Migration_Trajectory_Types_by_Age_Microdata_Radaris_Anonymized

Full documentation -> [Radaris Research](https://radaris.com/research/) A de-identified dataset for studying the shape of people's lifetime state-level migration paths - did they stay, move once, return to where they started, or move repeatedly - together with age band and the size of their recorded kinship network. Two files are provided: - Microdata (migration_trajectories_microdata.csv, ~787K rows, this dataset) - one row per person, so you can compute and verify everything yourself. - Summary table (migration_trajectories_summary.csv, 93 rows, companion dataset Migration_Trajectory_Types_by_Age_Summary_Radaris_Anonymized) - a ready-made cross-tabulation age_group x trajectory_type x rel_bin -> people, for quick use without processing the full file. Both are derived from the same records; the summary is just the microdata aggregated. ## Why this dataset exists (the gap it fills) Migration research has a documented blind spot around the direction and repetition of individual moves. Studies of migration and social networks "usually do not rely on the number of moves at the individual level, but use the aggregate level of internal migration at the country level," and consequently "existing research overlooks the direction of migration (i.e., onward and return) due to data limitations" (Thomas & Gillespie, 2024, Social Science Research). Return migration is especially constrained by "the lack of data on return migration," and this is sharpest at older ages, where "due to data limitations, theoretical frameworks and empirical evidence fail to provide clear indication of the ... mechanism of return migration, especially at older ages" (Sheftel, 2021). This dataset speaks to that gap: it labels stayers, single moves, returns, and repeat/onward moves at the individual level and crosses them with age and kinship-network size. Two advantages over the usual sources: - vs. retrospective surveys (the common source for migration biographies), which suffer from "recall of past migration events that may lead to biased estimates" (Thomas & Gillespie, 2024): here paths come from records, not memory - no recall bias (but see the censoring caveat). - vs. Census / IRS flow tables, which give yearly origin -> destination flows but not an individual's path (return vs. onward) or its link to kinship. ## Reproduce it yourself This is individual-level microdata, so the headline numbers are not something you have to take on trust - you can recompute them. For example, the share of movers who returned to their origin state, by age: import pandas as pd df = pd.read_csv("migration_trajectories_microdata.csv") movers = df[df.trajectory_type != "stayer"] print(movers.groupby("age_group").apply(lambda s: (s.trajectory_type == "return").mean() * 100).round(1)) Running this on the microdata returns the same figures reported below. The 93-row summary table is exactly this kind of aggregation, precomputed. ## Headline pattern: return migration rises with age Among people who moved at all, the share whose path is a return to their origin state increases with age (from the microdata; <25 is sparse after privacy filtering and is not reliably estimated): age_group | share of movers who returned 25-39 | 30.7% 40-54 | 36.9% 55-69 | 39.9% 70+ | 40.4% By age 70+, roughly two in five movers are returnees - a large, clean signal on exactly the older-age return migration the literature flags as under-measured. (Association, not a causal claim.) Overall trajectory mix (of people with a recorded history): stayer 56.9%, return 16.4%, single move 15.5%, repeat/onward 11.1%. ## Columns (this file: migration_trajectories_microdata.csv) - person_id: Random surrogate id - not reversible, not shared with other datasets (cannot be joined). Row key only. - age_group: <25, 25-39, 40-54, 55-69, 70+. - region: Current U.S. Census region: Northeast, Midwest, South, West. - trajectory_type: stayer (one state), single move (two states), return (ends in origin state after leaving), repeat/onward (3+ distinct states, no return). - rel_bin: Recorded-relatives bucket: none, 1, 2, 3-4, 5+. The companion summary file (migration_trajectories_summary.csv) has columns: age_group, trajectory_type, rel_bin, people - one row per cell (no region dimension). ## What you can do with it - Study return migration by age and kinship-network size - a phenomenon the literature calls data-starved, especially at older ages. - Compare trajectory types across the life course and across regions (microdata). - Test whether kinship-network size associates with path type. - Use as a large, records-based complement to small survey samples on migration biographies. ## How it was built 1. Trajectory type derived per person from the state sequence in the residential history (consecutive duplicate states collapsed; military codes AA/AE/AP removed). return = last state equals the first after having left it. region = current region. 2. Kinship count without links - only the number of filled relative slots; relative names and ids were never carried forward, so no relationship graph is produced. 3. De-identification - names, source ids, cities and full histories removed; birth year binned; a fresh random person_id assigned (not shared with other releases). 4. k-anonymity (k=5) over {age_group, region, trajectory_type, rel_bin}; people in cells below 5 were suppressed (69 rows, <0.01%). The summary table applies k=5 as well. ## Caveats - read before drawing conclusions - return is measured at the state level, not the town level: it means returning to the origin state, not necessarily the origin town. - Trajectories can be right-censored. If early moves were not recorded, some returns and repeat moves are missed, so the return share is a lower bound, not an overestimate. - <25 is sparse after privacy filtering and should not be read as a reliable estimate. - rel_bin is recorded relatives, not family size - it partly reflects profile completeness; none means "none found," not "no family." - Association, not causation - the data shows patterns; it does not explain why people return or move. ## Provenance & terms The underlying data for this project is provided by Radaris, a comprehensive people search platform with an extensive database of public records and demographic information in the United States. Leveraging Radaris's deep data infrastructure on individuals residing and moving across the country, this dataset captures broad domestic migration trends over time. Crucially, the source material has been stripped of all personal identity elements and synthesized into an aggregated, anonymous format. The resulting dataset is intended strictly for statistical, demographic, and academic research, offering a safe and compliant framework for studying population-level mobility without compromising individual privacy. ## Contact [e-mail]