A Vision of the Growth Process in a Technologically Progressive Economy: the United States,

Transcription

1 Economic Hisory Woring Papers No: 226/2015 A Vision of he Growh Process in a Technologically Progressive Economy: he Unied Saes, Gerben Baer London School of Economics Nicholas Crafs Universiy of Warwic Pieer Woljer Wageningen Universiy Economic Hisory Deparmen, London School of Economics and Poliical Science, Houghon Sree, London, WC2A 2AE, London, UK. T: +44 (0) F: +44 (0)

2 LONDON SCHOOL OF ECONOMICS AND POLITICAL SCIENCE DEPARTMENT OF ECONOMIC HISTORY WORKING PAPERS NO DECEMBER 2015 A Vision of he Growh Process in a Technologically Progressive Economy: he Unied Saes, Gerben Baer London School of Economics Nicholas Crafs Universiy of Warwic Pieer Woljer Wageningen Universiy Absrac We develop new aggregae and secoral Toal Facor Produciviy (TFP) esimaes for he Unied Saes beween 1899 and 1941 hrough beer coverage of secors and beer measured labor qualiy, and show TFP-growh was lower han previously hough, broadly based across secors, srongly varian ineremporally, and consisen wih many diverse sources of innovaion. We hen es and rejec hree prominen claims. Firs, he 1930s did no have he highes TFP-growh of he wenieh cenury. Second, TFP-growh was no predominanly caused by four leading secors. Third, TFP-growh was no caused by a yeas process originaing in a dominan echnology such as elecriciy. Keywords: Harberger diagram; mushrooms; produciviy growh; oal facor produciviy; yeas JEL Codes: N11, N12, O47, O51 Acnowledgemens: We are graeful o Herman de Jong for maing daa available o us and o Reize Gouma for help wih he esimaion of capial inpu. An earlier version was presened a seminars a he London School of Economics, Oxford Universiy, UC Davis, Sanford Universiy, and he Universiy of Groningen, where we received a number of very helpful commens. We han Sephen Broadberry, Herman de Jong, Alexander Field, and Nicholas Oulon for criical commens and advice on an earlier draf. Funding for a pilo projec was provided by a seed gran from he Sunory and Toyoa Inernaional Cenres for Economics and Relaed Disciplines (STICERD) of he London School of Economics, as well as a gran from he Neherlands Organisaion for Scienific Research (NWO Gran no.: ). Moron Jerven provided excellen research assisance a his iniial sage of he projec. The usual disclaimer applies.

3 A Vision of he Growh Process in a Technologically Progressive Economy: he Unied Saes, Gerben Baer (London School of Economics) Nicholas Crafs (Universiy of Warwic) and Pieer Woljer (Wageningen Universiy) December 2015 Absrac We develop new aggregae and secoral Toal Facor Produciviy (TFP) esimaes for he Unied Saes beween 1899 and 1941 hrough beer coverage of secors and beer measured labor qualiy, and show TFP growh was lower han previously hough, broadly based across secors, srongly varian ineremporally, and consisen wih many diverse sources of innovaion. We hen es and rejec hree prominen claims. Firs, he 1930s did no have he highes TFP growh of he wenieh cenury. Second, TFP growh was no predominanly caused by four leading secors. Third, TFP growh was no caused by a yeas process originaing in a dominan echnology such as elecriciy. JEL Classificaion: N11, N12, O47, O51. Keywords: Harberger diagram; mushrooms; produciviy growh; oal facor produciviy; yeas Acnowledgemens: We are graeful o Herman de Jong for maing daa available o us and o Reize Gouma for help wih he esimaion of capial inpu. An earlier version was presened a seminars a he London School of Economics, Oxford Universiy, UC Davis, Sanford Universiy, and he Universiy of Groningen, where we received a number of very helpful commens. We han Sephen Broadberry, Herman de Jong, Alexander Field, and Nicholas Oulon for criical commens and advice on an earlier draf. Funding for a pilo projec was provided by a seed gran from he Sunory and Toyoa Inernaional Cenres for Economics and Relaed Disciplines (STICERD) of he London School of Economics, as well as a gran from he Neherlands Organisaion for Scienific Research (NWO Gran no.: ). Moron Jerven provided excellen research assisance a his iniial sage of he projec. The usual disclaimer applies.

4 Inroducion One of he mos famous findings in growh economics is in Solow (1957), namely, ha 7/8 h of labor produciviy growh in he Unied Saes in he years 1909 o 1949 came from echnical change, or more precisely could no be aribued o capial deepening bu was a residual. This was soon confirmed in he landmar sudy of Kendric (1961) which esimaed ha during 1869 o 1953 and 1909 o 1948, he growh in oal facor produciviy (TFP) accouned for 80 and 88.5 per cen of labor produciviy growh, respecively. Modern research basically confirms hese findings and also shows ha TFP growh was by far he predominan source of labor produciviy growh even when refinemens are made o allow for improvemens in labor qualiy (cf. Table 1). The conex for hese esimaes is he episode of echnological advance which is ofen called he second indusrial revoluion which saw he Unied Saes overae Briain as he world s leading economy and in he process achieve unprecedened raes of TFP growh. 1 During he early decades of he wenieh cenury, he Unied Saes was in he forefron of he developmen of he mos imporan new echnologies including aviaion, he inernal combusion engine, mass producion, elecriciy, and perochemicals (Mowery and Rosenberg, 2000). Elecriciy, whose impac peaed in he 1920s (David, 1991), is widely recognized as one of hisory s mos imporan general purpose echnologies (GPTs). Gordon (1999; 2000) described TFP growh in he period beween 1891 and 1972 as one big wave based on a few major echnology clusers which delivered much more rapid advance han has been seen since or seems liely in fuure. Imporanly, TFP growh coninued o be rapid hrough he Grea Depression and Field (2003) labeled he 1930s as he mos echnologically progressive decade of he wenieh cenury. Undersanding how and why he Unied Saes was so successful in achieving rapid echnological progress a his ime is obviously imporan, all he more so given curren fears abou secular sagnaion. A ey aspec of his is o decide on he mos appropriae vision of he growh process, a challenge which was se by Harberger (1998) in his presidenial address o he American Economic Associaion. Harberger conrased he possibiliy of a yeas process and a mushroom process of TFP growh (or in his erminology real cos reducion ); he former would be based on a common source perhaps a GPT wih many spillovers applying across many secors, while he laer would enail muliple disparae sources. Agains his bacground, he main conribuion of his paper is o develop a much improved and exended daase which provides a much fuller descripion of he secoral paern of TFP growh across he American economy for he period 1899 o 1941 and sub periods wihin hose years. Compared wih he esimaes in Kendric (1961), which sill comprise he main source available o researchers, our growh accouning covers in deail abou 80 per cen raher han 50 per cen of he privae domesic economy (PDE), we provide deailed esimaes for 1929 o 1941 raher han 1929 o 1937, we adjus TFP growh for labor qualiy improvemen wihin occupaions, we esimae he conribuion of capial inpus on a capial services basis where feasible, i.e., for 1929 o 1941, and we obain a se of valueadded weighs which allow boh he aggregaion of secoral TFP growh raes and also he consrucion of diagrams o illusrae he degree of inequaliy of secoral conribuions o oal TFP growh. Overall, 1 No only was 2 per cen per year far in excess of wha he Unied Saes iself had achieved during mos of he nineeenh cenury, i was also well ahead of TFP growh in Briain during and afer he Indusrial Revoluion which never exceeded 0.8 per cen per year (Abramoviz and David, 2001; Crafs, 2004a; Mahews e al., 1982). 1

5 we find ha conribuions o TFP growh were spread widely across he American economy, so we perform an addiional growh accouning exercise relaing o he direc and indirec implicaions of TFP growh in elecrical capial in he 1920s when elecriciy had is maximum impac as a GPT. We use he resuls of hese invesigaions o challenge, or a leas o qualify, claims made by David (1991), Field (2011), Gordon (2000), and Harberger (1998). To his end, we address several specific quesions. Firs, we consider wheher he growh process was mushrooms (originaing in muliple disparae sources), or yeas (widespread and originaing in one dominan source), or superficially yeas bu acually mushrooms. Second, we quanify he impac of he echnology clusers ha are said o comprise he one big wave of TFP growh. Third, we re examine he issue of wheher he 1930s really was he mos echnologically progressive decade of he wenieh cenury. Fourh, we review he magniude and pervasiveness of TFP growh accruing from invesmen in elecric moors in manufacuring during he 1920s o accoun for he role of elecriciy as a GPT. In ouline, our approach is as follows. We use convenional neoclassical growh accouning assumpions aing accoun of labor qualiy in measuring labor inpus. We sar wih he esimaes in Kendric (1961) for and improve and exend hem where possible. We are able o add 5 secors consrucion, disribuion, FIRE (finance, insurance, and real esae), posal services, and specaor enerainmen. We produce esimaes for 1929 o 1941 using he Naional Income and Produc Accouns (NIPA) o obain nominal value added a he indusry level (which are deflaed using available price daa) and employmen. Capial inpus for 1929 o 1941 are consruced using a perpeual invenory mehod and invesmen daa from he BEA s (2010) fixed asses ables. The capial socs are hen aggregaed o he indusry level on he basis of impued renal prices. The resul is a measure of capial services, which accuraely capures he inpu flows derived from he capial in place. We replace Kendric s pre 1929 indices of labor inpus and develop new indusry level indices for 1929 o 1941, aing changes in he qualiy of labor ino accoun. The big advanage of our mehod for esimaing labor inpus compared wih ha of Kendric (1961) is ha i accouns for improvemens in educaional aainmen wihin occupaions. Moreover, since we now have boh capial and labor inpus measured on a similar basis o ha of he BLS our TFP esimaes for 1929 o 1941 are comparable wih he BLS esimaes for he pos 1948 period. Having consruced hese new growh accouning esimaes a he secoral level, we use hem o examine he concenraion and persisence of he indusry origins of TFP growh. We examine boh he rae of TFP growh and he inensive growh conribuion (TFP growh muliplied by share of oal value added) of each secor. This enables us o consruc diagrams for he PDE in he syle of Harberger (1998) and hus o see wheher he resuling picures loo lie mushrooms or yeas. We also compue ran correlaions o examine iner emporal persisence of produciviy performance a he secoral level. In order o pursue he issue of he role of elecrificaion of facories in TFP growh in American manufacuring, we obain esimaes of TFP spillovers by replicaing he regression analysis in David (1991) using our new TFP esimaes and also develop esimaes of elecrical capial in use a he secoral level. This permis boh growh accouning wih wo ypes of capial (elecrical and nonelecrical) and also he consrucion of Harberger diagrams for elecrical and non elecrical conribuions o TFP growh in manufacuring for he 1920s. The resuls of hese analyses are quie differen from hose of earlier wor in a number of imporan aspecs. Firs, we esimae a much higher rae of growh of labor qualiy han did Kendric (1961); for , our esimae is abou 0.8 per cen per year while his was abou 0.3 per cen per year. An 2

6 immediae corollary is ha we esimae a lower rae of TFP growh in he PDE a 1.3 per cen per year during compared wih Kendric s esimae of 1.7 per cen per year. Second, we esimae ha he echnology clusers associaed wih Gordon s one big wave accouned for jus under 40 per cen of TFP growh in he PDE during Their conribuion rose seadily from 0.3 per cen per year in o jus under 0.75 per cen per year in This is impressive bu possibly less overwhelming han a reader of Gordon (2000) migh imagine. Third, we do no agree wih Field (2003) (2011) ha he 1930s was he mos echnologically progressive decade if he crierion is TFP growh in he PDE; we esimae ha TFP growh was 1.87 per cen per year during compared wih 2.00 percen per year during and 2.23 per cen per year during Fourh, we find ha, including boh own TFP growh and TFP spillovers, elecrical capial conribued only 28 per cen of TFP growh in manufacuring during he 1920s and ha his conribuion varied more across secors han did non elecrical TFP growh, so our findings do no suppor he idea of a yeas process based on elecriciy as a GPT as he bes way o concepualize echnological progress in manufacuring a his ime. Fifh, we find ha inensive growh conribuions were generally broadly based wheher a he level of he PDE or of manufacuring, wih he excepion of he period However, we also find ha here is raher low persisence of TFP growh and inensive growh conribuions beween periods and he economic hisory lieraure reveals muliple sources of TFP growh. Togeher wih our findings on elecriciy, his leads us o conclude ha Harberger s vision ha TFP growh is really a mushrooms process is probably righ even hough, superficially, he appearance is yeas lie. Our resuls ae on added significance in he conex of he renewed ineres in selecive indusrial policy ha has emerged since he financial crisis of 2008 (Warwic, 2013). This migh be hough of as inervening o encourage a shif of producive resources owards secors wih prospecs of rapid TFP growh. Our invesigaion of TFP growh during underlines he dangers of selecive policies favoring echnologically progressive indusries. In paricular, our resuls sugges ha TFP growh over ime was highly unpredicable, ha here were muliple sources of TFP growh raher han a few ey echnologies, and ha large, unglamorous secors lie wholesale and reail rade, which can benefi from bu do no originae echnological progress, conribued much more o TFP growh han exciing new indusries lie elecric machinery. This suggess ha Harberger (1998) was righ o argue ha raher han o ry o bac winners, he ey role for policy is o enable firms o respond effecively o changing circumsances and no o obsruc creaive desrucion. 1. Key Ideas and Lieraure Review A naural saring poin for considering he spread of produciviy growh across secors is he seminal paper by Harberger (1998). This paper inroduced a graphical device similar o he Lorenz Curve o display he paern of TFP growh conribuions by indusry when indusries are raned by heir TFP growh rae from highes o lowes; his is reproduced in Figure 1. Here Figure 1a shows a case where all secors conribue o TFP growh and no indusry dominaes he growh process while Figure 1b shows a case where TFP growh is negaive in a significan number of secors and relaively few indusries accoun for mos of he oal. Harberger used he erminology of a yeas process and a mushroom process o describe hese paerns. 2 A yeas process would be based on very broad and general exernaliies whereas a mushroom process would resul from TFP growh wih a 1001 differen causes. However, i is also 2 The analogy comes from he fac ha yeas causes bread o expand very evenly whereas mushrooms pop up unpredicably and spasmodically. 3

7 possible, as Harberger acnowledged, ha a yeas lie paern migh be observed when disparae (unconneced) sources of TFP growh happen o be widespread and secors wih negaive TFP growh are absen or a leas have lile weigh. Harberger s own view was ha he mushroom diagram represened he usual configuraion and he produced esimaes o illusrae his for he Unied Saes in he period 1948 o 1985 where he diagrams resembled Figure 1b raher han Figure 1a. Moreover, he argued ha over ime differen indusries dominaed so ha conribuions were no only concenraed wihin each period bu exhibied lile persisence beween periods. He hough his migh be characerisic of a Schumpeerian process of creaive desrucion. Harberger did no provide esimaes for earlier in he wenieh cenury bu he saed ha he hough ha he mushroom paern prevailed hen as well and ha he 1920s was he decade of cars and rubber ires, he 1930s was he decade of refrigeraors, and he 1940s was he decade of pharmaceuicals (1998, p. 5). I may seem ha a General Purpose Technology (GPT), i.e. one which comes o have many uses, o be widely used and o have many Hicsian and echnological complemenariies (Lipsey, 1998) would generae a yeas process for TFP growh similar o Figure 1. This need no be he case, however. Consider a growh accouning formula of he ype used o analyze he impac of ICT on labor produciviy growh Y L Y L KO K ICT L L A A (1) KO K ICT A ICTM A NICTM L L This equaion decomposes he sources of labor produciviy growh ino conribuions from wo ypes of capial, ICT capial and oher capial each weighed by heir income shares, and α, and wo ypes of TFP growh in he manufacure of ICT equipmen and in he res of he economy each weighed by heir value added shares, η and φ. When applied o he U.S. non farm business secor in he period , he pea of he new economy boom, his resuls in conribuions from ICT capial and oher capial of 0.78 and 0.44 percenage poins per year, respecively, and from TFP growh in ICT producion and oher producion of 0.72 percenage poins and 0.90 percenage poins, respecively (Byrne e al., 2013). Since ICT producion was only a small secor, a Harberger diagram for TFP growh in his period would loo quie mushroomy even if (unremuneraed) TFP spillovers comprised a significan par of TFP growh in oher secors. If, however, a similar diagram were o be drawn for conribuions o labor produciviy growh, a yeas process would be more liely o emerge because invesmens in ICT capial have been spread across many secors. Indeed, his is wha David and Wrigh (1999) found prevailed during he heyday of an earlier GPT, elecriciy, where hey displayed a fla curve for labor produciviy growh beween 1919 and 1929 in American manufacuring and also repored a srong correlaion beween TFP growh and growh of horsepower in elecric moors which hey aribued o TFP spillovers. 3 A similar labor produciviy growh profile is repored for Swedish manufacuring for he period when elecrificaion diffused rapidly across many indusries (Prado, 2014). Even so, as Prado himself poins 3 In heir discussion of elecriciy as a GPT, David and Wrigh (1999) did no, however, presen a Harberger diagram for TFP growh eiher for manufacuring or he privae domesic economy. Nor did hey explicily quanify he secoral conribuions of his echnology o demonsrae ha i did indeed explain yeasiness. 4

8 ou, his would no always happen wih a GPT since adopion of he echnology across he economy as a whole migh be subjec o variable diffusion lags, as was he case wih seam power (Crafs, 2004b). And in any case, he GPT may no be he only game in own since is emergence may come a a ime when myriad oher sources of produciviy improvemen are available. Alhough he Harberger ype approach has generally been applied o manufacuring, his seems unduly resricive especially given he imporan role ha produciviy growh in services played in he wenieh cenury Unied Saes. This is implici in he well nown accoun of American economic growh given by Gordon (2000) who describes one big wave in TFP growh beween 1891 and 1972 wih an acceleraion peaing in He argued ha his was driven by four echnology clusers: elecriciy, he inernal combusion engine ogeher wih derivaive invenions such as inersae highways and supermares, rearranging molecules (chemicals and pharmaceuicals), and he enerainmen, communicaion and informaion secor. Surprisingly, Gordon did no aemp o quanify he growh conribuion of hese four echnologies over ime. The mos comprehensive recen descripion of he growh process which aemps a decomposiion of secoral conribuions o produciviy growh in he early wenieh cenury American economy is by Field (2011) who drew heavily on he wor of Kendric (1961). Field emphasized ha here were big differences beween he 1920s, in which TFP growh was dominaed by manufacuring, and he 1930s when TFP growh was spread more widely across much of he economy. In he period 1919 o 1929, TFP growh in manufacuring was very rapid a 5.1 per cen per year and exceeded 2 per cen per year in all bu one 2 digi manufacuring indusry; prima facie his does no seem o mach Harberger s caricaure and rapid elecrificaion of facories a ha ime migh seem o indicae ha a yeas process was a wor. In he period 1929 o 1937, however, TFP growh in manufacuring fell o 1.9 per cen per year and he range of TFP growh across manufacuring indusries was much greaer. An imporan innovaion made by Field was o examine TFP growh for he period 1929 o 1941 whereas Kendric (1961) based his disaggregaed esimaes on he period 1929 o 1937; i is he laer end dae which jusifies he label he mos echnologically progressive decade. The raionale for replacing 1937 by 1941 is ha his permis a view based on a year when economic recovery from he Depression was more complee. 4 Field (2011) provided a 4 secor breadown of TFP growh conribuions which suggess ha even if TFP growh wihin manufacuring was yeasy in he 1920s his may no be rue for he economy as a whole. On he oher hand, for he 1930s his emphasis on 4 As noed by Field (2011), 1941 was he firs year since 1929 ha unemploymen averaged less han 10 per cen, maing i far preferable o 1937 as a (peaceime) pea year. Sill, 1941 is no ideal, as unemploymen was higher (9.9 per cen) han in 1929 (3.2 per cen). Because of he pro cyclical naure of produciviy, Field argues ha TFP would have been higher in 1941 had he economy operaed a full capaciy, mainly because of he effecs of labor and capial hoarding. To address his, he suggess a one off adjusmen ha raises he average annual growh of TFP beween 1929 and 1941 from 2.31 o Inlaar e al. (2011) provide a more comprehensive es of he cyclical naure of produciviy in he inerwar US economy. They find robus evidence of shor run increasing reurns o scale, on he basis of which hey calculae a purified measure of echnological change which confirms ha he hoarding of producion facors was he dominan reason for he decline in measured Solow residual TFP in U.S. manufacuring beween 1929 and (Inlaar e al. 2011: 851) Beween 1933 and 1937, however, TFP grew much faser han echnology, resuling from a rapid expansion of he uilizaion of facor inpus. In he years leading up he Second World War he poenial bias beween TFP and echnology had vanished. I should also be noed ha, alhough unemploymen in 1941 was above he 1929 level, he peaceime oupu gap may have already closed. I is widely acceped ha he New Deal raised he NAIRU significanly; for example, Cole and Ohanian found ha he unemploymen rae was raised by 6 pecenage poins while Haon and Thomas (2010) esimaed ha he impac was probably even larger. Taen ogeher, hese poins mae he adjusmen suggesed by Field hard o jusify and we prefer o use he raw daa for

9 broadly based TFP growh suggess ha a yeas lie paern semming from various sources may have prevailed in he economy if no manufacuring. However, o explore hese poins properly requires a full decomposiion based on appropriae weighs and a finer breadown han he broad secors used by Field. The boom line is ha he lieraure does no conain an adequae quaniaive accoun of he indusry origins of produciviy growh during he early wenieh cenury and his implies ha exising visions of he growh process are somewha blurred. In paricular, i is no really clear wheher he process was always mushrooms, as Harberger would expec, or wheher here was a phase wih a yeas lie process a leas a he heigh of echnological progressiviy in he 1930s, as Field s accoun suggess. I is also unclear wheher elecriciy as a GPT can be credied wih promoing a yeas process of TFP growh in manufacuring in he 1920s. We see o clarify hese issues. In heoreical erms, hen, we can subdivide he underlying growh process ino wo differen ypes: generalized TFP growh, in which one dominan source of TFP growh wors wih a large effec across a large number of indusries (a yeas process); and specialized TFP growh in which, a he exreme, a source of TFP growh wors jus wihin one indusry (a mushroom process). 5 In urn, specialized TFP growh can be divided ino wo cases: universal specialized TFP growh, in which a very large number of indusries experience TFP growh from an indusry specific source; and sporadic specialized TFP growh, in which only a limied number of indusries experience TFP growh from an indusry specific source. This means ha he observaion of a yeas paern can reflec four differen underlying processes: zero TFP growh (universal sagnaion), universal specialized TFP growh only, generalized TFP growh only, and finally, a combinaion of generalized TFP growh and universal specialized TFP growh. In he conex of a Harberger diagram, hese processes would be observaionally equivalen and furher informaion would be required o discriminae beween hem. Of course, observing a mushroom paern is no consisen wih any of hese four underlying processes. Using he above erminology, i is clear ha Harberger s narraive accoun of mushrooms mixes wo differen underlying processes ha resul in observing eiher a mushroom or a yeas paern. When he menions 1001 differen causes of TFP growh, his appears very much o refer o a growh process of universal specialized TFP growh (he vas majoriy of indusries experience subsanial indusry specific TFP growh), which acually is liely o lead o observing a yeas paern. When Harberger remars ha he 1920s were he decade of cars, he 1930s of household appliances and he 1940s of anibioics, he appears o refer o a growh process of sporadic specialized TFP growh (a small number of indusries experience subsanial indusry specific TFP growh), which would lead us o expec a mushroom paern. 2. Daa and Mehods The definiive sudy on produciviy growh a he indusry level in he Unied Saes for he firs half of he wenieh cenury is Kendric (1961). Alhough Kendric offered subsanial deail, he esimaes presened in his boo fall some way shor of wha is required for a full empirical evaluaion of he ideas of Field and Harberger. Kendric provides esimaes of average annual TFP growh for 1899 o 1953 and for sub divisions of hese years for he aggregae of he privae domesic economy and for 5 secors 5 An inermediae ype of growh would be a process in which a source of TFP growh has an effec on more han one indusry, bu he number of indusries remains limied. 6

10 which in urn are divided ino 33 indusries. These secors covered 54 per cen of he privae domesic economy in TFP growh for he remaining 46 per cen (which includes consrucion, disribuion, finance, and mos of he res of he services secor) was obained arihmeically by comparing he oal of he covered secor wih he esimaes for he whole privae domesic economy; for 1899 o 1953, TFP in he covered secor was esimaed a 2.1 per cen per year, for he oal privae domesic economy a 1.7 per cen and he residual secor was calculaed as 1.3 per cen (Kendric, 1961, p. 137). Kendric s concep of TFP growh is based on he growh rae of real value added minus he facorshare weighed sum of he raes of growh of capial and labor inpus as in equaion (2) A Y A Y K L ( 1 ) (2) K L where α is he share of profis in value added. For he privae domesic economy and for he 5 main secors, labor inpus are based on man hours weighed by average hourly earnings o capure increases in labor qualiy resuling from shifs of worers beween differenly paid occupaions and indusries. Wihin sub secors, however, labor qualiy is assumed o remain unchanged. Kendric repors TFP growh raes by secor and by sub secor bu does no provide esimaes of value added shares or of he produciviy growh conribuion of each indusry (i.e., is TFP growh rae muliplied by is valueadded weigh). We ae Kendric s sudy as a saring poin bu exend and improve upon his wor. We provide esimaes of TFP growh for a more complee se of indusries and herefore have a much smaller residual secor. In paricular, we add esimaes for consrucion, disribuion, he financial (FIRE) secor, specaor enerainmen, and he pos office. However, we were unable o find sources ha would allow esimaes for he healh care, hoels and resaurans, and waerwors secors. Our exensions mean ha coverage goes up from 55 per cen o 80 per cen of he privae domesic economy on average in We also provide a full se of value added weighs and produciviy growh conribuions a he indusry level and we derive Harberger diagrams of produciviy growh conribuions boh for manufacuring and for he privae domesic economy. We also consruc esimaes for a sub period raher han he years o address he issues raised by Field (2006) and we consider he implicaions of replacing Kendric s periods and by and In maing hese esimaes, we ae fuller accoun of labor qualiy by allowing for he impac of he rapid increase in educaional aainmen in he firs half of he wenieh cenury and we do so a he indusry level basing our index of labor inpus on age, gender, and educaion of he worforce. For he inclusion of he five hard o measure secors namely, finance, insurance and real esae (FIRE), consrucion, wholesale and reail disribuion, specaor enerainmen, and posal services we esimaed capial, labor, facor income shares and oupu for he sar and end daes of our periods using a variey of sources ha included censuses, he Naional Income and Produc Accouns (NIPA), as well as a variey of secondary sources. From hese esimaes, TFP growh raes were calculaed and weighed by value added shares inerpolaed from he benchmar years, 1899, 1929, and 1939 o obain esimaes of each secor s inensive growh conribuion; see Appendix B for full deails. The second difficul as is o develop indusry level esimaes ha exend Kendric s esimaes for hrough Our approach here sars by obaining indusry level esimaes of nominal value added from NIPA and hen deflaing on he basis of wholesale prices from he U.S. Bureau of Labor 7

11 Saisics, producion prices from Hisorical Saisics of he Unied Saes (Carer e al., 2006), and, for some service secors, relevan price indices from NIPA. Labor inpus are based on NIPA for employmen adjused for hours of wor using Kendric (1961) and Hisorical Saisics and for qualiy using he mehod deailed below. Capial inpus are esimaed on he basis of capial services. We esimae he indusry level soc of capial for he privae domesic economy beween 1929 and 1941 using a Perpeual Invenory Mehod (PIM) and he invesmen and depreciaion series aen from he BEA s Fixed Asses ables. 6 Renal prices of asses a he indusry level are based on he impued indusry rae of reurn, he asse specific rae of depreciaion and capial gains and losses from changing asse prices. This allows he calculaion of capial compensaion weighs o aggregae he capial inpu. 7 We presen full deails in Appendix C. Three feaures of he mehods ha we employ deserve some commen. These relae o he way in which indusrial conribuions o aggregae TFP growh are measured, o he way in which we measure labor qualiy, and o he summary saisic used o describe he yeasiness of produciviy growh. Following Kendric (1961), we employ a growh accouning echnique based on value added raher han gross oupu. This also mirrors he approaches o examining conribuions o TFP growh adoped by Field (2011) and Harberger (1998). This implies ha we ae he conribuion o TFP growh of indusry j as ω j (ΔA j /A j ) where ω j is value added in indusry j divided by GDP and we sum hese individual conribuions of all n indusries o obain TFP growh for he aggregae privae domesic economy so ha A A GDP GDP n j1 A j A j j (3) This approach can be inerpreed as measuring an indusry s capaciy o conribue o economy wide produciviy bu he componens of his aggregae are no an accurae measure of disembodied echnical change (OECD, 2001). 8 An indusry s inensive growh conribuion (IGC) herefore depends no only on is rae of TFP growh bu also on is size and i follows ha inensive growh conribuions by indusry are no necessarily highly correlaed wih TFP growh raes. Our approach o measuring labor qualiy improves on ha of Kendric, in paricular, by aing accoun of he implicaions of he rapid increase in years of schooling on he qualiy of worers in each occupaion over ime bu also by allowing for changes in gender composiion and experience of he labor force. Our mehod also permis measuremen of labor qualiy a he indusry level. So for all indusries we conduc growh accouning wih measuremen of labor inpu based on qualiy as well as hours wored. No surprisingly, his mehod finds a higher rae of growh of labor qualiy; labor qualiy 6 Noe ha Gordon (2016) argues ha he official invesmen and depreciaion raes from he BEA severely underesimae he growh of capial inpus for he period beween 1925 and In paricular, he quesions wheher he depreciaion raes, which are fixed over ime, are represenaive for he Depression era. Appendix C explores he impac ha he adjusmen o he official depreciaion raes, along he lines suggesed by Gordon, would have on TFP growh beween 1929 and I is no possible o use his mehod o replace Kendric s esimaes of capial inpus for he pre 1929 period which are obained using a radiional capial socs mehod. In paricular, we lac asse by indusry capial socs and asse specific depreciaion raes. I should be noed ha he differences in TFP growh beween he wo mehods are generally fairly small; he implicaions are explored in Appendix C. 8 An alernaive which has his desirable propery is o do growh accouning on a gross oupu basis where he use of inermediae inpus is explicily aen ino accoun and aggregaion is based on Domar weighs (cf. Hulen, 1978). This is oo daa demanding o be aemped here, given he available hisorical sources. 8

12 in he privae domesic economy during is esimaed o have grown a 0.8 per cen per year compared wih Kendric s esimae of 0.3 per cen. To consruc an index of labor inpu for each individual indusry, we assume ha labor inpu ( ) for indusry j be expressed as a ranslog funcion of is individual componens. We form indices of labor inpu from daa on employmen by indusry, cross classified by gender, age and educaion. Dropping he indusry subscrip j for ease of noaion, he growh of labor inpu for indusry j can hus be represened as HK HK q l1 v l L L l l (4) where is employmen a he indusry level for a given se of q characerisics of he labor force l (gender, age and educaion) and is he wo period average of his employmen group s share in he oal labor income a he indusry level. The share of labor income ( ) is derived as he produc of he average wage ( ) and employmen ( ) for each combinaion of labor characerisic l, divided by he oal wage sum v l q l1 p L l l l p L l (5) So our measure of indusry labor qualiy growh is he difference beween he growh raes of he compensaion weighed index of labor inpu and oal employmen. LQ LQ q l1 v l L L l l L L (6) We follow a hree iered approach o he daa preparaion for he labor qualiy esimaion. Firs, we esimae educaional aainmen for individual worers for he pre 1940 census samples on he basis of he 1940 reurns. Second, we consruc an employmen marix for he enire period ha groups worers according o heir (prediced) educaional aainmen, gender, age and by indusry. Lasly, we derive he compensaion marix on he basis of average wages for each labor caegory aen from he 1940 census of populaion. These employmen and compensaion marices can hen be used o calculae labor qualiy on he basis of equaion (6). Full deails of our esimaion mehod are repored in Appendix D. We use Harberger diagrams along he lines of Figure 1 o address he quesion of yeas versus mushrooms. We follow Inlaar and Timmer (2007) in using a summary saisic, he Harberger 9

13 coefficien, o describe he degree of concenraion of indusry conribuions o oal TFP growh. 9 This measures he area beween he curve and he diagonal divided by he oal area under he curve, ha is V/(V + W) and will have a lower value he more equal are he indusrial conribuions o TFP growh. The Harberger coefficien can be expressed as 1 1 A f ( x) dx xdx f ( x) dx V 2 A H (7) 1 1 V W f ( x) dx f ( x) dx where x is he cumulaive share of value added when he inensive growh conribuions are raned from larges o smalles TFP growh rae, and f(x) is he cumulaive TFP growh conribuion of his cumulaive value added, and da/a is TFP growh for he whole economy, se a uniy in his diagram. 10 So he numeraor V in expression (7) is he area beween he curve and he diagonal, and W is he riangle under he diagonal. V is calculaed by subracing he riangular area W from he oal area under he curve. When TFP conribuions are exacly proporional o value added, he Harberger curve lies along he diagonal, and V equals zero. In order o evaluae he conribuion of elecriciy o produciviy growh in each manufacuring indusry in he 1920s, we use an augmened version of an equaion similar o (1) which is as follows: Y L Y L KO K ELEC HK K ELEC L L A A 1 L L (8) KO K ELEC A A HK K ELEC O ELEC L L L L where he addiional erms, (1 α ) Δ(HK/L) / (HK/L) and Δ(K ELEC /L) / (K ELEC /L), represen he conribuions of labor qualiy growh and of (unremuneraed) TFP spillovers resuling from he growh of he elecrical capial soc, respecively. These TFP spillovers are couned as par of oal TFP growh raher han he capial deepening conribuion of elecrical capial, and if no aemp is made o measure hem hey would accrue as par of φ(δa/a) O. We calculae he rae of growh of horsepower in elecric moors (boh primary and secondary) based on he insalled horsepower by indusry repored by DuBoff (1979) as an esimae of he rae of growh of elecrical capial and we obain an esimae of by assuming ha he share of profis accruing o elecrical capial corresponds o he share of elecrical equipmen capial in oal capial. We calculaed his share by esimaing oal capial per indusry and elecrical capial per indusry using he NIPA and daa on invesmen by asse ype from he BEA for 1921 o 1929 and hen aing he average of boh over hese years (see Appendix C). I has been srongly argued by David (1991) ha TFP spillovers were imporan in he 1920s when invesmen in secondary moors delivered uni drive on machinery and faciliaed improvemens in facory design. We follow David (1991) in esimaing by a cross secion regression of he change in 9 We have coined he erm Harberger coefficien as an analogue o he Gini coefficien which describes he degree of inequaliy associaed wih a Lorenz Curve. 10 In pracice, he diagrams are no as smooh as in Figure 1, as we have a discree number of indusries. Insead hey consis of piecewise linear plos. 10

14 TFP growh (he TFP growh for minus ha for ) agains he average annual rae of growh of horsepower in secondary moors in each manufacuring indusry during These esimaes can hen be used ogeher wih TFP growh in he elecrical machinery secor o consruc a Harberger diagram of oal elecrical TFP growh conribuions. We also use equaion (8) o consruc Harberger ype diagrams of conribuions o labor produciviy growh following he lead of David and Wrigh (1999) bu in addiion disinguishing beween elecrical and non elecrical conribuions. We calculae he equivalen of he Harberger coefficien (H*) from a modified version of equaion (7) which is Y f ( x) dx L f ( x) dx 2 xdx Y H L (9) 1 1 f ( x) dx f ( x) dx 0 0 where on his occasion f(x) is he cumulaive labor produciviy growh conribuion. 3. Resuls This secion repors our esimaes of value added weighs, growh of labor qualiy, and TFP growh a he indusry level ogeher wih our esimaes of inensive growh conribuions by indusry, i.e., he secoral decomposiion of TFP growh, which are derived from hem. A he same ime, we poin ou some noeworhy feaures of hese daa. Value added weighs are repored in Table 2. I is apparen ha wholesale & reail rade and FIRE, which have been added o he measured secor, are relaively large secors aen ogeher in he inerwar period hey are nearly as large as manufacuring. A he same ime, i is also sriing ha manufacuring accouns for only abou a quarer of oal value added hroughou he period. This maes clear boh ha confining a discussion of produciviy performance o ha secor alone would be poenially quie misleading and also ha srong produciviy growh for he whole economy would normally require oher secors o mae significan conribuions. I is also worh noing ha farming was sill quie sizeable which suggess ha, following Kendric, i is appropriae o base an analysis of produciviy in he mare economy primarily on he performance of he privae domesic economy (PDE) raher han he privae non farm domesic economy (PNE). Table 3 repors esimaes of labor qualiy growh by indusry for each period. As was noed above, hese esimaes are more deailed han Kendric s and hey also show a faser rae of labor qualiy growh because hey ae accoun of improvemens in labor qualiy wihin occupaions and secors which is imporan in an era of rapidly improving educaional aainmen. For example, Goldin and Kaz (2008) repor ha while only 10.6 per cen of hose aged 14 o 17 were enrolled in high school in 1900 by 1938 his had risen o 67.7 per cen. I is imporan o noe ha he much higher rae of growh of labor qualiy in he PDE han in he PNE is very largely explained by he impac of secoral reallocaion of labor, which mainly concerned worers moving ou of agriculure, as well as by increased educaional aainmen, which provided he mos imporan conribuion. In addiion, he increased average age of 11

15 worers raised labor qualiy over he long run while an increase in he proporion of females from 18 per cen a he sar of he period o 24 per cen a he end largely offse he age effec. 11 Table 3 presens a picure no only of rapid labor qualiy growh on average bu also one of subsanial variaion beween secors and over ime. The highes figure (elephone in ) is 1.14 per cen per year while he lowes (leaher producs in ) is 0.71 per cen per year. The range in successive sub periods is 1.64, 1.20, 0.98, and 1.08, respecively. This implies ha he correcion facors for labor qualiy applied o crude TFP will be quie variable and ha relaive secoral conribuions o TFP afer hese adjusmens are made will poenially loo quie a bi differen. Ineresingly, and perhaps conrary o some priors, he correlaions beween labor qualiy growh and refined TFP growh (repored in Table 4) are quie low across our 37 secors a 0.00, 0.03, 0.02, 0.18, respecively for each sub period, and 0.10 for he whole period, The fases labor qualiy growh over he period as a whole is found in paper, rubber producs, and exiles. The raes of TFP growh repored in Table 4 are generally lower han hose in Kendric (1961) in paricular because our adjusmen for he growh of labor qualiy is larger han his; over as a whole we esimae TFP growh in he PDE a 1.3 per cen per year compared wih 1.7 per cen according o Kendric. Obviously, his sill represens a very srong performance relaive eiher o he nineeenh cenury or rivals lie he Unied Kingdom. The fases TFP growh during hese years was in for he PDE bu no for manufacuring where TFP growh was much faser in he 1920s. I is clear ha srong performance in he 1930s was relaively broadly based and involved much of he services secor, including our residual secor. During , he op hree secors in erms of TFP growh were enerainmen, elecric uiliies and ranspor equipmen, all of which can be considered par of he second indusrial revoluion. Each of hese secors made regular appearances in he op 5 hroughou he period bu a furher 9 secors feaured a leas once in he op 5. More generally, ran correlaion coefficiens for secoral TFP performance beween successive periods were quie low (0.4, 0.0, and 0.2). There are 24 observaions (abou 16 per cen) wih negaive TFP growh; 13 of hese were for which may have been affeced by World War I. I is noiceable ha he 6 secors whose TFP growh fell by a leas 2.0 percenage poins beween and showed an average improvemen in TFP growh of 4.9 percenage poins beween and Table 5 displays esimaes of secoral inensive growh conribuions (IGC). I is noiceable ha wih he excepion of , he sum of negaive IGC of measured secors is very small below 10 per cen of oal TFP growh in each decade. The IGC depends boh on TFP growh and also on a secor s size. Indeed, he secor wih he fases TFP growh rae never had he larges IGC in any period alhough ran correlaions beween TFP growh and IGC are reasonably high mos of he ime, namely, 0.6, 0.9, 0.4, and 0.8 in successive periods. To faciliae comparisons, Table 6 provides raning of secors by TFP growh and by IGC in each period. The op 3 IGC secors during were wholesale and reail rade, railroads, and foods, none of which would be hough of as exciing new, echnologically progressive indusries. 12 Over he whole 11 The large number of secors wih negaive labor qualiy growh in reflecs endencies of he worforce becoming younger and more female a a ime when educaional aainmen was rising less quicly han in laer decades. 12 They were no idenified by Mowery and Rosenberg (1989) as especially R & D inensive. The leaders in ha respec were chemicals, peroleum, and elecrical machinery. 12

16 period , wholesale and reail rade (wih a value added weigh of 13.5 per cen), which benefied from improvemens in ranspor and communicaions and increased sore sizes (Field, 2011) bu was no a he hear of he second indusrial revoluion, provided he larges IGC bu raned only 24 h in TFP growh. Liewise, farming was a large, unglamorous secor (wih a value added of 10.1 per cen) which had low TFP growh bu raned sixh in IGC over he whole period, and hird in he 1930s, a leas parially by benefiing from second indusrial revoluion innovaions such as pesicides, animal medicines, he combusion engine, and elecriciy, which was needed in vas quaniies for he Haber Bosch process o mae arificial ferilizer (Olmsead and Rhode, 2008). Manufacuring s IGC dominaed in when i accouned for abou 70 per cen of all TFP growh bu his was excepional and is average conribuion over was only abou 40 per cen. This was, however, sill well above is value added weigh of abou a quarer of GDP. The four echnologies ha Gordon (2000) idenified as driving his one big wave of TFP growh conribued increasingly over he period, rising from percenage poins in o percenage poins in wih an average IGC over he whole period of percenage poins, almos as big as he conribuion of he enire manufacuring secor (38 vs. 41 per cen) The Mos Technologically Progressive Decade? Field (2011) repeaed his 2003 argumen ha he 1930s, defined as , were he mos echnologically progressive decade of he wenieh cenury. This was based on ha period having he fases TFP growh in he privae non farm economy (PNE) and also ha he 1930s saw unusually broadly based TFP growh wih he inensive growh conribuion (IGC) of ranspor and uiliies combined wih wholesale and reail rade being roughly equal o ha of manufacuring wih each accouning for abou 47 per cen of TFP growh in he PNE. In his secion, we re examine boh hese claims using he new esimaes ha we repored in secion 3. In Table 7, we repor TFP growh raes for he privae domesic economy (PDE) over he long wenieh cenury where, as in Field (2011), he pos 1948 esimaes are aen from he Bureau of Labor Saisics (2014). As remared in secion 3 above, we follow Kendric raher han Field in basing our comparisons on he PDE raher han he PNE. We believe our esimaes are comparable for wih hose of he BLS for pos 1948 in ha capial and labor inpus are esimaed on he same basis (bu differenly from Kendric on whose esimaes Field relied). I is clear from Table 7 ha he 1930s did no have he fases TFP growh bu were below he TFP growh of he periods 1948 o 1960 and 1960 o Raher han singling ou he 1930s as having he highes TFP growh, i may be preferable o see TFP growh in he peaceime American economy seadily increasing from he 1920s hrough he 1960s (Gordon 2000). 13 We ae he big wave secors o be Chemicals, Peroleum & coal producs, Rubber producs, Elecric machinery, Elecric uiliies, Transpor equipmen, Wholesale and reail rade, Local ransi, Telephone, and Specaor enerainmen. The laer has been added o he Kendric big wave secors and is largely based on Baer (2012). 14 If he period 1948 o 1973 is considered, as in Field (2011), he TFP growh rae for he PDE is 2.12 per cen per year which is also superior o If he variable reiremen mehod proposed by Gordon (2016) is applied (see foonoe 5 above and Appendix C, ables C5 and C6), hen he PDE growh rae for reduces from 1.9 o 1.7 percen per annum, even hough i says he same a 1.3 percen per annum for he whole period Use of his mehod would undermine Field s claim ha TFP growh in he 1930s was he fases in he wenieh cenury. Also, wih Gordon s mehod, he Harberger coefficien increases appreciably from 0.35 o 0.38 which would go agains Field s view ha he 1930s saw more broadly based TFP growh han he 1920s, see secion 5. However, i should be noed ha no evidence is available o validae Gordon s assumpions abou delayed reiremen of capial. 13

17 In Table 8, we re wor Field s secoral decomposiion of TFP growh in he PNE which compares he 1930s wih he 1920s. We use he esimaes of TFP growh and value added weighs se ou in secion 3 and mid period raher han end period weighs, parly because his is normal pracice and parly because he manufacuring secor s weigh increased appreciably as he American economy moved owards war in In his case, our disagreemen wih Field (2011) is ha he significanly undersaed he conras beween he wo decades. 15 Our esimaes show ha, whereas he IGC for manufacuring in he 1920s accouned for 74.1 per cen of TFP growh in he PNE, in he 1930s his had fallen o 36.0 per cen as TFP growh in manufacuring almos halved. Indeed, during 1929 o 1941, we esimae ha ranspor and public uiliies ogeher wih wholesale and reail rade were no on a par wih manufacuring bu raher accouned for a much larger share of TFP growh in he PNE a 52.0 per cen. However, in order o obain a clearer picure he exen o which he secoral paern of TFP growh in he 1930s differed from wha had gone before, i is imporan o develop a perspecive based on a less coarse decomposiion of inensive growh conribuions and o ae a longer erm view. This as is underaen in he following secion using he device of he Harberger diagram. 5. Yeas or Mushrooms? The difference beween yeasy and mushroomy paerns of inensive growh conribuions (IGC) was reviewed in he conex of Figure 1. In Figure 2, we presen Harberger diagrams based on TFP growh conribuions obained from our daase for he period To our nowledge, his is he firs ime his has been aemped. In drawing hese diagrams for he PDE, we base hem on our aggregae measured secor, i.e., we have excluded he residual secor. We believe his is appropriae because he residual secor comprises an odd assormen of secors which may well have experienced quie differen raes of TFP growh. Combining hem is liely o exaggerae he degree of yeasiness in when he residual secor s TFP growh was 1.4 per cen per year (Table 4). 16 The picure ha emerges from Figure 2 is ha is quie lie no only he 1920s bu also he firs decade of he wenieh cenury. In each case, he visual impression is relaively yeasy and he Harberger coefficiens are quie similar wih 0.35 for being marginally lower han 0.37 for bu slighly higher han 0.33 for In each period here are few secors wih negaive TFP growh and heir value added share in he measured secor is fairly small. In conras, , saw lower TFP growh han he oher periods and has a Harberger coefficien of 0.71, loos very mushroomy and is somehing of an oulier wih negaive TFP growh in many secors. If he residual secor had been included as an addiion o he Harberger diagram, his picure would essenially remain bu he deails would change wih Harberger coefficiens of 0.42, 0.59, 0.40, and 0.33 for he successive periods (and 0.30 for he enire period). In fac, here seems o be a srong inverse correlaion beween he size of he Harberger coefficien and he aggregae rae of TFP growh as Inlaar and Timmer (2007) sress in heir review of he evidence for he mare secor in OECD economies around he urn of he 21 s cenury. I seems ha relaively rapid TFP growh is ypically associaed wih a more balanced and broadly based paern across secors. Tha said, i is worh noing ha he Harberger coefficiens for he American economy in he inerwar period 15 I should, however, also be noed ha our esimae of TFP growh in he PNE a 1.82 per cen per year for 1929 o 1941 is below he BLS esimae of 1.88 per cen for he PNE in 1948 o 1973 repored in Table The residual secor includes chariies, healhcare, hoels & resaurans, waerwors and miscellaneous services. 14