American Economic Association

Productivity Growth, Technical Progress, and Efficiency Change in Industrialized Countries:

Reply
Author(s): Rolf Färe, Shawna Grosskopf and Mary Norris
Source: The American Economic Review, Vol. 87, No. 5 (Dec., 1997), pp. 1040-1044
Published by: American Economic Association
Stable URL: http://www.jstor.org/stable/2951341 .
Accessed: 17/07/2014 05:55

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

American Economic Association is collaborating with JSTOR to digitize, preserve and extend access to The
American Economic Review.

http://www.jstor.org

This content downloaded from 82.34.208.77 on Thu, 17 Jul 2014 05:55:53 AM

All use subject to JSTOR Terms and Conditions
ProductivityGrowth,Technical Progress, and EfficiencyChange
in IndustrializedCountries:Reply
AND MARYNoRRIs *
By ROLFFARE, SHAWNAGROSSKOPF,
In their comment, Subhash C. Ray and
Evangelia Desli (1997) (hereafter RD) point
out that the specification of the decomposition
of the Malmquist productivity index used by
Fare et al. (1994) (hereafter FGNZ) is not
unique, and propose and compute an alter-
native specification of that decomposition.
We will discuss additional decompositions at
the end of this note, but proceed here by com-
paring the RD decomposition with FGNZ
based on both conceptual and computational
grounds.
RD provide a discussion of the overall
Malmquist productivity index, including the
important issue of when this index is equiva-
lent to the traditionalnotion of total factor pro-
ductivity (TFP) -namely underthe condition
that the reference technology be consistent
with constant returnsto scale (CRS). As they
point out, this will yield a measure of TFP
even if the "true" underlying technology is
not CRS, for example. Both RD and FGNZ
use the CRS reference technology to compute
overall Malmquist productivity.
One of the key issues raised is the role of
the underlying scale properties of the bench-
mark technologies used to define and compute
both productivity and its components. In par-
ticular, two "reference" technologies are em-
ployed in both RD and FGNZ: what we refer
to as CRS and variable returnsto scale (VRS)
* Fare and Grosskopf: Department of Economics,
Southern Illinois University, Carbondale, IL 62901;
Norris: USAID Madagascar, Department of State,
Washington, DC 20521. While working on this reply,
Grosskopf was holding the Kerstin Hesselgren Visiting
Professorship,which is funded throughthe Swedish Coun-
cil for Research in the Humanities and Social Sciences.
Gothenburg University graciously hosted Fare and
Grosskopf during this time. We are grateful for their
support.
1040
This content downloaded from 82.34.208.77 on Thu, 17 Jul 2014 05:55:53 AM
All use subject to JSTOR Terms and Conditions
technologies.' By construction,these technolo-
gies are nested:the CRS technology "contains"
the VRS technology, as in Figure 1 in RD. This
nestednessprovidesthe logical basis for our de-
composition.At a very intuitivelevel, we would
arguethat these two benchmarkscan be used to
provide bounds on the underlying true-but
unknown-technology.2 Intuitively we see the
VRS technology providing a type of convex
"inner approximation,"whereasthe CRS tech-
nology providesa type of convex "outerapprox-
imation." Thus these two technologies provide
alternativebenchmarks;they do not requirethat
the datasatisfyeitherCRS or VRS. Anotherpos-
sible intuitiveinterpretation
is thatthe CRS cap-
tures a (perhapshypothetical) "long run" and
the VRS approximatesthe shortrun.
As a reference technology, the CRS tech-
nology has some very useful features; for ex-
ample, it captures the notion of maximal
'The two reference technologies may be written for
computationalpurposes as
S'(CRS)
K
(X Zkykm Ym 1 M,
k= I
K
.ztx xt, n = 1, N,
Zk _=0 k =1, ..,K.
S'(VRS) is defined identically with the additional con-
straint on the intensity variables that they sum to unity,
i.e., y K= l Zt = 1, where the observed inputs and outputs
are denoted by x' = (xt I, x.N) and yk = (yI .
YkM), k = 1,. K, respectively. In the CRS model,
S'(CRS) the intensity variables z4, k = 1,. K are re-
stricted to be nonnegative, while in the VRS model, they
are also requiredto sum to one, providing the "tighter fit"
and nested propertymentioned above.
2 RD apparentlytake the view that the VRS technology
is the "true" technology.
VOL 87 NO. 5 FARE ET AL: PRODUCTIVITYGROWTH,REPLY
average product (consistent with the minimum
point on a long-run, U-shaped average cost
curve) which provides a very nice benchmark
for identifying both optimal scale in the one
period case (also pointed out by RD), and
technical change in the multiperiod case (as
we discuss below).
Turning to the decomposition in RD versus
FGNZ, we note that the VRS efficiency
change term is identical in the two decompo-
sitions, so we turn to the remaining two com-
ponents: technical change and scale change.
The technical change term used in both RD
and FGNZ can be summarized as
(1) TECHCH(j)
[DJ(xo, yo)D(xl, yl )
LD((x,y yo)D (xi, yi)
where = CRS in FGNZ andj = VRS in RD.
In FGNZ we think of technical change as
change in maximal average product between
period t and t + 1, and we exploit the fact that
CRS technology captures maximal average
product to specify and compute technical
change.3Otherthan the nice intuition the max-
imal average product notion provides, we note
that this is also consistent with the notion of
technical change as defined by Robert A.
Solow (1957). RD define technical change us-
ing the VRS technology as a reference; they
capturethe shift in VRS technology evaluated
at the two different observed input levels, but
that shift is not the change in maximal average
product.Computationallythe RD method may
pose some difficulties (as in the case of
Ireland, for which an "infeasible solution" is
reported in Table 1 ) since their technical
change component includes mixed-period dis-
tance functions computed under VRS.4
' This interpretationholds even if the "true" technol-
ogy were VRS, for example.
4 Specifically, computing technical change requires
computing "mixed-period" distance functions like
D (xk' +, y' + 1j)
This content downloaded from 82.34.208.77 on Thu, 17 Jul 2014 05:55:53 AM
All use subject to JSTOR Terms and Conditions
1041
Perhaps the most controversial component
(due perhaps to our exposition) in the FGNZ
decomposition is the scale change component.
The discussion of identification and compu-
tation of scale efficiency in the single-period
case in RD is useful here. They provide an
intuitive discussion of the role of the CRS
technology as a benchmarkin identifying what
Rajiv D. Banker (1984) calls the most pro-
ductive scale size (MPSS), and how that may
be used to identify deviations from that opti-
mal size in a given period; for example, from
equation (8) in RD, scale efficiency would be
computed in period "0" as
(2) SE?(xo,yO) =D?(xo,yo)/D?(xo,yo),
112 where the c and v subscriptsrepresentCRS and
VRS technologies, respectively. Similarly,fol-
lowing Banker, one could compute the single-
period scale efficiency in period " 1 " as5
(3) SE'(x,,yI)=D'(xI,yI)/D'(xI,yI).
Using RD notation, we can write the FGNZ
scale change term as
(4) SCHFGNZ = SE' (xl, y,) / SE0(xo, yo).
Thus FGNZ scale change is simply the ratio
of deviations from optimal size in period 0
compared to period 1; it is the ratio of the (re-
ciprocal) of the Banker definition of scale ef-
ficiency in those two periods.6 The intuitive
where in FGNZ, j = CRS and in RD, j = VRS. Under
VRS (1) may not have a finite solution, even when all
observed inputs and outputs are strictly positive, making
it impossible to evaluate RD technical change for thatob-
servation k'. In contrast, under CRS, (1) has a solution
and an associated feasible value for technical change, as
seen, for example, in the case of Ireland in the FGNZ
decomposition in Table 2 of RD.
5 We note that RD define the period " 1" scale effi-
ciency relative to period "0" technology. The Bankerno-
tion of scale efficiency is, however, a cross-sectional or
one-period notion, and thus would not have technology
from a period different than the observation being
evaluated.
6 This provides a link between the decomposition of
productivity over time and the decomposition of technical
efficiency in one period. Fare et al. (1983) derive a de-
composition of DC(x, y) into scale efficiency and D,(x,
y), which is the cross-sectional counterpartof our effi-
ciency change and scale change components.
1042 THE AMERICANECONOMICREVIEW
interpretationof the scale change component
is whether an observation got closer to or far-
ther from most productive scale size between
the two periods, where MPSS may differ from
period to period. In contrast, RD define scale
change as
(5) SCHRD
= SE1(xl, yl) SE0(xl, yl) 11/2
[SEo(xo, yo) SEl(x0,y0) J
The first ratio in the squarebracketsis iden-
tical to the FGNZ scale component, and the
second ratio includes what look like scale ef-
ficiency terms defined across periods. The sec-
ond ratio thus requires the computation of
distance functions like D?(xj, Yl), in which
data from one period ( " 1") are evaluated rel-
ative to a VRS benchmark from the previous
period. Again, as seen in the case of Ireland,
this can cause computationalproblems. These
problems do not arise in the FGNZ decom-
position precisely because it contains no
mixed-period terms computed under VRS.
Finally, the RD scale term may incorrectly
identify the scale properties of the underlying
technology.7
To sum up, RD provide an alternative spec-
ification of the FGNZ decomposition of the
Malmquist productivity measure. The result-
ing overall measure of productivity is iden-
7 As the following example shows, the RD scale com-
ponent may reportvalues that do not reflect the scale prop-
erties of the data. Suppose we have two firms, A and B,
which use one input to produce a single output.
Firm (xo, yo) (xI, y1)
A (6,3) (5,5)
B (4,2) (3, 3)
Based on this sample, both firms are producing at a point
of CRS in each period, t = 0, 1. Since both firms were at
MPSS in both periods, the FGNZ scale component reg-
isters a value of one, reflecting no change in scale effi-
ciency, as it should. The RD decomposition gives a scale
change value for firm A of (2/1.66667)/2, which is ap-
proximately 1.095, suggesting an improvement in scale
efficiency, which did not occur. For firm B in the RD
decomposition, there is no solution.
This content downloaded from 82.34.208.77 on Thu, 17 Jul 2014 05:55:53 AM
All use subject to JSTOR Terms and Conditions
DECEMBER 1997
tical in RD and FGNZ; in that sense, both are
"internally consistent." Both RD and FGNZ
rely on both CRS and VRS technologies to
identify their components of productivity
change, although the specification and inter-
pretation of the technical change and scale
change components are different, as are, of
course, the empirical values of these compo-
nents. Computationally, the RD scale and
technical change terms may not always be
computable; the corresponding FGNZ terms
are computable.
One of the messages of RD is that there is
more than one way to decompose an index.
We would like to point out two other varia-
tions for the interested reader.Based on an al-
ternative distance function definition of
productivity which W. Erwin Diewert ( 1992)
dubbed the "Hicks-Moorsteen" index, Hans
Bjurek (1994) derives a decomposition into
technical change and efficiency change. Like
the FGNZ decomposition, the components of
this index are not susceptible to nonexistence
of solutions to the individual distance func-
tion problems.8C. A. Knox Lovell and Emili
Grifell-Tatje ( 1994) propose a similar form of
productivity index using distance functions,
with a slightly different decomposition.
Perhaps the main contribution of RD is to
raise the issue of the interpretationof the role
of the reference technology in computing pro-
ductivity. Perhapsmost importantly,they have
confirmed that the correct benchmarkin com-
puting overall Malmquist productivity (if it is
to be a measure of TFP) is a CRS benchmark,
a source of much confusion in the literature.
In pointing out how to interpretthe use of CRS
and VRS technologies in the computation of
scale efficiency, for example, they are, in ef-
fect, saying that we are using these technolo-
gies as benchmarks rather than arguing, for
example, that the true technology is CRS. In
fact we agree with this general viewpoint-
the "true" technology is unknown; however,
thanks to the techniques we use, we can con-
struct useful benchmarks based on observed
data to use in identifying productivity growth
and its sources.
8 For a more detailed discussion, see Fare et al. ( 1996).
VOL. 87 NO. S FARE ET AL: PRODUCTIVITYGROWTH,REPLY 1043

REFERENCES Technical Progress, and Efficiency Change

Banker,RajivD. "Estimating Most Productive
Scale Size Using Data Envelopment Anal-
ysis." European Journal of Operational
Research, 1984, 17(1), pp. 35-44.
Bjurek, Hans. Essays on efficiency and pro-
ductivity change with application to
public service production. Ekonomiska
Studier, No. 52, Goteborgs Universitet,
Sweden, 1994.
Diewert,W.Erwin."Fisher Ideal Output,Input,
and Productivity Indexes Revisited." Jour-
nal of Productivity Analysis, September
1992, 3(3), pp. 211-48.
Fare, Rolf; Grosskopf,Shawnaand Lovell,C. A.
Knox. "The Structure of Technical Effi-
ciency." Scandinavian Journal of Econom-
ics, 1983, 85(2), pp. 181-90.
Fare,Rolf;Grosskopf,Shawna;Norris,Maryand
Zhang, Zhongyang. "Productivity Growth,
in Industrialized Countries." American
Economic Review, March 1994, 84(1), pp.
66-83.
Fare, Rolf; Grosskopf, Shawna and Roos,
Pontus. "On Two Definitions of Produc-
tivity." Economics Letters, 1996, 53, pp.
269-74.
Lovell, C. A. Knox and Grifell-Tatje,Emili. "A
GeneralizedMalmquistProductivityIndex."
Paper presented at the Georgia Productivity
Workshopat Athens, GA, October 1994.
Ray, SubhashC. and Desli,Evangelia."Produc-
tivity Growth, Technical Progress, and Ef-
ficiency Change in IndustrializedCountries:
Comment." American Economic Review,
December 1997, 87(5), pp. 1033-39.
Solow, Robert A. "Technical Change and the
Aggregate ProductionFunction." Review of
Economics and Statistics, August 1957,
39(3), pp. 312-20.

This content downloaded from 82.34.208.77 on Thu, 17 Jul 2014 05:55:53 AM

All use subject to JSTOR Terms and Conditions
1044 THE AMERICANECONOMICREVIEW DECEMBER 1997

This content downloaded from 82.34.208.77 on Thu, 17 Jul 2014 05:55:53 AM

All use subject to JSTOR Terms and Conditions