Quantification, Inequality, and the Contestation of School Closures in Philadelphia
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Sociology of Education
2019, Vol. 92(1) 21–40
Quantification, Inequality, and Ó American Sociological Association 2018
DOI: 10.1177/0038040718815167
the Contestation of School journals.sagepub.com/home/soe
Closures in Philadelphia
Meg Caven1
Abstract
Public education relies heavily on data to document stratified inputs and outcomes, and to design interven-
tions aimed at reducing disparities. Yet despite the promise and prevalence of data-driven policies and
practices, inequalities persist. Indeed, contemporary scholarship has begun to question whether and
how processes such as quantification and commensuration contribute to rather than remediate inequality.
Using the 2013 closure of 24 Philadelphia public schools as a case study, I employ a mixed-methods
approach to illuminate quantification and commensuration as nuanced processes with contingent, dualistic,
and paradoxical relationships to inequality. The quantified approach to selecting schools for closure pre-
disposed poor and minority communities to institutional loss because academic underperformance, a key
selection metric, was correlated with disadvantage. Paradoxically, academic performance measures, cou-
pled with commensuration strategies, also enabled advocates to successfully overturn closure recommen-
dations. I offer an evidentiary account of how quantification can perpetuate inequality, and I complicate
prevailing understandings of quantification as a technology of power.
Keywords
quantification, inequality, education policy, community involvement, mixed-methods, perfor-
mance metrics, commensuration, school closure
Across the country, public education policy and Furthermore, scholars of quantification contend
practice are increasingly guided by the use of data that numbers, particularly as instruments of govern-
and quantitative metrics. Standardized assessments ment, are technologies of social control that pro-
of student proficiency mandated by federal law duce and maintain unequal power relations and
(Every Student Succeeds Act [ESSA] 2015; No enable domination (Colyvas 2012; Foucault 1977;
Child Left Behind Act [NCLB] 2001), teacher eval- Porter 1995; Rose 1999).
uations undergirded by measures of student learn- School closures exemplify the confluence of
ing (Chetty, Friedman, and Rockoff 2014), and quantification, education governance, and social
the allocation of educational resources based on inequality. At the highest levels of policy, closures
computations of merit and demand1 all evince the manifest quantification: The No Child Left Behind
heightened quantification and rationalization of Act specified that schools could be shut down for
the field. The shift toward quantification in public repeatedly failing to meet minimum academic
education was motivated by a desire to measure
and equalize disparate academic opportunities and
1
outcomes uncovered by studies like the 1966 Cole- Brown University, Providence, RI, USA
man Report, but some argue that rationalization has Corresponding Author:
at best failed to remediate these inequities and at Meg Caven, Department of Sociology, Brown University,
worst, exacerbated them (Darling-Hammond 108 George Street, Providence, RI 02912, USA.
2007; Hursh 2007; Jennings and Sohn 2014). E-mail: meghan_caven@brown.edu22 Sociology of Education 92(1)
proficiency standards. Simultaneously, research students, school closures invariably spark contro-
demonstrates that school closures are both inequita- versy and contestation. A report authored by a coali-
bly distributed across different types of communi- tion of activists across 21 cities argues that closures
ties and can precipitate negative consequences for represent a systematic dismantling of the public
students and neighborhoods (Billger and Beck institutions that serve communities of color (Journey
2012; Burdick-Will, Keels, and Schuble 2013; for Justice Alliance 2014). Research confirms that
Kirshner, Gaertner, and Pozzoboni 2010; Valencia closures are unequally distributed across communi-
1980; Witten et al. 2003). Much less is known ties. Numerous case studies describe the dispropor-
about how schools are selected for closure and tionate effect of closures on low-income and minor-
what implications the selection process has for ity communities (Kirshner et al. 2010; Kretchmar
equality of educational access, opportunity, and 2011; Valencia 1980, 1984; Walker Johnson 2012;
outcomes. Given the links between quantification Witten et al. 2001). Several quantitative analyses
and school closure and between school closure confirm these claims; the stratified distribution of
and inequality, a deeper investigation of how quan- school closures was documented three decades
tification relates to inequality is needed to better ago—one study found that New York neighbor-
understand its effects in other educational or policy hoods where schools had closed had higher popula-
settings and suggest ways that decision-making tion density, older housing stock, and higher rates of
processes might be optimized for equity. public assistance receipt (Dean 1983). Analyses of
Using the 2013 closure of 24 Philadelphia public contemporary closure projects like the ‘‘Renaissance
schools as a case study, this article examines how 2010’’ closures in Chicago (Burdick-Will, Keels, and
quantification variably reproduces and remediates Schuble 2013) and longitudinal statewide closure
inequality. The study was motivated by the preva- patterns (Billger and Beck 2012) demonstrate the
lence of quantification in public schooling and edu- persistent association between neighborhood disad-
cation reform initiatives and by recent scholarship vantage and school closure.
urging a better understanding of the relationship There is also reason to be skeptical that clo-
between quantification and social inequality. Com- sures achieve their stated objective of improving
pelled by economic inefficiency and academic educational opportunities and outcomes. Rather,
underperformance, Philadelphia’s closures present research suggests that closing schools can nega-
a particularly interesting puzzle. The district’s tively affect students and neighborhoods. Multiple
data-driven selection process initially recommended studies have found that school closure is associ-
the closure of 38 schools, yet over the course of the ated with learning loss for displaced students.
community engagement process, 14 of these recom- Although this loss is sometimes recouped, stu-
mendations were overturned. Using a unique dents’ achievement only improves above the level
school-level data set that merges district administra- of their prior school when their new school is sub-
tive data with national survey data, and qualitative stantially higher performing (Engberg et al. 2012;
data from public meetings and hearings, this analy- Kirshner et al. 2010; Sherrod and Dawkins-Law
sis weaves together logistic regression techniques 2013; de la Torre and Gwynne 2009). Notably,
and qualitative analyses to address two research one evaluation of Philadelphia’s recommended
questions. First, how did the quantified approach closures found that most displaced students would
to decision making influence the distribution of clo- transition to schools no better than the schools
sure recommendations across communities? And from whence they came (Research for Action
second, how did communities’ campaigns to pre- 2013). Other studies link closure with more
serve their schools succeed or fail in the context long-lasting negative academic consequences and
of quantified approaches to decision making? a heightened likelihood of dropping out (Kirshner
et al. 2010), and a related body of work suggests
that closures also affect the communities in which
SCHOOL CLOSURES: UNEQUAL they occur, dissolving social capital, feelings of
efficacy, and social resources (Dean 1983; Valen-
DISTRIBUTION AND PERNICIOUS cia 1984; Witten et al. 2001).
CONSEQUENCES The suggestion that school closures may rein-
force existing inequalities in the education system
Although they are often defended as a means of heightens the imperative to understand how schools
improving educational opportunities for underserved are selected for closure. Boyd (1983:255) writes,Caven 23
‘‘[A]t the heart of the politics and management of of written proposals, 14 schools were removed
declining public services is the question of who from the closure list. In the summer of 2013, 24
will bear the immediate and long-term cutbacks . . . schools were shut down after a vote by the state-
whose schools will be closed?’’ In the intervening appointed School Reform Commission (SRC).
years, few studies have interrogated how schools
are selected for closure; those that do investigate
these processes highlight the role of data in manag- QUANTIFICATION AND THE
ing closure decisions (Bartl and Sackmann 2016; (RE)PRODUCTION OF
Basu 2004; Bondi 1987; Burdick-Will et al. 2013; INEQUALITY
Paino et al. 2014). For example, several years before
Philadelphia’s closures, Chicago used quantitative Extant scholarship identifies several ways quanti-
data to close underutilized and underperforming fication may be implicated in the production and
schools and in so doing also preferentially closed reproduction of inequality. First, scholars argue
schools in disadvantaged and highly segregated that quantification constructs identities and classi-
neighborhoods (Burdick-Will et al. 2013). fications of measured entities. Far from offering
a neutral reflection of reality, ‘‘political judgments
are implicit in the choice of what to measure, how
THE CASE: PHILADELPHIA, often to measure it, and how to present and inter-
2012–2013 pret the results. . . . Numbers, like other inscrip-
tion devices, actually constitute the domains they
In December 2012, the school district of Philadel- appear to represent’’ (Rose 1999:198). Once clas-
phia announced recommendations to close 38 of sified through quantification, measured entities
its 241 schools. As in many urban centers across become subject to judgment, stigma, and material
the country, Philadelphia’s closures were proposed consequences. In the case of public schooling,
as a remedy for massive inefficiency. A declining quantification, through choices regarding what to
urban population, budget cuts, and the prolifera- measure and the institutionalization of measure-
tion of charter schools had left the district with ment practices, does not identify but instead cre-
more than 50,000 empty seats and a budget short- ates the ‘‘failing school,’’ a type of institution
fall of $1.35 billion over five years (Jack and Slud- that is constructed and perceived as being worthy
den 2013). of punitive sanctions, including closure (Deeds
With crisis looming, the district hired a consul- and Pattillo 2015; Walker Johnson 2012).
ting firm to guide the closure and consolidation Relatedly, scholars have demonstrated that
process. Following other urban districts, Philadel- choices regarding how to measure a construct
phia took a highly quantified approach to selecting can also have significant computational conse-
schools for closure. A 2011 draft report nominated quences. Using comparative case studies, Four-
two principal criteria for issuing recommenda- cade (2011) illuminates how different approaches
tions: utilization and a building condition metric to measuring the same construct (damages caused
called Facilities Condition Index (FCI) (URS by oil spills) in France and the United States
2011). By the time closure recommendations yielded vastly different remunerative outcomes
were presented to constituents in 2012, this list for communities in the two settings.
of criteria had expanded to include projected sav- As results of this study will show, the four met-
ings and academic performance. rics Philadelphia officials used to select schools
According to the superintendent and district for closure had different meaning-making and
representatives, an initial list of 180 schools was computational consequences. In their explanations
identified using the aforementioned metrics. This of individual closure recommendations, the dis-
list was then narrowed to 50 schools through con- trict deployed measures of academic performance,
versations with community members that incorpo- utilization, building condition, and financial con-
rated concerns about transportation, feeder pat- cerns interchangeably, yet it is important to note
terns, and neighborhood school options. Further that these four criteria reflect two distinct justifica-
review of data and individual cases narrowed tions for shutting down schools. Measures of
this list again to 38 schools.2 After an extensive building condition, utilization, and projected sav-
community engagement process that included ings reflect the district’s resource management
public meetings, formal hearings, and a solicitation imperative, a largely agnostic attempt to improve24 Sociology of Education 92(1)
efficiency by consolidating students into fewer and Bell 2014; Lamont, Beljean, and Clair
buildings. Academic underperformance, on the 2014). First, I bolster extant research on school
other hand, sanctions particular schools for their closure processes by presenting a detailed account
individual failings. The rationale for closure var- of how the social and mechanical facets of quanti-
ied across individual proposals, and distinctions fication collude to unequally distribute closure
between resource management and academic jus- recommendations across communities. Philadel-
tifications proved critical for the distribution of phia’s quantified process predisposed schools in
closure recommendations and communities’ abili- low-income and minority communities to closure,
ties to preserve their schools. and it obscured that predisposition as neutral,
Finally, commensuration, which Espeland and objective, and inevitable. Quantitative results rein-
Stevens (2008) describe as the rank-ordering of force the role of academic performance measures
objects using a common metric, compromises in concentrating school closures in disadvantaged
equity by obscuring significant contextual factors communities (Burdick-Will et al 2013), and qual-
underlying the data. Commensuration, they argue, itative analyses expose decision makers’ insis-
transforms ‘‘qualities into quantities, difference tence on the unassailable impartiality of data in
into magnitude’’ (Espeland and Stevens 2008: closure decisions.
316). Colyvas (2012:169), for example, character- Second, by illuminating the ways in which
izes achievement scores, upon which education quantification equipped some communities with
policy has come to rely heavily, as ‘‘formalized the tools to successfully resist closures, I expand
abstractions’’ that bring a ‘‘patina of objectivity’’ current understandings of quantification as a tech-
to comparison, benchmarking, and decision mak- nology of power. Although scholarship to date
ing. She argues that quantification often ‘‘under- consistently characterizes quantification and com-
min[es] the identity, mission, and, ironically, mensuration as instruments of power and domina-
accomplishments of organizations.’’ The repeated tion in the hands of authority (Espeland and Ste-
use of rankings and quantitative metrics reinforces vens 2008; Foucault 1977; Rose 1999; Salais
understandings of schools as isomorphic organiza- 2012), several authors suggest that laypersons
tions (Meyer and Rowan 1977; Rowan 1982); it might access power through quantification (e.g.,
lines schools serving elite youth from affluent Desrosières 2014). Espeland and Stevens (1998:
communities up against those contending with 332) posit that commensuration can ‘‘arm dis-
devastating socioeconomic disadvantage, arrang- senters,’’ and Porter (1995) hazards the democra-
ing them hierarchically as if they were equivalent tizing character of quantification, illustrating
and easily comparable. how new professional classes eke out legitimacy
These effects were evident throughout Phila- in the shadow of parochial regimes of expertise.
delphia’s school closure process. Raw compari- Despite these claims, however, both studies devote
sons of schools’ academic performance grouped the bulk of their attention to quantification’s predi-
all schools together regardless of populations lection toward technical inscrutability and distanc-
served, concealing contextual details such as the ing the subject from the agent, undercutting the
proportion of students who are English Language value of local knowledge. Moreover, the subjects
Learners, have special needs, or are living below Porter (1995) describes as empowered by quanti-
the poverty line, all factors that sociologists have fied objectivity are far from powerless: account-
linked to lower test scores (Coleman 1966; Ladd ants, technocrats, engineers, scientists. This pro-
2012; Reardon 2011). Consequently, as results of ject reconsiders the unmet ideals of these
this study will show, commensuration—especially scholars’ democratization hypothesis, and it con-
using academic performance metrics—contributes tributes an alternative understanding of the rela-
to the concentration of closure recommendations tionships linking constituents, quantification, and
in disadvantaged communities. the authority of decision makers to the equity of
distributive outcomes.
Finally, I offer a populated account of quantifi-
cation as a nuanced set of processes and structures
THE PRESENT STUDY
whose relationship to inequality is highly contin-
I offer several contributions to a growing literature gent. Specifically, I demonstrate just how much
examining the linkages between cultural processes metrics matter for the crystallization of relation-
(including quantification) and inequality (Asad ships between quantification and inequality. InCaven 25
Philadelphia, the coexistence of academic perfor- the Philadelphia Public Schools Facilities Master
mance and resource management metrics in a Plan (FMP), the NCES Common Core of Data
decision-making framework and the varying sig- (CCD; 2009–2010),3 and the decennial census
nificance of each factor across closure recommen- from 2000 and 2010. The district-produced FMP
dations created conditions that both stratified the data report a broad range of school-level variables
distribution of closure recommendations and fur- describing school facilities, climate, utilization,
nished advocates with effective tools of resistance. and academic performance. I linked student demo-
Absent academic performance as a metric for clo- graphic data from the Common Core of Data to
sure recommendation, proposed closures might Facilities Master Plan data using school identifica-
have been less concentrated among disadvantaged tion numbers, school names, and street addresses. I
communities. Yet, absent academic performance, used latitude and longitude coordinates for each
these same communities may also have been less school included in the Common Core of Data to
successful in saving their schools. link schools to neighborhood-level (census tract)
data on socioeconomic conditions drawn from
the 2000 and 2010 decennial census using GIS
DATA AND METHODS software.
The analytic sample does not include the 87
This project uses a mixed-methods approach to
charter schools in the data set as they are not sub-
investigate the relationship between quantification
ject to the same governance structures and pro-
and inequality through the lens of school closure
cesses as traditional public schools. For seven
decisions in Philadelphia. Qualitative and quanti-
schools with missing data on key independent var-
tative analyses inform each other in a multiphase
iables, I used the MI package in Stata 14 to create
design (Creswell and Plano Clark 2007; e.g., Des-
five imputed data sets using multivariate normal
mond 2012). My initial explorations of descriptive
regression to estimate missing values. The final
and geospatial data revealed racial and socioeco-
sample includes 240 public schools in the School
nomic inequalities between schools recommended
District of Philadelphia that were subject to the
and not recommended for closure. Subsequent
closure decision-making process in 2012.
investigations of the qualitative data illuminated
the importance of quantified decision-making
frameworks in making recommendations and the
variable use of resource management and aca- Qualitative Data
demic performance rationales in justifications of Qualitative data were generated by the FMP com-
individual closure recommendations. I proceeded munity engagement process and were accessed
to test the relationships between school and neigh- through the School District of Philadelphia’s web-
borhood disadvantage, quantitative selection crite- site.4 Sources include videos of 15 public meetings
ria, and the likelihood of closure recommendation (~30 hours total) and five formal hearings (~24
using logistic regression techniques. I found that hours total) as well as 43 written proposals and
closure logics and their corresponding selection 44 individual communications responding to clo-
criteria varied in their effect on the relationship sure recommendations. I uploaded these data
between closure recommendation and disadvan- into NVivo 10 for coding and analysis. Although
tage and that quantitative differences failed to some might argue that video misses key dimen-
explain variation in recommended schools’ clo- sions of a social setting, Jones and Raymond
sure outcomes. I then used qualitative data from (2012) contend that video presents researchers
the community engagement process to examine with opportunities to access moment-by-moment
whether and how communities’ participation in records of settings and interactions. Recordings
debates over school closures determined which of community meetings provide a detailed catalog
schools were preserved and which were shut of individual testimonies and district responses—
down. data crucial to the following analyses.
I began by taking comprehensive field notes
and transcribing relevant sections of recorded
Administrative and Survey Data meetings. Quantification and commensuration
The data set for the quantitative analysis is com- emerged early as important argument tactics for
posed of data from a variety of sources, including both district personnel and community members.26 Sociology of Education 92(1)
These constructs and associated subthemes, whether different metrics would have yielded a dif-
including academic- and resource-related claims, ferent set of closure recommendations and borne
formed the core of my codebook. Qualitative a different relationship to disadvantage, but here I
data were also coded to specific schools, which am concerned with the specifics of how the dis-
allowed me to use queries and matrices to compare trict’s actions and choices distributed closure rec-
schools individually and by closure status. ommendations across communities.
Utilization. Utilization is calculated by
MEASURES dividing school enrollment by school capacity.
Capacity for each school was calculated by multi-
Dependent variables plying the number of classrooms in the building
Recommendation for closure. I assigned all by the maximum students per classroom (28 for
schools a binary variable indicating whether they middle and high school, 26.5 for elementary
were recommended for closure. Because the Facil- school) and then adjusting the total downward
ities Master Plan made a broad range of recom- (75 percent) to account for classrooms that require
mendations, including grade reconfiguration, additional space for curricular activities or do not
merger, relocation, and closure, I define school operate at maximum capacity.
closure in this study by program closure. Qualita-
tive data indicate that building closure was signif- Facilities Condition Index. Facilities Con-
icant for reasons of convenience and community dition Index is a common facilities management
resource, but the real wound of school closures industry metric that measures building condition
was inflicted by dismantling the school as a social by dividing the cost to repair the building by the
organism: a faculty and student body who estab- cost to replace the building. To calculate repair
lished a shared culture and identity under a com- costs, URS Corporation inflated the required ren-
mon name. Thus, schools that retained their names ovation costs identified in a 2005 Facilities Condi-
and moved to new buildings with student and tion Assessment to 2010 values and then deducted
teacher populations intact are coded as not recom- investments made in facilities in the intervening
mended even if the building they left behind was years. They calculated replacement cost by multi-
shut down. Facilities where the program was dis- plying a cost per square foot value by the overall
banded but the building remained open to accom- square footage of the building. Buildings with low
modate a new school were coded as recommended FCI (\33 percent) are considered to be in good
for closure. condition; those with FCI above 75 percent are
in poor condition.
Preservation. Schools recommended for clo-
sure were assigned an additional binary variable Savings. Savings metrics are not included in
indicating their preservation status. Schools that the URS Corporation draft report on long-range
were recommended for closure but left open are planning, but FMP data include numerous savings
coded as preserved (1), and those ultimately shut measures for each school, including operational,
down are coded as not preserved (0). facilities, and total savings as well as per-pupil
computations of each. In the present analyses,
Independent variables. The four criteria the savings variable measures the district’s projec-
the district used to select schools for closure—uti- tion of total savings associated with closing each
lization, Facilities Condition Index (FCI), savings, school. This choice reflects the district’s focus
and academic performance—comprise key inde- on reducing the deficit by any means possible,
pendent variables for this analysis. The first two as opposed to a specific per-pupil, facilities, or
of these criteria and the background information operational savings goal.
in the following first appear in a draft report pub-
lished by URS Corporation, an engineering, design, Academic performance. Academic per-
and construction firm hired by the Philadelphia formance measures reflect the proportion of students
School District to guide their Imagine 2014 long- scoring proficient or above on Pennsylvania System
range facilities planning campaign (URS 2011). A of School Assessment (PSSA) reading tests. These
separate analysis could productively interrogate data were reported in the district’s FMP data set.Caven 27
Other key independent variables measure school communities losing their schools. I begin by using
and community disadvantage. At the school level, binary logistic regression to evaluate how school-
the proportion of the student body that is black and neighborhood-level indicators of disadvantage
and the proportion receiving free or reduced-price (Models 1 and 2, respectively) are associated with
lunch are drawn from the CCD. I include a measure a school’s likelihood of being recommended for
of the proportion of the student body who come closure. To test whether there is a spatial element
from within the attendance boundary as a proxy to the stratification of school closure recommen-
for school desirability and families’ capacity to dations, I introduce school- and neighborhood-
navigate the choice process. Additionally, commu- level measures of disadvantage sequentially in
nities’ capacity to organize may be influenced by Models 1 and 2. The two subsequent models intro-
geographic proximity or underlying political and duce the district’s selection criteria to test how the
human capital. Neighborhood (census tract) meas- relationship between disadvantage and recommen-
ures of median household income, proportions of dation for closure changes in the context of quan-
households headed by women, and proportions of tification. To distinguish between the effects of
adults without high school diplomas are drawn two underlying justifications for school closure
from the 2010 decennial census. (resource management and academic underper-
Table 1 presents descriptive statistics for the formance) on this relationship, I introduce the
analytic sample. The table presents separate statis- resource management criteria in Model 3 and
tics for schools recommended and not recommen- add academic performance in Model 4.
ded for closure to highlight the variation between The model used in this phase of the analysis
these two groups. As existing research on school can be expressed as follows:
closure would suggest, schools recommended for
closure appear less utilized (50 percent vs. 77 per-
Recommended
cent), less academically proficient (29 percent vs. log 5n
ð1 Recommended Þ
77 percent), in worse structural condition (FCI 44
n5b0 1b1 School Factors1b2 Neighborhood
percent vs. 38 percent), and more expensive on
a per-pupil basis ($2,126 vs. $1,160) than schools Factors1b3 SelectionCriteria1e;
not recommended for closure. However, recom-
where SchoolFactors represents multiple varia-
mended schools also received less investment
bles describing the school’s demographic charac-
between 2003 and 2012 ($1,036 vs. $2,706) and
teristics, and NeighborhoodFactors represents
serve a higher proportion of poor (90 percent vs.
multiple variables describing the surrounding
82 percent of students receiving free or reduced-
neighborhood’s social and demographic environ-
price lunch), minority (85 percent vs. 60 percent
ment. SelectionCriteria are school-level scores
of students are black), and neighborhood (69 per-
on the district’s four metrics used to identify
cent vs. 66 percent in-boundary) students than
schools for closure.
schools not recommended for closure. The charac-
teristics of the neighborhoods in which schools are
situated also show some variation between recom-
mended and not recommended schools. In neigh-
FINDINGS AND RESULTS
borhoods surrounding schools facing closure, the Quantification and the Unequal
median household income is lower by nearly
$10,000 ($26,430 vs. $36,310), and rates of family Distribution of Closure
poverty and other indicators of disadvantage are Recommendations
higher than in neighborhoods around schools not Table 2 shows results from four logistic regressions
recommended, although the magnitude of some predicting schools’ likelihood of recommendation
of these differences is small. for closure. Results from Models 1 and 2 affirm
prior findings that schools serving disadvantaged
METHODS populations are more vulnerable to closures than
their peer institutions. Model 1 demonstrates that
The quantitative analysis uses logistic regression schools with high concentrations of black students
techniques to establish whether and how quantifi- are disproportionately likely to face closure. In
cation influenced the likelihood of disadvantaged a school of 100 students, each additional black28 Sociology of Education 92(1)
Table 1. Descriptive Statistics: Philadelphia Public Schools Recommended and Not Recommended for
Closure, 2012–2013.
Recommended for Not Recommended
Closure N = 38 N = 202
School Characteristics Mean/Percent SD Mean/Percent SD
Facilities
Utilization (enrollment/capacity) 50.29 20.49 77.02 21.88
Capacity 826.81 475.87 776.31 421.36
Facilities Condition Index (FCI) 43.82 21.95 38.15 24.38
Capital investment 2003–2012 ($1,000) 1,036.04 2,440.70 2,706.04 9,336.24
Demographics
Percent black 84.97 24.42 59.61 32.2
Percent in boundary 69.26 28.26 66.8 33.02
Percent free/reduced lunch 89.78 13.01 82 22.68
Academics/climate
Percent students proficient in reading 29.08 10.1 47.2 19.04
Percent students absent 101 days 54.29 14.38 44.97 14.04
Percent students suspended 16.66 8.28 10.54 8.1
Savings
Total savings ($1,000) 742.06 441.11 831.25 369.02
Savings per student ($100) 2,126.71 974.62 1,610.18 621.84
School configuration
Elementary 24.32 25.12
K–8 37.83 39.9
Middle 13.51 8.87
High 24.32 20.19
Neighborhood characteristics
Median household income ($1,000) 26.43 11.81 36.31 71.27
Unemployment rate 17.73 6.93 14.57 7.77
Percent below poverty line 27.73 15.08 22.74 16.03
Percent female-headed households 13.7 7.58 12.48 8.52
Percent lower than high school 18.02 7.72 15.27 7.23
student would increase the school’s likelihood of several interesting findings. First, resource man-
being recommended for closure by 3.1 percent. agement criteria, at least in part, drive closure rec-
The addition of neighborhood-level characteris- ommendations. Consistent with Burdick-Will and
tics in Model 2 reveals that although school-level colleagues’ (2013) findings, school utilization is
poverty was not a significant predictor of recom- especially predictive of a school’s recommenda-
mendation for closure, median household income tion status. All else being equal, a 1 percentage
at the neighborhood level was negatively and sig- point increase in a school’s utilization decreases
nificantly associated with recommendation. A its likelihood of recommendation for closure by
$1,000 increase in median neighborhood household roughly 5 percentage points. In this model, neither
income reduced a school’s likelihood of recom- facility condition nor projected savings are predic-
mendation for closure by 4.5 percent. Note that tive of closure recommendation. Correlation
relationships between school disadvantage and like- between the three resource management measures
lihood of closure recommendation persist when is low (\ .17), suggesting that insignificance is not
neighborhood-level controls are added. the result of multicollinearity.
The addition of the district’s resource manage- In an exploratory set of models that predict build-
ment criteria to these regressions in Model 3 yields ing closure instead of program closure (AppendixCaven 29
Table 2. Odds Ratios Predicting Relationships between Indicators of Disadvantage, District Closure
Selection Criteria, and Likelihood of Recommendation for Closure.
Model 1 Model 2 Model 3 Model 4
School characteristics
Percentage black 1.031*** 1.030*** 1.018* 1.012
(.008) (.008) (.009) (.009)
Percentage free and reduced lunch 1.012 1.005 .997 .978
(.012) (.014) (.017) (.022)
Percentage students in boundary 1.372 1.639 3.033 .687
(.895) (1.171) (2.623) (.714)
Neighborhood characteristics
Median household income ($1,000) .954* .955* .958
(.018) (.019) (.021)
Percentage female-headed household .961 .968 .934
(.029) (.029) (.034)
Percentage no high school diploma 1.003 .987 .981
(.036) (.041) (.046)
District closure criteria
Facilities Condition Index (FCI) 1.016 1.020
(.010) (.011)
Utilization .952*** .957***
(.010) (.012)
Total projected savings ($1,000) .999 .998*
(.001) (.001)
Percentage proficient in reading .899***
(.024)
Constant .006 .068 5.694 12070.03
Source: NCES Common Core of Data (2009–2010), the Philadelphia Public Schools Facilities Master Plan, and the
decennial census (2000, 2010).
*p < .05. ***p < .001.
A), the Facilities Condition Index does predict rec- to close for underlying reasons, such as housing
ommendation. This suggests that although building expensive and difficult to move career and technical
closure decisions were tightly coupled to selection education (CTE) programs, recent large capital
criteria, recommendations of program closures dis- investments, or their political status as highly funded
rupted the relationship between building condition Promise Academies. When overall projected savings
and building closure recommendation. In other is replaced with a per-pupil measure of savings, the
words, schools recommended for program closure already small coefficient becomes insignificant.
were recommended for reasons beyond resource Second, the relationship between disadvantage
management; academic logics weighed more heavily and closure recommendation persists with the
on program closures than building closures. addition of resource management selection crite-
Negligible or weak relationships between savings ria. The coefficient on schools’ racial composition
or facility condition and closure recommendation is somewhat reduced in size, from a 3 percent to
persist in Model 4, where the projected savings coef- 1.8 percent increase in the likelihood of closure
ficient becomes significant, although in the opposite recommendation for each additional percent of
direction from what one would anticipate. We would the student body that is black; nonetheless, the
expect the district to recommend the most expensive relationship remains significant. The size and
schools for closure, but the likelihood of a school’s strength of the relationship between median
recommendation in fact decreases marginally as pro- household income and likelihood of closure rec-
jected savings increase. This may reflect a subset of ommendation holds constant with the addition of
highly funded or expensive schools that are unlikely the new variables.30 Sociology of Education 92(1)
Taken together, the results of Model 3 imply recommendations, but the preservation of 14
that resource management criteria are indeed schools over the course of the community engage-
used to make closure recommendations and that ment process raises questions about how schools
they are largely independent of school and com- were removed from the list. One possibility is
munity disadvantage. While it is simultaneously that the district used selection metrics to close
true that disadvantaged schools and communities the most underutilized, worst performing schools.
are likelier to be exposed to closure recommenda- Alternatively, underlying disadvantage may have
tions and that quantitative measures related to played a role in schools’ divergent outcomes.
resource management are used to drive those rec- However, a comparison of criteria scores and
ommendations, results indicate that these patterns demographics (Table 3) reveals few significant
are unrelated. differences between closed and preserved schools.
Results from Model 4 add nuance to our under- Data-driven decision-making processes may have
standing of the relationship between quantification largely predicted recommendation for closure,
and the unequal distribution of closure recommen- yet mechanical quantification processes failed to
dations. Academic performance, the final criteria govern final decisions. To explain the variation
used to select schools for closure, is strongly asso- in closure outcomes, I turn to the qualitative data
ciated with closure recommendation. For every generated by the community engagement process.
additional percent of the student body scoring pro-
ficient or above on state reading tests, the likeli-
hood of closure recommendation decreases by Pursuing Preservation: Advocates’
over 10 percent. However, in contrast to Model 3, Engagement with Quantification in
the addition of academic performance metrics to
the model causes previously significant relation- Struggles over School Closures
ships between disadvantage and closure recommen- In keeping with other quantified policy domains,
dation to disappear. This disappearance demon- school district representatives did their best to
strates a strong correlation between academic characterize the data-driven selection process as
performance and measures of disadvantage. How- streamlined, formulaic, and objective. At the
ever unsurprising, the implications of this correla- opening of one community meeting, Superinten-
tion and the theoretical considerations for quanti- dent Hite dispelled the notion that closure recom-
fied approaches to closure decision making are mendations were subjective, vindictive, or ran-
worth making explicit. Specifically, schools in dom. Using data to prove the imperative of
poorer neighborhoods serving higher proportions school closures and legitimate the process of clo-
of minority students will always be more likely to sure selection, he contrasted his administration’s
face closure when academic performance is used approach with the image of an autocratic district:
as a selection criterion (Burdick-Will et al. 2013). ‘‘This is not a superintendent exerting his will
Changes between Models 2, 3, and 4 illustrate upon a set of schools and communities, this is
how quantitative selection criteria operate as indirect a necessary process in Philadelphia. . . . If we
mechanisms for unequally distributing closures across don’t close schools now, we could be talking about
disadvantaged schools and communities. By abstract- closing the whole district.’’ He went on to explain
ing the relationship between school and neighborhood that the district was carrying too many empty seats
disadvantage and the likelihood of closure recom- (53,000), it could not pay for the programs stu-
mendation, benign-seeming metrics designed to facil- dents needed, and that ‘‘[t]his meeting will cover
itate decision making simultaneously construct and the criteria that were used to come up with [clo-
conceal the relationship between disadvantage and sure] recommendations.’’
school closure. Importantly, not all quantitative meas- The central role of numbers and the implica-
ures are equally responsible. These results show that tion of their infallibility were echoed by other dis-
whereas academic performance metrics ensure a dis- trict personnel throughout the community engage-
proportionate distribution of closure recommenda- ment process. When asked about individual
tions, allocation of closures based on resource man- closures in community meetings, district represen-
agement concerns occurs largely independent of tatives rattled off a string of metrics that justified
school and community disadvantage. the recommendation to close a given school as if
The preceding analyses affirm that quantified the outcome were self-evident. Yet, qualitative
approaches to selection directed closure analyses reveal two distinct fissures in theCaven 31
Table 3. Comparisons of Closed and Preserved Schools.
Closed Preserved One-tailed t Test
District closure criteria
Facilities Condition Index (FCI) 46.76 40.07 .185
Utilization 50 50.79 .456
Total projected savings ($1,000) 801.47 640.24 .142
Percentage proficient/advanced in reading 27.96 31.93 .124
School characteristics
Percentage black 81.47 90.95 .121
Percentage free and reduced lunch 90.58 88.51 .313
Percentage in boundary 65.75 75.28 .161
Neighborhood characteristics
Median household income ($1,000) 25.72 27.65 .316
Percentage female-headed household 15.18 11.15 .057
Percentage no high school diploma 18.71 16.83 .238
N 24 14
Source: NCES Common Core of Data (2009–2010), the Philadelphia Public Schools Facilities Master Plan, and the
decennial census (2010).
mechanical objectivity of the district’s quantified Commissioner Dworetzky: And it has weight, but
approach to decision making. First, comparing you can’t say what exactly the weight is
explanations for individual schools’ closure rec- because you didn’t have like a grid in which
ommendations reveals that authorities used the you put a certain amount of weight on it.
four selection criteria inconsistently across
schools. The relative importance of each metric Later, Commissioner Dworetzky returned to this
varied by case, with some closures ultimately point:
assured by resource management rationales and
others by academic underperformance. By the time we get to the end of these hear-
Second, the meaning and legitimacy of aca- ings, I hope we’ll have more clarity around
demic performance measures in directing closures how academic performance should be used
was unstable and contested. At one formal hear- in this context. I am not sure we have
ing, a commissioner interrupted proceedings to a meeting of the minds on what weight it
ask how performance was used to make recom- should be given, but I do observe that in
mendations and how it should be evaluated more a situation where it’s both a criteria for clo-
generally in making closure decisions. Mr. Kihn, sure, and then we’re going to replicate
a school district staff member, responded: much of the same scenario, it doesn’t
make a great deal of sense to me and I
Mr. Kihn: We did the original filtering of the four want to understand better what the thinking
areas of the school that we described, utiliza- is on that. . . . Closure of a building, to me,
tion, costs, academic performance, and the con- I’m not sure what that does for academic
dition of the facilities and we didn’t assign each performance. It’s a question then of the stu-
of those particular weight. So, it wasn’t as dents. Where are they going to go? What
though we were over indexing on one or the opportunities are they going to have in the
other, but we did try to look at the whole place that they go to? (Closure hearings,
picture. day two)
Commissioner Dworetzky: Well, I am not sure
what that means. Is it a factor in making the This exchange illuminates both the malleability of
proposal? the quantified decision-making process and the
Mr. Kihn: Yes, it’s a factor. contested and politically unstable nature of32 Sociology of Education 92(1)
Figure 1. Pathways to preservation in recommended schools.
academic performance as a closure criterion. Mal- quantification as a valid framework for decision
leability enabled closure rationales and the impor- making. Together, the variation in the district’s
tance of individual criteria to vary across schools. justifications for closures and communities’ argu-
Because officials used data to look at the ‘‘whole mentation strategies mattered for schools’ fates in
picture,’’ they could flexibly adjust the weight of the closure process. Figure 1 plots the various
a given factor in each case; some closures were interactions between the district’s underlying rea-
substantiated by resource concerns, whereas others sons for closure recommendation, communities’
were undergirded by academic underperformance. arguments against closure, and closure outcomes.
The political instability of using academic perfor- For communities to succeed in getting a school
mance to justify school closure in conjunction removed from the closure list, advocates had to
with the measure’s distinctiveness from and sim- wage a quantified campaign that aligned with the
plicity compared to the chorus of the three district’s underlying rationale for closing the
resource management metrics likely contributed school. Yet it was not always clear to advocates
to its vulnerability in the face of quantified resis- which factors drove the closure recommendation
tance campaigns. Together, variation in underly- of specific schools, rendering successful campaigns
ing rationales for individual closures and the con- all the more difficult to design, even when the
tested nature of academic performance measures imperative for leveraging quantification was clear.
in closure deliberations set the stage for communi- Successful campaigns consistently used aca-
ties’ variable success in their campaigns for demic performance measures, which enabled sim-
preservation. plistic commensurative claims. In cases where
Like the district’s rationalization of individual closure recommendations were justified by
closure recommendations, communities’ strategies resource issues, advocates could not secure preser-
for arguing against closure also varied. Some vation through commensurative academic argu-
deployed data and commensuration to argue for ments, even if there were an academic case to be
the preservation of their schools, making argu- made, because such claims did not align with the
ments that spoke to both resource and academic underlying rationale for closing that specific school.
rationales for closure, whereas others contested Yet, communities’ quantified resource managementCaven 33
arguments also failed to preserve schools slated for took a moment to consider the human
closure, even when resource issues undergirded rec- side, this conversation of school closures
ommendations, because the resource management would be over. The human side of this is
landscape of school closures was too complex to safety, strain on families, and education
contest with either normative or commensurative just to name a few. (Community activist
arguments. For example, claims that schools should and parent, closure hearings, day one)
be preserved because they had received significant
investment in recent years (normative) or their Others were less explicit in their opposition to
building was in better condition than other schools quantification, endeavoring instead to qualita-
(commensurative) failed because they oversimpli- tively construct their schools as worthy of preser-
fied the broader ecosystem of school resources. vation. They attempted to validate the contextual
This pattern is somewhat counterintuitive as factors and other ways of valuing entities that
‘‘school quality’’ is often thought of as difficult to quantification and commensuration obscure
measure, whereas economic indicators are less neb- (Espeland and Stevens 2008; Lamont 2012).
ulous. However, in this instance of data-driven Despite the emotional power and the persuasive-
decision making, academic performance constructs ness of these types of arguments, however, they
were simplistic and easy to commensurate, whereas failed to secure schools’ preservation. For example,
measures of schools’ efficiency were mathemati- advocates argued that there was far more to George
cally complex, interdependent, and drew on infor- Pepper Middle School than data conveyed:
mation that was beyond advocates’ reach. Finally,
given the critical role of data in both making and If you look at the first set of pictures after
refuting closure recommendations, it is not surpris- my testimony, you’ll see the multipurpose
ing that advocates attempting to secure preservation fields, tennis and basketball courts, and
through qualitative arguments were never effective the beautiful greenery surrounding the
in resisting closure. school . . . the beautiful Heinz National
Wildlife Refuge. This is Southwest’s equiv-
alent to the schools . . . in the suburbs. This
No Data, No Chance: Failures of is an experience that inner-city children
rarely have a chance to partake. Now,
Qualitative Arguments look at the picture of Tilden [proposed
School closures in Philadelphia evoked fervent receiving school]. This school has none of
pleas to preserve schools based on unquantifiable the amenities associated with Pepper. Why
aspects of their character. Advocates portrayed take students away from this environment
schools as historical and invaluable community to an environment that does not have this
spaces, families, bastions of safety in chaotic greenery and open air? (Community activ-
neighborhoods, and critical sources of stability in ist, closure hearings, day two)
students’ unpredictable lives; they challenged clo-
sures on the grounds of racial justice. In so doing, Advocates portrayed Pepper as a unique and
they promoted, sometimes explicitly, an alterna- pastoral school with unquantifiable value to the
tive to the quantification narrative. One advocate’s students and community. Data-driven approaches
testimony simultaneously acknowledged and chal- to decision making, their arguments implied, over-
lenged the dominance of the quantitative approach looked the value of experiencing a college-like
to decision making: setting, especially for low-income students. Nota-
bly, even this qualitative construction of Pepper’s
I had come to talk about budgets and the value is commensurative. Advocates established
economic side of this, but then I asked Pepper’s exceptionalism by comparing it against
myself—that would be useless . . . num- Tilden. Despite this commensuration, which sub-
bers can be manipulated to look any way sequent analyses will demonstrate was critical
that you want them to look. One thing you for securing preservation of some schools, the
can’t manipulate is the human side to these SRC voted to close Pepper.
proposed school closures. There is a human Pepper’s closure demonstrates the limits of
side to every argument. And if Dr. Hite, both qualitative argumentation and commensura-
Mayor Nutter, and the SRC Board actually tion as tools of resistance. Commissioners were34 Sociology of Education 92(1)
not entirely deaf to claims that schools were qual- For Logan, theirs was a dramatic drop
itatively valuable to their constituents, expressing from 2005 ’til now . . . Cooke’s scores had
concerns about closing a school that ‘‘offers things a slight drop, then a major increase. . . .
that no other school in the district does’’ (closure Their attendance is 96 percent for both staff
hearings, day three), but they would not preserve and students; two times more entries than
a school based on these qualities alone. Instead, withdrawals from 2004 ’til now. Steady
commissioners urged the district to reevaluate decreases in suspensions, violent incidences,
enrollment projections, transportation plans, and number of children with more than two sus-
engineering concerns in an attempt to quantita- pensions and an increase in the number of
tively justify preservation. students passing with Algebra at the 8th
Qualitative arguments failed even when they grade for their exam. Now, closing Cooke
relied on comparisons against receiving schools. school when you have the two schools that
Subsequent analyses will demonstrate that com- you’ve selected that are having a downward
mensuration was a successful tactic for preserving incline in their test scores, behavioral issues,
schools, but only when advocates used metrics and attendance issues and you’re keeping the
that the district deemed relevant in decision mak- school that should be open to close. (Field
ing. The district’s quantified approach to selecting notes, Facilities Master Plan meeting,
schools for closure dictated the things that mat- December 18, 2012)
tered (e.g., utilization, academic performance)
and how meaning was made (commensuration). Cooke Elementary’s written proposal to the dis-
As we will see, communities’ campaigns against trict included 12 slides of tables comparing its test
closure were sometimes successful when they scores in multiple grades and years to those at the
fully adopted quantification and commensuration, proposed receiving schools. Other schools’ support-
but they fell short of their goals when they adopted ers also successfully used this strategy. Advocates
the tactics without the terms. for the preservation of Tanner Duckrey vocifer-
ously argued that sending their students to Stanton,
Pathways to Preservation: the proposed receiving school, meant sending them
Commensurating Academic to a school that was worse than the one they pres-
ently attended. District personnel revised their clo-
Performance sure recommendations ahead of the SRC’s final
Quantitative analyses demonstrate that the use of vote, removing Cooke from the recommendation
academic performance measures in decision- list and reversing their decision to close Duckrey.
making processes concentrated school closures in They proposed closing Stanton instead. Asked
disadvantaged communities. Paradoxically, these about the latter change by the SRC, Superintendent
same metrics also enabled communities to suc- Hite explained that it was brought to the district’s
cessfully advocate for the preservation of their attention that Stanton was a lower performing
schools. In numerous cases, advocates gained trac- school than Duckrey and that there was space to
tion with the district and School Reform Commis- accommodate Stanton’s students at Duckrey
sion by using academic performance data to argue instead (SRC vote, March 7, 2013).
against their school’s closing. These arguments
relied heavily on commensuration, often compar-
ing the academic performance of the recommen- Succumbing to Closure: Argument
ded school against that of the school(s) slated to Misalignment and Resource
receive its displaced students. At an early commu-
nity meeting, a mother arguing against the closure Ecosystem Complexity
of Jay Cooke Elementary testified: Commensurative arguments that leveraged aca-
demic performance data did not always overturn
I went directly off of you guys’ website, and school closure recommendations. When these
the two schools you have selected for ele- arguments failed, it was often because of underly-
mentary grades is Steel and Logan, and the ing and wider reaching resource management
other school is Grover Washington. For issues that the community struggled to navigate
Steel, their scores have dropped 40 percent and could not overcome. For example, Edward
in four years of straight downward incline. Bok Technical High School was slated for closureYou can also read
