PART I
SECULAR TREND, 1900-1985

Chapter 1
Introduction

"There were giants in the earth in those days," says Genesis, drawing a powerful analogy between secular diminution in physique and degradation in moral fiber.^[1] Linking dramatic change in quality to decisive changes in physique as it does, the quotation is an apt starting point for this work. My two key theses—that improvements in net nutritional intake mainly attributable to a diminution in the demands placed upon it caused by a shift out of labor-intensive work and a decline in the frequency and duration of infection due to more effective medical and public health intervention led to the improvement in population quality interpreted in terms of enhanced capability and work capacity; and that in addition to supply factors like technology, demand factors working both through the market and through entitlements were crucial to this secular trend—are theses about a striking secular change that has rendered Japan's young adult population potentially more productive on a per capita basis and physically taller and heavier than its forbears. Exaggerating somewhat for the sake of emphasis we can say that in quality terms, Japan's population today, in comparison to her population in the past, consists of giants.

Just as the quote from Genesis argues by way of analogy and metaphor, so does the account offered here. But while analogy is important in this work, so is the marshaling and interpretation of historical evidence. In this study I scrutinize a considerable body of evidence, in an eclectic manner. My account is eclectic because it combines both

qualitative and quantitative data and because insofar as it has a quantitative focus it draws on a wide variety of techniques to process and present that data—for instance, regression analysis conventionally employed by economists and cross-classification of data favored by sociologists and anthropologists—in weaving an account that I hope will be comprehensible to historians as well as specialists in various social sciences. That I have adopted an eclectic approach reflects not only my own predilections in analytical matters (in particular, my belief that it is through an interaction of inductive and deductive techniques that progress is secured) but also my desire to communicate my findings to scholars in a wide range of disciplines. I believe this is important because the secular improvement in population quality in Japan was a broad social process going beyond the mere operation of markets as they evolve in response to changes in technology. Institutions at the local and national levels governing the distribution of entitlements over foodstuffs and health-enhancing public health and medical services and the demand for these institutions as voiced through political and social protest play a very important role in my account.

Of the analogies and metaphors central to this volume, the most important are those involving the definition and measurement of population quality and nutritional intake net of the demands placed on it. The remainder of this chapter briefly discusses these key metaphors and some of the thorniest conceptual issues surrounding them and clears the way for more detailed treatment in terms of the historical data for Japan that I analyze in the remainder of this book. I begin with the concept of population quality.

Population Quality and the Standard of Living

By the average quality of a population, I mean the average potential capability and potential work capacity of its members. Capability and work capacity are to be understood in both physical and mental terms. Whether a population of a given average capacity or capability chooses to exercise and develop the potential it has depends on the attitudes toward and the incentives that individuals face concerning the development of these capabilities. From a theoretical viewpoint my concept of average population quality is quite close to the notion put forward by A. Sen et al. (1987) concerning the proper definition of the standard of living. The similarity of my view is briefly discussed in the remainder of

this section. From a practical empirical viewpoint, I choose to use anthropometric measures concerning height, weight, and chest girth for children and young adults at various key ages. That these measures serve as useful proxies for potential capabilities and work capacity is a matter that, to the best of my knowledge, is not discussed by Sen. Discussion of the usefulness of employing these measures as proxies is taken up below.^[2]

Why should we focus on population quality as defined in terms of capabilities and work capacity? There are two lines of reasoning that lead us to this conclusion. The first is the potential feedback of labor productivity on future productivity. That there is a relationship between the productivity of the past and the productivity of the present arises from a variety of potential linkages. For instance, short-run linkages run through the level of nutritional intake or a sense of well-being to work capacity or the eagerness to work in the present period or in the immediate future. Or again this process may take place over several generations. The better fed and the healthier women are in any given period, the more likely they are to give birth to offspring with adequate birth weights and the more nutritious their breast milk is if they elect to breast feed their infants. That feedback appears to have been empirically important in the Japanese case is crucial to the argument I develop in Part II of this book concerning the motivation for and role of population quality-enhancing behavior on the part of households and non-household organizations like enterprises and governments.

The second reason to start with a definition centered on capabilities and work capacity is because it leads to a linkage between the measurement of population quality proposed in this volume and the measurement of the standard of living as proposed by Sen (1987). In my opinion Sen's definition has much to offer when it is contrasted with the more conventional definitions used by economists. For instance, Sen (1987:7 ff.) forcibly argues that the standard subjectivist utilitarian view runs into the objection not only that utility in and of itself cannot be directly measured but that definitions employing it fall short of what most people believe on logical grounds is the standard of living. Thus if we interpret utility in terms of pleasure and happiness, we must address the question of whether persons who are very poor and exploited in material terms may not, by dint of being socially conditioned into lacking ambition and material aspirations, find happiness in their poverty equal to or greater than that experienced by a wealthy individual who commands many more resources in consumption. Again if we think of

utility as defined in terms of the satisfaction of desire, we encounter the same objection raised immediately above that the downtrodden may find their desires so stifled through progressive disappointments that they abandon these desires one by one, retaining only a few modest ones that they are sure of fulfilling. Despite the fact that such a downtrodden individual may well be satisfying the few desires he or she is left with, is it not unreasonable to say that such an individual is well-off? Again if we conceive of utility in terms of choice, we encounter a host of problems: what about choices made by an individual which benefit others and not the individual per se? What about choices involving non-marketed goods like one's own children? Again if two individuals with identical time and income constraints choose different bundles of commodities, can we conclude anything about relative levels of standard of living?

It is theoretically possible to avoid the pitfalls of subjectivist definitions by using consumption, or what Sen (1987: 14 ff.) calls opulence. Now ignoring the question of constructing price indexes so that comparisons between two different economies or within one economy over time can be made, which plagues the actual empirical estimates of relative opulence that are made, there remain conceptual problems with using consumption of goods and services for individuals in the same economy facing identical prices. The two individuals may have different physical needs, for example, different demands for foodstuffs because one individual is ill and the other is not, or because one individual has a higher metabolic rate than the other, or because one lives in a colder climate than the other, and so forth. In practice there is no way to control for all background variables that condition demands placed on consumption so that unambiguous quantitative rankings of individuals can be arrived at. While it is reasonable to suppose that a doubling of income per capita is associated with some improvement in the standard of living, if this gain in income is at least partly purchased at the expense of increased crime or environmental degradation or an increase in stress, it is not at all clear to what extent the overall standard of living has actually improved.

Of course, there are many thoughtful advocates of the conventional approaches using either opulence or opulence taken together with other indicators of quality of life, the nonmaterial components of welfare implicitly being purchased from the resources made available by opulence. For the sake of my ongoing comparison between the measures of population quality that I espouse and the various facets of opulence, specifi-

cally, net nutritional intake, which I will make use of in my empirical analysis as causal factors explaining how levels of population quality are determined, I present some indicators of opulence and associated quality of life measures for Japan over the period 1881-1980 in table 1. Note that while death rates go down and life expectancies go up roughly in tandem with the rise in per capita income and consumption, reported illness rates for middle-aged males also go up. And note that the trend in reported illness has a direct bearing on some of Sen's objections to the conventional subjectivist/utilitarian approaches to measuring the standard of living. Is the trend due to increases in levels of income that allow people to "purchase" more leisure by declaring themselves ill in circumstances that in the past, when income was far lower, would not have served as socially acceptable justification for taking time off? Or are middle-aged men in Japan subject to more stress now than in the past because expectations about work capacity have risen faster than work capacity? Or is the trend a mirror for improved quality of medical care and diagnosis?

In any case, to return to the conceptual issue raised by Sen, he rejects both subjectivistic utilitarianism and opulence as criteria for defining the standard of living. As an alternative he proposes a definition that includes the various "doings" a person achieves; that is, he advocates use of a definition based on achievement or capability (the capacity to achieve). It is not necessary here to go into detail concerning the very interesting distinctions Sen makes between functionings that are achievements and capabilities that are the capacities to achieve and are conditioned, among other things, by freedom. What is relevant to our discussion here is that Sen focuses on what I think many, perhaps most, people mean when they talk about a life "well lived," or "richly lived." People have in mind some concept of the capacity to achieve things, whether it is because they are physically capable of doing things or because they are skilled or well educated or sensitive to opportunities, and so forth. Without taking up the thorny issue of political constraints on freedom to exert oneself, I take a narrow form of this concept of achievement and capacity to achieve and define it as I have done above in terms of population quality. Certainly there is a relationship between population quality as I define it and Sen's concept of the standard of living, but because my definition is narrower than his, populations of identical quality according to my definition may well enjoy quite different levels of the standard of living, say, because of differences in the capital-to-labor ratio or in the political system. For the purposes of this

study a narrow definition will suffice. Moreover, to actually measure the standard of living it is necessary to narrow the scope of our definition even further. Which is the issue to which we will now proceed.

Anthropometric Measures and the Secular Trend in Human Growth

I propose to measure population quality in terms of anthropometric measures for children and young adults, namely, in terms of average levels of height, weight, chest girth, and weight for height indexes, for males and females at various ages up to twenty. To understand why this approach can usefully serve to measure quality, it is necessary to briefly consider three major findings from the field of auxology on which I draw in selecting my measures. These three findings are as follows. (1) There has been a secular trend in the anthropometric measures for adults in all populations that have experienced a pronounced and sustained rise in per capita income, referred to hereafter as the secular trend in levels of human growth. (2) There has also been a secular trend in the tempo or timing of growth in children and youths as they mature toward their terminal adult heights and weights and chest girth. The mean age of maturation has declined; in particular, the mean age when children experience their greatest postinfancy growth has declined over time. This is referred to hereafter as the secular trend in the tempo of human growth. (3) While the gene pool and heredity are important at the individual level and at the average level over a period of many generations, evolution within the gene pool is a minor factor in the secular trend. I take up the third item in the next section and turn to the first two points in the remainder of this section.

Auxology is a field with a venerable tradition reaching back to the Renaissance. J. M. Tanner (1981), one of the leading authorities in the field, has provided us with an engaging account of the evolution of the study of human dimensions from its beginnings in the work of artists like Leonardo da Vinci and Albrecht Dürer to the studies by Montbeillard of his son's growth in the mid-eighteenth century to the anthropological work of Shuttleworth and Franz Boas. Among other findings to emerge from this literature is an appreciation that the process of human growth is quite uniform for each sex taken separately as evidenced by a regular age-specific profile for the growth process and by the simultaneous coordination or coincidence of organ and tissue development for the various components of the body, but that development age and chronolog-

ical age vary individually within populations and over time, in a secular sense, for populations. For instance, Tanner (1961) shows that the development of the brain and the ability to perform well on intelligence tests is related to physical maturation in other areas of the body, like the length of legs and arms and the size and functioning of the sexual organs. Now because of the long history during which auxologists have developed a set of precise measurements for exact calculation of height, weight, and other physical characteristics, and because auxological data is often generated by military and educational institutions as by-products of their examination of physical fitness for the individuals under their command or charge, we can document the secular trend in anthropometric measures for a number of national populations or sub-populations.

And wherever we can document these trends, it appears that they accompany economic and social modernization (see Eveleth and Tanner I990; Tanner 1978, 1994; and the various chapters and preface by Komlos in Komlos 1994). Hence researchers have been given the opportunity to document the secular improvement of living standards by using data on secular movements in anthropometric measures. And this brings us back full circle to the issue of the standard of living and to my position regarding how to interpret secular changes in the level and tempo of human growth.

Note that first of all I do not define or measure the standard of living in terms of the anthropometric measures; rather, I define population quality in terms of capabilities and work capacity, which I measure in terms of the anthropometric measures for children and young adults. By measuring population quality in terms of the properties of children and young persons, many of whom have not yet entered into employment, I explicitly focus on capabilities and capacity for future work as opposed to achievements and accomplishments. Thus population quality in my definition is very close to a narrow version of Sen's concept of the standard of living, which excludes the political and social constraints that may in practice limit the ability of individuals to turn capacities into realized achievements and concentrates on potential adult physical and mental work capacity. Other things being equal, the greater the population quality, the greater the standard of living in Sen's sense. The relationship between population quality and the standard of living defined in other terms—say, in subjective utilitarian terms or in terms of opulence—is complex and far less clear than is the relationship between population quality and Sen's standard of living concept. Suffice it to say

that there is no obvious connection between my concept of population quality and the utilitarian definitions, but that some aspects of the standard of living defined in terms of opulence do play a role in my account as determinants of population quality. It is important to keep in mind that when I refer to the standard of living, I am referring to Sen's concept, and when I refer to the determinants of population quality, I have in mind specific variables that are often incorporated into the opulence definition of the standard of living.

Coevolution

I hope it is by now clear that I do not subscribe to the view that the standard of living, particularly defined in terms of opulence, is equivalent to population quality. While I have already provided a number of grounds for reaching this conclusion, I have not yet considered one of the most compelling, namely, the influence of the gene pool on anthropometric measures. R. Steckel (1994b: 1, 9 ff.) argues that as long as we work with population averages and changes over time or differences between population averages, we can largely control for the influence of the gene pool. He does concede, however, that comparisons between populations of Asian and Western European descent are complicated by genetic factors. For instance, P. Eveleth and J. M. Tanner (1990: chap. 9) provide an abundance of evidence that physical proportions—for example, leg to trunk length as measured by the ratio of sitting to standing height—vary between different gene pools: Individuals of African descent tend to have long legs relative to trunk length; individuals of European descent tend to have moderate leg lengths relative to trunk length; and individuals of Asian descent tend to have short legs relative to trunk length. For this reason, other things being equal, adults of African descent tend to be taller than individuals of Asian descent. That there are observed height differentials does not necessarily speak to the question of whether long-standing or contemporary opulence-based standard of living differentials exist among these groups.^[3] Now it may be thought that while international comparisons are complicated by genetic factors, comparisons within a gene pool or secular changes within a gene pool are free of this problem. Is it not the case, for example, that secular change for, or differentials within, Japan's population (which is often said to be racially homogeneous due to its isolation from the Eurasian landmass) are unambiguously attributable to factors that are not genetic? Unfortunately, as we shall now see, this position cannot be sustained.

First, race itself is a questionable category in anthropological analysis. The prevailing view is that while genetic inheritance is important, there is no such thing as distinct "races." Within the boundaries of Japan live persons of Ainu, Korean, and Chinese descent and/or progeny descended from marriages between persons of different ethnic origin.^[4] For this reason, throughout this book I refer to "Japan's population" rather than to "the Japanese people."

Second, coevolution may exist. By coevolution, I mean the interacting evolution of culture with genes. Imagine that there is random and ongoing genetic change and that most of these genetic changes vanish over time but that some have adaptive or survival value because of the cultural environment in which the phenotypes carrying the genetic coding exist.^[5] The possibility of coevolutionary change has been extensively explored by anthropologists in the last several decades. A number of the arguments in this field are systematically reviewed and tested by W. Durham (1991). Several examples culled from the literature appear in chart 1, below. Note that two major arguments, both of a coevolutionary nature, have been advanced to explain why persons of Asian descent tend to have shorter legs than Africans. One is that because of random genetic changes selected for, because marriage ages were unusually youthful in Asia, or because of diet, Asians go through the adolescent growth spurt approximately a year earlier than non-Asians. Hence, because in the years leading up to the adolescent growth spurt legs are favored in growth, non-Asians have an extra year or more for leg development. The second main coevolutionary argument relates to climatic differences: gene pools that evolved in hot climates require greater heat loss per unit of volume; hence genetic changes that produce longer legs are favored. Insofar as coevolution does occur, does it not occur very slowly, over generations and hence over hundreds or even thousands of years? In the short span of a century or perhaps two centuries is it not reasonable to suppose that there is too little time for coevolution to occur? For instance, is it not reasonable to suppose that the secular gain in standing height that I will document for Japan's population over the period 1900-1985 is due to nongenetic factors?

At first glance the argument appears plausible, but there are problems with it. Taken literally, this thesis means that the typical male living in Japan at the turn of the century had the genetic potential to reach an adult height of, say, on average 170 centimeters but, due to the exertions of physical work, the ravages of infection, and inadequate nutrition, was only able to reach a level of around 160 centimeters. Perhaps this is in fact

the case, and the failure to reach putative genetic potential is a fully satisfactory explanation. But at the present stage of our knowledge of auxology we simply do not know whether coevolution can be totally ruled out, even for analysis covering a period as short as a century. Care must thus be exercised in interpreting statistical associations between secular movements in components of the standard of living defined in opulence terms and secular changes in height, weight, and related anthropometric measures. For simplicity in the analysis of secular trends in chapter 2, I will not explicitly discuss coevolution. I will argue that changes in net nutritional intake are the dominant factor in accounting for changes in population quality in Japan. But I must again warn the reader that we cannot completely simply dismiss the possibility that coevolution may be operating even over a period as short as eight and a half decades.

Gross and Net Nutrition

By net nutritional intake, I mean total (gross) nutritional intake net of the nutrients used to fuel physical and mental work and to fight off disease. In fashioning this definition I ignore the nutritional intake used up in states of pure rest like sleeping (the so-called basal metabolic rate).^[6]

The first of the two key hypotheses of this study is that a major cause of the improvement in population quality in Japan over the 1900-1985 period is a secular improvement in net nutritional intake. To put the hypothesis in simple mathematical terms my claim is that

where f stands for some mathematical function, Q is an indicator of population quality, and N is net nutrition. Now we can write net nutritional intake as gross nutritional intake GN minus the nutritional resources used up in staving off disease and in physical work. Let D stand for an index of the incident of disease and L for the demands placed on nutritional intake by physical labor. Then we can write

where l and d are parameters of negative value. Thus we can rewrite equation (1.1) as

where g is a mathematical function. I will devote chapter 2 to exploring variants of equation (1.3) in terms of a variety of proxy variables for Q , GN , D , and L .

Organization of the Study

We can now, by way of setting the stage for the analysis presented in the remainder of this book, draw together the various themes touched on in this chapter. The view advanced in this study is that population quality has vastly improved in Japan because of improvements in net nutritional intake. When analyzing changes in national aggregate averages, my view is that equation (1.3) (or variants of it) suffices for a quantitative analysis of the relationships involved. This type of analysis is the burden of chapter 2. While the net nutrition hypothesis finds significant support in the analysis presented in chapter 2, it is incomplete because it brushes over the social and economic context within which demand for population quality interacts with supply factors like technological improvements in food production and in medicine level in determining actual outcomes for population quality. To explore demand, we must consider the way demand is voiced both through markets and through nonmarket mechanisms like the demand for entitlements that redistribute demand among various socioeconomic groups. That is, we must undertake cross-sectional analysis comparing regions and socioeconomic groups and we must consider the ways in which various groups and regions voiced their demands for entitlements. Bringing entitlements into the analysis forces us to consider community and government. Governments regulate and set standards for foodstuffs and medicines. Governments also provide public goods and affect the levels of entitlements enjoyed by individual households through various mechanisms of redistribution. But governments are not the only organizational entities that shape the entitlements available to individual households. Enterprises also set standards for their workers and provide entitlements for employees and their household members. In Part III explore the development of the institutions affecting household entitlements over public health and medicine and foodstuffs and in doing so highlight the balkanization of entitlements in prewar Japan. The story I develop is very much the stuff of economic history since it stresses the strength of the market, and it is also the stuff of social history since it stresses social unrest aimed at voicing demand for entitlements. In short, recounting this story that turns on the social history of population quality in Japan underlines the point that we must never neglect the role of political and social factors in the shaping of the great secular trends in population quality evident for the industrialized nations over the last several centuries.

Chapter 2
Secular Trends in Anthropometric Measures of Human Growth and Their Relationship to Net Nutritional Intake

Here I take up the twin tasks of measuring and documenting the secular improvement in population quality in Japan over the period 1900-1985 and of examining the causal role played by improvements in net nutritional intake. The primary reason for focusing on this period is the availability of a long-term annual time series on height, weight, and chest girth for schoolchildren assembled by the Ministry of Education. Fortunately it also proved possible to construct time series for a variety of proxies for the factors underlying net nutritional intake over the same historical stretch of time: gross nutritional intake, the incidence of diseases likely to affect young persons, and the flow of child/youth labor services. The findings support the basic hypothesis of this study, that net nutritional intake is a major determinant of population quality.

I wish to stress, however, that I do not claim that the conclusions arrived at in this chapter have been proven in any absolute sense of that word. What I do show is that, using certain techniques and specific proxy variables, the hypothesis that net nutritional intake matters cannot (within reasonable bounds of confidence) be rejected. Thus the tests I use are based on what is usually known as the classical theory of statistical inference (see Maddala 1992; Pindyck and Rubinfeld 1991). By their very nature, the tests and evidence offered here are provisional: their validity and appropriateness depend on the choice of proxy variables, the degree to which the variables are subject to measurement

error, the design of the statistical procedures themselves, and so forth. That I feel my efforts are not in vain flows from my view that empirical work in economic and social history is fundamentally different from that in fields where one can conduct an ongoing, although finite, number of repeated and controlled experiments. Study of a historical process ultimately relies on processing some portion of the extant record left from actual realized experience in the past. One cannot redo the experiment over and over again. Rather, one can find new data from other periods, other countries, or other sources for the particular country or group of countries currently being examined which have not yet been explored, or one can select different ways of processing the data already in the hands of the community of researchers. Therefore, the field proceeds through an inductive process in which rejection, acceptance, or partial acceptance of hypotheses is used to motivate data gathering and to condition future hypothesis testing.

The chapter is organized as follows: first, I document the secular changes in the anthropometric measures for schoolchildren which serve as the measures for various facets of population quality in Japan; second, I discuss series on nutrition, public health and medicine, and child/ youth labor input that will serve as proxies for the factors underlying changes in net nutritional intake; third, I report the fruits of regression analysis using various econometric methods devised for the analysis of time series data. Because the regression techniques I use are probably not well known by all readers of this book and because some explanation must be given as to why I have selected the ones I have chosen, in an appendix to this chapter I discuss technical statistical issues in some detail.

The Secular Improvement in Height, Weight, and Chest Girth, 1901-1985

The anthropometric measures consist of a complex of related measures of physique: height, weight, weight for height (BMI, the body mass index, defined as weight in kilograms divided by the square of height calculated in meters), chest girth, and so forth. The most commonly utilized of the measures is (standing) height.^[1] The widespread use of height as a measure of the standard of living in the field of anthropometric history reflects the fact that for the last several centuries figures on height have been collected and published by and for the use of military organizations in a number of countries (see panel A of table 2). The degree

to which military data is representative of the terminal heights of young adult males is a matter of debate in particular cases (see the special methods developed by Flood, Walter, and Gregory [1990] to handle British data of this sort). But the fact remains that even when one discounts for problems of measurement, there are clear secular trends in adult height for the United States and select European countries. Indeed the estimates marshaled in panel A of table 2 suggest that while heights for males in the United States did not rise appreciably, heights in the Scandinavian countries showed a dramatic increase. Figures for male schoolchildren aged eighteen in Japan show increases that in percentage terms are comparable to those for Scandinavia. For instance, comparing the figures in panel B.1 of table 2 for the years 1981-1985 with those for 1901-1985, it is apparent the percentage gain in height for males is about 6.6 percent, which is roughly equivalent to the percentage gains for Sweden and Norway recorded over a historical period that is almost twice as long.

To a degree the figures on height suggest that the population has "caught up" with Western Europe or at least closed the gap that once existed. However, one must be careful in asserting this. A gap still remains; on average in 1985 males in Japan tended to be shorter than males in most Western European countries. During the 1981-1985 period height for males in Japan averaged about 170 centimeters. These average levels were reached by males in select European countries and the United States at the following dates: United States, 1715; Sweden, 1913; Norway, 1927; Denmark, 1930; Holland, 1950; France, 1960; and Italy, 1977 (Floud, Wachter, and Gregory 1990:26). Whether this increase in heights in Japan will continue and whether the gap will ever be eliminated is, of course, a matter that we cannot resolve here. But one point that was raised in chapter 1 must be kept in mind: the gene pool has a definite effect on standing heights in Japan. Note from the figures in panel B.2 of table 2 that most (but not quite all) of the gain in standing height has been due to a gain in leg length. As can be seen, the ratio of sitting-to-standing height has been declining as legs have become longer.

The other point that can be readily gleaned from a perusal of the figures for Japan in table 2 is the dominance of secular change in tempo over secular change in level. I give figures on ages 6 and 12 as well as on age 18 both to capture the dynamics of the growth spurt and to take advantage of the six-year differences between the ages. Creating a six-year standard interval is of considerable utility to the statistical analysis

taken up in section 2.4, but an additional benefit can be immediately grasped here. We can calculate the gain in average height for persons aged 12 in a given year t and the same cohort of persons aged 6 in year t - 6. That is, we take the average height for persons aged 12 in year t and subtract from that the average height for persons aged 6 in year t - 6 in order to estimate the gain for the cohort in year t - 6. A secular trend toward earlier maturation, that is, a downward drift in the mean age of the growth process and in the adolescent growth spurt in particular, can be seen from the fact that the anthropometric measures at age 6 and age 12 increase disproportionately in comparison to the gains at age 18. For instance, the percentage gains in height over the 1901-1910/1981-1985 period at age 6 is 8.8 percent and at age 12 is 12.2 percent. The fact that mean age of the growth spurt is shifting downward means that the secular trend in the gains in the anthropometric measures between age 6 and 12 is unusually great. For instance, the age 6 to age 12 gain in centimeters averaged over the 1901-1910 period is 27.2 and the gain averaged over the 1971-1980 period is 34.4. In short, the secular trend in the tempo of growth in height up to age 6 dominates over the secular trend in levels of height at age 18. That is to say, the improvement in population quality can be discerned both in the trends in levels and in the trends in tempo.

The dominance of secular trends in tempo over secular trends in levels is even more evident for females than it is for males. Figures are given in table 3. The percentage increases in height levels over the 1901-1910/ 1981-1985 period for girls aged 6, 12, and 18 are 9.3 percent, 12.6 percent, and 6.4 percent, respectively. The percentage gain at age 18 for females is less than that for males at age 18, but the percentage gains at age 6 and at age 12 exceed the gains for males. The difference in secular changes in tempo is surely linked to the fact that females mature at younger ages than do males, which can be easily gleaned by comparing the gains between ages 6 and 12 for females with those for males. Note that the increase in the amount of the gains from ages 6 and 12 between 1901-1910 and 1971-1980 is identical for both sexes (7.2 cm) and that the gains are always larger for females than for males. Note also that gains between ages 12 and 18 are far larger for males than for females. This reflects both earlier maturation for females and the fact that terminal average adult heights for females are less than those for males. The secular decrease in gains in height for females aged 6 and 12 is an unusually dramatic indicator of the secular trend in the tempo of growth.

While the differences between males and females in terms of the tempo of growth in and levels of height are of interest, the main point I wish to stress is the similarity in the secular trends between the sexes. The correlations between levels of male and levels of female height at ages 6, 12, and 18 are very high: over the 1900-1985 period the correlations are +.997, +.99, and +.97, respectively; and over the 1900-1940 period the correlations are +.97, +.98, and +.96, respectively.

That trends in tempo and level differ somewhat in magnitude does not mean that the heights at the various key ages that we focus on here are not highly correlated. For instance, over the 1900-1985 period the correlation between heights for males at age 6 and 12 is +.98; at ages 6 and 18, +.99; and at ages 12 and 18, +.97. For females the correlations over the same period are somewhat lower: +.96, +.90, and +.91, respectively. That correlations are lower for females probably reflects the striking dominance of secular trend of tempo. It is interesting and useful to keep in mind that over the period 1900-1940 the correlations between heights at ages 6 and 12, at ages 6 and 18, and at ages 12 and 18 are lower than over the entire period 1900-1985. For instance, for males the correlations are respectively +.96, +.90, and +.91; and for females the correlations are respectively +.91, +.88, and +.92. A final point to keep in mind in considering both the differences and similarities between secular trends in the tempo and in the levels of human growth is the greater sensitivity of six-year gains in height to changes in nutrition and other components of per capita consumption. To see this consider figure 1, which graphs annual figures on an index for nutrition (described in the next section), heights for females aged 6, and six-year gains in height between ages 6 and 12. Notice that the drop in nutritional intake occurring in the late 1930s and early 1940s due to Japan's growing military involvement is clearly mirrored with a slight lag in the case of six-year gains in height but that the impact on height levels is far less pronounced. In sum, tempo dominates over levels in terms of secular trend; it is also a more sensitive barometer of changes in the factors underlying population quality. To be sure, individuals who are deprived in the short run may "catch up" later on and therefore the overall impact of a short-term diminution in net nutrition may not be long lasting. Nevertheless, insofar as we are interested in ascertaining the causal connection between net nutrition and population quality, the greater sensitivity of gains in height and gains in other anthropometric measures is an important point that must not be forgotten.

Figure 1.
Indexes of Nutrition, Height for Females Age 6 and Growth for Females
from Age 6 to 12

I have begun my discussion of secular trends in the anthropometric measures in Japan with figures on height because height figures are the ones that are most commonly encountered in anthropometric history. While I do make use of some Japanese military conscription data later on in this study and T. Shay (1994) makes extensive use of it in his study, the bulk of my data is for schoolchildren and comes from surveys conducted in schools by the Ministry of Education. Why do I favor the use of this data? First, it is more comprehensive than the military recruitment examination data: it covers both sexes; it includes figures on weight and chest girth; and it is available throughout both the 1900-1940 and postwar periods (the military draft in Japan ended after World War II, although Japan currently maintains a small self-defense force). Second, the data are available for a variety of ages under 18 and hence allow analysis of tempo effects that, as noted above, are far more sensitive to short-run fluctuations in net nutrition than are levels, especially adult levels.

Despite the compelling virtues of the data set analyzed here, there are defects in it which we must confront before proceeding further. Chart 2 provides a summary background for a general discussion of the data used not only in this section but throughout this chapter. At this junc-

ture I will concentrate on the anthropometric measures covered in panels A, B, and C of the chart. As can be seen, the figures are for schoolchildren, and therefore children in the relevant age groups who did not attend school are excluded. Who was attending school? This depends on the period involved. S. B. Levine and H. Kawada (1980: 48-52) note that by 1900 four years of schooling (typically from ages 6 to 10), made compulsory in 1886, was virtually universal in practice. In 1907 it was decreed that compulsory education be extended to six full years, although this requirement was not successfully enforced until 1918. However, the extent of underregistration should not be exaggerated. For instance, in 1910, 98.8 percent of males and 97.4 percent of females of compulsory school age were reported as attending school (Japan Statistical Association 1988: table 22-1, p. 212). It is thus reasonable to suppose that by the early 1900s most children aged 6 and 12 are covered in the Ministry of Education data set. After the war nine years of education was made compulsory (from ages 6 to 15), and by the 1970s most children in Japan were in fact graduating from high school, which means that the coverage of individuals aged 18 in the Ministry of Education data set is by and large universal for all three ages analyzed here. The problem, of course, is the potential bias resulting from the selectivity of enrollment of children over age 12 before that period.

As can be seen from panel A of chart 2, there is a potential upward bias in heights, weights, and chest girth for males and females aged 18; and the earlier the date, the greater the upward bias. The reason is that the earlier the date, the lower the advancement rate past age 12 in the school system, and as a result the greater the selectivity of the population of those examined by the Ministry of Education. It is demonstrated in Part II of this volume that there are socioeconomic differentials in the anthropometric measures: in general the higher the social and economic status one is born into, the greater the height, weight, and chest girth and the higher the probability of advancing past age 12 in the school system. But as advancement rates increased the extent of this upward bias began to fall. It is likely that there is a contrary tendency at work on heights, weights, and chest girths for individuals aged 18, a contrary tendency that may allow us to conclude that this declining upward bias does not present a major problem for our analysis. The reason for the existence of a contrary tendency lies in the declining mean age of maturation. While I use age 18 as my oldest age for analysis (since advancement rates in the educational system after age 18 are so low that we

cannot usefully employ data on persons over 18 as auxological indicators for the population at large), in fact growth does not necessarily terminate at age 18. But as the mean age of maturation declines, the proportion of the population that continues to mature after age 18 declines. Hence it is likely that the secular trend toward earlier maturation, according to which there is a declining downward bias in the anthropometric measures (the earlier the year, the greater the downward bias), does counteract the upward bias due to a decline in the selectivity of the population receiving an education through age 18. For this reason I feel relatively comfortable using data for all three ages, 6, 12, and 18. But I must caution the reader that the problem does exist.

Now to return to a point made earlier about the virtues of using a data set that allows us to measure and analyze a variety of anthropometric characteristics of Japan's population, let us turn to figures on weight and the BMI. Figures for males appear in table 4 and figures for females appear in table 5. The virtues of having data on weights for both males and females (not just males, as is usually the case when one is using military recruitment data) can be seen from a comparison of trends in the BMI for males and for females. For if we calculate percentage gains in weight for the period 1901-1910 to 1981-1985, we get the following figures for percentage increase in weight and the BMI.

Note that the increase in weight for females at age 18, contrary to the increase at ages 6 and 12, is far less than that for males. The reason cannot be physiological: as can be seen from table 5, this is almost entirely a postwar phenomenon. The obvious explanation is dieting, and the reason for dieting is the concept of beauty for women that places strong emphasis on being slender. (The concept of sacrificing potential physical work capacity for beauty is undoubtedly much more typical of the urban nonagricultural population than of the farming population for whom potential for work is a virtue among young women on the

marriage market. Hence it is not surprising that the trend is much more evident during the postwar period when the farming population has been rapidly dwindling than during the prewar era.) Thus the weight and BMIs of young adults were apparently shaped by a cultural rule that does not appear to have been operating for younger children. This is an additional argument in favor of using data for young children as well as adults. It should be noted, however, that height of females at age 18 appears to be unaffected by the practice of dieting. Correlations between measures of height and weight for males and females presented in panels B and C of chart 1 testify to differences between height, on the one hand, and weight and the body mass index, on the other, which must be kept in mind in interpreting the anthropometric measures as measures of population quality.

Trends for a third major auxological measure, chest girth, are explored in table 6. Particularly striking for females is the secular trend in tempo that overshadows the secular trend in levels. For instance, in comparison with 1901-1910 when the six-year gain in chest girth for girls aged 6 is 11.6 centimeters, the average six-year gain during the 1981-1985 period is 18.2 centimeters.

Finally, the figures on anthropometric measures in table 7 illustrate several points about the drawbacks and strengths of the data set for school children on which my analysis in this chapter rests and provide additional background about prewar trends in height.^[2] The following observation supportive of use of the data on schoolchildren leaps out at us. Trends in average height and weight for persons examined for conscription examinations are much less striking—indeed, the trends in averages are almost ambiguous—than are the trends we secured for schoolchildren. To be sure, it may be argued that truth may lie on the side of an ambiguous or uncertain trend and that, therefore, finding this trend in the military recruit data is a virtue and not a deficiency. But there are reasons for thinking that the figures on schoolchildren more accurately represent the true underlying pattern. To see my point, note that the figures on percentages "small" (defined as 150 cm or less) and on percentages "tall" (defined as 170 cm or more) point toward an unfaltering increase in height that is consistent with what—using averages—we have found for male schoolchildren. Indeed, the fact that moments of the distribution of heights other than averages are available for military recruits is one of the most attractive features of the military recruitment data. But the shortness of the time period for which the data on distribution of heights is available (1915-1940), the lack of figures

on females, the absence of any measurements capturing the timing of the growth spurt, and so forth, make sole reliance on these data alone problematic. I do make use of them to a limited degree in Part II of this study.

Indexes of Nutrition, Public Health and Medicine, and Child/Youth Labor Input

The goal underlying construction of the measures discussed here is to secure proxy variable(s) capturing the three main factors underlying levels of net nutritional intake: gross nutritional intake and the uses to which human bodies put that nutritional intake other than in physical growth—physical exertion and work and fighting off diseases, especially infectious disease. The human body must either dip into its reserves of fat and muscle tissue or into flows of new nutritional intake, thereby potentially diverting the body's built-up, or freshly secured, reserves away from physical and mental growth, in generating energy that keeps it healthy and capable of sustaining physical exertion. The basic equation (1.3) decomposing gross nutritional intake into components that are used in human growth, into components that are utilized in fighting off disease, and into components that are consumed in physical work guides us in constructing the proxy variables. The extent to which the proxy variables capture or fail to capture the true factors underlying net nutritional intake for the average young person living in Japan during the 1900-1985 period is a matter that can be debated, but I believe my measures can be usefully employed in the time series analysis described in the next section.

The main assumptions underlying the proxy variables for the three components of gross nutritional intake are reviewed in chart 1. Here I will restrict my remarks concerning the variables to the following points concerning the philosophical approach that guided me in constructing the variables. One of my guiding principles was to construct variables that do not decisively depend on any one type of data or on any one type of assumption. For instance, it has been pointed out to me that there may be a bias in the figures on labor force participation for children and young adults used in the construction of the index of child/youth labor input during the first decade or so of the twentieth century (presumably the bias diminishes with time).^[3] Since the labor force participation rate only receives half of the weight in the overall index, this potential source of bias is reduced, albeit not completely eliminated. Another

guiding principle was, wherever possible, to construct two or more proxy variables for the underlying factor to see to what extent different assumptions about the appropriateness of a given proxy are valid. Hence in the case of gross nutritional intake I offer one series for the prewar period 1900-1940 and three for the postwar period up to 1985. Finally, since I am interested in relative and not absolute levels of the proxy variables, I constructed indexes (typically with the average value for 1900-1904 = 100) for each of the components of each overall index and then, by weighting each component index, calculated the sum of the components to arrive at an overall index. By construction, the resulting variable is a composite index.

Of the many potential problems that reduce the reliability of the series advanced here, one seems to me to be of sufficient concern to mention here: most of the series are per capita averages and hence may not accurately reflect the circumstances specific to children and young adults. This objection is not valid in the case of the proxy variable for child/youth labor input but is valid for the nutritional intake and public health and medicine series. Without doubt we do not know how food and public health and medical services were distributed within families. For instance, a bias in favor of males or eldest male sons among siblings has been extensively discussed in the literature on the Japanese family. Such a bias may exist, but the high correlations between the anthropometric measures for male and female children should be food for thought for those who would argue that such a bias undermines the validity of the measures used here. Moreover, it should be pointed out that experiments with dividing the nutrition series by a consumer unit weighted population—that is, by age- and sex-specific population weighted by numbers giving the relative consumption demand of the group compared to prime age males—did not reveal a significant difference in trends in nutrition between the per capita and the per consumer weighted population estimates.^[4]

A useful summary of the various component indexes underlying the gross nutritional intake variable appears in table 8. Since some data were available only for the postwar period (e.g., estimates of carbohydrate and fat and calcium intake), two nutritional indexes, a composite index with wide coverage (NUTI2) and an index for calcium and fat intake (CALCFATI), are calculated for the postwar years only. The index available for the entire 1900-1985 period, NUTI1, is a composite of indexes for per capita per diem consumption of calories, protein, and four types of vitamins (see panel F of chart 2 and notes to table 8). Underlying

the trends in all three indexes are two factors: increases in consumption of foods traditionally eaten during the preindustrial era before the 1880s, that is, an expansion in the quantity of nutrient intake with a given diet; and a change in the composition of food intake, that is, a structural shift in diet. As far as the composition of diet is concerned, it is important to recognize that the opening up of Japan to trade in the 1850s and the American Occupation after 1945 both left a mark on the dietary intake of the population of Japan. In particular, a shift toward calcium and fat partly because of a growing consumption of dairy products is directly related to internationalization of the Japanese diet after the 1850s and, again with greater force, after 1945.

To grasp the importance of the composition of diet, let us compare the Japanese diet in 1861 with selected years during the period 1874-1940. M. Umemura, N. Takamatsu, and S. Itoh (1983: 35) give the following percentage distribution of the main staples in 1861: rice, 47 percent; barley, 28 percent; assorted grains, 19 percent; and potatoes, 3 percent. Throughout most of Japan at the close of the preindustrial era, rice was the main source for caloric energy and for proteins.^[5] The role of rice declined until the 1930s, when it increased briefly; since the war consumption of rice has been in decline, especially in urban areas.^[6] Consider the calorie and protein intake arising from various sources (Mosk and Pak 1978).

The trend toward growing diversity in diet during the 1877-1920 period is apparent. However, during the interwar years there is a noticeable shift back toward rice, although the contribution of fish to overall protein intake seems to be on the rise during the interwar epoch. Perhaps even more dramatic in terms of a shift in the composition of diet antedating the American Occupation is the pronounced shift toward fruit as a source of vitamin A. The following figures give us important indicators of the impact of shifts in diet on the percentage composition

of vitamin intake by food type for the years 1874 and 1940 (Mosk and Pak 1978).

Of course, these trends in diet may well be overshadowed by the dramatic increase in calcium and fat intake following World War II conditioned by a growing American influence in dietary matters. There was a definite shift toward coffee, cereal, toast, and juice for breakfast and toward lunches based around sandwiches and milk. Moreover, school cafeterias served milk at lunch. But as important as the postwar shift is, the extent of change during the prewar era should not be ignored.

If the linkage between human growth and diet were more accurately understood, it might be possible to focus on more specific indicators of food intake. Lacking this kind of detailed information, my approach is to explore the impact that two composite indexes (NUTI1 and NUTI2) and one more specific index (CALCFATI) have on human growth.

My approach to the construction of an index of public health and medicine is based on slightly different reasoning. The basic indexes underlying the overall composite variable are discussed in panel D of chart 2 and estimates for the 1900-1985 period appear in table 9. My idea was to use two criteria in constructing an index of the impact of public health and medicine on a reduction in the frequency and severity of disease: the level of inputs per capita (a quantity measure) and the efficacy of the inputs (a quality measure). Before the 1940s and the development of antibiotic drugs, the ability of doctors and hospital staffs to effectively treat disease was much more limited than it was after 1905. Hence my approach is to build up an index that recognizes that the ratio of doctors (and by inference other medical personnel like nurses

and pharmacists) to population is important, but so is the effectiveness of the treatments offered. Of course, public health measures designed to stop the spread of microorganisms were also important. Hence I utilize two measures that implicitly take into account the impact of public health programs and the efficacy of therapy and inoculation offered by the medical community: the inverse of the death rate from four major infectious diseases—tuberculosis, pneumonia, bronchitis, and enteritis—which takes into account both incidence of infectious disease and treatment of it when it occurs; and cases per death for four other infectious diseases—cholera, dysentery, typhoid fever, and smallpox—which measures in inverted form the efficacy of the medical community in treating infectious disease when it occurs. I use inverses for death rates and the death-to-case ratio to construct an index that grows in value as efficacy improves and disease incidence declines. And I concentrate on infectious disease since I am mainly interested in children and young adults and it is infection that is the main source of illness for them.

What is striking about the variable measuring public health and medicine is its dramatic increase after the 1950s. This is testimony to the effectiveness of the antibiotic drugs in treating infection.

The last of the three proxy variables, that measuring child/youth labor input, is described in panel E of chart 2. Estimates of the composite variable and its two components for the 1900-1985 period appear in table 10. Two series exist, one covering the entire period and the other beginning after the war. There is little that needs to be said here about this variable. The trends in it are strongly affected by two factors: the growth of the educational system and the secular increase in the length of compulsory education within the system; and the shift from unpaid family work, exemplified by agriculture and forestry, to wage labor. Children were an important source of labor in small shops and on farms. Hence the shift away from this activity had a decided impact on the degree to which children were introduced into constant physical exertion from an early age. The variable is also an indirect indicator of the physical work levels to which mothers were accustomed, which presumably had some effect on birth weights of children.

In sum, in this section I have described the description of three composite variables for gross nutritional intake, for levels of and efficacy of public health and medicine, and for child/youth labor input. What impact these three variables had on net nutritional intake and hence on the levels and the tempo of human growth is the subject of the next section.

Long-Term Determinants of the Anthropometric Measures, 1901-1979

My goal here is to provide some measure of the relative success of the gross nutrition hypothesis in accounting for the secular trend in levels and tempo of human growth. It is important at the outset to be clear about the goals I have set for myself in the analysis that follows. What I show is that using the criterion of classical statistical theory one cannot reject the net nutritional hypothesis for Japan. I also show that given this provisional acceptance of the net nutritional hypothesis, changes in the demands placed on nutritional intake, rather than changes in the levels of gross nutritional intake, appear to be decisive to the secular trend in human growth and population quality.

Is such a modest aim satisfactory? There has been considerable debate on this point over the last two decades. There now seems to be some sort of consensus among scholars in the social sciences that an arbitrary distinction between the methods of science and those of the humanities, which was the pivotal notion of logical positivism, is not possible. Indeed, the prevailing view now seems to be that rhetoric and argument based on a notion of persuasiveness that goes beyond a mechanical recounting of statistical "tests" and "experiments" is the best we can aspire to in the social sciences (see Denton 1988; Klamer, Mc-Closkey, and Solow 1988; McCloskey 1994; Mirowski 1987). To accept this position is not to embrace an extreme version of relativism according to which there is no criterion by which we distinguish between competing hypotheses and viewpoints. But it does lead us to abandon the idea that any given statistical test is absolute and definitive. And it leads us to a pragmatic approach to the choice of statistical methods. In particular, it leads me to a strategy based on two criteria: use of forms for time series analysis, which allows me to estimate elasticity so I can differentiate between stronger and weaker effects, and use of cross-sectional and qualitative data as a supplement to time series analysis. Part II of this volume explores the latter kind of evidence.

In what follows I use log-log regressions to explore the impact of the three major factors underlying trends in net nutritional intake on the secular trend in human growth in Japan over the eight and a half decades between 1900 and 1985. The particular regression formats I employ are presented in chart 3. In the appendix to this chapter I provide a discussion of some of the technical reasons for using the particular functional forms underlying my results.

The results appear in table 11 (for height), table 12 (for weight), table 13 (for the BMI), and table 14. While there are differences between the findings for each of the measures taken separately, their consistency is striking. Indeed, as a practical matter, consistency is important to my pragmatic approach because I do not believe that any one "test" will ever be decisive. What is the main message to emerge from these results? It is twofold. First, the impact of gross nutritional intake as measured here is not great. For instance, results secured with the two broad composite variables for gross nutritional intake, NUTI1 and NUTI2, based on a wide variety of nutrients like calories, proteins, and vitamins, do not seem to have a consistent and positive impact on the levels and six-year gains in the anthropometric measures. And insofar as nutrition does seem to have a consistent statistically significant impact, it does so in terms of fat and calcium intake; it is the CALFATI

variable that gets the best results in this time series analysis. Thus one tentative conclusion that can be drawn from my analysis is that the shift toward dairy product consumption that was especially pronounced after World War II contributed to the secular trend in population quality. Second, secular trends in demands placed on nutritional intake seem to have dominated in the secular trend in population quality, especially during the prewar period. For instance, the elasticities on the proxy for child/youth labor input are especially large in the prewar period. After the war this index appears to be less important and the index of public health and medicine appears to be more important, although its estimated elasticity does not tend to be very large. These results accord with common sense. Before the war and the introduction of antibiotic drugs, the efficacy of public health and medicine was limited; and immediately after World War II, compulsory education was extended through to the end of middle school, which drastically reduced child/youth labor input.

Putting more social detail in these somewhat dry results is my aim in the second part of this book. But the results reported here are of interest and give support to the net nutritional hypothesis basic to this study.

Summary and Conclusions

During the first eight and a half decades of the twentieth century there was a dramatic improvement in population quality as measured by an assortment of anthropometric measures for schoolchildren. I have provided a wide range of data that document this secular trend for both males and females. It is the basis for my assertion at the beginning of this study that the person living in Japan today is in many ways a "giant" in comparison to a person living in Japan a century earlier.

I have also presented estimates for three composite variables—one for gross nutritional intake, one for public health and medicine, and one for child/youth labor input—whose trends over the last eight and a half decades strongly support the inference that net nutritional intake has dramatically improved in Japan. Finally, I have shown that the hypothesis that the trend in population quality is related to (caused by, I might assert) improvements in net nutritional intake cannot be rejected in time series analysis.

In short, I have a story to tell about the secular trend in population quality in Japan and evidence that supports that story. But the story is incomplete. In the real world demand matters as well; and in the give-

and-take of the real world, markets and nonmarket factors shape the way demand is voiced. Hence consideration of our second theme concerning demand brings us down from the realms of abstraction to what, I trust, the reader will find to be a more concrete story. This concrete story concerns the way in which particular social groups go about using the market and entitlements to secure the levels of population quality they demand. In particular, I argue that because the market played a major role in shaping the demand for population quality during the feudal period and because entitlements were balkanized during that era, a legacy was created which played a major role in shaping developments after the 1880s when the nation began industrialization. This is the story I turn to now.

Appendix
Choice of Functional Forms for Regression Analysis

As chart 3 shows, I use first differences of logarithms of my dependent and independent variables in carrying out the time series analysis described above. Why did I use first differences of logarithms? Because I wanted to cope with potential nonlinearities that might plague the analysis and also because I am interested in estimating elasticities. In addition, I found through the use of Dickey-Fuller tests that most of the variables had unit roots (see Maddala 1992: chap. 14; Pindyck and Rubinfeld 1991). The problem here is technical but can be described in a nontechnical manner as follows: do the variables analyzed by regression analysis satisfy the assumptions of classical regression theory? If the variables do not meet the requirements—and in time series analysis many variables do not meet the requirements because of unit roots and other problems—one must take some action to cope with potential difficulties in making inferences using the t -statistics and confidence bands on parameter estimates. In the case of the analysis discussed in section 2.4, I found that most of my variables failed to meet the stipulated requirements because they have unit roots. For this reason I used the standard procedure of taking first differences to cope with the attendant problems.

A more sophisticated approach was suggested to me: the use of vector autoregression (VAR) procedures. I decided to eschew this approach because it is most fruitfully used when one is interested in forecasting. In forecasting one is not very interested in individual parameter esti-

mates but rather in the relative reliability of the entire estimated structure of a set of equations for predicting one or several outcome variables. Forecasting population quality in terms of the anthropometric measures is of no interest to me. As I stated earlier, I have opted for an eclectic approach that does not rely on one particular set of tests and one particular data set. Indeed, it is my view that the critics of logical positivism in social science have established a valid methodological point by demonstrating that any assertion of validity based on a technique, no matter how sophisticated, is really a rhetorical device and therefore subject to debate. At any rate, this belief informs the pragmatic attitude toward statistical analysis employed throughout this study.