BRIEF HISTORY OF SOCSIM
The SOCSIM Demographic Computer Microsimulation Program has been developed and applied over nearly three decades, starting in 1970, by Eugene A. Hammel and Kenneth W. Wachter. It is maintained at the Department of Demography of the University of California, Berkeley. The National Institute on Aging, the National Science Foundation, and the National Institute of Child Health and Human Development have contributed support.
The early history of SOCSIM is described in the 1986 volume in honor of Peter Laslett, The World We Have Gained, edited by Smith and Bonfield (1986). SOCSIM formed the basis of the study of extended family households in pre-industrial England published by Wachter, Hammel, and Laslett (1978) as Statistical Studies of Historical Social Structure. In 1986, the first kinship predictions to take account of historically changing demographic rates were constructed using SOCSIM, appearing in "The Kin of the Aged in the Year 2000" in the volume entitled Aging edited by Kiesler, Morgan, and Oppenheimer (1986). Over the years, studies using SOCSIM have ranged from the origin of incest tabus to the adjustment of the Slavonian Census of 1698. Current use is focussed on projections of kin and step-kin for U.S. elderly in the coming century.
SOCSIM is a research tool rather than a packaged program, and a certain level of demographic expertise is required for its use. SOCSIM was first implemented in FORTRAN by David Hutchinson and later Herb Doughty. It was rewritten and extended in PASCAL by Chad McDaniel and Carl Mason. The current version in C is written by Marcia Feitel, and runs at Berkeley under UNIX on SUN SPARC workstations.
SOURCES OF FURTHER INFORMATION
The current working version of SOCSIM is written in the C computer language. Up-to-date SOCSIM Technical Documentation . is maintained on the website of the Department of Demography at U.C. Berkeley at http://www.demog.berkeley.edu/~marcia/c_doc.html. A simplified demonstration program is found at http://www.ceda.berkeley.edu/programs/socsim/index.html. Readers may also want to consult a selected SOCSIM bibliography .
Current working papers using SOCSIM, listed below, may be accessed in "Portable Data Format" (pdf) via netscape. These are prepublication drafts or in some cases updated reports on material previously published. These papers should not be cited, reproduced, or redistributed without the express permission of Professor Wachter. Please be strict about observing copyright provisions. Comments and suggestions (to wachter@demog.berkeley.edu) will be most welcome.
Kinship Resources for the Elderly: An Update
(text)
Testing the Validity of Kinship
Microsimulation: An Update (text)
(Figures available soon.)
THE MICROSIMULATION BACKGROUND
Reviews of demographic microsimulation may be found in the book Family Demography, edited by Bongaarts, Burch, and Wachter (1987) and in the article in Population Index by DeVos and Palloni (1989). An assessment of the broader field of social science microsimulation by a panel of the National Research Council is presented in two volumes edited by Citro and Hanushek(1991), especially pages 286 to 289 of Volume I. The SOCSIM program is a descendant of the POPSIM programs described by Rao, Lindsey, and coworkers ((1973) Its distant cousins include a pioneering program of Peter Kunstadter (1963), the REPSIM models described by Clague and Ridley (1973), and the AMBUSH model of Nancy Howell (1979). A close relation is CAMSIM developed by James E. Smith (1987). The first complete documentation of SOCSIM appeared in a manual by Hammel, Hutchinson, Wachter, Lundy, and Deuel (1976).
SOCSIM is a "closed" demographic simulation model, The terminology of the field distinguishes between "closed" and "open" models in terms of the structure of marriage. In a closed model, marriage partners are found from among individuals already present with life histories and kin within the simulated population. In an open model, marriage partners are created at marriage, with characteristics conditional on those of the marrying population member. Closed models are much more complicated, but they permit bilateral kinship reckoning to arbitrary depth, one of SOCSIM's most important features.
BRIEF TECHNICAL OVERVIEW
The following paragraphs give a brief technical overview of SOCSIM. Full details can be found in the SOCSIM Technical Documentation .
The internal data structure within SOCSIM consists of three flat arrays, the population roster P, the marriage roster M, and the socio-economic supplement to the population roster X. The P-array contains a row for each person who is alive at any point during the simulation. When a person is born during the simulation, a new row is added to the P-array. The row remains after the person's death, so that the person's kin and characteristics can be retrieved.
Columns of the P-array include the person's identification number, sex, the population group to which a person belongs, dates of birth and death, pointers to a person's eldest ever-born child, a person's next eldest siblings through the father and through the mother, and the person's most recent marriage in the M-array. Individual-specific multipliers for fertility are also included. The M-array contains a row for every marriage union established during the simulation, with a marriage identification number, date of marriage, date and cause of any marriage termination, pointers to the husband and wife of the marriage, and to the next previous marriage if any of husband and of wife, The X-array has the same person identifiers as the P-array and includes variables for specific applications such as wealth, income, and education which are specified by the user for each study.
Time is measured in integral months. The timespan of the simulation is divided into segments, generally of 5 or 10 years each, during which a given set of vital rates stored in input rate tables hold force. Transitions for members of the population include the initiation of a marriage search, childbirth, migration, and death. These transitions are called "events" and the type and waiting time for a person's next event are determined in a stochastic event competition whenever a previously-scheduled event is executed. An event competition for a person also takes place when the person is affected by death of a spouse, migration, or other events previously scheduled for other related individuals.
The model for the event competition is a standard model of competing risks. A candidate waiting time is generated for each event for which a person is eligible. The event with the shortest waiting time wins the competition and is scheduled. The losing waiting times are discarded. The distribution of each waiting time is piecewise exponential, meaning that the hazards are piecewise constant over intervals of months specified by the user.
Events, once scheduled, are listed in an event queue. In the simulation, control proceeds month by month. In each month events scheduled for that month are executed (in random order). Characteristics in the P-array and X-array are updated and pointers are reset in response to the events. For instance, when a child is born, a new row is added to the P-array, characteristics are generated by various random processes, the last-born-child pointers for the parents are reset, the next-elder sibling pointers for the next-younger siblings are reset, and new events are scheduled with event competitions for mother and for child. Lengthy code is involved, but the logic is mainly an expression of common sense.
The most elaborate part of the simulation is the marriage module. One of the competing events for people eligible to marry is the initiation of a marriage search. Those previously scheduled for marriage who have not yet been paired with a suitable partner are kept in a marriage queue. A person initiating a marriage search, searches in random order through the opposite-sex members of the marriage queue, checking each candidate against a set of criteria for acceptable age range and group membership. (The population is divided, if desired, into groups, which in various applications may be households, lineages, ethnic groups, social classes, etc.) The first candidate not precluded by incest taboos who meets all criteria is then selected as the marriage partner. If no one meets the criteria, the person joins the marriage queue and becomes eligible to be chosen by newly arriving searchers of the opposite sex. At the end of each segment, all members of the marriage queue are removed, have new events scheduled for them, and become eligible to initiate new searches. With populations of a thousand or more with historically realistic rates, searchers find partners quickly. In the jargon of economics, "the marriage market clears."
There are many special features of the model. Waiting times to next birth for women are adjusted conditional on the timing and survival of the preceding child. Fertility is parity-specific (it depends on the number of preceding births), and is also specific to woman's age (but not partner's age) and marital status. Fertility is heterogeneous; each woman has her own fertility multiplier raising or lowering her probabilities of childbirth at all ages, and a component of this multiplier can be inherited from her mother, introducing mother-daughter correlations in fertility. Divorce depends on duration of marriage. Migration is implemented as a change of group membership. Some groups are defined as groups "abroad", others as groups "at home". There are endogenous feedback mechanisms for adjusting fertility and migration to targets. Mortality rates can be specified in terms of age-specific probabilities of dying, separately by marital status, or in terms of Lee-Carter model forecasts with specified parameters. Group membership can be used to introduce observed heterogeneity in rates or to define household membership or other social structure.
Each run of the SOCSIM program produces a population list (.opop file) in the same format as the initial population, along with a marriage list (.omar file) and covariate value list (.opox file). From these files the demographic history and kinship connections of the population can be reconstructed. At Berkeley, these reconstructions are generally implemented using computer programs in the Splus Statistical Language, copies of which are available to interested users. (Splus is a trademark of Statistical Sciences, Inc., of Seattle, Washington.) The SOCSIM program itself generates for each run a set of statistical tabulations (.stat file) and age pyramids for each group and simulation segment (.pyr file). Typical runs of SOCSIM involve total populations of living and dead up to around 100,000. Usual practice is to run 50 or 100 replications of a given simulation, for up to 10 million simulated people in an experiment.
REFERENCES
Please also consult the selected SOCSIM bibliography .
John Bongaarts, T. Burch, and K.W. Wachter, editors( 1987) . Family Demography: Methods and their Applications. The Clarendon Press, Oxford.
Constance Citro and E. Hanushek, Editors (1991) Improving Information for Social Policy Decisions: The Uses of Microsimulation Modeling, two volumes, National Academy Press, Washington, D.C.
Alice Clague and J.C. Ridley (1973) The Assessment of Three Methods of Estimating Births Averted, in Computer Simulation in Human Population Studies, (Edited by B. Dyke and J. MacCluer), pp. 329-382, Academic Press, New York.
Susan DeVos, and A. Palloni. (1989) Formal Models and Methods for the Analysis of Kinship and Household Organization. Population Index 55:174-198.
Eugene A. Hammel, C. K. McDaniel, and K. W. Wachter (1979) Demographic Consequences of Incest Prohibitions: A Microsimulation Analysis. Science 205:972-977.
Eugene A. Hammel, K. W. Wachter, and C.K. McDaniel (1981) The Kin of the Aged in A.D. 2000, in Aging (Edited by by S. Kiesler, J. Morgan, and V. Oppenheimer), pp. 11-40, Academic Press, New York.
Eugene A. Hammel, D. Hutchinson, K. Wachter, R. Lundy, and R. Deuel (1976). The SOCSIM Demographic-Sociological Microsimulation Program Operating Manual. Institute of International Studies Research Monograph No. 27, University of California, Berkeley, California.
Eugene A. Hammel, C. Mason, K. Wachter, F. Wang, and H. Yang (1991) Rapid population change and kinship : the effects of unstable demographic changes on Chinese kinship networks, 1750-2250, in Consequences of Rapid Population Growth in Developing Countries (Edited by G. Tapinos, D. Blanchet, and D. Horlacher), pp. 243-271, Taylor and Francis, New York.
Nancy Howell (1979) Demography of the Dobe !Kung. Academic Press, New York, Chapters 5, 11, 14,15, and 16.
Peter Kunstadter R. Buhler, F. Stephan, and C. Westoff (1963) Demographic Variability and Preferential Marriage Patterns. American Journal of Physical Anthropology 21: 511-519.
Jiang Lin (1992) Parity and Security: A Simulation Study of Population Aging, Kinship Network, and Old Age Security in China. Ph.D. Dissertation, Department of Demography, University of California, Berkeley, California.
Jiang Lin.(1994) Parity and Security: A Simulation Study of Old Age Support in Rural China. Population and Development Review 20:423-448
A. V. Rao, Q. Lindsey, R. Bhavsar, B. Shah, D. Horvitz, and J. Batts (1973) The Evaluation of Four Alternative Family Planning Programs for Popland, a Less Developed Country, in Computer Simulation in Human Population Studies, (Edited by B. Dyke and J. MacCluer), pp. 261-304, Academic Press, New York.
Jaxk Reeves (1982) A Statistical Analysis and Projection of the Effects of Divorce on Future U.S. Kinship Structure. Ph.D. Thesis in Statistics, University of California, Berkeley
Jaxk Reeves (1987) Projection of Number of Kin, in Family Demography (Edited by J. Bongaarts, T. Burch, and K. Wachter), pp. 228-248, Clarendon Press, Oxford.
James E. Smith (1987) The Computer Simulation of Kin Sets and Kin Counts, in Family Demography (Edited by J. Bongaarts, T. Buch, and K. Wachter), pp. 249-266, Clarendon Press, Oxford.
Prof. Kenneth W. Wachter
Department of Demography, UCB
2232 Piedmont Avenue
Berkeley, CA. 94720-2120
phone 510-642-1578; 510-642-9800; fax 510-643-8558
e-mail wachter@demog.berkeley.edu