Migration & the Gravity Model

An idea borrowed from physics

Ernst Georg Ravenstein noticed in his 1885 "Laws of Migration" that migration between two places seemed to depend on their size and on how far apart they were, and in 1946 the sociologist George Zipf gave the idea a precise mathematical form directly modelled on Newton's law of gravitation: the flow between two places should scale with the product of their populations and fall off with the distance between them. That analogy — literal masses attracting each other in physics, populations "attracting" migrants in geography — gives the model its name and its enduring simplicity.

M_ij = k · Pop_i^α · Pop_j^β / dist_ij^γ

M_ij   predicted flow from place i to place j
Pop_i, Pop_j   the population (or economic mass) of origin and destination
dist_ij        distance between i and j (physical, or a travel-cost proxy)
α, β    mass exponents — usually close to 1, fitted to data
γ       distance-decay exponent — larger γ ⇒ migration is more local
k       calibration constant tying predicted flow to observed units

In its simplest symmetric form the exponents on both populations are 1 and the distance term is a straightforward power law, giving the familiar M_ij = k · Pop_i · Pop_j / dist^γ. Larger cities generate and attract proportionally more migration because they offer more jobs, more social ties, and more institutions worth moving for; distance suppresses flow because moving is costly — financially, socially, and in the effort required to gather information about a distant place.

live demo · migrant flow between cities sized by population● LIVE

Calibrating the decay exponent

The distance-decay exponent γ is where the model does its real work, and it is not a universal constant — it is fitted separately to whatever flow it is being asked to predict. A small γ implies migrants are almost indifferent to distance, appropriate for highly mobile, well-off populations choosing between distant metropolitan areas; a large γ implies migration is dominated by short local moves, more typical of lower-income populations or shorter-term relocation. Alternative "deterrence functions" beyond a simple power law — an exponential decay e^(−β·d), or piecewise functions for different transport modes — are common when the power law fits the data poorly at very short or very long ranges.

Beyond migration: the same equation everywhere

The gravity model's real popularity in regional science comes from how widely the same functional form applies. It underlies the classic "trip distribution" step of the four-step transportation planning model used by city and regional planners to predict commuting flows between zones; economists use essentially the same equation to model international trade flows between countries, where it is one of the most empirically successful relationships in the field; and epidemiologists use it to predict how a disease is likely to spread between connected cities, using flight or migration data as the mass and distance terms.

The radiation model: a parameter-free alternative

The gravity model's chief weakness is that its distance-decay parameters have to be recalibrated for every new context, country or dataset, with no principled way to transfer a fitted γ from one setting to another. In 2012, Filippo Simini, Marta González, Amos Maritan and Albert-László Barabási proposed the radiation model, which replaces distance decay with the idea of intervening opportunities: a migrant compares their home location against every possible destination and is more likely to settle at the nearest place that offers a genuinely better opportunity than what is available closer to home. The resulting formula requires no free parameters to fit at all, and in several benchmark studies it predicted commuting and migration flows as well as — or better than — a carefully calibrated gravity model.

Frequently asked questions

Why does the gravity model use population instead of some other measure of city size?

Population is a convenient proxy for how much economic and social activity a place generates — more people means more jobs, more social ties, more universities and more reasons someone elsewhere might want to move there. In practice, researchers often substitute or add other 'mass' variables such as GDP or employment when those better predict a specific flow, keeping the same multiplicative structure.

What does the distance decay exponent actually control?

It controls how sharply predicted flow falls off as distance grows. A small exponent means distance barely matters and people move about as readily to far cities as to nearby ones; a large exponent means migration is dominated by short, local moves. Empirically, the exponent differs by transport mode and migration type, and is typically fitted to observed flow data rather than assumed.

Is the gravity model still the state of the art for predicting migration?

No — it remains widely used because it is simple and interpretable, but it needs distance-decay parameters calibrated separately for every new context, which limits generalisation. The radiation model, proposed by Filippo Simini and colleagues in 2012, replaces the calibrated distance-decay function with a parameter-free formula based on intervening opportunities, and often matches or beats the gravity model without any fitting at all.

Try it live

Everything above runs in your browser — open Migration Model and change the parameters while it is running. Nothing is installed, nothing is uploaded, the whole model lives in one tab.

▶ Open Migration Model simulation

Migration & the Gravity Model: Borrowing Newton for People