Model Migration Schedules (1981)

MODEL MIGRATION SCHEDULES
Andrei Rogers and Luis J. Castro International Institute for Applied System Analysis, Austria
RR-8 1-30 November 1981
INTERNATIONAL INSTITUTE FOR APPLIED SYSTEMS ANALYSIS Laxenburg, Austria
International Standard Book Number 3-7045-00224 Research Reports, which record research conducted at IIASA, are independently reviewed before publication. However, the views and opinions they express are not necessarily those of the Institute or the National Member Organizations that support it. Copyright O 1981 International Institute for Applied Systems Analy sis All rights reserved. No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage or retrieval system, without permission in writing from the publisher.
PREFACE
Interest in human settlement systems and policies has been a central part of urbanrelated work at IIASA since its inception. From 1975 through 1978 this interest was manifested in the work of the Migration and Settlement Task, which was formally concluded in November 1978. Since then, attention has turned to the dissemination of the Task's results and to the conclusion of its comparative study: a quantitative assessment of recent migration patterns and spatial population dynamics in all of IIASA's 17 NMO countries. This report is part of the Task's dissemination effort, focusing on the age patterns of migration exhibited in the data bank assembled for the comparative study. It begins with a comparative analysis of over 500 observed migration schedules and then develops, on the basis of this analysis, a family of hypothetical schedules for use in instances where migration data are unavailable or inaccurate. Reports summarizing previous work on migration and settlement at IIASA are listed at the back of this report. They should be consulted for further details regarding the data base that underlies this study. ANDRE1 ROGERS Chairman Human Settlements and Services Area
CONTENTS
SUMMARY
1 2
INTRODUCTION AGE PATTERNS OF MIGRATION 2.1 Migration Rates and Migration Schedules 2.2 Model Migration Schedules A COMPARATIVE ANALYSIS OF OBSERVED MODEL MIGRATION SCHEDULES 3.1 Data Preparation, Parameter Estimation, and Summary Statistics 3.2 National Contrasts 3.3 Families of Schedules 3.4 Sensitivity Analysis ESTIMATED MODEL MIGRATION SCHEDULES 4.1 Introduction: Alternative Perspectives 4.2 The Correlational Perspective: The Regression Migration System 4.3 The Relational Perspective: The Logit Migration System CONCLUSION REFERENCES APPENDIX A Nonlinear Parameter Estimation with Model Migration Schedules APPENDIX B Summary Statistics of National Parameters and Variables of the Reduced Sets of Observed Model Migration Schedules APPENDIX C National Parameters and Variables of the Full Sets of Observed Model Migration Schedules
Research Report RR-8 1-30, November 198 1
MODEL MIGRATION SCHEDULES
Andrei Rogers and Luis J . Castro International Institute for Applied Systems Analysis, Austria
SUMMARY
7% report draws on the fundamental regulan'ty exhibited by age profiles of migration all over the world to develop a system of hypothetical model schedules that can be used in multiregional population analyses cam-ed out in countn'es that lack adequate migration data.
INTRODUCTION
Most human populations experience rates of age-specific fertility and mortality that exhibit remarkably persistent regularities. Consequently, demographers have found it possible to summarize and codify such regularities by means of mathematical expressions called model schedules. Although the development of model fertility and mortality schedules has received considerable attention in demographic studies, the construction of model migration schedules has not, even though the techniques that have been successfully applied to treat the former can be readily extended t o deal with the latter. We begin this report with an examination of regularities in age profile exhibited by empirical schedules of migration rates and go on to adopt the notion of model migration schedules to express these regularities in mathematical form. We then use model schedules to examine patterns of variation present in a large data bank of such schedules. Drawing on this comparative analysis of "observed" model schedules, we develop several "families" of schedules and conclude by indicating how they might be used to generate hypothetical "estimated" schedules for use in Third World migration studies - settings where the available migration data are often inadequate or inaccurate.
AGE PAITERNS OF MIGRATION
Migration measurement can usefully apply concepts borrowed from both mortality and fertility analysis, modifying them where necessary to take into account aspects that
A. Rogers, L.J. Cnstro
are peculiar to spatial mobility. From mortality analysis, migration studies can borrow the notion of the life table, extending it to include increments as well as decrements, in order to reflect the mutual interaction of several regional cohorts (Rogers 1973a, b, 1975, Rogers and Ledent 1976). From fertility analysis, migration studies can borrow well-developed techniques for graduating age-specific schedules (Rogers et al. 1978). Fundamental to both "borrowings" is a workable definition of the migration rate.
2.1
Migration Rates and Migration Schedules
The simplest and most common measure of migration is the crude migration rate, defmed as the ratio of the number of migrants, leaving a particular population located in space and time, to the average number of persons (more exactly, the number of personyears) exposed to the risk of becoming migrants. Data on nonsurviving migrants are often unavailable, therefore the numerator in this ratio generally excludes them. Because migration is highly age selective, with a large fraction of migrants being young, our understanding of migration patterns and dynamics is aided by computing migration rates for each single year of age. Summing these rates over all ages of life gives the gross migraproduction rate (GMR), the migration analog of fertility's gross reproduction rate. This rate reflects the level at which migration occurs out of a given region. The age-specific migration schedules of multiregional populations exhibit remarkably persistent regularities. For example, when comparing the age-specific annual rates of residential migration among whites and blacks in the United States during 1966-1971, one finds a common profile (Figure 1). Migration rates among infants and young children mirrored the relatively high rates of their parents, young adults in their late twenties. The mobility of adolescents was lower but exceeded that of young teens, with the latter showing a local low point around age 15. Thereafter migration rates increased, attaining a high peak at about age 22 and then declining monotonically with age to the ages of retirement. The migration levels of both whites and blacks were roughly similar, with whites showing a GMR of about 14 migrations and blacks one of approximately 15 over a lifetime undisturbed by mortality before the end of the mobile ages. Although it has frequently been asserted that migration is strongly sexselective, with males being more mobile than females, recent research indicates that sex selectivity is much less pronounced than age selectivity and is less uniform across time and space. Nevertheless, because most models and studies of population dynamics distinguish between the sexes, most migration measures do also. Figure 2 illustrates the age profdes of male and female migration schedules in four different countries at about the same point in time between roughly comparable areal units: communes in the Netherlands and Sweden, voivodships in Poland, and counties in the United States. The migration levels for all but Poland are similar, varying between 3.5 and 5.3 migrations per lifetime; and the levels for males and females are roughly the same. The age profiles, however, show a distinct, and consistent, difference. The high peak of the female schedule precedes that of the male schedule by an amount that appears to approximate the difference between the average ages at marriage of the two sexes. Under normal statistical conditions, point-to-point movements are aggregated into streams between one civil division and another; consequently, the level of interregional migration depends on the size of the areal unit selected. Thus if the areal unit chosen is a
Model migration schedules
FIGURE 1 Observed annual migration rates by color (- - - white, -black) and single years o f age: the United States, 1966-1971.
minor civil division such as a county or a commune, a greater proportion of residential location will be included as migration than if the areal unit cliosen is a major civil division such as a state or a province. Figure 3 presents the age profiles of female migration schedules as measured by different sizes of areal units: (1) all migrations from one residence t o another, (2) changes of residence within county boundaries, (3) migration between counties, and (4) migration between states. The respective four GMRs are 14.3, 9.3, 5.0, and 2.5. The four age profiles appear to be remarkably similar, indicating that the regularity in age pattern persists across areal delineations of different size. Finally, migration occurs over time as well as across space; therefore, studies of its patterns must trace its occurrence with respect to a time interval, as well as over a system of geographical areas. In general, the longer the time interval, the larger the number of return movers and nonsurviving migrants and, hence, the more the count of migrants will understate the number of interarea movers (and, of course, also of moves). Philip Rees, for example, after examining the ratios of one-year to five-year migrants between the Standard Regions of Great Britain, found that
. . . the number of migrants recorded over five years in an interregional flow varies from four times to two times the number of migrants recorded over one year. (Rees 1977, p. 247)
0.25
Netherlands, 1972
0.25 {
Poland, 1973
0.25 0.20
Sweden, 1968-1 973
0.25
United States, 1966-1 971
FIGURE 2 Observed annual migration rates by sex (--- females, -males) and single years of age: the Netherlands (intercommunal), Poland (intervoivodship), Sweden (intercommunal), and the United States (intercounty); around 1970.
0.45 0.40 0.35 0.30 0.25 0.20 0.15 0.10

Within counties
V)
CI a 3
..C
CI
0.05
nties es
10
20
30
40
50
60
70
80
90
FIGURE 3 Observed average annual migration rates of females by levels of areal aggregation and single years of age: the United States, 1966-1971.
2.2
Model Migration Schedules
From the preceding section it appears that the most prominent regularity found in empirical schedules of age-specific migration rates is the selectivity of migration with respect t o age. Young adults in their early twenties generally show the highest migration rates and young teenagers the lowest. The migration rates of children mirror those of their parents; hence the migration rates of infants exceed those of adolescents. Finally, migration streams directed toward regions with warmer climates and into or out of large cities with relatively high levels of social services and cultural amenities often exhibit a "retirement peak" at ages in the mid-sixties or beyond. Figure 4 illustrates a typical observed age-specific migration schedule (the jagged outline) and its graduation by a model schedule (the superimposed smooth outline) defined as the sum of four components: 1. A single negative exponential curve of the pre-labor force ages, with its rate of descent a, 2. A left-skewed unirnodal curve of the labor force ages positioned at mean age p , on the age axis and exhibiting rates of ascent h, and descent a,
A. Rogers, L.J. Castro
a, A, a, A, a,
= rate of descent of pre-labor force component = rate of ascent of labor force component = rate of descent of labor force component = rate of ascent of post-labor force component = rate of descent of post-labor force component c = constant
x , = low point xh = high peak x r = retirement peak
X = labor force shift A = parental shift B = jump
XI
x,,
x +A
Age, x
FIGURE 4
The model migration schedule.
3 . An almost bell-shaped curve of the post-labor force ages positioned at /.I, on the age axis and exhibiting rates of ascent A, and descent a, 4. A constant curve c , the inclusion of which improves the fit of the mathematical expression to the observed schedule
The decomposition described above suggests the following simple sum of four curves (Rogers et al. 1978):
The labor force and the post-labor force components in eq. (1) adopt the "double exponential" curve formulated by Coale and McNeil(1972) for their studies of nuptiality and fertility. The "full" model schedule in eq. (I) has 1 1 parameters: a , , a , , a , , p,, a , , A,, a , , p,, a , , A,, and c . The profile of the full model schedule is defined by 7 of the 1 1 parameters: cu,,p2,cu2,A,, p,, a , , and A,. Its level is determined by the remaining 4 parameters: a , , a , , a , , and c . A change in the value of the GMR of a particular model schedule alters proportionally the values of the latter but does not affect the former. As we shall see in the next section, however, certain aspects of the profile also depend on the allocation of the schedule's level among the pre-labor, labor, and post-labor force age components and on the share of the total level accounted for by the constant term c. Finally, migration schedules without a retirement peak may be represented by a "reduced" model with seven parameters, because in such instances the third component of eq. (1) is omitted. Table 1 sets out illustrative values of the basic and derived measures presented in Figure 4. The 1974 data refer t o migration schedules for an eight-region disaggregation of Sweden (Andersson and Holmberg 1980). The method chosen for fitting the model schedule to the data is a functional-minimization procedure known as the modified Levenberg--Marquardt algorithm (see Appendix A, Brown and Dennis 1972, Levenberg 1944, Marquardt 1963). Minimum chi-square estimators are used to give more weight to age groups with smaller rates of migration. To assess the goodness-of-fit that the model schedule provides when it is applied to observed data, we calculate E, the mean of the absolute differences between estimated and observed values expressed as a percentage of the observed mean:
This measure indicates that the fit of the model to the Swedish data is reasonably good, the eight regional indices of goodness-of-fit E being 6.87,6.41,12.15,11.01,9.3 1, 10.77, 11.74, and 14.82 for males and 7.30, 7.23, 10.71, 8.78,9.31, 11.61, 11.38, and 13.28 for females. Figure 5 illustrates graphically this goodness-of-fit of the model schedule t o the observed regional migration data for Swedish females. Model migration schedules of the form specified in eq. (1) may be classified into families according to the ranges of values taken on by their principal parameters. For example, we may order schedules according to their migration levels as defined by the values of the four level parameters in eq. (I), i.e., a , , a , , a , , and c (or by their associated GMRs). Alternatively, we may distinguish schedules with a retirement peak from those without one, or we may refer to schedules with relatively low or high values for the rate of ascent of the labor force curve A, or the mean age 5. In many applications, it is also meaningful t o characterize migration schedules in terms of several of the fundamental measures illustrated in Figure 4 , such as the low point x,,the high peak xh,and the retirement peak x,. Associated with the first pair of points is the labor force shift X, which is defined to be the difference in years between the ages of the high peak and the low point, i.e., X = xh - xl. The increase in the migration rate of individuals aged xh over those aged xlwill be called the jump B.
8 TABLE 1
A. Rogers, L.J. C ~ s r r c ,
Parameters and variables defining observed model migration schedules: outmigration from the 8 Region
Parameters and variablesa

GMR
01 a 1
1. Stockholm
Male Female
2. East Middle Male 1.44 0.035 0.088 0.079 20.27 0.090 0.406 Female 1.48 0.039 0.108 0.096 18.52 0.109 0.491
3. South Middle Male 1.33 0.032 0.096 0.091 19.92 0.104 0.404 Female 1.41 0.033 0.106 0.112 18.49 0.127 0.560
4. South Male Female 0.84 0.02 1 0.104 0.067 19.88 0.129 0.442
01
k
a1
A1
03 M3 a 3 A3
%(O-14) %(IS-64) %(65+)

61c
611 '31 1911 02 03
' ~ 1 1 parameters and variables are briefly defined in Appendix B and discussed more comprehensively in the %he G M R , its percentage distribution across the three major age categories (i.e., 0-14, 15-64, 65+), and
The close correspondence between the migration rates of children and those of their parents suggests another important shift in observed migration schedules. If, for each point x on the post-high-peak part of the migration curve, we obtain by interpolation the age (where it exists), x - A, say, with the identical rate of migration on the pre-low-point part of the migration curve, then the average of the values of A,, calculated incrementally for the number of years between zero and the low point x l , will be defined as the observed parental shift A . An observed (or a graduated) age-specific migration schedule may be described in a number of useful ways. For example, references may be made to the heights at particular ages, to locations of important peaks or troughs, to slopes along the schedule's age profde, to ratios between particular heights or slopes, to areas under parts of the curve, and to both horizontal and vertical distances between important heights and locations. The various descriptive measures characterizing an age-specific model migration schedule may be conveniently grouped into the following categories and subcategories:
Swedish regions, 1974 observed data by single years of age.

-
5. West
6. North Middle
7. Lower North
8. Upper North
Male
Female
Male
Female
Male
Female
Male
Female
7.69 29.57 0.023
7.07 27.42 0.027
7.37 29.92 0.042
5.89 27.01 0.059
7.37 30.15 0.053
5.05 26.94 0.077
7.26 31.61 0.040
5.08 28.30 0.063
following text. the mean age ii are all calculated with a model schedule spanning an age range of 95 years.
1.
Basic measures (the 11 fundamental parameters and their ratios) a , , a,, a,, c heights: locations: p , , p, slopes: a , , a , , A,, a , , A, ratios: S I C - a , / c , S , , = a , / a , , a = a,/a,, b,, = a , / a , , a, = A,/a,, ,
= A,/%
2.
Derived measures (properties of the model schedule) GMR, %(O-14), %(15--64), %(65+) areas: locations: ? , x, XI, xh, i distances: X, A , B
A convenient approach for characterizing an observed model migration schedule (i.e., an empirical schedule graduated by eq. (1)) is to begin with the central labor force curve
A. Rogers, I..J. Castro
2. E m Middle
FIGURE 5
continued on facing page.
0.09-
5.w-I
6. Nonh Middla
om.
00 .7
FIGURE 5 Observed (jagged line) and model (smooth line) migration schedules: females, Swedish regions, 1974.
12
and then to "add on" the pre-labor force, post-labor force, and constant components. This approach is represented graphically in Figure 6.
labor force pre-labor post-labor component force force component component

FIGURE 6
constant component
model schedule
A schematic diagram of the fundamental components of the full model migration schedule.
One can imagine describing a decomposition of the model migration schedule along the vertical and horizontal dimensions; e.g., allocating a fraction of its level to the constant component and then dividing the remainder among the other three (or two) components. The ratio S , , = a , / c measures the former allocation, and S , , = a , / a , and S , , = a,/a, reflect the latter division. The heights of the labor force and pre-labor force components are reflected in the parameters a, and a , , respectively, therefore the ratio a,/a, indicates the degree of "labor dominance", and its reciprocal, S , , = a , / a , , the index of child dependency, measures the pace at which children migrate with their parents. Thus the lower the value of 6 , , , the lower the degree of child dependency exhibited by a migration schedule and, correspondingly, the greater its labor dominance. This suggests a dichotomous classification of migration schedules into child dependent and labor dominant categories. An analogous argument applies to the post-labor force curve, and S , , = a,/a, suggests itself as the appropriate index. It will be sufficient for our purposes, however, to rely simply on the value taken on by the parameter a,,with positive values pointing out the presence of a retirement peak and a zero value indicating its absence. Labor dominance reflects the relative migration levels of those in the working ages relative to those of children and pensioners. Labor asymmetry refers to the shape of the left-skewed unimodal curve describing the age profile of labor force migration. Imagine that a perpendicular line, connecting the high peak with the base of the bell-shaped curve (i.e., the jump B), divides the base into two segments g and h as in Figure 7. Clearly, the ratio h/g is an indicator of the degree of asymmetry of the curve. A more convenient index, using only two parameters of the model schedule is the ratio a , = A,/a,, the index of labor asymmetry. Its movement is highly correlated with that of h/g, because of the approximate relation
FIGURE 7
A schematic diagram of the curve describing the age profde of labor force migration.
where o: denotes proportionality. Thus o, may be used to classify migration schedules according t o their degree of labor asymmetry. Again, an analogous argument applies t o the post-labor force curve, and o, = h,/a, may be defined as the index of retirement asymmetry. When "adding on" a pre-labor force curve of a given level to the labor force component, it is also important to indicate something of its shape. For example, if the mgration rates of children mirror those of their parents, then a, should be approximately equal to a,, and p,, = CY,/CY,, the index of parental-shift regularity, should be close to unity. The Swedish regional migration patterns described in Figure 5 and in Table 1 may be characterized in terms of the various basic and derived measures defined above. We begin with the observation that the outmigration levels in all of the regions are similar, with GMRs ranging from a low of 0.80 for males in Region 5 t o a high of 1.48 for females in Region 2. This similarity permits a reasonably accurate visual assessment and characterization of the profiles in Figure 5. Large differences in GMRs, however, give rise to slopes and vertical relationships among schedules that are noncomparable when examined visually. Recourse then must be made t o a standardization of the areas under the migration curves, for example, a general rescaling to a GMR of unity. Note that this difficulty does not arise in the numerical data in Table 1 , because, as we pointed out earlier, the principal slope and location parameters and ratios used to characterize the schedules are not affected by changes in levels. Only heights, areas, and vertical distances, such as the jump, are level-dependent measures. Among the eight regions examined, only the first two exhibit a definite retirement peak, the male peak being the more dominant one in each case. The index of child dependency S,, is highest in Region 1 and lowest in Region 8 , distinguishing the latter region's labor dominant profile from Stockholm's child dependent outmigration pattern. The index of labor asymmetry o, varies from a low of 2.34, in the case of males in Region 4 to ahigh of 4.95 for the female outmigration profile of Region 8 . Finally, with the possible exception of males in Region 1 and females in Region 6, the migration rates of children in Sweden do indeed seem to mirror those of their parents. The index of parental-shift regularity PI, is 1.26 in the former case and 0.730 in the latter; for most of the other schedules it is close to unity.
14
A COMPARATTVE ANALYSIS OF OBSERVED MODEL MIGRATION SCHEDULES
Section 2 demonstrated that age-specific rates of migration exhibit a fundamental age profile, which can be expressed in mathematical form as a model migration schedule defined by a total of 11 parameters. In this section we seek t o establish the ranges of values typically assumed by each of these parameters and their associated derived variables. This exercise is made possible by the availability of a relatively large data base collected by the Comparative Migration and Settlement Study, recently concluded at IlASA (Rogers 1976a, 1976b, 1978, Rogers and Willekens 1978, Willekens and Rogers 1978). The migration data for each of the 17 countries included in this study are set out in individual case studies, which are listed at the end of this report.
3.1
Data Reparation, Parameter Estimation, and Summary Statistics
The age-specific migration rates that were used t o demonstrate the fits of the model migration schedule in the last section were single-year rates. Such data are scarce at the regional level and, in our comparative analysis, are available only for Sweden. AU other region-specific migration data are reported for five-year age groups only and, therefore, must be interpolated to provide the necessary input data by single years of age. In all such instances the region-specific migration schedules were first scaled to a GMR of unity (GMR = 1) before being subjected to a cubic-spline interpolation (McNeil et al. 1977). Starting with a migration schedule with a GMR of unity and rates by single years of age, the nonlinear parameter estimation algorithm ultimately yields a set of estimates for the model schedule's parameters (see Appendix A for details). Table 1 in section 2 presented the results that were obtained using the data for Sweden. Since these data were available for single years of age, the influence of the interpolation procedure could be
TABLE 2 hameters defining observed model migration schedules and parameters obtained after a cubicRegion and width of age group
1. Stockholm 2. East Middle 1 ~r 0.026 0.108 0.065 18.52 0.109 0.491 0.003 5 yr 0.026 0.106 0.070 18.99 0.117 0.351 0.003 3. South Middle 1 ~r 0.023 0.106 0.080 18.49 0.127 0.560 0.003 5Y r 0.023 0.105 0.087 18.93 0.136 0.375 0.003 4. South 1 ~r 0.025 0.104 0.080 19.88 0.129 0.442 0.003
5 ~r
Parameters
01 a1 a1 Cla a1 A1
1 ~r 0.029 0.091 0.047 19.32 0.094 0.369 0.002 0.000 85.01 0.369 0.072
5v 0.028 0.089 0.049 19.69 0.098 0.313 0.002 0.000 81.20 0.364 0.080
c
a3
0.025 0.106 0.085 20.23 0.135 0.367 0.003
Cln
a3
A,
a~bserved data are for single yearsof age (1 yr); the cubic-spline-interpolatedinputsare obtained from observed
15
assessed. Table 2 contrasts the estimates for female schedules in Table 1 with those obtained when the same data are first aggregated to five-year age groups and then disaggregated to single years of age by a cubic-spline interpolation. A comparison of the parameter estimates indicates that the interpolation procedure gives generally satisfactory results. Table 2 refers to results for rates of migration from each of eight regions to the rest of Sweden. If these rates are disaggregated by region of destination, then 8' = 64 interregional schedules need to be examined for each sex, which will complicate comparisons with other nations. To resolve this difficulty we shall associate a "typical" schedule with each collection of national rates by calculating the mean of each parameter and derived variable. Table 3 illustrates the results for the Swedish data. To avoid the influence of unrepresentative "outlier" observations in the computation of averages defining a typical national schedule, it was decided to delete approximately 10 percent of the "extreme" schedules. Specifically, the parameters and derived variables were ordered from low value to high value; the lowest 5 percent and the highest 5 percent were defined to be extreme values. Schedules with the largest number of low and higli extreme values were discarded, in sequence, until only about 9 0 percent of the original number of schedules remained. This reduced set then served as the population of schedules for the calculation of various summary statistics. Table 4 illustrates the average parameter values obtained with the Swedish data. Since the median, mode, standard deviation-tomean ratio, and lower and upper bounds are also of interest, they are included as part of the more detailed computer outputs reproduced in Appendix B. The comparison, in Table 2, of estimates obtained using one-year and five-year age intervals for the same Swedish data indicated that the interpolation procedure gave satisfactory results. It also suggested, however, that the parameter A, was consistently underestimated with five-year data. To confirm this, the results of Table 4 were replicated wit11 the Swedish data base, using an aggregation with five-year age intervals. The results, set out in Table 5, show once again that A, is always underestimated by the interpolation procedure. This tendency should be noted and kept in mind.
spline interpolation: Sweden, 8 regions, females, 1974.'
5. West
6. North Middle
7. Lower North
8. Upper North
data by five-year age groups (5 yr).
16
TABLE 3 Mean values of parameters defining the full set of observed model migration schedules: Sweden, 8 regions, 1974 observed data by single years of age until 84 years and over.'
Males Parameters
01
01
Females With retirement peak (1 1 schedules) 0.025 0.080 0.050 21.52 0.096 0.439 0.002 0.0012 75.45 0.797 0.294 Without retirement peak (58 schedules) 0.027 0.114 0.078 19.13 0.133 0.525 0.003 With retirement peak (5 schedules) 0.023 0.087 0.051 19.20 0.101 0.377 0.003 0.0017 72.07 0.688 0.192
Without retirement peak (52 schedules) 0.029 0.1 26 0.066 21.09 0.113 0.459 0.003
0 '
Pa
(2.1
Aa
c
as
P3 03 A3
'Region 1 (Stockholm) is a singlecommune region; hence there exists no intraregional schedule for it, leaving 8' - 1 = 6 3 schedules.
TABLE 4 Mean values of parameters defining the reduced set of observed model migration schedules: Sweden, 8 regions, 1974 observed data by single years of age until 84 years and over.'
Males Parameters a1
01
Females With retirement peak (9 schedules) 0.026 0.085 0.05 1 21.25 0.093 0.416 0.002 0.0006 76.7 1 0.847 0.158 Without retirement peak (54 schedules) 0.026 0.108 0.076 19.09 0.127 0.537 0.003 With retirement peak (3 schedules) 0.024 0.093 0.055 18.87 0.106 0.424 0.003 0.0001 74.78 0.938 0.170
Without retirement peak (48 schedules) 0.029 0.1 24 0.067 20.50 0.104 0.448 0.003
a1
PI
0 1 A1
c
a3
Cg
03 A3
'Region 1 (Stockholm) is a singlecommune region; hence there exists no intraregional schedule for it, leaving 8' - 1 = 6 3 schedules, of which 6 were deleted.
It is also important to note the erratic behavior of the retirement peak, apparently due to its extreme sensitivity to the loss of information arising out of the aggregation. Thus, although we shall continue to present results relating to the post-labor force ages, they will not be a part of our search for families of schedules.
17
TABLE 5 Mean values of parameters defining the reduced set of observed model migration schedules: Sweden, 8 regions, 1974 observed data by five years of age until 80 years and over.a
Males Parameters Without retirement peak (49 schedules) With retirement peak (8 schedules) Females Without retirement peak (54 schedules) With retirement peak (3 schedules)
0.028 0.115 0.068 20.61 0.105 0.396 0.002
0.026 0.088 0.052 20.26 0.084 0.390 0.001 0.0017 77.47 0.603 0.148
0.026 0.108 0.080 19.52 0.133 0.374 0.002
0.026 0.077 0.044 19.18 0.089 0.341 0.002 0.0036 77.72 0.375 0.134
a ~ e g i o n (Stockholm) is a single-commune region; hence there exists no intraregional schedule for it, 1 ' leaving 8 -- 1 = 63 schedules, of which 6 were deleted.
3.2
National Contrasts
Tables 4 and 5 of the preceding subsection summarized average parameter values for 57 male and 57 female Swedish model migration schedules. In this subsection we shall expand our analysis to include a much larger data base, adding to the 114 Swedish model schedules another 164 schedules from the United Kingdom (Table 6), 114 from Japan, 20 from the Netherlands (Table 7), 58 from the Soviet Union, 8 from the United States, and 32 from Hungary (Table 8). Summary statistics for these 510 schedules are set out in
TABLE 6 Mean values of parameters defining the reduced set of observed model migration schedules: the United Kingdom, 10 regions, 1970.~
Males Parameters
a1
Females With retirement peak (23 schedules) Without retirement peak (61 schedules) With retirement peak (21 schedules)
Without retirement peak (59 schedules)
0.018 0.089 a2 0.048 P a 21.56 a1 0.153 A2 0.333 c 0.004 a3 0.002 P 3 71.84 a 3 0.583 As 0.403 a ~ intraregional migration data were included in the United Kingdom data; hence 1 ' - 10 = 90 o 0 schedules were analyzed, of which 8 were deleted.
a1
0.021 0.099 0.059 22.00 0.127 0.259 0.003
0.016 0.080 0.053 20.42 0.120 0.301 0.004 0.007 71.11 0.692 0.309
0.021 0.097 0.063 21.35 0.151 0.327 0.003
18
TABLE 7 Mean values of parameters defining the reduced set of observed model migration schedules: Japan, 8 regions, 1970; the Netherlands, 12 regions, 1 9 7 4 . ~ Japan Males Parameters Without retirement peak (57 schedules) Females Without retirement peak (57 schedules) Netherlands Males With retirement slope (10 schedules) Females With retirement slope (10 schedules)
a ~ e g i o n in Japan (Hokkaido) is a single-prefecture region; hence there exists no intraregional schedule 1 for it, leaving 8' - 1 = 6 3 schedules, of which 6 were deleted. The only migration schedules available for the Netherlands were the migration rates out of each region without regard to destination; hence only 12 schedules were used, of which 2 were deleted.
TABLE 8 Mean values of parameters defining the reduced set of observed total (males plus females) model migration schedules: the Soviet Union, 8 regions, 1974; the United States, 4 regions, 19701971; Hungary, 6 regions, 1 9 7 4 . ~ Soviet Union Parameters Without retirement peak (58 schedules) United States With retirement peak (8 schedules) Hungary Without retirement slope (7 schedules) With retirement slope (25 schedules)
aIntraregional migration was included in the Soviet Union and Hungarian data but not in the United States data; hence there were 8' = 64 schedules for the Soviet Union, of which 6 were deleted, 6' = 36 schedules for Hungary, of which 4 were deleted, and 4' - 4 = 12 schedules for the United States, of which 2 were deleted because they lacked a retirement peak and another 2 were deleted because of their extreme values.
Appendix B; 206 are male schedules, 206 are female schedules, and 98 are for the combination of both sexes (males plus females).*
*This total does not include the 56 schedules excluded as "extreme" schedules. During the process of fitting the model schedule to these more than 500 interregional migration schedules, a frequently encountered problem was the occurrence of a negative value for the constant c. In all such instances
19
A significant number of schedules exhibited a pattern of migration in the post-labor force ages that differed from that of the 11-parameter model migration schedule defined in eq. ( 1 ) . Instead of a retirement peak, the age profile took on the form of an "upward slope". In such instances the following 9-parameter modification of the basic model migration was introduced
M ( x ) = a , exp (-a,x)
The right-hand side of Table 7, for example, sets out the mean parameter estimates of this modified form of the model migration schedule for the Netherlands. Tables 4 through 8 present a wealth of information about national patterns of migration by age. The parameters, given in columns, define a wide range of model migration schedules. Four refer only to migration level: a l , a 2 ,a,, and c. Their values are for a GMR of unity; to obtain corresponding values for other levels of migration, these four numbers need to be multiplied by the desired level of GMR. For example, the observed GMR for female migration out of the Stockholm region in 1974 was 1.43. Multiplying a, = 0.029 by 1.43 gives 0.041, the appropriate value of a , with which to generate the migration schedule having a GMR of 1.43. The remaining model schedule parameters refer to migration age profile: a,, p,, a,, A,, p,, a,, and A,. Their values remain constant for all levels of the GMR. Taken together, they define the age profile of migration from one region to another. Schedules without a retirement peak yield only the four profile parameters: a,, p,, a,, and A,, and schedules with a retirement slope have an additional profile parameter a,. A detailed analysis of the parameters defining the various classes of schedules is beyond the scope of this report. Nevertheless a few basic contrasts among national average age profiles may be usefully highlighted. Let us begin with an examination of the labor force component defined by the four parameters a,, p, , a,, and A,. The national average values for these parameters generally lie within the following ranges:
the initial value of c was set equal to the lowest observed migration rate and the nonlinear estimation procedure was started once again.
20
A. Rogers, L.J. G s t r o
In all but two instances, the female values for a , , a , , and h, are larger than those for males. The reverse is the case for p,, with two exceptions, the most important of which is extubited by Japan's females, who consistently show a high peak that is older than that of males. This apparently is a consequence of the tradition in Japan that girls leave the family home at a later age than boys. The two parameters defming the pre-labor force component, a , and a , , generally lie within the ranges of 0.01 to 0.03 and 0.08 to 0.12, respectively. The exceptions are the Soviet Union and Hungary, which exhibit unusually high values for a , . Unlike the case of the labor force component, consistent sex differentials are difficult t o identify. Average national migration age profiles, like most aggregations, hide more than they reveal. Some insight into the ranges of variations that are averaged out may be found by consulting the lower and upper bounds and standard-deviation-to-mean ratios listed in Appendix B for each set of national schedules. Additional details are set out in Appendix C. Finally, Table 9 illustrates how parameters vary in several unaveraged national schedules, by way of example. The model schedules presented there describe migration flows out of and into the capital regions of each of six countries: Helsinki, Finland; Budapest, Hungary; Tokyo, Japan; Amsterdam, the Netherlands; Stockholm, Sweden; and London, the United Kingdom. All are illustrated in Figure 8 . The most apparent difference between the age profiles of the outflow and inflow migration schedules of the six national capitals is the dominance of young labor force migrants in the inflow, that is, proportionately moremigrants in the young labor force ages appear in the inflow schedules. The larger values of the product a , h, in the inflow schedules and of the ratio A,, = a , / a , in the outflow schedules indicate this labor dominance. A second profile attribute is the degree of asymmetry in the labor force component of the migration schedule, i.e., the ratio of the rate of ascent h, to the rate of descent a , defined as o, in section 2. In all but the Japanese case, the labor force curves of the capitalregion outmigration profiles are more asymmetric than those of the corresponding inmigration profiles. We refer to this characteristic as labor asymmetry. Examining the observed rates of descent of the labor and pre-labor force curves, a , and a , , respectively, we find, for example, that they are close to being equal in the outflow
TABLE 9 Parameters defining observed total (males plus females) model migration schedules for flows 1974; the United Kingdom, 1970. Finland Parameters
"I
QI
Hungary To Helsinki
0.024 0.170 0.130 22.1 3 0.198 0.231 0.003
Japan To Budapest
0.008 0.262 0.094 17.69 0.152 0.305 0.003 0.00005 0.059
From Helsinki
0.037 0.127 0.081 21.42 0.124 0.231 0.000 0.00027 99.32 0.204 0.042
From Budapest
0.015 0.239 0.082 17.10 0.130 0.355 0.003 0.00001 0.072
From Tokyo
0.019 0.157 0.064 20.70 0.111 0.204 0.003 0.00002 0.061
To Tokyo
0.008 0.149 0.096 15.74 0.134 0.577 0.002 0.00131 0.000
"1
Pl
a1
A1 c
"3
k
Q3
A,
21
schedules of Helsinki and Stockholm and are highly unequal in the cases of Budapest, Tokyo, and Amsterdam. In four of the six capital-region inflow profiles a, > a,. Profiles with significantly different values for a, and a, are said to be irregular. In conclusion, the empirical migration data of six industrialized nations suggest the following hypothesis. The age profile of a typical capital-region inmigration schedule is, in general, more labor dominant and more labor symmetric than the age profile of the corresponding capital-region outmigration schedule. No comparable hypothesis can be made regarding its anticipated degree of irregularity.
3.3
Families of Schedules
Three sets of model migration schedules have been defined in this report: the 11parameter scheduIe with a retirement peak, the alternative 9-parameter schedule with a retirement slope, and the simple 7-parameter schedule with neither a peak nor a slope. Thus we have at least three broad families of schedules. Additional dimensions for classifying schedules into families are suggested by the above comparative analysis of national migration age profiles and the basic measures and derived variables defined in section 2. These dimensions reflect different locations on the horizontal and vertical axes of the schedule, as well as different ratios of slopes and heights. Of the 524 model migration schedules studied in this section, 412 are sex-specific and, of these, only 336 exhibit neither a retirement peak nor a retirement slope. Because the parameter estimates describing the age profile of post-labor force migration behave erratically, we shall restrict our search for families of schedules to these 164 male and 172 female model schedules, summary statistics for which are set out in Tables 10 and 1 I . An examination of the parametric values exhibited by the 336 migration schedules summarized in Tables 10 and 11 suggests that a large fraction of the variation shown by these schedules is a consequence of changes in the values of the following four parameters , and derived variables: p,, 6,, , o , and PI,.
from and to capital cities: Finland, 1974; Hungary, 1974; Japan, 1970; the Netherlands, 1974; Sweden,
Netherlands From Amsterdam To Amsterdam
Sweden From Stockholm To Stockholm
United Kingdom From London To London
TABLE 1 0 Estimated summary statistics of parameters and variables associated with reduced sets of observed model migration schedules for Sweden, the United Kingdom, and Japan: males, 164 schedulesa summary statistics Parameters and variables GMR (observed) GMR (model) E
a1
5.
Lowest value
Highest value
Mean value
Median
Mode
Standard deviation
Standard deviation1 mean
9 "
' 4
a2
a 2
A2
n %(O-14) %(IS-64) %(65+)

&,c
612
Pn
02
9
Xh
X
A
B
A list of definitions for the parameters and variables appears in Appendix B. '
TABLE 11 Estimated summary statistics of parameters and variables associated with reduced sets of observed model migration schedules for Sweden, the United Kingdom, and Japan: females, 172 schedulesa Summary statistics Parameters and variables
GMR (observed) GMR (model) E
a1
Lowest value
Highest value
Mean value
Median
Mode
Standard deviation
Standard deviation/ mean
(I1
01
fil
(12
A,
c
FT
%(O-14) %(IS-64) %(65+)

61c
611
02 1
02
Xl
Xh
X A B
0.00388 1.00000 4.17964 0.00526 0.01585 0.02207 15.06610 0.05467 0.08367 0.00012 24.51402 9.37675 60.55278 1.46164 0.89359 0.02828 0.09121 0.38917 10.32012 17.03028 2.89007 23.73040 0.00831
1.59564 1.00000 60.83579 0.04496 0.41038 0.18944 37.76019 0.33556 1.49869 0.00685 37.86541 3 1A7480 81.17286 19.56255 192.60318 0.90435 2.48385 12.23371 21.79038 30.92059 15.09035 37.24700 0.09111
0.19909 1.00000 15.42092 0.02259 0.10698 0.07426 20.63237 0.14355 0.40032 0.00347 30.65265 20.93872 68.65491 10.40638 9.39987 0.34847 0.81472 3.26434 14.5 1330 22.49959 7.98629 28.50972 0.03118
A list of definitions for the parameters and variables appears in Appendix B. '
27
Migration schedules may be early or late peaking, depending on the location of p, on the horizontal (age) axis. Although this parameter generally takes on a value close t o 20, roughly three out of four observations fall within the range 17--25. We shall call those below age 19 early peaking schedules and those above 22 late peaking schedules. The ratio of the two basic vertical parameters, a , and a,, is a measure of the relative importance of the migration of children in a model migration schedule. The indexof child dependency, S,, = a,/a,, tends to exhibit a mean value of about one-third with 8 0 percent of the values falling between one-fifth and four-fifths. Schedules with an index of one-fifth or less will be said to be labor dominant; those above two-fifths will be called child dependent. Migration schedules with labor force components that take the form of a relatively symmetrical bell shape will be said to be labor symmetrical. These schedules will tend t o exhibit an index of labor asymmetry (a, = A,/ol,) that is less than 2. Labor asymmetric schedules, on the other hand, will usually assume values for a, of 5 or more. The average migration schedule will tend to show a a, value of about 4, with approximately five out of six schedules exhibiting a a, within the range 1-8. Finally, the index of parental-shift regularity in many schedules is close to unity, with approximately 70 percent of the values lying between one-third and four-thirds. Values of PI, = a,/% that are lower than four-fifths or higher than six-fifths will be called irregular. We may imagine a 3 X 4 cross-classification of migration schedules that defines a dozen "average families" (Table 12). Introducing a low and a high value for each parameter gives rise to 16 additional families for each of the three classes of schedules. Thus we may conceive of a minimum set of 60 families, equally divided among schedules with a retirement peak, schedules with a retirement slope, and schedules with neither a retirement peak nor a retirement slope (a reduced form).
TABLE 12
A cross-classification of migration schedules. Measures (averaee values) Peaking Dominance

(h,, = 1/31
Asymmetry
( a , = 4)
Regularity
( , = 1) s,
Schedule Retirement peak Retirement slope Reduced fonn
(r, = 20)
+ + +
+ + +
+ + +
+ + +
To complement the above discussion with a few visual illustrations, in Figure 9(a) we present six labor dominant profiles, with SIC fixed at 22. The tallest three exhibit a steep rate of descent a, = 0.3; the shortest three show a much more moderate slope of a, = 0.06. Within each family of three curves, one finds variations in p, and in the rate of ascent A,. Increasing p, shifts the curve to the right along the horizontal axis; increasing A, raises the relative height of the high peak. The six schedules in Figure 9(b) depict the corresponding two families of child dependent profiles. The results are generally similar to those in Figure 9(a), with the
I
# .
It
" "
I1
0
2 2 22
1,
11 11 I1
a",
11 11
30
exception that the relative importance of migration in the pre-labor force age groups is increased considerably. The principal effects of the change in 6,, are: (I) a raising of the intercept a , + c along the vertical axis, and (2) a simultaneous reduction in the height of the labor force component in order to maintain a constant area of unity under each curve. Finally, the dozen schedulcs in Figures 9(c) and 9(d) describe similar families of migration curves, but in these profiles the relative contribution of the constant componerlt to the unit GMR has been increased significantly (i.e., 6,, = 2.6). It is important to note that such "pure" measures of profiles asxl, x,, ,X, and A remain unaffected by this change, whereas "impure" profile measures, such as the mean age of migration ii, now take on a different set of values.
3.4
Sensitivity Analysis
The preceding subsections have focused on a comparison of the fundamental parameters defining the model migration age profiles of a number of nations. The comparison yielded ranges of values within which each parameter may be expected to fall and suggested a classification of schedules into families. We now turn to an analytic examination of how changes in several of the more important parameters become manifested in the age profile of the model schedule. For analytical convenience we begin by focusing on the properties of the double exponential curve that describes the labor force component:
We begin by observing that if a , is set equal to A, in the above expression, then the labor force component assumes the shape of a well-known extreme value distribution used in the study of flood flows (Gumbel 1941, Kimball 1946). In such a case x h = p, and the function f,(x) achieves its maximum y h at that point. To analyze the more general case where a, # A,, we may derive analytical expressions for both of these variables by differentiating eq. (4) with respect to x , setting the result equal to zero, and then solving to find
an expression that does not involve a,, and
an expression that does not involve p, . Note that if A, > a , , which is almost always the case, then xh that if a , = A,, then the above two equations simplify to
> p,.
And observe
and
31
Since p, affects xh only as a displacement, we may focus on the variation of xh as a function of a, and A,. A plot of xh against a,, for a fixed A,, shows that increases in a, lead to decreases in x h . Analogously, increases in A,, for a fixed a,, produce increases in xh but at a rate that decreases rapidly as the latter variable approaches its asymptote. The behavior of yh is independent of p, and varies proportionately with a,. Hence its variation also depends fundamentally only on the two variables a, and A,. A plot of yh against a,, for a fixed A,, gives rise to a U-shaped curve that reaches its minimum at a, = A,. Increasing A, widens the shape of the U. The influence of a, and A, on the labor force component may be assessed by examining the proportional rate of change of the function f, (x):
Equation (7) defines this rate of change as the sum of two components: -a, and the exponential A, exp[-A,(x - p , ) ] . To demonstrate how the actual rates of ascent and descent are related to A, and a, we may take, for example, a typical set of parameter values such as a, = 0.1, A, = 0.4, and p, = 20 and then proceed t o calculate the quantities presented in Table 13. The calculations indicate that, at ages above 30, the actual rate of descent is almost identical t o -a,. The actual rates of ascent are very different from the A, value, except for ages close to x = p,.*
TABLE 13 Impacts of A, and a, on the actual rates of ascent and descent of the lab01 force component: A, = 0.4, a, = 0.1, and p, = 20. Actual rates of ascent and descent Range of age In this range the impact of a, can be ignored
15
Age (x)
d x ) = A, exp [-A,
1192 161 22 3
(x
-p,)]
-a,
+ g(x)
1192 161 22 3
In this range the impact of A, can be ignored
0.007 0.001
4.093 4.100
*We are grateful to Kao-Lee Liaw for suggesting the examination of eq. (7) and for pointing out that the parameters A, and a, are not truly rates o f ascent and descent, respectively.
32
The introduction of the pre-labor force component into the profile generally moves x h t o a slightly younger age and raises y h by about a , exp(-a,xh), usually a neghgible quantity. The addition of the constant term c , of course, affects only y h , raising it by the amount of the constant. Thus the migration rate at age xh may be expressed as
A variable that interrelates the pre-labor force and labor force components is the parental shift A . To simplify our analysis of its dependence on the fundamental parameters, it is convenient to assume that a, and a, are approximately equal. In such instances, for ages immediately following the h g h peak x h , the labor force component of the model migration schedule is closely approximated by the function a, exp[-a,(x, - p,)] . Recalling that the pre-labor force curve is given by a , exp(-a,x,) when a, = a,, we may equate the two functions t o solve for the difference in ages that we have called the parental shift:
This equation shows that the parental shift will increase with increasing values of p 2 and will decrease with increasing values of a, and ti,,. Table 1 4 compares the values of this analytically defined "theoretical" parental shift with the corresponding observed parental shifts presented earlier in Table 1 for Swedish males and females. The two definitions appear t o produce similar numerical values, but the analytical definition has the advantage of being simpler t o calculate and analyze. Consider the rural-to-urban migration age profile defined by the parameters in Table 15. In this profile the values of a, and A, are almost equal, making it a suitable illustration of several points raised in the above discussion. First, calculating xh with eq. (5) gives
as against x h = 21.59 set out in Table 15. Deriving y h using eq. (6) gives
where a,/A, = 0.23710.270 = 0.878. Thus M(21.59) is approximately equal to y h + c = 0.069 + 0.004 = 0.073. The value given by the model migration schedule equation is also 0.073. Since a, # a,, we cannot adequately test the accuracy of eq. (8) as an estimator of A. Nevertheless, it can be used t o help account for the unusually large value of the parental shift. Substituting the values for p,, a,, and ti,, into eq. (8), we find
And although this is an underestimate of 45.13, it does suggest that the principal cause for the unusually high value o f A is the unusually low value of ti,,. If this latter parameter
TABLE 14
Observed and theoretical values of the parental shift: Sweden, 8 regions, 1974. Regions of Sweden
8
5. West 29.57 28.97 27.42 27.51 6. North Middle 29.92 29.43 27.01 28.54 7. Lower North 30.15 26.6 1 26.94 28.19 8. Upper North 31.61 29.89 28.30 28.95
Parental shift ~ b s e r v e dmales ,~ Theoretical, males observed: females Theoretical, females a ~ o u r c eTable 1. :
1. Stockholm 27.87 25.14 25.49 24.68
2. East Middle 29.99 29.24 27.32 26.85
3. South Middle 29.93 30.01 27.27 28.16
4. South 29.90 29.65 27.87 28.91
34
TABLE 15
A . Rogers, L.J. Castro
Parameters and variables defining observed total (males plus females) model migration schedules for urban-to-rural and rural-to-urban flows: the Soviet Union, 1974. Parameters and variablesa Urban-to-rural
0.74 0.005 0.313 0.1 27 19.26 0.177 0.286 0.005 33.66 8.63 78.30 13.07 0.977 0.038 1.77 1.61 11.09 20.94 9.85 42.30 0.045
Rural-to-urban
3.41 0.002 0.431 0.187 21.10 0.237 0.270 0.004 31.24 5.59 84.60 9.81 0.548 0.01 1 1.82 1.14 11.38 21.59 10.21 45.13 0.063
GMR
aI
&I
a2 Pz a2
A,
11
%(O-14) %1(15-64) %(65+) 6~c

6 ~ z
P2 1
01
X1
Xh X A
B
a~
list of definitions for the parameters and variables appears in Appendix A.
had the value found for Stockholm's males, for example, the parental shift would exhibit the much lower value of 22.52.
ESTIMATED MODEL MIGRATION SCHEDULES
An estimated model schedule is a collection of age-specific rates derived from patterns observed in various populations other than the one being studied plus some incomplete data on the population under examination. The justification for such an approach is that age profiles of fertility, mortality, and geographical mobility vary within predetermined limits for most human populations. Birth, death, and migration rates for one age group are lughly correlated with the corresponding rates for other age groups, and expressions of such interrelationships form the basis of model schedule construction. The use of these regularities to develop hypothetical schedules that are deemed to be close approximations of the unobserved schedules of populations lacking accurate vital and mobility registration statistics has been a rapidly growing area of contemporary demographic research.
4.1
Introduction: Alternative Perspectives
The earliest efforts in the development of model schedules were based on only one parameter and hence had very little flexibility (United Nations 1955). Demographerssoon
35
discovered that variations in the mortality and fertility regimes of different populations required more complex formulations. In mortality studiesgreater flexibility was introduced by providing families of schedules (Coale and Demeny 1966) or by enlarging the number of parameters used to describe the age pattern (Brass 1975). The latter strategy was also adopted in the creation of improved model fertility schedules and was augmented by the use of analytical descriptions of age profiles (Coale and Trussell 1974). Since the age patterns of migration normally exhibit a greater degree of variability across regions than do mortality and fertility schedules, it is to be expected that the development of an adequate set of model migration schedules will require a greater number both of families and of parameters. Although many alternative methods could be devised to summarize regularities in the form of families of model schedules defined by several parameters, three have received the widest popularity and dissemination: 1. The regression approach of the Code-Demeny model life tables (Coale and Demeny 1966) 2. The logit system of Brass (Brass 1971) 3. The double exponential graduation of Coale, McNeil, and Trussell (Coale 1977, Coale and McNeil 1972, Coale and Trussell 1974) The regression approach embodies a correlational perspective that associates rates at different ages to an index of level, where the particular associations may differ from one "family" of schedules to another. For example, in the Code-Demeny model life tables, the index of level is the expectation of remaining life at age 10, and a different set of regression equations is established for each of four "regions" of the world. Each of the four regions (North, South, East, and West) defines a collection of similar mortality schedules that are more uniform in pattern than the totality of observed life tables. Brass's logit system reflects a relational perspective in which rates at different ages are given by a standard schedule whose shape and level may be suitably modified to be appropriate for a particular population. The Code-Trussell model fertility schedules are relational in perspective (using a Swedish standard first-marriage schedule), but they also introduce an analytic description of the age profile by adopting a double exponential curve that defines the shape of the age-specific first-marriage function. In this study we mix the above three approaches to define two alternative perspectives for estimating model migration schedules in situations where only inadequate or defective data on internal (origin-destination) migration flows are available. Both perspectives rely on the analytic (double plus single exponential) graduation defined by the basic model migration schedule set out earlier in this study. Both ultimately depend on the availability of some limited data t o obtain the appropriate model schedule, for example, at least two age-specific rates, such as M(0-4) and M(20-24), and informed guesses regarding the values of a few key variables, such as the low and high points of the schedule. They differ only in the method by which a schedule is identified as being appropriate for a particular population. The first perspective, the regression approach, associatesvariations in the parameters and derived variables of the model schedule t o each other and then to age-specific migration rates. The second, the logit approach, embodies different relationships between the model schedule parameters in several standard schedules and then associates the logits of the migration rates in a standard to those of the population in question.
36
A. Rogers, L.J. Costro
4.2
The Correlational Perspective: The Regression Migration System
A straightforward way of obtaining an estimated model migration schedule from limited observed data is to associate such data with the basic model schedule's parameters by means of regression equations. For example, given estimates of the migration rates of infants and young adults, M(0-4) and M(20-24) say, we may use equations of the form
to estimate the set of parameters Qithat define the model schedule. The parameters of the fitted model scl~edules not independent of each other, however. Higher than average are values of A,, for example, tend to be associated with lower than average values of a , . The incorporation of such dependencies into the regression approach would surely improve the accuracy and consistency of the estimation procedure. An examination of empirical associations among model schedule parameters and variables, therefore, is a necessary first step. Regularities in the covariations of the model schedule's parameters suggest astrategy of model schedule construction that builds on regression equations embodying these covariations. Given the values for 6 , x , , and xh ,for example, one can proceed to derive p,, A,, o,, and PI,. Since a, = A,/a, we obtain, at the same time, an estimate for a,, which we then can use to fmd a,. With a , established, a , may be obtained by drawing on the definitional equation ti,, = a , / a , , and a, may be found with the similar equation P,, = a , / a , . A initial value for c is obtained by setting c = a , / 6 , , , where 6 , , is estimated by n regressing it on 6,,, anda,,a,, andc are scaled to give a GMR of unity. Conceptually, this approach to model schedule construction begins with the labor force component and then appends t o it the pre-labor force part of the curve. The value given for 6 , , reflects the relative weights of these two components, with low values defining a labor dominant curve and high values pointing to a family dominant curve. (The behavior of the post-labor force curve is assumed here to be treated exogenously.) We begin the calculations with p, to establish the location of the curve on the age axis; is it an early or late peaking curve? Next, we turn to the determination of its two slope parameters A, and a, by resolving whether or not it is a labor symmetric curve. Values of a , between 1 and 2 generally characterize a labor symmetric curve; higher values describe an asymmetric age profile. The regression of a , on a, produces the fourth parameter needed to define the labor force component. With values for p,, A,, a,, and a, the construction procedure turns to the estimation of the pre-labor force curve, which is defined by the two parameters a, and a , . Its relative share of the total unit area under the model migration schedule is set by the value given to 6 , , . The retirement peak and the upward slope are introduced exogenously by setting their parameters equal to those of the "observed" model migration schedule. The collection of regression equations given in Table 16 exemplifies a regression system that may be defined to represent the "child dependency" set, inasmuch as their central independent variable ti,, is the index of child dependency. It is also possible to replace this independent variable with others, such as o, or P,, for example, to create a "labor asymmetry" or a "parental-shift regularity" set. The regression coefficients were obtained using the age-specific interregional migration schedules (scaled to unit GMR) of Sweden, the United Kingdom, and Japan. Of the three variants, the child dependency set gave the best fits in about half of the female schedules tested, whereas the parental-shift

TABLE 16
A basic set of regression equations.
Regression coefficients of independent variables Dependent variables

P,
Intercept
-
6,,
X1
Xh
a 2
(males) (females) (males) (females) (males) (females) (males) (females) (males) (females) (males) (females)
-3.26 7.69
3.28 -2.14 0.15 0.13 5.59 6.05 1.33 1.42
-4.67 -0.53 0.08 0.08 0.89 0.63 4.03 0.04
1.39 1.63 4.09 4.09 -1.17 -0.85 4.04 4.04
A,
o,
1.31 1.19
16.43 10.97 1.90 1.82 0.03 0.04 9.41 0.19
P,,
a,
6, ,
13.83 26.4 3
regularity set was overwhelmingly the best fitting variant for the male schedules (see Rogers and Castro 1981). To use the basic regression equations presented in Table 16, one first needs to obtain estimates of til2, x l , and xh. Values for these three variables may be selected to reflect informed guesses, historical data, or empirical regularities between such model schedule variables and observed migration data. For example, suppose that a fertility survey has produced a crude estimate of the ratio of infant to parent migration rates: M = M(0--4)l M(20-24), say. A linear association between S12 and this M ratio, with the regression equation forced through the origin, gives
for females, and
for males. Figure 10 illustrates examples of the goodness-of-fit provided by the estimated schedules to the observed model migration data. Two sets of estimated schedules are shown: those obtained with the observed index of child dependency (S12)and those found with the estimated index (i12), both calculated using the above regressions. In each case x1 and xh were set equal to the values given by the observed model migration schedules.
4.3
The Relational Perspective: The Logit Migration System
Among the most popular methods for estimating mortality from inadequate or defective data, is the so-called logit system developed by William Brass about twenty years ago
39
and now widely applied by demographers all over the world (Brass 197 1, Brass and Coale 1968, Carrier and Hobcraft 197 1, Hill and Trussell 1977, Zaba 1979). The logit approach to model schedules is founded on the assumption that different mortality schedules can be related to each other by a linear transformation of the logits of their respective survivorship probab~lities.That is, given an observed series of survivorship probabilities l(x) for ages x = 1,2,...,w, it is possible t o associate these observed series with a "standard" series l,(x) by means of the linear relationship
where, say,
The inverse of this function is
The principal result of this mathematical transformation of the nonlinear l(x) function is a more nearly linear function in x , with a range of minus and plus infinity rather than unity and zero. Given a standard schedule, such as the set of standard logits, Y,(x), proposed by Brass, a life table can be created by selecting appropriate values for y and p . In the Brass system y reflects the level of mortality and p defines the relationship between child and adult mortality. The closer y is t o zero and p t o unity, the more the estimated life table is like the standard. The logit perspective can be readily applied to migration schedules. Let .M(x) denote the age-specific migration rates of a schedule scaled to a unit GMR, and let .M,(x) denote the corresponding standard schedule. Taking logits of both sets of rates gives the logit migration system
and
where, for example,
The selection of a particular migration schedule as a standard reflects the belief that it is broadly representative of the age pattern of migration in the multiregional population
40
A. Rogers, L.J. Gzstro
system under consideration. (Our standard schedules will always have a unit GMR; hence the left subscript on .Y,(x) will be dropped.) To illustrate a number of calculations carried out with several sets of multiregional data, we shall adopt the national age profile as the standard in each case and strive to estimate r.egiona1outmigration age profiles by relating them to the national one. Specifically, given an m X m table of interregional migration flows for any age x , we divide each origin-destination-specific flow Oii(x) by the population in the origin region Ki(x) to define the associated age-specific migration rate Mii(x). Summing these over all origins and destinations gives the corresponding national rate M. .(x), and scaling all schedules to unit GMR gives .Mii(x) and .M..(x), respectively. Figure 1 1 presents national male standards for Sweden, the United Kingdom, Japan, and the Netherlands. (We shall deal only with graduated fits inasmuch as all of our nonSwedish data are for five-year age intervals and therefore need to be graduated first in order to provide single-year profiles by means of interpolation.) The differences in age profiles are marked. Only the Swedish and the United Kingdom standards exhibit a retirement peak. Japan's profile is described without such a peak because the age distribution of migrants given by the census data ends with the open interval of 65 years and over. The data for the Netherlands, on the other hand, show a definite upward slope at the post-labor force ages and therefore have been graduated with the 9-parameter model schedule with an upward slope. Regressing the logits of the age-specific outmigration rates of each region on those of its national standard (the GMRs of both first being scaled to unity) gives estimated values for 7 and p. Reversing the procedure and combining selected values of 7 and p with a national standard of logit values, identifies the following important regularity: whenever 7 = 2 ( p - 1 ) then the GMR of the estimated model schedule is approximately unity (Rogers and Castro 1981). Linear regressions of the form
fitted to our data for Sweden, the United Kingdom, Japan, and the Netherlands, consistently produce estimates for do and d l that are approximately equal t o 2 in magnitude = -2, and = +2. Thus and that differ only in sign, i.e.,
Differences in the national standard schedules illustrated in Figure 1 1 suggest that a single standard schedule may be a more restrictive assumption in migration analysis than in mortality studies. It therefore may be necessary to follow the Code-Demeny strategy of developing families of appropriate schedules (Code and Demeny 1966). The comparative analysis of national and interregional migration patterns carried out in section 3 identified at least three distinct families of age profiles. First, there was the 1 1-parameter basic model migration schedule with a retirement peak that adequately described a number of interregional flows, for example, the age profiles of outmigrants leaving capital regions such as Stockholm and London. The elimination of the retirement peak gave rise to the 7-parameter reduced form of this basic schedule, a form that was used to describe a large number of labor dominant profiles and the age patterns of migration schedules with a single openended age interval for the post-labor force population,
42
for example, Japan's migration schedules. Finally, the existence of a monotonically rising tail in migration schedules such as those exhibited by the Dutch data led to the definition of a third profile: the 9-parameter model migration schedule with art upward slope. Within each family of schedules, a number of key parameters or variables may be put forward in order to further classify different categories of migration profiles. For example, in section 3 we identified the special importance of the following aspects of shape and location along the age axis: 1. Peaking: early peaking versus late peaking (p,) 2. Dominance: child dependence versus labor dominance (6,,) 3. Asymmetry: labor symmetry versus labor asymmetry ( 0 , ) 4. Regularity: parental-shift regularity versus parental-shift irregularity (/3,,) These fundamental families and four key parameters give rise to a large variety of standard schedules. For example, even if the four key parameters are restricted to only dichotomous values, one already needs Z4 = 16 standard schedules. If, in addition, the sexes are to be differentiated, then 32 standard schedules are a minimum. A large number of standard schedules would make the logit approach a less desirable alternative. Hence we shall examine the feasibility of adopting only a single standard for both sexes and assume that the shape of the post-labor force part of the schedule may be determined exogenously. In tests of our logit migration system, therefore, we shall always set the postlabor force retirement peak or upward slope equal t o observed model schedule values. The similarity of the male and female median parameter values set out in Tables 10 and 11 (for Sweden, the United Kingdom, and Japan), suggests that one could use the average of the values for the two sexes to define a unisexual standard. A rough rounding of these averages would simplify matters even more. Table 17 presents the simplified basic standard parameters obtained in this way. The values of a,, a,, and c are initial values only and need t o be scaled proportionately t o ensure a unit GMR. Figure 12 illustrates the age profile of this simplified basic standard migration schedule.
TABLE 17 The simplified basic standard migration schedule.

-
Fundamental parameters
Fundamental ratios
We have noted before that when 7 = 0 and p = 1 , the estimated model schedule is identical t o the standard. Moreover since the GMR of the standard is always unity, values of 7 and p that satisfy the equality 7 = 2(p - 1) guarantee a GMR ofunity for the estimated schedule. What are the effects of other combinations of values for these two parameters?
FIGURE 12 Simplified basic standard migration schedule.
Figure 13 illustrates how the simplified basic standard schedule is transformed when
-y and p are assigned particular pairs of values. Figure 13(a) shows that fixing -y = 0 and increasing p from 0.75 to 1.25 lowers the schedule, giving migration rates that are smaller in value than those of the standard. On the other hand, fixing p = 0.75, and increasing -y from -1 to 0 raises the schedule, according to Figure 13(b). Finally, fixing GMR = 1 by selecting values of -y and p that satisfy the equality -y = 2(p - 1) shows that as -y and p
both increase, so does the degree of labor dominance exhibited by the estimated schedule. For example, moving from an estimated schedule with -y = --0.5 and p = 0.75 to one with -y = 0.5 and p = 1.25 does not alter the area under the curve (GMR = l ) , but it does increase its labor dominance (Figure 13(c)). Given a standard schedule and a few observed rates, such as M(0-4) and M(20-24), for example, how can one find estimates for -y and p , and with those estimates go on to obtain the entire estimated schedule? First, taking logits of the two observed migration rates gives Y(0-4) and Y(20-24) and associating these two logits with the pair of corresponding logits for the standard gives
Solving these two equations in two unknowns givescrude estimates for -y and p , and applying them to the standard schedule's full set of logits results in a set of logits for the estimated schedule. From these one can obtain the migration rates, as shown earlier. Tests of such a procedure with the migration data for Sweden, the United Kingdom, Japan, and the Netherlands, however, indicate that the method is very erratic in the goodness-of-fits that it produces and, therefore, more refined procedures are necessary. Such procedures (for tlie case of mortality) are described in the literature on the Brass logit system (for example, in Brass 1975, Carrier and Goh 1972). A reasonable first approximation t o an improved estimation method for the case of migration is suggested by the regression approach described in subsection 4.2. Imagine a
A. Rogers, I..J. Olstro
45
regression of p on the M ratio, M(0-4)/M(20 -24).Starting with the simplified basic standard migration schedule and varying p within the range of observed values, one may obtain a corresponding set of M ratios. Associating p and the M ratio in this way, one may proceed further and use the relational equation to estimate ? from p^:
A further simplification can be made by forcing the regression line to pass through the origin. Since the resulting regression coefficient has a negative sign and the intercept exhibits roughly the same absolute value, but with a positive sign, the regressionequations take on the form
where M = M(0-4)/M(20- 24). Given a standard schedule and estimates for y and p , one can proceed to compute the associated estimated model migration schedule. Figure 14 illustrates representative examples of the goodness-of-fit obtained using this procedure. Two estimated schedules are illustrated with each observed model migration schedule: those calculated with the interpolated 85 single-year-of-age observations and the resulting least-squares estimates of y and p , and those computed using the above regression equations of p on the M ratio. Although the fits are moderately successful, it is clear that further study of this problem is necessary.
CONCLUSION
This report began with the observation that empirical regularities characterize observed migration schedules in ways that are no less important than the corresponding wellestablished regularities in observed fertility or mortality schedules. Section 2 was devoted to defining mathematically such regularities in observed migration schedules in order to exploit the notational, computational, and analytical advantages that such a formulation provides. Section 3 reported on the results of an examination of over 500 migration schedules that underscored the broad generality of the model migration schedule proposed and helped to identify a number of families of such schedules. Regularities in age profiles lead naturally to the development of hypothetical model migration schedules that might be suitable for studies of populations with inadequate or defective data. Drawing on techniques used in the corresponding literature in fertility and mortality, section 4 develops procedures for inferring migration patterns in the absence of accurate migration data. Of what use, then, is the model migration schedule defined in this study? What are some of its concrete practical applications? The model migration schedule may be used to graduate observed data, thereby smoothing out irregularities and ascribing to the data summary measures that can be used for comparative analysis. It may be used to interpolate to single years of age, observed migration schedules that are reported for wider age intervals. Assessments of the reliability
Stockholm
London
O-'"
Tokyo
0.~~1
Amsterdam
FIGURE 14 The fits of the relational approach when using the estimated parameters from 85 observations (- - -) and the parameter from the observed M ratio (. . . .) compared with the observed () data for the female populations o f Stockholm, London, Tokyo, and Amsterdam.
47
of empirical migration data and indications of appropriate strategies for their correction are aided by the availability of standard families of migration schedules. Finally, such schedules also may be used to help resolve problems caused by missing data. The analysis of national migration age patterns reported in this study seeks t o demonstrate the utility of examining the regularities in age profile exhibited by empirical schedules of interregional migration. Although data limitations have restricted some of the findings to conjectures, a modest start has been made. It is hoped that the results reported here will induce others to devote more attention t o this topic.
ACKNOWLEDGMENTS The authors are grateful to the many national collaborating scholars who have participated in IIASA's Comparative Migration and Settlement Study. This report could not have been written without the data bank produced by their collective efforts. Thanks also go to Richard Raquillet for his contributions to the early phases of this study and to Walter Kogler for his untiring efforts on our behalf in front of a console in IIASA's computer center. Kao-Lee Liaw, Philip Rees, Warren Sanderson, and an anonymous reviewer offered several useful comments on an earlier draft.
REFERENCES
Andersson, A. and I. Holmberg (1980) Migration and Settlement: 3. Sweden. RR-80-5. Laxenburg, Austria: International Institute for Applied Systems Analysis. Bard, Y. (1974) Nonlinear Parameter Estimation. New York: Academic Press. Benson, M. (1979) Parameter Fitting in Dynamic Models. Ecological Modelling 6:97-115. Brass, W. (1971) On the scale of mortality. Pages 69-1 10 in Biological Aspects of Demography, edited by W. Brass. London: Taylor and Francis, Ltd. Brass, W. (1975) Methods for Estimating Fertility and Mortality from Limited and Defective Data. Chapel Hill, North Carolina: Laboratories for Population Statistics, University of North Carolina at Chapel Hill. Brass, W. and A. Code (1968) Methods of analysis and estimation. Pages 88--150 in The Demography of Tropical Africa, edited by W. Brass et al. Princeton, New Jersey: Princeton University Press. Brown, K.M. and J.E. Dennis (1972) Derivative free analogues of the Levenberg-Marquardt and Gauss algorithms for nonlinear least squares approximations. Numerische Mathematik 18:289-297. Carrier, N. and J. Hobcraft (1971) Demographic Estimation for Developing Societies. London: Population Investigation Committee, London School of Economics. Carrier, N. and T. Goh (1972) The validation of Brass's model life table system. Population Studies 26 (1):29-5 1. Coale, A. (1977) The development of new models of nuptiality and fertility. Population, special issue: 131-154. Coale, A. and P. Demeny (1966) Regional Model Life Tables and Stable Populations. Princeton, New Jersey: Princeton University Press. Coale, A.J. and D.R. McNeil (1972) The distribution by age of the frequency of first marriage in a female cohort. Journal of the American Statistical Association 67:743 749. Coale, A. and J. Trussell (1974) Model fertility schedules: variations in the age structure of childbearing in human populations. Population Index 40(2):185-206. Fiacco, A. and G. McCormick (1968) Nonlinear Programming: Sequential Unconstrained Minimization Techniques. New York: Wiley.
48
A. Rogers, L.J. G s t r o
Gumbel, E.J. (1941) The return period of flood flows. Annals of Mathematical Statistics 12:163-190. Hill. K. and J. Trussell (1977) Further developments in indirect mortality estimation. Population Studies 31(2):313--334. KimbaU, B.F. (1946) Sufficient statistical estimation functions for the parameters of the distribution of maximum values. Annals of Mathematical Statistics 17:299--309. Levenberg, K. (1944) A method for the solution of certain nonlinear problems in least squares. Quarterly of Applied Mathematics 2:164-168. Marquardt, D.W. (1963) An algorithm for least-squares estimation of nonlinear parameters. SIAM, Journal of Numerical Analysis 11:431-441. McNeil, D.R., T.J. Trussell, and J.C. Turner (1977) Spline interpolation of demographic data. Demography 14:245--252. Rees, P.H. (1977) The measurement of migration, from census data and other sources. Environment and Planning A 1 :247-260. Rees, P.H. (1979) Migration and Settlement: 1. United Kingdom. RR-79-3. Laxenburg, Austria: International lnstitute for Applied Systems Analysis. Rogers, A. (1973a) The multiregional life table. The Journal of Mathematical Sociology 3:127--137. Rogers. A. (1973b) The mathematics of multiregional demographic growth. Environment and Planning 5:3-29. Rogers, A. (1975) Introduction to Multiregional Mathematical Demography. New York: Wiley. Rogers, A. (1976a) The Comparative Migration and Settlement Study: A Summary of Workshop Proceedings and Conclusions. RM-76-01. Laxenburg, Austria: lnternational lnstitute for Applied Systems Analysis. Rogers, A. (1976b) Two Methodological Notes o n Spatial Population Dynamics in the Soviet Union. RM-7648. Laxenburg, Austria: lnternational lnstitute for Applied Systems Analysis. Rogers, A., editor (1978) Migration and settlement: selected essays. Environment and Planning A 10 (5):469-617. Rogers, A. and L. Castro (1981) Model Schedules in Multistate Demographic Analysis: The Case of Migration. WP-81-22. Laxenburg, Austria: lnternational lnstitute for Applied Systems Analysis. Rogers, A. and J. Ledent (1976) Increment-Decrement life tables: a comment. Demography 13:287290. Rogers, A., R. Raquillet, and L.J. Castro (1978) Model migration schedules and their applications. Environment and Planning A 10:475-502. Rogers, A. and F. Wiekens (1978) Migration and Settlement: Measurement and Analysis. RR-78-13. Laxenburg, Austria: lnternational lnstitute for Applied Systems Analysis. United Nations (1955) Age and Sex Patterns of Mortality: Model Life Tables for Underdeveloped Countries. New York: United Nations. Willekens, F . and A. Rogers (1978) Spatial Population Analysis: Methods and Computer Programs. RR-78-18. Laxenburg, Austria: lnternational lnstitute for Applied Systems Analysis. Zaba, B. (1979) The four-parameter logit life table system. Population Studies 33(1):79- 100.
APPENDIX A
NONLINEAR PARAMETER ESTIMATION WITH MODEL MIGRATION SCHEDULES This appendix briefly illustrates the mathematical programming procedure used to estimate the parameters of the model migration schedule. The nonlinear estimation problem may be defined as the search for the "best" parameter values for the function
in the sense that a predefined objective function is minimized when the parameters take on these values. This problem is the classical one of nonlinear parameter estimation in unconstrained optimization. All of the available methods start with a set of given initial conditions, or initial guesses of the parameter values, in the search for better estimates following specific convergence criteria. The iterative sequence ends after a finite number of iterations, and the solution is accepted as giving the best estimates for the parameters. The problem of selecting an effective method has been usefully summarized by Bard (1 974, p. 84) as follows:
. . . no single method has emerged which is best for the solution of all nonlinear programming problems. One cannot even hope that a "best" method will ever be found, since problems vary so much in size and nature. For parameter estimation problems we must seek methods which are particularly suitable to the special nature of these problems which may be characterized as follows:
1. A relatively small number of unknowns, rarely exceeding a dozen or SO. 2. A highly nonlinear (though continuous and differentiable) objective function, whose computation is often very time consuming. 3. A relatively small number (sometimes zero) of inequality constraints. Those are usually of a very simple nature, e.g., upper and lower bounds. 4. No equality constraints, except in the case of exact structural models (where, incidentally, the number of unknowns is large) . . .
50
A . Rogers, 1.J. Casrro
For computational convenience, we have chosen the Marquardt method (Levenberg 1944, Marquardt 1963). This method seeks out a parameter vector P* that minimizes the following objective function:
where fp is the residual vector. For the case of a model schedule with a retirement peak, vector P has the following elements:
where T denotes transposition. The elements of the vector fp can be computed by either of the following two expressions:
where M(x) is the observed value at age x and Mp(x) is the estimated value using eq. (Al) and a given vector P of parameter estimates. By introducing eq. (A4) in the objective function set out in eq. (A2), the sum of squares is minimized; if, on the other hand, eq. (AS) is introduced instead, the chi-square statistic is minimized. In matrix notation, the Levenberg--Marquardt method follows the iterative sequence
where X is a non-negative parameter adjusted to ensure that at each iteration the function (A2) is reduced, Jq denotes the Jacobian matrix of +(P) evaluated at the q iteration, and D is a diagonal matrix equal to the diagonal of JTJ. The principal difficulty in nonlinear parameter estimation is that of convergence, and the method discussed here is no exception. The algorithm starts out by assumingsome initial parameters, and then a new vector P i s estimated according to the value of A, which in turn is also modified following some gradient criteria. Once some given stopping values are achieved, vector P* is assumed to be the optimum. In some cases, however, this P* reflects local minima that may be improved with better initial conditions and a different set of gradient criteria. Using the data described in this report, several experiments were carried out to examine the variation in parameter estimates that could result from different initial conditions (assuming Newton's gradient criteria).t Among the cases studied, the most significant differences were found for the vector P with 11 parameters, principally among the parameters of the retirement component. For schedules without the retirement peak, the vector P* shows n o variation in most cases.
t F o r a complete description o f gradient methods, see Fiacco and McCormick 1968, Bard 1974.
51
The impact of the gradient criteria on the optimal vector P* was also analyzed, using the Newton and the Steepest Descent methods. The effects of these two alternatives were reflected in the computing times but not in the values of the vector P*. Nevertheless, Bard (1974) has suggested that both methods can create problems in the estimation, and therefore they should be used with caution in order to avoid unrealistic parameter estimates. It appears that the initial parameter values may be improved by means of an interactive approach suggested by Benson (1979).
APPENDIX B
SUMMARY STATISTICS OF NATIONAL PARAMETERS AND VARIABLES OF THE REDUCED SETS OF OBSERVED MODEL MIGRATION SCHEDULES
Legend Observed gross migraproduction rate Unit gross migraproduction rate Goodness-of-fit index E (mean absolute error as a percentage of the observed mean) a,, level of pre-labor force component a , , rate of descent of pre-labor force component a,, level of labor force component p,, mean age of labor force component a,, rate of descent of labor force component A,, rate of ascent of labor force component a,, level of post-labor force component p,, mean age of post-labor force component a , , rate of descent of post-labor force component A,, rate of ascent of post-labor force component c, constant component 5, mean age of migration schedule Percentage of GMR in 0-14 age interval Percentage of GMR in 15-64 age interval Percentage of GMR in 65 and over age interval
a1 alpha1 a2 mu2 alpha2 lambda2 a3 mu3 alpha3 lambda3

C
mean age %(O-14) %(IS-64) %(65+ ) deltal c deltal 2 delta32 beta1 2 sigma2 sigma3 x low x high x ret. x shift a b
4, = a1lc 61, = a l l % 63, = a3/a, 01, = a l l % a, = A,la, 0 3 = &/a,

X,,
low point
xh ,high point x, ,retirement peak X, labor force shift A , parental shift

B,jump
Summary statistics for Swcdish males without a retiremcnt peak using single year of age data: 48 schedules.
1owes t value
highest value
mean v a l u e
median
mode
s t d . dev.
s t d . dev. / mean
grnr ( o b s ) gmr ( m m s ) mae%m a1 alpha1 a2 mu2 a 1p h a 2 1amb d a 2 a3 mu3 alpha3 1 ambda3

C
mean a g e %( 0-14) % ( 15-64) %(65+ ) del t a l c del t a l 2 del ta32 beta12 s i gma2 s i gma3 x low x high x ret. x shift a
b
Summary statistics for Swedish males with a retirement peak using single year of age data: 9 schedules.
1owes t value
gmr ( o b s ) gmr ( m m s ) maeXm
highest value
mean v a l u e
median
mode
s t d . dev.
mu2 alpha2 1amb d a 2 a3 mu3 a 1p h a 3 1 amb d a 3

C
mean a g e X ( 8-14) % ( 15-64) %(65+ ) del t a l c del ta12 del ta32 be ta12 s i gma2 s i gma3 x low x high x ret. x shift a
b
Summary statistics for Swedish females without a retirement peak using single year of age data: 54 schedules.
1owes t value gmr ( o b s ) gmr ( m m s ) mae%m a1 alpha1 a2 mu2 alpha2 1 amb d a 2 a3 mu3 alpha3 1ambda3
C
highest value
mean v a l u e
medi an
mode
s t d . dev.
s t d . dev. / mean
mean a g e % ( 0-14) % ( 15-64) %(65+ ) del t a l c del t a l 2 de 1 t a 3 2 be t a l 2 s i gma2 s i gma3 x low x high x ret. x shift a b
Summary statistics for Swedish females with a retirement peak using single year of age data: 3 schedules.
1 owes t
value
highest value
mean v a l u e
median
mode
s t d . dev.
gmr ( o b s ) gmr ( m m s ) maeZm a1 alpha1 a2

1 abb da2 a3 mu3 alpha3 1 ambda3
C
mean a g e % ( 0-14) % ( 15-64) %(65+ ) del t a l c del t a l 2 del ta32 be t a l 2 s i gma2 s i gma3 x low x high x ret. x shift a b
Summary statistics for males o f the United Kingdom without a retirement peak: 59 schedules.
1 owes t value
highest value
mean v a l u e
median
mode
s t d . dev.
s t d . dev. / mean
gmr ( o b s ) gmr ( m m s ) maeZm a1 alpha1 a2 mu2 alpha2 1ambda2 a3 mu3 alpha3 1ambda3
C
mean age % ( 0-14) % ( 15-64)
%(65+ ) del t a l c del ta12 del ta32 beta12 s i gma2 s i gma3 x low x high x ret. x shift a
b
59
Summary statistics for females of the United Kingdom without a retirement peak: 61 schedules.
1owes t value
gmr ( o b s )
highest value
mean v a l u e
medi an
mode
s t d . dev.
s t d . dev. / mean
gmr ( m m s ) maeXm a1 alpha1 a2 mu2 alpha2 1 amb da2 a3 mu3 alpha3 1amb da3
C
mean age % ( 8-14) % ( 15-64) %(65+ 1 del t a l c del t a 1 2 de 1 t a 3 2 be t a 1 2 s i gma2 s i gma3 x low x high x ret. x shift a b
Summary statistics for females of the United Kingdom with a retirement peak: 21 schedules.
5.
lowest value gmr (obs) gmr (mrns) mae%m a1 alpha1 0.04829 1 .00000 4.7497 1 0.00805 0.02459
highest value 0.3430 1 1 .00000 22.13955 0.04165 0.24502
mean value 0.14933 1 .00000 9.20055 0.01794 0.08924
median 0.13736 1 .00000 8.84962 0.01517 0.09505
mode
s t d . dev.
s t d . dev. / mean 0.5590 1 0.00000 0.46487 0.45765 0.61239 0.361 13 0.23037 0.59595 0.39 148 1 .42077 0.11586 0.80572 1 .05522 0.3504 1 0.10250 0.18305
9
2
7 '
! -i
1 a b da2 m a3 mu3 alpha3 1 ambda3

C
mean age % ( 8-14] % ( 15-64) %(65+ 1 del t a l c del t a l 2 de 1 t a32 be ta12 s i gma2 s i gma3 x low
x high x ret. x shift
a
b
62
A . Rogers, L.J. Costro
'no b(D0a 0. .
@-
4 0)
V P I
0C (d
eo r
aaNm am-b aamm aQ-P ao-a aNNa

. . a .
.4
-2
a-aw a-am aaaa amaN a(Daa .... avaa
Summary statistics for Japanese females without a retirement peak: 5 7 schedules.
lowest value gmr tobs) gmr (mms) mae%m a1 alpha1 a2 mu2 alpha2 1 ambda2 a3 mu 3 alpha3 1 ambda3
C
highest value
mean value
median
mode
s t d . dev. 0.3365 1 0.00000 5.10822 0.00874 0.03604 0.03158 4.98334 0.04493 0.16910 0.00000 0.00000 0.00000 0.00000 0.00135 2.41 142 4.7997 1 5.40643 2.97760 25.2697 1 0.16540 0.00000 0.29367 1 .57000 0.00000 2.6281 1 3.25665 0.00000 2.2461 1 2.18864 0.01340
s t d . dev. /mean
3
$
3 5-
1 .35027 0.00000 0.4595 1 0.42507 0.30852 0.37210 0.23370 0.29654 0.48352 0.00000 0.00000 0.00000 0.00000
mean age % ( 8-14) % ( 15-64) %(65+ ) del t a l c del t a l 2 de 1 t a32 beta12 s i gma2 s i gma3 x low x high x ret. x shift a b
Summary statistics for males of the Netherlands with a retirement slope: 10 schedules.
1owes t value gmr ( o b s ) gmr ( m m s ) maeXm a1 alpha1 a2 mu 2 a 1p h a 2 1 ambda2 a3 mu3 alpha3 1 amb d a 3
C
highest value
mean v a l u e
medi an
mode
s t d . dev.
s t d . dev. / mean
mean a g e % ( 0-14) % ( 15-64)
%(65+
del t a l c del t a l 2 de 1 t a 3 2 beta12 s i gma2 s i gma3 x low x high x ret. x shift a
b
Summary statistics for females of the Netherlands with a retirement slope: 10 schedules.
1 owes t value
highest value
mean value
median
mode
s t d . dev.
s t d . dev. /mean
$
gmr (obs) gmr (mms) me m aX a1 alpha1 a2 m2 u alpha2 1 a b da2 m a3 mu3 alpha3 l ambda3
C
mean age % ( 0-14) % ( 15-64) %(65+ ) del t a l c del t a l 2 del ta32 be t a l 2 s i gma2 s i gma3 x low x high x ret. x shift a
b
0.11394 0.00000 0.25556 0.10426 0.16562 0.15718 0.01345 0.11408 0.09280 1 .40559 0.00000 0.33289 0.00000 0.1675 1 0.02030 0.08764 0.02603 0.06620 0 . 18100 0.2343 1 1 .38423 0.17406 0.19596 0 . a0000 0.04906 0.02368 0.00080 0.03837 0.02714 0.12143
Summary statistics for the total population o f the Soviet Union without a retirement peak: 58 schedules.
1owes t value gmr ( o b s ) gmr ( m m s ) maeXm a1 alpha1 a2 mu 2 alpha2 1 ambda2 a3 mu3 alpha3 1 ambda3
C
highest value
mean v a l u e
median
mode
s t d . dev.
s t d . dev. / mean
mean age % ( 0-14) % ( 15-64) %(65+ ) del t a l c del t a l 2 de 1 t a 3 2 beta12 s i gma2 s i gma3 x low x high x ret. x shift a b
Summary statistics for the total population of the United States with a retirement peak: 8 schedules.
1 owes t value
gmr tobs) gmr t m ) m s me m aX a1 alpha1 a2 mu 2 a 1pha2 1 a b da2 m a3 m3 u alpha3 1 a b da3 m
C
highest value
mean value
medi an
mode
s t d . dev. 0.17155 0.00000 2.18925 0.0046 1 0.02920 0.01414 0.54137 0.03137 0.11553 0.00269 8.55974 0.16264 0.04554 0.00087 1 .53949 1 .33968 2.63248 2.24953 4.77822 0.09357 0.04826 0.29015 1 .82038 0.27618 0.38767 0.4622 1 3 . 9 1673 0.66965 1 .52249 0.00650
mean age % ( 8-14) % ( 15-64) %(65+ ) d e l talc del ta12 de 1 ta32 beta12 s i gma2 s i gma3 x low x high x ret. x shift a
b
Summary statistics for the total population of Hungary without a retirement peak: 7 schedules.
1owes t value gmr ( o b s ) gmr ( m m s ) mae%m a1 alpha1 a2 mu 2 alpha2 1 amb d a 2 a3 mu 3 alpha3 1 ambda3
C
highest value
mean v a l u e
median
mode
s t d . dev.
s t d . dev. / mean
mean age % ( 0-14) % ( 15-64) %(65+ ) del t a l c del t a l 2 del ta32 beta12 s i gma2 s i gma3 x low x high x ret. x shift a b
Summary statistics for the total population of Hungary with a retirement slope: 25 schedules.
3
mode s t d . dev. 1.15148 0.88000 2.26151 0.00448 0.05620 0.81350 1.04162 0.02715 0.03984 0.00039 0.00000 0.01448 0.00000 0.00098 2.51858 2.8066 1 2.70972 3 . 12025 4.57756 0.05060 0.00367 0.32345 0.59299 0.00000 0.54132 0.51739 0.00000 0.52667 3.58805 0.00349 s t d . dev. /mean
3
lowest value gmr (obs) gmr (mms) me m aX a1 alpha1 a2 m2 u alpha2 1 ambda2 a3 mu3 alpha3 1 ambda3
C
highest value
mean value
median
-$
;a
1 .2478 1 0.88000 0.26545 0.2993 1 0.29 168
mean age % ( 8-14) % ( 15-64) %(65+ ) del t a l c del ta12 del ta32 beta12 s i gma2 s i gma3 x low x high x ret. x shift a
b
APPENDIX C
NATIONAL PARAMETERS AND VARIABLES OF THE FULL SETS OF OBSERVED MODEL MIGRATION SCHEDULES C.l C.2 C.3 C.4 Sweden (1974) United Kingdom (1970) Japan (1970) Netherlands (1 974) C.5 C.6 C.7 Soviet Union (1974) United States (1970) Hungary (1974)
Legend
Observed gross migraproduction rate Unit gross migraproduction rate Goodness-of-fit index E (mean absolute error as a percentage of the observed mean) a,, level of pre-labor force component a , , rate of descent of pre-labor force component a,, level of labor force component p,, mean age of labor force component a , , rate of descent of labor force component A,, rate of ascent of labor force component a,, level of post-labor force component p,, mean age of post-labor force component a , , rate of descent of post-labor force component A,, rate of ascent of post-labor force component c , constant component Z, mean age of migration schedule Percentage of GMR in 0-14 age interval Percentage of GMR in 15-64 age interval Percentage of GMR in 65 and over age interval
&Ic= a , l c 612 = a , l a , &32 = ' 3 1 ' 2 012 = a1/az a2 = A 2 1 0 2 0 3 = A,/%
XI,
a1 alpha 1 a2 mu2 alpha2 lambda2 a3 mu3 alpha3 lambda3

C
mean age %(O-14) %(l5-64) %(65+ ) deltalc delta 12 delta32 beta12 sigma2 sigma3 x low x high x ret. x shift a b
low point
x, ,high point x,, retirement peak
X, labor force shift A , parental shift B , jump
72
APPENDIX C . l
Sweden (1974).*
( ,
8. Upper North
:J 6. North Middle
FIGURE C . l
Map of the regional aggregation o f Sweden used for this study.
*Input data are for single years of age. This is the only country in the comparative study for which this is the case.
Males.
gmr ( o b s ) gmr (mms) mae Z m a1 alpha1 a2 mu2 alpha2 1 ambda2 a3 mu3 alpha3 1 ambda3
C
mean a g e %( 0-14) X ( 15-64) %(65+ ) del t a l c del ta12 de 1 t a 3 2 beta12 s i gma2 s i gma3 x low x high x ret. x shift a b
1 2 3 4
sweden males sweden males sweden males sweden males
1 to 2 1 to 3 1 to 4 1 to 5
5 sweden males 6 sweden males 7 sweden males 8 sweden males
1 to 6 1 to 7 1 to 8 1 to the rest
APPENDIX C .1 (continued).
gmr ( o b s ) gmr (mms) maeZm a1 al'phal a2 mu2 alpha2 1 amb da2 a3 m3 u alpha3 1 amb da3
C
mean age Z( 0-14) % ( 15-64) %(65+ ) del t a l c del t a l 2 del ta32 beta12 s i gma2 s i gma3 x low x high x ret. x shift a b
1 2 3 4 5
sweden males sweden males sweden males sweden males sweden males
2 to 1 2 to 2 2 to 3 2 to 4 2 to 5
2 to 6 2 to 7 2 to 8 2 t o the rest
gmr Cobs) gmr (mms) maeZm
alpha2 1 ambda2 a3 mu3 alpha3 1 ambda3

C
mean age Z( 0 - 1 4 ) Z( 15-64) Z(65+ ) del t a l c del t a l 2 de 1 t a32 beta12 s i gma2 s i gma3 x low x high x ret. x shift a b
2 3 4 5
1 sweden males sweden males sweden males sweden males sweden males
3 to 1 3 to 2 3 to 3 3 to 4 3 to 5
APPENDIX C .1
(continued).
gmr ( o b s ) gmr ( m m s ) maeXm a1 alpha1 a2 mu2 a 1p h a 2 1ambda2 a3 mu3 a 1p h a 3 1ambda3

C
mean age X ( 0-14) X ( 15-64) del t a l c del t a l 2 del ta32 be t a 1 2 s i gma2 s i gma3 x low x high x ret. x shift a b
1 sweden males 2 sweden males 3 sweden males 4 sweden males 5 sweden males
4 to 1 4 to 2 4 to 3 4 to 4 4 to 5
6 sweden males 7 sweden males 8 sweden males

9 sweden males
gmr ( o b s ) gmr ( m m s ) maeZrn a1 alpha1 a2 m 2 u alpha2 1 ambda2 a3 mu3 a 1p h a 3 1ambda3

C
mean a g e Z ( 8-14) Z ( 15-64) Z(65+ 1 del t a l c del t a l 2 del ta32 beta12 s i gma2 s i gma3 x low x high x ret. x shift a b
1 2 3 4 5
5 to 1 5 to 2 5 to 3 5 to 4 5 to 5
6 7 8 9
sweden mates sweden males sweden males sweden males
gmr ( o b s ) gmr (mms) maeZm a1 alpha1 a2 mu2 alpha2 1 ambda2 a3 mu3 alpha3 1 ambda3
C
mean age X ( 0-14) X ( 15-64) X(65+ ) del t a l c del t a l 2 del ta32 beta12 s i gma2 s i gma3 x low x high x ret. x shift a b
1 2 3 4 5
6 to 1 6 to 2 6 to 3 6 to 4 6 to 5
6 7 8 9
gmr f o b s ) gmr (mms) maeh a1 alpha1 a2 mu2 alpha2 1 amb d a 2 a3 mu3 alpha3 1ambda3
C
mean a g e Z( 0 - 1 4 ) Z( 15-64) %(65+ ) del t a l c del t a l 2 del ta32 beta12 s i gma2 s i gma3 x low x high x ret. x shift
a
1 2 3 4 5
7 7 7 7 7
to 1 to 2 to 3 to 4 to 5
6 7 8 9
APPENDIX C. 1 (continued).
g m r (obs) g m r (mms) maeXm a1 alpha1 a2 mu2 alpha2 1 ambda2 a3 mu3 alpha3 1 ambda3
C
mean age X( 0-14) % ( 15-64) %(65+ ) del talc del tal2 de 1 t a 3 2 beta12 s igma2 s i gma3 x low x high x ret. x shift a b
1 2 3 4 5
8 to 1 8 to 2 8 to 3 8 to 4 8 to 5
6 7 8 9
Females.
gmr ( o b s ) gmr ( m m s ) maeZm a1
mu3 alpha3 1 ambda3

0
mean a g e %( 8-14) % ( 15-64) %(65+ ) del t a l c del t a l 2 de 1t a 3 2 beta12 s i gma2 s i gma3 x low x high r ret. x shift a b
1 2 3 4
sweden females sweden females sweden females sweden females
1 to 2 1 to 3 1 to 4 1 to 5
5 sweden females 6 sweden females

7 sweden females 8 sweden females
gmr ( o b s ) gmr ( m m s ) maeZm a1
a3 mu3 alpha3 1amb d a 3

C
mean a g e Z( 8-14] 2( 15-64) Z(65+ ) del talc del t a l 2 del ta32 be t a l 2 s i gma2 s i gma3 x low x high x ret. x shift a b
1 2 3 4 5
sweden females sweden females sweden females sweden females sweden females
2 to 1 2 to 2 2 to 3 2 to 4 2 to 5
6 sweden females
7 sweden females 8 sweden females 9 sweden females
s 01 E
1sal aq4 01 E 8 01 E L 01 E g 01 E sapuraj uapams saleuaj uapams salemaj uapams sapuraj uapams
6 8
L g
p 01 E
z 04
E 01 E E I01 E
sapuraj uapams s s a p a j uapams p sapuaj uapams E sapmaj uapams sapuraj uapams I
Z9IP0'0 69ELZ 'LZ P1866'S
1 P0E0' 0 ILE6L:'LZ ZI0EZ.S 00000'0 0P08Z' zz 820S0'LI 00000'0 a621'6 11898'0 00000'0 16MS'0 ZPI I Z ' 6 Z IS891 'P 69ZP0' 0L 0888L' SZ 60E6S ' 9Z 60100'0 00000'0 00000'0 00000'0 00000'0 9LEPL ' 0 LP180'0 109LP'61 L19S0'0 ZL0L0' 0 ELI&@'@ ZLISL'ZS 68888.1 EWE0 ' 0 8
690P0 ' 0 LE0LL' 9Z 910P8.9 00000'0 8E0ZL ' I Z CZ088 ' P I 00000'0 08L0 1 ' E Z69S8 ' 0 00000'0 EP80C ' 0 9Sl l P ' 0 1 00SSS' L 669SE' 0L 10880 ' ZZ Z661S'LZ 89200' 0 00000'0 00000'0 00000 ' 0 00000'0 9S6EP ' 0 PPIPI ' 0 EZI I Z ' 6 I 6c0643'0 0z1z1 ' 0 88LZ0'0 8P6PL ' S I 00000'1 ESZLC ' 0 S
9S1 P 0 ' 0 0LES I ' 9Z 11089'P 00000 ' 0 9E09Se0Z SZ088 ' S I 88888'0 PZ9SS ' P SEZP I ' 0 00000'0 82L61 ' 0 S18E9'L.Z 6 x 6 0 '8 8999L'ZL 1 L N 1 '61 LIESZ'62 SS000'0 00000'0 00000'0 00000'0 00000'0 9L09L' 0 L699 1 ' 0 LSP6S ' 8 1 SLLL0' 0 LLEZ0'0 ES10'0 L0L61'81 00000'1 ESILE'0 P
APPENDIX C.l
(continued).
gmr f o b s ) gmr f m m s ) maeZm a1 alpha1 a2 mu2 alpha2 1 ambda2 a3 mu3 alpha3 1 ambda3
C
mean age X ( 0-14) 2 1 15-64) 2(65+ ) del t a l c del ta12 del ta32 beta12 s i gma2 s i gma3 x low x high x ret. x shift a b
1 sweden females
2 sweden females
3 sweden females 4 sweden females 5 sweden females
4 to 1 4 to 2 4 to 3 4 to 4 4 to 5
6 7 8 9
gmr ( o b s ) gmr ( m m s ) maeXrn a1 alpha1 a2 mu2 alpha2 1 ambda2 a3 mu3 alpha3 1 ambda3
C
mean age X ( 8-14] X ( 15-64) 2(65+ ) del t a l c del t a l 2 del ta32 beta12 sigma2 s i gma3 x low x high x ret. x shift a b
1 2 3 4 5
5 to 1 5 to 2 5 to 3 5 to 4 5 to 5
6 7 8 9
5 5 5 5
to 6 to 7 to 8 to the rest
gmr Cobs) gmr ( m m s ) maeZm a1 alpha1 a2 mu2 a 1p h a 2 1ambda2 a3 mu3 alpha3 1amb d a 3
C
mean a g e Z ( 8-14) X ( 15-64) 2(65+ ) del talc del ta12 del ta32 beta12 s i gma2 s i gma3 x low x high x ret. x shift a b
1 2 3 4 5
6 to 1 6 to 2 6 to 3 6 to 4 6 to 5
6 7 8 9
gmr ( o b s ) gmr (mms) maeXm a1 alpha1 a2 mu2 alpha2 1 ambda2 a3 mu3 alpha3 1 ambda3
C
mean age % ( 0-14) 2 ( 15-64) %(65+ ) del t a l c del t a l 2 del ta32 beta12 s i gma2 s i gma3 r low r high r ret. r shift a b
1 2 3 4 5
7 to 1 7 to 2 7 to 3 7 to 4 7 to 5
6 7 8 9
APPENDIX C. 1 (continued).
gmr ( o b s ) gmr ( m m s ) maeXm a1
a3 mu3 alpha3 1ambda3

C
mean a g e X( 0-14) X( 15-64) %(65* ) del t a l c del t a l 2 del ta32 be t a l 2 s i gma2 s i gma3 x low x high x ret. x shift a b
1 2 3 4 5
8 to 1 8 to 2 8 to 3 8 to 4 8 to 5
6 7 8 9
8 to 6 8to7 8 to 8 8 to the rest
APPENDIX C.2 United Kingdom (1970).*
'side
FIGURE C.2
Map of the regional aggregation of the United Kingdom used for this study.
*Due to lack of data, Northern Ireland has been omitted as a region. Despite this we refer to the nation as the United Kingdom (and not Great Britain) in order t o maintain consistency with the IIASA case study report (Rees 1979).
90
A. Rogers, L.J. Gzstro
NN(?b
(D
Ega2 ~
-w
a :
a-zaaa
. .??? !a9 a a .a . . a m .YP P N a .Q?l m a m 9 a a . . a.m?b? a a a F .&s !-.
?#mzg
-8O
(?-a-
Q.J? .Y?
a l $. ~ . ~. . g. ~ s ~ i . ~ i 288i ~ ~ ~ ~ ~ 8 i
hl-a-(?(?
a - z a a a z a a a a 6 a-F ~ m ~ ~ a a a - = a ~ ~ a v ~ a a
.l?
888% ? % .??.? .P .?
gR
a-~aaa~aaaaaaa-ab-Qaa-rna=~aQ) (?maRa
a - ~ a a a z & a a a a a a bhlaPaaaCVa--aa
......
8-F
-CV
Ra
a--aaa
CV
aaaaaaa - ~ ~ m ~ a a a a = ~ a m ~ a
a VC a
(?
-hl
2 !
I-mm-5
0 0 0 0 0 C C C C C 0
a - ~ a a a ~ a a a a a a a ~ ~ p a . a a a a a a ~ a ( ? h l4 a
-(?
4 4 4 4
a-=aaa~aaaaaaagEs~vaaaaa~~a-aa
-hl
..
X J
Z b N--'a E %8P3""" f f f s m y .????l?? . . . . . . . . .
si,Z~pzsaf
--- ~ ~ ~ ~ ~ ~ ~ ~ ~
?l .P. .?P?
CIrnbrnW 0 0 0 0 0
C C C C C
P-
b(?
QPa
a - a a a acv a a a a a a (?-a-a ~ m & & - N- a -mm a - - a ~ a~b a ~
5 L:
8
a - @ a a a $ a a a a a a a m- a -b v ~ a a a - a f i ; a a E a m@
nn 4 4 4 4 -
U lA I
i:
E:
CP
CP
d d
o B
:-zs';N,?.?-CM ;: M
! ; 2s 99 : :-2w$?2:(?92:: c a a- a d weoam
h l
(?
a n n n
q$iV ;f2J sc,;c calnln---meaoG-ac
2qg
v l v l v l r n r n
B B B B
';a';aaaa
E E E E E
h U
~ u M P u ~ M
M n OP
gmr ( o b s ) gmr (mms) maeZm a1 alpha1 a2 mu2 alpha2 1 ambda2 a3 mu3 a 1 pha3 1 ambda3
C
mean age %( 8-14) %(IS-64) Z(65+ ) del t a l c d e l La12 de 1 t a 3 2 be t a l 2 s i gma2 s i gma3 x low x high x ret. x shift a b
1 2 3 4 5
u. k. u. k. u. k. u. k. u. k.
males males males males males
2 to 1 2 to 3 2 to 4 2 to 5 2 to 6
6 u. k. 7 u. k. 8 u. k. 9 u.k. 10 u. k.
2 to 7 2 to 8 2 to 9 2to10 2 to the rest
APPENDIX C 2
(continued).
gmr (obs) gmr (mms) maeZm a1 alpha1 a2 mu2 alpha2 1 ambda2 a3 mu3 a 1 pha3 1 ambda3
C
mean age
Z( 8-14)
Z( 15-64)
Z(65+
del talc delta12 de 1 t a 3 2 be t a l 2 si gma2 s i gma3 x low x high x ret. x shift a b
)
1 u.k. 2 u. k. 3 u. k. 4 u. k. 5 u. k.
3to1 3 to 2 3 to 4 3 to 5 3 to 6
u. k. u. k. u. k . u. k. 10 u. k.
6 7 8 9
3 to 7 3 to 8 3 to 9 3 to 10 3 to the rest
gmr ( o b s ) gmr (mms) maeZm a1 alpha1 a2 mu2 alpha2 1 ambda2 a3 mu3 alpha3 1 ambda3
C
mean a g e Z( 8-14) 2 ( 15-64) Z(65+ ) del talc del t a l 2 de 1 t a 3 2 be t a l 2 s i gma2 s i gma3 x low x high x ret. x shift a b
1 u. k. males 2 u. k. males 3u.k.males 4 u. k. males 5 u. k. males
4 to 1 4 to 2 4to3 4 to 5 4 to 6
6 7 8 9 10
u.k. u. k. u. k. u. k. u. k.
4to7 4 to 8 4 to 9 4 to 10 4 to the rest
g 01 p 0,
s
s
z 01 s
1 01
E 01
s s
sapm sapm sapm sapm sapm
.q .n s .q .n p .9 'n E .q .n z 'n 1
SSE10'0 EEI I L ' 9 Z SZ086.01
*Z*LS.
IP9P8'0 16CS0.0 LLS 1 0 ' 0
1 ~ ~ 8 ' 0IZLP8.0 1Z l P I ' 0 91ZCE'0 88008.0 S910P.0 L S S K '0 0L9SZ.0 c s a z 'L CSWL ' z LZ911 ' Z I 95681 ' 0 2 SZLSS'P9 t76681'Z9 8P966' ZZ 0S0Z9'Ll LSL8Z ' 6E: B I L E I 'ZC ZZZ00' 0 00000'0 00000'0 00000'0 80008.0 88891 '0 S1681'0 Z0886' CZ SEW0' 0 IL9Z0.0 Z0910'0 ZE00L.6 88888'1 68BS0' 0 6
6LLL 1 ' C 9 LE08C ' IZ PI081 '11 85901 ' I ZL0Z0' C 96L09.0 9L09L.0 L10ZC'0 PLPCI' z 808L9'LI BIBSS'P9 Z9ILL'Ll ZEL69' LC
=is. 6
1
jsqoj ~ m 8
* Y1
r - O o r n d " ,
* C C C
0 "
*
0 0 0 0 0
w w w w w
azzaa
B B B B B
E E E E E
- ~ m b v )
0 0 0 0 0
C C C C C
w w w w w
VI VI P 6
-~m
N a u
2.58
- - m ~ ~ m -c c a a m - m a LW-..E: \ c ce r e Camm-.-.-m66o.-cA L L e Q N Q E mQa m -Q----.wW-A L V I c---. c c c a,-.a 8 m--N r- a m r o E N N N U U U ~ 01 V I M M M M a e Q E aW W E m a Q E Q1
mm
t9 a u
-ZPP 0
Y l Y l Y l Y l V l
aaaaa
O N N
a. -
"
"
"
6 E E 6 6
-a
I
gmr ( o b s ) gmr (muis) mae%m a1 alpha1 a2 mu2 alpha2 1 ambda2 a3 m3 u a 1 pha3 1 ambda3
C
mean age %( 0-14) % ( 15-64) %(65+ del t a l c del t a l 2 del ta32 beta12 s i gma2 s i gma3 x low x high x ret. x shift a b
1 u. k. 2 u. k. 3 u. k. 4 u. k. 5 u. k.
7 7 7 7 7
6 7 8 9 10
u. k. u. k. u. k. u. k. u. k.
7 to 7 to 7 to 7 to 7 to
6 8 9 10 the rest
gmr ( o b s ) arnr (mms) maeZm a1 alpha1 a2 mu2 alpha2 1ambda2 a3 mu3 alpha3 1 arnbda3
C
mean age Z ( 0-14) 2 ( 15-64) 2(65+ ) del t a l c del t a l 2 de 1 t a32 beta12 s i gma2 s i gma3 x low x high x ret. x shift a b
u. k. u. k. u. k. u. k. 5 u. k.
1 2 3 4
8 to 1 8 to 2 8 to 3 8 to 4 8 to 5
u. k. males 10 u. k. males
6 7 8 9
u. k. males u. k. males u. k. males
gmr ( o b s ) gmr ( m m s ) maeZm
I 0.04453 1.88888 14.86221
mean age Z( 6-14) X ( 15-64) Z(65+ ) del t a l c del t a l 2 de 1 t a 3 2 beta12 s i gma2 s i #ma3 x low x high x ret. x shift a b
1 2 3 4 5
u. k. u. k. u. k. u. k. u.k.
9 to 1 9 to 2 9 to 3 9 to 4 9to5
6 u. k. males 7 u. k. males 8 u. k. males 9 u. k. males 10 u. k. males
gmr ( o b s ) gmr (mms) maeZm a1 alpha1 a2 mu2 a 1 pha2 1 ambda2 a3 mu3 alpha3 1 ambda3
C
mean a g e Z( 0-14) % ( 15-64) %(65+ ) del talc del t a l 2 de 1 t a 3 2 beta12 s i gma2 s i gma3 x low x high x ret. x shift a b
1 2 3 4
u. k. u. k. u. k. u. k. 5 u. k.
10 to 1 10 to 2 10 to 3 10 to 4 10 to 5
u. k. u. k. u. k. u.k. 10 u. k.
6 7 8 9
10 to 6 10 to 7 10 to 8 lot09 10 to the rest
100
P %-@$PbO
2
bN
%fa?~5ag~~ifgfbm-.RGg~g~~%18~&
N mN ,#,x,mSSw
ao, i m
NbP
aab
............
P????""?
. .I??
.?7?
a-2aaavaaaa&aawambvaa-aa-~aN m-lo-N
.............................
-IGa
P
lo
8 I a8- ~ 8 8 m ~ 3t fg8l $ w ~ 5g3 FRE%g f ~ s 8 ~ g ~ f ~ ~ f8 8 l ~

mN-o,?
(D
FPa P
N-
mo,
a - G a a a ~ a a a a & & a m wb - m - a a a a a - m a m a N m-
. . . . .?? . . .
IGa
(C,
~. .f. .~. .~. .~. .w. .n. .;. .$. .i . ~. . ~. . i. .~. .@ # ~ l ~ ~ ~ R $ I .

m-mRGP
N0 o
w--vwm-
wmw
-aN-P Nm mbb a - a a - z a a a z a a a a a a a ~ e m o , m a a - m-a v - a ~ g & m-loN
a-2aaa-aaaaaaaNo,bmmaaaaaP-abma N m-lo-N F.l
Model migrotion schedules
ff& 4 ..............
Ef&"i@ff
mv,
n3 RN wv f$$gfi~~f8$f$~8 .?P .P . .???

rnm
* 2
() I
c - m a - 5
0 0 0 0 0 * * * * *
ClClClClN
-a 8
P\'
CO CO
h l
. a
C C a C a * * a O N O 8 e e a--N r- am?=! **am a a 8 a- m a m -
I I m m a - m m? w e . E a Y , Y , . a r d 0 8 8 0.- U C a -w----**-c L. VI a-wa a a a,-.a N N P \ ' l J u u P VI L4 * * * *
gmr (obs) gmr (mms) me m aZ
I 0.07934 1.00000 9.02247
mean age Z(0-14) %(IS-64) %(65+ ) del t a l c del t a l 2 de 1 t a32 beta12 s i gma2 s i gma3 r low r high x ret. r shift a b
34.43119 18.85730 66.16161 14.98109 2.37994 0 . 16936 0.00256 0.24242 2.67630 0.21 157 15.70024 21.71038 62.21795 6.01014 25.44373 0.02946
1 2 3 4
u. k. females u. k. females u. k. females u. k. females 5 u. k. females
3 to 1 3 to 2 3 to 4 3 to 5 3 to 6
6 u. k. females 7 u. k. females 8 u. k. females

9 u. k. females 10 u. k. females
3 to 7 3 to 8 3 to 9 3 to 10 3 to the
gmr ( o b s ) gmr (toms) mae Z m a1
alpha3 1 ambda3
C
mean a g e Z( 8 - 1 4 ) Z( 15-64) Z(65+ ) del t a l c d e l La12 d e l La32 be La12 s i gma2 s i gma3 x low x high x ret. x shift a b
I 2 3 4 5
u. k. u. k . u. k. u. k. u. k.
females females females females females
4 to 1 4 to 2 4 to 3 4 to 5 4 to 6
6 u. k . 7 u. k . 8 u. k. 9 u. k. 10 u. k .
gmr ( o b s ) gmr (mms) maezm a1 alpha1 a2 mu2 a 1 pha2 1 ambda2 a3 mu3 a 1 vha3 1 ambda3
C
mean a g e Z( 8-14) Z ( 15-64) Z(65+ ) del t a l c del t a l 2 del ta32 beta12 s i gma2 s i gma3 x low x high x ret. x shift a b
1 u. k. females 2 u. k. females 3 u. k. females 4 u. k . females 5 u. k. females
5 to 1 5 to 2 5 to 3 5 to 4 5 to 6
u. k. u. k . u. k. u. k. u. k.
5 to 7 5 to 8 5 to 9 5 to 10 5 to the
gmr ( o b s ) gmr (mms) mae%m
mean a g e 0-14) X( 15-64) X(65+ ) del t a l c del t a l 2 de 1 t a 3 2 beta12 s i gma2 s i gma3 x low I high I ret. x shift a
%C
2 3 4 5
1 u. k. females u. k. females u. k . females u. k. females u. k. females
6 to 1 6 to 2 6 to 3 6 to 4 6 to 5
6 u. k. females
7 u. k. females
8 u. k. females 9 u. k. females 10 u,k. females
APPENDIX C 2
(continued).
g r (obs) m n r (mms) m
1 0.04829 1.00000
mean age % ( 0-14) % ( 15-64) X(65+ ) d e l talc d e l tal2 de 1 t a32 beta12 s i gma2 s i gma3 x low
x ret. x shift
0.0631 1 32.20114 20.49481 68.78252 10.72268 6.99362 0.30191 0.0005 1 0.76646 1 .77239 0.16780 14.15021 58.04865 9.63022 28.5661 1
1 u. k. 2 u. k. 3 u. k. 4 u. k. 5 u. k.
7 to 1 7 to 2 7 to 3 7 to 4 7 to 5
6 u. k. 7 u. k. 8 u. k. 9 u. k. 10 u. k.
7 7 7 7 7
to 6 to 8 to 9 to 10 to the rest
gmr (obs) gmr (mms) me m a2 a1 alpha1 a2 m2 u a l pha2 l ambda2 a3 mu3 a 1 pha3 l ambda3
0
mean age Z ( 8-14) 2 ( 15-64) 2(65+ ) del talc del t a l 2 del ta32 be tal2 s i gma2 s i gma3 x low x high x ret. x shift a
b
12.57829 31.42656 0.02362
18.11823 26.27616 0.82910
9.52022 29.51372 8.02188
13.55031 29.17042 0.81438
9.19021 31.0'3333 8.02262
7.83018 23.79711 0.01875
8.44019 28.17887 8.02720
13.54031 33.72836 8.82460
8.16019 25.93718 0.82388
8.846328 28.44961 0.82429
1 2 3 4 5
u. k. females u. k. females u. k. females u. k. females u. k. females
8 to 1 8 to 2 8 to 3 8 to 4 8 to 5
6 u. k. females
7 u. k. females 8 u. k. females 9 u. k. females 10 u. k. females
8 to 6 8 to 7 8 to 9 8 to 10 8 to the I
gmr (obs) gmr (mms) mae4rn
I 0.03536 1.88800 35.50578
mean age 4 ( 0-14) 4 ( 15-64) 4(65* ) deltalc del t a l 2 de 1 t a32 beta12 s i gma2 s i gma3 x low x high x ret. x shift a b
5.56013 32.45698 0.03227
1 2 3 4 5
u. k. u. k. u. k. u. k. u. k.
9 to 1 9 to 2 9 to 3 9 to 4 9 to 5
6 7 8 9 10
u. k. u. k. u. k. u. k. u. k.
9 to 6 9 to 7 9 to 8 9 to 10 9 to the
20920 '0 ZLCCfl 'LZ CZ0S8 ' 6 00000'0 8b8ZP'ZZ LI0LS'ZI 00000'0 LS0PI ' Z 0z00Z ' I 00000'0 PIPLP.0 86S8S'SI U S 2 9 '9 61950'89 808CC ' SZ L0SC6'9Z 81288'8 00000'8 00000-0 00000'8 00000'0 60CPZ ' 0 95C11'0 0ZPZS '61 181L0'0 0C%1 ' 0 S0PE0' 0 ISCP8.8 00000'1 9L6S0 ' 8 S
PICZ0'0 PL8ZE ' IC 0C086 ' 2 1 00000'0 LP0ZP' SZ LIBPP'ZI 00000'0 1989S.0 P0L9S ' 0 00000'0 96mC'0 8P9ZZ ' P 86WP'ZI S8E8& ' 89 81 112'61 S086L' ZC ICrn.0 00000'0 00000'0 00000'0 00000'0 SZLI 1 ' 0 1Z982'0 P168s'0& 696S0' 0 C6911 '0 0Z810'0 LSSW'0I 00000' 1 81 981 ' 0 C
61SZ0'0 9LC10.62 PZe)GC'01 00000' 0 8P098 ' SZ PZ0LP ' S 1 00000'0 Z60C8 ' I 08S6L' 0 00000'0 IPPIP.0 ZL098' L I 8ZC8C ' S 6L6PZ ' 8 9 C6992 '9Z 0C86C 'LZ LL100.0 00000'0 00000'0 00000' 0 08888'0 891ZZ'0 L01ZI ' 0 C0ZIP'CZ 6LZL0' 0 SC960' 0 L10C0'0 CW91 ' 6 88888' 1 90IL0.0
I
110
A. Rogen, L.J. Casrro
APPENDIX C.3 Japan (1 970).*
FIGURE C.3
Map of the regional aggregation of Japan used for this study.
*This regional aggregation of Japan varies slightly from the one used in the forthcoming IlASA case study report.
Males.
gmr ( o b s gmr ( m m s mae%m a1 alpha1 a2 mu2 a 1p h a 2 1ambda2 a3 mu3 alpha3 1arnbda3

C
mean a g e Z ( 0-14) % ( 15-64) %(65+ ) del t a l c del t a l 2 de 1 t a 3 2 be t a l 2 s i gma2 s i grna3 x low x high x ret. x shift a b
1 2 3 4
japan japan japan japan
males males males males
1 to 2 1 to 3 1 to 4 1 to 5
5 6 7 8
A. Rogers, L.J. Chstro
---E E E E
Vl
a n a n m m m m
Vl
Vl
Vl
E E E E
m m m m ._ .- .- .-
Z$$k
wc-am
-CIm*v,
C
0 C C C C 0 0 0 0
CICICICICI
gmr ( o b s ) gmr ( m m s ) mae X m
mean a g e Z ( 0-14) X ( 15-64) del t a l c del t a l 2 de 1 t a 3 2 beta12 s i gma2 s i gma3 x low x high x ret. x shift a b
1 2 3 4 5
japan japan japan japan japan
3 to 1 3 to 2 3 to 3 3 to 4 3 to 5
6 japan males
7 japan males 8 japan males 9 japan males
APPENDIX C.3
(continued).
gmr ( o b s ) gmr (mms) maeZm a1 alpha1 a2 mu2 a 1p h a 2 1ambda2 a3 mu3 a 1p h a 3 1ambda3

C
mean a g e Z ( 0-14) 2 ( 15-64) %(65+ ) del talc del t a l 2 del ta32 be t a l 2 s i gma2 s i gma3 x low x high x ret. x shift a b
1 japan 2 japan 3 japan 4 japan 5 japan
4 to 1 4 to 2 4 to 3 4 to 4 4 to 5
6 japan males 7 japan males 8 japan males 9 japan males
gmr ( o b s ) gmr (mms) mae %m a1
mean a g e % ( 8-14) % ( 15-64) %(65+ ) del t a l c del tal2 del ta32 beta12 s i gma2 s i gma3 x low x high x ret. x shift a b
1 2 3 4 5
5 to 1 5 to 2 5 to 3 5 to 4 5 to 5
6 7 8 9
APPENDIX C.3 (continued).
gmr ( o b s ) gmr (mms) mae%m a1 alpha1 a2 mu2 alpha2 1 ambda2 a3 mu3 alpha3 l ambda3
C
mean a g e %( 0-14) % ( 15-64) %(65+ ) del t a l c del ta12 del ta32 beta12 s i gma2 s i gma3 x low x high x ret. x shift a b
2 3 4 5
1 japan japan japan japan japan
6 to 1 6 to 2 6 to 3 6 to 4 6 to 5
6 7 8 9
6 to 6 to 6 to 6 to
6 7 8
the rest
lsai aql 07 L 80 L 3 L 07 L 9 01 L
sapm Slm ae Slm ae Slm ae
uede! 6 ud[ 8 ee ud[ L ee uede! 9
ES920 ' 0 CZSLE ' LE 1 Z0L0.6 00000'0 LE0SZ' I Z 91081 ' 1 2 68888'0 90619's S08PC ' I 00000'0 0001&'0 9&&0S' 8 SC888' L E0PEP ' 9L Z9LL9 ' 1 S 9PSEE'I & 06100'0 00000' 0 00000'0 0 1 0 '0 C00 00000' 0 L81LE.0 81990'0 6SPS8' 9 1 661S0'0 12680'0 21910'0 S9018'01 00000'1 SE86C ' 0
lEZP0'0 91012'86 P10Z1 ' 9 00000'0 0&01&'81 91061 ' 1 2 00000'0 6P6PP' 8 02096' 1 00000'0 IPELZ.0 SP089' S I 69616'P 86201 ' 1 8 CELL6 ' C I 66ZL6' LZ 8 100'0 1 00000' 0 00000'0 00w0-0 00000'0 16ZS9.0 LZLL0'0 P9E01 ' I S 0tL90' 0 LPISI '0 EP810'0 LLSI9'01 00000'1 0S96Z - 0 9
A. Rogers, I,.J. Cnsfro
v J v J V ) V )
3;s;
4 3 4 3 4 3 P )
E E E E
m m m m
C C C C
444% .- .- .- .-
V ) V J V ) V ) v J
a;; ;;
4 3 0 4 3 P ) P )
E E E E E
C C C C C
m m m m m
m m m m m .- .- .- .- .-
a a a a a
Females.
gmr ( o b s ) gmr ( m m s ) maeZm
mean age X( 8-14) X ( 15-64) X(65+ ) del t a l c del t a l 2 de 1 t a 3 2

s i gma2 s i gma3 x low x high x ret. x shift a b
1 2 3 4
females females females females
1 to 2 1 to 3 1 to 4 1 to 5
5 6 7 8
gmr ( o b s ) gmr ( m m s ) maeXm a1 alpha1 a2 mu2 alpha2 1ambda2 a3 mu3 a 1p h a 3 1ambda3

C
mean a g e X( 0-14) X( 15-64) %(65+ ) del t a l c del t a l 2 del ta32 be t a l 2 s i gma2 s i gma3 x low x high x ret. x shift a b
1 japan 2 japan 3 japan 4 japan 5 japan
2 to 1 2 to 2 2 to 3 2 to 4 2 to 5
6 7 8 9
IsaI agI 03 E 8 01 E L 01 E g 01 E
sapmaj sapmaj sapmaj sapmaj
uedej uedej uede! uedej
6
8 L
s 03 E P 03 E
z 01 E
1 01 E
E 01 E
sapmaj sapmaj sapmaj sapmaj sapmaj
uedej uedeF uede! E uedej uede! 1
8EIE0'0 I LEPS ' 6Z LIWE'L 00000'0 PP060 ' PZ LZ0SL ' 91 00000'0 EL99L ' Z 66001 ' 1 00000'0 08&Sb ' 0 109Z0'S P88SI ' P I 88PW ' S9 LZPPP ' 02 860% ' EE LPS00 ' 0 00000'0 00000 ' 0 00000'0 00000 ' 0 818W.0 ESLPI ' 0 60629' I Z 99LL0 ' 0 EPZ9I' 0 8PLZ0 ' 0 LEP9P ' E I 00000' 1 SE910'0 L
PPEZ0 ' 0 6LE06 ' LZ IZWE.6 00000'0 0S016'92 6Z0LS ' L I 00000'0 !308PE' I 69969 ' 0 00000'0 106437.0 E680Z ' 9 E6182'EI E19t70' 19 P61L9'SZ I L8ES ' Z S 90s00'0 00000'0 00000'0 00000'0 00000'0 L0LEZ ' 0 98SL 1 ' 0 SE%L ' SZ Z89L0 ' 0 ZSZZ I' 0 ZPIE0'0 t96SE '0 1 00000'1 900t70.0 9
E9EZ0 ' 0 PLEE I' 0E 8Z0SP'ZI 00000'0 S W L ' PZ LI0SZ'ZI 00000'0 L168&'0 U0L9'0 00000'0 29829 ' 0 9Z00C ' C Z6PSL ' 9 1 LS81E'S9 IS9Z6'LI 99019'SC 95900'0 88880.0 00000'0 00000'0 00000'0 EZI 11 ' 0 18S8Z.0 I82IZ'EC EPPE0' 0 0L161'0 B120'0 ZL8tZ.L 00000'1 W 6 S'I C
00620'0 ELE6 1 '82 0Z09L '8 00000'0 Ct4308'EZ EZW0 ' S I 00000'0 0S018' 1 8LE98 ' 0 88888.0 8 Z S K '0 866ZC'S IZtZ8'ZI 06x0'99 06PZI ' IZ 8XE% 'ZC P m - 0 00000'0 88880'0 88868'0 00000'0 SZS8z ' 0 LSLSI' 0 WZ6L' I Z PPP88'0 1 1 x 1'0 8LSZ0' 0 Z686L ' S 00000'1 6LL68.0 Z
ZLZZ0' 0 IP018'8Z 610SE.8 00000 ' 0 SP0EL ' PZ 9Z08&'91 00000 ' 0 6L89E ' Z LPIEL.0 00000 ' 0 8ZLIE.0 0661P.E 26901 ' S I L t 6 8 6 ' E9 19E06'0Z 9060L ' PE 06S00 ' 0 00000 ' 0 00880'0 00000 0 00000'0 SSEZE ' 0 6S9E 1 ' 0 8LL8 I' ZZ 9SE90 ' 0 16660 ' 0 L10Z0'0 88P0S ' CI 00000' I 80SE0' 0 1
APPENDIX C 3
(continued).
gmr ( o b s ) gmr (mms) maeZm
mu2 alpha2 1 ambda2 a3 mu3 alpha3 1 arnbda3

C
mean age Z( 8 - 1 4 ) % ( 15-64) %(65+ ) del t a l c del t a l 2 del ta32 beta12 s i gma2 s i gma3 x low x high x ret. x shift a b
1 japan females
2 3 4 5 japan japan japan japan females females females females
4 to 1 4 to 2 4 to 3 4 to 4
4 to 5
6 7 8 9
gmr ( o b s ) gmr ( m m s ) mae%m a1 alpha1 a2
mean age % ( 0-14) % ( 15-64) %(65+ ) del t a l c del t a l 2 del ta32 beta12 s i gma2 s i gma3 r low r high r ret. r shift a b
1 2 3 4 5
5 to 1 5 to 2 5 to 3 5 to 4 5 to 5
6 7 8 9
g m r (obs) gmr (mms) maeXm a1 alpha1 a2 mu2 a 1pha2 1 amb d a 2 a3 mu3 a 1p h a 3 1 ambda3
C
mean age
X( 0-14)
X ( 15-64)
X(65+
del talc del t a 1 2 del t a 3 2 beta12 sigma2 s i gma3 x low x high x ret. x shift a b
1 2 3 4 5
6 6 6 6 6
6 7 8 9
gmr (obs) g r (mms) m me m aZ a1 alpha1 a2 m2 u alpha2 1 ambda2 a3 m3 u alpha3 1 ambda3 o mean age Z(0-14) Z(15-64) Z(65+ ) d e l talc d e l ta12 de 1 t a32 beta12 s i gma2 s i gma3 x low
x high x ret. x shift
a
b
1 0.00913 1.00000 27.60073 0.03926 0.11792 0.00040 55.48544 0.37309 0.06338 0.00000 0.00000 0.00000 0.00000 0.00141 25.08721 27.68680 68.26564 4.04755 27.94174 98.07555 0.00000 0 . 3 1607 0.16988 0.00000 13.75020 27.3205 1 0.00000 13.57031 29.22816 0.02717
2 3 4 5
1 japan japan japan japan japan
7 to 1 7 to 2 7 to 3 7 to 4 7 to 5
6 7 8 9
gmr (obs) gmr (mms) maeZm a1 alpha1 a2 mu2 alpha2 1 amb d a 2 a3 mu3 alpha3 1 ambda3
C
mean age X( 0-14) X( 15-64) X(65+ 1 del talc del tal2 del ta32 beta12 s i gma2 s i gma3 x low x high x ret. x shift a b
1 2 3 4 5
8 to 1 8 to 2 8 to 3 8 to 4 8 to 5
6 7 8 9
APPENDIX C.4 Netherlands (1 974).*
12. ljsselmeerpold and Dronten
FIGURE C.4
Map of the regional aggregation of the Netherlands used for this study.
*All schedules are outmigration flows from each region to the rest of the country.
Males: outmigration from each region.
gmr ( o b s ) gmr ( m m s ) maeZm a1 alpha1 a2 mu2 alpha2 1 amb da2 a3 mu3 alpha3 1 amb da3
C
mean a g e % ( 8-14) % ( 15-64) %(65+ ) del t a l c del t a l 2 de 1 t a 3 2 beta12 s i gma2 s i gma3 x low x high x ret. x shift a b
1 2 3 4
netherlands males netherlands males netherlands males netherlands males region = 1 region = 2 region = 3 region = 4
5 6 7 8
netherlands netherlands netherlands netherlands
region = 5 region = 6 region = 7 region = 8
m-aaaaaaa&aaaaammmmaaa~ammaama m m-am - m -m
.............................
-
II
I 1
II
I 1
.g .g .g .? M M M M
2 E b b
B I I I a 7 7 7
E E E E
d3-s-s E K E C
BBB2
EE55 a c a c
O O Q B
E E E C
A. Rogers, L.J. Casrro
.o 0 .4! .4! 6b.G M M 2 2 2 2
C C C E
aaaa
0 5 0 0 r r r 0
b , C I
V l V l V l V l
9 9 5 9
h O,
S S S
E E E
'6b'6b '6b '6b
E C C C 0 0 0 0
2 2 2 2
a" a" a" a"

0 0 0 0
gg2EJ
C C C C
lgss
gmr ( o b s ) g m r (mms) mae%m a1 alpha1 mu 2 a 1p h a 2 1 amb d a 2
a2
a3
mu3 alpha3 1amb d a 3
C
mean a g e %( 0 - 1 4 ) %( 15-64) %(65+ ) d e l tal2 d e l ta32 b e tal2 si gma2 s i gma3 x low x high x ret.
a b
x shift
9 10 11 12
netherlands netherlands netherlands netherlands
region region region region
=9 = 10 = 11 = 12
132
A. Rogers. L.J. Castro
APPENDIX C .5
Soviet Union (1 974).*
7. Baltic Republics,
urban areas
urban areas
'"CJF'4. Central Asian Republics, urban areas
FIGURE C.5
Map of the regional aggregation of the Soviet Union used for this study.
*Total (male plus female) flows only. Regions 1-7 refer to the urban areas of the region; region 8 includes aN rural areas of the Soviet Union.
gmr ( o b s ) gmr (mms) maeXm al alpha1 a2 mu2 alpha2 1 ambda2 a3 mu3 a l pha3 1 amb da3
C
mean age X ( 0-14) X ( 15-64) %(65+ ) del t a l c del t a l 2 de 1 t a32 beta12 s i gma2 s i gma3 x low x high x ret. x shift a b
1 2 3 4 5
ussr ussr ussr ussr ussr
migration flow migration flow migration flow migration flow migration flow
1 to 1 1 to 2 1 to 3 1 to 4 1 to 5
6 7 8 9
ussr ussr ussr ussr
migration flow migration flow migration flow migration flow
APPENDIX C.5
(continued).
gmr ( o b s ) gmr (mms) mae%m a1 alpha1 a2 mu2 alpha2 1 ambda2 a3 mu3 alpha3 1 ambda3
C
mean age % ( 0-14) % ( 15-64) %(65+ ) del t a l c del t a l 2 del ta32 beta12 s i gma2 s i gma3 x low x high x ret. x shift a b
1 2 3 4 5
migration migration migration migration migration
flow flow flow flow flow
2 to 1 2 to 2 2 to 3 2 to 4 2 to 5
6 7 8 9
ussr ussr ussr ussr
migration migration migration migration
flow flow flow flow
gmr ( o b s ) gmr (rnms) maeZm a1
a3 m3 u alpha3 1 ambda3
C
mean a g e Z ( 0-14) 2 ( 15-64) Z(65+ ) del t a l c del ta12 del ta32 beta12 s i gma2 s i gma3 x low x high x ret. x shift a b
1 2 3 4
ussr ussr ussr ussr 5 ussr
3 to 1 3 to 2 3 to 3 3 to 4 3 to 5
6 7 8 9
ussr ussr ussr ussr
gmr ( o b s ) gmr ( m m s ) maeXm
mean a g e X( 0-14) 7 ( 15-64) X(65+ ) del t a l c del t a l 2 del ta32 beta12 s i gma2 s i gma3 x low x high x ret. x shift a b
0.00000 9.03021 59.77655 0.04335
ii. iG3j
1 2 3 4 5
4 to 1 4 to 2 4 to 3 4 to 4 4 to 5
6 7 8 9
ussr ussr ussr ussr
gmr ( o b s ) gmr (mms) mae%m
mean a g e X ( 8-14) % ( 15-64) del t a l c del t a l 2 del ta32 beta12 s i gma2 s i gma3 x low x high x ret. x shift a b
2 3 4 5
1 ussr ussr ussr ussr ussr
5 to 1 5 to 2 5 to 3 5 to 4 5 to 5
6 7 8 9
ussr ussr ussr ussr
APPENDIX C .5
(continued).
gmr ( o b s ) gmr (mms) maeZm a1 alpha1 a2 mu2 alpha2 1 ambda2 a3 mu3 a 1 pha3 1 ambda3
C
mean age Z( 8-14) Z ( 15-64) Z(65+ ) del t a l c del t a l 2 del ta32 beta12 s i gma2 s i gma3 x low x high x ret. x shift a b
1 ussr migration flow 2 ussr migration flow 3 ussr migration flow 4 ussr migration flow 5 ussr migration flow
6 to 1 6 to 2 6 to 3 6 to 4 6 to 5
6 7 8 9
ussr ussr ussr ussr
1sal aq1 01 L 8 01 L L 01 L g 01 L
m o uo!lerZy ~ MOD uo!le~@u MOD uoyex@u MOD uoyerZy
rssn lssn rssn rssn
6 8
L
s 01 L p 01 L E 01 L 2 01 L I 01 L
uo!1e1%!ur Issn s u o y e 1 8 y Issn p MOU uoyer8pu rssn E MOU uo!1e@m Issn z MOD UO!~BI%!LU lssn I
MOU
MOD
L8BS0.0 9SWZ'6& IZ011'6 00000'0 SE0L I ' 02 P1090'1 1 00000'0 I S6EP ' I Z80IP' I 00000'0 L9CZ0 ' 0 EM89 ' 0 6689C.CI CLCS6 ' LL ELL9 ' 8 ZZP81'1C LBS00.0 00000'0 00000'0 00000'0 00000'0 rnBE'0 18812'0 LZP96 ' 8 1 ZSLP I ' 0 IPL6Z.0 6K00'0 9SZ0I 'LI 00000'1 LILGL'0 8
6 E 7 '0 6t0 ZE080' I E IZME.6 00000'0 E090.02 E10EL'01 00000'0 Z9VS6.0 1 986L' 0 00000'0 0L8P0'0 S0ZSI ' I 6P6LI ' I S SILZG'ZL 9EC68'I I
q
8
00000'0 PE86Z ' I L6M.0 00000'0 0SEP0' 0 8Z16P'I SI6IZ'ZI PPLBS ' LL 8Z611'11 ZPELZ ' 0I 80V89'I E 9 ~ s ~ - a ISE00'0 LSP00 ' 0 00000'0 00000'0 00000'0 00000'0 00000'0 00000'0 00000'0 00000'0 11662'0 68062' 0 S68ZZ '0 S0Vzz'0 61LL9.81 E9SL6.81 CSSL 1 ' 0 8L9S1 '0 ZLPL I ' 0 0Z91Z'0 L1600.0 28900'0 6SP19'PI EIESL'PI 00000'1 00000'1 m1S0'0 6SIZ1'0 Z
P10ZI'll 00000' 0 ZPLZS ' I 68916'1 00000' 0 ZP6Z0' 0 S80E8'0 L981C'EI 60Z6S' LL PZ680 ' 6 SLL60' E E PBS00.0 00000'0 00000'0 00000'0 00000'0 8860E:0 ' 88Z0Z' 0 PL169'81 SZZPI '0 LIL9Z.0 81P00'0 16ZZ9.91 00888'1 S6LLP '0
IJ!VS x la^ x V! x 8V A01 x Eoms !s Zoms!s ZI8Iaq ZEBl I a P ZIBl IaP

31-1
(
lap
a88 uoam
+S9)% (P9-S1)ii (PI-0 )ii

3
EP B I EBVdl'J E""J C ZBP9 ' 1 "s ZVd 18

J '
znm Z' J
IBVdlB
uzaom
J ' I
(SUU) J U (sqo) Jus
A. Rogers, L.J. b s t r o
0 C C C 0 0 0
w w w m
.s .? .s .s
C C C C
Q I
.g.gJ .gJ .I
m m m
f i c c c c
3 P 3 3 3 m m m m m
0 0 0 0 0
% % % P P
a a a a a
e e 4 r n - t v l
APPENDIX C.6
United States (1970).*
-..
...........
1. North East
2. North Central
FIGURE C.6
Map of the regional aggregation of the United States used for this study.
*Total (male plus female) flows only.
APPENDIX C .6
(continued).
gmr ( o b s ) gmr (mrns) mae'lr a1 alphal a2 mu2 alpha2 1ambda2 a3 mu3 alpha3 1amb d a 3
C
gmr ( o b s ) gmr (mms) maeZm a1 alphal a2 mu2 alpha2 l ambda2 a3 mu3 alpha3 1ambda3
C
mean age %( 8-14) 7(15-64) Z(65+ ) del t a l c del ta12 del ta32 bets12 s i gma2 s i gma3 x low x high x ret. x shift a b
mean age % ( 8-14) %( 15-64) %(65* ) del t a l c del ta12 del ta32 beta12 s i gma2 s i gma3 x low x high x ret. x shift a b
1 2 3 4
u. s. u. s. u. s. u. s.
total total total total
1 2 3 4
u. s. u. s. u. s. u. s.
gmr (obs) gmr (mms) maelm a1 alphal a2 mu 2 alpha2 1 amb da2 a3 mu3 alpha3 1 amb da3
C
gmr (obs) gmr (mms) maelm a1 alphal a2 mu2 alpha2 1 ambda2 a3 mu3 alpha3 1 amb da3
0
mean age
Z 0-14) (
2 15-64) (
mean age
Z 0-14) (
% 15-64) (
%(65+ ) del talc del tal2 del ta32 beta12 s i gma2 s i gma3 x low x high x ret. x shift a b
Z(65+
del talc del tal2 del ta32 be tal2 s i gma2 s i gma3 x low x high x ret. x shift a b
1 2 3 4
u. s. u. s. u. s. u. s.
1 2 3 4
u. s. u. s. u. s. u. s.
4 to 4 to 4 to 4 to
1 2 3
the rest
144
APPENDIX C.7
Hungary (1 974).*
FIGURE C.7
Map of the regional aggregation of Hungary used for this study.
*Total (male
female) flows only.
Model migmtion schedules
a - a s a a ~ a a a a m-a-a ~ a a a ~ a a ~0 a ~ ~ a a a a aa ~
c c c .z '2 '2
0 0 0
..............................
-
. . . . . . . . . . . . . .a. a. . . . . . . . . . . . z . . a - ~ ~ a s ~ s a a a a am -a a - a m a a - a a ~v a a - a
.f.#;.
X X X X
zzzr:
M M M M c c c c 1 1 1 1
2 2 2 2
g m r (obs) gmr ( m m s ) mae%m a1 alpha1 a2
mu2
alpha2 1 amb da2 a3 mu3 alpha3 1 amb da3

C
mean age % ( 0-14) % ( 15-64) %(65+ ) del t a l c del ta12 del ta32 beta12 s i gma2 s i gma3 x low x high x ret. x shift a b
1 2 3 4
hungary hungary hungary hungary migration migration migration migration
2 to 1 2 to 2 2 to 3 2 to 4
5 hungary migration 6 hungary migration

7 hungary migration
2 to 5 2 to 6 2 t o the rest
mws 0 0 0
C C C
0 0 0
e e e .g .g .El * * *
.$ .B .B
E E E
---a N
P 8 0 8
-. -.
+
N d O d a a aa r C CP CP L.L.Q, Q N Q 6 m Q 8 6 8 d 4 + N 3- dm 3WWSd a d 6 a - m 6 a + o
ar
.P .9 .E:
C
c c c
C
m m m
X X X
232
%%% C C E
gmr ( o b s ) gmr ( m m s ) mae %m a1 alpha1 a2 mu2 alpha2 1 ambda2 a3 mu3 alpha3 1 amb d a 3
C
mean a g e % ( 0-14) % ( 15-64) %(65+ ) del t a l c del t a l 2 de 1 t a 3 2 be t a 1 2 s i gma2 s i gma3 x low x high x ret. x shift a b
1 2 3 4
hungary hungary hungary hungary migration migration migration migration
5 to 1 5 to 2 5 to 3 5 to 4
5 hungary migration 6 hungary migration

7 hungary migration
5 to 5 5 to 6 5 to the rest
A. Rogers, L.J. Costro
008-mdONWNa868P0-800-ddm88 686-0a-08N8Q,8mNaaPWP-6m8 N 6 8 P 0 0 0 W W 8 8 0 8 N N a - + 0 8 m a P 8 8 8 N N N 0 0 m 6 W 8 8 N 8 8 a a P ~ 0 0 6 8 P 6 8 a008-8-8N-N888880000Q66N800P8 m - W 8 8 8 0 0 8 8 6 8 8 8 8 N ~ W W 0 0 8 8 8 - 8
. . . . . .-. . . . . . . . . . . . . . . . .
m-a-
P 8 N W 0 d m W W 0 8 W 8 P m P 0 b b m P m 0 0 8 08mNW00Q0000-8688WWW008VQN88 08bbQ-4N-8888mbbPP00P--W8
W W 8 Q 8 - - P N N 8 8 8 8 8 6 a 0 6 c 0 ~ 8 0 0 0 8
a88-a--8888m888mmNmN8N08
8 d a 8 8 8 8 8 8 8 8 8 8 8 m 6 0 W 6 8 8 8 8 8 N m-a-
........................
W W W
.s .s .s
m m m
X X X
$&&
G G G
s s s
291, ~ 8 8
4-3
a .r ,
DO DO-
pvlvln
."
RELATED PUBLICATIONS OF THE MIGRATION AND SETTLEMENT TASK
THEORY AND MODELS Migration and Settlement: Selected Essays (Reprinted from a Special Issue of Environment and Planning A) Andrei Rogers (Editor) Migration and Settlement: Measurement and Analysis Andrei Rogers and Frans Willekens Spatial Population Analysis: Methods and Computer Rograms Frans Willekens and Andrei Rogers Migration Patterns and Population Redistribution (Reprinted from Regional Science and Urban Economics) Andrei Rogers Essays in Multistate Mathematical Demography (Reprinted from a Special Issue of Environment and Planning A) Andrei Rogers (Editor) Multidimensionality in Population Analysis (Reprinted from Sociological Methodology 1980) Nathan Keyfitz Advances in Multiregional Demography Andrei Rogers (Editor) NATIONAL CASE STUDIES Migration and Settlement: 1. United Kingdom Philip Rees Migration and Settlement: 2. Finland Kalevi Rikkinen Migration and Settlement: 3. Sweden Ake Andersson and Ingvar Holrnberg Migration and Settlement: 4. German Democratic Republic Gerhard Mohs Migration and Settlement: 5. Netherlands Paul Drewe Migration and Settlement: 6. Canada Marc G. Termote Migration and Settlement: 7. Hungary m a Bies and K h A n Tekse Migration and Settlement: 8. Soviet Union Svetlana Soboleva Migration and Settlement: 9. Federal Republic of Germany Reinhold Koch and Hans-Peter Gatzweiler
(continued overleafl
152
NATIONAL CASE STUDIES (continued)
Reloted publications of the Migration and Settlement Task
Migration and Settlement: 10.Austria Michael Sauberer Migration and Settlement: 11.Poland Kazimierz Dziewohski and Piotr Korcelli Migration and Settlement: 12.Bulgaria Dimiter Philipov Migration and Settlement: 13. France Jacques Ledent and Daniel Courgeau Migration and Settlement: 14.Czechoslovakia Karel Kihnl Migration and Settlement: 15.Japan Zenji Nanjo, Tatsuhiko Kawashirna, Toshio Kuroda Migration and Settlement: 16. United States William Frey and Larry Long Migration and Settlement: 17. Italy Agostino LaBella
RR-81-16 RR-81-20 RR-81-21

Forthcoming Forthcoming Forthcoming ,Forthcoming Forthcoming
THE AUTHORS
Andrei Rogers has been Chairman of the Human Settlements and Services Area at IIASA since October 1976. He came to IIASA from Northwestern University, Illinois, USA. His current research focuses on migration patterns and the evolution of human settlement systems in both developed and developing countries.
Luis J. Castro came to IIASA from the Universidad Nacional Autonoma de MBxico to work on the comparative study of migration and settlement in the 17 nations associated with the Institute and on a case study of urbanization and development in Mexico. His current research is focused on age patterns of migration and their underlying causespecific components.

Model Migration Schedules (1981)

Încărcat de

Informații document

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Model Migration Schedules (1981)

Încărcat de

Drepturi de autor:

Formate disponibile

MODEL MIGRATION SCHEDULES

RR-8 1-30 November 1981

INTERNATIONAL INSTITUTE FOR APPLIED SYSTEMS ANALYSIS Laxenburg, Austria

Research Report RR-8 1-30, November 198 1

MODEL MIGRATION SCHEDULES

AGE PAITERNS OF MIGRATION

A. Rogers, L.J. Cnstro

Migration Rates and Migration Schedules

Model migration schedules

Sweden, 1968-1 973

United States, 1966-1 971

Model migration schedules

0.45 0.40 0.35 0.30 0.25 0.20 0.15 0.10

Model Migration Schedules

A. Rogers, L.J. Castro

x , = low point xh = high peak x r = retirement peak

X = labor force shift A = parental shift B = jump

The model migration schedule.

Model migration schedules

Parameters and variablesa

%(O-14) %(IS-64) %(65+)

Model migration schedules

Swedish regions, 1974 observed data by single years of age.

7.69 29.57 0.023

7.07 27.42 0.027

7.37 29.92 0.042

5.89 27.01 0.059

7.37 30.15 0.053

5.05 26.94 0.077

7.26 31.61 0.040

5.08 28.30 0.063

A. Rogers, I..J. Castro

continued on facing page.

Model migration schedules

A. Rogers, L.J. Castro

labor force pre-labor post-labor component force force component component

Model migration schedules

A. Rogers, L.J. Castro

A COMPARATTVE ANALYSIS OF OBSERVED MODEL MIGRATION SCHEDULES

Data Reparation, Parameter Estimation, and Summary Statistics

0.025 0.106 0.085 20.23 0.135 0.367 0.003

Model migration schedules

spline interpolation: Sweden, 8 regions, females, 1974.'

data by five-year age groups (5 yr).

A. Rogers, L.J. Castro

Model migration schedules

0.028 0.115 0.068 20.61 0.105 0.396 0.002

0.026 0.108 0.080 19.52 0.133 0.374 0.002

Without retirement peak (59 schedules)

0.021 0.099 0.059 22.00 0.127 0.259 0.003

0.021 0.097 0.063 21.35 0.151 0.327 0.003

A. Rogers, L.J. Cnstro

Model migration schedules

Model migration schedules

Netherlands From Amsterdam To Amsterdam

Sweden From Stockholm To Stockholm

United Kingdom From London To London

A. Rogers, L.J. Cnstro

Model migration schedules

A. Rogers, L.J. Castro

Standard deviation1 mean

n %(O-14) %(IS-64) %(65+)

Standard deviation/ mean

%(O-14) %(IS-64) %(65+)