Hostname: page-component-8448b6f56d-mp689 Total loading time: 0 Render date: 2024-04-23T08:14:19.772Z Has data issue: false hasContentIssue false

Tombstones and Roman Family Relations in the Principate: Civilians, Soldiers and Slaves*

Published online by Cambridge University Press:  24 September 2012

Richard P. Saller
Swarthmore College
Brent D. Shaw
University of Lethbridge


Tombstones furnish perhaps three-quarters of the entire corpus of Latin inscriptions. Many of these give no more than the name of the deceased, but tens of thousands also offer the historian a few additional details, such as age at death and the name and relationship of the commemorator. Previous studies of the tombstones en masse have focused on nomenclature and age at death. In this study we wish to ask what conclusions can be drawn from the data about the commemorator's relationship with the deceased.

Research Article
Copyright © Richard P. Saller and Brent D. Shaw 1984. Exclusive Licence to Publish: The Society for the Promotion of Roman Studies

Access options

Get access to the full version of this content by using one of the access options below. (Log in options will check for institutional or personal access. Content may require purchase if you do not have access.)


1 The whole corpus of Latin inscriptions must now number c. 250,000 or more. MacMullen, R., ‘The Epigraphic Habit in the Roman Empire’, AJPh 103 (1982), 233–46, at p. 238Google Scholar, estimates 156,000 in CIL, but this number has certainly increased greatly. In two test cases (Britain: CIL VII (1873) compared with RIB and subsequent publications; Africa: CIL viii. 1–2 (1881–94) compared to ILAlg I, II. 1–2) we found that the number had about doubled in size. If so, funerary stones probably represent about 170–190,000 of the total.

2 See n. 7 below.

3 Mitterauer, M. and Sieder, R., The European Family (1982)Google ScholarPubMed, ch. 2, ‘The Myth of the Large Pre-Industrial Family’. R. Wall provides a subtle discussion of the difficulties of identifying appropriate criteria (e.g. household membership or participation in family production or consumption) for analyses of family type and points out that Le Play was not very precise on this issue (Wall, R. et al. , Family Forms in Historic Europe (1983), 163CrossRefGoogle Scholar).

4 Laslett, P. and Wall, R. (eds.), Household and Family in Past Time (1972)CrossRefGoogle Scholar. The various criticisms of Laslett's approach are summarized in Anderson, M., Approaches to the History of the Western Family, 1500–1914 (1980), 2738CrossRefGoogle Scholar. Despite the criticisms, Mitterauer and Sieder still conclude that ‘it cannot be maintained … that the dominant family form of pre-industrial times was the large family community in which several generations lived together’ (op. cit., 39).

5 Sailer, R., ‘Familia, domus and the Roman Conception of the Family’, Phoenix, forthcomingGoogle Scholar.

6 See p. 139 below.

7 Weaver, P. R. C., Familia Caesaris: A Social History of the Emperor's Freedmen and Slaves (1972)CrossRefGoogle Scholar; Rawson, B., ‘Family Life among the Lower Classes at Rome in the First Two Centuries of the Empire’, CPh 61 (1966), 7183Google Scholar; their approach has also been used by Treggiari, S., ‘Family life among the staff of the Volusii’, TAPA 105 (1975), 393401Google Scholar, and by Flory, M. B., ‘Family in Familia. Kinship and Community in Slavery’, AJAH 3 (1978), 7895Google Scholar.

8 The figure of c. 25,000 is a rough estimate of the total number of stones read for this survey; of these, a large number were not useful for tabulation because they were too fragmentary, illegible, or ambiguous in meaning. Hence only some 12–13,000 stones actually yielded data tabulated on our charts (A–D and 1–32).

9 Le droit des tombeaux romains (1963), chs. 6–8. Hopkins, K., Death and Renewal (1983), 205 fCrossRefGoogle Scholar. has come to a similar conclusion. The formal distinction is expressed by Gaius in Dig. II. 7. 5: ‘Familiaria sepulchra dicuntur, quae quis sibi familiaeque suae constituit, hereditaria autem, quae quis sibi heredibusque suis constituit’; cf. 11. 7. 6. pr. (Ulpian), CJ 3. 44. 4, 8, 13 (esp. the last, on the confusion and merging of the two types). Even in the Roman columbaria individualized commemoration is found, as for example in those of the monumentum Liviae (CIL vi, 3926–4307), the monumentum familiae Marcellae (ib., 4418–4708), the monumentum inter Appiam et Latinam (ib., 4881–5075), and the monumentum familiae Neronis Drusii (ib., 4327–4413). For the development of this type of tomb see J. Toynbee, M. C., Death and Burial in the Roman World (1971), 113–18Google Scholar and plates 27–8 (columbaria), and 132–43 (house-tombs); for the developments at Ostia see Meiggs, R., Roman Ostia, 2nd ed. (1973), 459–61Google Scholar.

10 Of all Noricum tombstones (N = c. 500) about 360 were clearly specified as ‘se vivo’ or ‘sibi’ types; another c. 100 not in these categories were deemed to be equivalent to them because of their context (i.e. no explicit sign of a deceased), as were ante-mortem commemorations.

11 Dig. 11. 7. 3–5 (Ulpian), cf. 11. 7. 14. 2, 6–7 (Ulpian).

12 Dig. 11. 7. 4 (Ulpian).

13 Dig. 11. 7. 14. 8 (Ulpian); on filial duty over-riding heirship see Dig. 11. 7. 14. 13 (Ulpian).

14 Dig. 11. 7. 12. 4 (Ulpian).

15 For a discussion of the qualities of affection attributed to the deceased and commemorators on Spanish tombstones, see Curchin, L. A., ‘Familial epithets in the epigraphy of Roman Spain’, in Mélanges Étienne Gareau (1982), 170–82Google Scholar. On the more general question of grief and emotional ties to the deceased, see Hopkins, op. cit. (n. 9), 217 ff.

16 Dig. 11. 7. 2. 6 (Ulpian); cf. 11. 7. 42 (Florentius, Institutes).

17 Petronius, Satyr. 71, ‘Horologium in medio, ut quisquis horas inspiciet, velit nolit, nomen meum legat. Inscriptio quoque vide diligenter si haec satis idonea tibi videtur …’ A real example of similar behaviour is attested by CIL xiii. 2. 1, 5708 (Germania superior, Andemantunnum, later the civitas Lingonum, Langres); for comment see Hatt, J. J., La tombe gallo-romaine (1951), 65 ffGoogle Scholar.

18 George, M. Dorothy, London Life in the Eighteenth Century (1925; reprint, 1976), 294 and 391Google Scholar, n. 94, quoting a London advertisement: one shilling entrance fee plus two pence a week, age restriction 14–60, with a full description of the funeral assured to the subscriber. Such ‘Friendly Societies’, as they were called, like Roman collegia, were an urban social phenomenon not limited to funeral concerns alone.

19 Hopkins, op. cit. (n. 9), 205 ff. discusses funerary collegia and mass graves. For collegia see also Waltzing, J. P., Éitude historique sur les corporations professionnelles chez les Romains (4 vols., 1895; reprint, 1970) 1, 141–53 and 256300Google Scholar; on their extension to the provinces see J. J. Hatt, op. cit. (n. 17), 77–84; for examples of their constitutions and practices see CIL VI, 10251–10423, and Meiggs, op. cit. (n. 9), 334, citing ILS 7212 (Lanuvium) and CIL xiv. 1, 4548 (Ostia). For inscriptions in columbaria see n. 9 above; for the burials at Isola Sacra see Calza, G., La necropoli del Porto di Roma nell'Isola Sacra (1940), 46, figsGoogle Scholar. 10 and 80; cf. Toynbee, op. cit. (n. 9), 82–7, 101–3, and plates 18–19 for Isola Sacra, and 22–5 for poor burial types elsewhere in the empire. As in life, the ‘houses of death’ of the poor are cramped in the spaces between the more magnificent homes of the rich. For the burial of the poor in other early modern societies see Ariès, P., The Hour of Our Death (1983), 56 f., 207 f., and 270 fGoogle Scholar. (and cf. 185 f. on ‘confraternities’).

20 For the Rome gladiators see CIL VI. 2, 10168–10202; for the Spanish gladiators see Bellido, A. Garciay, ‘Lapidas funerarias de gladiatores de Hispania’, AEA 33 (1960), 123–44Google Scholar (= AE 1962: 44–58). On family life amongst gladiators see Ville, G., La gladiature en Occident des origines à la mort de Domitien (1981), ch. 4. 5, 329–32CrossRefGoogle Scholar. He believes that most of these ‘marriages’ were ephemeral affairs with gladiatorial groupies (ludiae), but the consistent mention of husbands and wives using the formal terminology of marriage (i.e. uxor, maritus, coniunx) and of children shows that at very least the forms of nuclear family relationships were observed.

21 The price attested for CIL viii, 18162 = 3042 is 96 HS. For a survey of costs see Duncan-Jones, R., The Economy of the Roman Empire: Quantitative Studies, rev. ed. (1982), 99101Google Scholar, nos. 213–44; most are from Lambaesis (cf. p. 70: there are 49 prices from the city, more than from any other African centre; two-thirds of them refer to tomb costs). Since most of these prices are expressed as expenditures ex n. sesterces, it is not certain that the whole amount must refer to the cost of the monument alone; in one case (CIL viii, 3079) the funeral is mentioned as a separate item of cost, though its price is not noted separately. The claim that a 2,000 HS payment recorded in the charter of the collegium of the cornicines at Lambaesis (ILS 2354) is for burial costs seems unfounded. It seems to be merely the sum of money that would otherwise have been paid to the member on leaving the society because of promotion or retirement from army service; on his death the amount is to be paid to his heir, ‘Item si qui obitum naturae red(diderit), acc(ipiet) her(es) ips(ius) sive proc(urator) D’. On the variables of decoration and form that affect cost see Weynand, R., ‘Form und Dekoration der römischen Grabsteine der Rheinlande im ersten Jahrhundert’, BJ 1089 (1902), 185238Google Scholar.

22 For the size and description of the Cirtan stones see ILAlg. 1. 2, 845 f.; for Castellum Celtianum, ib., 2117f.; for Thubursicu Numidarum, ILAlg. I. I, 1337 f.; for those of Thugga see the periodic reports on them by L. Poinssot in CT 16 (1909) and 17 (1910). On the average the latter were 0·50 × 0·20 m. and the three lines or so of text include only DMS plus the name and age of the deceased.

23 Compare the burial costs noted for Africa in Duncan-Jones, op. cit. (n. 21) with those for Italy, pp. 161–71, nos. 550–636; for costs related to income see Table 2, p. 79 (for Africa) and Table 3, p. 130 (for Italy). We can say, for example, that the more expensive tombs erected by soldiers represented a cost of about 2% of their nominal annual income. For allowances made by funerary collegia in Italy see Duncan-Jones, op. cit., 131. Burial costs for poor unmarked graves are probably best reflected in the municipal laws governing costs of burials and execution (and other punishments) set for public libitinarii at Puteoli and Cumae, indicating a total cost of under 50–60 HS; see Bove, L., ‘Due nuove iscrizioni di Pozzuoli e Cuma’, RAAN, N.s. 41 (1966), 207–39Google Scholar, cf. Labeo 13 (1967), 22–48 (= AE 1971: 88–9). For the funeraticium of Nerva see Chron. Min. (ed. Mommsen) 1. 146, and A. Degrassi, ‘Nerva Funeraticium Plebi Urbanae Instituit’, ch. 52 in Scritti Vari di Antichità, vol. I (1962), 697–702. Pliny, Pan. 40. 1 mentions the possibility of spending the whole of a parva hereditas on a sepulchrum for the deceased.

24 In the rural regions influenced by Theveste the proportion of tombstones with named commemorators is quite high (67%), while at Cirta, an administrative centre not heavily influenced by the army, only 13% of commemorators mention a relationship with the deceased and most of those do so without giving their own names (12%).

25 See Clauss, M., ‘Probleme der Lebensalterstatistiken aufgrund römischer Grabinschriften’, Chiron 3 (1973), 395417, at 404–5Google Scholar, and Table VII, a phenomenon already noted by Kajanto, I., ‘On the Problem of the Average Duration of Life in the Roman Empire’, Annales Academiae Scientiarum Fennicae, ser. B, no. 153. 2 (1968), 1213Google Scholar.

26 Clauss, art. cit., Table X, pp. 415–17 for the towns concerned; those with low commemoration of under Ios have about 1–2% average attested servile population, whereas those with relatively high infant-child commemoration average about 10% or more servile population. The urban size differential is also evident, the former being towns of the 2–4,000 range, the latter larger cities of the 10,000 plus range.

27 Infant and child mortality for e0 = 25 would be about 530 out of 1,000 live births by age ten, see Coale, A. J. and Demeny, P., Regional Model Life Tables and Stable Populations (1966)Google Scholar, South Level 3. There is also a clear conflict with actual burials. At the Roman town of Sitifis (Sétif) in North Africa, for example, where funerary epitaphs yield a fairly normal distribution according to our ‘civilian’ pattern, about 40% of all burials (N = 228) were those of children one year of age and under (N = 88), and 62% were of children ten years of age and under (N = 141); see Février, P. A. and Guery, R., ‘Les rites funéraires de la nécropole orientale de Sétif’, Ant. Afr. 15 (1980), 91124CrossRefGoogle Scholar. The same problem of the discrepancy between actual burials and information on commemorations was noted by L. Henry for modern cemeteries: ‘La mortalité d'après les inscriptions funéraires’, Population 12 (1957), 149–52, and ‘L'âge du décès d'après les inscriptions funéraires’, ib. 14 (1959), 327–9.

28 On this skewing see Hopkins, K., ‘On the Probable Age Structure of the Roman Population’, Population Studies 20 (1966), 245–64, at pp. 251 fCrossRefGoogle ScholarPubMed. At least part of this skewing in some of the populations we have studied can be explained as the result of cultural patterns in the practice of commemoration that exist prior to the act of commemoration itself. The cultural pattern of heavy emphasis on parents' celebration of children, for example, that is evident in some of our samples, will necessarily skew the average age at death downwards considerably, irrespective of the actual demographic make-up of that whole population.

29 See n. 37 below.

30 There are some groups of dated tombstones as, for example, those from Sitifis (dated by the provincial era, most to the first half of the third century); others, such as the military stones from Britain, can be dated by the known movements of military units (see Birley, A. R., The People of Roman Britain (1979), chs. 3–5Google Scholar). Otherwise, one is left with changes in formulaic expressions to indicate a very approximate date, a method which does not as yet command great confidence, see Lassere, J.-M., ‘Recherches sur la chronologie des épitaphes païennes de l'Africa’, Ant. Afr. 7 (1973), 7152CrossRefGoogle Scholar; for the general temporal distribution of all dated stones in the empire see MacMullen, art. cit. (n. 1), graph p. 243 (derived from Mrozek).

31 For the inscriptions from Cirta, Castellum Celtianum, Regio XI in Italy, and Rome: Servile the first thousand stones were surveyed; for Rome: Lower Orders every fiftieth stone out of the c. 20,000 available in CIL vi was used.

32 It should be emphasized that, as a result, the tables do not give the number of tombstones examined.

33 CIL v. 2, 6091. A survey of the typical midrange Lambaesis sample reveals that about 80% of all stones contain 1–1 or 2–1 relationships (they represent an even higher proportion (95%) of all those in which relationships are actually expressed). Commemorations involving a group of more than three were rare among Lambaesis epitaphs: 4% involved four; less than 1% involved five; and less than 0·5% involved six.

34 Certain formulaic expressions of a cumulative and non-specific nature were also excluded from the count of personal relationships. For example, the formulaic expression ‘libertis libertabusque’ found most frequently on stones from Rome (N = 122 for our Familia Caesaris sample) and Ostia (N = 91) in our samples, was not included in our survey because it is not a specific expression of the commemoration of known individuals. For a different type of analysis of funerary groups in Greece see Humphreys, S. C., ‘Family Tombs and Tomb Cult in Ancient Athens: Tradition or Traditionalism?’, JHS 100 (1980), 96126CrossRefGoogle Scholar = ch. 5 in The Family, Women and Death (1983). 79–130.

35 Occasionally wives and children were also labelled ‘heredes’, in which cases they were counted only in the appropriate kinship category. Our assumption that commemorators designated solely as ‘heres’ were unrelated was often corroborated by differences in nomenclature or by the additional label of ‘amicus’.

36 Because we were seeking to measure the extent of attachment of servi and liberti to free populations, large or autonomous servile groups such as the Familia Caesaris were excluded from our civilian and military samples and were counted separately. There were, of course, a few ambiguous cases of servile (and other) relationships. Where there was any question or doubt as to the correct placing of such inscriptions, they were excluded from our count. Such cases were never numerous enough to make any significant difference to the ratios of personal relationships in the tables.

37 See Shaw, B. D., ‘Latin Funerary Epigraphy and Family Relations in the Later Empire’, Historia 33 (1984)Google Scholar.

38 Speidel, M., Die Equites Singulares Augusti (1965), 1–2, 1621Google Scholar.

39 The large proportion of memorials to wives set up by husbands is partly to be explained by the legal rules which place the financial responsibility on the person receiving the woman's dowry. Where no dos is left or it is inadequate to cover funeral expenses or an inheritance is also left, the woman's father or heirs may be held responsible (Dig. 11. 7. 16, 20, 22, 28). In such cases, as in others, the law did not require husbands to pay for inscriptions perpetuating the memory of their relationship: the fact that they did so is a sign of affection and sense of duty.

40 CIL vi, 3194, 3267, 3282, 3288, 3300.

41 La famille et l'amour sous le Haut-Empire remain’, Annales, E.S.C. 33. 1 (1978), 3563Google Scholar. The impact of the article is clear in the review of Goody, J., The Development of the Family and Marriage in Europe (1983)CrossRefGoogle Scholar by the eminent mediaeval historian, Georges Duby (TLS, Oct. 14, 1983, 1107).

42 ibid., 35. Stone, L., The Family, Sex and Marriage in England, 1500–1800 (1977)Google Scholar, argues for the increasing discreteness of the nuclear family, with a growth in the intensity of emotional bonds within that unit. For a summary of other recent work on this subject, together with a critique of the approach, see Anderson, op. cit. (n. 4), 39–64.

43 Veyne, art. cit. (n. 41), 36.

44 ibid., 40.

45 ibid., 60.

46 De rerum natura 3. 894 ff. In this passage the meaning of domus would appear to be ‘household’ including the immediate family (see Sailer, art. cit. (n. 5)). There seems little point in adding all the well-known evidence from Cicero's letters about his concern for Terentia, his children and his brother during his exile, or about his grief at Tullia's death, or about his efforts to keep together his brother's marriage and to discipline his nephew.

47 Though the arenas changed, the competition for offices, honours and wealth obviously continued. One of the clearest public indications of this continuity was the morning salutation—an open sign of social subordination of the client and of the power of the patron to provide goods and services, in particular the ability to dominate adversaries in the law courts (Tacitus, Dialogus, especially 6 and 11 where Maternus explains his withdrawal from public life on the grounds that he does not want to participate in the competition any longer). See Saller, R. P., Personal Patronage under the Early Empire (1982)CrossRefGoogle Scholar, chs. 3–4 for the fallacy of the ‘service aristocracy’ and art. cit. (n. 5) for the importance of the domus frequentata. We do not deny the possibility of developments in the institution of the family (e.g. toward an increasing recognition of cognates in law), but these were already largely in place by the late Republic when our evidence becomes adequate to discuss changes with any confidence. See Thomas, Y., ‘Mariages endogamiques à Rome. Patrimoine, pouvoir et parenté depuis l'epoque archaïque’, RD 58 (1980), 362 ff.Google Scholar, where he is forced to rely on legends from the regal period in his analysis of incest taboos. For a brief statement of the importance of the nuclear family similar to ours see Treggiari, S., ‘Libertine ladies’, CW 64 (1971), 198Google Scholar.

48 With the exception of the Noricum sample, extended family members (excluding alumni) never supply more than about one-third of those outside the nuclear family, and for most civilian populations the proportion is under one-fourth. The Noricum sample is extraordinary in its proportion from within the nuclear family (91%). and it is because of this that the extended family, though not a large proportion of all relationships (7%), provides the bulk of the commemorators outside the nuclear family. These features are to be explained by the prevalence of extended ante-mortem commemorations (i.e. ‘to myself, to my wife, to my sons and daughters’). This practice meant that hardly anyone in Noricum found him/herself in the position that a significant fraction in other areas did—of dying with no immediate family left alive to commemorate.

49 By our calculation on the basis of the life table cited above, n, 27, something like 12% of children at age 10 would have had a paternal grandfather alive and perhaps 5% at age 16. Since a large proportion of all dedications were set up to young children (one-third or more in some samples), we would expect paternal grandfathers to represent 3–4% of commemorators, if they had regularly participated when alive, rather than the 0–0·7% that we actually find. Patrui, whom we might also expect to find in commemorations if the extended family had been important, are no more common than paternal avi (a total of four scattered in the samples for the Roman lower orders, Latium, Regio XI, Narbonensis and Lambaesis, comprising 2,022 relationships). Some perspective on these figures can be gained by comparison with a society in which the extended family was normal: in Russia of the last century 66% of all households contained three or more generations (Mitterauer and Sieder, op. cit. (n. 3), 29).

50 In the Regio XI sample there are three paternal avi as compared with three maternal avi; two paternal avi as compared with six grandmothers and one maternal avus in Lambaesis; one paternal avus and four aviae in Latium; no paternal avus and four maternal grandparents in Narbonensis; no paternal avus and two maternal grandparent couples in the Roman lower orders.

51 Weaver, op. cit. (n. 7), 95, correctly noted the centrality of the nuclear family in Roman society. De Visscher, op. cit. (n. 9), 118, pointed out that most funerary dedications from the imperial period include only a very narrow circle of family.

52 De officiis 1. 58. Cicero's hierarchy has a pseudohistorical element, as well as a moral element in it.

53 Cicero, Ad Att. 12. 32. 2 (also 12. 7 concerning the practice of giving sons living allowances). In Pro Caelio 18 Cicero indicates that Caelius' separate household had been brought up by his accusers as a criticism, but he argues that at Caelius' age such behaviour was hardly reprehensible. It may be that this passage reveals the tension in the late Republic between the old values associated with the patriarchal household and the new values which accommodated the practical advantages of separate households.

54 Pro Sex. Roscio 43.

55 Seneca, , Cons. ad Marciam 24. 1Google Scholar; Pliny, Ep. 9. 9. 2 (in which Pliny's praise for the recently deceased Pompeius Quintianus as optimus filius, because he continued to live with his difficult father, would have little force if most adult sons did the same). Crook, J., ‘Patria Potestas’, CQ, N.S. 17 (1967), 119CrossRefGoogle Scholar, discusses the problem of reconciling the custom of father and adult son having separate households with the legal institution of patria potestas.

56 In the Roman lower orders sample 85 dedications with relationships (about one-third) give age at death: of the 85, 39% are under 10 years old, 21% are 11–20, 25% are 21–30, and 15% are older. In the Gallia Narbonensis sample 62 dedications with relationships (about one-sixth) give age at death: of the 62, 23% are under 10, 44% are 11–20, 23% are 21–30, and 10% are older. We do not suggest that these proportions hold for the majority of the dedications in which no ages are given, and consequently they are of very limited value.

57 Anderson, op. cit. (n. 4), 78.

58 See above, p. 126.

59 Strabo 3. 4. 18; see Baroja, J. Caro, ‘Organización social de los pueblos del norte de la península ibéirica en la antigüedad’, in Legio VII Gemina: Coloquio international, León, 16 al 21 de septiembre de 1968, ed. González, A. Viñayo (1970), 2630Google Scholar. For the difficulties of such structuralist interpretations of the ‘savage’ and the ‘civilized’ in local Mediterranean societies, see Pembroke, S., ‘Women in charge: the function of alternatives in early Greek tradition and the ancient idea of matriarchy’, Journal of the Warburg and Courtauld Institutes 30 (1967), 135CrossRefGoogle Scholar.

60 Tranoy, A., La Galice romaine: recherches sur le nord-ouest de la péninsule ibérique dans l'antiquité (1981), 106–7Google Scholar.

61 On their position in work see Strabo 3. 4. 17; 3. 2. 9; Sil. Ital., Pun. 3. 348–53; Justin 44. 3. 7; cf. Caro Baroja, op. cit. (n. 59), 28. For the practice of couvade (Spanish covada) see Caro Baroja, op. cit., 27; Tranoy, op. cit., 107, objects that women at work are not necessarily women in power (backed up with many examples of male dominance in Spanish society), but he misses the critical point that women involved in field work were exceptional in the Mediterranean (see J. Goody, op. cit. (n. 41), 30).

62 Carcopino, J., Daily Life in Ancient Rome, ed. Rowell, H. T. and transl. Lorimer, E. O. (1956), 96Google Scholar, laments the degeneration of the Roman family, as evidenced by the thousands of petite bourgeoisie epitaphs ‘where the deceased is mourned by hism freedmen without mention of children’. Our tables demonstrate that in every area this type of dedication constituted only a small fraction of all commemorations. Carcopino's statement shows the danger of making sweeping generalizations on the basis of an impressionistic study of tombstones.

63 We counted 244 tombstones for officers and ex-officers, as compared with 272 for milites and retired milites. Since milites outnumbered officers by eight or ten to one, the rate of finds of officers' tombstones is at least seven times greater per capita than of ordinary soldiers'.

64 In most civilian populations ‘brother-brother’ dedications represent 4–5% of all relationships. Although some military populations reveal similar ‘brother-brother’ proportions (notably Britain, 3–7%; Spain, 5–7%), most are higher (the mean is 9·5% of all relationships). In some highly militarized units, such as the equites singulares at Rome and the troops in Germany, the ‘brother-brother’ relationship accounts for almost half of all nuclear family relationships.

65 The figures are based on an analysis of the data in the study of Schober, A., Die römischen Grabsteine von Noricum und Pannonien, Sonderschriften des österreichischen archäologischen Institutes in Wien, Bd. 10 (1923)Google Scholar. For the aniconographic stones see nos. 16 f.; for the military iconography see nos. 153 ff.; cf. the earlier work by Hofmann, H., Römische Militärgrabsteine der Donauländer, Sonderschriften des österreichischen archäologischen Institutes, Bd. 5 (1905)Google Scholar for supplementary materials and illustrations.

66 A sample collected for Raetia showed that it belonged more to the Noricum-Upper/Middle Danube pattern of more ‘civilian’ type relations amongst the soldiers. It therefore represents the eastern boundary of the ‘military’ pattern on the British-Rhine frontier.

67 On the social status and mobility of centurions see Birley, E., ‘Promotions and Transfers in the Roman Army, II: the Centurionate’, Carnuntum Jahrbuch 7 (19631964), 2133Google Scholar, and Dobson, B., ‘The Centurionate and Social Mobility during the Principate’, Recherches sur les structures societies dans l'Antiquité classique (1970), 99116Google Scholar. Centurions and other high ranking officers represent about 20% of the British sample of conjugal relationships; they, together with veterans who settled in Britain (i.e. those who decided to remain in the country), form the great majority of the conjugal dedicators and celebrands.

68 Birley, op. cit. (n. 30), 34–81, is the most comprehensive and fully documented survey.

69 Birley, op. cit., 84–5, citing RIB 156 (Bath), a man described as a Belga, ‘pretty certainly he was from the British Belgae’; Holder, P., The Roman Army in Britain (1982), 47Google Scholar counts the possible pre-Hadrianic case of Pomponius Valens. For auxiliaries see Birley, 104 and Holder, 51; both accept two cases, one of which seems very questionable: Nectovelius, son of Vindex, a man nationis Brigans (RIB 2142), and a man with the cognomen [Satu]rninus from the colony of Glevum (Gloucester) who is probably of immigrant descent (CIL xvi, 130).

70 For a list of the known sources of external recruiting see Birley, op. cit., 84 f., 97–8, and Holder, op. cit., 47–51.

71 Birley, op. cit., 101.

72 Dobson, B. and Mann, J. C., ‘The Roman Army in Britain and Britons in the Roman Army’, Britannia 4 (1973), 201CrossRefGoogle Scholar, repeated at 204–5; echoed by Birley, op. cit., 95, 104–5, and by Holder, op. cit. (n. 69), 49. The thrust in all these arguments (e.g. Birley, 82 f.) is to demonstrate that the general pattern must also be true of the British instance.

73 Dobson and Mann, art. cit., 201.

74 ibid., 203; see Birley, op. cit. (n. 30), 95–6: ‘In the second century local recruitment must have gradually become normal, limited though our direct evidence is … The presumption really must be that men with Gallic, or colourless Latin names, were British, from the second half of the second century onwards, unless the contrary is stated’ (our italics). A dubious method also subscribed to by Holder, op. cit. (n. 69), 48.

75 Birley, op. cit. (n. 30), 104 does express some doubt about his conclusion concerning local recruitment because of the rarity of ‘cases of soldiers' parents or sisters on record’.

76 See Alföldy, G., Die Hilfstruppen der römischen Provinz Germania Inferior (1968), 100 f.Google Scholar, and Kraft, K., Zur Rekrutierung der Alert und Kohorten an Rhein und Donau (1951), 44 f.Google Scholar, arguments summarized by Dobson and Mann, art. cit. (n. 72), 193–5.

77 See Alföldy and Kraft, opp. citt. The troubles of A.D. 68–9 were crucial in ensuring a decisive change in imperial policy toward the north-western frontier. Africa and Spain were less affected by these events and tended, subsequently, to keep more of their auxiliary units locally, see Cheesman, G. L., The Auxilia of the Roman Imperial Army (1914; reprint, 1971), 164fGoogle Scholar. Dalmatia already experienced the effects of a similar policy after the revolt of A.D. 6–9, see Alföldy, op. cit., 88 f. and his Die Auxiliartruppen der Provinz Dalmatien’, AArchHung 14 (1962), 259–96Google Scholar, and indeed, it was probably being instituted along the north-western frontier before A.D. 69, see n. 80 below.

78 See Forni, G., Il reclutamento delle legioni da Augusto a Diodeziano (1953), 216–17Google Scholar (Legio I Minervia, Germania inferior), 225 (Legio VI Victrix, Germania superior, to A.D. 122), 235 (Legio XXII Primigenia, Germania superior), and 227 (Legio VIII Augusta, Germania superior). In all these units external recruiting predominates–some from Noricum and Thrace, but most from the Gauls.

79 Tac., Agr. 15. 3, ‘… nihil iam cupiditati, nihil libidini exceptum. In proelio fortiorem esse qui spoliet: nunc ab ignavis plerumque et imbellibus eripi domos, abstrahi liberos, iniungi dilectus, tam quam mori tantum pro patria nescientibus …’; 31. 1 (Speech of Calgacus), ‘Liberos cuique ac propinquos suos natura carissimos esse voluit: hi per dilectus alibi servituri auferuntur; coniuges sororesque etiam si hostilem libidinem effugerunt, nomine amicorum atque hospitium polluuntur’. In Hist. 4. 14. 1 Tacitus begins by explaining that the dilectus was regarded as a most serious burden ‘by nature’ because of the sexual aggression and greed of Roman recruiting officers; but later Civilis is made to state explicitly that ‘Instare dilectum quo liberi a parentibus, fratres a fratribus velut supremum dividantur’ (14. 4. 3).

80 Kraft, op. cit. (n. 76), 40 sees this as a crucial passage that illuminates our understanding of motives behind the local revolts in the region in A.D. 69; he takes it to refer both to the twenty-five-year period of service, and to the policy of transferring auxiliary units far away from their homelands, a process which he thinks to be well under way before A.D. 69.

81 Mócsy, A., Pannonia and Upper Moesia (1974). 154–8Google Scholar; for Upper Moesia see his Gesellschaft und Romanisation in der romischen Provinz Moesia Superior (1970), 166–75; a glance at Forni, op. cit. (n. 78), 222–3 (Legio IV Flavia) and 225–6 (Legio VII Claudia) reveals that all known recruiting was external for the latter province.

82 For recruiting to the legion in Spain see Hervas, J. M. Roldán, Hispania y el ejército romano. Contribución a la historia social de la España antigua, Acta Salmanticensia, Filosofía y Letras no. 76 (1974), 245–50 and maps 3–6, pp. 349–52Google Scholar (most are Spaniards); for the region of Galicia in which the legion itself was stationed, and for the recruitment of Asturians, see Yanguas, N. Santos, El ejército romano y la romanización de los Astures (1981)Google Scholar, who shows, with Roldan Hervas, 65–158, that most auxiliaries taken from the region were shipped out of the region to the Rhine-Danube frontier; see Forni, op. cit. (n. 78), 226–7 (Legip VII Gemina) 7/24 (c. 30%) of known post-Hadrianic recruits to the legion are from external sources.

83 For Spain see n. 82 above; for the Pannonias see Forni, op. cit. (n. 78), 216–17 (Legio I Adiutrix, 6/21 external), 228–9 (Legio X Gemina, 8/15 external), 230–1 (Legio XIV Gemina, 7/17 external), and 236 (Legio XXX Ulpia Victrix, all external).

84 As can also be seen in the patterns of known marriages as attested on the military diplomata of auxiliary soldiers. First, a large proportion seem to be unmarried (N = 40/78, 52%); secondly, those who have named wives (N = 21/78, 27%) have obviously brought them with them to their posting from their home region, since the wives are specified as being of the same ethnic origin as themselves; and lastly, the latter pattern would seem to be true even of those men with no specified wife, but whose children are noted (N = 17/78, 22%), to judge from their children's nomenclature. For the data see Roxan, M., ‘The distribution of Roman military diplomas’, Epigraphische Studien 12 (1981), 265–86, esp. 276–7Google Scholar, with attention to her caveats about the significance of the temporal distribution of the evidence—i.e. the patterns are significant for the period before A.D. 120; between A.D. 120–140/60 there would seem to be a trend towards more frequent marriage by auxiliary troopers.

85 Mommsen, T., ‘Die Conscriptionsordnung der römischen Kaiserzeit’, Hermes 19 (1884), 1–79, 210–34Google Scholar = ch. 3 in Gesammelte Schriften 2 vi (1910; reprint, 1965), 20–117, esp. 22–30.

86 For the literature, and a complete review of the evidence, see Shaw, B. D., ‘Soldiers and society: the army in Numidia’, Opus 2 (1983), 133–60Google Scholar, esp.

87 That is to say, if there are divergences from what is believed to be the pattern, then some ad hoc explanation is required. A case in point is the divergence from the normal pattern revealed by the publication of an inscription from Egypt detailing the origines of 133 soldiers of Legio II Traiana (AE 1955: 238); the men, discharged in 157, were recruited in the early 130s A.D. Since most of their origins are non-Egyptian, an explanation for the abnormal recruiting pattern was sought in the emergency created by the war in Judaea; see Gilliam, J. F., ‘The Veterans and Praefectus Castrorum of the Legio II Traiana in A.D. 157’, AJPh 77 (1956), 359–75Google Scholar. It is only fair to note that Gilliam stresses both the paucity of the other evidence on recruiting in Egypt and notes Mommsen's principle with the caveat that ‘We should not be too hasty in erecting “principles” on scanty evidence or assume inflexibility on the part of the imperial government; policies may have changed from decade to decade’ (p. 361). The limited point being made here is the use of the principle to measure ‘aberrant’ cases.

88 AE 1971: 534 (Banasa, Mauretania Tingitana) where the domus and familia culminate in the ethnic group, the gens, of the Zegrenses, in a context where the city is no longer the central organizational unit of the society; cf. Tac., Germ. 15 and 25 for the extended family as an integral part of local social structure. Both the Tabula Banasitana and the Germania are cited only as examples of different patterns of social organization that may have dominated in some rural areas of the empire.

89 See, e.g. Ariès, op. cit. (n. 19), a most influential work, some of whose more adventuresome statements are clearly contradicted by our analysis of funerary commemoration and family life in the Roman empire. For example, Ariès claims that the addition of the age at death to the tomb from the fourteenth to sixteenth centuries onwards ‘corresponds to a more statistical conception of human existence in which life is defined more by its length than by its content, a conception which is that of our bureaucratic and technological civilization’ (p. 222), a generalization which would seem to be refuted by the Roman data. And again (p. 230) he sees the rise of funerary commemoration on stone in the late Middle Ages, and its spread downwards to common people, as ‘the expression of a new feeling, the sense of family’. But clearly vast numbers of people over equally vast areas of western Mediterranean Europe had already felt this ‘new feeling’ centuries before the novel development noted by Ariès.