Family Reconstitution Research Paper

Family reconstitution revolutionized the study of historical population processes, allowing estimation of demographic rates that was virtually impossible by other means. Demographers would ordinarily estimate vital rates by comparing counts of vital events to counts of the population at risk. For example, the general fertility rate could be estimated by dividing the number of births in a year by the number of women aged 15–49. However, accurate censuses contemporary with records of vital events are rare in the past. Family reconstitution not only overcame this difficulty by counting both events and persons, but also afforded a methodological alternative that is superior in many ways to estimation from census data, because the reconstitution data are on individuals, so that the characteristics of persons at risk are more specifically known. For example, estimates of fertility can be specific for unmarried, married, first-married, and remarried women, can include information on duration of marriage as well as age, and so on. It would be very rare to find information tabulated in such detail in published censuses, even in modern times. The technique of family reconstitution also led to important insights. For example, it showed (in contrast to the expectations of Notesteinian theory) that the decline of fertility in some places occurred well before economic modernization and before any de-cline in mortality.

1.1 History And Scope Of The Technique

Family reconstitution as a demographic technique was developed initially by Swedish demographers (Edin 1915, Hyrenius 1942, 1958) and advanced by the French (Fleury and Henry 1985, Gautier and Henry 1958, Henry 1956, 1967, Henry and Blum 1988), then by English scholars (Wrigley 1966, 1969, 1972, 1973, 1997, Wrigley and Schofield 1973, Wrigley et al. 1997). It linked by names of persons the parish records of baptisms, burials, and marriages of individuals (hence the technical phrase, nominal data linkage). Genealogists had always used the linking of disparate records by names of individuals. Its large-scale use by historical demographers was an innovation. It is of course now commonplace, especially using unique identifiers, in the management of medical, insurance, taxation, police, and other data, especially by computer. In the hands of historical demographers, the ecclesiastical ceremonies noted in parish records served as proxies for the underlying births, deaths, and onsets of cohabitation. It must always be remembered that the events in parish records are not the same as the underlying demographic events, e.g., a baptism is not a birth. With that caveat, a family reconstitution should provide information on the dates of birth, marriage, widowhood, remarriage, dates of the birth of children, and of death for individuals. It is possible in principle to continue such linkages across succeeding generations, or across collateral lines. Historians and demographers can examine family reconstitutions for patterns of fertility, mortality, and nuptiality, and to a limited extent those of migration in historical populations. The technique can be extended to the study of any other type of social or biological phenomena supported by the particular dataset or other data such as medical, notarial, or tax records that can be linked to it, e.g., heritability of cause of death, marriage patterns between social groups, fertility by social class, etc. For example, family reconstitution can be used in investigation of genetic characteristics (Bean 1990), or the relationship between fertility patterns and socioeconomic and cultural characteristics (Hammel 1995), or in patterns of fertility change within and between family lines (Hammel and Goldstein 1996).

2. Related Methods And Expansions

2.1 Genealogies

Genealogies are records of the linked lives of persons related by consanguinity and affinity and can, like reconstituted families, be used to extract demographic information. They are usually very selective, however, and unlikely to be representative of the broader population. Aristocratic and wealthy families and lineages are more likely to be recorded than those more characteristic of the population. Within genealogies, persons who do not leave descendants are more likely to be forgotten. Males are more likely to be remembered than females. Females may be less well remembered if they have fewer male children of consequence with progeny, so that events linked through females are likely to be lost (especially if the females were in informal liaisons with the males of the lineage). We might expect such biases to shift in matrilineal societies, but no data on such differences have been brought to light.

2.1.1 Ortssippenbucher. An important exception to some selection biases of genealogies are the Ortssippenbucher of some German regions, compiled by local genealogists from parish registers primarily to demonstrate the ethnic purity of families. These are genealogies of the common folk (but of course selective as their purpose indicates), and have been used for demographic estimation (Knodel 1988, Knodel and Shorter 1976).

2.1.2 Qing Dynasty Genealogies. A second exception to some of these biases are the imperial genealogies of the Qing Dynasty in China, which were continuously compiled for all descendants of the founder and have also been used for demographic estimation (Lee et al. 1993). Females are relatively well recorded in these data.

2.2 Civil Vital Event Registers

Civil registers of vital events have the same potential as parish registers. They are intended to cover all faiths in a civil jurisdiction, but they may under-report some population segments such as vagrants or the poor. In some regions, civil registers can serve as a continuation of earlier parish registers and are especially valuable if secularization progresses and observation of ecclesiastical ceremonies declines.

2.3 Population Registers

Historically, civil and ecclesiastical population registers were periodic listings of the inhabitants of residential units. They have the advantage of preserving the identity of the residential unit (usually ‘house-holds’) within which the demographic events occurred, although just what that unit was is itself often problematic, since actual and legal residence might be confused, and even long-term out-migrants might sometimes be continued in the listings. (Such continuation is not unknown in some parish records as well, e.g., births to out-migrant former inhabitants.) Population registers accomplish what is most difficult in family reconstitution, namely the nominal data linkage within and between individual lives, because local recorders presumably knew the identities of the persons and would not ordinarily confuse one John Smith with another John Smith. Thus, the families discernible in population registers are not so much reconstituted as preconstituted. The occurrence of vital events is implied through changes in household membership. However, the longer is the cycle ac-cording to which the registers were updated (usually at best annually), the more uncertain are the precise dates of the events leading to changes in membership, each update being only a terminus ad quem, and thus the more uncertain the exact periods of risk.

2.3.1 Civil Population Registers. Perhaps the most celebrated historical population registers are the household registers of Tokugawa Japan c. 1650–1868 (Hayami 1969, Smith et al. 1977) and later of Japanese-occupied Taiwan c. 1904–45, but registers were also kept in some European countries, e.g., Belgium (Gutmann and van de Walle 1978) and Italy (Kertzer and Hogan 1989). Records in such registers may sometimes be linked to independently kept vital event registers, both civil and ecclesiastical. (The Tokugawa registers actually had an ecclesiastical purpose, namely surveillance of the spread of religious sects, i.e., Christianity.)

2.3.2 Libri Status Animarum. Libri status animarum (‘books of the state of souls’) were ecclesiastical population registers that often contained not only registration information (e.g., which members of a household had been catechized, or which had been confirmed) but also the dates of these events and of baptism, marriage, and burial. Known by various terms in Catholic and Protestant regions, they vary in continuity and the diligence with which they were kept. Among the best are the Swedish catechitical and similar records (Lext 1984).

3. Methods Of Family Reconstitution

The work of family reconstitution was from its inception strongly dependent on the paleographic and historical skills and intuition of the scholar. The earliest efforts were done by hand, starting with a marriage record, then attempting to find the burial and any remarriage records of the spouses, then the baptismal records of the children, then the marriages and deaths of those children. Such data were entered on a ‘marriage fiche’ or ‘family reconstitution form’ for later collation and analysis. Account had to be taken of the literacy of scribes, variations in spellings of the same names, variations in standardization of surnames, and of course the completeness of the records themselves. Later attempts were assisted by computers, but final judgments often still depended on intuition in deciding between two or more equally plausible linkages. Efforts have been made to completely automate the linkage procedure, including the use of mathematical programming methods to resolve ambiguities, although it is not clear whether the algorithms themselves may not simply substitute new for old biases (Wrigley and Schofield 1973, Welford 1989, Shannon et al. 1989, Skolnick 1973, Schofield 1992, Nault and Desjardins 1989, Bouchard 1992, Bengtsson and Lundh 1991, 1993). It is difficult to know when the correctness of the nominal links in a dataset has been maximized, especially when data are scanty or contain errors. For example, if surnames are often omitted in the parish documents, and especially if the stock of first names is small (e.g., ‘John married Mary, June 1, 1712’), the number of alternative linkages can be very large because many Johns may have married Marys on that date. In principle linear programming methods can be used to obtain the best solution, one that maximizes the density or quality of linkage, but that solution may not be historically correct. Estimates of central tendency (e.g., the general fertility rate) will not be biased by such presumably random error, but variances will be, as will be any associations with other variables if guesses are more likely to be bad for some classes of persons. The results of nominal record linkage may be compared to population registers, but population registers also have problems of accuracy. Such problems of optimizing linkages are of course less when data are detailed (e.g., Swedish parish and population register records) or contain ethnographically validated clues to linkage (e.g., interfamilial patterns of godparenting in some South Slavic contexts wherein children of a marriage tend to have the same godparents, who also serve as marriage witnesses, etc.).

4. Limitations Of The Technique

Although a powerful tool, family reconstitution has important limitations caused by missing data. Some stem from the nature of the sources.

4.1 Data Quality And Coverage

While some parish records (e.g., Florence, Istria) go back to the fifteenth century, most do not commence until the sixteenth, and many are not reliable until the eighteenth century. Civil event and population registers are generally no older than the nineteenth, but ecclesiastical registers are often as old as the corresponding event registers. Libri status animarum were only irregularly kept in many areas, and all ecclesiastical records were subject to variations in quality depending on the abilities and dedication of the recording priest, attempts by higher authorities (both ecclesiastical and civil) to collect sufficient and standardized information, and the vagaries of copying and preservation. Ecclesiastical records are known for Catholic and Protestant populations but seldom for Orthodox Christians (not to mention Muslims, Jews, or others). Thus in areas of mixed religion, for example, in nominally Catholic Croatia or parts of Bosnia, large non-Catholic proportions of the population would not be included in the data and investigation of regional patterns would be hampered. Remarkable discontinuities in the data can occur in periods of shifts in religious adherence, as for example sharp swings between conformity and nonconformity in English parishes, or the widespread conversion of Catholics to Islam in the western Balkans. Similar problems would arise in using missionary records outside of Europe and the Americas. Civil event and population registers are by their nature relatively independent of religious bias, although they may miss subpopulations whose members for other reasons are under-registered.

4.2 Statistical Biases Caused By Sample Selection

Probably the most serious problems for family reconstitution are selection bias and truncation, issues addressed early by L. Henry (1956, 1967).

4.2.1 Selection By Data Quality. Where data are in general of acceptable but varying quality, there is an inherent tendency to want to analyze records that are only of the best quality. The socioeconomic condition and demographic regime of a part of a population that has good records may be quite different from those of another part that has poor records. Analysis limited to only the best data may give a misleading picture of the whole, either with respect to entire parishes or to subpopulations within parishes. This kind of selectivity bias is similar to that experienced in the analysis of genealogical data.

4.2.2 Selection By Migration. Because the work of linkage is so tedious, even with computer assistance, most reconstitutions have been limited to single parishes. Out-migration from such parishes will ordinarily go unremarked in parish records (and even sometimes in household registers). Since the kinds of persons who complete their lives within a parish may be quite different from the kinds who leave it, building a demographic picture of a parish based on completed lives may be misleading. These and similar problems (see below) can be partly overcome by analysis of a set of contiguous parishes, especially if migration tended to be short-range. Migration effects may be powerful where migration flows were strong (viz. Ruggles 1992, Kasakoff and Adams 1995, cf. Wrigley 1994). Reconstitution data have been used to examine migration per se (Bideau and Brunet 1993).

4.2.3 Truncation And Censoring. The time actually spent in a parish by a person who was not born in it or did not die in it may be unknown, and thus the time at risk of some demographic event may not be known. Many demographic measures cannot be properly estimated if time at risk is not known. For example, a woman may have had recorded births but no recorded death. The amount of time she was actually at risk of having the next birth cannot then be accurately estimated. Thus, the fertility of women at her achieved parity or at ages higher than that of her last recorded birth cannot be estimated accurately. For example, estimating the total fertility rate for the population would demand knowledge of the age-specific fertility rates by the usual five-year age groups or by single years, and the loss of information on some women who died without record within such spans would deflate the denominators of the rate calculations.

In general, observation on individuals at risk of some event is said to be truncated if the size of the cohort at risk to which the individuals belong is unknown and censored if the number of individuals at risk is known but the events are unobserved or if observed events are recorded only in time intervals instead of accurately. Examples are: (a) death records lacking age so that date of birth is unknown and the number of persons in the birth cohort cannot be recovered from birth records (truncated on the left); (b) death records with age recorded but birth records unavailable so that information is obtained only on persons who died at each age, not on those who survived that age (truncated on the right); (c) no marriage records but death records with age information showing survival and marital status so that the birth cohort size is known but the date of marriage is not (censored on the left); (d) birth and marriage records but no information on death so that observation ceases on the date of marriage (censored on the right). If statistical methods take censoring into account, censoring induces no bias in the computation of rates or probabilities of occurrence of an event, or in event-history analyses. By contrast, truncation does induce such bias. As a general rule, the onset or cessation of observation of an individual at risk should not be correlated with the event of interest.

Solutions to the problem of missing data in sets of life histories are various but problematic. They include using the occurrence of another event to the same individual (e.g., next births if interest is in mortality), random censoring, occurrence of an event to a close family member (thus implying continued residence and at least potential registration of an event), or statistical imputation of the censoring point, etc. (Hammel 1993, Trussell and Guinnane 1993).

4.2.4 Bias From Delays In Recording. Bias may also be created by delays in the recording of demographic events or in the ceremonies that recognize them. An individual may die or move before an event is re-corded and therefore be missed. Such delay is especially problematic for births and infant deaths. If the date of recording is accepted as the date of birth, long delays will make infants seem younger than they are, biasing any age-related computations. If an infant dies before it can be baptized, there will be no record of the birth, and often there will be no record of the death since unbaptized infants often received no formal burial. If the death is recorded and an age at death is given, the date of birth can be imputed and thus recovered. If death is unrecorded, mortality may be underestimated. Where both birth and death are unrecorded, infant mortality will also seem to be lower than it really is. Since in many historical populations up to a third of infants died in their first year, and half of those died in their first month, such statistical biases can be very important.

5. Prospects

The most promising outlook for family reconstitution and related techniques include the extension and elaboration of computer-based mathematical programming methods for rigorous and testable nominal data linkage, the combination of multiple data sources such as parish event registers, population registers, tax and notarial records, etc. to increase the accuracy and scope of data, and the use of event-history techniques to overcome problems of missing event dates.

6. Summary

Despite difficulties imposed by the quality and cover-age of the historical data and the possibility of biased inference, family reconstitution has given demo-graphic and historical scholarship a remarkable window on the demographic past, unachievable by any other means. In so doing it provides information for much longer spans of time than do modern data and often for periods before any effects of industrialization or modernization have been felt. It thus allows examination of longer term trends under broader socioeconomic conditions, and like all historical enterprises allows scholarship to escape the constraints imposed by modern conditions.

Bibliography:

1. Bean L L 1990 The Utah Population Database: demographic and genetic convergence and divergence. In: Adams J, Lam D A, Hermalin A, Smouse P E (eds.) Convergent Issues in Genetics and Demography. Oxford University Press, New York, Oxford
2. Bengtsson T, Lundh C 1991 Evaluation of a Swedish computer program for automatic family reconstitution. Lund Papers in Economic History 8. Lund University, Department of Economic History, Lund, Sweden
3. Bengtsson T, Lundh C 1993 Name-standardisation and automatic family reconstitution. Lund Papers in Economic History 29. Lund University, Department of Economic History, Lund, Sweden
4. Bideau A, Brunet G 1993 The construction of individual life histories: application to the study of geographical mobility in the Valserine Valley in the nineteenth and twentieth centuries. In: Reher D, Schofield R (eds.) Old and New Methods in Historical Demography. Clarendon Press, Oxford, UK
5. Bouchard G 1992 Current issues and new prospects for computerized record linkage in the province of Quebec. Historical Methods 25: 67–73
6. Edin K A 1915 Studier i Svensk fruktsamhetsstatistik. Ekonomisk Tidskrift 9: 251–304
7. Fleury M, Henry L 1985 Nou eau manuel de depouillement et d’expoitation de l’etat civil ancien, 3rd edn. Institut National d’Etudes Demographiques, Paris
8. Gautier E, Henry L 1958 La population de Crulai, paroisse Normande au XVIIe et au XVIIIe siecles. Presses Universitaires de France, Paris
9. Gutmann M P, van de Walle E 1978 New sources for social and demographic history: the Belgian population registers. Social Science History 2: 121–43
10. Hammel E A 1993 Censored intervals in family reconstitution: a sensitivity test of alternative strategies with historical Croatian data. In: Reher D, Schofield R (eds.) Old and New Methods in Historical Demography. Clarendon Press, Oxford, UK
11. Hammel E A 1995 Economics 1: culture 0. Fertility change and differences in the northwest Balkans 1700–1900. In: Green-halgh S (ed.) Rethinking Reproduction: Culture and Political Economy in Demographic Behavior. Cambridge University Press, Cambridge, UK
12. Hammel E A, Goldstein J 1996 Dick Easterlin in Slavonia: patrilineal fertility decline on the Croatian frontier 1700–1900. Paper presented to the Population Association of America, New Orleans, May
13. Hayami A 1969 Aspects demographiques d’un village japonais, 1671–1871. Annales E.S.C. 3: 617–39
14. Henry L 1956 Anciennes familles Gene oises: Etude demo-graphique, XVIe–XXe siecle. Presses Universitaires de France, Paris
15. Henry L 1967 Manuel de demographie historique. Droz, Paris
16. Henry L, Blum A 1988 Techniques d’analyse en demographie historique, 2nd edn. Institut National d’Etudes Demographiques, Paris
17. Hyrenius H 1942 Estlandss enskarna. Demograkska studier. Lund University, Lund, Sweden
18. Hyrenius H 1958 Fertility and reproduction in a Swedish population without family limitation. Population Studies. 12: 121–30
19. Kasakoff A B, Adams J W 1995 The effect of migration on ages at vital events: a critique of family reconstitution in historical demography. European Journal of Population 11: 199–242
20. Kertzer D, Hogan D P 1989 Family, Political Economy, and Demographic Change: The Transformation of Life in Casalecchio, Italy, 1861–1921. University of Wisconsin Press, Madison, WI
21. Knodel J E 1988 Demographic Behavior in the Past. A Study of Fourteen German Village Populations in the Eighteenth and Nineteenth Centuries. Cambridge University Press, New York, Cambridge, UK
22. Knodel J E, Shorter E 1976 The reliability of family reconstitution data in German village genealogies (Ortssippenbucher). Annales de Demographie Historique 13: 115–54
23. Lee J, Campbell C, Feng W 1993 The last emperors: an introduction to the demography of the Qing (1644–1911) lineage. In: Reher D, Schofield R (eds.) Old and New Methods in Historical Demography. Clarendon Press, Oxford, UK
24. Lext G 1984 Studier i svensk kyrkobokforing 1600–1946. Goteborg University, Goteborg, Sweden
25. Nault F, Desjardins B 1988 Recent advances in computerized population registers. Historical Methods 21: 29–33
26. Nault F, Desjardins B 1989 Computers and historical demography: the reconstitution of the early Quebec population
27. Collection de Tires a Part, No. 266. Universite de Montreal, Departement de Demographie, Montreal, PQ, Canada
28. Reher D, Schofield R (eds.) 1993 Old and New Methods in Historical Demography. Clarendon Press, Oxford, UK
29. Ruggles S 1992 Migration, marriage, and mortality: correcting sources of bias in English family reconstitutions. Population Studies 46: 507–22
30. Schofield R 1992 Automatic family reconstitution: the Cambridge experience. Historical Methods. 25: 75–9
31. Shannon H S, Jamieson E, Walsh C, Julian J A, Fair M E, Buffet A 1989 Comparison of individual follow-up and computerized record linkage using the Canadian mortality data base. Canadian Journal of Public Health 80: 54–7
32. Skolnick M 1973 The resolution of ambiguities in record linkage. In: Wrigley E A (ed.) Identifying People in the Past. Edward Arnold, London
33. Smith T C, Eng R Y, Lundy R T 1977 Nakahara: Family Farming and Population in a Japanese Village, 1717–1830. Stanford University Press, Stanford, CA.
34. Trussell J, GuinnaneT 1993 Techniques of event-history analysis. In: Reher D, Schofield R (eds.) Old and New Methods in Historical Demography. Clarendon Press, Oxford, UK
35. Welford J A 1989 Nominal Record Linkage: the Development of Computer Strategies to Achieve The Family-Based Record Linkage of Nineteenth Century Demographic Data. University Microfilms International, Ann Arbor, MI
36. Wrigley E A (ed.) 1966 An Introduction to English Historical Demography from the Sixteenth to the Nineteenth Centuries. Basic Books, New York
37. Wrigley E A 1969 Population and History. McGraw-Hill, New York
38. Wrigley E A 1972 Some problems of family reconstitution using English parish register material: the example of Colyton. Proceedings of the Third International Conference of Economic History, Munich 1965, Section VII, Demography and Economics, La Haye, Paris
39. Wrigley E A (ed.) 1973 Identifying People in the Past. Edward Arnold, London
40. Wrigley E A 1994 The effect of migration on the estimation of marriage age in family reconstitution studies. Population Studies 48: 81–97
41. Wrigley E A 1997 How reliable is our knowledge of the demographic characteristics of the English population in the early modern period? Historical Journal 40: 571–95
42. Wrigley E A, Davies R S, Oeppen J E, Schofield R S 1997 English Population History from Family Reconstitution, 1580–1837. Cambridge University Press New York, Cambridge, UK
43. Wrigley E A, Schofield R S 1973 Nominal record linkage by computer and the logic of family reconstitution. In: Wrigley E A (eds.) Identifying People in the Past. Edward Arnold, London, pp. 64–101