Disorder, Social Control, and Opportunity

.title[
# Disorder, Social Control, and Opportunity
]
.subtitle[
## Advancing research on communities and crime
]
.author[
### Charles C. Lanfear Ross L. Matsueda & Lindsey R. Beach
]

---

# Today's Talk

The overview:

1. Community theories of crime: Broken windows and collective efficacy

2. Thinking about causality in communities research

3. Applying causal thinking: An embedded field experiment

4. Moving forward with communities research

---
class: inverse, cct

# Community Crime Theories

---

# Broken Windows

[Wilson & Kelling (1982)](https://www.theatlantic.com/magazine/archive/1982/03/broken-windows/304465/) describe a system where community disorder (1) increases crime by 
signalling low social control to offenders and (2) decreases actual social control by reducing willingness to use public spaces.
--

![Broken Windows](data:image/png;base64,#img/broken_windows_pretty.svg)

]

* B, E: Disorder and crime constrain social relations and impede control

* C, D: Social control inhibits both disorder and crime
]

---

## Keizer, Lindenberg, & Steg (2008)

Keizer et al. used multiple disorder manipulations and recorded norm violation with field experiments in Groningen, the Netherlands to test the direct micro-mechanism of broken windows.

![Keizer](data:image/png;base64,#img/keizer.svg)

]

* Disorder greatly increased all norm violation (A)

* Strong **internal validity** from experimental design

* Weak **external validity**: Single sites in one city, small-scale, minor norm violations

* Criticism: [Wicherts & Bakker (2014)](https://doi.org/10.1177/1368430213502); [Lanfear (2017)](http://hdl.handle.net/1773/40974)

]

---

# Collective Efficacy

[Sampson & Raudenbush (1999)](https://doi.org/10.1086/210356) contend that disorder and crime are both manifestations of low collective efficacy—the community's capacity for informal social control.
--

* C, D: Disorder and crime are products of low collective efficacy—social control—which is exogenous (B, E)
]

---

## Sampson & Raudenbush (1999)

Simultaneous equations of crime, disorder, and collective efficacy in Chicago neighborhoods, using survey-measured collective efficacy and objectives measure of disorder: Video of block faces coded by researchers.

]

* Control for crime impact on collective efficacy (E), but assume no disorder feedback (B)

* Strong **external validity**: *Actual* disorder, crime, neighbourhoods

* Weak **internal validity**: Cross-sectional, covariate adjustment
]

---

# Key Questions

* How do we reconcile these (and related) studies?

* How do we improve on them?

* Where should we go next?

---
class: inverse

# Causal Thinking

---

# Potential Outcomes

Causal effects are differences between *potential outcomes*

Fundamental problem: We only ever see *one* outcome

---

# Ignorability

We can compare different units if their differences are *ignorable*

We do this with *randomization* or *conditioning on variables*

* Most prefer randomization because communities are **complex**

---

# SUTVA1

* Only one treatment at each measured level
   * No *versions* of treatment
   * No *compound treatments*
* Assignment mechanism for treatment doesn't affect outcomes
]

.footnote[[1] [Stable Unit Value Treatment Assumption (Rubin 1986)](https://doi.org/10.1080/01621459.1986.10478355)]

---

# SUTVA1

* Only one treatment at each measured level
   * No *versions* of treatment
   * No *compound treatments*
* Assignment mechanism for treatment doesn't affect outcomes

### No interference

* A unit's outcomes are not affected by the treatment assignment of other units
]

.footnote[[1] [Stable Unit Value Treatment Assumption (Rubin 1986)](https://doi.org/10.1080/01621459.1986.10478355)]

---

# Consistency

Violated when different types of treatment are assumed to be the same

Commonly violated in experiments *and* observational research

* e.g., scales combining different forms of physical or social disorder1

---

## Consistency: Asymmetry

Occurs when increasing and decreasing treatment has different effects

Nearly all studies assume symmetry—but reality is often asymmetric1

* *Pulling the knife out doesn't heal the wound*
* Control conditions *may be a treatment* for some units

---

# Interference

Interference means some units' treatments affect other units' outcomes

These spillovers are common in theory and the real world

* Participants influence each other and neighbourhood borders are porous
* Micro-macro mechanisms, e.g. signalling, norms, shared meanings

---

# Interference

Interference gets complicated quickly

Can often be ruled out in experiments

* But this eliminates important, real mechanisms—even primary ones
* Experimental results rarely scale linearly to large-scale policies or natural changes ([Nagin & Sampson 2019](https://www.annualreviews.org/doi/abs/10.1146/annurev-criminol-011518-024838))

---

# The Gist

**Experimental Methods**

* Good at isolating direct causal effects (internal validity)
   * Some things cannot be manipulated
* Bad at measuring ecological causal processes (external validity)
   * Manipulations may also lack fidelity

**Observational Methods**

* Bad at isolating any mechanisms and ruling out confounding
* Wider range of "treatments" (e.g., real disorder) and outcomes (e.g., homicide)
* *Actually observing real ecological processes at work*

---
class: inverse, sncp

# The Seattle Neighborhoods and Crime Project

---

## The Experiments

Over 2000 person-hours of **field experiments**, conducted over four years, embedded in representative Seattle neighbourhoods with rich **ecological survey data** (2002-2003 SNCS)

* Balance strengths and weaknesses of experimental and observational methods

* Maximize internal and external validity

&zwj;1. **Mailbox Experiment** (today)

* Effect of disorder on crime and prosocial behavior (replicate Keizer)
* Association between collective efficacy and crime and prosocial behavior

&zwj;2. **Lost Letter**: Collective efficacy &#8594; prosocial behaviour

&zwj;3. **Littering Intervention**: Disorder &#8594; sanctioning

&zwj;4. **Litter Clean-Up**: Collective efficacy &#8594; disorder removal

---

![Experiments](data:image/png;base64,#img/experiments_pretty.svg)

---

# Number of Trials

---

---

---

# Mailbox: Method

* A letter containing a visible $5 bill is left near mailbox for passersby to encounter

---

* A letter containing a visible $5 bill is left near mailbox for passersby to encounter

* Litter and a sign board with graffiti are introduced to manipulate disorder

* Observe whether participants mail, steal, or ignore envelope

* Record gender, age, race, and grouping of participants to adjust for differences between and interference within trials

* Record if disorder present *prior to trial* to test asymmetry

* 368 trials with 3,481 participants in 20 Seattle census tracts

---

# Mailbox: Model

Objective: Causal test of direct effect of disorder on crime (A).

---

# Mailbox: Model

Our Experiment: Causal tests of direct effects of disorder on crime and prosocial behavior—and associations with collective efficacy

---

# Mailbox: Results

<table style="width:70%">
 <tr>
 <td style="font-size:115%" colspan="5">Mailbox: Participant Actions</td>
 </tr>
 <tr>
 <th>Condition</th>
 <th>Walk-By</th>
 <th>Mail</th>
 <th>Theft</th>
 </tr>
 <tr>
 <td rowspan="2">Control</td>
 <td>1496</td>
 <td>176</td>
 <td>28</td>
 </tr>
 <tr>
 <td>88.0%</td>
 <td>10.4%</td>
 <td>1.6%</td>
 </tr>
 <tr>
 <td rowspan="2">Treatment</td>
 <td>1617</td>
 <td>136</td>
 <td>28</td>
 </tr>
 <tr>
 <td>90.8%</td>
 <td>7.6%</td>
 <td>1.6%</td>
 </tr>
</table>

* Disorder has **no impact on theft**—Keizer et al. (2008) fails replication

* Disorder **reduces mailing**

* Collective efficacy predicts less **theft** and **prior disorder**

* Consistent with ecological studies and suggests external validity
--

* **Prior disorder** predicts thefts, but no difference in treatment effect

* Association persists once disorder removed—not local disorder effect

---

# Other Experiment Results

* Collective efficacy strongly predicts lost letter returns

* Contextual effects (letter addressee) exert strong effects
   * i.e., people don't like nazis

* Collective efficacy doesn't predict direct sanctioning of litterers

* But race and gender of litterer does
   * Sanctions rare, even in efficacious places

* Collectively efficacious neighbourhoods really hate my litter cameras

* Suggests large differences in clean-up speed

* Mailbox experiment replicated in Oxford in Summer 2022
   
   * Zero treatment effect on theft
   * High walk-by rate, likely due to foot traffic

---
class: inverse, littercam

# Moving Forward

---

# Ongoing Projects

* [Lanfear (2022) "Collective efficacy and the built environment." *Criminology, 60*(2), 370–396](https://github.com/clanfear/ccl_cv/raw/master/articles/Collective%20Efficacy%20and%20the%20Built%20Environment.pdf)

* Urban political economy, social control, and situational opportunity
   * Modeling long-term, large-scale, endogenous community processes

* Lanfear & Matsueda (Ongoing) "A micro theory of crime opportunities:
Symbolic interaction among motivated offenders, suitable targets, and capable
guardians"

* A micro-foundation linking informal control and opportunity theories
   * Generating testable multilevel propositions

* Lanfear (Under revision) "Collective efficacy and formal social control"

* Relationship between norms of intervention and police effectiveness
   * Role of collective efficacy in policing and vice versa

---

## New Empirical Directions

**Survey Research**

* Continued large-scale ecological surveys: e.g., PHDCN-CS 3
* Focus on:
   * Repeated measures, maximizing variation
   * New measures: mobility, sources of perceptions, collective action

**Embedded Experiments**

* More micro-mechanism tests: e.g. exploring asymmetry
* Environmental interventions: e.g., work by MacDonald & colleagues
* Policing interventions: e.g., [Weisburd et al. 2021](https://doi.org/10.1007/s11292-019-09401-1)
* *Manipulating collective efficacy / social control*

**Embedded Qualitative Research**

* Testing assumptions and observing mechanisms
* The original model: Shaw & McKay (1942)
* A new model: [St. Jean (2007)](https://press.uchicago.edu/ucp/books/book/chicago/P/bo5298963.html)

---

## New Theoretical Directions

**Analytical and Multi-Level Focus**

* Ecological theory mechanisms are often ill-specified and untested
   * e.g., *How does collective efficacy work?*
* Micro-macro and macro-micro transitions
   * The situation: Selection, emergence, learning, communication
* [Matsueda (2017)](https://doi.org/10.1111/1745-9125.12149); [Wikström & Kroneberg (2022)](https://doi.org/10.1146/annurev-criminol-030920-091320)

**Increasing Policy Relevance**

* Attention to social processes that will occur in application
   * e.g., communication, spillovers, cascades
* Attention to complex systemic relationships with urban political economy, neighbourhood change, and policing
   * e.g., investment, residential mobility, police demand and perceptions
* Addressing failures of community interventions (e.g., [Linning et al. 2022](https://doi.org/10.1186/s40163-022-00167-y))

---
class: inverse

# Acknowledgements and Q&A

Our experimental findings are the result of over 2000 person-hours of field work over four summers. This was made possible by the participation of graduate student researchers, grant funding, and UW institutional resources.

Anquinette Barry

Chris Hess

John Leverso

Kate O'Neill

Terrence Pope
]

National Science Foundation

UW Royalty Research Fund

### Support

UW Center for Statistics and the Social Sciences

UW Center for Studies in Demography and Ecology
]

---
count: false
class: inverse

# Appendix

---

count:false
# Mailbox Models

|                    |  Mail    |  Theft   |     Mail  |   Theft  | 
|:------------------:|:--------:|:--------:|:---------:|:--------:|
|  (Intercept)       | **-2.053** | 0.332    | **-1.467**  | -1.437   | 
|                    | *(0.538)*  | *(1.170)*  |  *(0.579)*  | *(1.329)*  | 
|  Treatment       | **-0.275** | -0.081   |**-0.263**   | -0.098   | 
|                    | *(0.118)*  | *(0.261)*  |  *(0.119)*  | *(0.263)*  | 
|  Coll. Eff.          |  0.265   | **-1.206** |   0.171   | -0.804   | 
|                    | *(0.173)*  | *(0.396)*  |  *(0.181)*  | *(0.424)*  |
|  Walk-by Rate   | **-0.564** | -**0.535** |  **-0.534** | **-0.429** | 
|                    | *(0.063)*  | *(0.132)*  |  *(0.064)*  | *(0.133)*  |
|  Male             |          |          |  -0.094   | **0.880**  | 
|                    |          |          | *(0.117)*   | *(0.296)*  |
|  Non-White         |          |          |  **-0.459** | 0.397    |
|                    |          |          | *(0.144)*   | *(0.272)*  |
|  In Group            |          |          |  **-0.876** | **-1.302** |
|                    |          |          | *(0.161)*   | *(0.475)*  |

---
count:false
class:inverse

# The Lost Letter

---
count:false

# Lost Letter: Method

Replication of [Milgram et al. (1965)](https://doi.org/10.1086/267344) and [Sampson (2012)](https://www.press.uchicago.edu/ucp/books/book/chicago/G/bo5514383.html).

2,938 letters were dropped on Seattle sidewalks for passersby to encounter.1

.footnote[[1] Drops conducted on foot in center city and by car in most tracts—at the cost of one car accident. ]

24 letters—8 per addressee—were dropped in each census tract:
  * Charles F. Landers Sr. (Personal)
  * Friends of Black Lives Matter (BLM)
  * American Neo-Nazi Party (Nazi)

Mailed letters were received at a PO Box monitored by the research team.

???

We still have every single returned letter in storage.

Letters were geocoded to drop locations and featured a small printed ID number for tracking.

---
count:false

# Lost Letter: Model

.pull-right[
* Returning a lost letter is a low-risk pro-social intervention highly correlated with collective efficacy (G)1.
]

.footnote[
[1] See [Sampson (2012)](https://www.press.uchicago.edu/ucp/books/book/chicago/G/bo5514383.html) and [Milgram, Mann, and Harter (1965)](https://doi.org/10.1086/267344).
 
 
 
]

---
count:false

# Lost Letter: Model

.pull-right[
* Returning a lost letter is a low-risk pro-social intervention highly correlated with collective efficacy (G)1.

* This may be used to update the 2002-2003 SNCS collective efficacy measures which may have shifted over time.2
]

.footnote[
[1] See [Sampson (2012)](https://www.press.uchicago.edu/ucp/books/book/chicago/G/bo5514383.html) and [Milgram, Mann, and Harter (1965)](https://doi.org/10.1086/267344).

[2] Note however that collective efficacy is persistent ([Sampson 2012](https://www.press.uchicago.edu/ucp/books/book/chicago/G/bo5514383.html)).
]

---
count:false

# Lost Letter: Results

| *Study* (Letter Address) | *Return Rate* | `$\rho(CE)$` 
|:--------------------------------|--------|------------|
|*Milgram et al. (1965)*1 | | |
| &nbsp;&nbsp; Medical Research Associates | 70% | -- |
| &nbsp;&nbsp; Friends of the Community Party | 70% | -- |
| &nbsp;&nbsp; Friends of the Nazi Party | 25% | -- |
|*Sampson (2012)*2 | 33% | .41 |
|*SNCP*3 | | |
| &nbsp;&nbsp; Charles F. Landers Sr. | 79% | .35 |
| &nbsp;&nbsp; Friends of Black Lives Matter | 71% | .32 |
| &nbsp;&nbsp; American Neo-Nazi Party | 24% | .03 |

* Our results closely mimic Milgram et al. (1965).
* Collective efficacy does not predict Neo-Nazi letter returns.

---
count:false

## .blue[Mailed] and .rust[Unmailed] Letters

---
count:false

# Updating Collective Efficacy

* Sampson (2012) found `$A = .73$` over 7 years.
* Assume `$A = .50$` over 12 years.
* We find `$\rho(Z_1Y) = .35$` for personal letters.
]

---
count:false

# Updating Collective Efficacy

* Sampson (2012) found `$A = .73$` over 7 years.
* Assume `$A = .50$` over 12 years.
* We find `$\rho(Z_1Y) = .35$` for personal letters.

Extrapolation:

* `$\rho(Z_1Y) = A*B$`
* `$.35 = .50 * B$`
* `$B = .70$`
]

Under some assumptions,

* Collective efficacy is highly correlated with letter returns.
* We have an updated measure of collective efficacy.

---
count:false

---
count:false
class: inverse

# The Litter Clean-Up

---
count:false

# Litter Clean-Up: Method

]

---
count:false

# Litter Clean-Up: Method

* A concealed camera and motion detector record when the litter is cleaned up and how many people pass by.
]

---
count:false

# Litter Clean-Up: Method

* A concealed camera and motion detector record when the litter is cleaned up and how many people pass by.
]

---
count:false

# Litter Clean-Up: Method

* A concealed camera and motion detector record when the litter is cleaned up and how many people pass by.
   * This permits capturing the relationship between collective efficacy and time to remove disorder.

]

---
count:false

# Litter Clean-Up: Method

* An alternate magazine rack defaced by graffiti and stickers is swapped in to manipulate disorder.
]

---
count:false

# Litter Clean-Up: Results

* The experiment was abandoned after a small number of trials.

---
count:false

# Litter Clean-Up: Results

* The experiment was abandoned after a small number of trials.

* In Ravenna, residents reported the magazine racks on Nextdoor as an "eyesore".

???

Ravenna, where I was living at the time, is a high collective efficacy, fairly affluent, residential neighborhood.

* Two magazine racks had locks cut and were removed.

* Thus we were unable to continue experiment *because of collective efficacy* in an affluent neighborhood.

* In contrast, one rack was left *for an entire year* in Capitol Hill without issue.

---
count:false

# Litter Clean-Up: Alternative

We supposed the speed of return of lost letters was correlated with clean-up time of trash.

As a pilot, in selected tracts, 24 pieces of litter were dropped in the same locations as lost letters and then checked three times over the course of a day.

Neither timing nor a binary measure of clean-up were found to be associated with either lost letter returns or overall collective efficacy.

---
count:false
class: inverse

# The Littering Intervention

---
count:false

## Littering Intervention: Method

A confederate walks down the street and flagrantly litters in view of passersby near a trash can with a no-littering sign.1

Researchers record if confederate is sanctioned—**risky prosocial behavior**—or litter disposed of by participant—**low-risk prosocial behavior**. *Content* of sanction is also recorded.

Litter and a sign board with graffiti are introduced to manipulate disorder.

Replicated with three confederates of varying gender and race to manipulate interpersonal context which may influence evaluations of risk:2
 * Black Man
 * White Man
 * White Woman

We conducted 576 trials in 12 Seattle census tracts.

???

8 trials per person per condition in 12 tracts. 8 x 2 x 3 x 12

---
count: false

# Littering Intervention: Model

---
count: false

# Littering Intervention: Model

* Norm-Violation is *fixed*—all relationships are conditional on the offense occurring.
]

---
count:false

# Littering Intervention: Model

* Norm-Violation is *fixed*—all relationships are conditional on the offense occurring.

* Test effect of interpersonal context of norm violation on sanctioning (L).
]

---
count:false

# Littering Intervention: Model

* Norm-Violation is *fixed*—all relationships are conditional on the offense occurring.

* Test effect of interpersonal context of norm violation on sanctioning (L).

* Captures association between collective efficacy and sanctioning (M).
]

---
count:false

# Results: Sanctions

<table style="width:70%">
 <tr>
 <td style="font-size:115%" colspan="5">Litter Intervention: Sanctions</td>
 </tr>
 <tr>
 <th> Condition</th>
 <th>Black Man</th>
 <th>White Man</th>
 <th>White Woman</th>
 <th> All</th>
 </tr>
 <tr>
 <td rowspan="2">Control</td>
 <td>2</td>
 <td>2</td>
 <td>9</td>
 <td>13</td> 
 </tr>
 <tr>
 <td>.021</td>
 <td>.021</td>
 <td>.103</td>
 <td>.045</td>
 </tr>
 <tr>
 <td rowspan="2">Treatment</td>
 <td>1</td>
 <td>8</td>
 <td>4</td>
 <td>13</td> 
 </tr>
 <tr>
 <td>.011</td>
 <td>.091</td>
 <td>.043</td>
 <td>.045</td>
 </tr>
 <tr>
 <td rowspan="2">Both</td>
 <td>3</td>
 <td>10</td>
 <td>13</td>
 <td>26</td>
 </tr>
 <tr>
 <td>.016</td>
 <td>.052</td>
 <td>.068</td>
 <td>.045</td>
 </tr>
</table>

Key Findings:

* No *unconditional* effect of disorder on sanctions—a risky prosocial action.
   * Disorder *interacts* with confederate race/gender.

--
* Collective efficacy and letter returns did not predict sanctioning (M).

--
* Sanctioning varies widely by confederate in both *quantity* and *content*.
   * White woman: *Confrontational* sanctions usually from men.
   * Black man: *Non-confrontational* sanctions only from women.
   * White man: *Non-confrontational* sanctions, men and women.

???

One of the big takeaways here is that sanctioning for littering is just really quite rare.

As a result, our test here is underpowered.

LB: 11/13 from men
TP: 3 sanctions from women
CL; 6/10 sanctions from men

---
count:false

# Results: Throwing Away

<table style="width:70%">
 <tr>
 <td style="font-size:115%" colspan="5">Litter Intervention: Throw-Aways</td>
 </tr>
 <tr>
 <th> Condition</th>
 <th>Black Male</th>
 <th>White Male</th>
 <th>White Female</th>
 <th> All</th>
 </tr>
 <tr>
 <td rowspan="2">Control</td>
 <td>3</td>
 <td>7</td>
 <td>7</td>
 <td>17</td> 
 </tr>
 <tr>
 <td>.032</td>
 <td>.079</td>
 <td>.079</td>
 <td>.059</td>
 </tr>
 <tr>
 <td rowspan="2">Treatment</td>
 <td>3</td>
 <td>3</td>
 <td>4</td>
 <td>10</td> 
 </tr>
 <tr>
 <td>.032</td>
 <td>.032</td>
 <td>.043</td>
 <td>.035</td>
 </tr>
 <tr>
 <td rowspan="2">Both</td>
 <td>6</td>
 <td>10</td>
 <td>11</td>
 <td>27</td>
 </tr>
 <tr>
 <td>.032</td>
 <td>.052</td>
 <td>.057</td>
 <td>.047</td>
 </tr>
</table>

We also evaluated if our interventions impacted throwing away the dropped litter—a low risk, non-confrontational prosocial action.

Key Findings:

* Disorder treatment reduces throwing away.

* This effect also conditional on confederate characteristics.
--

* Collective efficacy and letter returns do not predict throwing away.