“Under a spreading chestnut tree.
The village smithy stands;.
The smith, a mighty man is he,.
With large and sinewy hands;.
And the muscles of his brawny arms.
Are strong as iron bands”.
-Henry Wadsworth Longfellow.
Abstract
In nineteenth-century America, blacksmiths were a fixture in every village, town, and city, producing a diverse range of products from axes to wheels and services from repairs to horse shoeing. In constructing his historical GNP accounts, Gallman opted to exclude these “jacks-of-all-trades” from the manufacturing sector, classifying them instead as part of the service sector. However, using establishment-level data for blacksmiths from the federal censuses of manufactures for 1850, 1860, and 1870, we re-examine that choice and show that blacksmiths were an important, if declining, source of manufactured goods. Moreover, as quintessential artisan shops, a close analysis of their structure and operation helps resolve several key puzzles regarding industrialization in the nineteenth century. As “jacks-of-all-trades,” they were generally masters of none (except for their service activities). Moreover, the historical record reveals that several of those who managed to achieve mastery moved on to become specialized manufacturers of that specific product. Such specialized producers had higher productivity levels than those calling themselves blacksmiths producing the same goods, explaining changes in industry mix and the decline of the blacksmith in manufacturing.
This is a preview of subscription content, log in via an institution to check access.
Similar content being viewed by others
Notes
The 1900 census combined blacksmithing with wheelwrighting.
Collection of sample data from the extant manuscripts of the nineteenth century censuses of manufacturing was begun by Bateman and Weiss (see 1981) and completed by Atack and Bateman. The Atack and Bateman samples pertain to the 1850 through 1880 census years, but we do not use the 1880 sample. This is because, as explained in the text, we rely heavily on information that the Census collected on the specific products that blacksmith shops produced—information which was not collected by the 1880 census. The basic sample data are available for download from https://my.vanderbilt.edu/jeremyatack/data-downloads/. This paper also uses additional information on business organization (e.g., partnership, corporation) culled from the original Atack–Bateman data worksheets; see Atack (2014).
Gallman’s (1960) appendix gives the details of his estimation procedure. In the case of manufacturing, the basic sources are the federal censuses, starting in 1840. These were supplemented by various state censuses, which were used to interpolate to mid-points (e.g. 1854) between federal census dates.
As we discuss later in the paper, an obvious problem with this logic is that median establishment size in manufacturing in 1850 was two workers and approximately 80% of establishments had five workers or fewer (Margo 2015, p. 221). Moreover, a clear majority of all establishments through 1880 (and beyond) were sole proprietorships and corporations were rare—even if their products were not (Atack 2014, Tables 17.1 and 17.2). We return to this point later in the paper.
Specifically, we drop observations for which no labor, or capital, or inputs, or outputs were reported, if value-added (output value minus input value) was negative, if the business produced less than $500 worth of (nominal) annual output (such establishments were not supposed to be included in the census) and those whose estimated rate of return lay in the upper or lower 1% (on the grounds that these were outliers and must have suspect data).
Specifically, space was at a premium since the data had to be transferred to 80-column Hollerith punch cards after encoding for entry into the mainframe computer. Moreover, the primary scientific programming language of the time (FORTRAN) was not well suited to string manipulation.
A few individual worksheets are missing from their worksheet folders—presumably these were removed at some point over the past fifty years or so to check information and not returned (or improperly filed). In these cases, the “doing business as” field has been coded as missing.
These figures differ slightly from those reported in Atack (2014) because of the application of data screens here to eliminate observations with any missing or suspect data.
As previously noted, not all of this information made it into the original Atack–Bateman samples since the data were encoded on 80-column Hollerith punch cards—three cards per observation, one for labor, capital, power, location, etc., one for inputs, and one for outputs. Bateman and Weiss determined that no more than four inputs and output values, quantities and codes could be accommodated within the 80-column space of a single card. However, since few establishments reported more than four inputs or outputs, they opted to consolidate the additional data from those few observations rather than add more (mostly blank) input and output cards per observation. When there were more than four distinct inputs or outputs listed, the values of the least important raw material inputs and outputs were aggregated and coded as “miscellaneous” as the fourth input or output. A similar practice must also have been adopted by the enumerators as they sometimes listed a “miscellaneous” category as the last input or output in their enumeration.
In the public code book accompanying the Atack–Bateman sample (http://my.vanderbilt.edu/jeremyatack/files/2011/08/MFGDOC.pdf), a few products have multiple codes that survived the data cleaning process so that the number of different products or raw materials is slightly less than reported in the text. The multiple codes are allowed for in assigning broad product categories.
We refer to our upper bound as "plausible" in the text because we are assuming, plausibly, that blacksmiths who reported their activities as, for example, "jobbing" were disproportionately engaged in services. Our upper bound excludes values associated with these activities from the calculation, causing the manufactures share to be higher than its true value. Exclusions occur within observations (for example, a blacksmith listing "jobbing" as one of its product codes will have the value of this excluded from its total gross value) or across observations (the shop will be dropped from the calculation if all of its gross value is associated with product codes that cannot be clearly assigned to either manufacturing or services).
For this purpose, we use the lower bound measure because this is defined for all product codes—and therefore, all blacksmith shops, whereas, as previously noted, the upper bound measure excludes activities for which the product code is too vaguely worded (“blacksmithing”) to assign to manufactures or services.
At the suggestion of a referee, we conducted a sensitivity analysis in which we narrowed the sample in Panel A, Table 4, to blacksmiths that reported producing a specific agricultural good, whether this was the first, second, third, or fourth product listed. This is a narrower test of the small firm effect because it substantially restricts the product mix by construction, unlike the regressions in Panel A of Table 4.
There is only one good for which there are sufficient observations in the samples to estimate such a regression—plows. Specifically, we compute a variable, PLOWVAL, which is the sum of the total value of plows produced (first through fourth products listed), and restrict the sample to blacksmith shops for which PLOWVAL was positive (in any census year). There are 89 observations in this sample. The dependent variable is the log of the value of plows, and the critical independent variable is the small firm dummy (=1 if one or two; the regression also includes dummies for urban status, state, year, and linear terms in the log of the number of workers, the log of capital invested, and the log of the value of raw materials. The coefficient of the small firm dummy is positive (β = 0.147) which, consistent with the argument in the text, could be attributed to selection bias; however, the standard error is large (s.e. = 0.476), so the coefficient is (very) imprecisely estimated, and we cannot reject the hypothesis that it is statistically zero. We also conducted a similar exercise focusing on blacksmith shops that derived at least 50% of their gross revenue from the production of wagons; in this regression, the dependent variable is log of value added per worker, and the regression includes the small firm dummy, urban status, state, and linear terms in the log of the capital–labor ratio and the share of gross value derived from wagons. There are 50 observations in this sample. The coefficient on the small firm dummy is positive, and the coefficient of the share of gross value from wagons is negative, again consistent with the patterns observed in Panel A of Table 4; like the “plows” regression above, however, both coefficients have large standard errors, and we cannot reject the hypothesis that they are statistically zero.
There are other examples of well-known industrial firms that started as independent blacksmith shops, for example, Studebaker Brothers, which began as a blacksmith shop in the early 1850s, but soon specialized in wagons and carriages. The company grew dramatically during the Civil War as a consequence of military contracts with the Union Army (Erskine 1918), a couple of decades after Deere made the same kind of transition to specialist product producer.
Almost fifty years has passed since collection of these data began and it has been about 45 years since Atack did any product coding on them. No one remembers what the distinction was between the two “miscellaneous” codes, but they were assigned consecutively and very early in the project: 45 and 46. Initially, sequential numerical codes were assigned, began with “1.” After the 99th code had been assigned, subsequent codes were alphanumeric beginning with A0 (A-zero) through A9, then B0 through B9, etc., as the coding sheets and punch cards allowed for only two characters for each code. Once the 80-column Hollerith punch card constraint vanished (in the late 1970s with the switchover to terminals and eventually personal computers), all codes were translated into 4-digit numerical codes as entering only numerical data was faster, more accurate, and more consistent than a mix of numbers and characters. Atack’s best guess for the initial distinction between the two “miscellaneous” codes is that “45” was used where the census enumerator had classified the product as “Other articles” (aka, miscellaneous) while “46” was used where Bateman and Weiss (and their student helpers) had done the aggregation, but this distinction was lost at some point. Certainly, Atack only remembers using “46” for “miscellaneous” (or not specified).
The following final product codes were used for establishments describing themselves as blacksmiths (SIC 769): 1, 7, 10, 11, 13, 16, 27, 28, 29, 32, 45, 46, 47, 52, 53, 54, 55, 57, 63, 64, 68, 74, 83, 94, 96, 124, 130, 152, 164, 165, 168, 191, 192, 199, 203, 228, 257, 310, 346, 350, 351, 358, 366, 367, 370, 422, 446, 519, 533, 537, 564, 611, 628, 629, 630, 640, 649, 650, 651, 655, 703, 789, 822, 829, 852, 854, 935, 982, 985, 991, 1040, 1079, 1105, 1109, 1148, 1161, 1215, 1233, 1246, 1265, 1292, 1297, and 1308.
For example, Gallman considered “carpentering” to be a non-manufacturing industry, putting it into construction instead. It is important to keep in mind that none of the non-manufacturing totals were “lost”—they were simply put elsewhere in Gallman’s national accounts. In the case of the hand trades, these went into services, as we pointed out in the text of our paper.
For example, in 1850, blacksmithing accounted for 97.8% of total value of products in the six hand trades.
The ratio figures in the last column of Table 5 are still too large because we are using the upper bound shares of gross value, rather than, say the average of the upper and lower bounds. Further, it is likely that the share of manufactures in value added in the hand trades is lower still, because manufacturing used more raw materials per dollar of gross value than services.
We recognize that the typical blacksmith spent part of his time making manufactures and part of his time performing services; in effect, we are assuming that if the blacksmith spent half of his time making manufactures, this is the equivalent of 0.5 of a gainful worker.
Gallman's estimates of gainful workers and of value added per worker include mining as well as manufacturing (i.e., value added per gainful worker in manufacturing and mining).
References
Atack J (1976) Estimation of economies of scale in Nineteenth Century United States manufacturing and the form of the production function. Ph.D, unpublished dissertation, Department of Economics, Indiana University, Bloomington
Atack Jeremy (1977) Returns to scale in Antebellum United States manufacturing. Explor Econ Hist 14(4):337–359
Atack J (2014) America: capitalism’s promised land. In: Neal L, Williamson JG (eds) The Cambridge history of capitalism. Cambridge University Press, Cambridge, pp 533–573
Atack J, Bateman F (1999) Nineteenth Century American industrial development through the eyes of the census of manufactures: a new resource for historical research. Hist Methods 32(4):177–188
Atack J, Bateman F, Weiss Thomas J (2004) National samples from the census of manufacturing: 1850, 1860, and 1870 (ICPSR 4048). Inter-University Consortium for Political and Social Research, Ann Arbor
Atack J, Bateman F, Margo RA (2008) Steam power, establishment size, and labor productivity in Nineteenth Century American manufacturing. Explor Econ Hist 45(April):185–198
Bateman F, Weiss TJ (1981) A deplorable scarcity: the failure of industrialization in the slave economy. University of North Carolina Press, Chapel Hill
Bates C (1886) The law of limited partnership. In: Little, Brown, Boston http://galenet.galegroup.com/servlet/MOML?af=RN&ae=F101533948&srchtp=a&ste=14&locID=29002
Broehl WG (1984) John Deere’s company: a history of Deere and Company and its times. Doubleday, New York
Burdick, FM (1899) The law of partnership including limited partnerships. In: The students’ series. Little, Brown, Boston http://galenet.galegroup.com/servlet/MOML?af=RN&ae=F101544572&srchtp=a&ste=14&locID=29002
Erskine Albert Russel (1918) History of the Studebaker corporation. Poole Bros, Chicago
Gallman RE (1960) Commodity output, 1839–1899. In: The Conference on Research in Income and Wealth (ed) Trends in the American Economy in the Nineteenth Century, Columbia University Press for the NBER, New York, pp 13–72
Gallman RE (1966) Gross national product in the United States, 1834–1909. In: Brady DS (ed) Output, employment, and productivity in the United States After 1800, Columbia University Press, New York, pp 3–76
Gallman RE, Weiss TJ (1969) The service industries in the Nineteenth Century. In: Fuchs Victor R (ed) Production and productivity in the service industries. Columbia University Press for the NBER, New York, pp 287–381
Hilt E, O’Banion K (2009) The limited partnership in New York 1822–1858: partnerships without kinship. J Econ Hist No 69(3):615–645
Howard Stanley E (1934) The Limited Partnership in New Jersey. J Bus Univ Chic 7(4):296–317
Margo RA (2015) Economies of scale in Nineteenth Century American Manufacturing: a solution to the entrepreneurial labor input problem. In: Collins WJ, Margo RA (eds) Enterprising America: Businesses, Banks, and Credit Markets in Historical Perspective. University of Chicago Press, Chicago, pp 215–244
Potter N (1960) Comment. In: The Conference on research in income and wealth (ed) Trends in the American economy in the Nineteenth Century, Columbia University Press for the NBER, New York, pp 67–69
Sokoloff KL (1984) Was the transition from the artisanal shop to the nonmechanized factory associated with gains in efficiency? evidence from the U.S. manufacturing censuses of 1820 and 1850. Explor Econ Hist 21(4):351–382
United States. Census Office (1872) The statistics of the wealth and industry of the United States, embracing the tables of wealth, taxation, and public indebtedness; of agriculture: manufactures: mining: and the fisheries with which are reproduced, from the volume on population, the major tables of occupations. Compiled, from the original returns of the Ninth Census (June 1, 1870), under the direction of the Secretary of the Interior, by Francis A. Walker, Superintendent of Census. In: H.misdoc.46. Washington: [s.n.]. http://congressional.proquest.com/congcomp/getdoc?SERIAL-SET-ID=1475+H.misdoc.46
United States. Office of Management and Budget (1987) Standard industrial classification manual. For sale by National Technical Information Service, Springfield
Weiss TJ (1967) The service sector in the United States, 1839 through 1899. Ph D, University of North Carolina, Chapel Hill
Acknowledgements
We are grateful to Stanley Engerman; Thomas Weiss; seminar participants at Boston University, Carnegie-Mellon, NBER, and Yale University; and two referees for helpful comments.
Author information
Authors and Affiliations
Corresponding author
Appendices
Appendix 1
As indicated in the text, enumerators of the censuses of manufactures in 1850, 1860, and 1870 were instructed to list up to six raw materials used in the production of up to four individually identified final products. Specifically, the instructions stipulated that:
“Under the general heading, entitled “Annual products” is to be inserted the quantity, kind, and value of each produced during the whole year. It will require great care to fill this column properly. When several articles are manufactured, the first four only need be particularly specified, and the remainder classed under a general heading of “Other articles,” and the aggregate value of such articles carried out, the quantity being omitted; or, where otherwise impracticable in any case, the aggregate value, without the specific quantity or kind. In stating the value of the products, the value of the articles at the place of manufacture is to be given, exclusive of the cost of transportation to any market.” [emphasis in original] (Wright 1900, p. 314)
The Bateman–Weiss coding scheme kept the spirit of these instructions within the space constraints imposed by an 80-column Hollerith punch card. To achieve this, they reduced the number of individually identified raw materials and final products to a maximum of the four most important (by value). In those cases where more than four inputs or outputs were identified, only the three most important by value were identified by specific codes and the value of the remaining inputs or outputs was aggregated, reporting that value under a code for “Miscellaneous.”
Collectively, the products made by the blacksmiths in the individual Bateman–Weiss state samples were classified under 83 different final product codes, 82 of which were unique (in the sense of different descriptions or units of measurement—including none). The duplicate code is for “miscellaneous.”Footnote 17 In analyzing the activities of blacksmiths, we grouped these 83 final products (disregarding the units of measurement) into six broad groups (some of which represent judgment calls about what was meant by the product description).Footnote 18 Specifically:
“General blacksmithing work”: blacksmithing, custom work, horseshoes, jobbing, joiner work (presumably welding, etc.), miscellaneous (horse) shoeing/shoeing, etc.,/shoes, and stove fitting.
“Hardware”: copper, harnesses (presumably fittings thereof like bits, buckles, hame clips, and rosettes), hinges, iron cast, ironware, locks, locks, etc., millwork, nails, screws, shipwrighting (presumably fittings like oarlocks), spikes, springs, tableware, tinware, and wagon irons.
“Implements”: agricultural implements, axes, corn planters, cradles, cultivators, edge tools, etc., farm/plantation, hoes, machinery, mining, planers, plows, reapers, scythes, steel work, threshing machines, tools, and wheat drills.
“Iron work”: iron railings/rails, iron/ironwork, and wrought iron.
“Repairs”: guns/rifles (almost certainly confined to repairing items such as trigger guard, sight, etc.), repair work, and wagon work.
“Wagons and Carriages”: buggies, carriages, carts, coaches, wheel hubs, sleighs, wagons, wheels.
Appendix 2
We use our estimates of the share of blacksmiths’ gross value added represented by their manufacturing (as opposed to services) output to explore the bias in Gallman’s estimates of nominal value added in manufacturing for the census years 1850–1870. Gallman’s estimates of nominal value-added (in hundreds of millions of current dollars) can be found in Table A-1 of his 1960 article (Gallman 1960, p. 43). In his discussion of the construction of the estimates, Gallman (1960, p. 57) notes that “[c]ensus manufacturing totals were adjusted to exclude nonmanufacturing industries … included in the census of manufactures of [1850] through [1870].”Footnote 19 Among these were six industries that Gallman (p. 58) collectively referred to as the “hand trades”: blacksmithing and locksmithing (1850–1880), coppersmithing (1860–1880), whitesmithing (1850–1860), gunsmithing (1870–1880), and carriage-smithing (1860). For example, the 1860 census of manufactures includes a row pertaining to “carriage-smithing”; Gallman adjusts by excluding figures for this industry from his totals. The overwhelming majority of the totals for the hand trades pertain to blacksmithing.Footnote 20
In column 2 of Table 5 we reproduce Gallman’s estimates of nominal value added in manufacturing for 1850–1870. In column 3, we report total value added (“value of products” minus “value of raw materials”) for the six hand trades; and, in column 4, the ratio of value added in the hand trades to Gallman’s aggregates. Note that these ratios are absolutely small overall but smaller in 1870 than in 1850. This would indicate a modest upward bias in the aggregate growth rate of manufacturing value added in Gallman’s estimates, if we were to assume that all of the value added in the hand trades pertained to manufacturing. We know that this is not the case for blacksmithing, but we lack data on the manufactures share for the other hand trades. However, this does not matter, because as noted above, blacksmithing accounted for the vast majority of economic activity in the hand trades. As a practical matter, therefore, we can adjust value added in the hand trades downward by multiplying by the manufactures shares from Panel A of Table 3; for this purpose, we use the upper bound shares. In effect, we are assuming that, proportionately, manufacturing in the other hand trades was the same as in blacksmithing. These adjusted totals are shown in column 4, Table 5. The exclusion of manufacturing value added from the hand trades does bias upward Gallman’s estimates of the size of the manufacturing sector, more at the beginning of the period (1850) than at the end (1870). While this supports Potter’s (1960) conceptual criticism, the magnitude of the bias is trivial.Footnote 21
We can also use our results to explore the size of the bias in Gallman’s estimates of output per worker. To this end, we use the following equation, which pertains to the hand trades:
In this equation, V refers to value added, L to gainful workers, M to manufacturing, and S to services; β is the ratio of labor productivity in manufactures as opposed to services.Footnote 22 For the hand trades, we can estimate the V’s from Table 5; we know the total \(L \, = \, \left( {L_{\text{M}} + \, L_{\text{S}} } \right)\) from the census of manufactures; and we can estimate β from the regression in Panel A of Table 4, assuming a manufactures share of 1 (we use the regression coefficient of the manufactures share from last column in Panel A of Table 4: \(\beta = { \exp }\left( { - 0.132} \right) = 0.876\)). By rearranging the equation, we can estimate the ratio LM/LS; and because we know the total L, we can recover estimates of LM.
In Table 6, we report Gallman’s estimates of gainful workers in manufacturing (column 2); our estimates of LM in the hand trades (column 3); the ratio of our estimates of LM in the hand trades to Gallman’s estimates of gainful workers in manufacturing (column 4); Gallman’s estimates of nominal value added per worker (column 5); our adjusted estimates of output per worker, which include manufacturing output and estimated gainful workers (LM) from the hand trades (column 6); and the ratio of our estimates of output per worker to Gallman’s (column 7).Footnote 23 There is a slight upward bias to Gallman’s estimates of labor productivity, more so in 1850 than in 1870—again, consistent with Potter (1960)—but the magnitude of the bias is trivial (and literally zero in 1860).
Rights and permissions
About this article
Cite this article
Atack, J., Margo, R.A. Gallman revisited: blacksmithing and American manufacturing, 1850–1870. Cliometrica 13, 1–23 (2019). https://doi.org/10.1007/s11698-017-0165-x
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11698-017-0165-x