (standard certification vs. emergency, temporary or provisional
certification issued to those lacking the above credentials) is a measure of both
knowledge of the subject and of teaching skills. Linda Darling-Hammond
compiled data from all 50 states using the 1993-94 Schools and Staffing Surveys
(SASS) and the National Assessment of Educational Progress (NAEP). She
found that at the state level, the percentage of well-qualified teachers (with full
certification and a major in their field) was the strongest, consistently positive
predictor of student achievement (.61 < r < .80, p<.001) while the percentage of
newly hired, uncertified teachers was the strongest, most consistently negative
predictor of student achievement (-.63 < r < -.40, p<.05).
Scores on state licensing examinations: Another variable that combines
several important factors are scores on state licensing examinations, which test
both basic skills and teaching knowledge. Ronald Ferguson (1991) examined
900 Texas school districts, controlling for student background and district
differences, and found that a combination of teacher qualification variables –
scores on a licensing examination, education level, and years of experience --
explained more of the inter-district variation in students' reading and mathematics
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT INTRODUCTION
5
achievement gains than student socioeconomic status. The strongest of these
variables were the scores on the state licensing exam.
1.2.1 Differences with Our Study
These studies formed a context of our work. However, the goal of our study was
not to determine the effect of teachers’ credentials, qualifications, and experience
on student performance. Our goal, rather, was to discover whether daylighting in
classrooms remained a significant indicator of student performance even when
teacher characteristic variables were included in a statistical regression model.
Thus, our study differed from those discussed above in several important ways.
First of all, our data collection procedure of teacher variables was limited, due to
privacy concerns, to the variables we could reliably measure through self-
reporting. We had to exclude original sources such as transcripts, college or
licensing board test scores, or classroom observations.
Second, the data in other studies was often aggregated to the district or state
level. We, on the other hand, analyzed the data at the student and classroom
level, which may yield different results or emphasize different factors.
1.3 Summary of Previous Study
For the original schools study we identified three study sites of large school
districts that had a range of daylighting conditions in their classrooms. We
collected test scores and demographic information for all second through fifth
graders in the district, and classified their classrooms for the amount and quality
of daylight available. We choose to work with data on elementary school children
since they typically spend all year in one classroom. Thus, we could directly
isolate the effects of that one classroom. We also specifically selected districts
that had a number of classrooms lit from above with skylights or roof monitors
(“toplighting”). We reasoned that daylight provided through windows might have
a number of complicating factors, such as the quality of view, whereas daylight
provided from above typically had fewer other qualities that might influence
results, thus we would be more likely to be looking a pure “daylighting” effect.
The three districts were located in San Juan Capistrano, (Southern) California;
Seattle, Washington; and Fort Collins, Colorado. These three districts have very
different climates, different school building types, different curriculums and
different testing protocols. The districts also provided us with information about
student demographic characteristics, special school programs, size of schools,
etc.
We added information to these data sets about the physical conditions of the
classrooms to which these children were assigned. We reviewed architectural
plans, aerial photographs and maintenance records and visited a sample of the
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT INTRODUCTION
6
schools in each district to classify the daylighting conditions in over 2000
classrooms. Each classroom was assigned a series of codes on a 0-5 scale (see
Figure 1) indicating the size and tint of its windows, the presence and type of any
skylighting, and a holistic daylighting code indicating the overall quality and
quantity of daylight expected from both windows and toplighting combined. In
Capistrano, the skylights were given a variable type (A, AA, B, C, D) rather than
a scalar. The configuration of these skylight types is described in the original
report. The Daylight Code, which is used predominately for reporting findings in
this report, was based on the following qualitative criteria, with foot candle levels
at midday conditions are provided as an illustration rather than a criteria.
Daylight Code 5 Classroom is adequately and uniformly lit with daylight, such that
teacher could successfully instruct with electric lights off, for
most of the school year. 50± footcandles on most desks.
Daylight Code 4 Classroom has major daylight component, and could
occasionally be operated without any electric lights. Daylight may
have strong gradient. 30± footcandles on many desks.
Daylight Code 3 Classroom has adequate levels in limited areas, such as near
windows. Some, but not all, electric lights could occasionally be
turned off. 15± footcandles at some desks.
Daylight Code 2 Classroom has poor and/or very uneven daylight. Not likely to
ever operate without electric lights fully on. 10± footcandles in
limited areas.
Daylight Code 1 Classroom has minimal daylight. Very small and/or darkly tinted
windows or inadequate toplighting. Not possible to operate
without electric lights. 5± footcandles in limited areas.
Daylight Code 0 Classroom has no daylight.
Figure 1: Daylight Code Definitions
Ultimately the study analyzed test scores performance for 8000 to 9000 students
per district. We looked at both math and reading scores in all three districts, and
analyzed each separately, alternately using the holistic daylight code and the
separate window and skylight codes, for a total of twelve statistical models.
The Capistrano Unified School District proved to be our most interesting study
site for a number of reasons. The District administers standardized tests both in
the fall and spring, allowing us to compare the change in students’ math and
reading test scores while they spent the year in one classroom environment.
Because the District, like most in California, has a number of standardized
portable classrooms at every elementary site, we were able to use these
portables as a standardized condition controlling for the influence of individual
school sites or neighborhoods. We also collected additional information at this
district about the HVAC and ventilation conditions of the classrooms, which was
also included in the analysis.
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT INTRODUCTION
7
In Capistrano, using a regression equation that controlled for 50 other variables,
we found that students with the most daylighting in their classrooms progressed
20% faster on math tests and 26% on reading tests in one year than those with
the least. Similarly, students in classrooms with the largest window areas were
found to progress 15% faster in math and 23% faster in reading than those with
the least. Students that had a well-designed skylight in their room, one that
diffused the daylight throughout the room and which allowed teachers to control
the amount of daylight entering the room, also improved 19-20% faster than
those students without a skylight. Classrooms with a skylight that allowed direct
beam sunlight into the classroom and did not provide the teacher with a way to
control the amount of daylight were actually seen to have a negative association
with student performance. In addition, in three of the four Capistrano models, the
presence of an operable window in the classroom was also seen to have a
positive effect on student progress, associated with 7-8% faster learning. These
effects were all observed with 99% statistical certainty.
The Seattle and Fort Collins school districts administer only one standardized
test at the end of the school year. In these districts, the study used the final
scores on math and reading tests at the end of the school year and compared
the results to the district-wide average test score. In both of these districts we
also found positive and highly significant (99%) effects for daylighting. Students
in classrooms with the most daylighting were found to have 7% to 18% higher
scores than those with the least.
The three districts have different curricula and teaching styles, different school
building designs, and very different climates. And yet, the results of the studies
show consistently positive and highly significant effects. This consistency across
such diverse school environments persuasively argues that there is a valid and
predictable effect of daylighting on student performance.
These models explained from 25% to 44% of the variation in student scores (R2=
.25 to .44). Thus another 56% to 75% of the variation might be explained by
other factors not included in our equation such as teacher quality, home life,
health, nutrition, individual talents and motivation, etc. There always remains the
possibility that some other variable left out of the equation is influencing results
on the variable of interest.
Reviewers of the original school study specifically asked if “better” teachers were
more likely to be assigned to the more daylight classrooms, thus influencing the
results. Additionally, they asked if the analysis might be more accurate if
performed by grade level rather than aggregating data from four grade levels
together. This follow-on study addresses those concerns by re-examining our
most detailed models for the Capistrano district.
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT INTRODUCTION
8
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT TEACHER SURVEY
9
2. TEACHER SURVEY
The first task for the follow-up study was to collect additional information about
the teachers that could be added to the original models. We choose to work with
the Capistrano Unified School District for three reasons: they had provided us
with the most detail in the original study, they were willing to cooperate with us on
further studies, and they were physically the closest district to us.
2.1 Methodology
We asked the District the best way to compile additional information about the
specific teachers in the study that would be useful in our re-analysis. The District
was unable to provide us with information about their teachers directly due to
confidentiality restrictions. However, they agreed that we could solicit such
information from the teachers, in a survey. A survey gave the teachers an
opportunity to decline to participate, and allowed us to collect additional
information that could be kept confidential from the District.
We agreed that the District would review and approve the instrument, and also
help us to locate the teachers in our sample for distribution of the survey. A two-
page survey was developed and reviewed by the District and members of our
Technical Advisory Committee. A draft version was tested on a number of local
elementary teachers for ease of use and clarity.
The final survey, with a explanatory cover letter from the District office, was
distributed to a stratified sub-sample of teachers from our original data set. We
identified 14 schools with a balanced sample of all window and skylight
conditions found in the original 27 elementary schools included in the 97-98
database. Our goal was to achieve a sufficient population of teachers in each
daylighting condition, in order to have the best chance to achieve statistical
certainty in our new analysis. We provided the District a list of all teacher names
used in mapping the data for those 14 schools. The District then located these
teachers for us. Over the two year period, between the survey and the original
data mapping, about 17% of the teachers had left the district or moved to non-
teaching jobs and about 6% had re-located to a different school in the district. As
a result, our sub-sample of teachers now resided at every elementary school in
the district.
Surveys were mailed to each school office, with a list of teachers to whom they
were to be distributed. After two days, the responses were collected in a
confidential master envelope and returned to us for analysis. Ultimately, we
received completed surveys from 68% of the teachers on our distribution list, or
206 teachers, representing 3900 students in our data set. Some school offices
disregarded our list and distributed the survey to all of their teachers, so we
received responses from an additional 44 teachers who were not in our original
study, for a total of 250 responses.
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT TEACHER SURVEY
10
0%
10%
20%
30%
40%
50%
01233.544.55
Daylight Code
Pe rce ntage of Students
O r igin a l Surveyed
Figure 2: Surveyed vs. Original Population Distribution by Daylight Code
Figure 2 shows the resulting distribution of students by Daylight Code for the
surveyed population compared to the original population of the study. The two
populations are reasonably similar. There is a slight increase in the proportion of
teachers in the higher daylight codes (3.5+) due to our concern that our sample
include enough teachers to support statistically significant analysis. The
reduction in Daylight Code 2 reflects a lower sampling of teachers in portable
classrooms.
2.1.1 Survey Structure
The two-page survey instrument, provided in Appendix 7.1, contains both
structured and open-ended questions. The primary purpose of the survey was to
collect information about teacher characteristics that could be included in our
models of student performance in daylit classrooms. Thus, the survey first asked
for the classroom and grade assignment for both the current year and the 97/98
school year so that we could verify our data mapping. It then asked for the
teacher’s education level, certificates, additional coursework, special honors, and
years of teaching experience—in the current school, district and total.
In addition, we collected information about the teachers’ perception of any
“assignment bias,” their preferences for classroom selection, and additional
information about how they operated their classrooms. While this information was
not part of the primary intent of the survey, it was hoped that such information
might provide valuable insight in future analysis.
Thus, the survey was designed to answer the following questions:
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT TEACHER SURVEY
11
1. What are the educational qualifications and experience of the teachers
who taught in classrooms included in the 97-98 data set?
2. Did they believe that they have been allowed to choose their own
classroom or have any influence on where they are assigned?
3. If they could choose a classroom, what attributes of a classroom would
they give top priority in their selection?
4. How do these teachers operate the energy-using features of their
classrooms? For example, if they have operable windows, how often
do they open them?
2.2 Teacher Characteristics
The survey responses were categorized, cleaned and entered into a database.
Information from open-ended questions was coded for analysis. The teacher
characteristic information was eventually transformed into variables for inclusion
in the statistical models of later tasks in this study.
The Capistrano Unified School District tries to maintain uniformly high education
and training standards among its teachers, which tends to reduce the variation in
teacher quality across classrooms. In discussions with Capistrano administrators
prior to the survey, we were told that the District was not hiring teachers with
provisional or emergency credentials. Beyond requiring all of their teachers to be
certified, the district highly values continuing education for all teachers. A sliding
salary scale rewards additional college education, in addition to years of
experience. The District also provides opportunities for on-site training classes
that are specifically tailored to the curriculum needs of the district.
2.2.1 Years of Experience
The 250 teachers who responded to the survey varied in their teaching
experience from one or two years to more than 40 years. They averaged 11
years of teaching in the CSUD district and 13.5 years of teaching in total (see
Figure 3).
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT TEACHER SURVEY
12
Teachers Experience
0
5
10
15
20
25
30
35
40
45
No. of Years Teaching
Years teaching in district
Years teaching total
Figure 3: Number of Years Teaching for Survey Respondents
2.2.2 Education Level, Certificates and Honors
Teachers were asked to report their highest educational degree, plus additional
college course work, training programs, and special certificates and honors. This
information was described qualitatively by the teacher respondents, thus we
needed to classify the responses into meaningful categories that could be used
to analyze the data. The first step was to understand the educational
requirements for elementary school certification in California, and similarly the
District’s standards for hiring and promotion.
There are two levels of accreditation in California elementary schools. A
Preliminary Credential is good for the first five years of teaching. It requires as a
minimum completion of a bachelor's degree and a teacher preparation program,
knowledge of the US Constitution, plus additional certification in teaching
reading, passing a standardized test of knowledge (CBEST) and the multiple
subject assessment for teachers (MSAT). The second level of accreditation is
called the Professional Clear. It requires an additional fifth year of study beyond
the bachelor's including course work in computer, health and special education.
Based on interviews with the District personnel officers and review with our
Technical Advisory Group, we decided to group the teachers’ education levels for
analysis into two simple categories, BA and MA, with three sub-categories, as
follows:
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT TEACHER SURVEY
13
• "BA" indicated any teacher with a bachelors degree
• "Clear only" indicated teachers who had been teaching for 7 years or
more, but had not pursued any continuing education beyond that
necessary for their professional clear credential.
• "BA Plus" indicated teachers who listed college credits beyond the
minimum required for certification
• "MA" indicated those with a masters, or doctorate (one case)
• "MA Plus" identified teachers with college credits beyond a master's
degree.
In our sample of surveyed teachers (Figure 4), 58% had Bachelor degree, of
which 12% had only a BA and had taught for 6 years or less, 12% were grouped
in the Clear Only category, 34% were grouped in the Bachelor Plus category;
42% of the teachers reported having a Masters degree, of which 29% had just an
MA, and 13% were grouped in the Masters Plus category,
Teacher Education
MA Plus
13%
MA
29%
BA Plus
34%
Clear Only
12%
BA
12%
Figure 4: Teacher Education Level
In addition to their qualifications, teachers also reported other credentials that
identified if they have received any special certificates or honors. From this
information we defined two other analysis categories:
• The Certificates category included teachers, who reported special
certificates beyond those required for the CLEAR credential, such
as a certificate in bilingual or gifted and talented education.
• The Honors category grouped together all teachers who reported
special awards or honors, such as being named a mentor teacher
or Teacher of the Year.
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT TEACHER SURVEY
14
Figure 5 shows the proportion of surveyed teachers who were classified into
these two categories.
Teachers with Certificates
77%
Cetificates
23%
Teacher Recognition
60%
Recognition
40%
Figure 5: Teacher Certificates and Honors (Recognition)
2.2.3 Classroom Preferences
We pursued a number of different methodologies to understand if there was an
intentional or unintentional bias in assigning some teachers to more daylit
classrooms. In our original study we had interviewed administrators and
principals in the district, who assured us that there was no obvious mechanism or
practice of assigning "better" teachers to more daylit classrooms. Given the rapid
growth of the district, frequent reassignment of classrooms to accommodate new
school openings and added portable classrooms tended to randomize teacher
classroom assignments on a fairly regular basis. In addition, it was reported that
each school site follows its own administrative criteria in assigning teachers to
classrooms, using criteria such as clustering of grade levels or special interest
teaching teams.
From the Teacher Survey we found a slightly different story. Of the teachers
surveyed, 32% felt that they may have had some influence on the selection of
their classrooms within the past year (a yes answer to Question 14) and 41%
answered yes or maybe. Similar percentages reported that they may have had
past influence. Thus, the teachers seemed to feel that they could influence
classroom selection.
When asked to indicate their top criteria for selection of a classroom, if they were
to have a choice (Question 15), 8% of the sample ranked windows or natural light
as their top criteria, and 27% mentioned windows, natural light or view within
their top three choices. Lumped together, these three criteria would have placed
fourth in importance as a classroom selection criteria, after classroom size
(53%), convenient location (36%), and storage capacity (30%). (See Figure 6
and discussion in Section 2.2.4 below.) Thus, while windows and associated
qualities light natural light and view are important to teachers, they are not the
most important criteria that teachers claim drive their choices.
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT TEACHER SURVEY
15
In addition to the structured questions, teachers were given the opportunity to
write any comments they wished. Over two hundred, or 80%, took the
opportunity to write informative comments, while three complained about not
enough time to respond. (See Appendix 7.1.3 to read the un-edited comments)
Their comments read as a loud plea for better physical conditions in the
classroom. The reader should realize that many of the comments are referring to
class-size reduction measures that were instituted in the District to increase the
number of teaching spaces, but unfortunately have compromised physical
comfort and control. The passion for control of physical conditions--lighting,
acoustics, ventilation and thermal comfort--is also very evident in these
comments. The list of comments should make compelling reading for anyone
managing or designing school facilities.
2.2.4 Criteria for Classroom Selection
The survey, in an open-ended question, asked what were the three most
important criteria that the teacher would use to select a classroom, if they were
given the choice. We grouped the qualitative responses into the following
categories, reported in the order of their frequency of mention within the top three
criteria:
• Size indicated teachers’ preference for larger classrooms and was
most frequently listed in the top three criteria, mentioned by 53% by
respondents. It was also the most frequently listed as the top
preference.
• Location of the classroom within the school layout was the second
most common criteria in determining their classroom choice (36%),
and was also second as the top criteria. The location preferences
included close proximity to the school entrance, administrative offices,
playground, library, or other elements of the school plan.
• Storage space inside the classroom in the form of closets or cupboards
was the third most mentioned criteria.
• Water or the availability of a sink in the room was among the top four
most mentioned criteria. Comments typically emphasized the primary
importance of water in the classroom for student hygiene, and
secondarily for class projects.
• Quiet captured criteria such as “lack of noise” and “being in a quiet
zone.” It was the fifth most common criteria (23%) mentioned in any of
the top three preferred classroom attributes by teachers, and third
criteria in terms of teachers’ top preference (after classroom size and
location).
• Windows were mentioned by 20% of the respondents.
• HVAC indicated a preference for air conditioning in the classroom, or
control of temperature, or acceptable thermal comfort conditions.
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT TEACHER SURVEY
16
• Door indicated a preference for full enclosure or the availability of a
door to close off the classroom from other activity areas.
• Proximity indicated a preference for a classroom close to particular
colleagues, either by grade level or shared teaching responsibilities.
• Condition indicated a preference for better physical conditions, such as
new paint, furniture or carpet, or good maintenance.
• Ventilation indicated a preference for fresh air or good air circulation.
• Lighting indicated preference for a good lighting quality in the
classroom or control of the lighting levels.
• Natural light indicated a preference for natural light from windows or
skylights.
• Walls indicated a preference for lots of wall surfaces for display.
• Bathroom indicated a preference for a bathroom close by.
• Views indicated a preference of a good view from the classroom.
• Whiteboards indicated a preference for lot of whiteboard surfaces.
• Phone indicated a preference for a telephone available in the
classroom.
• Workroom indicated a preference for being adjacent to a teacher
workroom.
Classroom Selection Criterea
0%
10%
20%
30%
40%
50%
60%
size
locatio n
storage
water
quiet
w indow s
HVAC
door
proximity
condition
ventilation
lighting
natural light
w alls
bathroom
view s
w hiteboards
phone
w orkroom
Percentage of Responses
Top 3 choices
1st preference
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT TEACHER SURVEY
17
Figure 6: Most Preferred Attributes of Classrooms
It should be noted that the teachers' preferences for classroom features is largely
a function of what options are, or are not, currently available to them. For
example, a teacher in a school without windows but the option of moving to a
portable with a window may rank windows very high, while a teacher in a
classroom with large windows but no sink, may rank access to water highest.
Thus, we interpret these results to be particular to the context of the Capistrano
Unified School District and the status of current facilities.
2.2.5 Permanent vs. Portable Classroom Preference
The use of portable classrooms in California was mandated by the state for a
number of years as a strategy to accommodate rapidly shifting population
growth. As a consequence, every school site in our Capistrano study had a
substantial number of portables. Portable classrooms have also come under
recent scrutiny for possible poor indoor air quality or other health concerns such
as mold growth. A number of state and national studies are currently trying to
assess the health implications of portable classrooms. Our 1999 study did not
find any negative student performance impacts associated with portables.
Indeed, our models tended to find positive, but not statistically significant
impacts, associated with being in a portable classroom, once we controlled for
daylight, ventilation and all other variables in our equation. To learn more about
teacher’s perceptions of portables we included a question about preference of
portable or permanent classrooms in the survey (Question 16). The answers and
associated comments are fully presented in Appendix 7.1.2.
Permanent vs. Portable Preference
portable
15%
no answer
2%
permanent
68%
no opinion
15%
Figure 7: Permanent vs. Portable Classroom Preference
Sixty eight percent of the teachers surveyed preferred to teach in a permanent
classroom rather than a portable one. (Figure 7). Thirty percent of the
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT TEACHER SURVEY
18
respondents were divided equally between those who preferred portable
classrooms or reported no preference for either type.
48% of the teachers that preferred portables mentioned that the closed walls of
the portable solved the noise and distraction problems found in the permanent
classrooms of their school created by an open classroom plan or poor acoustics.
24% preferred portables because they were larger than the permanent
classrooms available at their school. Remaining comments mentioned the
presence of air conditioning, better bulletin boards, and better physical condition.
Teachers who preferred permanent classrooms had a much wider range of
reasons why. Larger size, better location, better amenities, less noise were
frequently mentioned. One teacher summed up a preference for permanent
classrooms in the comment: “Feels substantial and lets children know they are
important and that things are not temporary.” 22% of teachers preferring
permanent classrooms specifically mentioned indoor air quality concerns, such
as moldy or musty smells and increased incidence of allergies or colds in
portables.
2.2.6 Classroom Energy Management
In the survey, teachers were asked how they operated a number of energy using
features in their classrooms. The data that we have for the Capistrano
classrooms merely indicates the presence of a feature, such as operable
windows, not whether or how it is used. This set of questions was intended to
provide insight into how their might actually use these features, and provide
some baseline data, admittedly self-reported, that might allow us to estimate the
energy impacts of various features.
Figure 8 highlights the percentages of teachers’ responses for the ten energy
statements surveyed. Positive percentages indicate actions taken, while negative
percentages indicate inability to act, or no action.
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT TEACHER SURVEY
19
Classroom Energy Management
25%
46%
54% 55%
6%
24%
5% 9% 12% 2%
-60%
-41%
-22%
-10%
-42%
-11%-11%
-9%
-5%
-9%
-60%
-40%
-20%
0%
20%
40%
60%
80%
100%
adjust thermostat
open door (ventilation)
darken room (video)
close w /d for noise
som e lights off
open w indow (ventilation)
all lights off
block sun
fan on
dr aw curt ains
% of re spons es
>10x /y ear
>10x/w eek
Nev er Do
Can ' t Do
Figure 8: Teachers' Energy Management of Classrooms
HVAC control: Over 50% of the teachers’ surveyed reported adjusting the
classroom thermostat on a weekly basis, and almost 90% of them reported doing
this more than 10 times/school year (about monthly).
Acoustic control: Over 80% of the teachers occasionally close the windows or
doors (“close w/d for noise”) to avoid high noise levels from the outside, and 55%
do this frequently.
Ventilation control: 46% open the outside door for ventilation purposes on a
weekly basis and 84% do this at least 10 times every year. 25% of the teachers
surveyed reported doing this on a daily basis. More than 40% of the teachers
surveyed reported they can’t open a window for natural ventilation, while 42% of
the total sample open a window at least 10-times/school year. 12% of the
teachers report using a portable fan, which probably means they brought in their
own personal fan that they purchased themselves to solve a perceived ventilation
problem in their classroom.
In the comments section, one teacher summarized the teaching challenges faced
with in small, poorly ventilated portable: “The students do not have enough
space to move around. Most large projects are eliminated because of lack of
space and no access to water. The room is so small that we use the ramp
outside to set up centers. The door is always open because the poor circulation
in the room gets us sick. We have no water to wash our hands after sneezing
and coughing…we get sick more often and pass colds, flu to each other because
of our close proximity.”
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT TEACHER SURVEY
20
Lighting control: Darkening the room for TV or video is also very common, done
by over 80% occasionally and 25% frequently. Turning some or all lights off is
also a fairly common activity, while taking measures to block the sun, or close
curtains is much less frequent.
Figure 9 shows further detail on teachers' management of the electric lighting in
their classrooms. This graph shows 54% of the teachers turning some of the
lights off, and 37% of the teachers turning all of the lights off, at various
frequencies during the school year.
Teacher's Operation of Classroom Lights
0%
5%
10%
15%
20%
25%
30%
35%
40%
45%
50%
not possible
never do
occassionally, <10x/yr
often, 10+/year
often, 10+/year, per weather
very frequently, 1+x/wk
constantly, 1+x/day
all lights off
some lights off
Figure 9: Teachers' Lighting Management of Classrooms
2.3 Conclusions
The results of the teacher survey on preferences and operation of classrooms
suggest that daylighting and operable windows are indeed important to teachers,
but tend to be secondary to their most pressing concerns, such as adequate size,
location, and water (hygiene) availability in classrooms. Acoustic, thermal and
visual comfort and adequate ventilation are all frequently listed as top priorities.
The optional comments response to the survey was overwhelming. 98% of the
teachers surveyed took the time to write about what was good and bad in their
classrooms. The passion put into the comments on physical comfort in
classrooms makes it clear that teachers are very stressed by any type of poor
physical condition in classrooms where they must work every day with 20-30 very
active children. “Please help California get more square footage per child. It’s
crazy!” pleaded one. “Teaching … without running water makes me feel like it’s
the 1900’s. We carry pails of water!” exclaimed another. One teacher concluded
about the need for cross ventilation: “I believe it is good for myself and students
to breathe in some fresh air. It helps us all think.” While some teachers report
being pleased and comfortable with their classrooms, a sizable group feel they
have overwhelming physical challenges in their classrooms that routinely
interfere with their ability to teach.
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT TEACHER SURVEY
21
There are clearly some important energy use challenges revealed in the survey
that should be carefully considered by school designers and facility managers. In
Figure 8 it is clear that Capistrano teachers are actively trying to increase the
ventilation of their classrooms by opening doors, opening windows, and adding
portable fans. Furthermore, 54% claim to be adjusting the thermostat at least
once a week and 55% also claim to be closing windows or doors at least once a
week specifically to control noise in the classroom, implying that they had
previously opened them, most likely for ventilation. This suggests that teachers’
driving desires for good ventilation, thermal comfort and acoustic comfort tend to
be in conflict with the options allowed by their physical environment. Increasing
ventilation is likely to also increase ambient noise in the classroom and/or reduce
thermal comfort. One teacher summarized this problem with the comment: “I like
being able to adjust the a/c, heat and ventilation. The down side of this is the a/c
unit makes a lot of noise and makes hearing students and teacher more difficult,
so you have to raise your voice, ask for repeats or be very stuffy and
uncomfortable during oral readings and discussions.”
The Capistrano school district is in a relatively mild climate in Southern
California, where ambient temperatures are often in the comfort zone, allowing
natural ventilation without supplementary heating or cooling. However, even in
Capistrano, it is highly probable that substantial energy is wasted running heating
or cooling systems while classroom doors and windows are open. Simply
improving the efficiency of the heating and cooling systems will not solve this
problem. Rather, given teachers’ strong desire for more ventilation, classroom
design should include systems that allow increased ventilation without increasing
energy use for heating or cooling.
Lighting energy use is also an important issue for schools, constituting a large
percentage of overall energy use. The provision of daylighting in classrooms
only saves energy if electric lights are turned off when not needed, either
manually or automatically. The results in Figure 9 suggest that a manual lighting
control scheme has an likelihood of being operated by about half of the teachers
in a school. This behavioral element should be factored into any proposed
lighting control scheme. While automatic systems may be effective more often,
their cost-effectiveness should be compared to manual systems that are
occasionally operated by 50% of the teachers.
The information in the Capistrano teacher survey is not comprehensive enough
to draw any universal conclusions about teacher preferences or behaviors.
However, it is strongly suggestive that the physical environment is a key factor in
teaching effectiveness, and that teacher preferences for classroom operation
need to be given high priority in the design of comfort systems and classroom
controls.
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT TEACHER SURVEY
22
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT TEACHER BIAS ANALYSIS
23
3. TEACHER BIAS ANALYSIS
Once the information in the teacher survey was categorized and compiled into a
database, we were able to analyze the data for a potential bias in teacher
assignment to more daylit classrooms. This task was pursued with a variety of
analytic approaches.
3.1 Hypothesis
For this task we set out to test the hypothesis that the higher rates of learning in
daylit classrooms might be attributable to "better" teachers being located in more
daylit classrooms. For this discussion "better" teachers would be defined as
those who are responsible for faster learning rates in their students, as reflected
in the rate of progress measured by standardized math and reading tests. Daylit
classrooms would be defined by the Daylight Code assigned to each classroom
in the original study.
3.2 Methodology
In order to study this question we needed to 1.) find a way to identify potentially
"better" teachers 2.) determine if the "better" teachers were being differentially
assigned to more daylit classrooms and 3.) determine to what extent the
magnitude or significance of the daylighting effect would change if information
that could predict teacher quality could be included in the model.
Our first step was to define the specific teacher variables to be included in the
models, based on the data we had collected in the early Teacher Survey task. In
order to do this, we needed to understand the basic structure of educational
requirements for a California Elementary School Teacher's credential, along with
the hiring and promotional policies of the district. We collected this information
from the Department of Education web site, the Capistrano District personnel
office, and by interviewing various district administrators.
3.2.1 Teacher Credentials
In discussions with Capistrano administrators prior to the survey, we were told
that the District was not hiring teachers with provisional or emergency
credentials. Beyond requiring all of their teachers to be certified, the district
highly values continuing education for all teachers. A sliding salary scale
rewards additional college education. The District also provides opportunities for
on-site training classes that are specifically tailored to the curriculum needs of the
district.
In the teacher survey we asked teachers to report on their years of teaching in
the current school, district, and total; their highest level of education; additional
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT TEACHER BIAS ANALYSIS
24
course work or certifications; and special awards or recognition. This information
was all self-reported and described in the teacher's own words. We
subsequently categorized this information into the eight variable codes described
below.
Teaching Experience: We defined the variable of Log Yrs Teach as the natural
log of the total number of years teaching. By using a natural log we attempted to
account for the diminishing effect of additional years of experience reported in
the research literature.
Level of Education:
BA indicated any teacher with a bachelors degree. Reported as Teacher
1.
Clear only indicated teachers who had been teaching for 7 years or more,
but had not pursued any continuing education beyond that necessary for their
professional clear credential. Reported as Teacher 5.
BA Plus indicated teachers who listed college credits beyond the minimum
required for certification. Reported as Teacher 2.
MA indicated those with a masters, or doctorate (one case). Reported as
Teacher 3.
MA Plus identified teachers with college credits beyond a master's degree.
Reported as Teacher 4.
Certification: This variable was used to identify teachers who had received any
special certificates or credentials, beyond the minimum required for a California
elementary multi-subject credential. Special certificates for Bilingual Education,
Gifted and Talented Education, Special Education, etc. were grouped together
under one variable. Reported as Teacher 6.
Honors: Many teachers reported receiving special awards, such as Teacher of
the Year, or being selected to be mentor teachers. Because responses varied,
and because we had little way of measuring how prestigious the awards were,
any teacher that reported receiving an award or being chosen to be a mentor
teacher was indicated by the AwarMent variable. Reported as Teacher 7.
The teacher characteristics variables were added back into the master data set.
The surveyed population of teachers represented about 1/2 of the original data
set. Thus, for about 1/2 of the student records we added the information
characterizing their teacher’s years of experience, education level, special
certificates or honors. The remainder of the student records were given an
indicator variable for no teacher information.
3.2.2 Assignment Bias
Once we had defined the teacher characteristic variables, we looked to see if
there were any significant correlations between these teacher characteristics and
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT TEACHER BIAS ANALYSIS
25
the daylight conditions in the classrooms in our Capistrano data set. This was our
first statistical test for a teacher assignment bias. If we found a strong pattern of
correlation between a few teacher variables and a few daylight codes, then it was
likely that some types of teachers were being differentially assigned to more
daylit classrooms. In this first pass at the analysis we included all of the window
related variables, including the daylight code, window code, skylight codes, and
operable windows.
The analysis was based on the data collected in the teacher survey, described in
the preceding section. We used the data from surveys of 206 teachers. These
teachers taught 3,948 of the students included in the original study. To be
consistent with the original study, the first pass statistical analysis was carried out
at the student level. In other words, each student was been taken to be an
observation. Since the number of students per teacher in our data set varied
somewhat independently of the number of total students in a classroom, this
approach has the effect of weighting the results according to the study population
database. Because of the large number of student observations, it also tends to
exaggerate the significance of the correlations.
Looking the student level, we found a statistically significant (2-tailed, p<.10),
correlation among almost all of the variables (see Figure 10). We found no
obvious pattern of any variables less likely to have correlations than others.
Furthermore, the magnitude of correlation was minor throughout. The strongest
correlation, at p=.01, was between Teacher 7 and Skylight Type B (a Pearson
Correlation of .227), implying that 5% (.2272) of classroom assignments might be
explained by this correlation. Nine other combinations had a Pearson Correlation
between 0.1 and 0.2 and all others (61%) were below 0.11, indicating a very
weak magnitude of correlation.
Variable Daylight Window AA Skylight A Skylight B Skylight C Skylight D Skylight Oper. Win.
Teacher 1 Pearson Correlation 0.089 0.068 -0.062 0.111 0.026 -0.001 0.015 -0.106
Sig. (2-tailed) 0.000 0.000 0.000 0.000 0.012 0.905 0.144 0.000
Teacher 2 Pearson Correlation 0.044 0.020 -0.076 0.048 0.085 -0.041 0.064 0.019
Sig. (2-tailed) 0.000 0.050 0.000 0.000 0.000 0.000 0.000 0.067
Teacher 3 Pearson Correlation 0.069 0.112 -0.001 -0.021 0.028 -0.018 0.020 0.004
Sig. (2-tailed) 0.000 0.000 0.914 0.039 0.007 0.082 0.058 0.718
Teacher 4 Pearson Correlation 0.083 0.080 0.109 0.013 0.077 -0.021 -0.025 0.018
Sig. (2-tailed) 0.000 0.000 0.000 0.218 0.000 0.048 0.015 0.081
Teacher 6 Pearson Correlation 0.066 0.051 -0.026 0.087 0.047 -0.039 -0.035 -0.034
Sig. (2-tailed) 0.000 0.000 0.011 0.000 0.000 0.000 0.001 0.001
Teacher 7 Pearson Correlation 0.150 0.147 0.056 -0.096 0.227 -0.067 -0.012 -0.030
Sig. (2-tailed) 0.000 0.000 0.000 0.000 0.000 0.000 0.246 0.003
Log yrs teach Pearson Correlation 0.138 0.171 -0.005 -0.007 0.097 -0.071 0.022 0.015
Sig. (2-tailed) 0.000 0.000 0.659 0.499 0.000 0.000 0.033 0.143
N= 3948 students
Figure 10- Correlation of Teacher Variables to Daylight Variables, Student level
Analysis, Capistrano
1 The Teacher 5 variable had not been defined at this time, so was left out of this correlation table.
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT TEACHER BIAS ANALYSIS
26
In a second pass, we also re-calculated the correlations using the 206 teachers
as independent observations. Using the smaller teacher population, un-weighted
for student population in our data base, presented a more extreme test for
significance. Out of the 56 correlations that are reported in Figure 9, ten were
judged to be potentially significant with p-values of .10 or less at the teacher
level. None of the correlations with the Daylight Code were significant. Skylight
Type AA did show a pattern of correlations, but with only 5 surveyed teachers in
this group, we discounted this as a random result. The most interesting finding
was a slight indication that more senior teachers (Log yrs teach) had some
influence being assigned to classrooms with larger window areas, operable
windows, or skylight types A, and that Teacher 7 (honors) were more likely to be
assigned to skylit classrooms type A or B. The magnitude of a possible effect is
minimal, with only 1% to 5% of the variation in assignment to these classroom
types potentially explained by either of these variables.
We concluded from this exercise that there was indeed some potential for an
assignment bias relative to honors or years of experience, but that a two-
dimensional correlation analysis was not a sufficient tool to determine its
magnitude or influence on the results of the multi-variate regression models.
3.2.3 Decision to Focus on Daylight Code Only
For simplicity sake, we choose to work henceforth with just the Daylight Code.
Tracking the change in performance for one variable, instead of eight, reduced
the complexity of the task dramatically. We choose to focus on the Daylight Code
since it was the holistic code that combined the effects of the window and
skylight codes together. It had been very robust in the previous analysis, and
described the classroom characteristic of greatest interest.
By focusing our attention on just the change in the Daylight Code across models,
we were more likely to see patterns across models.
3.2.4 Daylight Code as a Dependant Variable
Next we ran a regression model with the Daylight Code as the dependant, or
outcome, variable and the teacher characteristic variables as the independent, or
explanatory variables. This model was run using only the surveyed teacher
population. This model would tell us more precisely if there was indeed an
"assignment bias," such that some teacher types were more likely to be assigned
to daylit classrooms. It was a more precise test than the correlation tables, since
it allowed the influence of each teacher characteristic variable to be assessed
simultaneously.
From this regression model, we found that there were NO teacher
characteristics, as defined by our variables from the survey data, that were
significant in explaining assignment to more daylit classrooms. The variable that
achieved the highest probability of influence was Teacher 7 (honors) at only 78%
likelihood of significance (p=.22) that there might be a 5% higher assignment in
Daylight Code (A teacher who had received an honor or award had a 78%
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT TEACHER BIAS ANALYSIS
27
probability of being assigned to a classroom rated 3.15 on the daylight scale
instead of a 3.0). The other variables had a 50% probability or less.
The R2 for this model was only 0.014, indicating that all of the teacher
characteristic variables could explain only 1% of the variation in assignment to
daylight classrooms. When we ran a similar model at the student level, the level
of explanation increased to 2%. Thus, from this exercise we conclude that the
Capistrano Unified School District did not have any marked bias in the
assignment of teachers to more daylit classrooms, based on the teacher
characteristics that we studied.
3.2.5 Teacher Assignment Bias Models
Our final step in the Teacher Bias Analysis was to re-run the original Capistrano
student performance models with the teacher characteristic variables added to
the list of potential explanatory variables. Again, we choose to focus our reporting
on the results of the Daylight Code for simplicity, although we did also run the
separate models with the window and skylight variables. The original models
were re-run for both change in reading and math scores at the student level.
Teacher characteristic variables were added for 42% of the population.
It should be noted that the performance of the observed students within a given
classroom may not be mutually independent. In the original research, we carried
out a special analysis to assess the effect of correlation between students within
a given classroom (See Appendix 6.2 to original report, dated 6/29/1998). This
analysis indicated that the statistical significance of some of our results was
somewhat overstated but the effects of interest were not substantially altered.
However, carrying out the analysis at the student level made it easier to explore
the relationship between characteristics of the student, teacher, room, and
school.
3.3 Findings
Figure 12 and Figure 11 display the findings of these two models, compared to
the original models without the teacher variables. The school site variables and
outliers have been left off of the equations shown here for simplicity, but are
included in the full model detail in the Appendix 7.2. A central column shows the
change in the B coefficient for each variable and the model R2.
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT TEACHER BIAS ANALYSIS
28
New Model Change Old Model
Capistrano, Teacher Analysis - Math Daylight new-old Capistrano, Original Analysis Math Daylight
28-2 (Original population) R^2 C17-md
Model R^2 0.259 0.003 Model R^2 0.256
B Std. Error p (Signif) BB Std. Error p (Signif)
(Constant) 9.045 0.464 0.000 (Constant) 8.026 0.407 0.000
Classroom characteristics Classroom characteristics
Daylight code 0.430 0.072 0.000 -0.075 Daylight code 0.504 0.067 0.000
Teacher characteristics
Teacher 3 -0.933 0.248 0.000
Teacher 5 -0.688 0.335 0.040
Log yrs teaching 0.373 0.077 0.000
Student characteristics Student characteristics
Grade 2 9.624 0.216 0.000 -0.088 Grade 2 9.711 0.215 0.000
Grade 3 5.949 0.220 0.000 0.018 Grade 3 5.931 0.219 0.000
Grade 4 1.802 0.216 0.000 -0.011 Grade 4 1.813 0.216 0.000
Absences unverified -0.263 0.123 0.033 0.000 Absences unverified -0.263 0.123 0.032
Absences unexecused -0.029 0.014 0.043 -0.003 Absences unexecused -0.026 0.014 0.069
GATE program -1.191 0.222 0.000 0.045 GATE program -1.236 0.223 0.000
Language program 0.488 0.205 0.017 -0.001 Language program 0.490 0.205 0.017
School characteristics School characteristics
School Pop-per 500 -0.995 0.000 0.000 -0.483 School Pop-per 500 -0.512 0.000 0.010
Figure 11 - Change in Capistrano Math Model with Addition of Teacher Variables
New Model Change Old Model
Capistrano, Teacher Bias Analysis - Rea ding Daylight ne w-old C apistrano, Or iginal Ana lysis Re ading D aylig ht
28-2 (Original population) R^2 C17-rd
Model R^2 0.248 0.002 Model R^2 0.246
B Std. Error p (Signif) BB Std. Error p (Signif)
(Constant) 3.009 0.303 0.000 (Constant) 3.025 0.298 0.000
Classroom characteristics Classroom characteristics
Daylight code 0.475 0.086 0.000 0.011 Daylight code 0.464 0.085 0.000
Operable windows 0.650 0.212 0.002 0.007 Operable windows 0.643 0.212 0.002
Teacher Characteristics
Teacher 3 -0.917 0.288 0.001
Teacher 5 -1.335 0.388 0.001
Log yrs teaching 0.221 0.090 0.014
Student characteristics Student characteristics
Grade 2 10.823 0.251 0.000 -0.037 Grade 2 10.860 0.251 0.000
Grade 3 4.368 0.255 0.000 0.069 Grade 3 4.298 0.254 0.000
Grade 4 0.944 0.252 0.000 0.008 Grade 4 0.937 0.252 0.000
GATE program -1.432 0.257 0.000 0.020 GATE program -1.452 0.257 0.000
LANG progra
m
0.827 0.239 0.001 -0.011 LANG progra
m
0.838 0.239 0.000
Figure 12 - Change in Capistrano Reading Model with Addition of Teacher
Variables
Even with the addition of the teacher characteristic variables into the original
models, the daylight variable stayed highly significant in both cases. For the
math model, with the outcome variable as the change in fall to spring math
scores, the magnitude of the daylight effect decreased slightly.
For the reading model, the magnitude of the daylight effect actually increased. In
the case of the reading model, operable windows also remained a significant
variable, and also increased slightly in magnitude.
Three of the eight teacher characteristic variables were found to be significant in
both models. (While the significant teacher variables here were consistent, they
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT TEACHER BIAS ANALYSIS
29
were not consistent in the models using window codes and skylight types as
explanatory variables, nor were they consistent in later models that we ran,
discussed later.)
With the addition of information about the teachers, the R2 of the models
increased, but only by a tiny amount, increasing their power of explanation by
less than 1%.
3.4 Conclusion
Thus, we conclude that the strength of the daylight variable showed in the
original analysis was not an inadvertent effect of a “teacher assignment bias.”
We have shown in the regression model of the Daylight Code versus the teacher
characteristic variables, that the teacher characteristics captured in our survey
only explained 1% of the variation of teacher assignment to daylit classrooms.
Furthermore, in the master student performance regression models adding
information about teacher characteristics for 42% of the population did not
reduce the significance of the daylight variables. As might be expected, the
magnitude shifted slightly; in one case down, in one case up.
3.5 Discussion
One potential weakness in the findings above is that we only had teacher
characteristic information for less than half of the study population. We decided it
would be a good test to re-run the models for just the population of students
represented by teachers who responded to the Teacher Survey. That way, we
could look at a model where 100% of the population had information about the
teachers. This “surveyed population” model included 206 teachers and 3948
students, or about 50% of the original population.
We were aware that if we shifted the sample population for a model, we ran the
risk of getting different results. But we wanted to examine the stability of the
daylighting coefficient in our models over different sample populations. We also
wanted to explore the stability of including the information about the teachers.
Thus, we decided to run similar models to the original Capistrano math-daylight
and reading-daylight models, looking at the change in the daylight variable from
one sample to another and with the addition of the teacher characteristic
variables.
We also had one other complexity to account for. In coding the data from the
Teacher Survey it was discovered that three schools had been inadvertently
dropped from the original study population. Criteria for inclusion of a student’s
record in the original analysis had included complete records for test scores,
attendance and demographic data. We did not observe at the time that we had
not been provided with attendance data for three entire schools. Thus, the data
cleaning procedures resulted in inadvertently dropping all students (and all
teachers) from those three schools from the analysis. We were particularly
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT TEACHER BIAS ANALYSIS
30
concerned since two of the dropped schools represented somewhat extreme
daylight conditions, one with many classrooms of Daylight Code 0, and the other
with many Classrooms of Daylight Code 4. Thus, we worried that the exclusion of
these schools from the original analysis may have skewed our results.
We noted that any effect due the missing attendance data could be absorbed to
some degree by the dummy variable that identified the school site that was
missing the attendance information. Thus, we decided to create a new
“expanded” population that included these three schools and provided a
“missing” indicator in the attendance record fields. This “expanded population”
model included 394 teachers and 9200 students, 13% larger than the original
study population.
(number of students in population)
Figure 13: Surveyed, Original, and Expanded Populations
We were interested to see if the daylight variable would remain significant in
models of student performance in these different populations, with and without
the addition of the teacher characteristic variables. The teacher survey
population would present the clearest test of the impact of the teacher
characteristics, since for this population we would have information about teacher
characteristics for 100% of the teachers. The expanded population was likely to
have the truest daylight results, since it represented the full 2-5 grade district
population in 1997/98 school year. For this population we had information on
50% of the teachers.
3.5.1 Findings of Different Study Population Models
Figure 14 and Figure 15 compare the results for the three sets of regression
models; the original model, the expanded model, and the teacher surveyed
model, for the reading and math models. Full detail of all models is included in
the Appendix. In addition to comparing the B coefficient for the Daylight Code,
the significance of the Daylight Code and the R2 of the model, we also report
here on the effective rate of change in the learning rate, and the confidence
interval for that rate.
Expanded
Population (9302)
Surveyed
Population (3949)
Original 1999
Population (8100)
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT TEACHER BIAS ANALYSIS
31
Key
Capistrano
Reading Model,
Study Population
Teacher
Variables
Included in
Model
B Coefficient
for Daylight
Code
p (Signif)
of B Model R2
% Change in
Learning
Rate
confidence
interval
A original no 0.464 0.000 0.247 26% ±10%
B original yes 0.475 0.000 0.248 27% ±10%
Shift from Model A to B 0.011 no change 0.001 1% no change
C expanded no 0.416 0.000 0.238 24% ±9%
D expanded yes 0.418 0.000 0.240 24% ±9%
Shift from Model C to D 0.002 no change 0.002 0% no change
E surveyed no 0.434 0.000 0.239 23% ±12%
F surveyed yes 0.463 0.000 0.243 25% ±12%
Shift from Model E to F 0.029 no change 0.004 2% no change
Figure 14: Daylight Affect for Different Populations, with and without Teacher
Variables, on Reading Tests in Capistrano
Key
Capistrano Math
Model, Study
Population
Teacher
Variables
Included in
Model
B Coefficient
for Daylight
Code
p (Signif) of
B Model R2
% Change in
Learning
Rate
confidence
interval
A original no 0.504 0.000 0.257 20% ±5%
B original yes 0.430 0.000 0.259 17% ±6%
Shift from Model A to B -0.074 no change 0.002 -3% 1%
C expanded no 0.351 0.000 0.250 14% ±5%
D expanded yes 0.301 0.000 0.252 12% ±5%
Shift from Model C to D -0.050 no change 0.002 -2% no change
E surveyed no 0.544 0.000 0.274 21% ±8%
F surveyed yes 0.497 0.000 0.277 19% ±8%
Shift from Model E to F -0.047 no change 0.003 -2% no change
Figure 15: Daylight Affect for Different Populations, with and without Teacher
Variables, on Math Tests in Capistrano
For the reading model, the most conservative estimate of a daylight effect would
be +11% for the surveyed population without teacher variables (23%-12%), while
the most optimistic would be +37% for both the original and surveyed population
with teacher variables (27%+10% and 25%+12% respectively). For the math
model, the most conservative estimate of a daylight effect would be +7% for the
expanded population with teacher variables (12%-5%), while the most optimistic
would be +29% for the surveyed population without teacher variables
(27%+10%). Thus, from worst to best case we can say with a high degree of
confidence, that children with the most daylighting in Capistrano are learning
somewhere from 7% to 37% faster on the District's math and reading curriculum.
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT TEACHER BIAS ANALYSIS
32
With the addition of teacher characteristics to the three sets of models, the
following changes were observed:
• Daylight variables were still significant across all models
• R
2 value increased by 0% to +2% indicating that the models with teacher
characteristics had a slightly better explanatory power for the studied
phenomena.
• Math models indicated a decrease in the effect of daylight on student
performance by 2% to 3%.
• Reading models indicated an increase in the effect of daylight on student
performance by 0% to 2%.
• In general the availability of daylight in classrooms was reliably associated
with an increase in student performance and learning rate of somewhere
within the bounds of 7% to 37%. The central tendency among all these
models would seem to be a 25% improvement in reading and a 16%
improvement in math, or a 21% general improvement between children in
classrooms with the most daylight (code 5) compared to those in
classrooms with the least (code 0). In summary, if the average student in
the district were moved from an average classroom (code 2.5) to a
classroom with maximum daylight (code 5), he or she would be expected
to increase his or her learning rate by 11% (10.5).
• All these results were observed with 99.9% statistical certainty.
In addition, we were interested to understand the change in daylighting effect
among the three populations, the original, expanded, and surveyed, before the
addition of the teacher variables. Figure 16 and Figure 17 compare the changes
when moving from the original population to the expanded population (13%
larger), and from the original to the surveyed (50% smaller) for both reading and
math. These changes were also very modest, with from a 3% to 6% shift in the
net impact of the daylight variable on student learning rates.
Key
Capistrano
Reading Model,
Study Population
Teacher
Variables
Included in
Model
B Coefficient
for Daylight
Code
p (Signif)
of B Model R2
% Change in
Learning
Rate
confidence
interval
A original no 0.464 0.000 0.247 26% ±10%
C expanded no 0.416 0.000 0.238 24% ±9%
Shift from Model A to C -0.047 no change -0.009 -3% -1%
A original no 0.464 0.000 0.247 26% ±10%
E surveyed no 0.434 0.000 0.239 23% ±12%
Shift from Model A to E -0.030 no change -0.008 -3% 2%
Figure 16: Teacher Variables and Daylight effect on Reading for the Three
Populations Compared
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT TEACHER BIAS ANALYSIS
33
Key
Capistrano
Math Model,
Study
Population
Teacher
Variables
Included in
Model
B Coefficient
for Daylight
Code
p (Signif)
of B Model R2
% Change in
Learning
Rate
confidence
interval
A original no 0.504 0.000 0.257 20% ±6%
C expanded no 0.351 0.000 0.250 14% ±5%
Shift from Model A to C -0.153 no change -0.007 -6% -1%
A original no 0.504 0.000 0.257 20% ±6%
E surveyed no 0.544 0.000 0.274 21% ±8%
Shift from Model A to E 0.040 no change 0.017 1% 2%
Figure 17: Teacher Variables and Daylight effect on Math for the Three
Populations Compared
Interestingly, the greatest variability between models, 6%, occurred from the
original to expanded populations for the math model. Earlier, in the Classroom
Level Analysis, included in the Appendix of the 1999 Detailed Report, we had
found much greater variability in the success of math instruction attributable to
individual teachers than reading instruction. Thus, we would also expect greater
volatility in the math results between population samples.
The following findings were observed when comparing the three populations
before adding the teacher variables to them:
• No change in significance of daylight variable
• The explanatory power of the statistical models (i.e., R2) in explaining the
data varies by less than 2%.
3.5.2 Conclusions of Different Study Population Models
The shift in model study populations actually had a greater impact on the R2 of
the models than the addition of the teacher characteristic variables. We also saw
the largest shift in the magnitude of the B coefficient for the Daylight Code
between study populations, rather than with the addition of information about the
teachers. Thus, we conclude that the selection of the study population is more
likely to impact findings about the effect of daylight than is the addition of
information about teachers.
We continue to believe in the importance of the addition of the teachers'
characteristics to the model, both to access the potential for a teacher bias and to
further refine the accuracy of the model. However, it is clear from this exercise
that the study population is likely to have an even greater effect on the results.
This once again argues for the importance of replicating the study in other
districts, and preferably in widely differing geographic regions and cultural
environments.
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT TEACHER BIAS ANALYSIS
34
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT GRADE LEVEL ANALYSIS
35
4. GRADE LEVEL ANALYSIS
The Grade Level Analysis task was intended to answer two of the questions that
were raised from a previous peer review1 of the Daylighting in Schools study.
One question was whether it was might be more appropriate to analyze the data
in single grade cohorts, rather than across grades. It was proposed that
especially in Seattle, for the Iowa Test of Basic Skills (ITBS), results could not
correctly be compared across grades. Creating separate models for each grade
level would solve this problem.
A second question asked whether the daylighting effect might vary by grade
level. The models used in the first analysis constrained the results to a simple
linear expression. It was argued that there might be a progressive effect, again
especially in Seattle, where children were exposed to fairly consistent daylighting
conditions for the duration of their career at a given school. In Seattle, where we
were looking at absolute test scores, exposure to good daylight conditions over
more than one year might result in a cumulative effect. This would be evidenced
by a progressively greater daylight effect in each higher grade. Again, separate
grade level models would allow the daylight effect to change by grade level,
allowing us to identify any patterns as children got older.
In Capistrano, we hypothesized that we would not find any progressive effects
since children are likely to be shuffled back and forth between traditional
classrooms and portable classrooms with each change in grade level. We
confirmed with the District that the churn rate in the Capistrano district is
reasonably low, with about 4% growth per year, and a similar number of students
who relocate to other districts per year. Thus, we estimate about 90% of the
students return to a given school each year. Typically, they would experience at
least two, if not three or four daylight conditions throughout their career at a given
school. Furthermore, since in Capistrano we were looking at the improvement in
schools in one year, from fall to spring, cumulative effects would be less likely to
show up.
4.1 Hypothesis
Given the main objective of this task, it was hypothesized that daylighting may
have a cumulative effect on student scores. This hypothesis would be likely true
if a pattern of progressively stronger effects by grade level was observed in
Seattle, where children typically remain under one school-wide daylighting
condition. A comparative analysis for the test scores in the Capistrano school
district, where students may change between high and low daylighting conditions
1 Daylighting and Schools Peer Review Report, sent to PG&E, July 21, 1999. Not released.
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT GRADE LEVEL ANALYSIS
36
during their stay at an elementary school, would corroborate our hypothesis if a
minor or no cumulative effect of daylight was observed in that district.
4.2 Methodology
We re-ran the student performance regression models for both Seattle and
Capistrano, this time allowing the daylighting effect to vary by grade level. This
was achieved by adding grade level interaction variables for each variable in the
model. This is statistically equivalent to running separate models, but simplifies
the reporting and interpretation.
Interaction variables between the grade level of the student and each
explanatory variable were created and added to the original Capistrano and
Seattle models. As in the original study, the Capistrano model used the
difference between fall and spring scores while Seattle’s used the absolute value
of the spring scores.
Since information regarding teacher characteristics was available for the
Capistrano school district, the teacher variables were also included in the
Capistrano math and reading models to strengthen their explanatory power.
4.3 Findings
The data from our interaction models did not show a significant effect for the
interaction variables between daylight and separate grade levels. This indicates
that, for our study populations, we could not support the hypothesis that daylight
has a different or cumulative effect on student performance by each grade. The
full model results are shown in the Appendix 7.3.
We also found that allowing the results to vary by grade did not improve the
accuracy of the models. The R2 of the models increased only very slightly with
the addition of the interaction variables, 4% for the Seattle reading model, and
less than 1% for the other three. (See Figure 18 and Figure 19)
It is important to note, however, that the daylighting effects remained highly
significant even after the addition of the interactive variables. This indicates that
daylight still provides a robust explanation of student performance in math and
reading tests across all grades. For the Capistrano reading model, the magnitude
of the effect (B) declined by 14%, but not the significance.
For the Capistrano math model, we saw a greater impact on both the magnitude
(45% decline) and significance (7% decline). This is the one incidence where the
daylight variable would not pass our threshold criteria of 95% significance or
greater for inclusion in the model. This decline in significance and magnitude
were probably caused most by the addition to this model of the one daylight-
grade level interaction variable that did prove significant: Daylight Code(2nd
grade). This interaction variable was found to increase the daylight effect
considerably for second graders, by more than twice (216%). The interpretation
here would be that second graders in more daylit classrooms were mastering the
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT GRADE LEVEL ANALYSIS
37
math curriculum dramatically faster then those who were not in daylit classrooms,
and also comparatively faster than children in other grade levels in daylit
classrooms.
While this finding about second graders learning math might seem potentially
interesting, the fact that we did not find any other significant interaction effects in
any of the other model tends to discount the validity of this finding. Out of twelve
opportunities, the interaction between grade level and daylight was found to be
significant in only one case. Thus, we tend to doubt that there is any differential
sensitivity to daylight by grade level.
Key Test Interactive
Variables B Model R^2 % impact error bound Signif.
A Reading N 0.464 0.247 26% +/-10% 100.0%
B Reading Y 0.396 0.239 22% +/-9% 100.0%
Shift from Model A to B -14% -0.008 -4% 0%
C Math N 0.504 0.257 20% +/-5% 100.0%
D Math Y 0.275 0.261 11% +/-12% 92.7%
Shift from Model C to D -45% 0.004 -9% -7%
Figure 18: Capistrano Grade Level Models with Interactive Variables Summary
In Seattle, when allowing for grade level interactions with all the other variables,
we saw no declines in significance, and also saw substantial increases in the
magnitude of the daylight effect. In the case of the Seattle reading model, the
magnitude of the daylight effect increased 26%, while in the math model the
magnitude of the daylight effect increased 12%. For the Seattle reading model,
the accuracy of the model (R2 ) increased 4%. This would tend to argue for the
validity of the increase in the magnitude of the daylight effect. Since some of the
significant interaction variables have to do with the physical conditions of the
classroom (school vintage, school size, classroom SF) it is possible that some of
the daylight effect was previously being masked by the imprecision of those
variables without the interaction effects.
Key Test Interactive
Variables B Model
R^2 % impact error bound Signif.
A Reading N 1.883 0.297 16% +/- 8% 100.0%
B Reading Y 2.533 0.337 22% +/- 7% 100.0%
Shift from Model A to B 26% 0.040 6% 0%
CMath N 1.391 0.258 12% +/- 7% 99.9%
DMath Y 1.585 0.257 13% +/- 7% 100.0%
Shift from Model C to D 12% -0.001 2% 0%
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT GRADE LEVEL ANALYSIS
38
Figure 19- Seattle Grade Level Models with Interactive Variables Summary
4.4 Conclusions
The grade level analysis did not increase the accuracy of the models. Further
more, while we did find interaction effects between grade level and other
variables, most notably the demographic variables, we did not find a consistent
interaction between grade level and a daylighting effect. This was true in both
Seattle and Capistrano.
From this exercise, we conclude that our original modeling approach, grouping all
of the data for grades 2-5, was sufficiently accurate. We also note that we did not
find any progressive effect for the daylighting variable, as postulated for Seattle,
nor any other pattern related to the age of the student.
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT ABSENTEEISM ANALYSIS
39
5. ABSENTEEISM ANALYSIS
The Capistrano data set includes information on absences and tardiness per
student. Both of these parameters were included as explanatory variables in our
original daylighting analysis, but not as dependant variables. We did not use
them as dependant variables at the time for two reasons. First, we did not have
this information for all three districts, and our original criteria included consistent
analysis across districts. Second, the absenteeism and tardiness data is much
thinner than student test performance data, since only about 10% of students
had a significant number of absences. Thus, it provided a much less sensitive
metric of performance.
However recent research findings by others, discussed below, suggested that we
should re-examine the Capistrano data set for similar effects. In a number of
studies increased ventilation rates have been found to reduce worker
absenteeism. There has also been increased interest in the effect of classroom
environments, particularly portable classrooms, on student health with a number
of epidemiological studies initiated to look for these links. Finally, many
daylighting proponents have been claiming the daylighting improves student
attendance, and thereby will also increase funding to the schools through
California’s system of ADA (average daily attendance) payments.
Milton et al of Harvard School of Public Health reported that increased ventilation
rates were associated with reductions in sick leave in the Polaroid Company
offices in Massachusetts1. They report: “Based on this latter analysis, 45% of the
sick leave among workers in lower ventilation areas was attributable to lower
outdoor air supply. Similarly, 41% of sick leave was [also] attributable to
humidification, and 39.2% of sick leave…was attributable to the presence of
(IAQ) complaints. This corresponded to 1.4 – 1.5 days of increased sick leave
per person per year attributable to ventilation, and 1.2 – 1.3 days per person per
year attributable to humidification, and 1.1 – 1.2 days per person per year
attributable to IAQ complaints, depending on age and gender.”
Teculescu et al. 2 recently reported that occupants of an air-conditioned building
were more likely to have multiple absences from work than were persons in a
naturally ventilated building. This study was limited, however, by the use of only
two buildings (in northeastern France), and by lack of control for ventilation rates
and individual and group factors that may have confounded the relationship
between building and sick leave.
1 Milton DK, Glencross PM,Walters MD. Risk of Sick Leave Associated with Outdoor Ventilation Level,
Humidification, and Building Related Complaints, Harvard School of Public Health, August 1999
2 Teculescu DB, Sauleau EA, Massin N, Bohadana AB, Buhler O, Benamghar L, Mur JM. Sick-building
symptoms in office workers in northeastern France: a pilot study. Int Arch Occup Environ Health 1998;
71:353-6.
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT ABSENTEEISM ANALYSIS
40
The interest in the effect of classroom construction and maintenance, particularly
portable classrooms, on student health has peaked in recent years. Current
projects in progress include: HP-Woods Institute is studying the relationships
between indoor environment and occupant performance in two elementary
schools, funded by Air Conditioning and Refrigeration Technology Institute’s 21-
CR program; the California Department of Public Health is beginning a study of
the environmental health conditions in portable classrooms, funded by Air
Research Board; a pilot study of indoor air quality in portable classrooms is being
done in Los Angles Count, funded by US EPA; another CEC PIER project is also
looking at exposure to VOCs and thermal comfort in four new portable
classrooms.
Given this level of interest, we concluded that it would be worthwhile to see if our
original Capistrano data set would allow us to make any correlations between
classroom physical conditions and student health. The absenteeism and
tardiness data could be used as a proxy measure of student health, while
daylighting, operable windows, air conditioning, age of classroom and type of
classroom (portable, modular, open, semi-open, traditional) could be used as
explanatory variables.
We choose to look at absences or tardies data as a reasonable potential proxy
for student health. However, our study could not distinguish reasons for
absences or tardies. There are many other powerful factors influencing
elementary school attendance besides the health of the student, such as dentist
or orthodontist appointments, outside activities, poor transportation, parental
health, family obligations, etc. Thus, our absenteeism and tardiness variables
cannot be interpreted as a strong metric of student health, but rather simply as
the best proxy for student health that we had available in our data set.
5.1 Hypothesis
In our earlier Capistrano study, we found that daylight was consistently
associated with enhanced learning rates, and operable windows were associated
(>95% certainty) with enhanced learning rates in three of the four models. In that
original analysis, neither portable classrooms nor the presence or type of air
conditioning had a statistically significant effect.
Based on this finding we hypothesized that daylighting and operable windows
might also be associated with a reduction in student absenteeism and tardiness
in the Capistrano school district.
If this hypothesis were true, operable windows and daylight, as explanatory
variables, would appear to be significant and negative in a regression analysis
with student absenteeism and tardiness as dependant variables.
Since the models also included other descriptions of the physical conditions of
classrooms, we could simultaneously test for the significance of those variables
in relationship to absenteeism or tardiness. We were particularly interested in
the portable classroom (port) and modular classroom (pport) variables. If
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT ABSENTEEISM ANALYSIS
41
portable or modular construction does indeed impact student health, then we
would expect to see these variables show up as significant in the regression
analysis.
5.2 Methodology
A multivariate regression model, using the original data from the 1999 study of
the Capistrano school district with all the school data, including daylight, operable
windows, as well as the addition of the new teacher and school variables, was
run. The student characteristics, teacher characteristics, and school and
classroom characteristics were run as independent exploratory variables against
absenteeism data the dependant outcome variable. A similar model was run with
the same variables against tardy data as an outcome variable.
The data set was redefined to include all those students who attended at least 40
days at the same school. The students, however, were not required to have test
scores. As a result, the population shifted slightly, including more students who
were not present for either the fall or spring tests, but excluding any records
missing attendance data. Thus, the three schools from which we had never
received attendance data were dropped from the population. The resulting
analysis population was 8808 students.
41%
33%
15%
3% 1% 1% 0% 0%
71%
4% 2% 2% 1% 1% 1%
6%
12%
6%
0%
10%
20%
30%
40%
50%
60%
70%
80%
0-4 5-9 10-14 15-19 20-24 25-29 30-34 35-39 40+
Number of Absences and Tardies
% of Students
Absences Tardies
Figure 20- Distribution of Absences and Tardies
The absence variable was defined as a function of the sum of three fields in our
data set: unverified absences, excused absences, and unexcused absences.
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT ABSENTEEISM ANALYSIS
42
Absences due to school function were not included. Only the sum of absences
per student was available. We did not have information on the distribution of
absences over time.
Plotting the attendance data in Figure 20 we noted a very strong curve, where
74% of the population were found to have both fewer than 10 absences, and
83% fewer than 10 tardies. In order to properly model this data distribution we
choose to use a natural log function, as expressed in the equation shown in
Figure 21 below. We normalized the absenteeism and tardiness data across the
whole population by adding a ratio of days enrolled to maximum possible days
enrolled:
Ln_Abs =
×
=Tardies)(or Absences ofnumber
40) (minimum enrolled days ofnumber
days) enrolled ofmaximum(180
ln
Figure 21- Equation for natural log of attendance data
5.3 Findings
The regression models with the log of absences or tardiness as dependant
variables did NOT support the hypothesis that daylight variables, or any other
physical characteristics of the classrooms, have a significant effect on student
absenteeism or tardiness.
While these models included all of the same explanatory variables used in
previous analysis, they proved to be comparatively weak models. The R2 of the
absences model was only 0.05, and that of the tardiness model 0.10, indicating
that only 5% and 10% respectively of the variance in the data was explained by
all of the variables included in the models.
5.3.1 Absenteeism Findings
Physical classroom variables that were considered and found to have NO
significance in the absenteeism model included: daylight code, operable window,
type of classroom (portable, open, traditional), air conditioning, and size of
classroom. In addition, none of the teacher characteristics were found to be
significant.
Variables that were significant included: grade level, student socio-economic
characteristics, special programs, school site, school vintage, and school
population.
Thus, we conclude that student demographic characteristics and school level
characteristics (which might include neighborhood effects, special programs, or
size of school) have the greatest relationship to student absenteeism.
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT ABSENTEEISM ANALYSIS
43
5.3.2 Tardiness Findings
The Tardiness model did find that three physical characteristics of classrooms
had a slight, significant effect on the pattern of tardiness:
• Daylighting had a modest, positive effect p=.000
• 5% reduction
• No Air Conditioning had a slight, negative effect p=.032
• 11% increase
• Portable classrooms had a slight, negative effect p=.037
• 5% increase
R2 = 0.097
These results could be interpreted to predict that the students in the most daylit
classrooms would be likely to have one less tardy per year than those in the least
daylight classrooms (5 daylight codes *.05 per code =25% reduction in norm of 5
tardies per year, or 4 tardies per year.) Likewise, no air conditioning was found to
be associated with a slight increase in tardiness, 11% from the norm of 5 to 5.5
tardies per year, and portable classrooms were found to be associated with a
slight increase in tardiness by 5%, up to 5.25 tardies per year. .
Since tardies are a somewhat subjective measure of student performance (not all
teachers mark a student tardy at the same point of lateness) and since tardies do
not have as a strong economic tie to the performance of the school as does
absenteeism data, we chose to discount these results as not particularly
interesting.
5.4 Conclusions
Student attendance, as measured by absences and tardies, was not predicted by
with the daylight conditions of the classrooms in the Capistrano Unified School
District. Likewise, other physical conditions of the classrooms were not found to
be reliable predictors of student attendance.
From this exercise, we concluded that attendance data is a very difficult outcome
metric to work in trying to understand the effects of the physical environment on
the performance of students, or the productivity of people in general. There are
two basic reasons for this difficulty. First, attendance data can only be a loose
proxy for the health of the student, since so many other events can cause a
student to be absent or tardy besides health effects caused by the physical
environment. Secondly, it is not a very sensitive metric. There is not a very big
range in attendance values among students, with only about 10% of the student
population showing much variation in number of days absent or tardy.
A summary of the findings from the absenteeism analysis is as follows:
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT ABSENTEEISM ANALYSIS
44
• Daylighting variables were not significant indicators of Absenteeism.
Similarly neither operable windows nor portable classrooms variables
were significant.
• Student demographic variables were the only reliable predictors of
absenteeism
• Physical characteristics of classrooms were not predictors of student
attendance
• Attendance data is not particularly useful as a performance metric,
providing meaningful variation for only 10% of students in our fairly large
samples (n= ~ 8800).
• A slight effect of daylight on student tardiness was observed, but not
considered interesting.
5.5 Discussion
Our study could not distinguish reasons for absences or tardies. It was assumed
that overall absence and tardy data might serve as a reasonable proxy for
student health. However, there are many other powerful factors influencing
elementary school attendance besides the health of the student, such as dentist
or orthodontist appointments, outside activities, poor transportation, parental
health, family obligations, etc. Thus, our absenteeism and tardiness variables
cannot be interpreted as a strong metric of student health, but rather simply as
the best proxy for student health that we had available in our data set.
Improved physical conditions in a workplace or school have been postulated by
many to be associated with reduced absenteeism. Indeed, this is a fairly common
assertion made in presentations advocating “green” or “sustainable” buildings—
that an improvement in the quality of the physical environment will result in fewer
absences and thus higher productivity. These claims are most frequently made
for improvements in indoor air quality (IAQ)1, but also variously for natural
ventilation, ventilation rates2, thermal comfort, ergonomic furniture, electric
lighting quality and the presence of daylight.
Our study can only speak to a few of these issues: the potential link between
poor indoor air quality in portable classrooms and increased absenteeism. It is
important to note that this re-analysis study of the Capistrano data did not
substantiate any of these claims.
1 Fisk WJ (2000). Health and productivity gains from better indoor environments and their relationship with
building energy efficiency. Annual Review of Energy and the Environment 25(1): pp. 537-566
2 Milton DK, Glencross PM, Walters MD (2000). Risk of sick leave associated with outdoor ventilation level,
humidification, and building related complaints. Indoor Air, 10(4): pp. 212-21
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT ABSENTEEISM ANALYSIS
45
• Portable classrooms are currently under investigation by a number of
researchers for poor indoor air quality1, which might reduce overall student
health.
• Our study did not find that there was any significant association between
portable classrooms and increased absenteeism among students.
• Operable windows have been associated with a reduction in indoor air quality
complaints2.
• We did not find that operable windows were significantly associated with
any improvement in attendance among elementary school students.
• Claims have been made that daylit schools are associated with improved
attendance among students3.
• We did not find that increased daylight in classrooms was associated with
better attendance.
1 Per Jed Waldman, CA Department of Public Health
2 MP Callahan, DS Parker, WL Dutton, and JER McLivaine, 1997. “Energy Efficiency for Florida Educational
Facilities: the 1996 Energy Survey of Florida Schools.” FSEC-CR-951-97, Florida Solar Energy Center,
Cocoa, Fl.
3 M Nicklas and G Bailey, “Analysis of the Performance of Students in Daylit Schools,” Proceedings of the
American Solar Energy Society, 1997.
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT ABSENTEEISM ANALYSIS
46
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT RE-ANALYSIS CONCLUSIONS
47
6. RE-ANALYSIS CONCLUSIONS
6.1 Grade Level Analysis
The data did not show a significant effect for the interaction variables between
daylight and separate grade levels. Likewise, we did not observe any consistent
patterns of an increase or decrease in daylight effects by grade level. Thus, we
conclude that there do not seem to be progressive effects as children get older,
nor do younger children seem to be more sensitive to daylight than older
children.
Allowing the results to vary by grade did not improve the accuracy of the models;
with one exception, the R2 of the models increased less than 1%. Therefore, we
believe that the extra analysis did not add significantly to our understanding and
future research can proceed looking at data across grade levels.
Furthermore, the daylighting effects remained highly significant even after the
addition of the interactive variables. This indicates that the Daylight Code still
provides a robust explanation of student performance in math and reading tests
across all grades.
6.2 Absenteeism Analysis
The student attendance record regression models did not support the hypothesis
that daylight variables or any other physical characteristics of the classrooms
have a significant effect on student absenteeism or tardiness. Notably,
daylighting conditions, operable windows, and air conditioning were not
significant in predicting absences. The models were comparatively weak; the full
set of 57 variables for the Capistrano data explained only 5% and 10% of the
variance in absences and tardies, respectively.
We chose to look at absences and tardiness data as the best proxy for student
health that we had available. Absenteeism and tardiness cannot be interpreted
as a strong metric of student health, since many other powerful factors influence
elementary school attendance. However, to the extent that attendance data does
reflect student health, our study may indicate only a weak connection between
physical classroom characteristics and student health.
6.3 Teacher Survey
Although the Teacher Survey task was primarily aimed at providing additional
information for other Re-analysis tasks, we did learn some useful information
about teacher preferences, attitudes and behaviors. For example, while the
teachers we surveyed clearly had a preference for windows, daylight and views
in their classrooms, these preferences were not likely to be driving classroom
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT RE-ANALYSIS CONCLUSIONS
48
selection. Far more important in classroom selection was an almost universal
desire for large classrooms, lots of storage and water supply in the classroom.
Environmental control is also an important issue for teachers, especially when
they find that they don't have it in their classroom. Teachers seemed to hold a
basic expectation that they would be able to control light levels, sun penetration,
acoustic conditions, temperature and ventilation in their classrooms. When
control of one or more of these environmental conditions was not available to
them in the classroom, they were passionate and outspoken in their outrage.
We also found that teachers reported using their optional control features
frequently enough to make significant impacts on classroom energy use. Use of
these features by a dedicated minority would seem to be sufficient to justify their
cost effectiveness in terms of energy savings. Of course, their value should also
be considered in terms of classroom comfort and productivity.
In their freely offered comments, the teachers were desperate to be heard about
the need for better physical environments in their classrooms. It is worth taking
the time to review these comments included in the Appendix. Class-size
reduction, in particular, has been responsible for many of their current
challenges. The teachers clearly resent the many inconveniences posed by sub-
optimal classrooms. Capistrano is a well-managed school district with many
beautiful new facilities, a mild climate and a world-class location on the Southern
California coast. Imagine what kind of responses might come from a district
facing far more extreme physical challenges!
6.4 Bias Analysis
We did find that a few types of teachers, those with more experience or honors,
were slightly more likely (1% to 5%) to be assigned to classrooms with larger
window areas, skylights or operable windows. However, a full multivariate
regression of teacher characteristics against the Daylight Code found that none
of the teacher characteristics that we identified were significant in explaining
assignment to daylit classrooms. This model explained only 1% of the variation in
assignment to daylit classroom. We concluded that this assignment bias, while it
does exist, is extremely small.
Similarly, we found that the daylight variables remained highly significant in the
student performance models, even after the addition of information about the
teachers. While a few teacher characteristics did show up as significant
variables in our models of student performance, the daylight variables remained
extremely robust in all models.
Comparing across twelve different models of student performance in Capistrano,
we conclude that the central tendency is for a 21% increase in learning rate
between children in classrooms with minimal daylight compared to those with
maximum daylight.
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT RE-ANALYSIS CONCLUSIONS
49
6.5 Re-Analysis Report
Overall, the strength of the daylight variable in predicting student performance
stands out sharply across all of these re-analysis efforts. The addition of more
information to the models did very little to change the predicted impact of the
Daylight Code on student performance.
Only the exercise to link the Daylight Code to student attendance was
unsuccessful. This is also an extremely important finding, since it contradicts so
many claims have been made about the health effects of daylight or other indoor
environmental conditions, as reflected in absenteeism rates of building
occupants. In this study, in this school district, we did not find that any of the
physical attributes that we had available to us to classify the classrooms could be
linked significantly with student attendance.
It is also very clear from these efforts, as we re-analyzed the original data sets
with additional information, that the findings of these models are much more
strongly dependant upon the particular population studied in the analysis than
upon the subtleties of all the variables included in the models. Thus, we conclude
that it will be much more informative to try to replicate this study with a
completely different population, at a different school district, such as we will
attempt to do in Task 2.4 of this project, than it would be to continue to try to
refine the models and with further detail in the explanatory variables. This
process has been informative as a sensitivity analysis and methodological study.
We look forward to applying these lessons in the next study.
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT RE-ANALYSIS CONCLUSIONS
50
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT APPENDICES
51
7. APPENDICES
7.1 Statistical Terminology
The following briefly describes key statistical terms in the report.
Table 1
Term Name Definition
r Correlation
Coefficient
Or
Pearson correlation
Measures the strength of the linear relationship
between two variables
It can take on the values from -1.0 to 1.0, where
-1.0 is a perfect negative (inverse) correlation,
0.0 is no correlation, and 1.0 is a perfect positive
correlation.
On page 6, r is the correlation between well-
qualified teachers, and student performances.
When .61<r<.80, a strong positive relationship is
predicted.
p p-value A p-value is a measure of how much evidence
you have against the null hypothesis, i.e. that the
hypothesis is not true. (In the report on page 6,
the null hypothesis could be interpreted as: r=0).
The smaller the p-value, the more evidence you
have. (On page 6, a very small p-value indicates
that one has very high evidence that the given
correlation is significantly different from 0). The
probability of a false rejection of the null
hypothesis in a statistical test is called the
significance level.
A p-value can vary from >.00 to <1.0. The
significance level is 1-p, expressed as a
percentage. So if a p-value is .01, the
significance level is 99%.
One may combine the p-value with the
significance level to make a decision on a given
test of hypothesis. In such a case, if the p-value
is less than some threshold (usually .05,
sometimes a bit larger like 0.1 or a bit smaller
like .01) then you reject the null hypothesis.
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT APPENDICES
52
Term Name Definition
R2 Regression
correlation
coefficient
A value between 0 – 1.0 that indicates how well
an X value (or the independent or explanatory
variables in the regression) explains a Y value
(the dependent variable). Technically, the
regression equation is: Y= B0+B1X1+ B2X2+…+
BnXn+e
where B0= intercept, e=error,
so as Xs change, Y, the dependent variable,
also changes., and variations in X values cause
variations in Y.
R2 is defined as the percentage of total variation
in Y explained by the independent variables.
If R2 is equal to 1, then entire variation in Y is
explained by the independent variables, i.e. the
model is very good, and the X variables have
perfect explanatory power (for explaining Y).
So, the higher the value of R2, the better the
model is for that set of data. Models explaining
data that have a high degree of inherent
variation, such as individual behavior, will have a
much lower R2 than models explaining more
predictable events, such as group averages.
B B Coefficient Technically, the regression equation is:
Y= B0+B1X1+ B2X2+…+ BnXn+e
where B0 is the intercept (constant), and
B1 ,B2 ,…,Bn are the slopes of the regression
equation, or the coefficients of the Xs, (or the
independent variables), and e is error.
A particular Bi (i=1,2,…,n) shows how a
particular Xi variable is related to Y. If a Bi
coefficient is a positive number, an increase in Xi
by one unit increases Y by the amount of the Bi
coefficient.
Please refer to Figure 11 for a list of the B
coefficients for each independent variable.
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT APPENDICES
53
7.2 Teacher Survey
(format changed slightly to fit two pages in this appendix
CLASSROOM SURVEY
Dear CUSD Teacher,
The Heschong Mahone Group, an architectural consulting firm, has been working with
the Capistrano Unified School District on an innovative study of the relationship of the
physical classroom environment and student performance. We have been funded by the
California Energy Commission to do a follow up study to examine a few methodological
questions. To do this, we need your assistance to collect information about CUSD
teachers and their classrooms.
Please fill out this brief two-page questionnaire and return it today. All individual
responses will remain strictly confidential, and will not be released to the District, or to
anyone outside of our immediate research team. Only summary data will be reported.
Thank you for your help!
Lisa Heschong, Partner, Heschong Mahone Group
A. Please tell us about yourself:
1. Your Name: Grade Level:
2. Your current room number (location): 99/00 School:
3. How many years have you been in this classroom?
(answer questions 4 and 5 below if you have moved your classroom in the past three years)
4. Your room number from 2 years ago (97/98): Grade Level:
5. How many years in that (97/98) classroom? 97/98 School:
6. How many years have you been teaching at this school?
7. How many years have you been teaching in this district?
8. How many years have you been teaching total?
9. Your Gender: Male Female
10. Your Age: 20-39 40-59 60+
11. Your College Degrees:
12. Additional Coursework:
13. Teaching Awards:
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT APPENDICES
54
B. Please tell us about your classroom:
14. Do you feel that you had any influence on the selection of your classroom location?
This past year: Yes No Maybe/not sure
When I first started here: Yes No Maybe/not sure
Anytime in between: Yes No Maybe/not sure
15. If you could select your own classroom, what would be the three most important criteria you
would use to choose? If possible, put them in rank order (1,2,3)
16. Do you prefer teaching in a permanent or portable classroom?
Permanent classroom: Portable classroom: No opinion:
Why?
17. In general, while school is in session, how often do you:
Never Always
(*Please use the scale described below:) N/A 0 1* 2* 3* 4* 5
Open a window for ventilation
Open a door for ventilation
Close a door or window to reduce noise
Turn on a portable fan
Adjust the thermostat
Teach with the curtains or blinds closed
Teach with all the electric lights off
Teach with some of the lights off
Darken the room for TV or computer use
Do something in order to block the sun
N/A This is not possible in my current classroom
0. I could do this in my room, but I never do
1. I do this occasionally, a few days a year
2. I do this often, more than 10 times per year, depending on the weather
3. I do this often, more than 10 times per year, independent of the weather
4. I do this very frequently, about once a week or more, all year
5. I do this about once a day or more, all year
18. Any comments?
Thank you very much for your time!
If you have any questions about this survey, please contact Lisa Heschong at the address below.
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT APPENDICES
55
7.2.1 Three Most Important
Criteria in Selection of Classroom
(Answers to Question 15)
fresh paint
location
matching/appropriate furniture
my own 4 walls
water in classroom
more storage
Heating/ventilation/air conditioning
natural light
sound proofing
Quiet
room and light
storage space
walls to separate from other rooms
air conditioning/heater
noise level-
air conditioning
clean air
proximity to facilities (bathroom, cafeteria)
a door that closes
full size walls
equitable room size
brightness/airflow/lighting
size
available water
A good location, off the street and parking lot
Enough room
ventilation, temperature control (see notes)
In main building
air conditioning that works quietly
close proximity to restrooms
quiet
your are in control of noise level
limited distractions i.e. window
windows for natural ventilation and lighting.
bulletin boards,
access to water
Access to water
a 2nd window for cross ventilation/light
sufficient storage
Size
sink
windows
permanent classroom
located near grade level team
noise
size
water in classroom
storage for supplies
in the building
light
new
Inside school
close to team
close to playground access
4.away from noise
in the building
away from the lunch area
in the same pod as the grade level I'm teaching
quality health standards i.e. no asbestos
safety close proximity to school
sink
size
location in school
storage space
quiet
spacious
close to supplies
quiet environment not near the lunch area
good lighting
good ventilation, air circulation
enough space and storage
inside where the main bldg. Provides water, sinks and
center work area
easier computer printer access and classrooms are
better maintained
size
storage boards and white boards/bulletin boards, 4
cleanliness
How large is the room
Is it clean and safe
Does it have communication to 911 or office staff
Proximity to MPR for music activities I do
ventilation - airflow (catches prevailing breeze
size and brightness (windows and skylights
larger in size
keep playground noise to minimum
storage
Adequate lighting
ventilation of fresh air into classroom
room size
sink-washing hands -
science, art
white boards to eliminate dust -
safe/noise
size
water
phone
air conditioned
sink
size
air
storage
close to office
full view of street for safety during weekends
near bathrooms
size,
location
who neighbor teachers are
DAYLIGHTING IN SCHOOLS, RE-ANALYSIS REPORT APPENDICES
56
Working air conditioner
Big room
water
size
location
quality of ac
more, much more room. My current room has no room,
it is a misnomer
cupboards that opened more than a 30 degree acute
angle
built-in shelves attached to freestanding walls
location
has windows
size
size
storage
noise level
size
location, proximity to same grade teachers, playground,
office
shape
square footage (storage, too)
quiet
access to water, elect. Etc. no water in my portable
quiet surroundings
windows, yet not looking out onto playground
sink with water
location
size
noise level
permanent classroom-completely enclosed
permanent classroom with minimal noise from
neighbors
portable with adequate ventilation
Quietness
space
near bathroom
light-windows
sink
noise level quiet
size
cupboards for storage
location
size
noise level
window
Large room(space for desks, floor space & small group
space
single desks (not large tables or trapezoids
sink and storage area
quiet
sink
larger size
cabinets
sink
room size
4 closed walls
large
windows
Balanced - behavior
academic abilities and
study skills - -
quiet (solid walls
sink
built-in shelves
light-natural
outside door
size/space
outside door
sink
built in cupboards
student friendly
ample room
location
large
quiet
good a/c
size
window
outdoor passage
self contained
adequate space -
self contained classroom /4 walls doors and quiet
an air conditioner that works
larger room to allow for centers
space, present size
windows tinted
storage space closed off by moving white boards
open windows, light
quiet, insulation from other rooms
nearby work room/office
with grade level
windows