ECC3841 Chap.7 Education and Peer Identification
Education and Peer Identification
Education networks illustrate the identification problem behind peer effects. Students outcomes can co-move because they influence one another, have peers with similar characteristics or share teachers and environments. Friendship selection and simultaneous influence create further threats. Random assignment, fixed effects, timing and instruments address different rivals and require explicit assumptions.
Network position can shape exposure, but measurement must precede the outcome for a clean interpretation.
What this chapter covers
- 01
Endogenous achievement effects
- 02
Contextual peer characteristics
- 03
Shared school and teacher inputs
- 04
Reflection and simultaneity
- 05
Friendship selection
- 06
Random peer assignment
- 07
Fixed effects and instruments
- 08
Network position and heterogeneous exposure
Worked example · free
Interpret a randomized-peer coefficient
- +1Multiply 0.15 by twelve to obtain 1.8 points.
- +1State the outcome and peer-score units.
- +1Use random assignment to explain independence from focal motivation in expectation.
- +1Limit the claim to the assigned group and design.
Key terms
- Reflection problem
- Simultaneous influence between an individual and peers that obscures causal direction.
- Peer selection
- Nonrandom formation of ties among people with related observed or unobserved traits.
- Leave-one-out mean
- A peer average calculated without the focal individuals own value.
- Exclusion restriction
- The assumption that an instrument affects the outcome only through the endogenous peer exposure.
- Network fixed effect
- A control absorbing factors shared by all members of the same network.
- Assigned peer
- A classmate or group member allocated by a rule outside the focal students choice. Assignment can reduce selection when it is implemented and retained, but cross-group interaction can still contaminate exposure.
- Pre-treatment network
- A set of ties measured before the intervention or outcome begins. It preserves temporal order for centrality and exposure, though earlier friendship formation may still reflect unobserved similarity.
- Network boundary
- The rule determining which people and relationships count as peers. Classroom, friendship and neighborhood boundaries represent different mechanisms and can produce different estimates from the same students.
- Spillover exposure
- The treatment or behavior reaching an untreated person through network connections. Measuring it requires the treatment assignment, the link definition and a rule for combining multiple treated neighbors.
- Peer composition
- The distribution of peers characteristics or prior outcomes around an individual. Composition can generate contextual effects even when peers current choices do not causally influence one another.
- Outcome simultaneity
- A condition where focal and peer outcomes are determined together. It creates the reflection problem and prevents a contemporaneous correlation from revealing which direction influence runs.
- Treatment contamination
- Exposure received outside the assigned group or condition. It weakens the intended comparison and can hide network spillovers unless external ties and actual interactions are measured.
- Practical magnitude
- The size of an estimated effect relative to an educational benchmark or policy cost. Statistical precision does not establish that the consequence is large enough to justify intervention.
- Identification variation
- The part of peer exposure used to estimate an effect after controls and design restrictions are applied. A credible answer states where this variation comes from and why it is independent of unobserved outcome determinants.
- Attrition bias
- Distortion created when students leave the sample or group in a way related to peer exposure and outcomes. Initial random assignment does not protect the final comparison when attrition differs systematically.
Education and Peer Identification FAQ
Why not regress grade on class average?
The class average can reflect the students own grade, simultaneous responses, sorting and shared teacher quality. A leave-one-out measure and a credible design are needed. State the causal threat, design assumption, coefficient units and remaining boundary in the same answer.
What does random assignment solve?
It can make assigned peer composition independent of unobserved student traits in expectation, reducing selection. Compliance and cross-group spillovers still need attention. State the causal threat, design assumption, coefficient units and remaining boundary in the same answer.
Can fixed effects solve reflection?
Fixed effects absorb shared stable factors but do not automatically resolve simultaneous outcome determination. State the causal threat, design assumption, coefficient units and remaining boundary in the same answer.
Why does friendship direction matter?
A nominated friend may influence differently from someone who names the focal student. Direction changes the exposure measure and mechanism. State the causal threat, design assumption, coefficient units and remaining boundary in the same answer.
Exam move
Organize every empirical answer under five labels: threat, design, assumption, estimate and limitation. Begin with the network boundary. A classroom average, nominated friendship network and residential neighborhood define different peers and different mechanisms. State whether links are directed, when they were measured and how isolated students are treated. Construct peer exposure without the focal students own outcome.
A leave-one-out mean removes mechanical overlap but does not solve reflection, selection or common inputs. List endogenous effects, contextual effects and correlated effects before choosing a method. Reflection arises because my outcome affects peers while peers affect me. Selection arises because similar students form ties or choose schools. Teachers, neighborhoods and resources create shared shocks.
Random assignment can reduce selection if compliance holds; fixed effects absorb specified shared factors; instruments require a defended exclusion restriction; earlier network measurement improves timing but does not automatically make ties random. For any design, say exactly which variation identifies the coefficient. Interpret the estimate with units.
Multiply the coefficient by a meaningful peer change, state the outcome scale and restrict the claim to the studied population and network. Separate precision from magnitude: a narrow interval can surround a trivial effect, while an imprecise estimate may not support policy despite a large point value. Compare specifications by the variation that remains, not by whether a coefficient keeps the same sign.
If school effects shrink an estimate, explain whether confounding was removed or between-school peer variation was discarded. Check spillovers outside assigned groups, differential attrition and network changes caused by treatment. Position-based heterogeneity needs pre-treatment centrality; post-treatment centrality can be an outcome itself.
Build one revision table in which each row names an identification threat and a design response. Then practise a short table comment that describes a coefficient, questions causal interpretation and proposes one credible robustness check. In policy answers, explain how grouping or targeting might change the network, not just individual exposure. Preserve uncertainty about generalization and distributional effects.
The final sentence should say what evidence would change your conclusion. This structure earns marks because it makes the causal logic visible rather than burying it inside a literature summary.
Working through Education and Peer Identification in ECC3841? Sia is AskSia’s AI Economics tutor — ask any ECC3841 Education and Peer Identification question and get a clear, step-by-step explanation grounded in how ECC3841 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.
