NY City School Teacher Evaluation Algorithm Contested

An algorithm used to rate the effectiveness of school teachers in New York has resulted in thousands of disputes of its results.

Risk classification

• Primary risk domain: 7 AI system safety, failures, & limitations
• Primary risk subdomain: 7.3 Lack of capability or robustness

The VAM system failed to perform reliably, exhibiting massive margins of error and generating highly inaccurate ratings that led to unfair evaluations and potential job losses.

Additional risk subdomains

• 7.4 Lack of transparency or interpretability: The VAM model operated as an opaque 'statistical black box' that teachers and administrators could not easily understand or verify.
• 1.3 Unequal performance across groups: Teachers with higher numbers of students with disabilities or English-language learners received systematically lower ratings due to the model's bias.

Causal factors

• Entity: AI
• Intent: Unintentional
• Timing: Post-deployment

The VAM algorithm's flawed evaluations and high margins of error were unexpected outcomes of deploying the system to assess teacher performance.

EU AI Act risk tier

• Risk tier: 2 High Risk

High Risk: The VAM system is an educational and employment evaluation system used to assess worker performance and make decisions affecting employment access and retention.

AI system and alleged parties

• AI system: value-added analysis
• AI purpose: Workforce Monitoring and Evaluation; Automatic Skill Assessment
• Behaviour type: Assistant
• Alleged developer: New York city Dept. of Education
• Alleged deployer: New York city Dept. of Education
• Alleged harmed parties: Teachers

Harm severity

Severity ratings are based on this scale.

Highest direct severity in any category: Substantial. Severity is scored from Negligible to Catastrophic in each harm category, for harm the reports describe as caused directly or indirectly by the AI system.

• Physical: direct Negligible, indirect Negligible
• Infrastructure: direct Negligible, indirect Negligible
• Property: direct Negligible, indirect Negligible
• Financial: direct Minor, indirect Negligible
• Environmental: direct Negligible, indirect Negligible
• Malicious content: direct Negligible, indirect Negligible
• Differential treatment: direct Minor, indirect Negligible
• Civil rights: direct Minor, indirect Negligible
• Democracy: direct Negligible, indirect Negligible
• Privacy: direct Substantial, indirect Negligible
• Psychological: direct Minor, indirect Negligible
• Epistemic: direct Minor, indirect Minor
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Financial

Reported: Yes

Directly caused: While no specific dollar amounts are quantified, the ratings directly threatened teachers' livelihoods, as two consecutive 'ineffective' ratings could result in termination.

Inferred additional harm: Some teachers likely suffered actual loss of income or employment, and school districts incurred administrative and legal costs handling thousands of appeals and lawsuits.

Differential treatment

Reported: Yes

Directly caused: Teachers with more students with disabilities or English-language learners fared worse under the evaluation system, and teachers in non-tested subjects were unfairly judged by other teachers' students.

Inferred additional harm: The system likely created a strong disincentive for high-quality teachers to work in struggling schools or with high-needs student populations, worsening educational inequality.

Civil rights

Reported: Yes

Directly caused: Teachers' labor rights and due process were violated through unfair performance evaluations based on arbitrary and flawed algorithmic metrics.

Privacy

Reported: Yes

Directly caused: The individual Teacher Data Reports of over 12,000 teachers, containing sensitive performance evaluations, were publicly released to media organizations.

Psychological

Reported: Yes

Directly caused: The public release of individual Teacher Data Reports containing names and flawed ratings caused significant distress, anxiety, and public shaming for thousands of teachers.

Inferred additional harm: It is likely that many of the thousands of teachers rated 'ineffective' or 'developing' experienced severe professional anxiety and stress regarding potential job loss.

Epistemic

Reported: Yes

Directly caused: The VAM system generated highly inaccurate and misleading ratings, such as labeling an exceptional teacher as 'ineffective' based on flawed statistical predictions.

Indirectly caused: The public release of these flawed ratings in major newspapers misled parents and the public about the actual competence of thousands of teachers.

People affected

• Occurrences reported: 1
• People reportedly harmed: 12000
• People reportedly exposed: 13000

Potential causes

Management

• Prioritizing Metrics over Fairness: Administrators pushed VAM to meet federal grants without validating accuracy.
• Lack of Internal Data Verification: Management failed to verify class sizes and rosters before running models.

Technology

• Impossible Predicted Scores: VAM algorithm predicted scores higher than the maximum possible exam points.
• Extremely Large Margins of Error: Confidence intervals for VAM scores averaged 35 to 53 percentile points.
• Statistical Black Box Model: VAM used complex, opaque computer models that educators could not verify.

Data Inputs

• Inaccurate Student-Teacher Linkages: Teachers were evaluated based on students or subjects they did not teach.
• Inflated Baseline Test Data: The underlying state test scores were inflated, skewing VAM predictions.
• Failure to Verify Class Data: Omissions like unverified class sizes and assignments skewed VAM results.

Human Factors

• Over-reliance on Algorithmic Metrics: Reformers embraced VAM as a data-driven solution despite expert warnings.
• Demoralization of Excellent Teachers: High-performing teachers were labeled ineffective, causing stress and lawsuits.
• Disincentive for Struggling Schools: VAM metrics discouraged teachers from working in lower-performing schools.

Process and Methods

• Rushed VAM Implementation: The evaluation system was rushed alongside the new Common Core curriculum.
• Public Release of Flawed Data: Publishing raw VAM scores publicly misled parents and shamed teachers.
• Lack of Authentic Assessments: Over-reliance on multiple-choice tests instead of qualitative measures.

Regulatory Environment

• Federal Funding Pressures: Race to the Top grant requirements forced rapid adoption of VAM evaluations.
• State-Mandated Evaluation Laws: State laws mandated connecting teacher performance to student test scores.

Information quality

• Classification confidence: High
• Reason for confidence: The reports are detailed, from reputable sources (Politico, Washington Post, Huffington Post), and clearly describe the VAM system, its flaws, and the consequences.
• Ambiguities identified: The exact mathematical formula of the VAM is not fully detailed (only described as a complex computer model/statistical black box).

The implementation of flawed value-added modeling (VAM) algorithms for teacher evaluations in New York caused significant local labor disputes, privacy violations, and psychological distress. However, as a domestic educational administrative failure, it has negligible implications for national security.

• Overall national security impact: Minor
• Scope: Single nation
• Primary target: United States
• Alleged perpetrator: New York City and State education departments

Threat characteristics

• Imminence: Long-term. The incident represents a policy and administrative failure rather than an active national security crisis requiring urgent response.
• Autonomy: Human-supervised. The AI system acted as an assistant to human administrators who retained final authority over employment decisions and the appeals process.
• Novelty: Established threat. The use of flawed statistical models and algorithms for organizational performance evaluation is an established issue with existing precedents.

Impact by dimension

• Physical security: Negligible. No kinetic threats, physical damage, or critical infrastructure compromise resulted from this administrative educational algorithm.
• Information security: Negligible. No intelligence compromise, classified data theft, or foreign disinformation campaigns were associated with this incident.
• Sovereignty: Negligible. While the incident affected local public school administration, it did not threaten state authority, territorial control, or core national government functions.
• Economic security: Negligible. The incident involved local labor disputes and administrative costs, but posed no threat to strategic industries, financial systems, or national technological advantage.
• Societal stability: Minor. The public release of sensitive performance data for 12,000 teachers and flawed evaluations violated privacy and labor rights, causing localized distress but no threat to national stability.