What predicts a school's ties to outside organisations

13,167 schools · 8 organisations · 6 surveys · 2005-06 to 2021-22 · survey-weighted

Press Run. The first run takes a moment while R downloads.

# the analytic file: 13,167 school-years, six surveys, 33 columns
base <- "https://school-connectivity.pages.dev/data/"
schools <- read.csv(paste0(base, "school_connectivity.csv"))

# every statistic is weighted with the survey weight; raw percentages are wrong
weighted.mean(schools$police, schools$w)      # 0.809
weighted.mean(schools$deg, schools$w)         # 4.00

# the paper's largest coefficient, rebuilt: the three standing arrangements as one index
z <- function(x) as.vector(scale(x))
schools$capacity <- z(z(schools$security) + z(schools$parent_process) + z(schools$equipment))
coef(lm(deg ~ capacity + z(crime_num) + factor(wave) + factor(level) + factor(urban),
        data = schools, weights = schools$w))["capacity"]
# +0.489. The paper reports +0.477 for the same index: it also holds school size, and
# uses both crime questions averaged rather than the one asked about the school's area.

# standard errors cluster on sw, the stratum crossed with the survey: 365 clusters
length(unique(schools$sw))                    # 365

# pandas
import pandas as pd
base = "https://school-connectivity.pages.dev/data/"
schools = pd.read_csv(base + "school_connectivity.csv")
edges   = pd.read_csv(base + "edges.csv")      # one row per school and organisation

# weighted, always
w = schools["w"]
(schools["police"] * w).sum() / w.sum()        # 0.809
(schools["deg"] * w).sum() / w.sum()           # 4.00

# the published coefficients, with their clustered standard errors and caveats
import json, urllib.request
json.load(urllib.request.urlopen(base + "coefficients.json"))

curl -s https://school-connectivity.pages.dev/data/school_connectivity.csv -o schools.csv
curl -s https://school-connectivity.pages.dev/data/tie_rates.json | jq .overall
curl -s https://school-connectivity.pages.dev/data/coefficients.json | jq .capacity_index

# everything the site serves, static, CORS open, no key
data/school_connectivity.csv    # the analytic file: 13,167 school-years, 33 columns
data/school_connectivity.json   # the same, as records
data/explorer.csv               # the lean subset this page parses
data/edges.csv                  # the two-mode structure: one row per school and organisation
data/codebook.json              # every column, its SSOCS variable, and the six coding traps
data/tables.json                # all 19 generated tables, rows and notes
data/coefficients.json          # the headline coefficients, each with its caveat
data/tie_rates.json             # weighted shares of the eight, overall and by survey

# READ codebook.json FIRST. Yes/no items are 1 and 2, the crime items are reversed,
# security staffing has a seam at 2015-16, and the cluster variable is sw, not strata.

# the source files are free from NCES; this site redistributes a subset
https://nces.ed.gov/surveys/ssocs/data_products.asp

# The working paper is under review and its authors are not listed here yet.
# Write to ron.levi@utoronto.ca for the current draft and how to cite it.
# To cite this site and its data:
@misc{school_connectivity_site,
  author = {{Community Safety and Policing Research Program, Global Justice Lab}},
  title  = {What predicts a school's ties to outside organisations},
  year   = {2026},
  note   = {University of Toronto},
  url    = {https://school-connectivity.pages.dev}
}

# and the source data
National Center for Education Statistics. School Survey on Crime and Safety (SSOCS),
public-use data files, 2005-06 through 2021-22. U.S. Department of Education.

What predicts how many organisations a school works with

Introduction

Schools work with outside organisations. On one side there are parents' groups, civic groups, local businesses and churches. On the other there are social services, mental-health providers, juvenile justice and the police. We study 13,167 US public schools, surveyed six times between 2005 and 2022, and ask what predicts how many of the eight a school works with.

Research on organisational partnership usually asks what predicts one relationship at a time: whether this actor deals with that one, whether this school works with the police. Metz, Leifeld and Ingold (2019) show what that misses. Actors do not decide on their relationships one at a time, so an explanation has to take the whole set at once. They model the set as a two-mode network, actors on one side and the things they connect to on the other; the line of work behind their model runs back through Bressers and O'Toole (1998), who set out why policy instruments are selected in sets, to Breiger's (1974) observation that one table of memberships defines the people by their groups and the groups by their people at the same time. Their case is thirty actors and fifteen policy instruments in the Swiss debate over micropollutants in water. We take the approach to a field of the opposite shape: thirteen thousand schools with no dealings with one another, answering the same eight questions six times over sixteen years.

The field turns out to have one dimension. Schools differ enormously in how many of the eight organisations they work with, far more than they would if each connection were decided separately. They differ hardly at all in which ones. There is no usual order of taking them on and no pair that schools treat as alternatives; the one structure that survives holding the margins fixed is faint, a pairing within the community and government sides, and it becomes part of the paper's apparatus. Once you know how many organisations a school works with, you can still barely guess which ones, and no measure in the survey improves that guess by much. The same holds in primary schools and high schools, in cities and in the countryside. So a school varies in a single respect: how far it reaches outside itself. Lin (2001) calls that extensity, the number of different kinds of position an actor can reach, and counts it as social capital in its own right. The paper asks what predicts it.

Two things do. The first is the school's own machinery: staffing, formal processes, equipment. Schools that do more inside work with more organisations outside, at +0.48 of an organisation per standard deviation, in every one of the six surveys and at every grade level. This is the larger source, and it cannot decide anything on its own, because a school doing well and a school that simply has more people and procedures to do the connecting would both produce it. The second source is the area as the principal describes it. Schools whose principals call the area high-crime work with more outside organisations, by about +0.08 once the machinery is accounted for. Here the theories can be told apart. Resource dependence predicts it: an organisation under pressure builds outside ties (Pfeffer and Salancik 1978). The social capital literature reads reach as accumulated advantage and predicts the reverse for actors in hard places: less reach, narrower access, the isolation Wilson (1987) described. For the people around these schools the data show the second. Where principals report more crime, parents turn up to school events considerably less often. For the institution the data show the first: it reaches further out. The people and the institution among them run in opposite directions, in the same places at the same time.

The environmental association is not policing under another name. Take the police out of the count and it gets stronger, not weaker. Take out juvenile justice as well and it holds. Count only the parents' groups, civic groups, businesses and churches, with no government organisation in the measure at all, and it is still there. In a rougher-seeming neighbourhood a school works with six of the eight organisations more often, most of all mental-health providers and social services. Ties to parents' groups fall; in those areas fewer parents turn up to anything, and once that turnout is taken into account the fall disappears.

The association is not the same in every survey. It is clear in three of the six, weaker in two, and absent in 2021-22, the year schools returned from the pandemic closures, so every pooled figure in this paper is an average over six years that differ. It shows no decline across the pre-pandemic surveys; the pandemic year is an interruption of the outcome, not a fading of the association. Over the sixteen years schools changed who they work with, not how much: the share working with juvenile justice agencies fell from 40 percent to 26, the share working with mental-health providers rose from 52 to 76, and the average number of organisations and its spread across schools ended where they began.

The scale is also what the two-mode tradition gains here. Earlier two-mode work asked why two actors resemble each other, and it answered from inside the field: shared beliefs, shared goals, working together. Thirty organisations in one consultation have no surroundings that differ between them, so the surroundings cannot explain anything about them. Thirteen thousand schools sit in thirteen thousand different places, and those surroundings predict how far each school reaches into its organisational field. The finding bears on an argument Sharkey has made about how communities respond to violence. Sharkey, Torrats-Espinosa and Takyar (2017) show that the growth of local nonprofit organisations, not policing, accounts for a large part of the American crime decline, and Sharkey (2018) argues that the organisational fabric of a place carries the response to violence. Their outcome is crime and their arrow runs from organisations to it. Ours runs the other way, and it is not causal: a school in a violent area holds more ties to the same kind of organisation Sharkey identifies, and no more to the police.

The abstract

We study how US public schools connect to eight kinds of outside organisation: parents' groups, civic groups, local businesses, churches, social services, mental-health providers, juvenile justice agencies, and the police. Across 13,167 schools surveyed six times between 2005 and 2022, schools differ mainly in how many of the eight they work with, and hardly at all in which ones; no types of school appear in any survey. Two things predict the number. The stronger one is the provision a school maintains in-house (security staff, a formal process for consulting parents, and security equipment), at about half an organisation per standard deviation. The second is the crime level the principal reports, for the school's area and, more strongly, for the students' neighbourhoods. That rating keeps predicting once the in-house provision is accounted for, weakening as that provision rises, and it is not a stand-in for recorded incidents, trouble inside the school, or the make-up of the student body. The association varies across the surveys but shows no decline before the pandemic; it is absent only in 2021-22, when the closures cut ties of every kind. The added partners span community and government alike. The exceptions are parents' groups, less common because fewer parents take part, and the police, no more common. Over sixteen years juvenile justice gave way to mental-health providers while the number and its spread held steady. Two theories answer the question. Social capital says organisations doing well end up with wide networks. Resource dependence says organisations under pressure build wide networks. The results fit the second: the crime rating predicts a school's connections, and the make-up of its student body does not. Every number here is an association: each survey measures everything at one moment, and no school is followed over time. The data cannot show what causes what.