All content of this website is under copyright and subject to all laws thereof. If you are unsure how to properly cite copyrighted material, refer to your style manual or feel free to e-mail me at email@example.com.
BETWEEN INDUSTRY STRUCTURE AND INCOME
Jack P. Gibbs and Janice [Stensrude] Philpot
Population Research Center, The University of Texas
published February 1965 in Land Economics, 41: (1), 74-80
This paper reports a study of the relationship between industry structure and level of income among the United States, 1890-1960. The focus is on changes in the relationship over time, and as such, the findings have certain implications for anticipating not only changes in industry structure but also the consequences of such changes with regard to income. Although no forecasts are attempted, the findings point to the need for planners to consider theoretical and methodological issues in making projections.
The fact that planners necessarily must consider future conditions confronts them with a dilemma. Only rarely can they control change in such a way that their conception of a future condition is realized; and this is all the more true for city and regional planning. Consequently, planners must resort to the projection of trends but this poses a major problem. It is one thing to assert that in the year 2000 land use in the United States will be such and such but it is a quite different thing to state that this condition will prevail if current trends continue. The latter is much more tenable because of its "if X then Y" quality but the qualification does not make the planner's problem less difficult. For one thing, although the "if" statement is intellectually comforting, it is a gross oversimplification because the "currrent" trend is unknown and unknowable. We can only know what past trends have been, not current trends, and the results of projecting past trends obviously depend on the period of time considered. Further, should one take action on the basis of a projection, the "if" qualification is set aside, i.e., as though we have actually accepted a prophecy.
Once the inevitability of working with projections is accepted, the central question becomes: what criterion of confidence should be applied to a projection? Our concern in this instance is general and not technical (e.g., analyzing the relative merits of the various formulae for extrapolating trends). To illustrate, suppose we examine urbanization trends in numerous countries since 1900, and suppose further that we find that in practically all cases the trend is either linear, logistic, or exponential. In contrast, suppose we analyze trends in fertility and find no definite pattern. Our argument is that these findings would be relevant in assessing the confidence that can be placed in projections of urbanization versus projections of fertility. However, one could argue that what has happened in other times and in other countries is in no way indicative of what will take place in a particular case. This hard-nosed approach is appealing, but it borders on the rejection of all projections, a luxury which the purist can afford but not the planner.
We counter this argument with the observation that all science rests on the assumption that what has happened in various places and times is indicative of what may happen in a particular instance. In this context one frame of reference with regard to projections is summed up in the notion of the form of change. Specifically, if it can be shown that change in a phenomenon typically conforms to a pattern (e.g., linear, exponential, logistic, cyclical), then this suggests how much confidence can be placed in projections of trends in that phenomenon in a particular instance.
The notion that change may conform to a particular pattern, not only in one instance but the same form in different instances, is linked to the idea of evolutionary change; that is, the idea that change, at least in some phenomena, is continuous in direction or occurs through a series of stages or states. Despite widespread criticism of 19th century theories on social and cultural evolution,1 there is still evidence of a continuous direction in certain kinds of change, particularly in demographic, ecological, and economic phenomena. A central fact, and one beyond question, is that urbanization, the standard of living, and productivity have steadily increased in numerous countries since the Industrial Revolution. Such trends characterize the United States, where change has followed a linear or exponential course toward a higher level of urbanization, education, real income, and productivity. This pattern holds for the nation as a whole and for each state.2
The fact that change in some phenomena often does follow an evolutionary course is of no little significance to the planner. Rather than merely assume that past trends in a given city, region, or nation will continue in the same direction, he can and should conduct research to determine (1) the frequency with which change in a phenomenon has exhibited a certain form and (2) the association of variation in the form with the characteristics of cities, regions, and nations. If induction has any value, such research could not fail to be relevant in evaluating projections for a particular case.
Apart from concern with the "accuracy" of a projection, the planner is confronted with still another problem—estimating the probable consequences of a projected change. As an example, we typically speculate not just about the future size of the population but also the probable consequences of the population reaching a given number, i.e., what it means in the way of traffic conditions, land use, etc.
As to speculation on the consequences of an anticipated change, the planner is not nearly so much in the dark as in the case of the projection itself. This is necessarily true since he can examine relationships between the phenomenon and the variables considered as possible correlates. Thus, to illustrate, since urbanization is associated with other variables (e.g., industry structure, standard of living, level of education), then one can make plausible predictions as to the characteristics of a given region or country if urbanization does reach a certain level. The major problem in this case is readily apparent. Derived projections necessarily assume that a particular relationship will hold at a future point in time. Yet it is recognized that the relationship between variables (e.g., urbanization and fertility) may change over time. However, even if we grant that relationships may change over time, it is possible that the change is regular and therefore can be anticipated. Thus, comparative research promises to be helpful not only in evaluating projections but also in assessing the consequences of certain kinds of changes.
Most of the foregoing observations on projections are illustrated by the present study of the relationship by states between industry structure at each census year back to 1890 and the 1959 median income. Industry structure was considered because it is a basic feature of any community, region, or nation, meaning that it is closely related to many other variables—urbanization, standard of living, level of education, etc. Further, changes in industry structure tend to follow an evolutionary course, with a shift in the predominant industry from primary (agriculture, forestry, and fishing) to secondary (manufacturing, mining, and building) to tertiary (commerce, transport, services, and other economic activities).3
Median income4 was considered as a second variable, not only because it is believed to be closely linked to industry structure5 but also because it is perhaps the major connecting link between industry structure and its correlates.
The findings are relevant to projections because we have examined the relationship between the industry structure of states at various points in time with current income. Obviously, the industry structure of a state in 1890 does not determine 1959 income; however, if industry changes do follow an evolutionary course, the characteristics of the industry structure in 1890 are indicative of what the characteristics will be after several decades, with the latter being the major factor in the determination of income.
Procedure. Any investigation of historical trends in the industries of states is hampered by the fact that census data, the major source of information on the subject, are not comparable over time. By combining categories and making certain adjustments a roughly comparable series of industry data can be derived for each state back to the census of 1890. Perloff, et al., have published such a series up to 1950 for the following industries: agriculture, mining, forestry, logging, fishing, manufacturing, and a residual category which is predominantly services.6 Using the method of adjustment applied to the 1890-1950 statistics, the authors extended the series up to 1960.7
One major deficiency in historical statistics on industries by states is that roughly comparable data can be obtained only for extremely broad industry categories. This is a crucial point because two states may have the same percent of the labor force in a broad industry category and yet be quite different with regard to sub-industries within the larger category.8
Given the number of persons employed in agriculture, extraction (forestry, logging, fishing, and mining combined), manufacturing, and services, one can compute the percent of the labor force employed in each category and relate these percentage figures for any census year to the 1959 median income of the states. In this case the relationship by states between industry and income measures were expressed in the form of Pearsonian product-moment coefficients of correlation. The coefficients for each industry and each census year since 1890 are shown in Table I.
Results. Table I reveals three patterns in the relationship by states between industry structure and 1959 median income. First, none of the 1960 industry measures are closely related to the 1959 median income. Second, without exception, the 1890 industry measures are much more closely related to 1959 median income than are the 1960 measures. And third, with very few exceptions the coefficients of correlation (which express the relationship by states between the percent employed in a given industry and the 1959 median income) decline regularly from 1890 to 1960.9
The fact that 1890 industry measures are more closely related to 1959 income than are the same industry measures for 1960 is so contrary to what one might expect that it deserves special attention. Two explanations of the anomaly have been explored. The first explanation considers alternative measures of industry structure, and the second focuses on evolutionary trends in industry structure.
Cumulative Industry Measures. One possible explanation of the findings reported in Table I is that 1960 measures are not closely related to 1959 income because the industry at any point in time does not have an immediate effect on income. Stated otherwise, it is possible that only measures which describe industry structure over a long span of time are closely related to income. To consider this possibility the percent of a state's labor force in a given industry category was summed over the 70-year period. Thus, to illustrate, the cumulative measure for agriculture is the percent of a state's labor force in agriculture in 1890, plus the percent in agriculture in 1900, and so on, to the percent in agriculture in 1960.
Each of the four summary measures (agriculture, extraction, manufacturing, and services) were correlated with the 1959 median income. The results are shown in Table II.
Although two of the cumulative measures (agriculture and services) are closely related to 1959 median income, a comparison of Table I and Table II reveals a most significant fact. On the whole, the cumulative measures are not as closely related to 1959 income as some of the 1890 measures considered alone. In fact, only in the case of agriculture is the cumulative measure more closely related to 1959 income and even here the difference between the coefficients is minute (-.856 for the cumulative measure as compared to -.839 for the 1890 measure).
Still another possibility is that the basis of the relationship between industry structure and income changed between 1890 and 1960. Thus, one could argue that with the passing of time the income of a state has become more and more a function of all industries considered simultaneously and less and less a function of any one particular industry. In other words, income is related to current and past trends in all industries. To consider this possibility, a coefficient of multiple correlation (R) was computed between 1959 income and the four cumulative industry measures as independent variables (i.e., R1.2345). Surprisingly, R proved to be only .88, a value which is just slightly higher than the coefficient of correlation between 1890 percent in agriculture and 1959 median income (-.839). Thus, it appears that some of the 1890 industry measures for a state are about as closely related to 1959 median income as any possible combination of industry measures over the years 1890-1960.
Evolutionary Trends. The second explanation of the findings in Table I is that the pattern stems from the use of broad industry categories and evolutionary developments within each industry category. The explanation rests on the following assumptions: (1) with technological changes, sub-industries develop within each major industry; (2) the most recent emerging sub-industries expand more rapidly and generate more income per employee than do the older sub-industries; (3) some states enter into a certain industry earlier than do other states; and (4) as states come to differ more and more with regard to sub-industries, they come to differ less and less with regard to the over-all industry category.10
Table III illustrates the way in which an evolutionary development of sub-industries may bring about a greater similarity among states with regard to a broad industry category.
State X in this instance entered into the industry before state Y and consequently in 1890 it had a larger percent of the labor force in the industry than did state Y. However, as sub-industries appear, the primary sub-industry G (which was once the whole of the industry) declines. State Y on the other hand, for one reason or another, entered the industry later than X; its primary sub-industry (G) is expanding and the percent of the labor force in the total industry category expands. Thus, the differences between X and Y decreased between 1890 and 1920 as far as the total industry category is concerned but differences between the two with regard to sub-industries increased. Between 1920 and 1960 the G sub-industry of state Y continued to expand and by 1960 the states were identical with regard to percent of the labor force in the total industry category, but in 1960 state X had a much larger pecent of the labor force in the sub-industries that generate high incomes.
The hypothetical situation depicted in Table III is particularly relevant with regard to the relationship between industries and income. Assuming that the 1959 median income for state X is $4,500 and $4,000 for state Y, there would be a direct relationship by states between the percent of the labor force in the industry in 1890 and 1959 median income; but there would be no close relationship between the 1960 percent in the industry and 1959 median income.
We do not have the necessary historical series of detailed industry statistics by states to confirm the interpretation of the findings in Table I directly but there is some relevant evidence. First of all, if the interpretation is substantially correct, we should find that the industry structures of states have come to resemble each other more and more with each decade since 1890. In other words, with the passing of time there has come to be less and less variation among the states with regard to the percent of the labor force in each of the four broad industry categories. This possibility has been investigated by computing the standard deviation of the percent in each broad industry category for each census since 1880. The results are shown in Table IV.
The pattern in Table IV is substantially as predicted, i.e., a tendency for each standard deviation in a column to be higher than the standard deviation for a later census year. In fact, whereas one would expect 56 exceptions on the basis of chance, there are only 15 in the 112 comparisons. Stated otherwise, the ratio of correct predictions to incorrect predictions is 97 to 12 or about 8 to 1.
Still other evidence in support of the interpretation of the findings is found in the relationships by states between certain sub-industries and 1959 median income. The sub-industries considered are those judged to be of the recently emerging type. It must be admitted at the outset that there is no definitive criterion of an emerging sub-industry, and the choice of sub-industries is therefore somewhat arbitrary. In the case of manufacturing, only those sub-industries in which the number employed increased more than 100 percent for the country as a whole between 1940 and 1960 were considered. The emerging sub-industries in manufacturing according to this criterion are "machinery, except electrical," "electrical machinery, equipment, and supplies," "transportation equipment, except motor vehicle," and "chemical and allied products."
The criterion of expansion is less applicable in the selection of emerging service sub-industries. Some of the service sub-industries (e.g., education and public administration) which expanded more than 100 percent between 1940 and 1960 are not new sub-industries by any means. Of the remaining expanding service sub-industries, the outstanding common denominator is professional services. Accordingly, the measure adopted is the percent of the experienced civilian labor force in three service categories: "engineering and architectural," "accounting, auditing, and bookkeeping," and "miscellaneous professional and related."11
Figure 1 shows the interrelationships by states among 1960 "emerging sub-industry" measures,12
Product-moment Coefficients of Correlation by States Among
Selected Industry Measures and 1959 Median Income
The coefficients of correlation shown in Figure 1 are fairly consistent with expectations. Both the 1890 industry measures and the 1960 "emerging sub-industry" measures are much more closely rleated to 1959 median income than are the 1960 major industry measures. Further, the fairly close relationships between the 1890 major industry figures and the 1960 emerging sub-industry measures are consistenet with the notion of an evolutionary course in the development of sub-industries within broad industry categories.
The absence of any close association by states between 1959 median income and the percent of the labor force in the broad industry categories of agriculture, extraction, manufacturing, and services in 1960 is clearly contrary to the widespread belief that income and industry structure are in some way related. However, other findings clearly suggest that the industry structures of the United States cannot be analyzed adequately in terms of broad industry categories. But such categories are meaningful if considered in a historical context. The distribution of the labor force among broad industry categories at some distant point in the past is closely related to current income and the only reasonable explanation is that the industry structure at some distant point in the past is indicative of the future structure of sub-industries. The relationship by states between emerging sub-industries in 1960 and the 1890 industry measures provides further confirmation of an evolutionary course in industry structure.
Some of the above conclusions have implications for planning that deserve special consideration. Given a multitude of unforeseen circumstances, such as wars and depressions, the projection of trends may appear at first glance to be rather pointless. But the fact that there is a very close relationship between the characteristics of a state in 1890 and in 1960 clearly suggests that change is by no means purely random and erratic. The findings further suggest that speculation as to some future condition need not be based on the extrapolation of past trends alone; specifically, the condition of a system at any point in time may be indicative of its future state. However, the planner or anyone concerned with projections should realize that existing concepts and perspectives may be inadequate for describing a future condition. Thus, in the case at hand, the distribution of the labor force among broad industry categories in 1890 is closely related to variation in 1959 state income; but currently the same categories neither reveal sharp differences among the states nor do they have a close relationship to income differentials. The crucial differences among the states is now the structure of sub-industries, and the findings clearly indicate the need to reject broad industry categories as a basis for speculation about future changes. It is possible that a similar process is now occurring with regard to urbanization, education, and land use, meaning that existing concepts and traditional measures cannot reveal what will eventually become the distinguishing features of communities and regions with regard to these variables. Here, as in the case of industry structure, an evolutionary frame of reference does not ensure valid forecasts but it does give us a new perspective for describing the salient features of an anticipated condition.