The World Technology Frontier: What Can We Learn from the US States?*
| Author | Jakub Growiec |
| DOI | http://doi.org/10.1111/j.1468-0084.2011.00686.x |
| Published date | 01 December 2012 |
| Date | 01 December 2012 |
777
©Blackwell Publishing Ltd and the Department of Economics, University of Oxford 2012. Published by Blackwell Publishing Ltd,
9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA.
OXFORD BULLETIN OF ECONOMICS AND STATISTICS, 74, 6 (2012) 0305-9049
doi: 10.1111/j.1468-0084.2011.00686.x
The World Technology Frontier: What Can We
Learn from the US States?*
Jakub Growiec
Institute of Econometrics, Warsaw School of Economics, al. Niepodleg`o´sci 162, 02-554
Warswawa, Poland and Economic Institute, National Bank of Poland, Warszawa,
Poland (e-mail: jakub.growiec@sgh.waw.pl)
Abstract
We re-estimate the world technology frontier non-parametrically using a dataset covering
OECD country-level data and US state-level data on GDP per worker and the stocks of
physical capital, unskilled labour and skilled labour. The auxiliary use of US state-level
data significantly reduces the upward bias in cross-country estimates of technical effi-
ciency, and so does allowing for imperfect substitutability between skilled and unskilled
labour. We then use our adjusted estimate of the world technology frontier in a series of
decompositions of productivity differences and sources of economic growth in the OECD
in 1970–2000, including also ‘appropriate technology vs. efficiency’ decompositions.
I. Introduction
Is it possible to use production factors more efficiently than in the US? Studies based
on aggregate cross-country data provide, almost unanimously, the negative answer: since
World War II, the US level of per-worker productivity has always been high enough to
guarantee that the US was one of the countries spanning the world technology frontier
(WTF).1Consequently, all post-war improvements in US productivity have been identi-
fied as either due to factor accumulation or technological progress at the frontier.We claim,
however, that since the US is a huge country with substantial internal heterogeneity, we
can learn more about the evolution of US productivity if we cease considering it as a single
data point, as habitually done in earlier studies. US state-level data show that it is possible
ÅI am grateful to Aleksandra Iwulska and Łukasz Mar´c for their help with the data and to Mateusz Zawisza for
performing the bootstraps. The article has benefitted considerably from the useful comments and suggestions of the
Editor Jonathan Temple, as well as Maciej Bukowski, Andrzej Cie´slik, Jacek Osiewalski, Małgorzata Pawłowska
and Artur Pr¸edki. Most of this research has been done at the Institute for Structural Research, Warsaw, Poland. The
working paper version of this text was circulated under the title: ‘Productivity Differences Across OECD Countries
and US States, 1970–2000: The WorldTechnology Frontier Revisited’. All errors are my responsibility.
JEL Classification numbers: E23, O11, O14, O33, O47.
1This is one of the conclusions of non-parametric studies by Kumar and Russell (2002), Henderson and Russell
(2005), Jerzmanowski (2007) and Badunenko, Henderson and Zelenyuk (2008). In Caselli and Coleman (2006) as
well as Badunenko, Henderson and Russell (2009), the US is found to fall behind the frontier, albeit very slightly.
778 Bulletin
to produce more efficiently than the US does on average. Even more interestingly,thanks
to the generally high productivity across US states, disaggregating the single US data point
into its constituent states should also lead to significant improvements in the precision of
estimates of the entire WTF. Hence, this study will combine information from US states
with country-level data to provide new estimates of the WTF.
The contribution of this study to the literature is threefold. First and foremost, as
announced just above, we shall use US state-level data to revisit the economic debate
on the shape of the WTF as well as on the sources of economic growth and cross-
country productivity differences. The most important insight here is that appending the US
sub-national dataset to the international one leads to a marked increase in the precision of
non-parametric [data envelopment analysis (DEA)-based] WTF estimates, especially in the
range of factor endowments observed across the US. Furthermore, thanks to a more accurate
approximation of the WTF, the reliability of earlier growth and development accounting
exercises can be substantially improved, too. These accomplishments are complementary
to the ones obtained thanks to known DEA bootstrap techniques (Simar and Wilson,1998,
2000; Kneip, Simar and Wilson, 2008): the advantage of our approach is that we add new
valid data points to the dataset considered, carrying genuine additional information.
The second novelty of the current study with respect to the established literature is
that we allow for imperfect substitutability between skilled and unskilled labour.2This
leads to a further refinement of results presented in earlier studies. As far as we know, this
decomposition has never been used before in non-parametric analyses of the kind adopted
here.
The third contribution of the current study is to propose a novel decomposition of coun-
tries’ productivity growth rates, indicating the extent to which the observed productivity
changes represent shifts of the WTF, or movements along the WTF.
As far as the territorial coverage of the current study is concerned, we focus only on
highly developed OECD countries located in Europe and North America (plus Australia
and Japan), and set aside developing and transition economies as well as small open econ-
omies such as the Asian Tigers.This will reduce precision in the estimation of the WTF in
the region of low capital and/or human capital endowments, where numerous developing
countries are located. Such an approach does not compromise the precision of efficiency
estimates in the range relevant to our study, though, under the rather innocuous assumption
that non-OECD countries do not operate at the same factor ratios as OECD countries or,
if they do, they are less efficient than at least one OECD country. In consequence, rather
than attempting to identify the whole WTF, we aim at obtaining its best possible estimates
in the range associated with factor ratios observed in the OECD countries (or US states).
At the same time, this approach also makes our results less vulnerable to poor data quality
(see the discussion about Sierra Leone spanning the WTF in Kumar and Russell, 2002).3
The time period considered is 1970–2000, and technologies from all earlier years are
allowed to span the WTF in the given year alongside current ones (cf. Henderson and
Russell, 2005). Indeed, it turns out that even some technologies used in 1970 remain effi-
2See Caselli and Coleman (2006) and Pandey (2008).
3We also use bootstrapping techniques to adjust for the inherent bias in efficiency estimates, thus somewhat neu-
tralizing the impact of outlying observations (cf. Simar and Wilson,1998, 2000; Kneip et al., 2008; Badunenko et al.,
2009).
©Blackwell Publishing Ltd and the Department of Economics, University of Oxford 2012
To continue reading
Request your trial