U.S._Inter-State_Migration_by_Age_Radaris_Anonymized

Full documentation -> [Radaris Research](https://radaris.com/research/) An anonymized, state-level view of where people move across U.S. states, broken down by age band. One row per person: an age group, the state they started in, and the state they live in now. ## Why this dataset exists Where people move, and how that differs across the life course, is a core question in demography, economics, and urban policy - yet clean, ready-to-use micro-level data on individual migration is hard to come by. This dataset offers a simple, privacy-safe view of internal migration across U.S. states, broken down by age. It is built to answer one question in particular: do migration patterns differ between younger and older people - and if so, how. It is deliberately small in width and large in depth of care: four columns, hundreds of thousands of people, and a transformation pipeline designed so that the result reveals population-level patterns while revealing nothing about any single person. ## What's in it One row per person, four columns: - person_id: A random surrogate ID. Not derived from any real identifier and not reversible. A row key only, not a feature. - age_group: Age band from year of birth: <25, 25-39, 40-54, 55-69, 70+. - first_state: The person's earliest recorded state of residence (origin), as a 2-letter code. - last_state: The person's current state of residence (destination), as a 2-letter code. If first_state == last_state, no interstate move was recorded (a "stayer"). If they differ, the pair encodes a directional flow origin -> destination. ## How it was built 1. Sampling. A uniform random sample of ~500,000 records was drawn from the full source database, so the sample's distributions reflect the source population. 2. Endpoint extraction. Each source record carried a residential history. We reduced each history to its two endpoints - the earliest state and the current state - and dropped everything in between. 3. Cleaning. Military postal codes (AA, AE, AP, used by APO/FPO/DPO overseas addresses rather than real states) were removed before extracting endpoints, so they never contaminate origin or destination. 4. De-identification. All direct identifiers - names, source IDs, cities, and full address histories - were removed. Year of birth was generalized into five age bands. The original ID was replaced with a random surrogate. 5. Re-identification control. The file enforces k-anonymity with k = 5 over the combination {age_group, first_state, last_state}: every published combination is shared by at least five people. The rare combinations that fell below this threshold (~0.6% of rows) were removed prior to release. ## Representativeness Because the 500,000 rows are a uniform random sample of the source database, the sample's marginal distributions are unbiased estimates of the full population: geographic composition, age composition, and interstate-mobility rate (overall and by age). Two boundaries: the k-anonymity step thins the rarest origin->destination corridors (do not treat tail frequencies as population estimates), and the data only speaks at the state level (no cities, intermediate stops, or move timing). ## Limitations - State-level only; nothing below states is recoverable by design. - Endpoints only; intermediate states, number of moves, and return migration are not represented. - No dates; this is a cross-sectional snapshot, not a time series. - "Stayer" means no interstate move was recorded, not necessarily no move at all. - Sampling + suppression slightly thin the rarest flows. ## Privacy approach This is a de-identified, state-level aggregate: no names, no cities, no full trajectories. The k = 5 threshold guarantees that no row corresponds to a rare or unique age-plus-origin-plus-destination profile, so the file cannot be used to single out or re-identify an individual. ## Source & terms The underlying data is provided by Radaris, a people-search platform with an extensive database of public records and demographic information in the United States. The source material has been stripped of all personal identity elements and synthesized into an aggregated, anonymous format, intended strictly for statistical, demographic, and academic research. Mirrors: [Kaggle](https://www.kaggle.com/datasets/zaramann/u-s-inter-state-migration-by-age), [Hugging Face](https://huggingface.co/datasets/ZaraMann/U.S._Inter-State_Migration_by_Age), [Zenodo](https://zenodo.org/records/21321002).