# Data Documentation This document describes the data sources and variables used in the fourth Anthropic Economic Index (AEI) report. ## Claude.ai Usage Data ### Overview The core dataset contains Claude.ai usage metrics aggregated by geography and analysis dimensions (facets). **Source files**: - `aei_raw_claude_ai_2025-11-13_to_2025-11-20.csv` (pre-enrichment data in data/intermediate/) **Note on data sources**: The AEI raw file contains raw counts and percentages. ### Data Schema Each row represents one metric value for a specific geography and facet combination: | Column | Type | Description | |--------|------|-------------| | `geo_id` | string | Geographic identifier (ISO-3166-1 country code for countries, ISO 3166-2 region code for country-state, or "GLOBAL"). Examples: "USA", "AGO-LUA" (Angola-Luanda), "ALB-02" (Albania-Fier) (raw version uses 2- instead of 3-letter country codes) | | `geography` | string | Geographic level: "country", "country-state", or "global" | | `date_start` | date | Start of data collection period | | `date_end` | date | End of data collection period | | `platform_and_product` | string | "Claude AI (Free and Pro)" | | `facet` | string | Analysis dimension (see Facets below) | | `level` | integer | Sub-level within facet (0-2) | | `variable` | string | Metric name (see Variables below) | | `cluster_name` | string | Specific entity within facet (task, pattern, etc.). For intersections, format is "base::category" | | `value` | float | Numeric metric value | ### Facets **Geographic Facets:** - **country**: Country-level aggregations - **country-state**: Subnational region aggregations (ISO 3166-2 regions globally) **Content Facets:** - **onet_task**: O*NET occupational tasks - **collaboration**: Human-AI collaboration patterns - **request**: Request complexity levels (0=highest granularity, 1=middle granularity, 2=lowest granularity) - **multitasking**: Whether conversation involves single or multiple tasks - **human_only_ability**: Whether a human could complete the task without AI assistance - **use_case**: Use case categories (work, coursework, personal) - **task_success**: Whether the task was successfully completed **Numeric Facets** (continuous variables with distribution statistics): - **human_only_time**: Estimated time for a human to complete the task without AI - **human_with_ai_time**: Estimated time for a human to complete the task with AI assistance - **ai_autonomy**: Degree of AI autonomy in task completion - **human_education_years**: Estimated years of human education required for the task - **ai_education_years**: Estimated equivalent years of AI "education" demonstrated **Intersection Facets:** - **onet_task::collaboration**: Intersection of O*NET tasks and collaboration patterns - **onet_task::multitasking**: Intersection of O*NET tasks and multitasking status - **onet_task::human_only_ability**: Intersection of O*NET tasks and human-only ability - **onet_task::use_case**: Intersection of O*NET tasks and use case categories - **onet_task::task_success**: Intersection of O*NET tasks and task success - **onet_task::human_only_time**: Mean human-only time per O*NET task - **onet_task::human_with_ai_time**: Mean human-with-AI time per O*NET task - **onet_task::ai_autonomy**: Mean AI autonomy per O*NET task - **onet_task::human_education_years**: Mean human education years per O*NET task - **onet_task::ai_education_years**: Mean AI education years per O*NET task - **request::collaboration**: Intersection of request categories and collaboration patterns - **request::multitasking**: Intersection of request categories and multitasking status - **request::human_only_ability**: Intersection of request categories and human-only ability - **request::use_case**: Intersection of request categories and use case categories - **request::task_success**: Intersection of request categories and task success - **request::human_only_time**: Mean human-only time per request category - **request::human_with_ai_time**: Mean human-with-AI time per request category - **request::ai_autonomy**: Mean AI autonomy per request category - **request::human_education_years**: Mean human education years per request category - **request::ai_education_years**: Mean AI education years per request category ### Core Variables Variables follow the pattern `{prefix}_{suffix}` with specific meanings: **From AEI raw file**: `*_count`, `*_pct` #### Usage Metrics - **usage_count**: Total number of conversations/interactions in a geography - **usage_pct**: Percentage of total usage (relative to parent geography - global for countries, parent country for country-state regions) #### Content Facet Metrics **O*NET Task Metrics**: - **onet_task_count**: Number of conversations using this specific O*NET task - **onet_task_pct**: Percentage of geographic total using this task - **onet_task_pct_index**: Specialization index comparing task usage to baseline (global for countries, parent country for country-state regions) - **onet_task_collaboration_count**: Number of conversations with both this task and collaboration pattern (intersection) - **onet_task_collaboration_pct**: Percentage of the base task's total that has this collaboration pattern (sums to 100% within each task) #### Occupation Metrics - **soc_pct**: Percentage of classified O*NET tasks associated with this SOC major occupation group (e.g., Management, Computer and Mathematical) **Request Metrics**: - **request_count**: Number of conversations in this request category level - **request_pct**: Percentage of geographic total in this category - **request_collaboration_count**: Number of conversations with both this request category and collaboration pattern (intersection) - **request_collaboration_pct**: Percentage of the base request's total that has this collaboration pattern (sums to 100% within each request) **Collaboration Pattern Metrics**: - **collaboration_count**: Number of conversations with this collaboration pattern - **collaboration_pct**: Percentage of geographic total with this pattern **Multitasking Metrics**: - **multitasking_count**: Number of conversations with this multitasking status - **multitasking_pct**: Percentage of geographic total with this status **Human-Only Ability Metrics**: - **human_only_ability_count**: Number of conversations with this human-only ability status - **human_only_ability_pct**: Percentage of geographic total with this status **Use Case Metrics**: - **use_case_count**: Number of conversations in this use case category - **use_case_pct**: Percentage of geographic total in this category **Task Success Metrics**: - **task_success_count**: Number of conversations with this task success status - **task_success_pct**: Percentage of geographic total with this status #### Numeric Facet Metrics For numeric facets (human_only_time, human_with_ai_time, ai_autonomy, human_education_years, ai_education_years), the following distribution statistics are available: - **{facet}_mean**: Mean value across all conversations - **{facet}_median**: Median value across all conversations - **{facet}_stdev**: Standard deviation of values - **{facet}_mean_ci_lower**: Lower bound of 95% confidence interval for the mean - **{facet}_mean_ci_upper**: Upper bound of 95% confidence interval for the mean - **{facet}_median_ci_lower**: Lower bound of 95% confidence interval for the median - **{facet}_median_ci_upper**: Upper bound of 95% confidence interval for the median - **{facet}_count**: Total number of observations for this facet - **{facet}_histogram_count**: Count of observations in each histogram bin (one row per bin, bin range in cluster_name, e.g., "[1.0, 1.0)") - **{facet}_histogram_pct**: Percentage of observations in each histogram bin (one row per bin) For numeric intersection facets (e.g., onet_task::human_only_time), the same metrics are available per category (e.g., per O*NET task), with cluster_name containing the category identifier: - **{base}_{numeric}_mean**: Mean value for this category - **{base}_{numeric}_median**: Median value for this category - **{base}_{numeric}_stdev**: Standard deviation for this category - **{base}_{numeric}_count**: Number of observations for this category - **{base}_{numeric}_mean_ci_lower/upper**: 95% CI bounds for the mean - **{base}_{numeric}_median_ci_lower/upper**: 95% CI bounds for the median #### Special Values - **not_classified**: Indicates data that was filtered for privacy protection or could not be classified - **none**: Indicates the absence of the attribute (e.g., no collaboration, no task selected) ### Data Processing Notes - **Minimum Observations**: 200 conversations per country, 100 per country-state region (applied in enrichment step, not raw preprocessing) - **not_classified**: - For regular facets: Captures filtered/unclassified conversations - For intersection facets: Each base cluster has its own not_classified (e.g., "task1::not_classified") - **Intersection Percentages**: Calculated relative to base cluster totals, ensuring each base cluster's percentages sum to 100% - **Country Codes**: ISO-3166-1 format for countries, three letter codes in the enriched file (e.g., "USA", "GBR", "FRA") and two letter codes in the raw file (e.g., "US", "GB", "FR"); ISO 3166-2 format for country-state regions (e.g., "AGO-LUA", "ALB-02" in enriched file, or "US-CA" in raw file) - **Variable Definitions**: See Core Variables section above ## 1P API Usage Data ### Overview Dataset containing first-party API usage metrics along various dimensions based on a sample of 1P API traffic and analyzed using privacy-preserving methods. **Note**: Unlike Claude.ai data, API data has **no geographic breakdowns** (no country or country-state facets). All API metrics are reported at global level only (`geography: "global"`, `geo_id: "GLOBAL"`). **Source file**: `aei_raw_1p_api_2025-11-13_to_2025-11-20.csv` (in data/intermediate/) ### Data Schema Each row represents one metric value for a specific facet combination at global level: | Column | Type | Description | |--------|------|-------------| | `geo_id` | string | Geographic identifier (always "GLOBAL" for API data) | | `geography` | string | Geographic level (always "global" for API data) | | `date_start` | date | Start of data collection period | | `date_end` | date | End of data collection period | | `platform_and_product` | string | "1P API" | | `facet` | string | Analysis dimension (see Facets below) | | `level` | integer | Sub-level within facet (0-2) | | `variable` | string | Metric name (see Variables below) | | `cluster_name` | string | Specific entity within facet. For intersections, format is "base::category" or "base::index"/"base::count" for mean value metrics | | `value` | float | Numeric metric value | ### Facets **Content Facets:** - **onet_task**: O*NET occupational tasks - **collaboration**: Human-AI collaboration patterns - **request**: Request categories (hierarchical levels 0-2 from bottom-up taxonomy) - **multitasking**: Whether conversation involves single or multiple tasks - **human_only_ability**: Whether a human could complete the task without AI assistance - **use_case**: Use case categories (work, coursework, personal) - **task_success**: Whether the task was successfully completed **Numeric Facets** (continuous variables with distribution statistics): - **human_only_time**: Estimated time for a human to complete the task without AI - **human_with_ai_time**: Estimated time for a human to complete the task with AI assistance - **ai_autonomy**: Degree of AI autonomy in task completion - **human_education_years**: Estimated years of human education required for the task - **ai_education_years**: Estimated equivalent years of AI "education" demonstrated **Intersection Facets:** - **onet_task::collaboration**: Intersection of O*NET tasks and collaboration patterns - **onet_task::multitasking**: Intersection of O*NET tasks and multitasking status - **onet_task::human_only_ability**: Intersection of O*NET tasks and human-only ability - **onet_task::use_case**: Intersection of O*NET tasks and use case categories - **onet_task::task_success**: Intersection of O*NET tasks and task success - **onet_task::human_only_time**: Mean human-only time per O*NET task - **onet_task::human_with_ai_time**: Mean human-with-AI time per O*NET task - **onet_task::ai_autonomy**: Mean AI autonomy per O*NET task - **onet_task::human_education_years**: Mean human education years per O*NET task - **onet_task::ai_education_years**: Mean AI education years per O*NET task - **onet_task::cost**: Mean cost per O*NET task (indexed, 1.0 = average) - **onet_task::prompt_tokens**: Mean prompt tokens per O*NET task (indexed, 1.0 = average) - **onet_task::completion_tokens**: Mean completion tokens per O*NET task (indexed, 1.0 = average) - **request::collaboration**: Intersection of request categories and collaboration patterns - **request::multitasking**: Intersection of request categories and multitasking status - **request::human_only_ability**: Intersection of request categories and human-only ability - **request::use_case**: Intersection of request categories and use case categories - **request::task_success**: Intersection of request categories and task success - **request::human_only_time**: Mean human-only time per request category - **request::human_with_ai_time**: Mean human-with-AI time per request category - **request::ai_autonomy**: Mean AI autonomy per request category - **request::human_education_years**: Mean human education years per request category - **request::ai_education_years**: Mean AI education years per request category - **request::cost**: Mean cost per request category (indexed, 1.0 = average) - **request::prompt_tokens**: Mean prompt tokens per request category (indexed, 1.0 = average) - **request::completion_tokens**: Mean completion tokens per request category (indexed, 1.0 = average) ### Core Variables #### Content Facet Metrics **O*NET Task Metrics**: - **onet_task_count**: Number of 1P API records using this specific O*NET task - **onet_task_pct**: Percentage of total using this task **Request Metrics**: - **request_count**: Number of 1P API records in this request category - **request_pct**: Percentage of total in this category **Collaboration Pattern Metrics**: - **collaboration_count**: Number of 1P API records with this collaboration pattern - **collaboration_pct**: Percentage of total with this pattern **Multitasking Metrics**: - **multitasking_count**: Number of records with this multitasking status - **multitasking_pct**: Percentage of total with this status **Human-Only Ability Metrics**: - **human_only_ability_count**: Number of records with this human-only ability status - **human_only_ability_pct**: Percentage of total with this status **Use Case Metrics**: - **use_case_count**: Number of records in this use case category - **use_case_pct**: Percentage of total in this category **Task Success Metrics**: - **task_success_count**: Number of records with this task success status - **task_success_pct**: Percentage of total with this status #### Numeric Facet Metrics For numeric facets (human_only_time, human_with_ai_time, ai_autonomy, human_education_years, ai_education_years), the following distribution statistics are available: - **{facet}_mean**: Mean value across all records - **{facet}_median**: Median value across all records - **{facet}_stdev**: Standard deviation of values - **{facet}_mean_ci_lower**: Lower bound of 95% confidence interval for the mean - **{facet}_mean_ci_upper**: Upper bound of 95% confidence interval for the mean - **{facet}_median_ci_lower**: Lower bound of 95% confidence interval for the median - **{facet}_median_ci_upper**: Upper bound of 95% confidence interval for the median - **{facet}_count**: Total number of observations for this facet - **{facet}_histogram_count**: Count of observations in each histogram bin (one row per bin) - **{facet}_histogram_pct**: Percentage of observations in each histogram bin (one row per bin) #### Indexed Facet Metrics (API-specific) For indexed facets (cost_index, prompt_tokens_index, completion_tokens_index), values are normalized so that 1.0 represents the average: - **{facet}_index**: Re-indexed mean value (1.0 = average across all categories) - **{facet}_count**: Number of records for this metric #### Intersection Metrics For categorical intersections (e.g., onet_task::collaboration): - **{base}_{secondary}_count**: Records with both this base category and secondary category - **{base}_{secondary}_pct**: Percentage of the base category's total with this secondary category For numeric intersections (e.g., onet_task::human_only_time): - **{base}_{numeric}_mean**: Mean value for this category - **{base}_{numeric}_median**: Median value for this category - **{base}_{numeric}_stdev**: Standard deviation for this category - **{base}_{numeric}_count**: Number of observations for this category - **{base}_{numeric}_mean_ci_lower/upper**: 95% CI bounds for the mean - **{base}_{numeric}_median_ci_lower/upper**: 95% CI bounds for the median ## External Data Sources We use external data to enrich Claude usage data with external economic and demographic sources. ### ISO Country Codes **ISO 3166 Country Codes** International standard codes for representing countries and territories, used for mapping IP-based geolocation data to standardized country identifiers. - **Standard**: ISO 3166-1 - **Source**: GeoNames geographical database - **URL**: https://download.geonames.org/export/dump/countryInfo.txt - **License**: Creative Commons Attribution 4.0 License (https://creativecommons.org/licenses/by/4.0/) - **Attribution note**: Data in the data/intermediate and data/output folders have been processed and modified from original source; modifications to data in data/intermediate include extracting only tabular data, selecting a subset of columns, and renaming columns; modifications to data in data/output include transforming data to long format - **Download date**: September 2, 2025 - **Output files**: - `geonames_countryInfo.txt` (raw GeoNames data in data/input/) - `iso_country_codes.csv` (processed country codes in data/intermediate/) - **Key fields**: - `iso_alpha_2`: Two-letter country code (e.g., "US", "GB", "FR") - `iso_alpha_3`: Three-letter country code (e.g., "USA", "GBR", "FRA") - `country_name`: Country name from GeoNames - **Usage**: Maps IP-based country identification to standardized ISO codes for consistent geographic aggregation ### ISO Region Code Mapping Region-level geographic data uses ISO 3166-2 standard subdivision codes. Some countries were excluded from region-level analysis due to mapping issues between source data codes and ISO 3166-2 standards. Country-level data remains available for all countries.