Z

Zelnick, Melvin, on mortality rates, 334 -335

---

1. See Laslett 1989 and the writings of Matilda White Riley (e.g., Riley 1987). On Europe, see particularly Laslett, "The aging of European populations and its social effects" (in press).

2. Tables 1.1 and 1.2 indicate that the original claim is nearer the truth for eastern Europe (including Germany) and also in some respects for southern Europe but is likewise too high for Britain's neighbors in western and northern Europe. If the whole depth of time before the 1880s is taken into account for England, however, longevity at birth has indeed doubled, and proportions of elderly have in fact trebled.

3. Nevertheless, they have been assembled under one heading and described as "The Second Demographic Transition," continuing falls in fertility and mortality being the leitmotiv. See van de Kaa 1987 and Lesthaeghe 1991-1992. It is interesting that demographers seem to have hesitated to include aging in the second demographic transition, though it is surely more clearly demographic than some of the other developments.

4. The terms "secular shift in aging," "lower and higher aging plateau," and "the Before" and "the After" were introduced in an attempt to distinguish aging from the other constituents of "modernization" and the aging process from the demographic transition or transitions. It was hoped to refer aging to the appropriate dates, periods, and stages of national development. See Laslett [1976] 1985 and 1984.

5. It is this which makes it inadvisable to use for properly historical purposes the demographic statistics ingeniously worked out by students of contemporary tribal communities. The remarkable results published in 1979 by Nancy Howell, Demography of the Dobe !Kung , for example, include aging estimates (see her chaps. 2 and 5) for an extant hunter-gathering society. But they do not necessarily indicate the situation during that era of universal hunter-gathering supposed to have preceded the agricultural era in which our own peasant ancestors lived.

6. For archaeodemography, see Acsádi and Nemeskéri 1970. Its inaccuracies and limitations are discussed in Petersen 1975.

7. For a specifically demographic work citing Ulpian's life table (considered by other demographers as lacking all basis in observation) and other historical fragments, see Dublin, Lotka, and Spiegelman [1936] 1948. It could be claimed that the parish registrations that yield the data for the demography of aging before the statistical era (see below) were likewise innocent of any demographic intent (see Laslett 1966) as were such sources as the censuses of China. It is the much more systematic character of the parish registers of the Christian church in European and Europeandecended societies and their persistence over long periods of early recordings that make them susceptible to satisfactory numerical analysis.

8. See my review of Georges Minois's History of old age (1989), ending "the hell with the Seven Ages of Man" (Laslett 1990).

9. Countries vary in this respect: the official U.S. census is closed for 50 years after being carried out, the British for 100 years. The Scandinavian countries are the most generous of access and their state-collected records go back as far as any of those elsewhere. Unfortunately, the original documentation for the censuses, even in countries like these, is very unevenly preserved. As for census-type documents drawn up before the statistical era, the situation is extremely uneven. In England, for example, whereas parochial registration of births, marriages, and deaths goes back to the 1530s and has made it possible to estimate age composition and life expectation, only about a dozen reliable census-type documents specifying ages are available before the official census began to record ages at the decennial census of 1831.

10. For examples, see Kertzer and Hogan 1989, Janssens 1993, and Bulder 1993.

11. These results are contained in Wrigley and Schofield [1981] 1989.

12. For an explication of demographic aging, particularly in relation to the demographic transition, see Chesnais [1986] 1992, esp. 274-289, and 1990. In accepting for the purposes of explanation the well-established model of the demographic transition, I have no wish to underwrite the functionalist theoretical framework or the assumption of inevitability as part of the modernization process which so often attaches to it. See the perceptive study by Simon Szreter (1993).

13. The recordings of baptisms, marriages, and burials from which the data used by Wrigley and Schofield [1981] (1989) for their monumental reconstruction were collected by several hundred volunteers (see their preface). Because gender is unspecified in the English parish registers that these volunteers used, they would have had to rely on given names and other clues to decide it. This was too laborious and uncertain a procedure to be worthwhile.

14. See the discussion in Laslett 1989, rev. ed., chap. 4, where possible projections for the twenty-first century in western Europe are considered. If substantiated, the more extreme of these forecasts would put life expectation at a maximum of 100 years at the end of that century rather than 85 years as indicated in figure 1.1. Under these extraordinary conditions, nothing like a plateau of the kind shown before the secular shift could be supposed.

15. Jacques Vallin (see Schofield, Reher, and Bideau 1991) suggests that the English evidence is not as reliable as the French because it depends to such an extent on model life tables, whereas the French is based on family reconstitutions. The English figures are, however, at present being checked against similar reconstitutions for English parishes. They are in general well confirmed. But it has to be accepted that English reconstitution evidence is never as reliable for this purpose as the French because English parish registers are less detailed in their descriptions of baptisms, marriages, and burials.

16. As has apparently been done by Vallin (1991: 48, fig. 3.3). The impression given by this figure is of a considerably less impressive climb at the end of the nineteenth century, and the appearance is more of a steady slope upward toward a substantial rise in the earlier 1900s. If the archaeodemographic evidence had been included in the analysis of longevity and a very much longer lapse of time thus taken into account, the impression of unceasing constancy would be very strong indeed.

17. See, for example, Kertzer and Schaie 1989, which includes historical studies of retirement both in Europe and America (Plakans 1989 [Europe], Vinovskis 1989 [America]).

18. Studies of these effects are particularly associated with Glen Elder and Tamara Hareven. See Hareven's chapter (with P. Uhlenberg) in this volume, with its references, and Elder 1974 and Hareven 1978.

19. See Fries 1980 and Fries and Crapo 1981. The critique in Laslett 1989, rev. ed., cites other discussions of their position, mostly dissenting from the claims made for the progressive compression of mortality and of morbidity in the final years. For the distribution of episodes of illness over the life course in the past as in the present and the drift toward their clustering in the later years, see the recent work of James C. Riley and others (e.g., Riley 1989). A research project on the statistical, demographic, and especially the historical aspects of life and death at the latest ages, concentrating on maximal length of life, is in progress at the Cambridge Group for the History of Population and Social Structure. It is working in conjunction with the Oldest Old Data Archive in Odense, Denmark, using the data so far assembled there. It has been incontrovertibly established that survival is decidedly improving at the latest ages at the present time, a circumstance not easy to reconcile with current compression of mortality in the oldest age group. See Kannisto 1994.

20. Estimates based on family reconstitutions at the Cambridge Group, subject to considerable uncertainty because of the difficulties of being confident of making reliable matches between personal names over periods of time greater than 50 years. The Oxfordshire villages were studied by Bridget Taylor.

21. On low life expectancy among British peers, see Hollingsworth 1964 and 1977. His estimates are now under further development at the Cambridge Group, where he has been kind enough to deposit the data.

22. Houston 1992, a very useful comparative analysis.

23. See Laslett 1977: 186. Iceland in 1729 did not have a population of what was even then thought of as on a national scale; there were only about 40,000 people. The aging peak of 1729 came in a year of disaster from smallpox, which seems to have affected younger people far more than the elderly. The Icelandic census gives 9.2 percent over 60 in 1703 and 8.7 percent in 1787.

24. See Houston 1992, where the issue is handled with considerable skill.

25. See the appendix for the surprisingly high survival of those who reached later life in the Before. Since this part of the chapter was written, evidence has been accumulating at Cambridge suggesting that the general expectation of elite groups having had longer lives than the rest of the population in the past may have to be revised, at least in respect to persons past infancy and childhood. An estimate by Roger Thatcher puts the total of centenarians in England and Wales at about 40 in 1861, and present evidence can be said to demonstrate what is claimed in the text: that it was possible to continue until very late life on the lower aging plateau and even to attain age 100.

26. The values for numbers alive at successive ages and the pooled total of years of life still left for them to live are standard features of the life table. Even in the clearest exposition of the life table, the major instrument of demographic analysis and particularly of mortality (Dublin, Lotka, and Spiegelman [1936] 1949) fails to explain to the uninitiated reader why expectation of life behaves in this way. The figures for England and Wales in 1989-1991 come from a provisional life table of the Government Actuary and those for Canada from Bourbeau and Legaré 1982. An even more elementary error about expectation of life is occasionally found in his-

---

torical discussions, that supposing longevity at birth of 70, then a person of 50 has twenty years to live. That this is a gross misconception should be obvious from the discussion in the text.

27. See Wrigley and Schofield 1989: 530, table A-3-2, and the Chester Beatty, life tables. Japanese period life tables for dates over the century since 1891 have been used to work out generational (i.e., cohort) life tables. See Kobayashi and Nanjo 1988.

28. See Mitterauer 1992.

29. See Laslett, "Is there a generational contract?" in Laslett and Fishkin 1992, particularly the chapter in that volume by David Thomson.

30. On this topic, the supposition that an elderly population hinders national performance, see Laslett 1989, chap. 3; in relation to France, see Dyer 1977 and especially Bourdelais 1993, chap. 9.

31. For an attempt at a general theory of the proper use of evidence of this kind, especially literary evidence, when it cannot be supplemented and checked from other sources, see Laslett 1976.

32. It should be added that numbers of such specialist studies have avoided these tendencies, as, for example, those in this volume or in Pelling and Smith 1991 or Conrad and Kondratowicz 1993. Although the argument that France was already "modern" in terms of aging under the Enlightenment is not spelled out by Troyansky, it is not absent from all discussion of the topic.

33. See Laslett 1989: 105, quoting from Eric Midwinter.

34. See Szreter 1994 and Laslett 1987a , 1987b , and 1992 for the historical critique of modernization. Carmen Avalos del Pino is at present engaged at the Aging Unit of the Cambridge Group in the analysis of the results of attempting to apply the modernization model to aging in Catalonia during the secular shift. For the historical dissociation of constituents of modernization in the history of the West, see Laslett [1976] 1985, and for its relation with the so-called sexual revolution of the mid-twentieth century, see Laslett 1991.

35. See Anderson 1985. The accompaniment of the elongated vertical kin links, that is, the attenuation of lateral ones, is the result not of lengthened life but of declining fertility, an example of how intricate are the distinctions that have to be made. For the simulation of kin links over the life course, see Bongaarts, Burch, and Wachter 1987, especially the contributions of J. H. Reeves and J. E. Smith, and Laslett 1988b . Compare Ruggles 1994 and Zhao in press. The chapter by E. A. Hammel in this volume represents an advanced and sophisticated example of the use of simulation methods.

36. For England, see Laslett 1989, chap. 7, and for the United States, see, for example, Smith 1981 and the references in his contribution to this volume. On the two views of the change in the position of the elderly in America, see Fischer 1978 and Applebaum 1978.

37. On all these issues, see the considered discussion in Conrad 1991, with its references, and the facts set out in Laslett 1993.

38. On this paragraph, see Laslett 1989, chap. 8; 1977, chap. 5; and 1993; and compare the statements made by Hammel and the other authors in Part II, Living Arrangements, in this volume. Research is in progress at the Cambridge Group on the detailed social structure of thirteen English communities between 1891 and 1921 which may determine what effects the early path of the secular shift could have

---

had on the familial position of the elderly. Compare also Janssens 1992 and Bulder 1993 for the Netherlands with the work of Avalos del Pino (see note 34) for Catalonia, where she has shown that co-residence of parents with married children did indeed increase against the expectation raised by the modernization model.

39. See Laslett 1992; on the recent rapid rise in the primary individual, see Kobrin 1976. As for the United States, Steven Ruggles has laid it down that toward the end of the nineteenth century, all parents who had a married child to reside with took up that familial position. It was only after that that the rise of the primary individual began. See Ruggles 1994 and compare Laslett 1989, rev. ed., chap. 9.

40. See Laslett 1988a and its references, especially to the work of David Thomson and Richard Smith. The amount provided for the dependent elderly seems to have varied extensively from country to country and locality to locality. It was almost never sufficient to support a penniless individual by itself. For the Dutch case, in which the support was particularly exiguous, see Bulder 1993. A somewhat skeptical view as to the collective support of the elderly is put by Thane 1993 and in her forth-coming book there cited.

41. See Gordon 1988.

42. See the chapter by Ransom and Sutch in this volume, and compare Riddle 1992. These developments are summarized in Laslett 1989, chap. 9.

43. The four-stage division of the life course is set out in Laslett 1989, where the Fourth Age is defined as one of final decrepitude and dependency, not a necessary stage of the life course of any individual and evidently subject to reduction and alleviation by medical progress and social policy. Along with the Third Age, it was intended to replace the blanket term "old age," which can now be considered obsolete. But the concept of a Fourth Age has been rejected by some authorities (see Young and Schuller 1991). In my own view the Fourth Age does represent a historical reality and its retention as a concept is required by the definition of the Third Age. See Laslett 1994, a review article on the Carnegie Enquiry into the Third Age in Britain, defending the Third Age against that Enquiry's decision to give it a chronological connotation covering a period of life between 50 and 74, a decision unacceptable to the originator of the concept and to those in the Third Age themselves.

44. The whole issue of life in the Third and Fourth ages has to be seen in relation to the analysis now proceeding of expectation of healthy active life as distinct from expectation of life as such. See discussion in Laslett 1989, rev. ed., chap. 5, and its references especially to the works of J.-M. Robine.

45. Nevertheless, there was a multiplicity of conventional ages of withdrawal, especially from public affairs ("Stepping Down," see Plakans 1989), affecting the better off for the most part, withdrawal into a retirement that might come in middle life and be ended by taking another office. This complicated situation is classically described by Keith Thomas (1976). The greater flexibility of transition into the final stage during the Before together with elements already present, which are characteristic of the Third Age, are discussed in Laslett 1989, rev. ed., and 1993. Bonfield (1989) considers the question of the Third Age during the Before in legal terms.

46. In Laslett 1987a and 1989, the proportional condition to be satisfied was different—at least 10 percent of the population over 65 years of age. Other demographic indicators associated with the longevity criterion for the appearance of the Third Age are found in the appendix.

47. For details of these projections and their results, see Ogawa 1988 and the following references for readers of Chinese: Yao Shuben 1989, Zeng Yi 1990, and Wu Cangping and Du Peng 1992.

48. Laslett and Fishkin 1992. The joint introduction bears this phrase as its title, and it is claimed that political theorists have a quite inadequate conception of justice over time and lack the concept of processional justice as well.

49. See Laslett 1988a and the introduction to Pelling and Smith 1991. Some scholars suppose that the Western familial system did specifically provide for old age dependency by requiring married children to take in aging parents (see the preface and Kertzer's concluding chapter to this volume). In my view the situation of such persons, many of whom had no surviving child, or none in a position to support them, was too miscellaneous for this to be so. Provision for support was ad hoc on the part of both family and collectivity, and the relationship may have been contractual, contractual between generations (see Laslett and Fishkin 1992).

50. It would not have been advisable to go over to "West" altogether for the later figures, because this would have sacrificed the continuity throughout the whole series which alone makes comparison possible over time.

51. Similar cross-comparisons can be made to estimate roughly other values of interest to the reader.

52. Values have been taken from the Stable Population Series for Model North, using the columns for a female gross reproduction rate of 2.25. This is only an approximate average for English historical experience and is badly out at higher levels. No correction has been made to remedy this for the reason given in note 50 above.

53. See especially Bourdelais 1993, chap. 7. His argument is that measures of this kind should be employed for preference in measuring longevity and that when converted into proportions over certain ages by an operation not easy to describe in the present context, should replace conventional aging indicators. The intriguing outcome is that countries like France, whose great age in proportional terms has been a major theme of this chapter, can actually be said to have grown younger in the twentieth century when his suggested indicators are applied.

1. In the view of some, perhaps exactly the rate of population growth.

2. Rapid fluctuation in fads of child rearing may make them seem unuseful, but this is misleading. They are just a little out of phase.

3. Ruggles (1987) has made just this point.

4. A full exposition of these matters is given in Wachter, Hammel, and Laslett 1978. Details of the SOCSIM microsimulation programs are given in Hammel, Mason, and Wachter 1990. I am indebted to Carl Mason for the programming of these simulation runs.

5. I use here the demographic rates employed by Hammel, Wachter, and McDaniel (1981) in a first exploration of some of these issues, as improved by Reeves (1982, 1987) to include divorce, and further modified for this exercise to adjust for an improved marriage market algorithm. Details of these rates, as employed in this analysis, may be obtained from the author on request.

6. I use here the demographic rates employed by Hammel (1990), based on the work of Acsádi and Nemeskéri (1970), Russell (1958), Coale and Demeny (1983), and Laiou (1977), and taken as typical of a harsh premodern environment. Details of these rates, as employed in this analysis, may be obtained from the author on request.

7. More detailed specifications can be obtained from the author.

8. More detailed specifications can be obtained from the author.

9. See Hammel and Laslett 1974 for detailed definitions. Brief definitions, including those types omitted from the analysis here, are

10. The reader will note that the proportions of the different household types do not sum to 1, because the rare types are omitted from the table. A full table of results can be obtained from the author. Similarly, the standard deviations of these proportions are not given (see the section on confidence intervals later in the text). The standard deviations of these proportions as estimated by bootstrap techniques are very close to the theoretical expectations and can be calculated as

.

11. We should never find multiple lineal households under a system of nuclear reincorporation, except by bizarre combinations of divorces or spousal deaths and remarriages. We did our best to prevent the occurrence of such bizarre events.

12. See, e.g., Hammel 1972, 1980.

13. This is not a very deep proposition, ethnographically.

14. Of course, one person's child-in-law is another person's child, and that second person might move in with the child, if both of the co-in-laws were widowed. Each is eligible to move in with a married child if their child survives (although only one is permitted to do so, since co-in-laws may not themselves co-reside); only one of them is eligible if the other has no surviving children. We would expect these differences in extended lineal proportions to diminish if, under stem family rules, a surviving widowed parent were ejected by a surviving widowed child-in-law. We do not implement that heartless rule, even if there are no grandchildren.

15. Much the same point can be made in the classification of households themselves. The decision that specifies the head of household for the purpose of classification is crucial to the outcome.

16. The number of households ranged from about 2,000 under MED(S) conditions to about 7,000 under LPM(R) conditions.

17. See Wachter, Hammel, and Laslett 1978.

18. See Hammel 1990.

19. See Hammel et al. 1991.

1. A similar tendency of growing household complexity was observed in two Italian communities: Fagagna (Laslett 1983) and Casalecchio (Kertzer 1989).

2. Originally one unit of land held in villenage, called a "serf parcel," was considered to be necessary for supporting one peasant family.

3. In the villages of German ethnicity, the households were on the average smaller, and simple family households predominated (Andorka and Faragó 1983).

4. I am grateful to Sandor Balázs-Kovács for locating and giving me access to the population listings of Sárpilis and to Laszlo Kosa for finding and giving me access to the Mezocsoknya* listing.

5. The rate of mortality of ages 0-9 is used instead of the infant mortality rate, as it is often not clear from the register whether mortality cases given as "one year old" were really somewhat below or over the first birthday.

6. Therefore, the famous "one child family system" or the "sin of Transdanubia" (Vásáry 1989) still had no influence on the conditions of the elderly in 1804 in Sárpilis.

7. When evaluating the migration rates, it is worth noting that only Pinkenhof had a larger population than Sárpilis. The other villages were smaller.

8. The absence of single persons in the population over age 60 was so surprising that the four lists were very carefully reviewed to find out whether single persons were not "hidden" among the widowed. This review, however, proved that there were indeed no single men and women over age 60 in these lists.

9. The age of several wives in the listing of Sárpilis could not be identified.

10. The husband and the wife in these two couples were aged 78 and 47 and 67 and 47.

11. Tables 4.11 and 4.12 present the distribution of households in which persons over age 60 lived (i.e., not the distributions of the persons over age 60). The notes to these tables give additional data on the households in which two old persons lived, so that the distributions of persons over age 60 can be calculated.

12. The Laslett (1972) typology is supplemented by subdividing the multiple family household (= type 5.b), secondary unit(s) down category, into stem (5.b.1) and joint (5.b.2) subcategories. In the stem households, one parental conjugal family unit and one nuclear unit of an ever-married son or daughter or his or her widowed spouse lived together. In the joint households, one parental conjugal unit and more than one unit of ever-married sons or daughters or their widowed spouses lived together.

1. The crude external migration rate, CMP_e , and the corresponding crude internal migration rate, CMR₁ , are calculated as

Age-specific rates, indicated as ₅₅₊ CMR_e , for example, are equivalent to age-specific birth- or deathrates. Rogers and Castro (1986) employ a similar age-specific, although less intuitive, measure.

2. Our decision to view age 55 as the onset of old age is not altogether arbitrary. It does not rest on the Pinkenhofers' understandings of the beginnings of old age, to be sure, because we have not examined the qualitative evidence, such as folklore, that might reveal their understandings of age grading. However, in 1804, the provincial Landtag adopted a peasant law that required the creation of Wackenbücher , or registers of all serf labor obligations, that included age-specific categories of workers. The registers defined "able-bodied" (arbeitsfahige ) males as those under 60 years of age and "able-bodied" females as those under 55 (Kahk 1982; Plakans and Wetherell 1992).

3. In an intriguing comment in chapter 3 in this volume, Eugene Hammel suggests that "household . . . regimes [can] be distinguished at relatively small sample sizes." If sustained, Hammel's finding may obviate, or at least reduce, the inevitable worries of historians who are forced to deal with very small numbers of the aged in single historical communities.

4. In 1850, the age heaping index, which reveals rounding of ages, was 103.8 for males and 90.5 for females; with no systematic rounding, the index would be 100 (Shyrock and Siegel et al. 1973: 1:205-206).

5. We address this larger issue of support in the context of kinship and community in Wetherell, Plakans, and Wellman 1994.

1. Unfortunately, it is not possible to ask the same questions for the American Indian population or for the African-American population before the end of slavery.

2. The exception is the population of Webb County, which is dominated by the small city of Laredo. Laredo had a more varied economy.

3. There is a considerable background literature about residential arrangements in the United States in the nineteenth and early twentieth century. Unfortunately, not enough of it deals with rural, as opposed to urban, populations. For the best general overview and a novel interpretation, see Ruggles 1987.

4. See Smith 1979, 1984, and 1986; and Smith, Dahlin, and Friedberger 1979.

5. For an overview of the project, see Gutmann and Fliess 1989. For methodological considerations, see Holmes and Gutmann 1989, Gutmann and Holmes 1988, and Gutmann et al. 1989.

6. For an example of other work in this project, see Gutmann 1991.

7. The sampling scheme involves taking a sample of households headed by individuals whose surnames begin with certain letters of the alphabet, in order to be able to link from document to document. See Vetter 1990 for a discussion of the representativity of this method. The linkage techniques used are described in Vetter, Gonzalez, and Gutmann 1992.

8. Webb County existed by 1850, but it was enumerated together with other Rio Grande Valley counties, in which most of the Mexican-American population lived.

9. The sample was designed to maximize the possibility of linking individuals between census years, so there is some likelihood that an individual sampled in the 1870 census would also be sampled in the 1880 census. By combining 1870 and 1880, the likelihood that all the cases in the combined sample are independent of each other is reduced. This problem is minimized by the low level of linkage obtained in all but the German-American population (Gutmann et al. 1990) and by the fact that we divide the population into ten-year age groups. Thus persons aged 65 in 1870 would be aged 75 in 1880, and they would be found in different cells in most of our analyses.

10. Whether he would have lived with a child at the very end of his life is open to question. His living only with his wife at age 79 is a clear sign of a preference for living alone, especially in a strong kinship community such as Gillespie County.

11. In this we follow the work of Smith (1979) and Smith, Dahlin, and Friedberger (1979). Like Shanas and colleagues, we give priorities to some residential arrangements, in the event more than one occur simultaneously. First, the categories used here give preference to close kin over more distant kin or nonkin. Second, the categories give preference to residence with married children over all others and to residence with unmarried children over all remaining possibilities. Thus, if an individual lives with a spouse, an unmarried child, and a married child, he or she would be placed in the category of living with a married child. If an individual lives with an unmarried child, a spouse, and other kin, the appropriate category would be "unmarried child." The literature is vague about whether widowed children are considered married or unmarried. Here they are treated as unmarried.

12. These figures are drawn from the Texas historical demography project database, described above. Smith (1979: 287) reports results about the marital status of men and women in the United States as a whole in 1900. His results are similar to those reported here for men but higher than those reported here for women. Women in Texas were probably more likely to have married, because sex ratios were different in Texas in the nineteenth century than in the whole United States, improving the likelihood of women marrying.

13. A proportional weight is applied to each case so that for the total sample for any year the proportion in each ethnic group in the sample population is the same as that for the state as a whole as reported in census publications. In general, the weights for the European and Mexican origin populations are less than one, and those for all other groups are somewhat greater than one. No other corrections have been applied to the data.

14. For differential migration once they arrived in Texas, see Gutmann et al. 1990.

15. This was accomplished by weighting each household head by the size of the household he headed. If instead of all residents of the sample counties we look to

---

the economic activities of the households of the older population, the pattern is little different from that in table 6.4. In all ethnic groups the proportion of people aged 50 and older living in farm owner households and living in no occupation households is greater than in the population as a whole.

16. In 1860 and 1870, tenants were defined as those who were enumerated as farmers but who reported no real property. In 1900, tenants were those who were enumerated as farmers but who declared themselves as renters in the question about home ownership or rental. Also, see Houdek and Heller 1986 and Ransom and Sutch 1977.

17. For an explanation of such approaches, see Agresti 1990. The analysis was done using the logit program of the STATA statistical analysis package.

18. A small number of people appear more than once in this database, in those cases where a person stays in the sample after age 55 for two or more censuses. We have not corrected for the lack of independence this implies, because we estimate that fewer than 10 percent of the cases are individuals who appear more than once. The logit estimation procedure makes use of the data weights we have used elsewhere in the analysis reported here.

19. The categories were (1) farmer or farm tenant, (2) farm and nonfarm worker, (3) white-collar worker, (4) skilled craftsperson, and (5) no profession. The farmer and farm tenant category was used as the reference. These categories were significant for models in which living alone and living with unmarried children were the dependent variables but not for one in which living with married children was the dependent variable. Where they were significant, the three occupational categories (numbers 2, 3, and 4 above) had roughly similar values and similar signs. They were positively related to living alone and negatively related to living with unmarried children.

20. An analysis of ages of heads of household of extended and multiple households showed that the most likely age of head was under age 35 and that in most cases, the co-resident kin was the sibling of the head and not the parent.

21. The role played by property ownership and its passage can be studied using our cross-references linking census data to property tax data and linking census data across time periods. Those data are exceptionally complex, and we have not yet gotten far enough in the underlying project to exploit them for the study of property transfers from generation to generation.

1. There are measurement issues concerning how the ownership information was elicited in the American census. In 1890, the enumerator's question was, "Is the home you live in hired, or is it owned by the head or by a member of the family?" The question for the 1910, 1920, and 1930 censuses was explained to enumerators as follows:

Home owned or rented .—This question is to be answered only opposite the name of the head of each family. If a dwelling is occupied by more than one family, it is the home of each of them, and the question should be answered with reference to each family in the dwelling. If the home is owned , write opposite the name of the head of the family "O."

---

If the home is rented , write "R." Make no entries in this column for the other members of the family. . . . Owned homes .—A home is to classed as owned if it is owned wholly or in part by the head of the family living in the home, or by the wife of the head, or by a son, or a daughter, or other relative living in the same house with the head of the family. It is not necessary that full payment for the property should have been made or that the family should be the sole owner. (U.S. Bureau of the Census 1979: 30, 46)

The basic philosophy was to identify owner-occupied units, generally ignoring precisely who in the household owned the unit. This viewpoint has continued up to the present. In the past, ownership was attributed to the head of household. Thus it is possible that an aged person might have been given headship out of deference or respect, while not actually owning the dwelling. Similarly, a younger (and usually male) relative might have been attributed headship while the property was actually owned by an elderly parent. It is probable that these effects largely offset each other, but there remain real issues of measurement.

2. For some significant exceptions, see Venti and Wise 1989, 1990; Feinstein and McFadden 1989; Stahl 1989; Ai et al. 1990.

3. A national sample of the 1880 U.S. census is being prepared at the Social History Research Laboratory at the University of Minnesota. The ultimate goal is to sample the 1850-1870 censuses as well as the 1920 census.

4. The results from the 1890 U.S. census were omitted from figure 7.1 because they were very similar to the data from the 1900 census public use sample. The 1930 U.S. census data had to be interpolated to the five-year age categories of the other data sets.

5. The 1900 U.S. census asked four questions on ownership: "Owned or rented," "Owned free or mortgaged," "Farm or home," and "Number of farm schedule" (U.S. Bureau of the Census 1979: 34). The tabulations of the public use sample were done only for those heads of households who replied that they owned or rented homes. Those who owned or rented farms were excluded.

6. Other data not presented demonstrate the highest rates among the rural farm population.

7. It is possible to reconstruct this from published state- and city-level data in 1930. Tabulations can also be done from the various public use samples.

8. The Northeast (North Atlantic Region) was 59 percent urban in 1890, as compared to 33 percent in the Midwest, 16 percent in the South (South Atlantic and South Central regions), and 37 percent in the West (U.S. Bureau of the Census 1975: Series A172, 178-179).

9. The 1890 census also has information on the amounts of mortgage principal and interest rates, unlike the 1900 census, which only has information on whether the property was mortgaged or held free and clear. It is truly unfortunate that the enumerators' manuscripts of the 1890 census have been lost.

10. Direct standardization was done using the age structure of all heads of household in 1900.

11. An experiment was done to design a measure of the mean age at home ownership similar to John Hajnal's (1953) singulate mean age at first marriage. Unfortunately, the results appear to be very sensitive to the precise age categories used. The best that can be said is that the mean age of home ownership was rising from the late nineteenth century, along with the mean age of the population. The mean

---

age at home ownership for the overall nonfarm population calculated in this way rose from 33.3 years in 1890 to 39.3 years in 1900 to 43.7 years in 1930.

12. This approach imposes some structure on the age-ownership profiles and allows parameters from the structure to describe the data. A simple solution for this is a logit, whose basic formulation is

where ln is the natural logarithm, P_j is the probability of owning a home for the jth age group (0 << P << 1), b₀ is the constant, b₁ is the slope with respect to age, AGE is estimated mean of the jth age group, and e_j is an error term. The estimation was done using weighted least squares and a minimum chi-square criterion for fitting. The procedure used was glogit in STATA with aggregated (grouped) data. Since the age-ownership curves are clearly nonlinear, the logit specification is a means to attempt linearization. In addition, it transforms a variable (P) from one bounded at 0 and 1 to one bounded at

and and also has certain other desirable statistical properties (Maddala 1983: chap. 2).

The results suggest some increase in the steepness of the age-ownership profile from the mid-nineteenth century up to about 1900 and then rough stability between 1900 and 1930. The cohort phenomena that had led to the flattening out of the age-ownership profile in the middle of the nineteenth century had dissipated by the early twentieth century. The elderly in successive age cohorts began to be more likely to own property, which should not be surprising given earlier results. (When more complete data sets from successive censuses are assembled, a full cohort analysis can be done.) The upward shift in the curves between 1900 and 1930 (seen in fig. 7.1) is direct evidence of that cohort effect. The increased regularity of the age-ownership relation is accounted for by measures of goodness of fit—the adjusted R-squared values and F-ratios increase from the 1865 New York State and 1889-1890 survey data to the census data for 1890, 1900, and 1930. The differentials previously observed within censuses are not, however, immediately apparent in the slopes. In 1900, the black population definitely had a flatter profile, resembling those for 1865 New York and the 1889-1890 U.S. Commissioner of Labor Survey, but this difference had disappeared by 1930. Rural-urban and nativity differences in the slopes of the logits were not large after 1900, however. The major result remains the increased steepness of the age-ownership relation in the twentieth century.

The elasticity of ownership with respect to age is defined as

where

is the elasticity of ownership (P) with respect to age (X), and Z = b₀ + b₁ X is from the logit equation.

1. A further factor relevant here is age at marriage, historically high in western Europe, which affects the age of the widow(er)'s children at the time of widowhood. See R. Smith 1984: 424-425.

2. On the dynamics of the sharecropper household in Italy, see Barbagli 1984 and Kertzer 1984, 1989.

3. For a fuller analysis of mortality in Casalecchio, see Kertzer and Hogan 1989: 62-65.

4. A detailed analysis of fertility changes and fertility differentials in Casalecchio in these years is provided in Kertzer and Hogan 1989: 157-173.

5. Here we make use of the household composition typology described by Hammel and Laslett (1974).

6. Demographic and economic data on the Casalecchio population are drawn from the Casalecchio database. This consists of a reconstruction of the approximately 19,000 individuals who ever lived in Casalecchio during the period 1865-1921, plus the inclusion of individuals who appeared in the 1861 census. Data sources include manuscript censuses, the population register, annual household tax registers, conscription registers, and other sources. For a full description, see Kertzer and Hogan 1989: 189-208.

7. The agricultural wage laborers actually showed an increasing likelihood of living in multiple family arrangements over this period, passing from 14 to 36 percent. The nonagricultural wage laborers maintained a constant 18 to 20 percent multiple family proportion in these years. For more details, see Kertzer and Hogan 1989: 53-57.

8. The view that widowhood commonly affected younger women in the preindustrial past is disputed by Keyssar (1974), on the basis of his study of Woburn, Massachusetts, in the eighteenth century. Simulation methods are employed by Bongaarts (1989) to examine the impact of changing rates of mortality, remarriage, and divorce on the proportion of people widowed by age in the United States in 1800, 1900, and 1980.

9. No census was taken in Italy in 1891, and the 1901 manuscript census forms were not retained by the communes.

10. On household dyamics in southern Italy in the past, see Da Molin 1990.

11. In the period 1864-1870, 8.6 percent of the marriages in Italy involved a widow, while 15.2 percent involved a widower. In the period 1891-1900, the figures were 6.2 percent and 11 percent, respectively.

12. Much of this discussion of remarriage in Casalecchio is adapted from Kertzer and Hogan 1989: 145-146.

13. Of the 61 widowers over age 40 who remarried, 74 percent married women who had never previously been married.

14. The procedure also takes into account other forms of right censorship, most notably, the end of our period of reconstitution in 1921. For the use of these methods in historical migration research, see Hogan and Kertzer 1985. For other historical family applications, see Alter 1988 and Kertzer and Hogan 1989.

15. This procedure did not take into account subsequent changes in marital status. Thus, in fact, by the time many of these individuals emigrated, they may well have become widowed. The result of this procedure is to underestimate any difference in migration propensity between the widowed and the married.

16. For those widowed at ages 16-45 vs. 46-59, p << .005; for those widowed at ages 16-45 vs. 60+ and 46-59 vs. 60+, p << .0001.

1. The original schedules are not extant. Also returned by towns were the numbers of unmarried men from age 16 to 60; married men from age 16 to 60; boys aged 16 years and under; men aged 60 years and upward; and male and female slaves (Rossiter [1909] 1966: 149-151, tables 76 and 77). The first federal census of 1790 only recorded information on white males under and over age 16, white females, all other free persons, and slaves; the extant manuscript returns have been published (U.S. Bureau of the Census 1907). Tax data for 1790 are from Belknap (1970): II:226-242. I am indebted to Charles Wetherell for a machine-readable copy of the colonial censuses of 1767 and 1773 and the 1790 U.S. census returns for New Hampshire towns.

2. The figures are as follows:

1. Sampling strategies for the three public use samples (PUS) differed: the 1910 PUS is a 1-in-250 sample of census manuscript schedules, the 1940 PUS is a 1-in-100 sample of households and group quarters (sampling weights are used to get a representative sample of individuals), and the 1970 PUS is a 1-in-20 sample of census records. It was possible to exclude residents of New York City from the 1910 and 1940 samples but not from the 1970 sample. We excluded residents of New York City because of the atypicality of New York City's population.

2. Data on the Manchester cohorts in this chapter were based on a longitudinal historical data set that Hareven constructed for the parents' cohort and on which Family Time and Industrial Time is based (Hareven 1982). She linked this historical data with a new data set on the children's cohorts, which she generated from intensive interviews, demographic histories, work histories, and migration histories of the children's cohort during the period 1981-1985. The real names of people discussed or quoted in this paper have been substituted with invented names.

3. While most elderly widows lived with an adult child, most adults did not live with a widowed mother. Under mortality conditions existing around the turn of the century, only 40 percent of the mothers alive at age 35 would still be living at age 75. For every 100 mothers over age 70, there were 352 living children. Thus even if all older mothers lived with a child, only about 10 percent of all adults would ever have a 75-year-old mother living with them (1/3.52 × 0.4 = 0.11).

1. Estimates by Spurgeon Bell (1940: 11) and Alvin Hansen (1922: 504) put the industrial workers as a fraction of all gainfully employed workers in 1900 at between 32.4 and 38.4 percent; Stanley Lebergott (1964: 512-513, tables A3 and A4) estimates nonfarm employees (excluding domestic service employees) at 37.4 percent.

2. A sampling of prior views can be found in Bancroft 1958: 38-39; Mushkin and Berman 1947; and Achenbaum 1978. In Ransom and Sutch 1986a , we have traced the historiography of pre-1940 retirement in an attempt to understand the persistent tendency to underestimate the extent of retirement during the sixty years prior to World War II. This issue has more than a purely historical relevance; it has profoundly influenced the thinking of economists on current policy issues. The under-

---

estimate of retirement is explicit in the work of Michael R. Darby (1979) and implicit in that of Laurence J. Kotlikoff and Lawrence H. Summers (1981). For a corrective, see Ransom and Sutch 1986b .

3. The modern concept of labor force participation did not exist prior to 1940. Estimates for the years 1900 to 1930 had previously been based on backward extrapolations of the 1940 data to earlier years (Mushkin and Berman 1947). The new estimates of labor force participation we have made are based on published census data for 1870 and 1880, data from the manuscript censuses for 1900 and 1910, and a special study by the U.S. Bureau of the Census for 1930. For a discussion of these estimates and the methodology, employed for our estimates, see Ransom and Sutch 1986a , 1988, and 1989a , and Ransom, Sutch, and Williamson 1991.Jon Moen (1987) has challenged our approach, but we believe his critique was based on a misunderstanding (see Ransom and Sutch 1989a ; Ransom, Sutch, and Williamson 1991).

4. The retired are not in the labor force; thus the proportion retired is defined as one minus the labor force participation rate. Census figures report a fall in the rate of labor force participation of men 60 and older from 54.5 percent in 1950 to 32.2 percent in 1980; see figure 11.1. This development is one of the most dramatic changes in labor force behavior to have occurred since World War II. Only the increase in the labor force participation of married women during the same period rivals the change in male retirement in quantitative significance.

5. A private saving plan that protected workers with term life insurance and also provided a lump sum payment after twenty years of contributions was very popular in the late nineteenth century. These instruments were called "tontine insurance policies" and were sold by most insurance companies until they were outlawed in 1906. For a discussion of these plans and circumstances surrounding their abolition, see Ransom and Sutch 1987. Perhaps more than the current era, this period of history seems to fit the assumptions that underlie the life cycle model of saving. That model sees the need to make provisions for retirement as the major motive for private saving (Modigliani 1986).

6. For a more detailed account of Wright's pioneering effort in data collection and its impact on the Massachusetts Bureau of the Statistics of Labor, see Williamson 1967.

7. An index of all of the state reports was published by the U.S. Bureau of Labor (Wright 1902). Wright became the informal leader of the movement to establish the various state bureaus and to further the collection of survey data. In 1885, when Congress established the federal Bureau of Labor Statistics, Wright became its first director. In that capacity he coordinated the efforts to collect data at the state level throughout the 1880s and 1890s. For a more complete summary of the role of Wright in encouraging the collection of this data, see Carter, Ransom, and Sutch 1991.

8. Joan Hannon (1978, 1982a , 1982b ) was one of the first to make extensive use of state labor survey data in computerized format in her analysis of ethnic discrimination in Michigan. More recent research efforts include work by Barry Eichengreen and Henry Gemery relying on a survey of Iowa teachers (Eichengreen 1984, 1987; Eichengreen and Gemery 1986); our own work using data from Michigan, Maine, and Kansas surveys (Ransom and Sutch 1986a , 1989b ); and used by Susan Carter, Peter Philips, and Elizabeth Savoca of a survey from California (Carter 1988; Carter and Philips 1990; Carter and Savoca 1990, 1991, 1992).

9. For a fuller discussion of the objectives and data sets to be collected by the Historical Labor Statistics Project , see Ransom 1990 and Carter, Ransom, and Sutch 1991.

10. This point has great significance for evaluating the trends in retirement in the 1870-1940 period evident in figure 11.1, since the fraction of the labor force engaged in agricultural occupations fell steadily from over 50 percent at the outset to less than 25 percent at the end of the period. This shift away from agriculture should have increased the overall proportion of the elderly population that was retired. The fact that such a trend is not evident in the aggregate data implies that retirement among the nonagricultural segment of the population was becoming less common before 1940 (Ransom and Sutch 1988).

11. We also have a survey of 935 farm proprietors from Michigan. Unfortunately, the age of the owner was not asked in this survey, and no use of these data is made in this chapter.

12. There may also have been a bias toward household heads that could have systematically led to the exclusion of older men who were living as dependents in the homes of their grown children.

13. The technique employed in the construction of these curves is explained in the appendix.

14. Data for the census of 1900 was from a public use sample; see Ransom and Sutch 1988. As figure 11.1 indicates, there was a relative stability in the labor force participation of older men during the last quarter of the nineteenth century. This is also true of adult mortality experience (Ransom and Sutch 1989a : 5). Thus we believe it is reasonable to extrapolate backward the age distribution of male labor force participation observed in 1900 to the time of these worker surveys.

15. Teaching may not be as different as usually supposed. Susan Carter and Elizabeth Savoca (1992) have analyzed the information in the Iowa survey in more detail. They argue that a teacher's lifetime job tenure was surprisingly long.

16. It is interesting to note that the earnings profile for farm laborers exhibits the least pronounced decline of earnings with age.

17. It is worth noting that this survey was completed prior to the impact of the depression of 1893-1894.

18. We have argued elsewhere that a variety of such "strategies" might be employed, ranging from a dependence on the more traditional family based systems of economic support for older family members to what we characterize as a "life cycle" strategy of saving for old age retirement (Ransom and Sutch 1989b ).

19. The methods for smoothing the data on days lost is similar to that for the age and earnings distributions presented earlier. See the appendix for details.

1. The evidence presented here shows that early-twentieth-century old age mortality rates in the United States were biased downward by age misreporting in censuses and vital registration. This was well known at the time, and actuaries regularly used approximations to close their life tables. Old age mortality in official U.S. life tables was derived by a combination of smoothing, extrapolation, and graduation techniques. For example, the life tables for 1890, 1901, and 1910 arbitrarily assume that q115 = 1 (Glover 1921: 348-350). Thus the rectangularization of survival curves discussed by Fries is primarily an artifact of the computational methods used to construct life tables and may not reflect any trend in old age mortality.

2. The 1918-1919 influenza epidemic may explain increases in mortality at ages 40 to 60 (Metropolitan Life Insurance Company 1929a ). The epidemic removed from the population a group of high-risk individuals who would otherwise have died in later years. This selective mortality reduced death rates immediately after the epidemic, and the rising rates in the 1920s could be seen as a return to the earlier trend. This hypothesis fails to explain rising mortality rates at older ages, however. The impact of the epidemic was concentrated in the early adult ages, and death rates above age 55 did not increase during these years.

3. Assume that the same pattern of age misreporting was present in the census as in the vital registration, namely, that the probability that a person age x would be reported as age y was the same. Then the reported age-specific death rates are really weighted averages of the form

---

where Oj is the reported death rate at age j, Mi is the true death rate at age i, Pij is the proportion of those reported at age j who are really age i.

The first term represents persons who are younger than age j but are erroneously reported to be age j>where O_j is the reported death rate at age j, M_i is the true death rate at age i, P_ij is the proportion of those reported at age j who are really age i. The first term represents persons who are younger than age j but are erroneously reported to be age j. The second term, p_jj M_j , represents those whose age is correctly reported at age j. The last term in the equation captures those who are incorrectly reported at age j from ages older than j. The reported death rate at age j can change either because the proportions who are misreported from other ages have changed (P_ij ) or because the levels of mortality at different ages have changed (M_i ).

4. The numbers of retirements, deaths, and pensioners in the Pennsylvania Railroad retirement fund on December 31 in selected years, 1900 to 1936, are as follows:

5. Since we do not have the records in which dates of birth were originally reported, it is not possible to test these data for age heaping. However, it was possible to test for heaping on certain calendar years, such as those ending in 0 or 5. Ratios were computed comparing the number of pensioners at age 71 in each year to a five-year moving average. These ratios were small and did not show identifiable five- or ten-year cycles.

6. Railroad occupations were highly segregated by race. Railroad porters, the major occupation filled by blacks, were mostly employed by the Pullman Company rather than the various railroads themselves.

7. The methods used to produce these estimates are described in Douglas C. Ewbank 1989. I am grateful to Dr. Ewbank for providing me with these special tabulations.

8. An additional adjustment was made to data for 1933 to compensate for the withdrawal of 125 pensioners of the West Jersey and Seashore Railroad, who were withdrawn from the pension fund when the railroad was leased to the Pennsylvania-Reading Seashore Lines. It was assumed that West Jersey and Seashore pensioners had the same age distribution as others in the pension fund at that time and that they suffered the same mortality during the six months that they were included in the fund in 1933. Since the numbers involved are very small, this adjustment has almost no effect on estimated mortality rates.

9. More formally, the number of person years by age at last birthday is a quantity analogous to the stationary population in the life table, L_x . If we assume that deaths are distributed evenly in each year of life, then

and

10. As explained above, the age-specific mortality rates were adjusted to exact ages by averaging adjacent estimates for ages at last birthday. Similarly, the weights used in the logit regression models were derived by averaging the lives at risk for the ages from which the mortality rates were derived.

11. The Gompertz model approximates mortality by the formula

in which x is age and A and B are estimated coefficients. In contrast, the logit model assumes

Many years of actuarial practice have shown that the Gompertz model fits most life tables well over limited ranges of adult ages. There is a controversy, however, about the shape of the mortality curve at the highest ages (Bayo 1972; Bayo and Faber 1983; Kestenbaum 1989; Pakin and Hrisanov 1984). Some estimates suggest that death rates above age 90 increase at a slower rate than the Gompertz curve. The logit model is mathematically similar to the Gompertz model, but it increases less rapidly at higher ages. W. Perks (1932) argues that the simple logit model is superior to the Gompertz model, but he prefers a more complicated four parameter model derived by taking the logit of the logit.

12. The survey obtained records on only 621 female pensioners (Federal Coordinator of Transportation 1940: 20).

13. Preston (1976) shows that the rate of decline of infectious diseases is similar at all ages in an international collection of mortality tables by cause. Comparable statistics are not available for the United States because of changes in the death registration area. However, a similar conclusion can be drawn from the experience of industrial insurance policyholders described by Dublin and Lotka (1937).

14. Wiehl (1930), however, found greater increases in male mortality than in female mortality when she compared 1926 to 1921.

15. This subject needs more investigation before a firm conclusion can be drawn, because the American diet in the nineteenth century was not clearly healthy by present standards. Atack and Bateman (1987: 209) write, "Much of the meat was consumed as fat salt pork; yet despite its unpalatability by modern standards, meat consumption in the 1850s was close to record high levels. . . . The diet was heavy on fats and salt but light on fluid milk, fresh fruit, and green vegetables. Indeed, vegetables were often referred to as 'sauce,' suggesting a role limited to providing flavor or as a garnish, rather than as foodstuffs in their own right."

1. An example of this can be found in Kumagai's (1987: 225) characterization of the rapid increase in life expectancy (from the mid-40s to around 80) in Japan over the past half century as a "longevity revolution." Since the bulk of the increase simply involves the increasing proportion of babies and small children who survive, it may be misleading to portray this as a revolution in "longevity."

2. These estimates are not universally accepted, and prehistorical demography remains a poorly developed field.

3. See also Preston 1970.

4. Although changes in the age structure produced by other forces could also have played a role.

5. These figures are for 1980.

6. According to figures from the United Nations, the proportion of elderly in Third World countries dropped from 7.1 percent to 6 percent over the period 1950-1975 (Okojie 1988: 3). African census figures regarding older ages must be read with some caution, due to problems of age misreporting.

7. Further comparative data of this sort may be found in Hoover and Siegel 1986: 22.

8. The way in which some proponents of this view present their data may be misleading, however. Carole Haber (1983: 11), for example, quoting Fischer's classic study of aging in American history, provides a table entitled "Demographic Life Cycle in America, 1650-1950," which at half-century intervals shows for men and women a series of average ages for marriage, last birth, and so on. For example, the table shows an average age at death for women in 1700 of 52, when the woman's last child is born at age 42. This gives the impression that the average mother would die when her youngest child was only age 10. Of course, the life expectancy of a woman still alive at age 42 would be well beyond age 52, leading to significantly different conclusions about the life course in the American past.

9. For a statement and defense of modernization theory and its implications for aging, see Cowgill 1986: 181-200.

10. For a critical view of the usage of the modernization concept in historical family research, see Laslett 1987.

11. South and East Asian societies also have the virtue for my purposes of being the locus classicus of Hajnal's joint household systems, which he contrasts with the simple household systems of northwest Europe (Hajnal 1983).

12. The continued importance that expectations of reliance on children for support in old age have on continued high fertility in much of the Third World is documented in Bulatao 1979.

13. This was based on a random sample of residents of metropolitan Delhi, aged over 60, conducted in the mid-1970s. The heavy patrilocal bias is evident in a study of a Bangladeshi village, conducted at the same time, that found 62 percent of those over age 60 living with a married son, another 16 percent with an unmarried son over age 15, and only 2 percent with a married daughter (Cain 1981: 378).

14. The joint family household, in which all sons live together with their parents, remains the cultural ideal in this community.

15. The proportion dipped from 82 percent in 1960 to 65 percent in 1984.

16. The proportion of those living alone rose steadily from 12 percent among those aged 60-64 to 39 percent for those over age 80.

17. From the publication of the now classic volume, Household and family in past time , Laslett and his colleagues at the Cambridge Group have noted the diversity of household systems found in Europe, especially through attention paid to Serbia and, later, Russia. However, this aspect of their message tended to get lost in the enthusiasm of many scholars for the finding that European peasantry and nuclear family co-residence were not incompatible.

18. This pattern extended far beyond Italy, into southern and central France and Iberia as well. Bourdelais (1986), for example, shows in his study of the village of Prayssas in Agen in the nineteenth century that the multiple and extended family system ensured that older people would live in households with their children until their death.

19. Richard Smith (1984: 419) has commented on the remarkable stability of the proportion of English elderly living with married children from preindustrial times into the twentieth century.

20. See also Smith 1984.

21. Of course, among the reasons for this increasing sex difference in remarriage rates at older ages is the sex ratio in the American population at these ages, for there are many more available women than men.

22. Life expectancy of males at birth is actually higher in a number of South Asian countries today than is female life expectancy, and even at age 65 men in Bangladesh and Pakistan can expect to live longer than women (Martin 1990: 96).

23. But see Laslett's (1988) discussion of "nuclear hardship," an important stimulus to the proposal I make in these paragraphs.

24. A similar northwestern European example has recently been provided for the Belgian city of Vervier in the nineteenth century, showing that older widows and widowers tended to live with a married child, especially when no unmarried children were available (Alter, Cliggett, and Urbiel 1992).

25. A good example of a neolocal system in southern Europe is provided by Da Molin's (1990a , 1990b ) study of a series of communities in southern Italy from the seventeenth through the nineteenth century. It also provides a good example of my thesis, for while she describes these communities as demonstrating the "nuclear structure of the southern Italian family" (1990a : 524), she also tells us that "a widow who was needy was never abandoned by her family . . . but rather taken into their homes and assisted" (ibid.: 52).

24. A similar northwestern European example has recently been provided for the Belgian city of Vervier in the nineteenth century, showing that older widows and widowers tended to live with a married child, especially when no unmarried children were available (Alter, Cliggett, and Urbiel 1992).

25. A good example of a neolocal system in southern Europe is provided by Da Molin's (1990a , 1990b ) study of a series of communities in southern Italy from the seventeenth through the nineteenth century. It also provides a good example of my thesis, for while she describes these communities as demonstrating the "nuclear structure of the southern Italian family" (1990a : 524), she also tells us that "a widow who was needy was never abandoned by her family . . . but rather taken into their homes and assisted" (ibid.: 52).

26. The demographic constraint debate, first raised in this context by the influential work of Levy (1965), has led to a series of attempts to disentangle the impact of demographic forces on household composition patterns through the use of microsimulation, the latest of which is found by Hammel in this book. On the role of demographic constraint in limiting the frequency of complex family households, see Kertzer 1989.