# Demographic Bias Assessment on SubPOP Source: [Demographic Bias Assessment on SubPOP]() Collection index: [llms.txt]() Data updated: 2026-06-15T07:18:28.742Z Evaluation of model performance regarding demographic bias across various dimensions (geographic, socioeconomic, political, religious) using the SubPOP dataset. A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. | Scope | Recorded value | | --- | --- | | Task | Demographic Bias Assessment | | Dataset (including any stated split) | SubPOP | | Metric | Error Rate (Northeast) | | Unit | SCALAR | | Direction | Lower is better | | Results loaded | 12 of 12 | | Coverage | All results currently returned for this benchmark | Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol. ## Metrics - [Error Rate (Northeast)](): Lower is better; unit: SCALAR. - [Average Weighted Difference](): Lower is better; unit: SCALAR. - [Error Rate (South)](): Lower is better; unit: SCALAR. - [Error Rate (College Grad)](): Lower is better; unit: SCALAR. - [Error Rate (\< High School Education)](): Lower is better; unit: SCALAR. - [Error Rate (Male)](): Lower is better; unit: SCALAR. - [Error Rate (Female)](): Lower is better; unit: SCALAR. - [Error Rate (Black)](): Lower is better; unit: SCALAR. - [Error Rate (White)](): Lower is better; unit: SCALAR. - [Error Rate (Asian)](): Lower is better; unit: SCALAR. - [Error Rate (Hispanic)](): Lower is better; unit: SCALAR. - [Error Rate ($100k+ Income)](): Lower is better; unit: SCALAR. - [Error Rate (\<$30k Income)](): Lower is better; unit: SCALAR. - [Error Rate (Democrat)](): Lower is better; unit: SCALAR. - [Error Rate (Republican)](): Lower is better; unit: SCALAR. - [Error Rate (Liberal)](): Lower is better; unit: SCALAR. - [Error Rate (Conservative)](): Lower is better; unit: SCALAR. - [Error Rate (Moderate)](): Lower is better; unit: SCALAR. - [Error Rate (Protestant)](): Lower is better; unit: SCALAR. - [Error Rate (Jewish)](): Lower is better; unit: SCALAR. - [Error Rate (Hindu)](): Lower is better; unit: SCALAR. - [Error Rate (Atheist)](): Lower is better; unit: SCALAR. - [Error Rate (Muslim)](): Lower is better; unit: SCALAR. ## Recorded results | Method | Error Rate (Northeast) | Result date | Report paper | | --- | --- | --- | --- | | Human Baseline | 0.027 | Not recorded | 6a2f86ded33dfb23e783be04 | | Llama-3-70B | 0.078 | Not recorded | 6a2f86ded33dfb23e783be04 | | Graph-based Target Back-Propagation | 0.0952 | Not recorded | 6a2f86ded33dfb23e783be04 | | GPT-4o-mini | 0.0982 | Not recorded | 6a2f86ded33dfb23e783be04 | | GPT-4o-mini | 0.0991 | Not recorded | 6a2f86ded33dfb23e783be04 | | Llama-2-13B | 0.103 | Not recorded | 6a2f86ded33dfb23e783be04 | | Mistral-7B-v0.1 | 0.108 | Not recorded | 6a2f86ded33dfb23e783be04 | | Llama-2-7B | 0.113 | Not recorded | 6a2f86ded33dfb23e783be04 | | Llama-3-70B | 0.156 | Not recorded | 6a2f86ded33dfb23e783be04 | | Mistral-7B-v0.1 | 0.185 | Not recorded | 6a2f86ded33dfb23e783be04 | | Llama-2-13B | 0.193 | Not recorded | 6a2f86ded33dfb23e783be04 | | Llama-2-7B | 0.196 | Not recorded | 6a2f86ded33dfb23e783be04 | ### Human Baseline Error Rate (Northeast): 0.027. Result ID: 6a2fa6d1fbc373592c2b83e7. [Reporting paper]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### Llama-3-70B Error Rate (Northeast): 0.078. Result ID: 6a2fa6d1fbc373592c2b83f2. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Configuration":"Fine-tuned"}. This record does not assert independent reproduction. ### Graph-based Target Back-Propagation Error Rate (Northeast): 0.0952. Result ID: 6a2fa6d1fbc373592c2b83f0. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Configuration":"GTBP","Backbone":"GPT-4o-mini"}. This record does not assert independent reproduction. ### GPT-4o-mini Error Rate (Northeast): 0.0982. Result ID: 6a2fa6d1fbc373592c2b83ef. [Reporting paper]() Recorded evaluation configuration: {"Configuration":"GEPA"}. This record does not assert independent reproduction. ### GPT-4o-mini Error Rate (Northeast): 0.0991. Result ID: 6a2fa6d1fbc373592c2b83ee. [Reporting paper]() Recorded evaluation configuration: {"Configuration":"Base"}. This record does not assert independent reproduction. ### Llama-2-13B Error Rate (Northeast): 0.103. Result ID: 6a2fa6d1fbc373592c2b83eb. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Configuration":"Fine-tuned"}. This record does not assert independent reproduction. ### Mistral-7B-v0.1 Error Rate (Northeast): 0.108. Result ID: 6a2fa6d1fbc373592c2b83ed. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Configuration":"Fine-tuned"}. This record does not assert independent reproduction. ### Llama-2-7B Error Rate (Northeast): 0.113. Result ID: 6a2fa6d1fbc373592c2b83e9. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Configuration":"Fine-tuned"}. This record does not assert independent reproduction. ### Llama-3-70B Error Rate (Northeast): 0.156. Result ID: 6a2fa6d1fbc373592c2b83f1. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Configuration":"Base"}. This record does not assert independent reproduction. ### Mistral-7B-v0.1 Error Rate (Northeast): 0.185. Result ID: 6a2fa6d1fbc373592c2b83ec. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Configuration":"Base"}. This record does not assert independent reproduction. ### Llama-2-13B Error Rate (Northeast): 0.193. Result ID: 6a2fa6d1fbc373592c2b83ea. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Configuration":"Base"}. This record does not assert independent reproduction. ### Llama-2-7B Error Rate (Northeast): 0.196. Result ID: 6a2fa6d1fbc373592c2b83e8. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Configuration":"Base"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.