Preferences for Redistribution and Economic Mobility Within Generations in the United States and Great Britain,

Transcription

1 Preferences for Redistribution and Economic Mobility Within Generations in the United States and Great Britain, Oriol Pons-Benaiges Stanford University This Version: October 28, 2017 Latest Version: web.stanford.edu/ oriol/mobility.pdf Abstract The United States stands out among the world s largest and richest countries for its high income inequality after redistribution. Previous research has associated the apparent preference for low redistribution in the United States with exceptional socioeconomic circumstances in the end of the 19th century. Following a sample of workers across the 1850/51 and 1880/81 censuses, this paper measures economic mobility within generations in late 19th century United States and Great Britain. Different measures of economic mobility indicate that the United States had lower economic mobility within generations than Great Britain, but this fact was driven entirely by farmers. For non-farmers, economic mobility within generations was higher in the United States than in Great Britain. I would like to thank Ran Abramitzky for his advice and support. I would also like to thank Davide Malacrino, Gavin Wright, Fulya Ersoy, Önder Polat, Daniel Bennett, David Yang, Caroline Hoxby, Tim Bresnahan, and Kalina Manova for their helpful comments and suggestions. oriol@stanford.edu. Department of Economics, Stanford University, 579 Serra Mall, Stanford, CA

2 1 Introduction The United States has a level of income inequality before redistribution that is similar to that of other large rich countries. Table 1 contains descriptive statistics on income inequality for a sample of 20 countries with more than 5 million citizens and an income per capita of more than $25,000 (adjusted for purchasing power). Most countries in the sample, including the United States, have a Gini coefficient close to 50% before redistribution. After redistribution, however, the United States stands out for its high level of income inequality. Even in the small sample in Table 1, the United States is two standard deviations away from the mean, with a Gini coefficient of 38%. Only one country, Israel, shows similar levels of inequality after redistribution while most other countries in the sample have a Gini coefficient around 30%. Why do we observe a relatively high level of income inequality in the United States? While this is still an open question, one of the arguments that is often put forward is American Exceptionalism, that is, the idea that the United States is (or has been) qualitatively different from other nations. A range of alternative explanations exists, such as 1) the large size of the United States and the strength of its army, which makes social revolts difficult, 2) the heterogeneity of the population in the United States, which makes people less willing to redistribute, 3) the negative perception that voters in the United States have about government intervention, which tends to minimize government size, or 4) the bipartisan political system in the United States, which hampers the emergence of a socialist party. 1 Exceptional socioeconomic circumstances in the past can play an important role in guiding policy debates today. For example, the United States is often perceived as a country with high economic mobility, despite a growing evidence documenting that its economic mobility is similar to Europe s, or even lower. 2 Alesina, Di Tella, and MacCulloch (2004) provide survey evidence showing that Americans and Europeans react differently to income inequality, affecting their preferences for redistributive policies. The authors consider that their findings are consistent with the perception (not necessarily the reality) that Americans have been living in a mobile society, where individual effort can move people up and down the income ladder, while Europeans believe that they live in less mobile societies. Since the perception of exceptional circumstances can outlive the circumstances themselves in the form of a popular belief, it is important to understand them. 3 This paper attempts to measure the degree of American Exceptionalism by the end of the 19th century with respect to one dimension: economy mobility within generations, that is, the likelihood that a young worker will end up in a different economic status when he or she is old. 4 To do so, I compare 1 See Alesina, Glaeser, and Sacerdote (2001) for a study of economic, political, and behavioral factors that can explain the different preferences for redistribution observed in the United States and Europe. 2 For evidence supporting this claim, see Erikson and Goldthorpe (1992), Solon (2002), Ferrie (2005), Corak (2006), Beller and Hout (2006), Jäntti, Røed, Naylor, Björklund, Bratsberg, Raaum, Österbacka, and Eriksson (2006), d Addio (2007), Isaacs, Sawhill, and Haskins (2008), or OECD (2010). 3 By popular belief I mean an idea that is accepted by most people, regardless of the facts, but not necessarily by the experts. It is, in a way, similar to the conventional wisdom popularized by John K. Galbraith, which may apply to both experts and non-experts. 4 I use the terms economic mobility across generations and economic mobility within generations instead of the most common terms inter-generational economic mobility and intra-generational economic mobility to avoid confusion. 2

3 Country Income Population Gini Coefficient Before Redistribution After Redistribution Difference Australia 43, Austria 42, Belgium 37, Canada 43, Czech Republic 27, Denmark 37, Finland 35, France 35, Germany 40, Israel 34, Italy 30, Japan 36, Netherlands 41, Norway 54, South Korea 33, Spain 29, Sweden 41, Switzerland 46, United Kingdom 37, United States 53, Mean 39, Median 37, St. Dev. 7, Table 1: Descriptive statistics on income inequality for a sample of 20 large rich countries. Notes: Sample of 20 large rich countries with their income per capita, Gini coefficients before and after redistribution, and the difference between Gini coefficients. This sample includes all countries with more than 5 million citizens and more than $25,000 income per capita (adjusted for purchasing power) except Taiwan, Saudi Arabia, and the United Arab Emirates; not all necessary data on Gini coefficients was found for these countries. Income is measured in per capita US$ and population is measured in millions of people. Sources: Official national estimates (population in 2013 or 2014), the IMF (income per capita in 2013), and the OECD (Gini coefficients in 2010). 3

4 economic mobility within generations in the United States and Great Britain (England, Wales, and Scotland) between 1850/51 and 1880/81. The socioeconomic circumstances of the first half of the 20th century are of particular interest because, during that period, the United States was laying the foundations of its modern welfare state. The demands for legal protection for female workers and benefits for poor mothers that resulted in the Maternalist Reform, or the social programs of the Roosevelt administration to palliate the effects of the Great Depression, are examples of the wave of social activism and political reform that took place during the Progressive Era. 5 By focusing on the end of the 19th century, I hope to capture the economic mobility experienced or perceived by those who were alive in the beginning of the 20th century. However, it is outside the scope of this paper to understand the relationship between economic mobility within generations and preferences for redistribution. The choice of Great Britain as a comparison country for the United States is motivated by the availability of comparable historical census data on occupations. Fortunately, there are important historical similarities between the two countries since the industrialization of the United States, such as their legal traditions, property rights systems, sources of labor, capital, and technology, language, and culture. This makes Great Britain a reasonable reference point to asses the degree of American Exceptionalism by the end of the 19th century. 2 Related Literature This paper benefits from a large literature measuring the degree of American Exceptionalism with respect to social and economic mobility. Long and Ferrie (2013) is the closest contribution, comparing occupational mobility across generations in late 19th century United States and Great Britain. The authors track a sample of male workers across censuses and compare their occupations and the occupations of their fathers at a similar age. Long and Ferrie find that occupational mobility across generations was significantly higher in the United States than in Great Britain by the end of the 19th century, and suggest that this can explain current preferences for low redistribution in the United States. In particular, Long and Ferrie argue that policies in the United States reflect a belief that high rates of economic mobility leave little need for substantial redistribution by the state and that in the experience of those who created the United States welfare state in the 1930s, the United States had indeed been exceptional. Additionally, Long and Ferrie find that the United States and Great Britain had similar occupational mobility across generations by the end of the 20th century, and attribute this change to a decline in United States occupational mobility during the 20th century. This finding contradicts the predominant view among sociologists: relative social mobility has been constant or trendless in all industrialized countries during the 20th century. 6 This disagreement generated two responses to Long and Ferrie 5 See Wilkinson (1999) for a literature review on the Maternalist Reform. 6 Some studies supporting this claim are Featherman, Jones, and Hauser (1975), Hauser, Koffel, Travis, and Dickinson (1975), Grusky and Hauser (1984), Guest, Landale, and McCann (1989), Grusky and Fukumoto (1989), or Eriksson and Goldthorpe (1992). 4

5 (2013). Xie and Killewald (2013) argue that data quality is more limited than Long and Ferrie acknowledge, and that calculating odds ratios is not appropriate to measure the occupational mobility of farmers. After correcting for this, Xie and Killewald find that occupational mobility across generations did not decline in the United States during the 20th century. Hout and Guest (2013) argue that Long and Ferrie focus almost exclusively on exchange mobility, but that the actual experience of social mobility reflects both structural and exchange mobility. 7 In their analysis, Hout and Guest find very low exchange mobility in Great Britain by the end of the 19th century, and high overall mobility in the United States by the end of the 20th century. Most of the literature on economic mobility in the United States and Great Britain in the late 19th century studies these countries separately, due to the lack of appropriate data for direct comparisons. A notable exception is Kaelble (1986), which makes an effort to bring together research on social mobility in Europe and America during the 19th and 20th centuries. However, throughout the book Kaelble acknowledges important limitations of comparing studies that use different metrics and datasets, or that have different purposes in minds. Fortunately, recent efforts in digitizing the whole 1880 United States census and 1881 Great Britain census, as well as small samples of other historical censuses, allow us to begin making such comparisons. Long and Ferrie (2013) exemplify this new trend of empirical studies complementing an old qualitative literature that typically views 19th century United States as amoremobilesocietythangreatbritain. 8 Methodologically, this paper is related to the work by Abramitzky, Boustan, and Eriksson (2012), who study the self-selection and economic outcomes of immigrants moving from Norway to the United States. As part of their work, the authors estimate the occupational upgrading over the life cycle, which is a measure of upward occupational mobility within generations. To do so, they compute the proportion of unskilled, blue-collar workers in 1850 that remain unskilled, blue-collar workers in The authors find that only 18% of unskilled, blue-collar United States workers in the 1850 remained so in 1880, as opposed to 47% in Norway. This suggests that upward occupational mobility within generations, as measured by this metric, was significantly higher in the United States than in Norway in the end of the 19th century. In a later work, Abramitzky, Boustan, and Eriksson (2014) compare the economic outcomes of United States immigrants and natives in the late 19th century, and find that the average United States immigrant did not face a substantial earnings penalty upon arrival and experienced the same rate of occupational advancement as natives. Aconsiderablenumberofstudiesmakeinternationalcomparisonsofeconomicmobilityinthelate 20th century. 9 These studies benefit from the availability of accurate data on income, and avoid the inconveniences of using proxies like occupation. Most of the attention has been devoted to economic 7 Overall mobility can be separated into structural and exchange mobility. Structural mobility is caused by the fact that rows and columns in the mobility matrix have different marginal distributions. Exchange mobility is unrelated to differences in the marginal distributions of rows and columns in the mobility matrix, and can be calculated as a residual after controlling for structural mobility. 8 See Tocqueville (1835), Sombart (1906), or Marx (1972) for examples of this qualitative literature. 9 See Treiman and Ganzeboom (2000) for a survey of this literature. Economic mobility within generations in the late 20th century is often studied in conjunction with other variables of interest, such as health or education. For examples of this literature, see Blane, Harding, and Rosato (1999), Power, Manor, and Li (2002), Langenberg, Hardy, Kuh, Brunner, and Wadsworth (2003), Novak, Ahlgren, and Hammarstrom (2012) or Plewis and Bartley (2014). 5

6 mobility across generations, perhaps because it is closely related to the idea of equality of opportunity. Kerckhoff, Campbell and Winfield-Laird (1985) compare the United States and the United Kingdom (England, Wales, Scotland, and Northern Ireland), and find that economic mobility across generations is significantly higher in the United States, mainly because of higher structural mobility. Examining OECD countries, Causa and Johansson (2009) find that the United States and the United Kingdom are among the least economically mobile societies in their sample, followed by some southern European countries. This paper complements the existing literature on occupational mobility in late 19th century United States and Great Britain by bringing new evidence on economic mobility within generations. My analysis focuses on overall mobility, which is arguably closer to what workers experienced or perceived than exchange or structural mobility, which are useful but technical artifacts. 10 Social mobility in historical settings is often measured as changes in occupations, which are used as a proxy for income. To be able to measure upward and downward mobility, however, it is important to rank occupations is a meaningful way. In what follows, I will attempt to construct such a ranking of occupations using historical income scores associated to each occupation, and then compute statistics summarizing upward and downward economic mobility within generations. 3 Data Economic mobility within generations can be measured by following the income of a sample of workers throughout their lives. To do so for the United States and Great Britain in the end of the 19th century, IusedatafromtheNorthAtlanticPopulationProject. 11 This dataset contains digitized census records from Canada, Great Britain, Germany, Iceland, Norway, Sweden, and the United States, from 1801 to In the case of the United States, the entire 1880 census, 1% samples of the 1850, 1860, 1870, and 1910 censuses, and a 5% sample of the 1900 census are available. For Great Britain, the entire 1881 census and a 2% sample of the 1851 census are available. Unfortunately, old censuses do not provide personal identification numbers, making it difficult to follow individuals across censuses. To solve this problem, this paper uses a matching procedure to link workers from the 1850/51 censuses (when the workers are young) to the 1880/81 censuses (when the workers are old) using observable characteristics. I restrict my sample to men because women had no widespread right to vote until 1920, their labor force participation was very low in the end of the 19th century and beginning of the 20th, and the change of name resulting from marriage would complicate the matching procedure. The variables that I use to match workers from one census to another are their name and surname, age, gender, birthplace (state in the United States on county in Great Britain), and race (only for United States workers). These are all the variables in the dataset that should not change 10 The focus on overall mobility as opposed to exchange mobility is in line with one of the main criticisms that Hout and Guest (2013) make of Long and Ferrie (2013). 11 The North Atlantic Population Project data set can be found at 6

7 Country United States Great Britain Year Male 101,323 25,512, ,520 14,534,818 Right age range 12,300 1,150,385 21, ,815 Not Duplicated 11, ,739 18, ,791 No missing value 11, ,221 18, ,054 Table 2: Sample sizes at each stage before the matching. Notes: The first stage drops observations that are not years old in 1850/51 and not years old in 1880/81. The second stage drops observations that are not unique, that is, that have one or more other observations with the exact same characteristics. The third stage drops observations with missing values for any of the variables used in the matching. over time, and age (which changes in a predictable way). 12 Before matching individuals across samples, it is necessary to drop some observations. First, I keep only males who report being between 20 and 25 years old in 1850/51 and between 50 and 55 years old in 1880/81, because the goal is to match the same workers in the beginning and end of their careers. Using a range of ages is necessary to take into account misreporting, both intentional and unintentional. Second, I drop all observations that are duplicated in each sample, that is, observations that are identical to other observations in the same sample. Duplicated observations represent individuals with the same value for all variables used in the matching, so they cannot be confidently matched with themselves in another sample. Third, and for the same reason, I drop observations that have missing values for any of the variables used in the matching. Table 2 shows the number of observations that are male, are years old in 1850/51 and years old in 1880/81, are not duplicated, and have no missing values. For the United States, the dataset contains census records for 101,323 males in 1850 and 25,512,728 males in 1880, but only 11,629 and 556,221 are suitable to be matched across censuses, respectively. For Great Britain, the dataset contains 194,520 males in 1851 and 14,534,818 males in 1881, of which only 18,332 and 258,054 are suitable to be matched across censuses, respectively. Matching male individuals from the 1850/51 to the 1880/81 censuses according to their name and surname, age, gender, birthplace, and race (only for United States workers) results in 2,226 successful matches for the United States (19% success rate) and 3,972 successful matches for Great Britain (22% success rate). 13 A match is considered successful when, across censuses, the individual has the same name and surname, the same gender, the same migrant status, is born in the same state (in the United 12 I do not use migration status because it contains strictly less information than birthplace. The migration status in the North Atlantic Population Project database is either resident for people born in the United States or Great Britain, or international for people born elsewhere. 13 These success rates are in line with those of other papers applying similar matching procedures, such as Long and Ferrie (2013). The success rate is computed as the number of successful matches over the number of observations in the smaller sample being matched. In our case, the smaller samples are those of 1850 in the United States and 1851 in Great Britain. 7

8 United HH City Ducan Age Urban Farm Labor States size size status force Native White Obs Original % 51% 88% 84% 98% 11,629 Matched % 53% 89% 91% 98% 2, Original % 87% 96% 85% 82% 556,221 Matched % 82% 97% 91% 98% 2,226 Table 3: Descriptive statistics for the original samples and matched sub-sample for the United States. Notes: Columns show the average age of individuals, the average household size, the average city size, the average Ducan socioeconomic status indicator, the proportion of individuals living in an urban area, the proportion of households with one or more members working as a farmer, agricultural laborer, or cottar, the proportion of individuals who are in the labor force, the proportion of individuals who are native born, the proportion of individuals who are white, and the number of individuals in each sample or subsample. States) or county (in Great Britain), is years old in the 1850/51 census and years old in the 1880/81 census, and has the same race (in the United States only). 14 To assess how representative the sub-samples of matched individuals are, we can compare their average characteristics with the original 1850/51 and 1880/81 samples. Table 3 shows descriptive statistics for the United States samples. The sub-sample of matched workers is similar to the original 1850 sample except for its lower average city size, and the higher proportion of native individuals. With respect to the original 1880 sample, the sub-sample of matched workers has a lower proportion of individuals associated with farming, a higher proportion of natives, and a higher proportion of white individuals. Table 4 shows similar descriptive statistics for Great Britain samples. 15 The sub-sample of matched workers contains less individuals living in urban areas in 1851 than the original sample, and is very close to the characteristics of the original 1881 sample. Overall, the sub-sample of matched individuals for the United States and for Great Britain are quite representative for the original samples, but in the analysis that follows we have to keep in mind the few systematic differences that were just described. Having constructed a sample of workers to be followed from 1850/51 to 1880/81, I now turn to the issue of measuring economic outcomes. One of the variables recorded in the 1850/51 and 1880/81 censuses of the United States and Great Britain is self-reported occupation. I will use occupation as a proxy for income, which was not recorded in these censuses. Using occupation as a proxy for income to measure economic mobility within generations has two important limitations. 16 First, it assigns the same income to all workers with the same occupation. This ignores the dispersion of income that undoubtedly exists within any given occupation. Second, changes in income that do not imply changes in occupation are also ignored by this 14 Before the matching, names were coded with the Soundex algorithm to take into account misspellings. Since these censuses were recorded orally, unfamiliar accents and pronunciations often lead to confusion. 15 The original 1850/51 and 1880/81 samples used in Table 3 and Table 4 only include males who report being between 20 and 25 years old in 1850/51 and between 50 and 55 years old in 1880/81, are not duplicated, and have no missing values for any of the variables used in the matching. In other words, these observations are comparable to the ones being matched. 16 For studies discussing the limitations of using occupation as a proxy for income, see McMurrer, Condon, and Sawhill (1997) or Björklund and Jäntti (2000). 8

9 Great HH Labor Age Urban Farm Britain size force Native Married Obs. Original % 57% 97% 94% 24% 18,332 Matched % 55% 97% 96% 25% 3,972 Original % 31% 99% 92% 82% 258,054 Matched % 34% 100% 96% 83% 3,972 Table 4: Descriptive statistics for the original samples and matched sub-sample for Great Britain. Notes: Columns show the average age of individuals, the average household size, the proportion of individuals living in an urban area, the proportion of households with one or more members working as a farmer, agricultural laborer, or cottar, the proportion of individuals who are in the labor force, the proportion of individuals who are native born, the proportion of individuals who are married, and the number of individuals in each sample or subsample. method. Therefore, using occupation as a proxy for income relies on the assumption that occupations reported in the census are detailed enough to capture significant changes in income. Despite these limitations, occupation is the only variable in the dataset that can be used to measure economic standing in both countries. To allow for reasonable comparisons over time, the North Atlantic Population Project database provides a set occupation codes that consistently classify all occupations recorded in late 19th century censuses in the United States and Great Britain. These occupation codes were devised to solve comparability issues generated by frequent reorganizations of the occupational and industrial classification systems by census administrators. For the United States, occupations are coded according to the 1950 Census Bureau occupational classification system. 17 coded according to a criterion devised by the General Register Office. 18 In the case of Great Britain, occupations are In addition to occupations and occupation codes, the North Atlantic Population Project database contains a constructed variable assigning an income score to each United States occupation code. This income score is calculated as the median total income (not only wage and salary) of all individuals with each particular occupation in 1950, measured in thousands of 1950 dollars. 19 Using income scores requires caution for two reasons. First, income scores are based on the relative economic standing of occupations in 1950, which might be different from that in 1850 and Second, the data used to build income scores represents a 3.33 percent sample of the United States population, including both men and women. However, the sample used in this paper contains only men. Despite these 17 This variable is labeled OCC50US in the North Atlantic Population Project database and it contains over 280 occupation codes. The documentation in the dataset notes that in , any laborer with no specified industry in a household with a farmer is recoded into farm labor. For further details about the classification of United States occupations, see Ronnander (1999). 18 This variable is labeled OCCGB in the North Atlantic Population Project database and it contains over 410 occupation codes. For further details about the classification of Great Britain occupations, see Woollard (1999a) and Woollard (1999b). 19 This variable is labeled OCSCORUS in the North Atlantic Population Project database. It is calculated using data from a published 1950 census report. The documentation in the dataset describes this variable as simply a tool for economically scaling occupations [... ] a way of turning occupation into a continuous measure. An occupation with a high score is a well-rewarded and probably high-status occupation, but note that the measure is an economic score, not a socioeconomic one. 9

10 Year Same Similar Unknown occupation occupation occupation 1850/51 2, /81 2,781 1, Table 5: Relationship between occupations for matched individuals. Notes: Cells show the number of Great Britain workers in the sub-sample of matched workers whose occupation code could be matched with the an occupation code in the United states with the exact same or a similar job description. Some observations are reported as having unknown occupations, and are therefore dropped from the sample. limitations, income scores provide a way to rank occupation codes in a meaningful way, allowing the crucial distinction between upward and downward economic mobility. To my knowledge, there is no income score systematically associated to occupation codes in Great Britain, so I construct one in the following way. Since each United States occupation code has an income score associated with it, I can attribute an income score to each Great Britain occupation code by associating each Great Britain occupation code with an equivalent United States occupation code. 20 Table 5 shows the number of Great Britain workers whose occupation code could be associated with an occupation code in the United States with the exact same job description, and the number of Great Britain workers whose occupation code could be matched to an occupation code in the United States with a similar job description. Given that there is a large number of occupation codes, and that there are many more occupation codes for Great Britain than for the United States, nearly all occupation codes in Great Britain can be associated with occupation codes in the United States with similar or, in most cases, identical job descriptions. There are 140 individuals in the sub-sample of matched workers for Great Britain that have occupations that were recorded but of unknown meaning, blank field or illegible either in 1851 or in 1881, and need to be dropped from the sample. This means that the sub-sample of matched individuals for Great Britain used in the empirical analysis of the following section contains 3,832 individuals, all of which have known occupations. Imputing United States income scores to Great Britain workers based on their occupation requires making the critical assumption that workers in Great Britain received relative economic compensations that were similar to the ones received by workers in the United States. In other words, I construct income scores for Great Britain workers assuming that, in most cases, well paid occupations in the United States were also well paid in Great Britain, and poorly paid occupations in the United States were also poorly paid in Great Britain. Using data from the original sample of United States workers in 1850, Table 6 shows how occupations in the United States were ranked by average income score (in 1950 dollars). It is, in my view, reasonable to assume that occupations in Great Britain were ranked 20 I have assigned an income score of $0 to all students. Army and navy pensioners have an income score equal to half of the income score of their counterparts on duty. Great Britain merchants and dealers have the same income score as United States store heads and purchasing agents, that is, $42,000. Great Britain contractors have the same income score as United States foremen, that is, $38,

11 Occupation Group Income Score in 1850 Income Score in 1880 Professional and Technical $45,000 $50,000 Managers, Officials, and Proprietors $42,000 $41,000 Clerical and Kindred $28,000 $27,000 Craftsmen $26,000 $27,000 Operatives $24,000 $24,000 Sales Workers $23,000 $24,000 General Laborers $20,000 $20,000 Service Workers $18,000 $17,000 Farmers $14,000 $14,000 Farm Laborers $10,000 $9,000 Table 6: Average income score by occupation group in the United States. Notes: Income scores are expressed in 1950 dollars. The sample only contains male workers. in a similar manner, and that United States income scores can be used to impute income scores for Great Britain occupations. 4 Empirical Analysis For a given sample of workers, economic mobility within generations can be represented with a mobility matrix. Such a matrix has rows representing the economic standing of workers when they are young (in our case, between 20 and 25 years old) and columns representing the economic standing of the same workers when they are old (between 50 and 55 years old). Table 7 and Table 8 show such mobility matrices for the United States and Great Britain, respectively, between 1850/51 and 1880/81. Income, which is imputed through occupation as explained in the previous section, is expressed in thousands of 1950 US dollars, and separated into 8 intervals, each with a range of $10,000. Before comparing the mobility matrices of the United States and Great Britain, it might be useful to analyze the characteristics of each mobility matrix separately. In late 19th century United States, most workers in the sample had income scores between $0 and $30,000 in the beginning and in the end of their careers. However, the number of workers in the lowest income category decreased considerably from 1850 to 1880, while the number of workers in the income category ranging from $40,000 to $50,000 almost tripled. Also, the number of workers in the two top categories, with income scores between $60,000 and $80,000, almost doubled between 1850 and In the case of Great Britain, there was a similar concentration of workers with income scores between $0 and $30,000, representing almost 90% of the sample. The number of workers in the bottom category, with income scores between 0$ and $10,000, halved from 1851 to 1881, while the number of workers with an income score between $40,000 and $50,000 more than doubled, and the number of workers in the top two categories more than tripled. 11

12 Income of workers aged (in 1850) Income of workers aged (in 1880) Total % , Total , % Table 7: Matrix of economic mobility within generations in the United States. Notes: Rows represent the income score (in thousands of 1950 dollars) of workers aged in 1850, and columns represent the income score of the same workers when they were years old in Each cell represents the number of workers that fall into each category. Additional rows and columns show the total number of workers per row and column, and the percentage they represent over the sub-sample of matched workers. 12

13 Income of workers aged (in 1851) Income of workers aged (in 1881) Total % , , , Total ,551 1, % Table 8: Matrix of economic mobility within generations in Great Britain. Notes: Rows represent the income score (in thousands of 1950 dollars) of workers aged (in 1851) and columns represent the income score of the same workers when they were years old (in 1881). Each cell represents the number of workers that fall into each category. Additional rows and columns show the total number of workers per row and column and the percentage they represent over the sub-sample of matched workers. 13

14 Diagonal cells in a mobility matrix represent economic immobility, that is, workers that were in the same economic category when they were between 20 and 25 years old and when they were between 50 and 55 years old. In contrast, cells outside the diagonal represent workers who changed their economic standing from the beginning to the end of their careers, that is, workers who experienced economic mobility within generations. For this reason, the portion of workers that changed their economic standing throughout their careers can be used as a rough measure of overall economic mobility within generations. For any M M mobility matrix A (with cells denoted by a ij ), define the following statistic: M 1 =1 P M i=1 a ii P M P M i=1 j=1 a ij We can use the M 1 statistic to compare economic mobility within generations for the United States and Great Britain based on the mobility matrices shown in Table 7 and Table 8. For the United States, this measure of mobility is M 1,US = 51%, andforgreatbritainitism 1,GB = 56%. This indicator suggests that overall mobility within generations was slightly lower in the United States than in Great Britain in the end of the 19th century. In other words, a random worker was more likely to be in the same income group in the beginning and the end of his or her career in the United States than in Great Britain. The M 1 statistic is a very limited measure of economic mobility. For example, it does not distinguish between upward and downward mobility. Out of the 2,226 workers in the United States sample, 685 experienced upward mobility (31% of the sample) and 450 experienced downward mobility (20%) throughout their careers. In Great Britain, out of 3,832 workers, there were 1,370 who experienced upward mobility (36%) and 795 who experienced downward mobility (21%). Thus, the average worker in the United States was slightly less likely to experience upward mobility than the average worker in Great Britain, and had similar chances of experiencing downward mobility. The analysis above does not take into account the magnitude of movements across income categories. 21 Anaturalextension,then,istoweightmovementsacrossincomecategoriesaccordingtothemagnitude of the movement, that is, the number of cells away from the diagonal in the mobility matrix. To this end, we can give a weight of 0 to observations in the diagonal of the mobility matrix, a weight of 1 to observations one cell away from the diagonal, a weight of 2 to observations two cells away from the diagonal, and so on. This gives rise to a second statistic of overall mobility that takes into account the magnitude of movements across income categories. For any M M mobility matrix A (with cells denoted by a ij )andweightsw ij = i - j, define the following statistic: M 2 = P M P M i=1 j=1 w ija ij P M P M i=1 j=1 a ij 21 By income categories I mean the $10,000 income score intervals used to construct mobility matrices in this paper. 14

15 This measure of weighted economic mobility within generations is exactly the same for the United States and Great Britain, M 2,US = M 2,GB = 83%. This suggests that, after taking into account the magnitude of movements across income categories in this particular way, the United States and Great Britain had very similar levels of economic mobility within generations between 1850/51 and 1880/81. Additional information about the nature of economic mobility can be obtained from computing the expected number of income categories that a worker crossed throughout his or her career. Conditional on experiencing upward mobility, United States workers expected an increase of 1.76 income categories; conditional on experiencing downward mobility, they expected a decrease of 1.41 income categories. 22 For Great Britain, these expectations are 1.53 and 1.34, respectively. This indicates that United States workers experienced, on average, larger upward and downward movements across income categories than Great Britain workers, conditional on moving upwards or downwards, respectively. Previous work on occupational mobility has typically given special treatment to farmers. 23 It might be the case that comparisons of economic mobility between the United States and Great Britain are driven almost entirely by farmers. To check for this possibility, Table 9 and Table 10 display mobility matrices of late 19th century United States and Great Britain, excluding workers who were farmers in 1850/ Using these new tables, I recalculate the M 1 and M 2 mobility statistics defined earlier. Excluding farmers, the fraction of United States workers who experienced economic mobility is M 1,US = 62%. In Great Britain, the fraction of non-farming workers who experienced economic mobility is M 1,GB = 56%. This indicates that economic mobility within generations was significantly higher for non-farming workers in the United States than in Great Britain. Thus, the fact that late 19th century United States shows lower economic mobility within generations than Great Britain is driven by farmers. 25 Among the 1,453 non-farming workers in the 1850 United States sample, 511 experienced upward economic mobility, and 395 experienced downward economic mobility throughout their careers, representing 35% and 27% of all non-farming United States workers, respectively. Similarly, the 1851 Great Britain sample has 3,694 non-farming workers in 1851, 1,275 of which experienced upward economic mobility while 793 experienced downward economic mobility, representing 35% and 21% of non-farming workers in Great Britain, respectively. This means that non-farming workers in the United States and Great Britain had similar chances of experiencing upward economic mobility throughout their careers, 22 The expected increase in income categories, conditional on experiencing upward mobility, results from dividing the sum of weighted upward movements by the sum of upward movements. Similarly, the expected decrease in income categories, conditional on experiencing downward mobility, results from dividing the sum of weighted downward movements by the sum of downward movements. 23 See Hout and Guest (2013) and Xie and Killewald (2013). 24 I exclude only farmers, and keep farm laborers. Among matched United States workers, the data set contains 773 farmers in 1850 and 1,053 farmers in That represents 34.7% and 47.3% of the sample of matched United States workers, respectively. Among matched Great Britain workers, the data set contains 138 farmers in 1851 and 262 farmers in 1881, which represents 3.6% and 6.8% of the sample of matched Great Britain workers, respectively. Thus, farmers represented a much larger proportion of male workers in the United States than in Great Britain by the end of the 19th century. 25 This finding is consistent with (but not necessarily related to) the emergence of the Granger movement between 1867 and According to the tables of economic mobility reported in this paper, United States farmers experienced less economic mobility than non-farming workers, so farmers in the United States had greater incentives to favor redistribution than non-farming workers. 15

16 Income of workers aged (in 1850) Income of workers aged (in 1880) Total % Total % Table 9: Matrix of economic mobility within generations in the United States, excluding farmers. Notes: Rows represent the income score (in thousands of 1950 dollars) of workers aged in 1850, and columns represent the income score of the same workers when they were years old in Each cell represents the number of workers that fall into each category. Additional rows and columns show the total number of workers per row and column, and the percentage they represent over the sub-sample of matched workers. 16

17 Income of workers aged (in 1851) Income of workers aged (in 1881) Total % , Total ,520 1, % Table 10: Matrix of economic mobility within generations in Great Britain, excluding farmers. Notes: Rows represent the income score (in thousands of 1950 dollars) of workers aged in 1851, and columns represent the income score of the same workers when they were years old in Each cell represents the number of workers that fall into each category. Additional rows and columns show the total number of workers per row and column, and the percentage they represent over the sub-sample of matched workers. 17

18 but non-farming workers in the United States had a higher probability of experiencing downward mobility. Taking into account the magnitude of movements across income categories, the M 2 weighted measure of economic mobility for non-farming workers is M 2,US = 103% for the United States and M 2,GB = 82% for Great Britain. This indicates that, taking into account the magnitude of movements across income categories, United States non-farming workers experienced significantly higher economic mobility within generations than their Great Britain counterparts. Non-farming United States workers expected an increase of 1.79 income categories, conditional on experiencing upward mobility, and a fall of 1.46 income categories, conditional on experiencing downward mobility. In Great Britain, non-farming workers expected an increase of 1.54 income categories, conditional on experiencing upward mobility, and a fall of 1.34 income categories, conditional on experiencing downward mobility. In other words, conditional on experiencing upward mobility, United States non-farming workers experiences larger increases in economic standing than their Great Britain counterparts, but conditional on experiencing downward mobility they experienced larger decreases in economic standing. So far, I have only used indicators of overall economic mobility within generations. This has been intentional; this paper aims at capturing the degree of economic mobility that workers in the United States and Great Britain experienced or perceived by the end of the 19th century. However, it is possible to use more sophisticated indicators to decompose economic mobility into different kinds, usually exchange and structural mobility, which allows understanding the nature of differences in economic mobility across countries. I will close this section with a brief analysis of exchange mobility within generations in late 19th century United States and Great Britain. 26 One way to measure exchange mobility is to measure the association between rows and columns by calculating the cross-product ratio. For the a 2 2 square matrix P = " p 11 p 12 p 21 p 22 # the cross-product ratio is: CPR P = p11 p 12 p21 p 22 In the special case of a 2 2 matrix, the cross product ratio can be interpreted as the ratio of (a) the odds that a young worker in economic category 1 will end up in category 1 rather than 2 when old over (b) the odds that a young worker in category 2 will end up in category 1 rather than 2 when 26 Exchange mobility is mobility that is not related to the fact that rows and columns have different marginal distributions in a mobility matrix. 18

19 old. Perfect exchange mobility implies a cross product ratio equal to 1. As shown in Long and Ferrie (2013), it is possible to compare the cross product ratio of two matrices larger than 2 2 using the Altham statistic, which summarizes the difference in all cross product ratios between matrices. 27 For two matrices A and B, each with r rows and s columns, the Altham statistic is defined as: 2 rx sx rx sx d (A, B) = 4 aij a lm b im b lj log a im a lj b ij b lm i=1 j=1 l=1 m= The Altham statistic measures how far the association between rows and columns in matrix A is from the association between rows and columns in matrix B. This statistic has the advantage that it can be used to perform a chi-squared likelihood ratio test. 28 The null hypothesis of this test is that d(a, B) =0,thatis,thatmatricesA and B have the same degree of association between rows and columns. 29 matrices A and B. Thus, rejecting this null indicates that exchange mobility is statistically different in The Altham statistic can also be used to measure exchange mobility for a given mobility matrix, by comparing it to an artificial J matrix with independent rows and columns. An example of such a J matrix is a matrix with all entries equal to 1. For the United States, the Altham statistic comparing the mobility matrix in Table 7 with a J matrix is d(us,j) = 88. Astatisticaltestofthenullhypothesis that rows and columns in the United States mobility matrix are independent (testing d(us,j)=0) confidently rejects the null hypothesis. In the case of Great Britain, the Altham statistic comparing the mobility matrix in Table 8 with a J matrix is d(gb, J) = 95, andthenullhypothesisthatd(gb, J) =0 is confidently rejected in a statistical test. The fact that d(us,j) is lower than d(gb, J) suggests that the United States had higher exchange mobility than Great Britain. In other words, it suggests that the economic standing of old United States workers was less related to their standing when they were young compared to their Great Britain counterparts, once we control for differences in marginal distributions of rows and columns. Finally, we can use the Altham statistic to measure the difference in exchange mobility between the United States and the Great Britain directly. The value of the Altham statistic comparing Table 7 with Table 8 is d(us,gb) = 43 and the null hypothesis that d(us,gb)=0has a p-value of 7%. This suggests rejecting the null hypothesis that exchange mobility was the same in the United States and Great Britain, although the evidence is not very strong. The same statistical analysis of exchange mobility can be carried out for non-farming workers, which are represented in Table 9 and Table 10. The distance between the mobility matrix of United States non-farming workers and the J matrix is d(us,j) = 89, andthenullthatd(us,j) =0is confidently rejected. Similarly, the distance between the mobility matrix of non-farming workers in Great Britain 27 See Altham (1970). 28 See Agresti (2002), page See Long and Ferrie (2013). 19