Population Change and Farm Dependence: Temporal and Spatial Variation in the U.S. Great Plains, 1900-2000*
By White, Katherine J Curtis
I investigate the relationship between county population change and farm dependence in the Great Plains region during the twentieth century, using spatial data analysis techniques. This research is rooted in a long-standing sociological and demographic interest in population responses to economic transitions and informs the theoretical understanding of urbanization processes. Using census and environmental data, the analysis challenges earlier assertions of a simple transition in the relationship between farm dependence and population change that accompanied modern technological advancements, namely tractors (the mechanization thesis). Rather than observing the proposed positive-to-negative shift, study results show a negative association throughout the pre- and post- mechanization periods. Partial support is found if the thesis is revised to consider the relationship between population change and the change in farm dependence rather than the level of farm dependence. Findings show mixed support for an alternative argument that nonfarm industries moderate the influence of farm dependence (the industry complex thesis). In contrast to earlier applications of the thesis, industrial relations in the Great Plains context are characterized by specialization rather than cooperation. Although the United States has witnessed considerable population increase over the twentieth century, growth has been uneven and episodic across subregions of the United States. The history of urbanization in the United States can be summarized as development originating on the East Coast, followed by expansion into today’s Midwest and the West Coast. The middle part of the country was generally skipped over; the Oregon Trail cut through the Great Plains but led to little settlement along the path. Economic interests motivated the development of transnational railroads to connect the easternmost and westernmost populations and markets of the country. Settlement of the Great Plains was promoted by the federal government directly through homestead acts, and indirectly through land grants to railroads that were sold to settlers. The remaining decades of twentieth-century U.S. urbanization are summarized by concentrated growth along the East and West Coasts, followed by growth in the South and Southwest. The main “losers”-the subregions that experienced considerable population loss during the twentieth century-are the Rust Belt, due to deindustrialization, and the Great Plains, presumably due to changes in the agricultural industry.
Two economic sectors changed dramatically during the twentieth century: agriculture and manufacturing. Yet only occasional and somewhat superficial attention has been given to the productivity side of the United States’ rural sector that promoted the industrial sector. In this study, I focus on the former while considering its relationship to the latter and suggest that looking back on the agricultural transition might inform the theoretical understanding of large-scale and long-term U.S. urbanization. Advancements in farm mechanization that coincided with industrialization had significant implications for the agricultural industry. I investigate the dynamic relationship between one part of the agricultural industry (farming) and population change in the agriculturally dependent region known as the Great Plains. I evaluate widely held beliefs about the relationship between farm dependence, an indicator of agricultural activity, and county population change through a systematic analysis of the entire twentieth century. I also account for underlying spatial autocorrelation that would otherwise bias study results. This stream of investigation is closely related to long-held demographic and sociological interests in urbanization and population issues.
In this study, I examine what changes occurred in the agricultural sector. Specifically, I ask, how is farm dependence associated with county population change in the Great Plains? My investigation of this relationship generates data about what was happening during this remarkable period in U.S. history and development. Study results reveal a complex and unstable relationship, in terms of magnitude and statistical significance, between farm dependence and county population change that is not adequately described in the dichotomized pre- and post- mechanization terms characteristic of much recent scholarship on the subject. The mechanization thesis is supported if the thesis is revised to consider change in farm dependence rather than the absolute level of farm dependence. The findings also challenge research that suggests that nonfarm sectors and their relations to the farm sector moderate the influence of farm dependence on population change. Mixed support for this industry complex thesis likely is due to industry specialization in the Great Plains.
Using decennial population censuses, agricultural censuses, and environmental data from 1900 to 2000, I employ spatial data analysis techniques to assess the relationship between farm dependence and population change among the Great Plains counties during the twentieth century. This analysis offers a comprehensive and spatially integrated investigation of classic theoretical arguments pertaining to the relationship between economic transitions and aggregate implications that are treated in the current context as a shift in local economic base and the accompanying population response.
Mechanization and Technological Innovation
One body of literature investigating agricultural economics and population dynamics focuses on the “pre-mechanization” period and the “post-mechanization” period, with 1940 marking the decade of transition (Albrecht 1986, 1993; Beale 1988; Bender et al. 1985; Rathge and Highman 1998). According to this research, the Great Plains experienced considerable population growth during the pre- mechanization period that coincided with a booming agricultural industry, specifically the farming industry (in contrast to ranching) and, even more specifically, dryland farming. In the terms of the theoretical schematic employed in earlier studies of this transition, the economic boom created a demand for more labor, which in turn created more production, since farm outputs were rooted heavily in human labor prior to the technological advancements made later in the twentieth century.
The post-mechanization period, in contrast, was one of considerable population loss in the Great Plains that coincided with the transformation of the U.S. agricultural industry. The theoretical framework adopted by these earlier studies argues that greater production (rooted in the pre-mechanization period and extended by technological innovations in the post-mechanization period) created greater economic competition, which resulted in less demand for labor. The implication of the decline in labor is population loss.
Farm mechanization occurred well before 1940. Indeed, by the mid- 1800s, John Deere was producing more than 10,000 “singing plows” per year (Cochrane 1993:190). Threshing rigs, grain combines, and steam engine tractors appeared on the American agricultural landscape before the close of the nineteenth century, and the gasoline engine tractor emerged during the first decade of the twentieth century. Still, the implications of these machines were not dramatically felt until considerable fine-tuning enhanced their efficiency and, in turn, widespread appeal of the more cumbersome early iterations of the farm implements.
For example, drawing from agricultural census data, Cochrane (1993) found a 90% reduction in the number of labor-hours it took to produce 100 bushels of wheat between 1830 and 1930. American farmers reduced the required labor-hours by another 80% in only 45 years (between 1940 and 1975) and produced more bushels on fewer acres (see also Friedmann 1978:565-66). The machines were invented prior to 1940, but improvements and adoption rapidly improved after 1940. The pre- and post-mechanization terminology used throughout the remainder of the study refers to one wave of technological change in farming, but a wave with considerable momentum and impact.
The general argument about the relationship between farm dependence and population change proposed in much of the earlier research, called the mechanization thesis in the remainder of this study, draws from human ecological theory and suggests that area population size is associated with its capability of economically supporting its inhabitants (Duncan 1964; Gibbs and Martin 1959; Hawley 1950, 1986). Exogenous factors disrupt the balance or alter the relationship between economic base and population size, namely technological innovations and environmental factors. Research on the Great Plains and the twentieth-century agricultural transition primarily focuses on technological innovations and draws on earlier assertions that mechanical innovations are accompanied by changes in the organization of relationships (Hawley 1986:60).
Studies of farm mechanization within the Great Plains claim general support for this argument. Researchers have suggested that population trends during the post-World War II era are significantly associated with the degree of farm dependence and that the nature of the relationship is negative (Albrecht 1986, 1993; Beale 1988; Bender et al. 1985; Rathge and Highman 1998). For example, in his analyses of nonmetropolitan Great Plains counties since 1940, Albrecht (1986, 1993) found a consistent negative relationship between the proportion of the population employed in farming and county population growth. The negative relationship persisted even during the dramatic population turnaround of the 1970s; counties with high farm dependence were no more likely to experience growth during the turnaround than they were in previous decades. Other, descriptive research has documented higher loss and lower growth in recent decades among farm-dependent nonmetropolitan counties relative to all other nonmetropolitan counties (Fuguitt and Heaton 1995; Johnson 1989; Johnson and Fuguitt 2000). This theoretical framework suggests that as the agricultural industry became increasingly mechanized, it required less human labor. Economics of scale led to larger and fewer farms, which in turn fostered disequilibrium between population size and economic opportunities. The response to this imbalance was, or is, population loss; the economic need for fewer hands is met by population re-adjustments that ultimately produce fewer hands to employ. Population re- adjustment, manifested as population decline, has far-reaching consequences that extend beyond farm jobs and affect the service and professional sectors of a local area. Other services and sectors include seed and parts distributors, as well as educational and financial providers. For example, between 1940 and 1970, many rural communities within the United States lost upwards of 50% of their residents (Beale 1978, 1980; Larson 1981), not all of whom were directly engaged in farming. Consequences of changes in the agricultural sector are not limited to the agricultural population, but extend to the entire area population.
Farming-Manufacturing Industry Complex
A second body of research proposes a more complex relationship between farm dependence and population change. According to this perspective, the relationship between farm dependence and total population change fluctuates historically and regionally and depends on global and national market conditions as well as nonfarm employment opportunities.1
For example, in her study of the rise of the wheat market in the United States and Great Britain, Friedmann (1978) found evidence of transformations in the type of production during the late nineteenth and early twentieth centuries: family farms versus wage labor. In the late 1800s, family farms competed against and, in effect, were replaced by large, mechanized farms that were able to out-produce the smaller family operations. When the global market for wheat changed and prices fell, however, family farms were left standing “long after wage laborers and capitalists  abandoned wheat production for better wages or profits elsewhere” (1978:568).
Technology plays a role in Friedmann’s analysis and in the theoretical perspective that claims a complex relationship between farm dependence and population change. Indeed, the adoption of farm technology by the family farm is a central source for successful competition. Mechanization makes competitive production possible for family farms by reducing labor inputs while simultaneously permitting acreage expansion. Yet also embedded in this analysis is the influence of the labor market and wage alternatives.
In their analysis of regional development in the Midwest, Page and Walker (1991) examined the interdependent relationship between the agriculture and manufacturing industries during the nineteenth century. Page and Walker asserted that, at this time, agriculture and manufacturing interacted with one another to produce a unique competence in both industries, and that mutually reinforcing activities of the industries were located within the single national subregion. The cooperation between the industries enabled each to prosper. The authors argued that the Midwest is unique from other areas within the United States and that the distinctiveness of the region contributed to the exceptional relationship between agriculture and manufacturing. The Midwest, of course, is not the Great Plains in terms of geography or functionality (e.g., the Great Plains is often treated as the “hinterland” of the Midwestern regional complex, even in Page and Walker’s analysis [1991:305]). There is manufacturing on the plains, yet it is not as well developed nor did it develop as early as in the Midwest. Still, the general nature of the relationship may apply to the understanding of farm dependence and population change on the plains.
According to this second, temporally and regionally varying perspective, a nonfarm economic alternative is the mechanism through which farm dependence influences population change. There are two basic scenarios. First, when family farms are able to reproduce themselves at a rate on par with wage labor, employment in the nonfarm sector is not attractive enough to draw household members away from the farm sector. Where farm income is not competitive with wage labor, and where a nonfarm wage labor market exists, household members will be inclined to leave the farm. Under these conditions, the amount of farm population would fluctuate dramatically, yet the total population would remain stable due to the alternative labor market. The argument suggests that the effect of farm dependence is moderated by the concentration of nonfarm labor opportunities.
Second, where (or when) agriculture and manufacturing are mutually dependent, farming families supplement household income with employment in the manufacturing industry, and the farm sector provides a market for manufactured goods. In this scenario, population growth, total or farm, is maintained or at least stabilized by the economic contributions of the manufacturing industry to the farm family and industry. Manufacturing, then, is expected to positively contribute to population growth by moderating the influence of farm dependence.
In either scenario, the stabilizing effect on the total population is predicated on the nonfarm market being located within the same area as the farmer. Under conditions in which farm income is not competitive with wage labor and a wage labor market exists elsewhere, the total population will be affected due to out- migration from the farm sector and the farm-dependent community to, presumably, the nonfarm sector and the non-farmdependent community. According to the mechanization thesis, farm dependence should be positively associated with county population change prior to mechanization. Once mechanization is refined and adoption is widespread, however, the farm industry is transformed, and the relationship between farm dependence and population change is altered; it is no longer positive, but should become negative. Of course, additional factors, such as agricultural restructuring and trade relationships, can also disrupt the relationship between economic base and population size through direct and indirect impacts on the key factors of the conceptual model. Regardless, the theoretical framework emphasizes that economic base, technological innovation, and environmental conditions are the main factors through which population changes. The second perspective, hereafter referred to as the industry complex thesis, suggests that alternative employment will moderate the influence of farm population on population change.
There are several methodological limitations of previous research on this topic that motivate the current analysis. First, the quantitative studies have investigated only the post-mechanization period and not the pre-mechanization period. This is problematic because the focus of the inquiry is motivated by a theorized transition between the two periods. Second, although the qualitative studies considered the pre- and postmechanization period, this research did not explicitly address the relationship between farm dependence and population change. There is no measure of the direction or magnitude of the influence of farm dependence. Third, previous studies have consistently identified the tractor as the major force that most dramatically mechanized the agricultural industry, yet they have not measured mechanization directly. Finally, earlier studies have not accounted for the underlying spatial autocorrelation that threatens to bias statistical estimation. Failure to account for spatial autocorrelation has substantive consequences: study findings used to inform theories of population change may be based on incorrect estimation. This study makes important theoretical and technical contributions by addressing each of these limitations.
DATA, MEASUREMENT, AND METHODS
Defining the Great Plains
The long-rooted reliance on the agricultural industry makes the Great Plains a unique and well-suited case to examine the theorized changing relationship between farm dependence and population change. Other industries were present in the Great Plains prior to and during the twentieth century (e.g., mining and manufacturing), yet farming was a major catalyst for the region’s settlement during the late 1800s and early 1900s, and much of the region is heavily embedded in farming even in recent decades. Moreover, discussions of contemporary population loss in the Great Plains suggest that dependence on the farm industry is the major contributor.
The study sample consists of 836 counties that are aggregated to the 1900 county boundaries to allow for comparable units of analysis across all decades under study. This aggregation yields 741 county clusters. My definition is more inclusive than those used in other studies of the Great Plains and is based on the U.S. Geological Survey definition that relies on area grassland and semi-arid status.2 Data are drawn from three sources. The main source is Historical, Demographic, Economic, and Social Data: The United States, 1790-1970, made available by the Inter-University Consortium for Political and Social Research (ICPSR 1976). The data were supplemented with published population and agricultural censuses to gain complete information for all decades between 1900 and 2000, as well as environmental data made available through the Vegetation/ Ecosystem Modeling and Analysis Project (VEMAP), provided by the VEMAP data group within the Ecosystem Dynamics and the Atmosphere Section, Climate and Global Dynamics Division, National Center for Atmospheric Research (Kittel et al. 1997; Schimel et al. 2000).
Counties are the selected unit of analysis because they are governmental units that unify the populations within their boundaries. They also are units of geography rich with social and economic data that are reasonably consistent over time. Government taxes and programs involving social, economic, and political institutions operate at the county level. County borders, however, change over time, and they tend to change for political reasons mainly associated with population size. Using a template developed by Horan, Hargis, and Killian (1989), each county is converted into its 1900 form and given a unique county cluster code, producing 741 county clusters for analysis according to the 1900 boundaries.3 Some county boundaries do not change, but others are dramatically different today than they were in 1900. For example, most of the county boundaries in Iowa have not changed since 1900, whereas almost every county in Oklahoma has changed boundaries. In fact, in 1900, Oklahoma had not yet become a state and was largely considered “Indian Territory.” The southern and northeastern parts of the state were divided between two sizeable areas, and smaller county divisions were made in the northwestern part of the state. The result is a few relatively large county clusters in the analysis.
County (or county cluster) population change is the dependent variable and is measured as the natural log [1 + ((Pt + 1 - P t) / Pt)], where Pt is the population at Time 1 and Pt + 1 is the population at Time 2, and 1 is added to correct for negative growth. This correction permits the use of the natural log to normalize the distribution of the outcome variable. Overall population change, rather than net migration, is selected as the dependent variable for three worthwhile reasons. First, measures of the separate components of population growth (deaths, births, and migration) are not consistently available at the county level until after 1950. Second, overall population change is consistent with previous work concerning the relationship between farm dependence and population change in the Great Plains during the post-mechanization period. Therefore, results from the current analysis speak directly to those obtained in previous studies by using a consistent outcome measure. Third, the theoretical model considers overall population size, or change in size, rather than change due specifically to migration. The population is perceived as an organizational unit rather than as a mere collection of individuals (Hawley 1986:6). Rather than addressing the individual and her migration behaviors, the present focus centers on the social conditions under which a given behavior might occur.
The key predictor variable is farm dependence and is measured as the proportion of the county population that resides on a farm at the first decade of the period under analysis. 4 Estimates for the farm population are culled from various sources, and a consistent measure was not available in all years. For example, complete data was not available for all counties in 1900, 1910, or 1920. Data from IPUMS (Ruggles et al. 2004) was used to create a state average farm household size that was then multiplied by the number of farms in each county. Estimates for the farm population in 1930-1960 were taken from Historical, Demographic, Economic, and Social Data: The United States, 1790-1970 (ICPSR 1976), and estimates for 1970-1990 were accessed through the GeoLytics CensusCD for the respective years (GeoLytics, Inc. 2004).
The measure treats farm dependences as a static rather than a dynamic variable because the research question, as informed by the theses, concerns the change in the relationship between the level of farm dependence and the change in population. The relationship is anticipated to change as the farming industry experiences widespread mechanization.
A key independent variable is mechanization and is measured as the number of wheeled tractors per farm in 1940 and the following decades (it was not consistently available until this year). The variable is included to account for the direct influence of technology and mechanization on population change. Analyses were conducted with and without the measure of mechanization to maintain model comparability. No substantive differences, with regard to the other covariates, were found, and the reported results include the measure of mechanization.
Two industry alternatives are explicitly considered in the analysis: manufacturing and livestock production.5 The proportion employed in manufacturing is included in the analysis as a measure of nonagricultural labor market alternatives. A consistent measure at the county level was not available for the entire study period, but some measure is available in all decades except 1910. The measure of proportion employed in manufacturing is derived from the approximated total number of reported wage earners in manufacturing divided by the county population in 1900, 1920, and 1930 (ICPSR 1976). For 1940 through 1970, data on the reported proportion employed in manufacturing was made available through the City and County Data Book Consolidated File, 1947-1977 (U.S. Department of Commerce 1978). Finally, data for 1980 and 1990 were accessed through the GeoLytics CensusCD (GeoLytics, Inc. 2004).
Grazing is the most extensive land use in the plains (Cunfer 2005:68), and livestock production has had a long and important role in the development of the Great Plains. In his agricultural and environmental study of the Great Plains, Cunfer (2005) described the steady profitability of cattle production on lands where crop production was impossible. Livestock production is measured in the current analysis as the total value of domestic animals, poultry, and bees in thousands of dollars.
Additional controls for population growth are included in the analysis, although they are given little attention in the discussion. A measure of sex ratio (the excess male ratio), whether the county contains a city of 10,000 or more, population density (initial population divided by county acreage), and a binary measure of county settlement date (whether the first year in which the county appears in the decennial census is before 1880) are included to address natural increase, growth potential, and urbanization. The possible influence of environmental conditions is measured as temperature range (the difference between the January and July monthly average in degrees Fahrenheit) and total annual precipitation (in inches).
I begin the exploration of variation in population change with an assessment of average growth rates, or percent population change, across the century. Emphasis is placed on the temporal aspect of variation in patterns of population change, and the county averages are compared to national figures in order to situate the Great Plains’ level of change in a broader context.
Several spatially oriented methods of analysis are also employed. Researchers must be sensitive to potential underlying spatial relationships or spatial autocorrelation when analyzing georeferenced data (Anselin 1988; Cliff and Ord 1973, 1981). Spatial autocorrelation arises because location proximity is typically accompanied by value similarity. The spatial dynamics of the dependent variable are explored through GIS and Moran’s I statistics (Moran 1950) using the software packages ArcGIS, GeoDa, and SpaceStat. These approaches are descriptive and enable the analyst to visualize the data in a spatial context and, thus, give the researcher some insight into the nature of the spatial process within the data, including the extent to which the spatial distribution of the dependent variable is reflected in the distribution of the independent variables. Finally, hypothesis testing about the relationship between population change and farm dependence is conducted through spatial lag regression using GeoDa and SpaceStat. Spatial regression is necessary when spatial autocorrelation is found in the error structure of the data because assumptions of independence are violated (see, e.g., Cliff and Ord 1973, 1981).6
Variation in Population Change
I address how patterns of population change vary, referring to counties as having either declined, grown, or remained stable in each decade. A county cluster is considered stable if the difference between two time points shows no increase or decrease by more than 5% of its value in the preceding decade. The classifications, reported in Table 1, are accompanied by regional and national growth rates in order to place the patterns of population change in a more general context.
The distribution of growth in the Great Plains was unstable during the years before mechanization. The majority or near majority of the region’s counties grew during the first two decades of the twentieth century, with about 59% and 46% experiencing growth in 1900-1910 and 1910-1920, respectively. Among the growing counties, more than 50% increased by over 10% in 1900-1910 and 34% in 1910- 1920, indicating that the high concentration of growth during these two decades is not due to a low-level increase. Yet these decades of considerable growth were principally followed by periods of decline. An equal proportion of counties remained stable as grew in 1920- 1930, but an estimated 45% of all Great Plains counties lost more than 5% of their populations between 1930 and 1940. A likely culprit for the noted decline of the 1930s is the devastating combination of the Dust Bowl and the Great Depression (Gregory 1989; Gutmann and Cunfer 1999). During the post-mechanization period-the years following 1940-around half of all counties declined, with the exception of only two decades. More counties gained population than lost during the 1970s and 1990s (more than 40% grew, whereas 21%- 22% declined in each of these decades). This pattern is consistent with discussions regarding the population turnaround of the 1970s (Fuguitt 1985; Long and DeAre 1988; Wardwell and Gilchrist 1980) and increased manufacturing and energy developments during the 1990s (McGranahan 1998; Murdock, Leistritz, and Schriner 1980; see also Johnson et al. , who found modest in-migration of middle-aged and older adults among farm-dependent counties in the 1990s). Neither of these growth-producing industries point to farm dependence as a contributor to positive growth. Further, studies of nonmetropolitan population change in more recent decades shows that areas that are dependent on agriculture have suffered population loss, whereas nonmetropolitan counties offering recreational amenities have experienced in-migration and overall population growth (Beale and Johnson 1998). The Great Plains’ nonmetropolitan counties tend to be rich in farm dependence and poor in nationally perceived recreational amenities.
Garnered from these distributions are general patterns that involve distributional reversals within as well as between periods demarcated by mechanization. The pre-mechanization period began with most counties experiencing growth, but ended with most counties suffering population loss. The opposite is true for the post- mechanization period; the beginning found most counties losing population, whereas growth was the modal pattern of change at the period’s end.
The pattern of population change in the Great Plains can be placed in context by comparing the region’s percent change in population to that of the United States. As evidenced in Table 1, the growth rates for the Great Plains are generally less consistent with the rates observed for the United States in the early part of the century relative to the degree of correspondence observed toward the end of the 100-year period, especially the final three decades. Researchers have suggested that growth rates for smaller, more rural geographic units become increasingly similar to national rates in later periods as economic structures become more similar (Bender 1980). In the later periods, economic patterns in rural areas became more closely linked to national economic activity, and, of central focus in the present study, economic activity is presumably associated with an area’s population growth.
The Great Plains’ growth rates exceed the United States’ in 1900- 1910 and 1920-1930 by 6%. The magnitude of the growth over these 10- year periods is substantial. The first decade corresponds with westward expansion, and the second follows World War I and accompanies continued seed strain advancements. Yet between 1910 and 1920, during the war and dismal environmental conditions, the Great Plains grew by 3% less than the United States overall. Harsh weather conditions and the accompanying depressed farm production has been highlighted in discussions of population change during the 1910s. North Dakota experienced drought conditions in 1917, while Nebraska suffered a severe winter between 1916 and 1917. Both of these states, in addition to South Dakota, experienced poor production in 1918 and 1919 (Ottoson et al. 1966).7
Between 1950 and 1960, the United States experienced the largest population increase since the beginning of the century (19%). Relative to the adjacent decades, the Great Plains also experienced a notable population increase (12% versus 6%-8%). The overall growth, however, was not as dramatic as that observed for the United States as a whole. This was a decade of marked suburban growth for the United States in general (Edmonston and Guterbock 1984; Guest 1978; Schnore 1962). The divergence between trends in the Great Plains and overall United States suggests that suburbanization may have occurred within the Great Plains, although to a lesser degree or at a later date than other subregions of the country. Toward the end of the century, especially by 1970, the growth rates between the Great Plains and the nation became more similar, deviating from each other by only three points at most. Importantly, these deviations typically favored the Great Plains; the region experienced a slightly higher percentage increase in population relative to the entire United States. For example, between 1990 and 2000, Great Plains counties grew by 16%, while the U.S. population increased by 13%.
Spatial Variation in Population Change
The spatial distribution of the growth patterns described above is illustrated in Figure 1 for three theoretically relevant decades: 1900-1910, 1940-1950, and 1990-2000. The lightest shade indicates the most extreme loss, whereas the darkest shade indicates the most extreme growth. Without the aid of any formal spatial statistic, one can see spatial patterns in the distribution of growth and decline in each of the decades. These maps also reveal how the distribution changes over time; the spatial patterns are not static.
In 1900-1910, growth was concentrated in the western half of the region, indicating the extent of continued westward expansion in the first decade of the twentieth century. In contrast, in 1940-1950, the region was riddled with decline, although a few pockets of growth emerged on the easternmost and westernmost boundaries. There was greater loss than growth, and the east-west demarcation of the early part of the century was replaced by a central-boundary division by the first decade of the post-mechanization period. Finally, in 1990-2000, decline was almost exclusively isolated to the center of the region. A central-boundary division persisted where growth was found among the counties on the fringe of the Great Plains, although at the opposite angle of that observed from 1940 to 1950. Moreover, pockets or clusters of growth and loss are evident in 1990-2000, whereas a large-scale pattern is observed in 1900- 1910. The type of spatial patterning is noticeably different across the century.
The visualized patterns can be reduced to a summary statistic that indicates the extent and direction of spatial autocorrelation. The Moran’s I statistic, reported in Tables 2 and 3, is one such statistic and can be considered a spatial counterpart to the nonspatial Pearson’s correlation coefficient.8
The Moran’s I (univariate) statistics for population change among the Great Plains counties reveal two important features of the spatial relationship in the data. First, a positive correlation is found in each decade. This indicates that counties in close spatial proximity share similar patterns of population change relative to more distant counties. Second, the magnitude of the positive correlation changes over time. For example, in 1900-1910, a 1- standard-deviation difference in population change corresponded with a 0.67-standarddeviation difference in adjacent counties. In contrast, a 1-standard-deviation difference corresponded with a 0.18- standard-deviation difference in population change in 1950-1960. The degree of spatial autocorrelation in the data was very high at the beginning of the twentieth century, fell to a nearly negligible level during the two decades following World War II, and then increased during the remaining three decades of the century.
The fluctuations in the strength of spatial autocorrelation coincide with important historical events. The diminished spatial relationship emerges at a point in the historical context where the Great Plains and United States experienced widespread, intense urbanization and mechanization. The spatial relationship strengthens during the decades of the population turnaround and farm crisis; these instances of growth and loss are more spatially concentrated. The results show that in addition to variation in the distribution of the spatial patterns, there is also variation in the strength of the spatial relationships during the century. (See Johnson et al.  for age-specific visualizations during the post-1950 decades.)
Farm Dependence and Population Change
Analysis of the relationship between farm dependence and population change over the twentieth century is motivated by previous research suggesting that agricultural mechanization occurring around 1940, specifically the widespread adoption of tractors, altered the relationship between farm dependence and population change (Albrecht 1986; Cunfer 2005; Grant 2002; National Agricultural Statistics Service 2003; Rathge and Highman 1998); mechanization presumably shifted the relationship from positive to negative in the post-1940 period. In contrast, the industry complex thesis anticipates that nonfarm employment, specifically manufacturing, moderates the association between farm dependence and population change.
Several iterations of modeling were conducted for each decade to arrive at the results reported in Tables 2 and 3. The iterations consisted of a series of sequential models estimated through standard OLS and spatial regression techniques that examined (1) the bivariate relationship between farm dependence and population change, (2) this plus the covariates, (3) a geographic trend surface only (based on the longitude and latitude of the county cluster centroid), (4) a multivariate model plus the geographic trend surface, (5) a multivariate model using the detrended residuals as the outcome variable, and (6) a geographically weighted regression analysis testing for structural stability in the parameters. These iterations were used to address the underlying spatial heterogeneity in the data before turning to methods appropriate for the analysis of spatial “dependence.” Ultimately, the multivariate model was analyzed using spatial regression techniques. Spatial lag regression is used for the analysis of the pre-mechanization and post- mechanization periods and suggests a spatial spillover effect (i.e., population change in a county affects population change in its neighboring counties). Spatial lag models attribute spatial autocorrelation to a small-scale, interactive association, in which a social process occurring in one county influences the same social process in neighboring counties. This modeling approach satisfies the assumption of independent errors and includes an additional parameter that captures the strength of the spatial effect.9 Discussion of the results focuses on the key relationship between farm dependence and population change, with emphasis on the covariates when the association is of direct relevance to the central theoretical relationship.
Pre-mechanization. The findings in Table 2 show that, contrary to the mechanization thesis, the association between farm dependence and county population change is not positive across the pre- mechanization era. Instead, the relationship is most accurately characterized as negative and unstable; the direction of the association is consistently negative throughout the period, and the associations are statistically significant at the beginning and end of the pre-mechanization era.
The relationship between farm dependence and population change is negative during the first decade of the 1900s, a decade when the agricultural industry was rapidly expanding and a time of massive settlement on the Great Plains. Between 1900 and 1910, nearly 60% of the counties experienced growth, with 52% growing by more than 10% (see Table 1 and Figure 1). A review of the distribution of the farm population (the measure of farm dependence) in 1900 reveals a heavy concentration in the easternmost counties and a low distribution among the western counties, illustrated in Figure 2. The negative association between farm dependence and population change at the beginning of the century is due to the very low concentration of a farm population, or any population, in the westernmost counties which experienced the greatest amount of positive population change between 1900 and 1910.
The reported unstandardized beta coefficient for the farm population in 1900 indicates that counties with average farm dependence experienced an estimated 46% change in population. By comparison, counties with a 1-standard-deviation increase in farm dependence changed by almost 32%, whereas counties with a 1- standard-deviation decrease in farm dependence changed by 62%. There is a 30-percentage-point difference between the two county types, and counties with less farm dependence have the growth advantage.
Contrary to previous postulations, the relationship between population change and farm dependence is nonsignificant (and negative) between 1910 and 1930. Poor environmental conditions have been identified as a potential factor that negatively affected farming and population growth during the 1910s (Ottoson et al. 1966). The results imply that temperature range negatively influenced population change, yet it does not attenuate the relationship between farm dependence and population change because there is no association to explain.
The 1920s is an interesting decade with historical circumstances that both impeded and promoted population growth in the Great Plains through farm dependence. Among the factors that impeded growth is the contracting agricultural industry during the mid- to late-1920s that preceded other industries in the depression. The factors that promoted growth include the agricultural industry profiting from World War I during the early years of the decade, many young men motivated to return to the family farm or establish their own by government incentives and a burgeoning national and international demand for U.S. crops, and the economic downturn of nonfarm industries that encouraged the “return to the land” movement-people living in more metropolitan places moved to rural areas in an effort to eke out an existence. While this movement took hold in the 1930s, there is evidence of rural-bound movement in the late 1920s (Woofter 1936). The influence of farm dependence in this decade is attenuated by manufacturing and livestock production (although these associations, in turn, are attenuated by measures of growth potential).
By the close of the pre-mechanization era, the relationship between farm dependence and population change was statistically significant. At this point in U.S. history, the Great Depression was under way, and counties with greater farm dependence were more likely to suffer population loss. Even though the Dust Bowl was regionally specific (Gutmann and Cunfer 1999) and the Depression had far-reaching impacts on all industries and regions, study results suggest that counties with high farm dependence were more hard-hit than those with less farm dependence.
A closer look at farm dependence. The results of the pre- mechanization era do not support the mechanization thesis and suggest that the framework needs careful revision. There is no evidence of a positive association between the level of farm dependence and population change during the pre-mechanization period. Contrary results, however, are yielded by an analysis of the change in farm dependence. Analysis of the relationship between change in farm dependence and change in population does not necessarily contradict the mechanization thesis and is, in fact, motivated by the theoretical argument. An emphasis on the change in farm dependence can be inferred from the thesis even though prior research has concentrated analytical focus on the level of farm dependence. The plains experienced growth in the farming industry, which created a demand for more farm labor, which then contributed to population growth.
In the bivariate context, a positive correlation between change in farm population and change in population is evidenced between 1900 and 1930, as reported in Table 4. The general pattern holds in the multivariate context, also reported in Table 4.10 Counties with an increase in farm dependence were more likely to experience population growth from 1900 to 1930. For example, in 1900-1910, there is a 72-percentage-point difference between counties with plus or minus 1 standard deviation in the change in farm dependence, and, in contrast to the results analyzing level of farm dependence, counties with a greater increase in farm dependence have the growth advantage. The mechanization thesis is supported when it is modified to address the relationship between changes in farm dependence and population change, not the level of farm dependence. Results also suggest that 1930, rather than 1940, was the transitional year for the relationship between farm dependence and population change.
Results from the analysis of the pre-mechanization period provide moderate support for the industry complex thesis. The industry complex thesis anticipates that nonfarm opportunities moderate the association between farm dependence and population change; the influence of farm dependence will be weaker in counties that also contain more nonfarm opportunities. Tests for an interactive association between manufacturing and farm dependence were conducted in all decades. Manufacturing has no direct association with population change (see Tables 2 and 3), yet, in some years, manufacturing moderates the negative influence of farm dependence. The moderating influence, however, is not in the hypothesized direction in each of the decades.
The estimated percentage change in population is reported in Figure 3 for the four decades in which a significant interactive association is found. The values reflect the calculated population change for counties with an average level of farm dependence and varying levels of manufacturing. The anticipated direction of the interactive association is found in only two of the four decades; the negative influence of farm dependence is weaker in counties with a higher level of manufacturing in 1920-1930 and 1960-1970, but the negative influence of farm dependence is aggravated in 1900-1910 and 1980-1990.
Review of the bivariate associations reveals an important shift in the relationships between the industry variables and population change and sheds light on the unstable conditioned influence of farm dependence, as reported in Table 5. Prior to the 1930-1940 decade, all industry variables were negatively associated with population change and positively related to one another. Counties with higher farm population were more likely to have higher livestock production and a larger proportion of the population employed in manufacturing. Yet counties with a stronger industry presence were more likely to lose population. The directions of the relationships are consistent with the earlier discussion of the location of development, the patterns of westward expansion, and the implications of analyzing levels rather than changes in industry.
A transition in the associations emerged in 1930. First, farm population (level, not change) remains negatively correlated with population change, yet manufacturing becomes positively correlated with population change and negatively correlated with farm population. Counties with higher manufacturing dependence were more likely to have lower farm dependence and greater population growth in and after 1930. Second, livestock production is no longer correlated with population change or manufacturing in 1930, and it is negatively associated with farm population. The results indicate that counties with a larger presence in any industry were likely to have a greater presence of other industries until 1930, at which point industry specialization emerged. After 1930, counties with greater farm dependence were less likely to engage in nonfarm industries, and vice versa. Manufacturing weakened the negative influence of farm dependence on population growth when manufacturing was positively associated with growth. The unanticipated direction observed in the 1980s requires further elaboration. The farm crisis was in effect, yet greater population loss is estimated for counties with greater nonfarm opportunities, perhaps because the type of manufacturing was dependent on farm production or centered in slow- growth or declining industries (e.g., Bernat 1994). Post- mechanization. Previous research has found a stable, statistically significant, negative relationship between farm dependence and population change during the postmechanization period. The results from this analysis show that, indeed, farm dependence is negatively associated with population change throughout the post-mechanization period (1940-2000). Consistent with prior findings, counties with high farm dependence appear significantly less likely to experience population growth during the post-mechanization era. Yet these results also indicate that the relationship changed somewhat over time. A stronger influence is noted in 1950-1960 relative to the other decades, and a nonsignificant association is observed for 1960- 1970 (see Table 3). The slackening of the negative influence of farm dependence on population change starting in the 1960s may reflect the beginning of mass suburbanization, in which many previously farm- dependent counties were settled by suburbanites.
A negative and statistically significant relationship between farm dependence and population change is found for the first two decades, between 1940 and 1960. This is a period of intense urbanization, mechanization, agricultural “corporatization” or commercialization, and increased foreign competition (Grant 2002; Lobao and Meyer 2001). Each of these factors would have a negative impact on the farming industry. During these decades, counties with greater farm dependence were at a population growth disadvantage compared to counties with less farm dependence. For example, in 1940- 1950, the reported unstandardized beta coefficient for the proportion employed on farms indicates that counties with average farm dependence experienced an estimated 3% loss in population, and counties with a 1-standard-deviation higher farm dependence declined by 11%, whereas counties with a 1-standard-deviation lower farm dependence grew by 6%.
A statistically nonsignificant relationship is found for the 1960s, but a relationship reemerges in the 1970s and persists throughout the remaining decades of the postmechanization era. The influence of farm dependence on population change for 1960-1970 is attenuated by covariates aimed to capture growth potential, including the presence of a city and settlement date. Competing forces, in terms of population dynamics, were at play in the remaining decades and included the rural renaissance of the 1970s and the farm crisis, beginning in the 1970s and taking hold in the 1980s. Both events put farm-dependent counties at a disadvantage; the population turnaround was generally experienced by nonfarm- dependent counties (Albrecht 1986, 1993), and the farm crisis differentially impacted farm-dependent counties. The implications of the farm crisis extended into the 1990s, and counties with greater farm dependence continued to be at a population disadvantage relative to counties with less farm dependence.
A closer look at mechanization. A direct measure of mechanization was omitted in earlier studies but included in the current analysis. The results show that the presence of tractors was positively associated with population change for the first half of the post- mechanization period. The findings are in direct contrast with the mechanization thesis, which argues that greater mechanization leads to population decline by creating a lower demand for labor. Analysis of the bivariate associations reveals a negative and statistically significant association between tractors and farm population in 1940- 1950 only (results not reported). The correlation is positive, although somewhat weak (p
Mechanization itself did not attenuate the influence of farm dependence, nor did it have much of a direct impact on population change in the final three decades of the post-mechanization era. Indeed, for 1990-2000, there is an estimated 2-percentage-point difference in the estimated population change for counties with plus or minus 1 standard deviation in the number of tractors (4% versus 6%, respectively). No difference is observed for 1970-1980 or 1980- 1990.
SUMMARY AND DISCUSSION
The study reveals mixed support for the mechanization thesis and the industry complex thesis. I suggested at the outset of the analysis that looking back on the agricultural transition would inform our theoretical understanding of long-term and large-scale U.S. urbanization patterns of the twentieth century. Results from the analysis of levels of farm dependence reveal that the transition was not a simple process, as characterized in earlier research. Farm- dependent counties generally lost more population than less-farm- dependent counties during the post-mechanization era, and farm dependence was not necessarily advantageous in the pre- mechanization years. The findings suggest that the proposed positive- to-negative shift between the pre- and post-mechanization periods is overly simplistic and requires reconsideration. A single shift in the nature of the relationship between economic base and population change that accompanied the introduction of technological innovation is not observed.
Partial support is found for the mechanization thesis when the influence of farm dependence is measured in terms of change rather than level. A positive association is observed in 1900-1910 and 1910- 1920 and suggests that increases in farming contributed to the westward expansion characterizing the early decades of the twentieth century, and, also consistent with the mechanization thesis, the association becomes negative in the later decades. However, the 1930s, rather than the 1940s, emerges as the pivotal decade. Only partial support for the thesis is garnered, however, since the influence of tractors was positive during the early decades of the so-called post-mechanization period. The mechanization thesis anticipates a negative association; mechanization presumably replaces the demand for farm labor, which, in turn, reduces population.
There is mixed support for the industry complex thesis. The premise of the industry complex thesis is that the associations are temporally and regionally dependent. Indeed, the nature of the relationship between farm and nonfarm industries shifted during the twentieth century, as did the moderating influence of manufacturing on the association between farm dependence and population change. Farm and nonfarm industries were positively associated with one another and negatively associated with growth on the plains during the pre-mechanization period; where one industry existed, others were likely to coexist. But the direction changed in 1930. Counties with greater farm dependence were less likely to have a strong manufacturing or livestock presence in the later decades, and manufacturing became positively associated with growth in 1930. In contrast to the Midwest region of the nineteenth century, the twentieth-century Great Plains can be characterized as a region with industry specialization rather than cooperation.
The finding that the level of farm dependence negatively influences population change throughout the twentieth century is important for demographic theory of urbanization and population growth and charged debates concerning settlement patterns and community development. Some claim that the settlement of the Great Plains is one of the nation’s greatest mistakes, measured in federal and private dollars as well as individual and community heartbreak (e.g., Kristof 2002).
Study findings encourage future research on the influence of state policy on population growth, including the role of economic subsidies for agricultural producers and incentives for nonagricultural industry development. The state has a long-standing presence in the development of the Great Plains. In the current period, senators from states with disproportionate population loss, a large proportion of which are plains states, have been summoned to defend the survival of their constituents through the New Homestead Act. The act would provide incentives to individuals and businesses willing to live in underpopulated areas of the United States. In contrast to the original Homestead Act, the incentives are centered on tax relief rather than land acquisition. The intention of the contemporary iteration, however, is the same: population growth.
*Katherine J. Curtis White, Department of Rural Sociology, 350 Agricultural Hall, 1450 Linden Drive, University of Wisconsin- Madison, Madison, WI 53706; E-mail: firstname.lastname@example.org. This work was supported by Center Grant P30 HD05876 and Training Grant T32 HD07014 from the National Institute for Child Health and Human Development, National Institutes of Health at the Center for Demography and Ecology, University of Wisconsin-Madison. The author acknowledges analytical and technical assistance from Nick Fisher, David D. Long, Paul R. Voss, and Jeremy White, in addition to substantive guidance from Kyle D. Crowder, Glenn V. Fuguitt, Jess Gilbert, Gary P. Green, Avery M. Guest, Jerald R. Herting, Stewart E. Tolnay, and Halliman H. Winsborough at various stages of development, and the helpful comments of the Demography editors and anonymous reviewers. 1. An additional counter-thesis to the technological argument was posited by Mann and Dickinson (1978; see also Mann 1990). According to this perspective, it is not technology per se that explains continued family farm production, but it is the “logic” of capitalism-the gap between production time and labor time. Family farms are able to survive within a capitalist system because the unpredictability and biological rhythms of nature are difficult to synchronize with industrial production and, therefore, the gap between production time and labor time persists in agricultural production. When the conditions of production are sufficiently altered by advances in technology, however, capitalism will move in and conquer agriculture. The implication for population is that until capitalism dominates agriculture, more family farms will persist and contribute to population stability, if not growth. While germane to the current topic, an adequate test of the Mann- Dickinson thesis would require a different level and scope of analysis than that used in this study (i.e., comparison of commodity labor requirements).
2. There is no consistent geographical definition of the Great Plains among historians, geographers, demographers, or sociologists. Inclusion ranges from all counties within six U.S. states to select counties within 13 states plus Canadian and Mexican territories (see, e.g., Adcock 1995; Albrecht 1986; Cromartie 1998; Gutmann et. al 1998). Yet all scholars agree that the Great Plains is distinct from its surrounding areas by its semi-arid quality; some years it is dry and desert-like, while in other years it is very wet, with great unpredictability. The sample for this study follows the U.S. Geological Survey definition. I have selected a broader definition than some studies of the Great Plains in an attempt to remain as inclusive as possible. While places within the easternmost states may receive more than 20 inches of rainfall, evaporation maintains semi-arid conditions among the western counties within these states (Kraenzel 1955).
3. Use of the template has the following implications: if counties A and B are not involved in the formation of a new county, there would be no change in the county code, and each would be assigned a separate county cluster code. But if a portion of county A split off to produce county B, the two would share the same county cluster code. Similarly, if county B merged with county A, the two would also have the same county cluster code.
4. This measure of farm dependence differs from other studies. The measure constructed and used by the Economic Research Service (ERS) is based on the proportion of county income that is based in farming (see Cook and Mizer 1994). Other researchers have used the proportion employed in farming as a measure of farm dependence (e.g., Albrecht 1986). This measure was explored in the present analysis but was not used given limited availability of industry- specific employment data until 1950. Analysis was conducted using the number of farms per capita, and the results varied somewhat. For example, a positive association was found in 1900-1910 and 1920- 1930. The contrary findings may be reconciled by the potential measurement error in the per capita measure of farms.
5. Two measures of crop production were considered but not included because both were highly collinear with farm dependence. This is intuitive given that farm dependence (measured as population or employment) suggests crop production, not grazing or livestock production. Measures for crop production that were considered include acres of cropland and the value of crops (per $1,000).
6. The employed spatial methods require the use of a weights matrix that defines the nature of the spatial relationship between counties by specifying which neighboring areas are assumed to affect dynamics in each given county cluster. A first-order, row- standardized queen matrix was selected for this analysis. This is a contiguity matrix that includes information on the counties neighboring each county. Its selection is based on previous research demonstrating that the spatial extent of population growth patterns reaches about 40 miles (Wheeler 2001). In the Great Plains, 40 miles captures the first neighbor. Row standardization guarantees convergence in the iterative MLE process used in spatial regression analysis.
7. This explanation is the only one found in the previous literature reviewed when preparing this manuscript. Although undocumented, alternative explanations, including the disproportionate involvement of Great Plains farm men in the war effort, could contribute to the comparatively low growth of the 1910s.
8. The Moran’s I statistic can range in value from -1 to 1, where 0 indicates no relationship and, therefore, no correlation between dynamics in neighboring county clusters. In contrast, a negative value suggests negative spatial autocorrelation, and a positive value indicates positive spatial autocorrelation. Positive autocorrelation is the most common form because observations tend to be similar to neighboring observations (Tobler 1970).
9. The spatial lag regression is represented in matrix notation as y = (I – nW)-1 Xa + (I – nW) – 1aa. The spatial error regression, a standard alternative spatial dependence model, is represented in matrix notation as y = Xa + (I – nW)-1 aa. Sensitivity analyses of both spatial error and spatial lag models were estimated for all decades. The results do not dramatically change, but the conceptual implications of the spatial lag suit the current subject; population change in one area lik