Survival Analysis of First Migration
Also known as: Hazard Model of First Move, Time-to-First-Migration Analysis, Age-at-First-Migration Survival Model, First-Move Event-History Analysis
Survival analysis of first migration treats the move out of one's place of origin as a timed event and asks not whether but when a person first migrates. Rather than modeling a binary 'migrated or not' outcome, it follows individuals from the moment they become at risk and models the instantaneous hazard of a first move as a function of age and changing life circumstances. The approach, codified for the social sciences by Blossfeld and Rohwer's event-history framework and applied to migration biographies by Clara Mulder, handles the two features that defeat ordinary regression: censoring, because most people in a sample have not yet migrated when observed, and time-varying covariates, because the things that trigger a move — finishing school, finding a job, forming a union — themselves change over time. The result is an estimate of how the risk of a first move rises and falls across the life course and how it responds to time-dependent conditions. It can be fitted nonparametrically with a Cox model or with a parametric baseline when the shape of age dependence is of interest.
Key highlights
- Models the timing of the first move directly through the hazard, recovering the life-course age pattern of leaving home rather than a flat yes/no outcome.
- Handles right-censoring correctly, using the information in not-yet-migrants instead of misclassifying them as non-migrants.
- Accommodates time-varying covariates, so triggers such as a new job or a union can be linked to the move as they actually occur.
- The Cox specification needs no assumption about the baseline age curve, while a parametric baseline is available when the shape of age dependence is itself the question.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Use survival analysis of first migration when you have longitudinal or retrospective biographical data that record the timing of a first move and you want to explain when, not merely whether, people leave their origin. It is the right tool whenever a large share of the sample has not yet migrated (so the outcome is censored) and when the suspected drivers — jobs, partnership, education, housing — change over time and should enter as time-varying covariates. It suits leaving-home, first internal moves, and first international departures alike, and it is preferable to logistic regression on an 'ever migrated' indicator because the latter ignores timing and mishandles not-yet-migrants. It is less appropriate when only a cross-sectional snapshot exists with no timing information, when repeated moves rather than the first are the object of interest (use a multi-episode design instead), or when the at-risk start time cannot be defined.
Strengths & limitations
- Models the timing of the first move directly through the hazard, recovering the life-course age pattern of leaving home rather than a flat yes/no outcome.
- Handles right-censoring correctly, using the information in not-yet-migrants instead of misclassifying them as non-migrants.
- Accommodates time-varying covariates, so triggers such as a new job or a union can be linked to the move as they actually occur.
- The Cox specification needs no assumption about the baseline age curve, while a parametric baseline is available when the shape of age dependence is itself the question.
- Requires individual-level data with reliable event and censoring times, which retrospective surveys often record with recall error.
- The proportional-hazards assumption can fail when a covariate's effect changes with age, demanding interactions with time or stratification.
- Unobserved heterogeneity (frailty) can bias estimates of duration dependence, making the hazard look as if it falls when it is selection at work.
- It models only the first event; analysts interested in repeat or onward moves must move to multi-episode or competing-risks designs.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
Why not just run a logistic regression on whether someone has ever migrated?
Because that throws away the timing and mishandles people who have simply not migrated yet. A 20-year-old who has not moved is not equivalent to a 50-year-old who never moved; the first is still at high risk and the second has largely passed through it. Survival analysis treats the not-yet-migrant as right-censored, contributing information up to the moment we lose sight of her, and models the hazard so that the same covariate can have different effects at different ages. Logistic regression on 'ever moved' confounds exposure time with propensity and cannot represent the life-course rhythm of first migration.
Should I use a Cox model or a parametric survival model?
It depends on whether the shape of age dependence is itself of interest. The Cox model leaves the baseline hazard unspecified and estimates only the covariate effects via partial likelihood, which is robust and convenient when you care about hazard ratios rather than the absolute age curve. A parametric baseline (Gompertz, Weibull, or piecewise-constant) is preferable when you want to describe how the risk of a first move rises and falls with age, to extrapolate, or to simulate. Blossfeld and Rohwer treat both as part of one toolkit; many migration studies fit a flexible piecewise-constant baseline to capture the strong age pattern explicitly.
How do time-varying covariates enter the model?
The data are organized so that each person contributes one or more intervals during which their covariate values are constant, and the values update when circumstances change — a new job, a marriage, the birth of a child. The hazard at any instant is then evaluated using the covariate values in force at that instant. This is essential for migration, because Mulder's work shows the first move is triggered by such transitions; entering only fixed background traits would miss the mechanism. Care is needed to ensure covariates are lagged appropriately so that values measured after the move do not leak into the prediction of it.
Sources
- 1.Blossfeld, H.-P., & Rohwer, G. (2002). Techniques of Event History Modeling: New Approaches to Causal Analysis (2nd ed.). Lawrence Erlbaum.ISBN 9780805840919
- 2.Mulder, C. H. (1993). Migration Dynamics: A Life Course Approach. Thesis Publishers, Amsterdam.ISBN 9789051701814
You have read it. What now?
Cite this page
ScholarGate. (2026, June 23). Survival Analysis of First Migration. ScholarGate. https://scholargate.app/migration-studies/survival-analysis-of-first-migration