You are on page 1of 11

Centre for Efficiency and Productivity Analysis

Working Paper Series


No. WP01/2010







DPIN Version 1.0: A Program for Decomposing Productivity Index Numbers

C.J. ODonnell




Date: March 2010



School of Economics
University of Queensland
St. Lucia, Qld. 4072
Australia


ISSN No. 1932 - 4398
DPINVERSION1.0:APROGRAMFORDECOMPOSINGPRODUCTIVITYINDEXNUMBERS

by

C.J.ODonnell

DPIN Version 1.0 is a fast userfriendly program for computing and Decomposing Productivity Index Numbers. The
program uses the conceptual framework developed by O'Donnell (2008) and the data envelopment analysis (DEA)
programsdevelopedbyO'Donnell(2009).Thispaperdescribestheconceptualframeworkandthemainfeaturesofthe
software.

1.TOTALFACTORPRODUCTIVITYINDEXES

In the case of a firm


1
that uses a single input to produce a single output, productivity is usually measured as the
outputinput ratio. In the more general case of a firm that produces several outputs using several inputs, it is
commontomeasuretotalfactorproductivity(TFP)astheratioanaggregateoutputtoanaggregateinput.Several
methods for computing these aggregates are described by O'Donnell (2008). Let Q
t
and X
t
denote the aggregate
outputandinputofafirminperiodt.ThentheTFPofthefirminthatperiodissimply

(1) = .
t
t
t
Q
TFP
X

TheassociatedindexnumberthatmeasurestheTFPofafirminperiodtrelativetoitsTFPinperiod0is:

(2)
0
0
0 0 0 0
/
/
= = =
t t t t
t
t
TFP Q X Q
TFP
TFP Q X X

where
0 0
/ =
t t
Q Q Q isanisanoutputquantityindexand
0 0
/ =
t t
X X X isaninputquantityindex.Thus,TFPgrowth
canbeviewedasameasureofoutputgrowthdividedbyameasureofinputgrowth.

AnyTFPindexthatcanbeexpressedintermsofaggregatequantitiesasinequation(2)issaidtobemultiplicatively
complete (ODonnell, 2008). The class of multiplicativelycomplete TFP indexes includes the wellknown Paasche,
Laspeyres, Fisher, Tornquist and HicksMoorsteen TFP indexes. Of these, only the HicksMoorsteen index can be
computedwithoutpricedata(allthatisrequiredisdataonquantities).AnotherTFPindexthatcanbecomputed
withoutpricedataistheMalmquistTFPindexofCaves,ChristensenandDiewert(1982).However,theMalmquist
TFP index is not multiplicativelycomplete and cannot, except in restrictive special cases, be regarded as a valid
measureofproductivitychange.TheDPINsoftwarecomputestheHicksMoorsteenTFPindex.

2.THECOMPONENTSOFPRODUCTIVITYCHANGE

O'Donnell(2008)demonstratesthatallmultiplicativelycompleteTFPindexescanbedecomposedintoameasureof
technicalchangeandseveralmeasuresofefficiencychange.Figure1illustratesthebasicideainaggregatequantity
space. In this figure, the TFP of a firm in period 0 is given by the slope of the ray passing through the origin and
point A, while the TFP of the firm in period t is given by the slope of the ray passing the origin and point Z. Let
lowercase a and z denote the angles between the horizontal axis and the rays passing through points A and Z.
Then the TFP index that measures the change in TFP between the two periods can be compactly written
0
tan / tan . =
t
TFP z a ThisabilitytowriteamultiplicativelycompleteTFPindexastheratioof(tangent)functionsof
anglesinaggregatequantityspaceisusedbyO'Donnell(2008)toconceptualiseseveralalternativedecompositions
of TFP change. For example, let tan e denote the TFP at any nonnegative point E. Then it is clear, both
mathematically and from Figure 1, that the change in the TFP of the firm between periods 0 and t can be
decomposedas
0
(tan / tan )(tan / tan ). =
t
TFP z e e a

1
`Firm'isagenerictermusedtorefertoanindividual,firm,business,state,countryoranyotherdecisionmakingunit.

2

Figure1. MeasuringandDecomposingTFPChange

Within this framework, a potentially infinite number of points E can be used to effect a decomposition of a
multiplicativelycomplete TFP index. O'Donnell (2008) focuses on those points that feature in measures of
efficiencythatarecommonintheeconomicsliterature.Expressedintermsofaggregatequantities,theefficiency
measuresthatfeatureprominentlyininputorienteddecompositionsofTFPchangeare:

- Inputoriented Technical Efficiency (ITE) which measures the difference between observed TFP and the
maximumTFPthatisispossiblewhileholdingtheinputmix,outputmixandoutputlevelfixed.Thisconceptis
illustrated in Figure 2, where the curve passing through points B and D is the frontier of a mixrestricted
productionpossibilitiesset.Theproductionpossibilitiessetismixrestrictedinthesensethatitonlycontains
(aggregates)ofinputandoutputvectorsthatcanbewrittenasscalarmultiplesoftheinputandoutputvectors
at point A. ITE is a ratio measure of the horizontal distance from point A to point B. Equivalently, it is a
measureofthedifferenceinTFPatpointsAandB:
0
tan / tan . = ITE a b

- InputorientedScaleEfficiency(ISE)whichmeasuresthedifferencebetweenTFPatatechnicallyefficientpoint
andthemaximumTFPthatispossiblewhileholdingtheinputandoutputmixesfixed(butallowingthelevels
to vary). This measure of efficiency is represented in Figure 2 as a movement from point B to point D:
0
tan / tan . ISE b d = O'Donnell(2008)referstopointDasthepointofmixinvariantoptimalscale(MIOS).

- Residual Mix Efficiency (RME) which measures the difference between TFP at a point on a mixrestricted
frontier andthe maximum TFP possible when input and output mixes (and levels) canvary. This measure of
efficiency is represented in Figure 2 as a movement from point D to point E:
0
tan / tan . RME d e = The curve
passingthroughEisthefrontierofanunrestrictedproductionpossibilitiesset(unrestrictedinthesensethat
therearenorestrictionsoninputoroutputmix). Theuseofthetermmixinthis contextisselfevident
themovementfrompointDtopointEisamovementfromanoptimalpointonamixrestrictedfrontiertoan
optimal point on a mixunrestricted frontier, so the difference in TFP is essentially a mixeffect. O'Donnell
(2008)alsousesthetermresidualherebecausei)thismovementmayalsoinvolveachangeinscaleandii)
when comparing TFP at point A with TFP at the point of maximum productivity (point E), RME is the
componentthatremainsafteraccountingforpuretechnicalandscaleefficiencyeffects.

A
Aggregate
Output
0
0
X
E
0
Q
e
t
Q
t
X
Z
z
a
= =
= =
= = =
0 0 0
0
0
/ tan
/ tan
tan tan tan
tan tan tan
t t t
t
t
TFP Q X a
TFP Q X z
TFP z z e
TFP
TFP a e a

Aggregate
Input

3


Figure2. AnInputOrientedDecompositionofTFPEfficiency

- InputorientedMixEfficiency(IME)whichmeasuresthedifferencebetweenTFPatatechnicallyefficientpoint
onthemixrestrictedfrontierandthemaximumTFPthatispossiblewhileholdingtheoutputlevelfixed.This
measureofefficiencyisrepresentedinFigure2asamovementfrompointBtopointU: =
0
tan / tan . IME b u

- ResidualInputorientedScaleEfficiency(RISE)whichmeasuresthedifferencebetweenTFPatatechnicallyand
mixefficient point and TFP at the point of maximum productivity. The term scale is used because any
movementaroundanunrestrictedproductionfrontierisamovementfromonemixefficientpointtoanother,
soanyimprovementinTFPisessentiallyascaleeffect.However,thetermresidualisalsousedbecause,even
thoughallthepointsontheunrestrictedfrontieraremixefficient,theymayneverthelesshavedifferentinput
and output mixes. Thus, what is essentially a measure of scale efficiency may contain a residual mix effect.
The term residual is also appropriate in the sense that if we are interested in decomposing the difference
between TFP at the observed point A and TFP at the point of maximum productivity E, then residual scale
efficiency isthe component that remains after we have accounted forpure technical andpure mix efficiency
effects. This measure of efficiency is represented in Figure 2 as a movement from point B to point U:
=
0
tan / tan . RISE u e

- TFP Efficiency (TFPE) which measures the difference between observed TFP and the maximum TFP possible
using the available technology. This measure of efficiency is represented in Figure 2 as a movement all the
wayfrompointAtopointE:
*
0 0 0
/ tan / tan = = TFPE TFP TFP a e where
*
0
TFP denotesthemaximumTFPpossible
usingthetechnologyavailableinperiod0.

Figure2illustratestwoofmanypathwaysfromAtoEandthereforetwoofmanydecompositionsofTFPefficiency:

(3)
0
0 0 0 0
0
*
TFP
TFPE ITE ISE RME
TFP
= =

(4)
0
0 0 0 0
0
= =
*
TFP
TFPE ITE IME RISE
TFP


A
Aggregate
Output
0
0
X
E
0
Q
e
B
u a
0 0 0 0
= = =
tan
t
tan
t
ta
an
n
ta
tan
a an t n n
a
I
a b
IS
d
RME
e
TFP E E
e
TE
d b

0 0 0 0
= = =
tan
t
tan
tan
tan
tan
t n
t an
a
an
u a b
IM
a
T E FPE ITE
b e e
RISE
u

D
b
U
d
Aggregate
Input

4

O'Donnell(2008)alsopresentsseveraloutputorienteddecompositionsofTFPefficiencyintermsofpathwaysfrom
point A to point E. Such decompositions provide a basis for an output or inputoriented decomposition of any
multiplicativelycompleteTFPindex.Theeasiestwaytoseethisistorewrite(3)as
*
0 0 0 0 0
. TFP TFP ITE ISE RME =
Asimilarequationholdsforthefirminperiodt: =
*
.
t t t t t
TFP TFP ITE ISE RME Itfollowsthat

(5)
*
0 *
0 0 0 0 0
t t t t t
t
TFP TFP ITE ISE RME
TFP
TFP TFP ITE ISE RME
| | | |
= =
| |
\ . \ .

Thefirstterminparenthesesontherighthandsideofequation(4)measuresthedifferencebetweenthemaximum
TFP possible using the technology available in period t and the maximum TFP possible using the technology
available in period 0. Thus, it is a natural measure of technical change. The economy/industry experiences
technicalprogressorregressasthistermisgreaterthanorlessthan1.Theotherratiosontherighthandsideof
(5) are measures of technical efficiency change, scale efficiency change, and residual mix efficiency change.
O'Donnell (2008) derives the outputoriented counterparts to equations (3) and (4) and demonstrates that the
inputandoutputorientedmeasuresoftechnicalchangeareplausiblyidentical.DPINcomputesthecomponents
of(5)togetherwiththecomponentsofthefollowingalternativedecompositionsofTFPchange:

(6)
| | | |
= =
| |
\ . \ .
*
0 *
0 0 0 0 0
t t t t t
t
TFP TFP ITE IME RISE
TFP
TFP TFP ITE IME RISE

(7)
| | | |
= =
| |
\ . \ .
*
0 *
0 0 0 0 0
t t t t t
t
TFP TFP OTE OSE RME
TFP
TFP TFP OTE OSE RME

(8)
| | | |
= =
| |
\ . \ .
*
0 *
0 0 0 0 0
t t t t t
t
TFP TFP OTE OME ROSE
TFP
TFP TFP OTE OME ROSE

whereOTE,OSE,OMEandROSEareoutputorientedmeasuresoftechnical,scale,mixandresidualscaleefficiency
thatareanalogoustotheinputorientedmeasures.

3.THEDPINSOFTWARE

The DPIN software uses the DEA programs developed by O'Donnell (2009) to compute and decompose the Hicks
MoorsteenTFPindex.Themethodologyandthesoftwarecanbeused

- whennopricedataisavailable(i.e.,whenonlyinputandoutputquantitydataisavailable);
- for firms that operate in any market environment (e.g., regulated industries, perfectly competitive
industries);
- forfirmsthathaveanybehaviouralobjective(e.g.,maximiseprofit,maximiseoutput);
- whenthetechnologyexhibitsconstantorvariablereturnstoscale;and
- toeitherallowfortechnicalregressorprohibittechnicalregress.
Twoversionsofthesoftwareareavailable:

- astandard(free)versionwhichisavailableathttp://www.uq.edu.au/economics/cepa/dpin.htm.The
standardversioncananalyseabalancedpanelcomprisingnomorethan45firmsandnomorethan350
observations(i.e.,firmsxperiods).
- aprofessional(commercial)versionthatcanbeobtainedbycontactingc.odonnell@economics.uq.edu.au.
Theprofessionalversioncananalyseabalancedpanelwithnolimitonthenumbersoffirmsor
observations.

5

The program was written
2
in C++ and compiled using the mingw compiler within the DevC++ integrated
development environment (IDE). The program is designed to run on a Windows XP or Vista platform. Installing
DPINinvolvesextractingthefilesinDPIN.zipintoanydirectory.DPIN.zipcontainsthefollowingfiles:

DPIN.exe theexecutablefile
DPINTermsandConditions.pdf disclaimer
DPINUsersGuide.pdf thisdocument
eg1_input.csv anexampleinputfile(seebelow)
eg2_input.csv asecondexampleinputfile(seebelow)
rice_input.csv aninputfiletoanalysethericedatainCoellietal.(2005).

RunningDPINinvolvestwosteps:creatinganinputfileandrunningtheexecutablefile.

CreatingtheDPINInputFile

DPINinputfilesmustbecreatedwithinMicrosoftEXCELandsavedina.csv(commadelimited)format.Theinput
file contains both commands and data. To illustrate, Figure 1 is a screenshot of the file eg1_input.csv required to
runthenumericalexampleinO'Donnell(2008).Therowsbeforetheendcommandarecommandrows;therow
immediatelyaftertheendcommandisaheaderrow;theremainingrowsaredatarows.DPINwillreadthetext
anddataincolumnsAandBofeveryrowuntilitencountersanendcommand.Itwillthenskiptheendcommand
andtheheaderrowandstartreadingthedata.Theprogramiscasesensitiveandwillonlyrecognisethefollowing
textstringsinthecommandrows:

Periods tospecifythenumberofperiods
Firms tospecifythenumberoffirms
Inputs tospecifythenumberofinputs
Outputs tospecifythenumberofoutputs
NoTechRegress toprohibittechnicalregress(optional)
CRS toimposeconstantreturnstoscale(optional)

The command rows can be listed in any order (e.g., the number of outputs can be specified first, second, or last).
Theprogramwillnotreadthetextintheheaderrowthisrowissimplyusedtoconvenientlyrecordthenamesof
theoutputsandinputs.Theoutputandinputdatamustbestoredinthedatarowsinaparticularformat:

- thefirstoutputvariablemustbestoredincolumnD
- alloutputvariablesmustbestoredtotheleftoftheinputvariables;
- thedatamustbeabalancedpanel;
- thedatamustbesortedfirstbyperiodandthenbyfirm;
- theobservation,firmandperiodidentifiersincolumnsA,BandCmustbenumeric(e.g.,1,2,...,not
Jan,Feb...);and
- to avoid numerical problems it is advisable to normalise all output and input variables to have unit
means.

The DPIN default settings are to allow for technical regress and variable returns to scale. To prevent technical
regress, insert a command row with NoTechRegress in column A. To impose constant returns to scale, insert a
commandrowwithCRSincolumnA.Figure2isascreenshotoftheinputfileusedtobothpreventtechnicalregress
andimposeconstantreturnstoscale.

RunningtheExecutableFile

The DPIN executable file is DPIN.exe. Doubleclicking DPIN.exe will open windows similar to those depicted in
Figures3to5.Figure3isacommandwindowthatreportsprogramandruntimeinformation;Figure4isawindow
usedtobrowseandselecttheinputfile;andFigure5isthecommandwindowasitappearsshortlyaftertheinput
fileeg1_input.csvhasbeenselected.ResultsarewrittentoEXCELoutputfileshavingextensions_output.csv.The

2
TheDPINprogramwasoriginallywritteninGAUSSbyC.ODonnellandwastranslatedintoC++byF.WrathmallandC.ODonnell.

6

name assigned by DPIN to the output file is reported in the command window prior to exit. In the professional
versionofDPIN,distancesarewrittentofileshavingextensions_distances.csv.

RuntimeErrors

In both versions of DPIN, error diagnostics are written to log files having extensions _log.txt. In log files (and the
commandwindow)thegeneralerrorcodeforfailuretosolvealinearprogramis1.ODonnell(2009)explainswhy
some linear programs may be infeasible and fail to solve. The error code for exceeding the maximum number of
simplexiterationsis5.Themaximumnumberofiterationsmaybereachedifthelinearprogramisdegenerate.For
detailsondegeneracyandcyclinginlinearprogramsseeWinston(2004,p.168171).

InterpretingtheDPINOutput

DPIN computes and reports measures of output and inputoriented technical, scale and mix efficiency for every
firm in every time period in the sample. It also computes and reports indexes of TFP change, technical efficiency
change,andoutputandinputorientedmeasuresoftechnical,scaleandmixefficiencychange.Toillustrate,Figure
6 presents the output file eg1_output.csv obtained by running DPIN on the input file eg1_input.csv. For an
interpretationoftheseresults,seeO'Donnell(2009,p.20,21,34,35).NotethatColumnKintheoutputfilereports
newpointsofmaximumproductivity.IftechnicalregressisprohibitedandthereisnoMaxTFPvaluereportedfora
particularperiodthenthisisbecausethepointofmaximumproductivityisunchangedfromthepreviousperiod.In
suchcases,theindexoftechnicalchangeis1.Themissingvaluecodeusedinoutputfilesis9999.Missingvalues
occurwhenalinearprogramcannotbesolved.

FurtherInformation

Forfurtherinformation,emailc.odonnell@economics.uq.edu.au

7

Figure1.InputfileforaModelthatPermitsTechnical
RegressandAssumesVRS.

Figure2.InputfileforaModelthatProhibits
TechnicalRegressandAssumesCRS.

8

Figure3.InitialCommandWindow

Figure4.InputFileSelectionWindow





Figure5.CommandWindowafterDataProcessing





Figure6.OutputFile


References

Caves, D. W., L. R. Christensen and W. E. Diewert (1982). "The Economic Theory of Index Numbers and the
MeasurementofInput,Output,andProductivity."Econometrica50(6):13931414.
Coelli,T.J.,D.S.P.Rao,C.J.O'DonnellandG.E.Battese(2005).AnIntroductiontoEfficiencyandProductivityAnalysis.
NewYork,Springer.
O'Donnell, C. J. (2008). An Aggregate QuantityPrice Framework for Measuring and Decomposing Productivity and
ProfitabilityChange.CentreforEfficiencyandProductivityAnalysisWorkingPapers.UniversityofQueensland.
O'Donnell, C. J. (2009). Measuring and Decomposing Agricultural Productivity and Profitability Change. 53rd Annual
ConferenceoftheAustralianAgriculturalandResourceEconomicsSociety.Cairns,Australia.
Winston,W.L.(2004).OperationsResearch:ApplicationsandAlgorithms.BelmontCA,BrooksCole.