Roger S. Bagnall and Bruce W. Frier, The Demography of Roman Egypt. Cambridge: Cambridge University Press, 1994. Pp. 354. ISBN 0-521-46123-5. $49.95.
Reviewed by Tim Parkin, Victoria University, Wellington, NZ (firstname.lastname@example.org).
"Egyptian affairs," wrote Philo of Alexandria in the first century A.D. (in Flaccum 3), "are intricate and diversified, hardly grasped even by those who have made a business of studying them from their earliest years." Like the subject it studies, this book is a complex one, but in its detailed demographic analysis of the census returns that have survived from Roman Egypt, it repays careful reading. The fine fruit of a joint effort between an expert in papyrology and a well-versed exponent of the intricacies of modern demographic methods, this volume in the prestigious series of Cambridge Studies in Population, Economy and Society in Past Times marks a significant advance in our knowledge of the demography of the era. It is a book to be read closely, not one that can be merely mined for basic answers to such questions as average life expectancy or levels of fertility. The authors aim to be "user-friendly", and at most points their discussion is clear and well-referenced. But a mass of figures, together with such phenomena as the coefficient of determination and Weiss' index of dissimilarity, may daunt the less mathematically minded. This would be an unfortunate effect, as the impact of the book is derived from its logical development of a complex demographic picture. My aim in this review is to react to that picture, in particular in terms of the census data themselves and the use to which they have been put.
For the last forty years papyrologists and social historians alike have relied on the now classic study of the census returns by Hombert and Préaux.1 That volume has stood the test of time well, but will now be less indispensable. Bagnall and Frier have more returns to deal with -- the total is now over 300 -- and it is significant that the increase in data has not markedly altered the basic findings about the operation of the census, or for that matter about the overall nature of the population patterns, derived from Hombert and Préaux's investigation. What marks out this volume above all is the new level of sophistication brought to the analysis of the data.
Having said that, all will find the catalogue of census declarations, which takes up nearly half of the book (pp.179-312), of enormous value, worth the price of the book on its own. Bagnall has looked anew at all the declarations that are extant; elsewhere, in a series of articles, he has published amended readings where applicable.2 Here, rather than simply publish all the texts, the basic information that each return gives us is summarised, in particular the persons declared and their ages. The body of information continues to grow: even after going to press a further four declarations came to light, added at the end of the catalogue but not taken into consideration in the data analysis.
The rest of the book is what will most concern us here. The first two, lengthy chapters furnish important background information. Chapter one outlines the workings of the census (such as they are known to us, from the returns themselves). Most of this is familiar from Hombert and Préaux, but it is worth having again; at any rate, a reassessment of Hombert and Préaux was long overdue. Standard questions are dealt with concisely and effectively, such as on the census process and the form of the returns (pp.11-26). Some new information comes to light, for example about the evidence for the beginning and end of the 14 year cycle.3 The second chapter turns to a consideration of the potential value of the returns as sources of demographic information. Of the 300+ returns, 233 give data of varying quality on household members, and it is on these that the analysis focuses.
Both chapters along the way highlight many of the potential perils in using these returns as sources of demographic information. The sample is minute when the chronological and geographical range is considered. It is well known that the returns are not evenly spread chronologically -- 87.6% of the 233 come from between A.D. 103 and 215. This is due to the nature of the survival of the returns, and the pattern roughly matches the chronological spread of all the dateable papyri we have. Inevitably little allowance for changes over time -- significant as these are in demographic terms -- can be made on the basis of the returns, though no one would argue for a static demographic picture over the course of three centuries.4 It is also noteworthy that the very form of the declarations varies over time, as well as over space (cf., e.g., pp.17, 19, 22-24). Geographically, the unevenness in the spread of the sample is even more striking. Urban centres are vastly overrepresented; the Arsinoite and Oxyrhynchite nomes account for some three-quarters of all the returns; and, it is worth stressing, no returns at all have survived from the capital, Alexandria. There is no way of knowing to what extent such geographical unevenness causes distortion in the data. The message is simple, and not underestimated by the authors in these opening two chapters: the data derived from the census returns are far from perfect (a point one might too easily forget as the demographic analysis proceeds), but they are the best we have got from the time. With such a message, I have no quibble.5 In fact, the census returns that have survived from Roman Egypt offer us perhaps the best demographic data in western history before the fifteenth century A.D.
But further problems with the database are worth highlighting, as they have been in the past.6 Infant mortality rates cannot be measured from the returns (cf. p.34 n.10). Furthermore, young girls often go unreported, as do young males of around the age when they become liable to the poll-tax. The levels of such underreporting, and of misrepresentation of age, appear to vary between villages and metropoleis. It is often forgotten that the returns we have are "private" declarations, not yet checked or corrected by government officials (some discussion of the measures taken to ensure all were declared would not have been out of place in this context; cf., e.g., Gnomon of the Idios Logos 44, 58-63, and PSI 1326). But the ages given in the returns -- 1084 in total -- are notable for not suffering from age-rounding (with some striking exceptions, such as PSI I.53, col. i); of course, precision need not equate with accuracy. Certainly there seems to be marked exaggeration of older adult ages, unless the proverbial fertility of Egypt really did produce a whole spate of mothers in their forties, and one aged 51. At the other end of the spectrum we also find a nine-year-old mother. There is something very wrong with the age-spread of the figures (see table 7.1, p.143). Even modern censuses are far from perfect; it is worth noting in that context that to ensure a good degree of demographic reliability it is recommended by the United Nations that censuses be held at least every ten years.7 A fourteen-year gap in Roman Egypt, determined by the requirements of poll-tax collection, offers less assurance, especially in view of high infant and childhood mortality. Most errors in age statements, whether intentional or otherwise, must now go undetected, but many may be spotted, even within such a small sample (see pp.39-44 for a variety of examples). The crucial question of the degree of social bias in the sample as we have it I shall return to later. The urban bias is fully appreciated by Bagnall and Frier, and one of the significant methodological advances in their analysis is the allowance they make for this bias. The demographic differences between village and metropolis would have been quite substantial. By weighting the results towards villages, which despite the extant census data probably had at least twice the population of the cities,8 a more balanced picture is derived.
In passing, it is worth commenting that this book deserves a much wider audience than just ancient demographers and papyrologists. It holds much to fascinate all with an interest in the ancient world. Incidental information abounds to reward the attentive reader: for example, a link is suggested between age awareness and the practice of astrology (p.46 n.52), though the month, day and hour of one's birth (or rather perhaps one's conception) may have been more important than the year (cf. Pliny, Letters 2.20.3); then again, concern about anni climacterici might have made a difference. The bold assertion is made (p.48 n.58) that adult males who are declared as atechnos, idiotes or argos were not indigent unemployed but rather were men of leisure, living off accumulated wealth. One would like to see this discussed at considerably greater length. Rather more substantially, there is very useful discussion on economic aspects of demography and Egyptian society (e.g. pp.69-70, 73-74), and also much of value on slavery (cf. pp.70-71, 156-59, though the argument is less compelling on the slave sex ratio and the apparent difference between village and metropolis, especially when one notes that out of a total of 118 slaves known from the returns, 45 are of uncertain age -- 32 from metropoleis, 13 from villages).
To return to the main thrust of the demographic material: chapters 3 to 8 contain the demographic analysis of the data, in a logically developed argument which I shall not attempt to relate in any detail here. Starting in chapter 3 with probable population figures for Egypt9 and with discussion of the urban/rural ratio, Bagnall and Frier go on to consider the types of households which appear in the returns. The usable "sample" by this stage stands at 167. The diversity of the household structures is astounding, ranging from the most simple (one person living alone) to highly complex multiple family and extended family groups. Shorter returns have a greater chance of being preserved, but even so 36% of households in the sample can be classified as complex rather than simple. Less than 23% of the sample of 167 are made up of a father, mother and children -- and, interestingly enough, such couples are on average much older than one might expect (see p.61). When one considers the village/metropolis split between simple and complex, there is clearly a marked difference: metropolitan households have a greater tendency towards simple; village households also tend to be a bit bigger (complex, by the way, does not necessarily mean big), though lodgers and slaves boost the metropolite numbers. The urban/rural divide ties in with recent findings on the Roman family. The nuclear family has been established as the norm in western, urbanized areas; these census returns reinforce the suspicion that outside of the urbanised West and lower down the social ladder, at least at some points in the family life cycle more complex family groupings were much more common than many may have assumed.10 If one gives weighting to the rural sample, one can conclude that at least 3/5ths of the population that the census returns represent lived in complex households.
It is easy to forget that we are dealing not with numbers but with people, and when reading this book it is often useful to step back from the analysis to consider the social history. Fascinating examples of households abound: the catalogue is well worth browsing in this respect. For example, in P.Berl.Leihg. III.52B (p.226), A.D. 147, there are at least five co-resident siblings, all declared as apatores (i.e. illegitimate), ranging in age from 30 to 14, two married to each other, and the oldest married to an outsider. In P.Berl.Leihg. I.17 (pp.234-35), A.D. 161, three brothers and two sisters live in a house that they rent from two wards of one of the brothers; all five siblings are (or have been) married, two of them to each other (but now divorced), and two of the brothers to two sisters. All have children; a divorced or widowed sister lives in the house with her child. The ex-wife of one of the brothers also lives there, with a child. Finally there are two slaves (mother and son, not daughter as the stemma on p.235 wrongly indicates) of that child. Compare P.Ryl. II.111, A.D. 161, for a similarly complex situation, where a man's former wife (his sister) and their children co-reside with the man's new wife and their children; in total 27 individuals are in the household. In P.Oslo III.99 (p.246), A.D. 161, a married couple, the husband 72, the wife 57, have two children, aged 40 and 8 years; if (and it is a big "if") the ages as stated are accurate, the woman was married by the age of 16, the marriage has lasted at least 40 years, and the second child was born to the woman when she was 49 years old. It is also striking that apparently the 40-year-old daughter is living in her parents' home and is not currently married. With P.Brux. 1.10 (p.260), A.D. 174, we have four brothers living together in a frérèche with their wives (one of whom is also a sister) and children. Two of the brothers also have children in the house from previous marriages. There are 17 family members in all. One of my favourites is BGU I.115 i (pp.266-68), A.D. 189, where a 50 year old male lives with his wife (who is also his sister and who is 54); they have eight children, five boys and three girls, the youngest 7, the oldest 29. The 29 year old son has married his sister, and their two one-year-old sons live in the house; another brother, aged 26 years, has married, this time outside of the family, and also has two sons, the older 13 years old (which means this brother became a father at the age of 13). In this household 27 people reside (at least five of them named Heron!), including some lodgers who are also related by blood. Note also P.Tebt. II.322 (pp.274-75), and BGU II.577 (pp.287-88), a four-generation household.
Quite apart from such fascinating examples, the evidence in total makes for at least one inevitable conclusion: on the basis of these returns, newly married Egyptian couples did not regularly leave their parents' homes and form new households -- again an important finding in contrast to much recent scholarship on the Roman family. The complexities of the households are further emphasised when one remembers that the returns give us a static picture of an ever-changing scene, in which as the life cycle of each member of the household evolves, the household structure continually changes, even from simple to complex and vice versa. But at times the sample simply becomes too small to allow convincing demographic analysis. Just as conclusions cannot convincingly be drawn about changes over time, so too in some cases the differences between urban and rural may be due to the vagaries of survival of returns rather than any real demographic trends. So, for example, no conclusions can be drawn about the differences in slave-holding between simple and complex rural households (p.71).
From households to life tables and life expectancy. Bagnall and Frier construct a female life table, using the Coale-Demeny West tables to make up for deficiencies in the data base. The method is laudatory; the results are highly plausible. Life expectancy at birth of the order of 22.5 years, with a growth rate of a little over 0, is in line with what one expects from comparative evidence (and, for what it is worth, is in line with what little other evidence may with caution be adduced from antiquity). On a few points I might quibble. For example, Bagnall and Frier argue that the annual female birth rate could have been as high as 54 per thousand. Even if the authors are right, as I think they are, to argue that contraceptive and abortive practices had little demographic effect (see below), 54 is too high in any long term regime. Bagnall and Frier criticise my view that a birth rate approaching 50 would be difficult to attain; in their view it would be easy. It is certainly possible, but in the long term (a phrase deleted when quoting from me) unlikely. In fact in the final analysis Bagnall and Frier plump for a birth rate in the low- to mid-forties per thousand, which is plausible enough.
If the female life table produced inspires confidence, and it does, the male figures are less assured. The problem basically comes down to the sex ratio in the data, which, when looked at over age groups, is certainly strange. The difference in sex ratio between villages and metropoleis is also striking, with males much more numerous than females in the cities, but females outnumbering males in the villages. With weighting, the ratio becomes almost balanced, though if one were to argue for a much heavier proportion of the population living in the villages, the ratio would be much more in favour of females overall. Bagnall and Frier make the important observation that male ages are less frequently preserved in the census documents than are female ages, and a corrective is needed for that. They also observe that more males than females appear to be lost from incomplete returns (p.93 n.6) -- one wonders how they can know that! One is similarly left doubting by their assertion that males outnumber females in the free population (p.94); that is probable, but with all the uncertainties and the biases present in the data, one is reluctant to base such a conclusion on a relatively small sample. Similarly, one is hesitant to accept conclusions about the change in sex ratio over time (pp.94-95). An ingenious case is developed, in chapter 5 and again in more detail in chapter 8, for explaining sex ratio difference over age, and between metropoleis and villages, in terms of the migration of young males from the villages to the cities. This is one explanation, and such migrational drift, for employment or for marriage, is quite credible. But defects in the database are another, more immediate explanation. Overall, the sex ratio is a major headache in terms of the census returns and results in one of the less compelling features of this book.11
At any rate the male life table produced is, as Bagnall and Frier admit, much more tentative; the general assumption is that male life expectancy is roughly the same as for females.12 Rather than attribute the improbable features of the age-specific sex ratios to inadequacies in the Coale-Demeny models (pp.106-7), the problem must lie with the returns as they have survived. Bagnall and Frier argue strongly that the female figures are accurate, the male figures more distorted: males of taxable age are underreported and older males' ages are exaggerated. But so are older females' ages. That female ages are systematically distorted too is possible, but difficult to detect or substantiate.
In chapters 6 and 7 Bagnall and Frier move on from the basic life table functions to consider fertility in more depth. Marriage is of course the focus, though it needs to be remembered that it is often unclear from the returns whether people are married (or were once married) or not. More distortions in the figures emerge, again not unexpected in such a relatively small sample. In the early twenties the percentage of married women drops sharply and unexpectedly. Males are marrying earlier (in the early 20s on average) than appears to have been the norm in the western empire. The sample is too small to allow any study of changes over time in marriage patterns, a considerable drawback, or for that matter between metropoleis and villages. More assured findings, however, concerning the age gap between husband and wife are interesting. Young women are marrying older men, and a fascinating scenario is developed where young men are finding it difficult to find a wife, and hence perhaps are migrating to the metropoleis (but was the situation any different there?). One option, close to home, was to marry one's sister, and a good, concise discussion of brother-sister marriage is offered. It is also striking that men are more likely to remarry than women. The census returns make it clear that older women were not remarrying after the death of a spouse or divorce, and that this acted as a control on population growth. The complexities of the remarriage scene, by the way, where spouses bring in children from previous marriages, add much to the picture Keith Bradley has developed of complex family groupings in the urban West.13
Chapter 7, perhaps the most complex chapter of all, breaks new ground in its treatment of fertility. I will not attempt to reproduce its arguments here, but I do want to point to one of the significant findings. John Riddle's recent book on contraception and abortion in pre-modern history has attracted much attention in the last few years and has advanced an alluring and innovative thesis. But Bagnall and Frier have, I think, effectively disposed of his arguments on demographic lines. Contrary to Riddle, it is clear on the basis of the returns that contraception and induced abortion were not widely and effectively practised as means of family limitation. Whatever exaggeration of ages was taking place, it is still evident that women in the later years of their reproductive careers were continuing to produce children. Bagnall and Frier's section on natural fertility is not as clear as it might be, but the findings are important.14 Breastfeeding,15 postpartum abstinence, and non-remarriage of females were the main means of restraining fertility. Infanticide was not frequently practised, but what exposure or infanticide was occurring appears from the figures to have been more common in the metropoleis than in the villages, especially for girls. But here again one comes up against the problem with the sex ratio. And even if a reflection of reality, the sex ratio may be due to the undervaluing of girls and subsequent higher mortality in younger years.
The brief chapter on migration I have already mentioned. Migration remains the most mysterious element of the demographic equation for antiquity, so the potential value of the analysis here is considerable, even if the tempting hypothesis developed about the migration of young males from the villages to the metropoleis is speculative. More could have been said here on anachoresis, the flight from taxes.16
The concluding chapter raises in a few pages some important final questions. As I said earlier, some consideration is given to changes over time, in particular in regard to the Antonine plague. But more significantly in terms of the book as a whole, the question is raised of the applicability of the findings to the wider Roman empire. The authors are cautious to an extent, but clearly believe that their findings have wider applicability than to just Roman Egypt. At one level it might be argued that the results relate only to the Arsinoite and Oxyrhynchite nomes, since it is from there that the vast majority of the returns derive. And indeed the demography of Middle Egypt may have been quite different in certain respects to the rest of Egypt -- for example in levels of urbanisation and in the practice of close-kin marriage. Perhaps the results could be held to be good for Egypt as a whole. Or do they in fact relate to the Roman empire at large? It is true that Roman Egypt would not have differed markedly in many demographic features from the rest of the empire. Average life expectancy at birth, for example, would probably have been at a fairly consistent level, with some regional and temporal variations, and possibly some difference over social class. That almost all people, male and female, married at some point in their lives, and usually at a relatively young age, would also be generally true. Such demographic features are relative constants, with little variation over the pre-modern Mediterranean world.
But regional differences, cultural and otherwise, are important. That the Egyptian census evidence relates to the Roman population as a whole is possible but unproven. It is not that I deny the wider applicability, only that it cannot be assumed. To uncover the demographic realities of the entire Roman empire, with a population at its height of roughly 60 million and covering a vast and diverse geographical area, seems to me a heavy burden to place on 300 census returns (or 233 on ages, or 167 on households) from parts of Roman Egypt over the course of several centuries, with data imperfectly preserved and distorted by various biases. Some other features are worth remembering too. Egypt enjoyed, if that is the right word, a high level of urbanisation, higher than for the rest of the empire apart, perhaps, from Italy. The demography of urban centres may tend (and this appears to be the case in the census returns) to have differed quite markedly in some features from rural areas. It has already been seen, furthermore, that males in the census returns appear on average to be marrying earlier than their western counterparts. Adoption does not figure at all in the census returns, but we are increasingly becoming aware of its importance as a feature of the Roman family. This all brings us back to questions of social bias. It has usually been held in the past, most notably by Keith Hopkins,17 that the census returns, for all their imperfections, are representative of the Egyptian population as a whole. Apart from geographical problems (Arsinoite and Oxyrhynchite nomes predominate; metropoleis are overly represented; Alexandria does not feature at all) it is also possible that the returns that have survived have survived precisely because they belonged to a slightly more affluent than average group of people.18 And this brings us to an important omission from the book. In discussing brother-sister marriage Bagnall and Frier mention in passing Brent Shaw's recent article on the subject.19 But they give no indication of or detailed response to Shaw's broader contention in relation to the census returns, namely that they do not represent the Egyptian population as a whole at all, but rather in the main a select group within that population, namely the direct descendants of Greek settlers. This may be overstating the case, but in any event Bagnall and Frier's comment (p.133 n.83) that "[i]n the census returns, chances are better than one in two that if a 15 year old woman married, her spouse will be either her sibling or a much older man" strikes me as far from typical of the Roman empire as a whole.
The book is well produced and clearly laid out, as always with Cambridge University Press. Typographical errors are enviably rare (I noted only p.116 n.20 "On the [sic] calculating", p.127 "fer tile", p.139 n.17 "signficantly", and p.223 "Adddenda"). It is perhaps worth pointing out, however, that some figures have become rather garbled, though not to any serious extent. For the record, I have picked up the following. In table 2.1, p.34, and table 4.2, p.77, q(65) should read 0.37118, not 0.37188, and l(65) should read 9,883, not 9,833. In table 5.3, p.100, l(10) should read 51,714, not 51,517; L(35) should read 174,240, not 174,420; q(55) should read 0.22024, not 0.22059; and l(90) should read 45, not 43. By p.63 the sample of 167 Egyptian households (p.58) seems to have shrunk inexplicably to 166; it returns to 167 by p.67. Furthermore something has gone amiss in totalling the numbers of female reported ages. In table 4.1, p.75, for 5-9 years, metropolitan females number 17, not 18, and the total is thus 35, not 36; for 25-29 years, the total metropolitan females number 18, not 17, and the total is thus 31, not 30; for 40-44 years, village females number 10, not 9, and the total is thus 19, not 18; for 45-49 years, village females number 12, not 11, and the total is thus 15, not 14; and the totals at the bottom of that table should now read 213 village females, and 339 total females. In table 5.2, p.97, some of these figures are this time correctly given, except for females aged 40-49; thus the total for females again should be 339, not 337. The errors may be traced to table A (pp.334-36), where, on adding the columns, one can see that the total for village females should be 213; 339 for all females; 391 for all villagers; and 712 for all persons. Presumably the final totals need also to be corrected, unless the adjustment is accounted for with uncertain ages. In table C (p.341) the total number of ages of paternity is 155, not 154. In table D (p.342) under "All persons", the nine-year-old is a metropolite, not a villager. And on p.235 the slave Ammonios appears as a female in the stemma, and is the slave of Sarapion, not of Zoidous.
Such trivial points aside, my comments here should not be taken as scepticism about the value of this book. I do think that the data are often pushed beyond their limits, but I would be more critical if the authors had not pursued their arguments as far as they did. As the eminent demographer Ansley Coale states in his brief foreword (p.xiii), the picture that emerges of the demography of Roman Egypt is not completely clear, due to "the limited size of the sample, and because of various kinds of bias in what is written on the returns." But the end result is a considerable increase in our understanding both of the census declarations and of the likely demographic reality of Roman Egypt, and perhaps also in many respects of the Roman world as a whole.
 M. Hombert and C. Préaux, Recherches sur le recensement dans l'Egypte romaine (P. Bruxelles inv. E 7616) (Leiden, 1952).  Five articles in Bulletin of the American Society of Papyrologists 27-30 (1990-93). This book includes a brief appendix, pp.313-19, giving minor corrections to the census texts.  See R. S. Bagnall, "The beginnings of the Roman census in Egypt", Greek, Roman and Byzantine Studies 32 (1991) 255-65.  Some attempt to consider change over time, particularly in relation to the Antonine plague, is made in the concluding pages, pp.173-77. More sophisticated is D. W. Rathbone, "Villages, land and population in Graeco-Roman Egypt", Proceedings of the Cambridge Philological Society 216, n.s. 36 (1990) 103-42, especially at 114ff.  Bagnall and Frier overstate the case that the census data have been all but shunned to date, though it is true that they have been underutilised. In Demography and Roman Society (Baltimore and London, 1992), I stated (see pp.19-22, 27, 58-59) that the Roman Egyptian census returns are the best source of demographic data from the ancient world (best of a bad bunch, to be sure), and I specifically referred (pp.166-67 n.61) to Bagnall and Frier's forthcoming study. Incidentally, Bagnall and Frier (p.xvi) say that their methods could be applied to other sources for ancient history; one wonders what other sources they have in mind.  Especially by K. Hopkins, "Brother-sister marriage in Roman Egypt", Comparative Studies in Society and History 22 (1980) 303-54; cf. Parkin 21.  C. Newell, Methods and Models in Demography (London, 1988) 15. This book is frequently cited (and with good reason) by Bagnall and Frier as a good introduction to demographic methods and models.  This is probably much too conservative: cf. Rathbone, especially p.123.  Following Rathbone in most regards. Most recent, but poor on Roman Egypt, is M. A. El-Badry, "Historical population estimates for Egypt: a critical review," Population Bulletin of the United Nations 31/32 (1991) 70-88.  In this context I might point to my brief comments relating to the relevance of Roman Egypt to the study of the Roman family in Journal of Roman Studies 84 (1994) 178-85, written before reading this book.  A useful critique of recent studies of the sex ratio is D. F. Sieff, "Explaining biased sex ratios in human populations", Current Anthropology 31 (1990) 25-48.  As Newell (as at note 7) 138 makes plain but as seems little appreciated by many demographers, the Coale-Demeny "female" life tables need not be related solely to the females of a population, nor should the "male" counterparts be automatically assumed as applicable to the same population. It is something of a pleasant irony, by the way, and an unusual feature for the ancient historian to discover that in the data derived from the census returns we apparently know more for certain about women than about men.  K. Bradley, Discovering the Roman Family: Studies in Roman Social History (Oxford, 1991). On Egyptian remarriage and other aspects of relevance to this book, see now J.-U. Krause, Witwen und Waisen im römischen Reich (Stuttgart, 2 vols., 1994), with appendices on Roman Egypt.  J. M. Riddle, Contraception and Abortion from the Ancient World to the Renaissance (Cambridge, Mass., 1992). See now B. W. Frier, "Natural fertility and family limitation in Roman marriage", Classical Philology 89 (1994) 318-33, a clearer and more compelling argument.  Cf. Parkin 129-32; also p.182 n.26, where I point out that the practice of exposing children may have had a positive effect on fertility levels, since mothers would not then be breastfeeding and would therefore be more likely to conceive in the immediate future. On the other hand, note also the possibility of amenorrhoea through undernourishment, particularly in periods of poor harvest; cf. M. W. Flinn, The European Demographic System, 1500-1820 (Baltimore, 1981) 31.  On anachoresis note most recently Rathbone 114-19, and S. Link, " Anachoresis. Steuerflucht im Ägypten der frühen Kaiserzeit", Klio 75 (1993) 306-20.  Hopkins. Bagnall and Frier are wary of this to a degree (p.50 n.65).  See D. Hobson, "House and household in Roman Egypt", Yale Classical Studies 28 (1985) 211-29, with Hopkins 315-16; cf. Bagnall and Frier p.49.  B. Shaw, "Explaining incest: brother-sister marriage in Graeco-Roman Egypt", Man 27 (1992) 267-99, responding to Hopkins.