7-VanZyl.qxd


The aim of this paper is to present the methodolog y,

structure and results of a dual cost-production model where

changes in the level of labour productivit y of a firm and an

industry can be measured. The analysis is based on the

application of the principles of dualit y where co-integrated

production functions for each sample time series are derived

from dual cost functions in order to derive the marginal

efficiency coefficients which measures productivit y changes.

The South African motor vehicle manufacturing industry is

used as a case study.

A decision had to be made whether to estimate a more flexible

or a more restricted form of a production function for each

chosen time series. A more restrictive form was chosen because

it is assumed to be more stable and robust than a flexible form

and is thus more suitable for incorporation into a production

model, which is utilised for the purpose of this particular

analysis (Wynn 1974). 

THE DUAL COST FUNCTION

The opt imal value of the object ive f unct ion of the 

primal problem is equal to the optimal value of the

corresponding dual problem. Optimal economic behaviour

of business entities is most commonly measured by the

principles of profit maximisation and the principle of cost

minimisation. The principle of the dualit y theorem thus

states that any constrained profit maximisation problem has

a dual problem in constrained minimisat ion of cost

(Salvatore 1996). 

The max imum product ion out put level is normally

expressed as a function of the input base and a certain level

of technological know-how (the result of economies of scope

and the learning curve effect). In terms of dualit y principles

the minimum cost function can be expressed as a function

of the combined production output of the firm or industries,

the input base, the input prices and the given level of

technological know-how. The applicable average cost

function is defined as the minimum cost level divided by the

given level of manufact uring output. The applicable average

cost function is important in determining ret urns to scale

for the purpose of the compilation of the dual model

(Hirschey 1995).

The cost minimisation problem is expressed as a constrained

optimisation problem where minimum cost is expressed as

the sum of input prices plus a Lagrange multiplier times the

difference bet ween the levels of production output on the

one hand and the levels of inputs and technical know-how on

the other hand. The first order condit ions for the

minimisation of production cost are simply the partial

derivatives of the minimum cost function in terms of 1)

input prices and 2) the difference bet ween manufacturing

output and the level of inputs and technological know-how.

The cost function dual to the production function is

determined by substituting each level of input combination

into the first order derivative of the difference bet ween

manufact uring out put and the level of inputs and

technological know-how. The optimal levels of inputs are

substituted into the cost equation (expressed as the sum of

the product of the level of inputs and input prices).

THE PROPERTIES OF THE DATA 

USED IN THE ESTIMATION

An important aspect of data collection is to establish the

properties of each data set by analysing each data series

thoroughly. Properties such as stationarity or the order of

integration are important.

The following time series are used in the estimation of cost

functions for the different sample time series and the

consequent derivation of the production functions:

G VAN ZYL
Professor in Economics

Department of Economics

Rand Afrikaans University

ABSTRACT
Positive real increases in the prices of inputs, as the result of an inefficient level of productivity, can no longer be

accepted in competitive markets. The survival and more specifically the competitiveness of businesses and industries

is critically dependent on a higher level of cost efficiency and in particular a more productive input base.

The aim of the paper is to present the methodology, structure and results of a new dual cost-production instrument

where changes in the level of labour productivity can be measured. A scientific derived and well-tested instrument

is developed. This instrument should enable human resource managers to monitor productivity changes more

effectively and make timely decisions on ways and means that could improve the productivity of labour.

OPSOMMING
Positiewe reële toenames in die pryse van insette as die resultaat van ’n oneffektiewe vlak van produktiwiteit kan

nie langer in mededingende markte geduld word nie. Die oorlewing en meer spesifiek die mededingendheid van

ondernemings en industrieë is krities afhanklik van ’n hoër vlak van koste-doelmatigheid en in die besonder van ’n

meer produktiewe inset basis.

Die doel van die publikasie is om die metodiek, struktuur en resultate van ’n nuwe gelyktydige koste-produksie

instrument bekend te stel waarmee veranderinge in die vlak van arbeidsproduktiwiteit gemeet kan word. ’n

Wetenskaplik-afgeleide en getoetste instrument is ontwikkel. Diè instrument behoort personeelpraktisyns in staat te

stel om produktiwiteitsveranderinge meer effektief te kan monitor en gevolglik tydige besluite te kan neem oor

wyses waarop arbeidsproduktiwiteit verbeter kan word.

THE COST-PRODUCTION DUALITY APPROACH TO THE

MEASUREMENT OF LABOUR PRODUCTIVITY: A DYNAMIC TOOL

FOR EFFECTIVE HUMAN RESOURCE MANAGEMENT 

Requests for copies should be addressed to: G van Zyl, Department of Economics,

RAU University, PO Box 524, Auckland Park, 2006

50

SA Journal of Human Resource Management, 2004, 2 (1), 50-53

SA Tydskrif vir Menslikehulpbronbestuur, 2004, 2 (1), 50-53



1. The cost of production is expressed as cost of labour plus the

cost of capital.

The cost of capital is calculated as real capital stock

multiplied by real user cost of capital. The user cost 

of capital is calculated in such a way that it not only

includes the acquisition cost of capital but also the return

forgone by using capital rather than rent ing it,

depreciation and any capital gains/losses associated with

holding the particular t ype of capital (Maurice 1985). For

the purposes of the estimations the user cost is expressed

as: user cost of capital = qk(i+�) ; where qk is the

acquisition cost of capital stock, i the real rate of interest

and d the rate of depreciation. The acquisition cost of

capital is deflated by the production price index in order

to account for differences in the price level at the time of

acquisit ion. The t ime series is transformed to its

logarithmic form.

The cost of labour for each time period is calculated as labour

employed times the real cost of labour. The real cost of labour

is calculated as the total wages & salaries paid deflated by the

production price index. The time series is transformed to its

logarithmic form.

After the cost of production is calculated, the time series is

transformed into its logarithmic form.

2. Manufacturing output at real cost

The manufacturing output at factor cost time series is simply

transformed into its logarithmic form.

3. A technology know-how index

This index is included in the function in order to capt ure

changes in technological know-how. The index is

expressed as a function of a socio-education index, an

energ y index and a labour skills index for the industry. The

weight of each of the series in this equation is derived by

performing a linear regression of the three indices in the

equat ion on manufact uring out put. The est imated

coefficients are then used as an indication of the relative

importance of each series in the technolog y index. The

ultimate choice of the explanatory indices is based on their

substantial influence on the efficiency of manufacturing

product ion. This index is not transformed into its

logarithmic form.

4. The demand for labour

This variable is represented by a time series of workers

employed in the manufact uring industries per time 

period. Each time series is transformed into its logarithmic

form.

5. Real fixed capital stock 

This variable is represented by a time series of real fixed

capital stock per time period in the manufact uring

industries. The time series is transformed to its logarithmic

form.

6. Dummy variables

The aim of the dummy variables is to avoid bias in the

estimation. Dummy variables are introduced to firstly

cover the period 1982-1993 when it became difficult to

import capital stock (as a result to exogenous political

restrictions) and secondly to cater for tense labour

relations in the period 1987-2001 that resulted in strikes

and work stoppages.

7. All the data series are tested for stationarity and the level of

integration of each of the variables in the regression for

each time period. The tests are performed on the data in

their logarithmic form. All variables are tested at 5% level

of significance and all the variables are integrated of the

order of 1. 

THE EMPIRICAL ESTIMATION OF COST

FUNCTIONS
The empirical function for the cost of production in the industry

for each time period is expressed as:

Cost of production = f (Production output, user cost of

capital, real cost of labour, technological know-how index)

The methodology that is used to estimate the cost functions is to

apply a roll-up approach where the first data point of the

previous data series is omitted and where a new data point is

chosen in a sequential manner for each successive data series.

The long-run coefficients of the cost functions are then

estimated using ordinary least squares. These functions are

estimated with a restriction on the coefficients of the input

prices to sum to unity. The restriction follows from economic

theory that states that a cost function must be homogeneous of

degree 1 in prices, that is, by doubling prices costs should be

doubled (Maurice 1985:135).

The resulting coefficients of the cost estimation per time period

are summarized in Table 1:

TABLE 1

ESTIMATION RESULTS OF THE CO-INTEGRATION COST

EQUATION PER TIME PERIOD

Independent variable Coefficients Coefficients Coefficients

1975-1999 1976-2000 1977-2001 

Production output (est (1)) 0.64210 0.61230 0.5960 

User cost of capital (est (2)) 0.38120 0.39340 0.4188 

Real cost of labour (est (3)) 0.61880 0.62960 0.6452 

Technology index (est (4)) -0.45912 -0.4788 -0.4890

R2 0.9231 0.9324 0.9245

Adjusted R2 0.9001 0.9133 0.9167

Dependent variable: Cost of production

Source: Own estimations

The long-run co-integrated cost functions for each time period

are as follow:

� 1975 – 1999

Cost of production = 0.64210*Manufacturing output +

0.38120*user cost of capital + 0.61880*real cost of labour -

0.45912*technological know-how index

� 1976 – 2000

Cost of production = 0.61230*Manufacturing output +

0.39340*user cost of capital 

+ 0.62960*real cost of labour - 0.4788*technological know-

how index

� 1977 – 2001

Cost of production = 0.5960*Manufacturing output +

0.4188*user cost of capital 

+ 0.6452*real cost of labour - 0.4890*technological know-

how index

The co-integrated cost functions have economic significance in

the sense that the signs of the estimated coefficients are as

expected. The magnitude of the estimated manufacturing

output parameter represents a decreasing degree of decreasing

returns to scale.

The residuals of the estimated cost function are tested for

stationarit y. Econometric tests are used to test whether 

the residuals obtained from the est imated product ion 

cost regressions are stationary (Makridakis 1998). These 

tests confirm that the residuals are stationary at a 5% level 

of significance. The error terms of the regressions co-

MEASUREMENT OF LABOUR PRODUCTIVITY 51



integrated and are confirmed by a correlogram test on the

residuals. The conclusion is that these est imated cost

functions could be used to derive long-run co-integrated

production functions in order to establish the changes in 

the marginal efficiencies of the inputs and more specifically

labour productivit y. 

THE DERIVATION OF CO-INTEGRATED

PRODUCTION FUNCTIONS

By applying dualit y theory it is possible to derive 

production functions from the estimated cost functions. 

The derivation of the long-run production functions is 

based on the methodolog y and calculations of Berndt (1991).

From the estimated cost equations it follows that production

costs always increases (as indicated by the estimated cost

f unct ion) and therefore technolog y in this part icular

industry displays decreasing returns to scale. This result 

was confirmed by using the dualit y methodolog y and

calculations of Berndt (1991). The calculation for returns to

scale is expressed as:

r = 1/1+est(1)

(where r is the returns to scale and est(1) is the estimated

coefficient of production output in the cost function). 

The estimated returns to scale coefficient for each data series is

indicated in the following Table 2. 

TABLE 2

RETURNS TO SCALE COEFFICIENT PER TIME PERIOD

Time period Returns to scale coefficient 

1975 – 1999

1976 – 2000

1977 – 2001     

Source: Own calculations

By simply multiplying each of the input price coefficients

and that of technolog y by ret urns to scale (r), the 

coefficients of the inputs (a, b and d) in the production

function (production output = aK�L�tei�) are obtained. These

calculations are based on the principles of dualit y and are as

follows: 

� = est(2) * r  (est(2) = coefficient user cost of capital)

� = (1- est(3)) *r  (est(3) = coefficient for labour cost)

� = est(4) * r (est(4) = coefficient for the technology index)

The associated constant for the production function is

estimated simply by estimating an equation with the

calculated coefficients, multiplied with the associated

production variables, and a constant as an explanatory

variable. The result of the least square estimation is a value for

the unknown constant which can be used in the long-run co-

integration equation. The calculated coefficients of the inputs

(�, �, and �) based on the principles of duality are indicated in

the Table 3.

TABLE 3

CALCULATED VALUES FOR THE PRODUCTION COEFFICIENTS

Time period �� �� �

1975 – 1999 0.2322 0.3768 0.2796 

1976 – 2000 0.2440 0.3905 0.2970 

1977 – 2001 0.2624 0.4043 0.3064

Source: Own calculations

The co-integration production functions when the variables are

not used in their logarithmic form are thus:

� 1975 – 1999

Manufacturing output = 8.0154*real fixed capital 0.2320

demand for labour 0.3768 tei 0.2798

� 1976 – 2000

Manufacturing output = 8.7345*real fixed capital 0.2440

demand for labour 0.3905 tei 0.2970

� 1977 – 2001

Manufacturing output = 9.3466*real fixed capital 0.2624

demand for labour 0.4043 tei 0.3064

The successful derivation of the production functions is

clearly illustrated by acceptable root mean squared errors,

mean absolute errors and mean absolute percentage errors.

Dynamic simulations were performed on the long-run co-

integrated production functions as an indication of the

goodness of fit of the model. From the dynamic simulations

the assumption can be made that each of the estimated

production functions delivers a very satisfactory fit and is a

stable function.

THE MEASUREMENT OF CHANGES IN THE LEVEL

OF PRODUCTIVITY

The changes in the marginal efficiencies (productivit y) of the

various components of the input base are simply calculated as

the percentage increase or decrease in the marginal products

of these components (a, b, and d) over the specified sample

periods.

�e in capital efficiency (productivity) = 

�e in labour efficiency (productivity) = 

�e in technology efficiency (productivity) = 

The results of these calculations are summarized in Table 4.

TABLE 4

PERCENTAGE CHANGES IN THE EFFICIENCY ESTIMATES

Time period �� �� �

1975 – 1999 +5.082 +3.640 +6.220

to

1976 – 2000 

1976 – 2000 +7.540 +3.530 +3.1650

to

1977 – 2001        

Source: Own calculations

� �

�

t t

t

x�1
100

1

–

� �

�

t t

t

x�1
100

1

–

1 – 100

1
t t

t

x+
α α

α

1
0.6266

1.5960
=

1
0.6202

1.6123
=

1
0.6090

1.64210
=

VAN ZYL52



Positive percentage increases in the marginal efficiencies of all the

inputs contributed to the improvement in the returns to scale of

the industry. It is also clear from Table 4 that the improvement in

labour productivity is the lowest of all the inputs and that capital

is substantially more efficient (productive) than the labour input.

The optimal combination of inputs is an important consideration

for a firm/industry. The estimated efficiency criterion coefficient

(�) enables the decision maker to determine the required

changes in input combinations in order to strive for a more

optimal (efficient) input base. The results of the calculations on

the optimality coefficient (�) are summarized in Table 5 (the

improvement in the level of technology is taken as a given).

TABLE 5

OPTIMALITY OF THE INPUT BASE

Sample period Optimality coefficient

� = �(K/L) – �(w/r) 

1974 – 1999 -8.215 

1975 – 2000 -7.312 

1976 – 2001 -7.115               

Source: Own calculations

From the Table 5 it is clear that in each of the sample periods the

optimality coefficient clearly indicates an over-utilisation of

labour (albeit at a slower pace). This implies that the greater level

of capital efficiency combined with a likely increase in capital

stock could push the input mix to a more optimal level. The

message to human resource management is clear namely that

labour productivity in this particular industry must increase

substantially in order to avoid greater levels of capital

substitution for the labour input. 

CONCLUSION

A scientific derived and well tested instrument in a managerial,

econometric and statistical sense was developed. The instrument

measures the changes in the level of productivity of the input

base more accurately. It can easily be applied on firm level over

weekly and/or monthly sample periods. This instrument should

enable management to monitor productivity changes more

effectively and make timely decisions that should result in a

more optimal usage of resources.

REFERENCES

Berndt, E.R. (1991). The Practice of Econometrics; Classic and

Contemporary. New York: Addison-Wesley.   

Hirschey, M & Pappas J.L. (1996). Managerial Economics.

Orlando: Dryden Press.               

Maurice, S.C. & Smithson S.W. (1985). Managerial Economics;

Applied Microeconomics for Decision Making. Homewood:

Irwin.

Salvatore, D. (1996). Managerial Economics in a Global Economy.

New York: McGraw-Hill.

Wynn R.F. & Holden K. (1974). An Introduction to Applied

Econometric Analysis. New York: Wiley & Sons.

MEASUREMENT OF LABOUR PRODUCTIVITY 53