The Early History of Migration Research

Revisited

Michael J. Greenwood Department of Economics

University of Colorado, Boulder, CO michael.greenwood@colorado.edu

Gary L. Hunt Department of Economics

University of Maine, Orono, ME Gary_Hunt@umit.maine.edu

This paper provides a history of the early contributions to the scientific study of migration. We begin with Ravenstein (1880s) and also feature the work of D.S. Thomas (1930s). Moreover, the development of the gravity model as applied to migration research (1930s and 1940s) is discussed. We discuss the historical reasons for interest in various migration phenomena, and we briefly treat the development of migration data sources.

Although internal migration has been a common aspect of human behavior, the scientific

study of such migration has been fairly recent. Ravenstein’s papers in 1880s provide the earliest

examples of what can be called scientific studies of internal migration. However, not until the

1930s did migration research “take off” as a subject of widely based social scientific inquiry.

Data limitations were certainly a factor in discouraging the study of internal migration, but such

limitations could not have been the sole factor. Lifetime migration data similar to those studied

by Ravenstein were available in the 1850 U.S. Census and in the 1851 Census of the United

Kingdom, and yet it was not until over 30 years later that Ravenstein conducted his work, and no

immediate follow-up to his research appears to have occurred, with the possible exception of

Welton (1911).

The phenomenon most responsible for initial interest in migration as a field of scientific

study appears to have been urbanization. During the late nineteenth and early twentieth centuries,

not only did U.S. and Western European urban areas grow rapidly, but also the percentage of the

population of the various nations living in urban areas (urbanization) grew rapidly. Urban growth

and urbanization were fueled by rural-to-urban migration. (In the United States, immigration

from abroad also encouraged urban growth because most of the immigrants settled in cities.)

Issues of rural depopulation and the adjustments of rural-to-urban migrants to life in the city

were prominent. An important subtopic in the United States was the movement of African

Americans, as well as whites, out of the South to northern cities. Questions relating to rural-to-

urban migration and South-to-North migration were the focus of what attention was directed to

internal migration in the United States during the 1920s.

A second societal force that encouraged interest in migration research was the Great

Depression. The importance of the Depression in this respect should not be underestimated, as

3

issues concerned with the Depression dovetailed with those regarding rural-to-urban migration.

During the 1930s, urban employment opportunities were hard hit and unemployment was great.

Urban unemployment was swelled in part by migrants from rural America, which caused

academicians and policy makers alike to ask if the urban in-migrants would be better off in their

rural origins. In terms of internal migration, no period has seen greater emphasis on policy than

the 1930s.

In her Research Memorandum on Migration Differentials, published in 1938, Dorothy

Swaine Thomas provides two appendices with annotated bibliographies that emphasize the

burgeoning of migration research during the 1930s. Bibliography I (Appendix A) contains 119

citations of U.S. and British work, and Bibliography II (Appendix B2) contains 72 citations to

German work. Thomas indicates that the various works were selected for inclusion because they

related either to migration differentials or to methodological issues, but in general we have no

good reason to believe that the works included in the two bibliographies were not representative

of the body of migration research as it existed during the late 1930s.

In the bibliography regarding American and British work, 90 of the 119 entries (75.6

percent) are for the 1930s. Another 23 (19.3 percent) are for the 1920s. In the bibliography

regarding German work, 40 of the 72 entries (55.6 percent) are for the 1930s. In Germany,

serious research on migration (involving primarily rural-to-urban migration) must have begun

earlier than in the United States and Great Britain because 12 German References (16.7 percent)

are for the 1890s, and Thomas has only a single American and British reference dating to this

period. Even allowing for obsolescence of knowledge and decay in citations to older works, the

1930s appear to have been when migration research began in earnest as a broadly-based field of

study.

4

Migration research has a strong interdisciplinary history. Although Ravenstein was a

geographer, it is clear that sociologists/demographers were the first to pursue the study of

internal migration in significant numbers. Lack of detailed migration statistics did not prevent

them from developing their own data using indirect methods, nor did it discourage them from

finding all sorts of other data relating to migration (e.g., population registry data). By the 1920s

certain sociologists/demographers were already prominent in this field of research, during which

time economists were just beginning to take an interest. The Depression of the 1930s was a

major impetus for economists to focus on migration as a topic of research. Since then,

individuals from many other academic disciplines have joined the study of migration, including

physicists, geographers, anthropologists, political scientists, epidemiologists, and others.

In this paper, migration research is viewed in an historical context that emphasizes the

roles of major theories, data sources, quantitative methods, and their interplay in the

development of this body of scholarship. We begin with a well-known sequel of studies

undertaken in the 19th Century.

RAVENSTEIN’S “LAWS OF MIGRATION”

In The Wealth of Nations Adam Smith paid little attention to migration. Apparently, at his

time internal migration was not a particularly important phenomenon because he states that “it

appears evidently from experience that a man is of all sorts of luggage the most difficult to be

transported” (Smith, 1776, p. 75). In the same paragraph he notes what appear to be enormous

money wage differentials: in London 18 pence a day, a few miles away 14 to 15 pence a day, in

Edinburgh 10 pence, and in the low country of Scotland 8 pence for “common labour.” He

indicates that “such a difference of prices, which it seems is not always sufficient to transport a

5

man from one parish to another” (p. 75) would, if it were for commodities, elicit so great a

transportation as to arbitrage the difference. Of course, the agricultural revolution was just

getting under way, the industrial revolution would not begin earnest for a few years, and the

great international migrations of the following century were not foreseen. Moreover the wage

differentials observed by Smith may not have reflected opportunities for equivalent utility gains

because of cost of living differences (importantly including rents) and other factors. Although we

know that major migrations occurred thousands of years before Smith’s time, it would not be

until over 100 years after he wrote The Wealth of Nations that serious migration research would

begin with Ravenstein.

Referring to the later years of the nineteenth century, the prominent British economist,

Alfred Marshal, summarized his views regarding migration: “the large towns and especially

London absorb the very best blood from all the rest of England; the most enterprising, the most

highly gifted, those with the highest physique and strongest characters go there to find scope for

their abilities” (1949; p. 199). Clearly, this statement bears on the selectivity of migration. Not

only does Marshall suggest that large cities tend to attract the most enterprising and the most

talented, but he also suggests that these same cities attract the physically strongest, as well as,

presumably, the morally and ethically strongest. The former ideas about migrants tending to be

the most enterprising and the most talented have come up repeatedly since Marshall’s time, but

the latter ideas about physical and moral strength have found little support (not withstanding

what D.S. Thomas writes about “physical and mental health differentials”). Marshall does not

support his assertions with hard data. He was familiar with E.G. Ravenstein because he cites

another, later paper (1890) in his Principles, but he never references Ravenstein’s work on

migration, which referred to the same period about which Marshall wrote.

6

One of the first systematic studies of migration was by Ravenstein (1885). He states that

the motivation for his study “was a remark made by the late Dr. William Farr [1876], to the

effect that migration appeared to go on without any definite law….” (Ravenstein, 1885; p. 167).

Ravenstein examined 1881 British census data on (i) nativity of the population and (ii) place of

residence enumerated in 1881. The geographic detail on nativity and residence included country

and county, and in some cases, town or city. Ravenstein utilized these data along with vital

statistics and immigration records to analyze “lifetime place-to-place” migration and net

migration. From the compilation of net migration figures, he was able to identify areas of

absorption (i.e., net in-migration) and dispersion (i.e., net out-migration). The sex detail in the

1881 census also permitted him to describe the sex composition of these migration patterns.

It is perhaps noteworthy that Ravenstein never came to grips with the limitations of the

lifetime migration data he studied. In the beginning of his earlier (1885) paper he discussed some

data shortcomings, such as different sizes of counties that could give rise to a “migrant” crossing

a county boundary with a move of 25 miles in one instance, whereas in another a move of 95

miles may not cross a county line. This discussion entirely missed the major criticisms of

lifetime migration data, such as their inability to pick up multiple migrations and return

migration. Such “problems” were not to be seriously discussed until much later in the United

States and Great Britain (especially during the 1920s and 1930s), but in Germany they appear to

have been a topic of interest during the 1890s.

From his various and detailed analyses, Ravenstein inferred seven “laws of migration”

which are as follows:

“1. We have already proved that the great body of our migrants only proceed a short distance, and that there takes place consequently a universal shifting or displacement of the population, which produces ‘currents of migration’ setting in

7

the direction of the great centres of commerce and industry which absorb the migrants.”

“In forming an estimate of this displacement we must take into account the number of natives of each county which furnishes the migrants, as also the population of the towns or districts which absorb them.”

“2. It is the natural outcome of this movement of migration, limited in range, but universal throughout the country, that the process of absorption would go on in the following manner:–”

“The inhabitants of the country immediately surrounding a town of rapid growth, flock into it; the gaps thus left in the rural population are filled up by migrants from more remote districts, until the attractive force of one of our rapidly growing cities makes its influence felt, step by step, to the most remote corner of the kingdom. Migrants enumerated in a certain centre of absorption will consequently grow less with the distance proportionately to the native population which furnishes them, and a map exhibiting by tints the recruiting process of any town ought clearly to demonstrate this fact. That this is actually the case will be found by referring to maps 3, 4, 8, and 9. These maps show at the same time that facilities of communication may frequently countervail the disadvantages of distance.”

“3. The process of dispersion is the inverse of that of absorption and exhibits similar features.”

“4. Each main current of migration produces a compensating counter-current.”

“5. Migrants proceeding long distances generally go by preference to one of the great centres of commerce or industry.”

“6. The natives of towns are less migratory than those of the rural parts of the country.”

“7. Females are more migratory that [sic] males.”

Considering the complete text of Ravenstein’s seven laws provides insight into the

richness of his analyses. The first law that he delineates incorporates three separate notions: (i)

that distance appears to deter migration, ceteris paribus, (ii) that large cities with vibrant

economies are the direction of much migration, and (iii) that place-to-place flows depend on the

population sizes of both the origin and destination. In combination with the second law, in part,

8

and the fifth law that he delineates, the contours of the gravity model of spatial interaction are

present. Moreover, it is a gravity model of the modified type in which differential economic

opportunities play an important role.1

A review of his detailed discussions involving the third and sixth laws reinforces the

notion that he viewed differential economic opportunities as important “pull” and “push” factors

in the context of rural-to-urban migration flows (Ravenstein, 1885; pp. 186-87). Spatial

variations in economic opportunities also are highlighted in his discussion of female migration

and choice of residence location, and in its comparison to male migration and choice of residence

location.

Woman is greater migrant than man. This may surprise those who associate women with domestic life, but the figures of the census clearly prove it. Nor do women migrate merely from the rural districts into the towns in search of domestic service, for they migrate quite as frequently into certain manufacturing districts, and the workshop is a formidable rival of the kitchen and scullery (Ravenstein, 1885; p. 196).

The only towns which have proved more attractive to males than to females are … towns [in which] male labour is more sought after than female labour (Ravenstein, 1885; p. 197).

The counties on the other hand which have retained a larger proportion of their county-born females than of males are either those which in their textile and similar industries afford employment to numerons [sic] females, or those which, owing to geographical position, are more or less remote from female labour markets, or, what brings about the same result, hold out inducements to male migrants in search of work in neighbouring iron works or coal mines (Ravenstein, 1885; p. 198).

Ravenstein emphasized the role of spatial variations in relative economic opportunity as a

key determinant of migration in general:

When we inquire into the motives which have led these migrants to leave their homes, they will be found to be various too. In most instances it will be found that they did so in search of work of a more remunerative or attractive kind than that afforded by the places of their birth (Ravenstein, 1885; p. 181).

9

This fundamental principle underlies a number of Ravenstein’s laws. In fact, during the

discussion of his paper at the Statistical Society’s meeting several individuals criticized

Ravenstein for the use of “law” in the title. The basic point was that Ravenstein had uncovered

certain empirical regularities but that the particular patterns exhibited were not necessarily

immutable. Moreover, any law concerning these regularities involved the basic underlying

processes that generated the observed regularities.

Mr. Rowland Hamilton argued that

The work of investigation had to be carried on by reversed processes, sometimes by working out from a law to its consequences, and sometimes by taking consequences, and drawing from them inferences which aided the discovery of ‘law,’ but did not in themselves constitute a ‘law’…. A natural ‘law’ worthy of the name could not admit of any exceptions. The conditions laid before them were but a portion of a very large and intricate question, and would constantly be modified by other facts (Ravenstein, 1885; p. 233).

As Mr. S. Bourne put it:

It did not appear to him from the result of the investigation that there was anything, strictly speaking, in the way of law to be discovered which regulated the migration of the population from one part of the country to another, unless it was the simple law of demand and supply (Ravenstein, 1885; p. 233).

The comments of Rev. I. Doxsey reinforce these points by providing examples of

migration flows created by changes in relative economic opportunities resulting from decisions

of farmers to plant their field in grass rather than crops; the commercial failure of a major

shipbuilder; and the relocation of firms due to labor disputes regarding the use of machinery

(Ravenstein, 1885; p. 232).

At the outset of the sequel, Ravenstein acknowledged these points: “of course I am

perfectly aware that our laws of population, and economic laws generally, have not the rigidity

of physical laws, as they are continually being interfered with by human agency.” Moreover, he

goes on to write that “instances of towns and even of countries having been stunted in their

10

growth through a mistaken policy can be found as easily as can instances in which a wise

legislation has partly neutralised natural disadvantages. But notwithstanding all these exceptions,

I hope to be able to prove that similar conditions produce similar migratory movements …”

(Ravenstein, 1889; pp. 241-42; emphasis added).

His analysis of migration in continental Europe and America (Canada and the United

States) provides further support for the empirical regularities found for the United Kingdom in

his initial work. Moreover, the underlying motivation specified for migration remains primarily

an economic one. Interestingly, he now frames his laws in terms of the structure of migration

responses to economic opportunity differentials.

I do not question for a moment that the principal, though not the only cause of migration, has to be sought for in over–population in one part of the country, whilst there exist elsewhere undeveloped resources which hold out greater promise for remunerative labour. It is obvious that this is not the only cause. Bad or oppressive laws, heavy taxation, an unattractive climate, uncongenial social surroundings, and even compulsion (slave trade, transportation), all have produced and are still producing currents of migration, but none of these currents can compare in volume with that which arises from the desire inherent in most men to ‘better’ themselves in material respects. It is thus that the surplus population of one part of the country drifts into another part, where the development of industry and commerce, or the possibility of procuring productive land still in a state of nature, call for more hands to labour. But how is this call supplied? (Ravenstein, 1889; p. 286).

Ravenstein’s answer to this rhetorical question is that the structure of the migration

response determines how the call is supplied and that the structure is summarized more or less by

his laws of migration. Ravenstein’s scholarship extends beyond the cataloging of seven laws of

migration which seem to be empirical regularities in Europe and America. It also encompasses a

model of the migration process that views spatial differentials in economic opportunity as the

impetus for migration, although to a lesser degree other factors such as physical and social

climate and public policies also play a role.

11

The structure of the response of migration to such spatial differentials is conditioned by

basic features of geography: i.e., by the relative sizes and locations of population centers; by the

transportation system infrastructure which impacts on the ease of travel and communication; and

the experience and knowledge of the population about alternative locations.

It should be emphasized at this point that this conceptual view, or model, of the migration

process was not original to Ravenstein. As he notes at the outset of his first paper, the registrar-

general wrote in his general report on the 1871 census that

The improved roads, the facilities offered under the railway system, the wonderful development of the mercantile marine, the habit of traveling about, and the increasing knowledge of workmen, have all tended to facilitate the flow of people from spots where they are not wanted to fields where their labour is in demand. The establishment of a manufacture or the opening of a new mine rallies men to it, not only from the vicinity, but from remote parts of the kingdom. The great towns afford such extraordinary facilities for the division and for the combination of labour, for the exercise of all the arts, and for the practice of all the professions, that they are every year drawing people within their limits (Ravenstein, 1885; pp. 167-68).

Ravenstein immediately adds that “farther inducements to migrate are offered by educational

facilities, salubrity of the climate or cheapness of living” (Ravenstein, 1885; p. 168).

One important aspect of this model of the structure of the response of migration is what it

leaves out: namely, the way in which the response is conditioned by the demographic structure of

the population. This is not meant as a criticism of Ravenstein and his contemporaries because the

census material available to them did not have any usable demographic detail related to

migration beyond sex. Consequently Ravenstein’s laws of the structure of migration responses to

spatial variations in economic opportunity involve primarily the geographic considerations that

are accessible to one who can rely only on aggregate lifetime place-to-place and net migration

data.

12

Ravenstein’s work incorporates both a spatial utility (economic opportunity) differential

view of the impetus to migration and a framework in which the migration response is mediated.

As such, it looks very similar to what later will become known as the modified gravity model of

migration. Interestingly, this basic conceptual framework enables one to consistently interpret

most of Ravenstein’s laws, but not all of them.

Perhaps the most significant issue involves the fourth law concerning “counterstream”

effects. If spatial utility differentials are driving migration, why do these essentially offsetting

effects occur? They are hard to explain without people and place heterogeneity, and the data

Ravenstein and his contemporaries had were insufficiently detailed in these respects to permit

any related analyses. It should be added at this point that Ravenstein does not account for any of

the “counterstream” effects by reference to the data on return migration.

In his classification of migrants, the concept of a temporary migrant is delineated but this

includes military personnel, college students, migrant agricultural labor, and prisoners (pp. 181-

84). No reference is made to return migrants. However, during the discussion following the

presentation of the paper, Mr. Hamilton refers to the phenomenon without response from

Ravenstein.

Some twelve years ago, for example, there was a considerable demand in the northern mining countries, especially Durham, for labour, and employers sent far and wide to get it. They attracted a considerable number of agricultural labourers from Norfolk, and the result was curious, and by no means that unsatisfactory. Some of those men came back again and said that the change was all a delusion, that they had to work harder and to spend a good deal of the enhanced wages in the food necessary for their support. Others of a more vigorous class, even when admitting this, preferred their new way of life (Ravenstein, 1885; p. 233).

In addition to these issues, there is also the general issue of whether the differential utility

notion could not be applied to interpret the structure of the migration response as well as the

impetus to migration. Of course this is what unfolds when the human capital model is developed,

13

as discussed. Anticipating this material a bit, the distance structure of the migration response is

consistent with this larger application of utility ideas. For example, the disutility of moving is

consistent with the first law, as well as with the aspect of the second law that concerns the

staging sequence of migration from remote districts to areas closer to large cities to replace the

previous residents from such areas who migrated to the cities.2 However, the “counterstream”

effects still require heterogeneity to be added.

U.S. MIGRATION RESEARCH: 1890-1920

After Ravenstein’s work was published in England, work on migration continued in

Europe. In contrast, there was a 30-year hiatus in research on internal migration in the U.S. One

of the main reasons was that immigration to the U.S. and its large eastern seaboard cities was

enormous during the first decade and a half of the Twentieth Century.

As reported by Borjas (1994; Table 1, p. 1668), immigration to the U.S. during the

decade 1901-1910 totaled nearly 8.8 million persons and accounted for nearly 54 percent of U.S.

population change during the period. In the following decade, which includes the war,

immigration totaled over 5.7 million persons and accounted for nearly 41 percent of the U.S.

population change for the corresponding decade. Immigration continued to slow during the

1920s, to about 4.1 million persons (24.6 percent of U.S. population change), partly in response

to changes in U.S. immigration law. During the depression decade of the 1930s, immigration

slowed to a trickle (about 0.5 million).

During the early decades of the twentieth century, research on internal migration was

displaced by a social and policy induced focus on immigration and the consequences of it,

particularly in large cities. In addition to recognizing this key factor, Dorothy Swain Thomas, in

14

her outstanding Research Memorandum on Migration Differentials, sets forth several additional

reasons.

The lack of interest in internal migrations is probably due in part to the apparently unlimited possibilities of economic and population expansion in America in the decades preceding the war. Although cities were expanding rapidly, the agricultural population was not only holding its own but producing a surplus. There was no widespread apprehension about the possible dysgenic aspects of this situation, about ‘rural exodus’ as a concomitant of ‘urbanization,’ or about the possible effects on national well-being of growth of cities at the expense of the country regions. In Germany and other older European nations, the real or imagined ill effects of urbanization had brought internal migration to the fore as a problem of importance long before the turn of the 20th century, for rural areas were experiencing difficulties in holding their population not only against the attractive forces of rapidly expanding cities but also against the forces operating in favor of emigration to America and other relatively new and free lands. The apparent menace of ‘depopulation’ led to an enormous activity among social scientists even though, in some instances, this activity was more speculative than scientific (D.S. Thomas, 1938; p. 2).

U.S. CONTRIBUTIONS, 1920-1940

Several changes occurred after World War I that stimulated U.S. research on internal

migration. First of all, the war had greatly decreased immigration to the U.S. and immigration

law changes establishing the national-origins quota system legally restricted immigration.

Second, there was contemporaneously a strong demand for labor in the northern industrial

centers. Third, African-American migration from the South to the North “swelled to unheard of

proportions” (D.S. Thomas, 1938; p. 2). And fourth, there was the decline in the birth rate both

in urban and in rural areas.

Finally, the dislocations produced by droughts and the Great Depression added to the

urgency of understanding the role of migration in labor market adjustments and in the

consequences for rural areas and cities. As noted by Thomas, “the depression migrations have

15

attracted widespread attention, and have stimulated research into the incidence and types of

migration and the kinds of people migrating” (D.S. Thomas, 1938; p. 3).

In 1924 the Social Science Research Council appointed the Committee on Scientific

Aspects of Human Migration. As an outgrowth of the Council’s interest in migration, in1934 C.

Warren Thornthwaite published Internal Migration in the United States. The Social Science

Research Council’s interest in migration clearly predated the Depression, but like other

migration research supported by it, the Thornthwaite work was inspired by the experience of the

early 1930s:

Except in isolated areas population has achieved a remarkable mobility. Fewer than half of the families of the United States are bound to a locality by the ties of home ownership, and the automobile has destroyed all respect for distance. But since the beginning of the depression, migration has assumed the form of bewildered aimlessness (Thornthwaite, p. 3).

During the 1920s, city economies were booming and significant rural to urban migration

occurred. Thornthwaite notes that the “disappearance ... of the prospects for enlarged economic

opportunity ... aroused an interest in migration, and has led to this present study” (p. 3). He goes

on to lament the meager migration data available for study: “Quantitative data dealing

specifically with the phenomenon of internal migration are practically nonexistent” (p. 4).

Thornthwaite, a geographer, used then available lifetime-migration data from various

censuses to develop a series of maps indicating what he called the “birth-residence index of

population movement,” which was net lifetime migration for each state. One of his conclusions

was that “the greatest single movement in the entire history of the country, one of the greatest in

the world, has been the migration to California in the last decade” (i.e., 1920-1930)

(Thornthwaite, 1934, p. 18). He attributes this migration to the retirement of people in the Corn

Belt and Wheat Belt who became wealthy during World War I. He offers an opinion that, in the

16

light of history, is curious: “Since the movement is abnormal in most respects, it is inconceivable

that it will continue” (p. 18).

Because the 1930 Census did not cover the severe depression years of the early 1930s,

Thornthwaite had to seek other means by which to measure migration over this period. He made

innovative use of school census data to develop estimates of county populations, and he used

these figures in combination with vital rates to measure net migration. Moreover, from the same

school census data, he was able to identify “routes of intrastate migration.” This is an example of

a migration scholar going beyond the obvious data sources to develop new data on migration and

new insights into migration phenomena, such as year-to-year changes in migration.

During the Depression a considerable flurry of additional migration research occurred.

One of the major efforts of this period was the work of Carter Goodrich (University of

Pennsylvania) and his colleagues, who, after producing several interim reports, published

Migration and Economic Opportunity in 1936. The charge of the “Study of Population

Redistribution” was “to consider what movements of population within the United States might

be necessary and desirable, and what part, if any, the Government should take in encouraging or

guiding them” (Goodrich, et al., 1936, p.v). The emphasis on policy and possible direct

government intervention is noteworthy. The study became one of the earliest, perhaps the

earliest, comprehensive surveys of the existing literature on internal migration, but with the goal

of informing public policy. As Goodrich put it, the commission of the Study of Population

Redistribution was “to make a reconnaissance of the field of internal migration in the hope of

discovering bases for the determination of public policy” (p. v).

Goodrich, et al. (pgs. 4-5) addressed two main questions clearly shaped by the experience

of the Depression: “If men and women are out of work, can they hope to gain employment by

17

migration? If their living is precarious, would they be more secure in other locations?” The first

question essentially asks if the unemployed would be more likely to find a job if they migrated.

The second, closely related question asks whether the working poor would be better off if they

migrated.

In answering these questions, the Study Group devotes approximately 30 percent of the

manuscript to an assessment of the regional distribution of wealth and employment in the United

States. Four regions were identified as precarious: (1) the Southern Appalachian Coal Plateaus;

(2) the Old Cotton Belt; (3) the Cutover Region of the Great Lakes (upper portions of Minnesota,

Wisconsin, and Michigan); and (4) the Great Plains. The analysis yields three conclusions:

1. “There can be little doubt that the chances of future livelihood are in general better in the North and West than in the South” (Goodrich, 499).

2. “The chances appear distinctly better among certain of the ‘center’ activities than in the ‘field’ occupations of agriculture and the other extractive industries” (Goodrich, 500).

3. “The chances of employment appear better in a relatively small number of urban and industrial districts – and particularly in their expanding peripheries – than in remote towns and in rural areas” (Goodrich, 500).

It is interesting to note, however, that the study found that the Depression did not cause any

remarkable change in the regional distribution of American wealth. The relatively poor regions

were relatively poor before the Depression.

As a discussion of migration policy, Migration and Economic Opportunity presents a

wealth of information on early efforts to control migration in the United States, Russia,

Germany, and Great Britain. As a survey of migration research, it comes nowhere near the value

provided just two years later by a Yale University migration scholar, Dorothy Swain Thomas in

her Research Memorandum of Migration Differentials. Before we consider the Thomas work in

18

more detail, two observations made by the Goodrich team and elaborated on by Thomas are

worth noting.

The first observation deals with age composition. With respect to rural-to-urban

migration, Goodrich, et al. (p. 691) note that the age composition of the urban in-migration

differed according to the “character of the city.” However, they did not have in mind large cities

versus small, high-income cities versus cities with lower average incomes, or any factors along

these lines. Rather, they were focused on Detroit during the period 1910-1920 compared to 1920-

1930: “Between 1910 and 1920 there was a significant net movement into Detroit of males of all

ages up to sixty-five, but between 1920 and 1930 the age limit of immigrants was about forty-

five years” (Goodrich, et al, p. 691). They attribute the change to a drop of 20 years in the

retirement age of industrial workers. After comparing migrant age composition for several cities,

the Study Group concluded that “the appeal of the city has been strongest, and its opportunities

have been greatest, for the younger people” (p. 693). This age selection of rural-to-urban

migration was a direct follow-up of the earlier work of Thornthwaite.

The second observation concerns sex composition. Again in terms of rural-to-urban

migration, Goodrich, et al. (p. 691) note young people predominated and that young females

exceeded young males. Nevertheless, during the 1910-1920 period the sex composition of in-

migrants to Detroit strongly favored males, but during the 1920-1930 period overall sex

composition was very similar due to a significant drop in the number of male in-migrants

(especially ages 20-39) and an appreciable rise in the number of female in-migrants (especially

15-24). The overall sex ratio of in-migrants fell from 147 (males per 100 females) to 100. That

for 20 to 24 year olds fell from 170 to 117. The study group notes that relatively fewer of the

recent arrivals are married and that their freedom from “family responsibility” accounts for their

19

mobility. Whatever changing marriage rates and birth rates may have had to do with the change,

it seems far more likely that women were replacing men on the assembly lines, and thus female

employment opportunities were growing significantly in Detroit at that time. However, the

reference to “family responsibility” is noteworthy.

Let us turn now to the work of Dorothy Thomas, whose 1938 Research Memorandum on

Migration Differentials followed closely on the heels of the work of Carter Goodrich of the

University of Pennsylvania. In her research memorandum, Thomas focuses on “migration

differentials.” In her words, “the problem of migration differentials is essentially the same as the

problem of selective migration, for the very question of whether persons who migrate are

differentiated from those who do not implies a process of selection” (D.S. Thomas, 1938; p. 4).

The differentials for which she critically summarizes the literature include: age, sex, family

status, physical and mental health, intelligence and education, occupation, and differentials in

motivation and assimilation. Her study covers the demographic structure context in which the

impetus to migrate occurs, and as such, complements and rounds out Ravenstein’s (1885, 1889)

attention to the geographical aspects that condition the migration response to spatial utility

differentials. In two appendices noted above Thomas provides annotated bibliographies that

contain 119 American and English contributions and 72 German contributions to the migration

literature. Her entries, which may have been influenced by her discipline (sociology) suggest that

prior to her writing in the 1930s, migration research was primarily in the domain of sociology,

which is a valid conclusion in any case.

Following Ravenstein’s (1885, 1889) descriptive and inferential tone, Thomas (1938)

states that her analysis of the selective aspects of migration “does not, however, imply anything

at all about the extent to which the selected qualities are innate predispositions or

20

environmentally modified characteristics” (D.S. Thomas, 1938; pp. 4-5). Like Ravenstein (1885,

1889), D.S Thomas (1938) viewed the impetus, or motivation, to migrate as related to the desire

by migrants to improve their lives.

The analysis … has suggested that migration may be viewed as a sort of goal-seeking, in which economic motives are involved (the migrant attempts, or hopes, to better his economic condition), but in which hedonistic motives also play an important role (the migrant attempts to escape the monotony of a limited environment, to exercise his right of choosing a new environment, and hopes, through migration, to attain and enjoy a richer environment). It also suggests the possibility that some migrations are, or appear to be, unmotivated, random movements.

Defining differentials in connection with these problems requires much more than meticulous application of statistical methods. Data obtained as administrative by-products are, in the main, unsuitable for throwing much light on these problems. The behavior of the migrants must be observed before and after migration; the migrants’ ‘own stories’ must be obtained; the environmental setting and the conditions of life in the communities of origin and destination must be described. This does not mean that statistical methods are inapplicable; on the contrary, they will be essential in the initial process of sampling and in the final evaluation of the observed differentials (D.S. Thomas, 1938; pp. 141-42).

Thomas (1938) reveals a rich understanding of the migration process. She recognizes the

role of spatial variations in utility in creating the impetus for migration and that the differentials

in response to such variations are conditioned by the demographic structure of the population. As

will be demonstrated in the selective review of the studies that she treats in her memorandum,

she also has an appreciation of the role of geographical structure in conditioning the migration

response. Moreover, she treats a multiplicity of important issues concerning data quality and the

strengths and weaknesses of alternative methodological approaches in a penetrating and valuable

manner. We now turn to a selective treatment of the studies reviewed in her memorandum.

AGE DIFFERENTIALS

The one generalization about migration differentials which can be considered definitely

established, although even this one cannot be stated precisely, is the following: there is an excess

21

of adolescents, and young adults among migrants particularly migrants from rural areas to towns,

compared with the non-migrating or the general population (D.S. Thomas, 1938; p. 11).

There were two general approaches to estimating the age differentials. The indirect

approach has several variants. The basic idea is to obtain two consecutive census enumerations

and use these to compute for particular areas the total population change between the census

enumeration dates. Next, one relies either on vital statistics on births and deaths for the particular

areas, or on the application of lifetables, to get an estimate of the net natural increase component

of total population change. The residual component is an estimate of net migration. Tabulation of

such residuals by age is purported to yield estimates of the differentials. Early studies using these

indirect methods by Hill (1925), Beck (1934), Hamilton (1934), Thornthwaite (1934), Dorn and

Lorimer (1936), and Hart (1921) find maximum migration rates in the 10-19 and 20-29 age

groups.

In addition to the usual issues regarding the accuracy of the enumerated population

counts and the vital statistics, and the representativeness of the lifetable values to the population

of the particular area under study, additional problems are pointed out by Thomas (1938). The

key one is that these procedures produce an age profile of the migration balance (i.e., of net

migration). This profile does not necessarily reflect the age composition of the migrants

themselves because in- and out-migrants may potentially cancel one another to different degrees

across age groups.

Another indirect approach is to look at concentrations and deficiencies in certain age

groups of populations known to have increased or lost by migration. For example, assuming it is

known that rural areas are losing migrants and urban areas are gaining, then peaks in the urban

distribution corresponding to troughs in the rural distribution probably are associated with

22

migration. The age composition of these peaks and troughs suggest the age composition of

migration.

This approach has been used in the context of African-American migration and rural-to-

urban migration because, in these cases, the necessary information was available to make

accurate assumptions regarding the sending and receiving area populations, as required by the

approach. The African-American studies include Frazier (1932), Kennedy (1930), Kiser (1932),

and Thompson (1935). The rural-to-urban ones are authored by Beck (1934), Clark and Roberts

(1936), Dorn (1934), Gee and Corson (1929), Hamilton (1934), Landis (1936), Lively and Foott

(1937), McKenzie (1933), Truesdell (1926), and E.C. Young (1924).

Thomas summarizes the findings of these studies in the following way.

These findings show a considerable degree of consistency. Urban-rural comparisons indicate urban excesses for quinquennial age groups from about 20-24 to 40-44, the greatest excesses occurring in the earlier groups. Northern Negro groups (known to have increased rapidly by migration) show excesses of about the same age groups in comparison with native whites in the same area and with Negroes in the South. These comparisons, then, indicate that a continuing process of migration tends to produce an excess of persons of ages 20-44 in gaining areas and deficiencies of the same group in losing areas (D.S. Thomas, 1938; p. 19).

Thomas concludes that the main difficulty of this indirect approach is that

It is even more difficult, however, to make reasonable inferences as to the age of migrants from comparisons of this sort than from migration balances, for these total distributions are affected by accumulations of survivors of past migration balances and by temporal variations in the rate of migration, as well as by differentials in fertility and mortality (D.S. Thomas, 1938; pp. 19 and 23).

In contrast to indirect approaches, the direct approach directly surveys the migrants

themselves or their families or neighbors. In some cases, the high school graduation records are

used. Many of these surveys were performed in conjunction with particular case studies; and

others were based on government population censuses and registration systems in which

migration and age are recorded in a systematic way for the population.

23

In absence of any system of registration of migrants at the time of migration, American and English investigators have had to rely upon reports from migrants themselves or from members of their families or from their neighbors in attempts to determine directly the age distribution of migrants at migration. The investigations where this approach has been employed are handicapped by inadequate and often unreliable data (D.S. Thomas, 1938; p. 24).

Case studies employing this direct approach include Beck and Lively (1931), Eugenics

Survey of Vermont (1931), Garey (1936), Gee and Corson (1936), Hamilton (1936), Hill (1925),

Kiser (1932), Lively and Foott (1937), Smick and Yoder (1929), E.C. Young (1924), and

Zimmerman (1926). Thomas assesses these studies as having “very little consistency in methods

or results. The samples are all small, the time period covered is often wide, the definitions of

migrants are various, and the tabulated age classes are not comparable from one study to

another” (D.S. Thomas, 1938; p. 25).

The Swedish census of 1930 reports data on resident population of communities (about

2,500) by age and sex and migration status for the five preceding years. Notwithstanding the

typical issues of repeat and return migration affecting such figures and other circumstances

tending to impart a downward bias to migration figures,3 Thomas (1938, p. 40) concludes that

“when recent Swedish migrants are isolated from other population groups, in the census of 1930,

they are found to be highly concentrated in the early adult ages.” These age composition patterns

can be readily seen on tables (particularly Table 3) and charts presented by Thomas (1938) that

report the Swedish census figures.

Thomas observes that “the Swedish census of 1930 affords a unique opportunity for

analyzing age selection…” (1938; p. 26). It is interesting to note that tabulations of migration

status were unavailable in the U.S. decennial censuses until the 1940 census in which data on

residence five years ago were presented without tabulation by age. In the 1950 census, residence

one year ago (versus five years ago because of World War II) was tabulated by age. Starting in

24

the 1960 decennial census, data were presented on residence five years ago by age. Prior to this,

the U.S. decennial censuses for 1850 through 1930 presented only lifetime migration figures for

the total population (i.e., place of birth tabulated by place of residence on the census enumeration

date).

Thomas (1938) presents two sets of migration data with age detail based on continuous

population registers “where the age of every person entering or leaving an area is recorded at the

time of migration” (p. 41). One of these existed in Amsterdam and the other in Germany.

Although the continuous register system of population accounting is used in a number of countries, data on age at migration are seldom tabulated. The only satisfactory tabulations for recent years are those published by the Statistical Bureau of the City of Amsterdam (D.S. Thomas, 1938; p. 42).

In her Table 4 Thomas tabulated the in- and out-migration rates by age-sex class for Amsterdam

during six periods. This is the first time series detail of which we are aware in such studies. Her

materials demonstrate the strong age selectivity of migration to and from Amsterdam during the

periods of time covered.

In commenting on age differentials in general, Thomas (1938) argues that age selection

may vary by (i) characteristics of origin and destination, (ii) distance of migration, (iii) family

status of migrants, and (iv) phase of the business cycle. She notes that a Dutch study begins this

line of inquiry based on the Amsterdam population register data and advocates more research

along these lines. Note that Thomas has gone well beyond Goodrich, et al. in that she has

identified several causal factors underlying migration age differentials. She has the beginnings of

a theory consisting of several hypotheses.

SEX DIFFERENTIALS

The approaches to analyzing sex composition of migration include both indirect and

direct approaches similar to those discussed in the context of age composition. Indirect

25

approaches appear in Dorn (1934), Hamilton (1934), Hill (1925), and Truesdell (1926). The

findings indicate selection in favor of females and are therefore consistent with Ravenstein’s law

in this regard.

The use of field surveys as a direct approach underlies a number of studies including

Eugenics Survey of Vermont (1931), Gee (1933), Gee and Corson (1929), Gee and Runk (1931),

Gist and Clark (1933), Lively and Foott (1937), Reuss (1937), Smick and Yoder (1929), O.E.

Young (1930), and Zimmerman (1926). These studies have the problems noted by Thomas

above in conjunction with survey-based age selectivity findings.

The Swedish census of 1930 and the Amsterdam register data are both direct sources and

as D.S. Thomas (1938) observes “the Swedish and Dutch sources again afford greater

possibilities of analyzing the extent and nature of sex-selective migration than do the American

or English sources” (p. 60). Thomas provides tables for Sweden and Amsterdam showing the

ratio of males per 100 females for migrants and other population components. In Thomas’ words

“the inescapable inference from these [Swedish data] is that migration to town and cities as a

whole, and particularly recent migration, has been strongly selective of females” (D.S. Thomas,

1938; p. 65). Moreover, “the Amsterdam data on cases of migration show the same general

picture …” (p. 65).

Thomas (1938) sums up the sex differential literature review as follows.

The available data suggest that the ‘rural exodus’ has been sex-selective of females, particularly in the younger adult ages, but that ‘urbanization’ per se is not selective of females. The pull upon the two sexes to the city probably varies with the type of city, the employment opportunities, and the distance from the area of origin. The variation in sex-selective migration during the course of the business cycle has not been studied, nor has the analysis of sex selection in relation to the economic and social structure of the communities of origin and destination been carried through satisfactorily” (D.S. Thomas, 1938; p. 68).

26

Again Thomas has gone beyond then existing studies by developing several underlying causes of

migration sex differentials, but the hypotheses and corresponding theory remain fuzzy and

undeveloped. Even today the theory underlying sex differentials remains undeveloped.

FAMILY STATUS DIFFERENTIALS

Regarding family-status differentials, Thomas’ view was that “obviously no

generalization can be made about the relative incidence of characteristics of lone migrants and

family migrants on the basis of an analysis of a single city, and American, English, and German

studies bearing on this point have been too limited and fragmentary to further enlighten the

situation” (D.S. Thomas, 1938; pp. 86-87). Significant developments along these lines awaited

the theory of family migration developed by Mincer in 1978. Moreover, substantive empirical

work on family migration, required micro and panel data not available until many years after

Thomas prepared her migration-differentials book.

PHYSICAL AND MENTAL HEALTH DIFFERENTIALS

The indirect evidence implied that rural-to-urban migration is selective of the “better

lives,” which seems to support Marshall’s assertion that the physically strongest migrate to cities.

This observation, if valid, has implications for rural stability and community life. The discussion

of this material raises issues concerning the consequences as well as the causes of migration.

INTELLIGENCE DIFFERENTIALS

A number of studies have attempted to determine if migration is selective of more

intelligent, highly educated persons. These include Anderson (1935), Eugenics Survey of

Vermont (1931), Gee and Corson (1929), Gee and Runk (1931) , Leybourne (1937), McCormick

(1933), Raper (1936), Smick and Yoder (1929), O.E. Young (1930), Zimmerman, Duncan, Frey

(1927), and Zimmerman and Smith (1930). All of these studies except Leyborne (1937) and Gee

27

and Corson (1929) find that “the better educated are selected in cityward migration” (D.S.

Thomas, 1938; p. 111); but, “the results are not additive: the samples are small, the techniques

vary, and there is little or no attempt at control of the variables noted above” (D.S. Thomas,

1938; p. 111) that relate both to relative opportunities to obtain or utilize in employment various

degrees of education and to the slippage between educational attainment and knowledge, skills,

and intelligence.

Moreover, an issue is raised concerning the direction of causation regarding the relative

performance of migrants on intelligence tests. Are migrants more intelligent than non-migrants

or do they go to places that have a more stimulating intellectual environment that increases their

relative performance? Thomas reviews in detail a study by Klineberg (1935) which she

convincingly demonstrates is “the outstanding attempt to clarify this situation and to develop

objective methods for evaluating intelligence selection...” (p. 111).

The sample was constructed from high school records of African-American children in

three southern cities: Nashville, Birmingham, and Charleston. These “were examined for

information as to whether those Negro children who had migrated to the North represented in

their school work a superior or inferior group when compared with the total Negro school

population in those cities” (D.S. Thomas, 1938; p. 112). Additional data were collected on scores

on a number of separate intelligence tests administered to children residing in New York with a

northern-born control group and southern-born treatment group.

The idea underlying the study is that if the environment is the key to performance in

school and on these types of tests, instead of a selectivity in migration favoring those who can

perform better in these ways, then two results should be observed. First, the school performance

of those who later in life were stayers and those who were migrants should be insignificantly

28

different. Secondly, the test scores of migrants should increase with length of residence in the

destination.

Klineberg felt that there was ‘no reasonable doubt as to the conclusion of this study. As far as the results go, they show quite definitely that the superiority of the northern over the southern Negroes, and the tendency of northern Negroes to approximate the scores of the Whites, are due to factors in the environment, and not to selective migration.’

‘There is, in fact, no evidence whatever in favor of selective migration’ (D.S. Thomas, 1938; p. 120).

Thomas (1938) reports in detail on another study of this issue completed by Gist and

Clark (1938) focusing on 2,544 persons who were students in 40 rural Kansas high schools

during the academic year 1922-23 and for whom residence in 1935 could be obtained. These

authors conclude that

‘in so far as the tests are indicative of ability ... the superior persons in the sample are tending to migrate more frequently to the cities than those in the inferior or average classes. On the basis of these data two questions logically suggest themselves: (1) Is such qualitative selection characteristic of rural-urban migrations in general? (2) What are the possible consequences of such selectivity to community life?’ (D.S. Thomas, 1938; p. 24).

A good example of Thomas’ careful analysis and critique of the literature that she

reviewed is illustrated by her comments on this study:

Two of the points raised in connection with Klineberg apply equally to Gist and Clark: they place undue reliance on small differences in averages, although they do present the distributions, and they take no account of the variability in the measuring instrument. Furthermore, they fail to control time of migration or to indicate whether settlement was permanent, or to differentiate the successive stages between original residence and residence at the time of investigation. Also, a group of high school students ‘who had already undergone a selective process in the education system’ was not the most appropriate sample for testing selective migration. The authors recognized most of these points and, while hesitating to make ‘broad generalizations,’ concluded that migration was selective of the more intelligent (D.S. Thomas, 1938; p. 125).

29

OCCUPATIONAL DIFFERENTIALS

The issue of interest to Thomas (1938) was whether or not migration and occupational

change (particularly upgrading) tend to occur together and not whether migration is selective of

particular occupations. Thomas (1938) reviews a study by Littmarch published in Swedish that

focuses on the migration patterns of certain migrant agricultural laborers in a particular region of

Sweden over the period 1925-29. Interestingly, Littmarch was a pastor in the Swedish Church.

His Archbishop proposed the study out of concern over the “high frequency of migration of

agricultural workers in the Malar Valley” (D.S. Thomas, 1938; p. 130). The study’s importance

lies more in the area of methodology than in its generalizable findings. One thousand family

annual migration histories were constructed for the period 1925-29 based on population registry

data. The study demonstrates the feasibility of constructing such histories from population

register data. Moreover, this study also points out the role of migration as a means to improve

one’s life. Thomas (1938) again emphasizes this behavioral aspect of the impetus or motivation

for migration.

DIFFERENTIALS IN MOTIVATION AND ASSIMILATION

Thomas (1938) considered in this section a study by Kiser (1932) of the migrations of

African-Americans from St. Helena Island off the coast of South Carolina during the period

immediately following the Civil War to 1928. In her view, this study makes a noteworthy

beginning in analyzing the place and people characteristics surrounding migration in the context

of migration as spatial search for improved levels of well-being. In addition, the study exhibits

strengths in the care with which data were collected and analyzed.

The study began with a complete enumeration of the island population, by household, in

1928 with records made of the age, sex, marital status, and date of departure and destination of

30

all those missing from the household. The attention to accuracy is attested to by the following

description of part of the data collection procedure:

In order to check the enumeration, all the cards [one for each resident and one for each migrant] were taken by the writer to Washington and compared in detail with the original schedules of the 1920 United States Census. This procedure, it was thought, would not only afford a more complete enumeration of residents but would also make it possible to secure a full list of individuals who left home after 1920. The Island enumeration of missing members was known to be unsatisfactory, especially insofar as the migration of whole families was concerned.... [This check disclosed a number of names representing] either non-reported migrants, non-reported residents, or deceased persons whose deaths had not been recorded. In order to follow up and clarify such information for specific individuals, the writer returned to St. Helena and consulted the native enumerators. In practically all cases these Islanders or others were able to state readily whether the individuals concerned were still living in St. Helena, had migrated, or had died (D.S. Thomas, 1938l p. 143).

The next step was to locate migrants in several destinations including Savannah,

Charleston, Boston, Philadelphia, and New York. The New York group was by far the largest

(233 persons) and the survey of these individuals took 9 months. Various information was

collected on the migrants and on the stayers on the island including age, sex, marital status, time

of migration, family composition, personality patterns (desires and motives), racial attitudes,

income, and education.

The geographic path of migration was determined in terms of direct migration from the

island to New York (or the final destination) or a sequence of moves to the final destination.

Return migration also was recorded. Distance and direction of migration and the factors

facilitating or impeding migration were analyzed. The characteristics of the origin and the

destination, and changes over time, were related to the impetus to migrate from the origin and the

process of assimilation in the destination. The author draws a number of conclusions from the

study that indicate that “unsatisfactory Negro-white relations and farm tenancy” and “poor

31

educational facilities offered to Negroes in the South" (D.S. Thomas, 1938; p. 157) do not appear

to be important factors in the migration decision of the Islanders.

The study of migration from St. Helena indicates, furthermore, that the underlying causes for departure are practically the same as those which are well known to be responsible for the general drift of young people–white or colored–from the farms of various sections of the country. Increased industrialization has given rise to large cities and the corresponding development in means of communication and transportation has served to break the isolation of rural areas. The dissatisfaction of young people with the conditions at home is intensified by the knowledge that friends are making good in the cities. After hearing stories of the economic and recreational opportunities available in the cities, the uneventful life on the farm seems dull to many.

General dissatisfaction with economic conditions or monotony of life are often predisposing causes but the immediate occasions for migration are usually specific and concrete incidents.

In some cases, the immediate cause itself may give rise to the motive for migration and the departure is made immediately. Ordinarily, however, it appears that Islanders have been rather slow to make up their minds to leave home. There were relatively few immediate permanent departures after the two storms and after the infestation of the boll weevil (D.S. Thomas, 1938; p. 157).

It is impossible to state categorically whether or not the Islanders have benefitted themselves by leaving home.

In broad perspective, the results of the migration may be summarized according to those which appear to be losses and those which seem to be gains. In the process of adjustment, at least, the migrant loses the independence which was his in St. Helena by virtue of farm ownership and self-direction in work. He leaves an assured livelihood for one far more precarious. He is deprived of the advantages of close group and family life. He leaves the open country and enters neighborhoods where congestion and bad housing are encountered.

On the other hand, there are benefits such as the money wage and the chance to gain a higher standard of living.... In pursuit of leisure-time activities, the individual is freed from the limitation of a rural environment.... Such liberation ... enables the migrant to develop his interests in vocational and cultural fields. In the northern cities, especially, there are increased opportunities for education (Thomas, 1938; p. 158).

D.S. Thomas’s (1938) work in the memorandum contributed significantly to advancing

the understanding of migration phenomena beyond the earlier work of Ravenstein (1885, 1889),

32

especially by adding the demographic component to the migration response structure. At the

same time, she reinforces and extends the geographical component to the migration response

structure and further emphasizes the role of spatial utility variations in motivating migration. Her

attention to issues of data accuracy and methodological sophistication are hallmarks of her

dedication to the use of the scientific method in pursuit of the study of the causes and

consequences of migration.

However, she also provided a compelling case for the recognition of the subtleties of and

interactions among migration motivations and geographical (and especially) personal

characteristics through which migration motivations are mediated into acts of migration. Her

work, along with Ravenstein's, is squarely in the vein of thinking and research that culminates in

the human capital model of migration that was introduced formally in the early 1960s.

THE DEVELOPMENT OF MORE FORMAL CAUSAL MODELS

The research discussed to this point dealing with the determinants of migration employed

neither formal models of the migration decision process nor formal statistical techniques. Rather,

these studies were mainly descriptive in style and inferential in tone; however, as noted above,

their findings were in certain instances powerful, insightful, and anticipatory of later migration

research.

During the 1920s and especially during the 1930s, considerable interest was directed at

migration phenomena not only in the United States, but also in the United Kingdom, and the

nature of this research began to take on a decidedly more formal tone, although descriptive

analysis remained the primary research methodology. For example, Makower, Marschak, and

Robinson published a series of papers in Oxford Economic Papers that was extremely insightful.

33

These authors used data from the Oxford Employment Exchange, which reported the number of

persons who entered the unemployment insurance system in various specific places other than

Oxford and who were residing in Oxford in 1936.4

Two “incentives” to migrate received particular attention in the work of Makower,

Marschak, and Robinson: unemployment differentials and distance. They defined the “relative

unemployment discrepancy” as “the ratio of the difference between the unemployment rate in the

county (or Division) and the unemployment rate in the whole country, to the unemployment rate

in the whole country” (1939, p. 81). They argued further that "the greater the mean deviation of

relative unemployment rates...the greater would be the total amount of migration measured as a

percentage of the population of the whole country” (1939, p. 81). Makower, Marschak, and

Robinson went on to show that “there was a very clear correspondence between variations in the

relative unemployment of the county and variations in the gains and losses by migration” (1939,

p. 82).5

Although Makower, Marschak, and Robinson are not credited with the development of

the gravity law of spatial interaction, they clearly laid out the same concepts: “Quite a close

relationship was found between discrepancies in unemployment rates and migration of labour

where allowance was made for the size of the insured population and the distance over which

migrants had to travel” (1938, p. 118). These authors even computed a distance elasticity of

migration, which they called the “coefficient of spatial friction”: “an increase of distance by 1

per cent reduces migration by from 1.6 per cent. to 2.1 per cent.” (1938, p. 106).

Interestingly, Makower, Marschak, and Robinson started with individual records that

indicated sex, age within broad age class (14-15, 16-18, 19-21, 21-65), industry of employment

before and after the move, and county of origin. Today such microdata no doubt would be

34

analyzed themselves, but these investigators aggregated the data by county of origin, presumably

because microdata were rare, techniques for their analysis were not yet developed, and

consequently the econometric analyses of microdata were also very rare.

Makower, Marschak, and Robinson (1939) also considered the time lags inherent in

migration, as well as the relationship between aggregate economic activity and migration. They

concluded that the lag between “incentive to move and migration” was short, certainly not more

than 18 months, but more likely about six months for Great Britain as a whole, and between zero

and six months for migration to Oxford. Moreover, their analysis indicates that mobility declined

during slumps and rose during recoveries and that moves over short distances were less sensitive

to national economic conditions than moves over long distances.

GRAVITY MODEL OF SPATIAL INTERACTION

We now turn to a discussion of the application of the unmodified gravity model to the

analysis of migration. Although the gravity model idea predates Ravenstein's work, its

application to the analysis of migration occurred thereafter. In the context of Ravenstein's

conceptual framework, the gravity model focuses on the structure of the migration response in

terms of geographic considerations (i.e., the relative sizes and locations of places).

To elucidate this contrast, consider the description of the gravity model given by H.C.

Carey (1858-59):

‘Man, the molecule of society, is the subject of Social Science ...’ ‘The great law of Molecular Gravitation [is] the indispensable condition of the existence of the being known as man ...’ ‘The greater the number collected in a given space, the greater is the attractive force that is there exerted ...’ ‘Gravitation is here, as everywhere, in the direct ratio of the mass, and the inverse one of distance’ (as quoted in Carrothers, 1956, p. 94).

35

Although several of Ravenstein's laws of the structure of migration response are

consistent with the gravity model, it apparently was not until the 1920s that the concept was

explicitly applied to migration. In a study of rural migration in New York, E. C. Young (1924)

hypothesized that origin-destination flow volumes were related to one another as predicted by

the gravity model.

Twenty years later, during the 1940s, J.Q. Stewart (1941, 1947) and G.K. Zipf (1942,

1946) elaborated the model and applied it to migration and other human spatial interactions. This

work involved determining to what extent the Newtonian formula required modification when

applied to human settings. As described by Carrothers (1956), Stewart, Zipf, and others analyzed

the deviation from unity of the exponents on the two population masses, whose product formed

the numerator; and the extent to which the exponent on distance separating the two areas,

specified in the denominator of the Newtonian formula, was different in value from two. Stouffer

(1940) examined the role of “intervening opportunities” on the effect of distance.

Stewart ‘s work has a distinctly physical-mechanistic flavor to it. In one paper (Stewart,

1950), he specified “molecular weights” for the population terms to capture the capacity of

individuals in the two areas to interact. One of his papers, “The Development of Social Physics”

appeared in the American Journal of Physics (Stewart, 1950) ; and another entitled “A Basis for

Social Physics” was published in the Impact of Science on Society (Stewart, 1952).

As Carrothers has pointed out, the strictly Newtonian formulation of the gravity law of

spatial human interaction is based on the behavior of molecules:

The behavior of molecules, individually, is not normally predictable, but in large numbers their behavior is predictable on the basis of mathematical probability. Similarly, while it may not be possible to describe the actions and reactions of the individual human in mathematical terms, it is quite conceivable that interactions of groups of people may be described this way (Carrothers, 1956; p. 99).

36

He continues in a more cautionary tone.

But it is important to keep in mind that, although the use of analogy in developing a concept may be attractive, it may defeat its purpose if strict and inflexible adherence is insisted upon. In this case, a fundamental difficulty arises from the different nature of the two basic units of measure involved: the individual human being can make decisions with respect to his actions, while the individual molecule (presumably) cannot. This does not imply that interaction of humans in large numbers cannot be described mathematically, but it does mean that the threshold where the power of individual decision-making critically affects the results must be determined before the concepts can be broadly applied in practice (Carrothers, 1956; p. 99).

This statement reveals the essential difference between the gravity model of migration

and the behavioral models of Ravenstein and others. Free individuals choose to migrate.

Ravenstein and those who developed the modified gravity and human capital models postulate

that this choice is conditional on spatial utility (economic opportunity) variations. It should be

noted that the flavor of Zipf's (1942, 1946) use of gravity concepts is more in tune with this

behavioral approach as evidenced by his reference to the principle of least effort whereby human

choice is related to economizing behavior.

The gravity model, as applied to migration, is not salvaged in a more recent attempt by

Niedercorn and Bechdolt (1967) to derive it within a constrained utility optimization framework

consistent with modern consumer theory. The reason is that utility in their approach is obtained

directly from making “trips.” Clearly, few if any individuals, derive utility from the act of

migrating. Instead migration consumes resources and is therefore costly. The benefits that

warrant incurring such costs stem from the arbitrage of spatial utility differentials through the act

of migration. The act of migration is therefore an investment act which yields utility not in and of

itself, but through the higher utility attained as a result of changing locations.

37

THE 1950s AND 1960s

The premier migration study conducted during the 1950s and early 1960s was also

carried out at the University of Pennsylvania. The three volume work, Population Redistribution

and Economic Growth United States, 1870-1950, had a number of widely-recognized

participants, including C.P. Brainerd, R.A. Easterlin, H.T. Eldridge, S. Kuznets, E.S. Lee, A.R.

Miller, and D.S. Thomas.6 The study, which was directed by Kuznets and Thomas, is a detailed,

state-level analysis that like most earlier migration research is descriptive in style and inferential

in tone. The basic notion that underlies the study is that “the distribution of a country's

population at any given time may be viewed as a rough adjustment to the distribution of

economic opportunities” (Kuznets and Thomas, 1957, p. 2). In this sense, the study is a logical

extension of Goodrich's earlier work, which had the same type of theme. One of the main

hypotheses of the study is that technological progress critically affected the sectoral and spatial

distribution of economic activity. Moreover, the rapidity with which the changes occurred

allowed little room for the vital processes of births and deaths to play an important role in

adjusting population to altered economic opportunities. Migration had to provide the main

impetus. Another important aspect of this study was its emphasis on the selective nature of

migration, especially long-distance migration and migration between dissimilar places.

MODIFIED GRAVITY MODELS

During the 1960s, the main thrust of migration research began to take on a

decidedly more formal tone that has continued to the present. Most of the research was not

formal in a theoretical sense, but rather intuitively generated hypotheses were at first tested

formally in an econometric sense with aggregate data, typically but not always with place-to-

place migration data. These aggregate models of migration frequently were specified in the

38

context of modified gravity models. The models are “gravity type” in that migration is

hypothesized to be directly related to the size of relevant origin and destination populations, and

inversely related to distance. The models are "modified" in the sense that the variables of the

basic gravity model are given behavioral content, and additional variables that are expected to

importantly influence the decision to migrate are included in the estimated relationship.

Modified gravity models that became common in the migration literature beginning

during the 1960s add several additional variables to those of the basic gravity model. Thus, we

now commonly find studies of place-to-place migration that take the following form:

lnMij = lnβ0 + β1lnDij + β2lnPi + β3lnPj + β4lnYi + β5lnYj

+ inm

n n Xln1∑ =α + jn

m

n n Xln1∑ =γ + eij , (1)

where the Y terms refer to income. Other variables that are commonly included (as reflected in

the “X” terms above) are unemployment rates, degree of urbanization, various climatological

amenity variables, various measures of public expenditures and/or taxes, and many other factors.

For certain variables, some models contain only origin characteristics, such as median age or

median number of years of schooling, which are meant as proxies for the characteristics of the

population from which the migrants are drawn. Modified gravity models hold an important place

in the migration literature because their formulators tried to incorporate behavioral content in the

context of the gravity-model approach. These efforts subsequently led to formal models of the

migration decision process such as those reflected in many studies that incorporate microdata.

Moreover, such models included a mix of disequilibrium and equilibrium notions that anticipated

the later, more rigorous development of the equilibrium hypothesis as related to migration.

The connection between modified gravity models and the migration decision process has

not always been tight. The dependent variable in modified gravity models is meant to proxy the

39

probability of moving from i to j. However, the denominator of the dependent variable frequently

has been population measured at the beginning or end of the migration interval. Such a measure

falls short of reflecting the population at risk to make a move from i to j. For example,

beginning-of-period population includes persons who die during the period over which migration

is measured, as well as those who emigrate from the country, and who are thus not available to

be counted as migrants. The end-of-period measure includes in-migrants who were not at risk to

be out-migrants from the area and also introduces simultaneity between migration and the

population measure.

Modified gravity models are frequently estimated in double logarithmic form,

presumably because this functional form yields reasonably good fits and the coefficients

obtained from it can be directly interpreted as elasticities of migration's response to changes in

the various independent variables of the estimated models. Goss and Chang (1983) show

considerable differences in the estimated coefficients of a migration model depending upon the

precise nature of the functional form that is specified. Moreover, common use of the double-

logarithmic functional form to estimate modified gravity models has led to a criticism by T.P.

Schultz (1982), who argues for the adoption of nonlinear maximum likelihood logit methods

over the double-log form of the model. In part his argument hinges on the geographic size of the

regions for which migration is measured. If all regions had the same population and land area,

migration and nonmigration probabilities would reflect the costs and benefits of the various

locational choices. However, the regions of any country differ greatly in population and land

area. A larger share of all moves will tend to occur within the boundaries of larger regions.

Consequently, more nonmigration will appear to exist for such regions. The result is that

nonmigration is spuriously correlated with origin population size and land area.

40

In the polychotomous logistic model the migration probabilities are expressed as ratios,

and the probability of not migrating is used in the denominator of the expression to normalize the

“flows.” That is, the dependent variable is ln[mij/(1 - mij)], which is sometimes called the

logarithm of the odds-ratio. Here mij refers to the probability of migration from i to j, and is thus

measured as Mij, or the actual number of movers from i to j divided by the population at risk to

migrate from i. The model can be estimated in one of two-ways that make sense in the migration

context. First, again assuming the double-log form of the model, the log of the ratio of various

destination-to-origin characteristics can be used. This approach, referred to as “uniform

symmetric,” implies that coefficients on variables for corresponding origin and destination

characteristics are the same except for sign. Second, a two-step decision process can be assumed

in which the decision maker first decides whether to migrate based on origin characteristics and

then decides where to migrate based on destination characteristics and perhaps other variables

(such as distance) that link the areas. In this case, origin and destination characteristics are

introduced separately. In an analogous fashion, nested decision trees could be constructed for

other levels of the decision process (e.g., to move, where to move, what house to select, etc.).

Some dissatisfaction has been expressed with the notion that an individual can decide whether to

move independently of where he/she might move. Thus, whether to move and where to move are

seen as joint decisions and not discrete and independent decisions (Linneman and Graves, 1983).

Schultz sees the standard gravity approach as inefficient because it fails to incorporate

information on the relative frequency of nonmigration (1 - mij). He argues, however, that “in the

limit, as the unit of time diminishes over which migration is measured, differences between the

two specifications of the migration model might be expected to diminish” (1982, p. 576). The

reason is that the population at risk to migrate becomes a better measure of the nonmigrating

41

population when the migration interval is very short. In any case, the logit approach provides a

more natural transition from the gravity model to the more behaviorally-grounded modified

gravity model.

MORE RECENT MODELING DEVELOPMENTS

SYSTEMIC MODELS

Both the gravity and the modified gravity models of spatial interaction treat migration

decisions in terms of an origin (i) and a destination (j). Migration between i and j depends on the

characteristics of i and j and the distance between them. In contrast, systemic models of

migration take into account not only the characteristics of an origin and a single destination, but

also the characteristics of all other relevant destinations in the geographical system of regions

under study. This consideration of the system as a whole is a systemic application of the notion

of intervening opportunities (Stouffer, 1940). There are three alternative approaches to systemic

migration modeling: (1) Alonso’s (1978, 1986) systemic model, (2) McFadden’s (1978, 1981)

nested logit model, and (3) the systemwide net migration approach of Greenwood and Hunt

(1984a, 1984b, 1985) and Foot and Milne (1984).

Alonso (1978, 1986) generalizes the modified gravity model to incorporate systemic

effects as shown in Equation (2):

Min = vi1−i

iDα wj

1−jjCβ

tij , (2)

where Mij is migration from origin i to destination j; vi is a function of the characteristics of i or

its population; Di is the “draw” or “pull” of the system upon i and is a function of the relation of i

to the rest of the system, per unit vi; αi is the elasticity of response of migration from i to j to the

system’s “draw” upon i; wj is a function of the characteristics of j or its population; Cj is the

42

“competition” (or “crowding” or “congestion”) related to the potential pool of movers into j and

is a function of the relation of j to the rest of the system, per unit of wj; βj is elasticity of response

to Cj; and tij is the ease of transitioning between i and j (e.g., of moving from i to j). Notice that if

we set Di = Cj = 1 and tij = an inverse function of distance, we have a modified gravity model,

which is a special case of Alonso’s systemic model. It should be noted that although Alonso’s

model was initially interpreted as a theory of movement, Alonso (1986) later indicated that he

preferred to view it not as a theory with specific falsifiable predictions, but as an encompassing

structure for a variety of less general spatial interaction models.

The modified gravity model stated in Equation (1) can be rewritten as a systemic model

as follows:

lnMij = lnβ0 + ijjijDln

1∑ ≠β + β2lnPi + jjij

Pln3∑ ≠

β + β4lnYi + jjijYln

5∑ ≠β

+ inn

m

nXln

1∑ =α + jnij

m

n jn Xln1∑ ∑≠ =γ + eij , (3)

McFadden (1978, 1981) develops the nested logit model with several levels. In a two-

level set up applied to migration, the top level models the unconditional decision to migrate from

or to stay in the origin. The lower level models the choice of destination conditional on the

decision to migrate. The destinations and origins incorporated in McFadden’s nested logit model

include all of the areas specified in the entire geographic system being studied. Consequently,

this approach is a systemic one. Hunt (2000) elaborates the particular structure of a two level

nested logit model of migration showing that it is partially degenerate because the origin is the

only lower-level choice consistent with an upper level choice of staying.

Some confusion is evident about the implications of such an “hierarchical” approach to

modeling decision making. The nested logit approach does not imply that individuals first decide

whether or not to migrate and then decide where to migrate. At the top level, the unconditional

43

choice of whether to migrate or not depends on individual characteristics and on the area

characteristics in both the origin and all destinations. This systemic information is passed to the

top level from the lower level(s) through inclusive values. So it is incorrect to conclude that the

top-level decision of whether or not to migrate incorporates only information on origin

characteristics and does not incorporate information on the area characteristics (e.g., local labor

market conditions, amenities, housing rental prices) in the system of destinations. The inclusive

values pass this systemic information from the lower conditional levels to the upper

unconditional level. So, the nested logit model approach is consistent with the view that the

decisions of whether to move and to where to move are joint decisions and not discrete and

independent decisions (Linneman and Graves, 1983).

A relationship exists between the approach of Alonso and that of McFadden. As Smith

(1991) has formally demonstrated, Alonso’s model can emerge from random multinomial spatial

events that impinge on the desirability of interaction as the asymptotically most probable

interaction structure. This seems to suggest an underlying commonality between the Alonso and

McFadden models. It would be a useful topic for further theoretical research to explore formally

this potential commonality in order to gain perspective on the relationship of Alonso’s model to

the nested logit model.

Several examples are of path breaking applications of microdata are available, including

DaVanzo (1978), who uses microdata to make progress on measuring the effects of

unemployment at the individual level on migration. Nakosteen and Zimmer (1980) investigate

the effects of self-selection at the individual level on the returns to migration using a Heckman

selection model approach. Borjas, et al. (1992) find comparative advantage selection in

individual U.S. domestic migration decisions using a Roy selection approach, as do Hunt and

44

Mueller (2004) for individual migration choices within and between Canada and the U.S. The

latter study also employs a nested logit model that enables the unconditional decision to migrate

to depend on economic and amenity conditions in both the home region and other regions in the

choice set. This avoids the inappropriate implicit assumption, in models that condition on

migrating, that the unconditional choice to migrate or stay is unaffected by the utility features

across all regions.

The systemwide net migration approach has several variants. In Greenwood and Hunt

(1984a), systemic information is supplied indirectly by a zero net migration constraint

implemented as a model closure rule across 171 net migration equations defined for Bureau of

Economic Analysis areas in the United States. In simulations of the model, the zero systemwide

net migration constraint implicitly incorporates systemic information. In later work, Greenwood

and Hunt (1984b, 1985) use a more direct approach. Each region’s net migration equation

incorporates the economic characteristics extant in that particular region and in all other regions.

The parameters on this set of systemic variables are estimated subject to adding up restrictions

across regional net migration equations that insure zero net migration systemwide.

The direct approach can be very consumptive of degrees of freedom which are often in

relatively short supply in time series data settings. The alternative of Foot and Milne (1984) is

similar to the direct approach of Greenwood and Hunt (1984b, 1985), but conserves degrees of

freedom by aggregating the systemic information on economic characteristics in all regions to a

single measure through the use of a distance-weighting scheme.

45

DATA LIMITATIONS AND ADVANCES

Data were clearly a problem that prevented migration research from blossoming. The

availability of only life-time migration data in the U.S. Census was especially limiting. Not until

1940 did the Census include a question on past residence at a fixed date (1935), and not until

1950 did the Census report migration cross-tabulations by age.7 Spatial detail at the county level

required that migration be estimated by the survival method. The resulting net migration data

caused a strong orientation during the 1930s and 1940s toward migration of the rural-farm

population. Although the creation of these net-migration data required some sophistication,

analyses remained descriptive (Bernert, 1944a, 1944b).

During the late 1940s, the Current Population Survey (CPS) began asking place of

residence one year earlier, which resulted in an annual series of U.S. migration data that could be

disaggregated to the four Census Regions. In many respects, the period of descriptive migration

research reached its peak in 1964 with the publication of Henry S. Shryock, Jr.’s Population

Mobility within the United States (Chicago: University of Chicago). This book provided a truly

detailed descriptive analysis of existing CPS data, as well as 1940 and 1950 Census data.

However, even before this book was published, a new day had begun to dawn on migration

research. The first study of which we are aware that yielded usable microdata on migration was

conducted in 1960.

Prior to 1975 virtually all migration research was based on aggregate data.8 In addition to

the problem noted above, modified gravity models were characterized by other problems and

shortcomings frequently associated with the use of such data. Aggregate data often were used to

proxy the characteristics of the population at risk to move, resulting in empirical estimates that

did not reflect accurately the influence of personal characteristics on the decision to migrate.

46

With some notable exceptions (e.g., DaVanzo, 1976b; Kau and Sirmans, 1977), studies of

aggregate migration failed to account for different types of moves, such as primary (or new),

return, and other repeat migration. Aggregate data also concealed differences in the underlying

determinants of migration of various population subgroups, although stratification by age and

race was not uncommon. Such data failed to account for the institutional population, of which

the military was especially important, and they made the study of family migration decisions

difficult. Another problem with modified gravity models was that the variables used to explain

migration often were measured at the end of the migration interval and were thus subject to

simultaneity bias (Greenwood, 1975b). During the 1970s, several simultaneous equations models

were developed to explain the causes and consequences of migration within the same empirical

framework, but for the most part during the last 20 years these models have not been further

developed.

CONCLUSIONS

In this paper we have provided a review of the early literature on internal migration,

beginning with the work of Ravenstein in the 1880s and continuing through the development of

the gravity model and certain developments related to this model. Because we feel that much of

the early work on migration is not widely recognized or appreciated, we have focused particular

attention on Ravenstein and on the premier migration researcher of the early middle years of the

twentieth century, Dorothy Swain Thomas. We make a special effort to relate the ideas of these

and other migration researchers to modern theories of and approaches to migration.

We have argued that the scientific study of internal migration grew along with urban

growth and urbanization, which clearly were fueled by rural-to-urban migration. The depression

47

of the 1930s provided a spark to migration research, and especially to policy-oriented migration

research, because the urban dwellers of the time included large numbers of unemployed who had

moved from rural areas. As new sources of migration data became available, perhaps most

notably beginning with the 1940 U.S. Census, migration research became more accessible, and

scholars from many disciplines joined its study.

A number of important methodological developments also characterized the history of

migration analysis, and these developments often went hand-in-hand with advances in data

availability. Alfred Marshall states that

It is the business of economics, as of almost every other science, to collect facts, to arrange and interpret them, and to draw inferences from them. ‘Observation and description, definition and classification are preparatory activities. But what we desire to reach thereby is a knowledge of the interdependence of economic phenomena’ (Marshall, 1948, p. 29).9

This statement reflects the methodological development of migration research. In the early

period, when census data on lifetime migration were the only data resource available, the

approach was descriptive and interpretative.10 Ravenstein’s (1885, 1889) famous studies are a

prime example of such an approach. The Ravenstein studies are, especially for his time,

extremely detailed descriptive analyses of internal migration. His “laws of migration” are

essentially hypotheses about migration behavior and are based on simple observation, but

certainly his laws indicate that he drew many inferences from his observations. In no sense were

these hypotheses tested in any formal way until many years later. Clearly, during the period of

Ravenstein’s work, and for many years after his work, migration research and migration

methodology were in the “observation and description” phase.

In a sense, the descriptive and interpretative phase of migration research peaked in the

famous work of Kuznets and Thomas sponsored by the American Philosophical Society,

48

Population Redistribution and Economic Growth in the United States, conducted during the

1950s and published in 1960. These authors and their collaborators organized a massive amount

of data on internal U.S. migration. Such descriptive and interpretative research continues to this

day, and in spite of the tremendous current emphasis on formal modeling, remains a highly

valuable migration research methodology. Making intuitive sense of data not only assists our

understanding of various migration phenomena, but it also contributes to the development of

testable hypotheses concerning migration.

The second phase of migration research entailed efforts to develop new and more detailed

data. In this sense, methodologically migration research remained in the observation and

description phase. Lifetime migration measures remained common in reported census data, so

migration researchers sought measures that reflected movements that occurred over more

narrowly defined geographies, for more specific population subgroups, and for various

intercensal time intervals. During this period, indirect methods of measuring migration were

developed, and descriptive migration research advanced in many important respects, especially

with respect to age and sex detail. One of the most important methodological developments

during this phase of migration research was the use of case studies that entailed surveys.

Although these surveys were characterized by many of the same problems that plague

retrospective migration histories today, they were important both for the types of questions asked

and for new insights cast on various migration phenomena. D.S. Thomas (1938) provides a fairly

detailed summary of many of these case studies, including an annotated biography. In an

important sense, these case studies and the underlying surveys they entailed were a precursor of

more recent availability of micro- and panel data, such as the Panel Study of Income Dynamics

(PSID) and the National Longitudinal Surveys (NLS).

49

The third phase of migration research began with the series of studies by Makower,

Marschak, and Robinson (1938, 1939, 1940). These authors provided the results of very simple

regressions. The need to develop the regression estimates by hand severely limited the

sophistication of the regressions that could be estimated, but for the first time hypotheses were

being tested formally. The importance of the basic variables of the gravity model (i.e., distance

and population) was stressed. With these studies, migration researchers took two major

methodological steps forward: (1) They expressed the basics of the gravity model of spatial

interaction, which was to become, in its modified from, a pillar of migration research, especially

for studies of place-to-place migration; and (2) they used regression analysis to test hypotheses

and study migration.

The next phase of migration research awaited the advent of computers. Modified gravity

models that included more than just distance, origin population, and destination population now

became the norm. For many years migration was an important topic of research in disciplines

like sociology and geography. After the publication in 1961 of Schultz's classic paper in the

American Economic Review, soon followed by Becker's paper on investment in human capital

and, in the same special issue of the Journal of Political Economy, Sjaastad's paper on migration

as an investment in human capital, migration research by economists really began to blossom.

The human-capital perspective provided a paradigm that caught the attention of economists and

provided a convenient theoretical framework for their work.

In the early days of its development, the migration literature produced by economists

emphasized the link between income differentials and interregional migration. This emphasis

certainly was understandable given both the emphasis on the human capital perspective and

long-standing theoretical ideas from international and interregional trade that goods and factors,

50

including labor, flow geographically in response to differential prices and rates of return,

respectively. The idea in those days was that migration occurred in a disequilibrium system in

which wage differentials reflected real utility differentials that gave rise to migration. One form

of investment in human capital is migration, presumably from low- to high-income (or wage)

areas, so variables such as origin and destination income or wages provided a major

methodological advance in modified gravity models. Other variables likely to influence inter-

regional migration (such as unemployment rates) also were included in these models. Certain

“amenity” variables, such as various measures of climate, also were introduced.

As a reflection of how far the migration literature as progressed, more recently much

emphasis has been placed on the idea that migration occurs within an equilibrium system in

which wage differentials are compensated and therefore do not reflect opportunities for real

utility gains. The basic idea is that areas with attractive amenities, such as climate in Southern

California, will have wages and rents that reflect these amenity values. If these notions are

correct, migration occurs in response to changes incomes or other lifetime events (e.g.,

retirement) that cause individuals to demand more access to location-specific amenities.

Before 1975, in formal models of migration, surprisingly little emphasis was placed on

employment opportunities as a determinant of migration. This failure occurred in spite of the fact

that Goodrich, et al. (1936) and Kuznets and Thomas (1960) placed great emphasis on job

opportunities as a major driver of interregional migration. A few studies had been conducted that

placed migration in a simultaneous-equation framework with employment change, but these

studies were exceptions. Since about 1975, the importance of employment change has received

much more attention.

51

A second major advance during this phase of migration research and one that was a

natural outgrowth of the emphasis on the importance of job opportunities in migration decisions,

involved the study of the consequences of migration. The consequences were viewed as

aggregate consequences for both the region of origin and the region of destination. How was

regional employment affected by in- and out-migration? Did migration cause an increase or a

decrease in average income in origins and/or in destinations? Such models often were estimated

by means of simultaneous-equation techniques in which both migration and its consequences

were treated as endogenous. Such methodological advances, at least in terms of estimation, could

not have occurred in the absence of powerful computers.

At about the same time that the development of computers was allowing major

methodological advances in the study of aggregate migration data, a new and potentially

revolutionary data source was being developed. The earliest sources of microdata that contained

migration information were beginning to trickle out to researchers. Before 1975 few studies even

attempted to study microdata in a descriptive context. Exceptions are Lansing and Morgan

(1967) and Wertheimer (1970). These studies incorporated microdata, but only in a descriptive

fashion. Neither study attempted to incorporate individual or household characteristics in a

causal model of migration.

Like the regression analyses based on aggregate data that awaited the availability of

powerful computers, the microdata themselves, not to speak of their analysis, required

tremendous computer storage capacity that did not exist in earlier computers. With the

availability of microdata, econometricians began to develop new methods of studying such data.

Now the study of migration could break from the vestiges of its early history. Now many new

issues like selectivity bias must be taken into account. Even in earlier studies, such as those of

52

Makower, Marshak, and Robinson, where microdata were available, the researchers aggregated

the data both because they did not know how to study such data and because, lacking computers,

they did not have the capacity to do so. Now much migration research is based on microdata.

These studies have allowed major advances in our understanding of various migration

phenomena because now individual and household characteristics are available for study.

53

NOTES

1. Greenwood (1975) coined the term “modified gravity model” to indicate the infusion of

relative economic opportunity (or more generally, utility) into the original and more mechanistic

version of the gravity model of spatial interaction.

2. As Greenwood (1975; p. 405, fn. 12) notes, migration staging is distinguished from “chain”

migration which involves migration streams later in time linked to those earlier in time in terms

of the geography of movement. The reason offered for such chaining is information flows and

help with migrant assimilation in destinations. Of course this additional aspect of the structure of

migration response is also consistent with human capital theory.

3. “Because of the concentration of maternity hospitals in towns and cities, children born in these

hospitals to women resident in other communities will not appear as migrants, when entering the

community at a later date” (D.S. Thomas, 1938; p. 27, fn. 15).

4. B. Thomas (1934, 1937) had previously used Employment Exchange data.

5. “Gains and losses” refers to net internal migration (corrected for international migration).

6. The three volumes are Lee at al. (1957), Kuznets et al. (1960), and Eldridge and D.S. Thomas

(1964).

7. Shryock (1964) provides a useful history of the collection and reporting of U.S. migration

data.

8. Greenwood’s 1975 survey article in the Journal of Economic Literature reflects this

orientation. Of the 251 publications cited in this paper, only three made direct use of microdata

and none used longitudinal data.

9. Marshall attributes the quote (in the quote) to Schmoller.

54

10. Although population registry data were available for many years before the 1880s, they do

not appear to have been used to study migration until much later.

55

REFERENCES

Alonso, W. 1978. A theory of movements. In systems approach to human settlements, edited by N. Hansen.

Alonso, W. 1986. Systemic and log-linear models: from here to there, then to now, and this to

that. Discussion paper 86-10, Center for Population Studies, Harvard University Anderson, W.A. 1935. Mobility of rural families. II. Changes in residence and the occupation of

sons and daughters in rural families in Genesee County, New York. Ithaca: New York Agricultural Experiment Station Bulletin 623.

Beck, P.G. Recent trends in the rural population in Ohio. Wooster: Ohio Agricultural

Experiment Station Bulletin 533. Beck, P.G. and C.E. Lively. 1931. Movement of open country population in Ohio II. The

individual aspect. Wooster: Ohio Agricultural Experiment Station Bulletin 489 Clark, C.O. and R.L. Roberts. 1936. People of Kansas: as demographic and sociological study.

Topeka: Kansas State Planning Board. Bernert, E.H. 1944a. Volume and composition of net migration from the rural-farm population,

1930-40, for the United States, major geographic divisions and states. Washington: U.S. Department of Agriculture, Bureau of Agricultural Economics.

Bernert, E.H. 1944b. County variation in net migration from the rural-farm population, 1930-40.

Washington: U.S. Department of Agriculture, Bureau of Agricultural Economics.

Borjas, G, S.G. Bronars, and S. J. Trejo. 1992. Self-selection and Internal Migration in the United States. Journal of Urban Economics, 32: 159-185.

Borjas, G.J. 1994. The economics of immigration. Journal of Economic Literature 32: 1667-

1717. Carrothers, G.A.P. 1956. An historical review of the gravity and potential concepts of human

interaction. Journal of the American Institute of Planners 22:94-102 DaVanzo, J. 1976. Differences between return and nonreturn migration: an econometric analysis.

International Migration Review 10:13-27. DaVanzo, J. 1978. Does unemployment affect migration?—evidence from microdata. The

Review of Economics and Statistics, 60: 504-514. Dorn, H.F. 1934. The effect of rural-urban migration upon death-rates. Population 1: 95-114.

56

Dorn, H.F. and F. Lorimer. 1936. Migration, reproduction, and population adjustment. Annuals of the American Academy of Political and Social Science 188: 280-289

Eldridge, H.T. and D.S. Thomas. 1964. Population redistribution and economic growth in the

United States. III. Demographic analysis and interrelationships. Philadelphia: The American Philosophical Society. Eugenics Survey of Vermont. 1931. Selective migration from three rural Vermont towns and its significance. Fifth annual report. Burlington: University of Vermont.

Foot, D.K. and W.J. Milne. 1984. Net migration estimation in an extended, multiregional gravity

model. Journal of Regional Science 24: 119-133. Frazier, E. F. 1932. The Negro family in Chicago. Chicago: University of Chicago Press. Garey, L.F. 1936. Stranded farmers in urban centers. Social Forces 14: 388-394. Gee, W. 1933. A qualitative study of rural depopulation in a single township: 1900-1930.

American Journal of Sociology 39: 210-221. Gee, W. and J.J. Corson. 1929. Rural depopulation in certain tidewater and Piedmont areas of

Virginia. University of Virginia: Institute for Research in the Social Sciences Monograph No. 3.

Gee, W. and D. Runk. 1931. Qualitative selection in cityward migration. American Journal of

Sociology 37: 254-265. Gist, N.P. and C.D. Clark. 1938. Intelligence as a selective factor in rural-urban migrations.

American Journal of Sociology 44: 36-58. Goodrich, C., B.W. Allin, C.W. Thornthwaite, H.K. Brunck, F.G. Tryon, D.B. Creamer, R.B.

Vance and M. Hayes. 1936. Migration and economic opportunity. Philadelphia: University of Pennsylvania Press.

Goss, E. and H.S. Chang. 1983. Changes in elasticities of interstate migration: implication of

alternative functional forms. Journal of Regional Science 23: 223-232. Greenwood, M.J. 1975a. Research on internal migration in the United States: a survey. Journal

of Economic Literature 13: 397-433. Greenwood, M.J. 1975b. Simultaneity bias in migration models: an empirical examination.

Demography 12: 519-536. Greenwood, M.J. and G.L. Hunt. 1984a. Migration and Interregional employment redistribution

in the United States. American Economic Review 74: 957-969.

57

Greenwood, M.J. and G.L. Hunt. 1984b. Econometrically accounting for identities and

restrictions in models of interregional migration. Regional Science and Urban Economics 14: 113-128.

Greenwood, M.J. and G.L. Hunt. 1985. Econometrically accounting for identities and

restructions in models of interregional migration: further thoughts. Regional Science and Urban Economics 15: 605-614.

Hamilton, C.H. 1934. Rural-urban migration in the Tennessee Valley between 1920 and 1930.

Social Forces 13: 57-64. Hamilton, C.H. 1936. The annual rate of departure of rural youths from their parental homes.

Rural Sociology 1: 164-179. Hart, H.N. 1921 Selective migration as a factor in child welfare in the United States, with special

reference to Iowa. Iowa City: University of Iowa Studies in Child Welfare, Vol. 1, No. 7 Hill, A.B. 1925. Internal migration and its effects upon death-rates: with special reference to the

county of Essex. London: Medical Research Council Special Report Series, No.95 Hunt, G.L. 2000. Alternative nested logit model structures and the special case of partial

degeneracy. Journal of Regional Science 40: 89-113. Hunt, G.L. and M.J. Greenwood 1985. Econometrically accounting for identities and restrictions

in models of interregional migration: further thoughts. Regional Science and Urban Economics 15: 605-614.

Hunt, G. L. and R. E. Mueller. 2004. North American migration: returns to skill, border effects, and mobility costs. The Review of Economics and Statistics, 86: 988-1007.

Kau, J.B. and C.F. Sirmans. 1977. The influence of information cost and uncertainty on

migration: a comparison of migrant types. Journal of Regional Science 17: 89-96. Kennedy, L.V. 1930. The Negro peasant turns cityward: effects of rural migrations to northern

centers. New York: Columbia University Studies in History, Economics, and Public Law No 329.

Kiser, C.V. 1932. Sea island to city: a study of St. Helena islanders in Harlem and other urban

centers. New York: Columbia University Studies in History, Economics, and Public Law No. 368.

Klineberg, O. 1935. Negro intelligence and selective migration. New York: Columbia University

Press.

58

Kuznets, S., A.R. Miller, and R.A. Easterlin. 1960. Population redistribution and economic growth in the United States. II . Analyses of economic change. Philadelphia: The American Philosophical Society.

Landis, P.H. 1936. Rural population trends in Washington. Pullman: Washington Agricultural

Experiment Station Bulletin 333. Lee, E.S., A.R. Miller, C.P. Brainerd, and R.A. Easterlin. 1957. Population redistribution and

economic growth in the United States, 1975-1950. I. Methodological considerations and reference tables. Philadelphia: The American Philosophical Society.

Leybourne, G.G. 1937. Urban adjustments of migrants from the southern Appalachian plateaus.

Social Forces 16: 238-246. Linneman, P.E. and P.E. Graves. 1983. Migration and job change: a multinominal logit

approach. Journal of Urban Economics 14: 263-279. Makower, H.J., J. Marschak, and H.W. Robinson. 1938. Studies in mobility of labour: A

tentative statistical measure. Oxford Economic Papers 1: 83-123. Makower, H.J., J. Marschak, and H.W. Robinson. 1939. Studies in mobility of labour: analysis

for Great Britain, Part I. Oxford Economic Papers 2: 70-97. Makower, H.J., J. Marschak, and H.W. Robinson. 1940. Studies in mobility of labour: analysis

for Great Britain, Part II. Oxford Economic Papers 4: 39-62. Marshall, A. 1948. Principles of economics, 8th edition. New York: The Macmillan Company.

Nakosteen, R. A. and M. Zimmer. 1980. Migration and Income: The Question of Self-selection. Southern Economic Journal, 46: 840-851.

Niedercorn, J.H. and B.V. Bechdolt, Jr. 1967. An economic derivation of the “gravity law” of

spatial interaction. Journal of Regional Science 9: 273-282. Lansing, J.B. and J.N. Morgan. 1967. The effect of geographical mobility on income. Journal of

Human Resources 2: 449-460. Linneman, P. and P.E. Graves. 1983. Migration and job change: a multinomial logit approach.

Journal of Urban Economics 14: 263-279. Lively, C.E. and F. Foott. 1937. Population mobility in selected areas of rural Ohio, 1928-1935.

Wooster: Ohio Agricultural Experiment Station Bulletin 582. McCormick, T.C. 1933. Urban migration and educational selection – Arkansas data. American

Journal of Sociology 39: 355-359.

59

McKenzie, R.D. 1933. The metropolitan community. New York: McGraw-Hill Book Co. McFadden, D. 1978. Modelling the choice of residential location. In Spatial interaction theory

and planning models, edited by A. Karlquist, L. Lundquist, F. Snickars, and J.W. Weibull. Amsterdam: North-Holland.

McFadden, D. 1981. Econometric models of probabilistic choice. In Structural Analysis of

Discrete Data with Econometric Applications. Edited by C.F. Manski and D. McFadden. Cambridge: the MIT Press.

Raper, A.F. 1936. Preface to peasantry: a tale of two black belt counties. Chapel Hill: University

of North Carolina Press. Ravenstein, E.G. 1885. The laws of migration. Journal of the Royal Statistical Society 48: 167-

235. Ravenstein, E.G. 1889. The laws of migration. Journal of the Royal Statistical Society 52: 241-

305. Reuss, C.F. 1937. A qualitative study of depopulation in a remote rural district: 1900-1930.

Rural Sociology 2: 66-75. Schultz, T.P. 1982. Lifetime migration within educational strata in Venezuela: estimates of a

logistic model. Economic Development and Cultural Change 30: 559-593. Shryock, H.S. 1964. Population mobility within the United States. Chicago: University of

Chicago Press. Smick, A.A. and F.R Yoder. 1929. A study of farm migration in selected communities in the

State of Washington. Pullman: Washington Agricultural Experiment Station Bulletin 233. Smith, A. 1937. An inquiry into the nature and causes of the wealth of nations. New York: The

Modern Library. Smith, T.E. 1991. A simple theory of Spatial Interaction: the Alonso model revisited.

Environment and Planning A 23:1247-1268. Stewart, J.Q. 1941. An inverse distance variation for certain social influences. Science 93: 89-90. Stewart, J.Q. 1947. Empirical mathematical rules concerning the distribution and equilibrium of

population. Geographical Review 37: 461-485. Stewart, J.Q. 1950. The development of social physics. American Journal of Physics 18: 239-

253. Stewart, J.Q. 1952. A basis for social physics. Impact of Science on Society 3: 110-133.

60

Stouffer, S.A. 1940. Intervening opportunities: a theory relating mobility and distance. American Sociological Review 5: 845-867.

Thomas, B. 1934. The movement of labour into South-East England 1920-32. Economica 1:

220-241. Thomas, B. 1937. The influx of labour into London and the South East 1920-36. Economica 4:

323-336. Thomas, D.S. 1938. Research memorandum on migration differentials. New York: Social

Science Research Council. Thompson, W.S. 1935. Population problems. New York: McGraw Hill Book Co. Thornthwaite, C.W. 1934. Internal migration in the United States. Philadelphia: University of

Pennsylvania Press. Truesdell, L.E. 1926. Farm population of the United States 1920. Washington: Census

Monographs VI. Welton, T.A. 1911. Englands recent progress: an investigation of the statistics of migration,

mortality, etc. in the twenty years from 1881 to 1901 as indicating tendencies towards the growth or decay of particular communities. London: Chapman and Hall.

Wertheimer, R.F. III. 1970. The monetary rewards of human migration. Washington: The Urban

Institute. Young, E.C. 1924. The movement of the farm population. Ithaca: New York Agricultural

Experiment Station Bulletin 426. Young, O.E. 1930. Migratory trends of graduates of Indiana high schools, 1900-1930.

Unpublished Ph.D. dissertation, Indiana University. Zimmerman, C.C. 1926. The migration to towns and cities. Journal of Sociology 32: 450-455. Zimmerman, C.C. 1927. The migration to towns and cities II. American Journal of Sociology 33:

105-109. Zimmerman, C.C., O.D. Duncan, and F.C. Frey. 1927. The migration to towns and cities. III.

American Journal of Sociology 33: 237-241. Zimmerman, C.C. and L. Smith. 1930. Migration to towns and cities. American Journal of

Sociology 36: 41-51. Zipf, G.K. 1942. The unity of nature, least-action and natural social science. Sociometry 5: 48-

62.

61

Zipf, G.K. 1946. The [P (1) P(2) / D] hypothesis: on the intercity movement of persons.

American Sociological Review 11: 677-686.