Mediating effect of advertising expenditure on labour productivity: a case of manufacturing industries in Pakistan.
Mahmood, Tariq
This paper explores the possibility that the labour productivity
enhancing effects often ascribed to capital intensity may partly act
through some mediating variable. The paper uses a mediation model to
estimate direct and indirect effects of capital intensity on labour
productivity in Pakistan's manufacturing industries. The data
involve 229 industries at five-digits level of aggregation. The data are
taken from Census of Manufacturing Industries for the year 2005-06.
Using capital intensity as an independent variable and advertising
expenditure as a mediating variable, the paper estimates total, direct,
and indirect effects on labour productivity. Approximately 18 percent of
total effects on labour productivity are found to be mediated through
advertising expenditure. The statistical significance of indirect
effects is tested using standard normal tests as well as bootstrap
method, and these effects are found to be significant.
JEL classification: D24, C31, M37, L60
Keywords: Productivity, Mediation, Advertising, Industries
1. INTRODUCTION
In productivity literature capital intensity is often regarded as
one of the most important determinants of labour productivity. As
pointed out by Arrow, et al. (1961), in a linearly homogeneous
production function marginal productivity of labour is an increasing
function of capital-labour ratio. Usually some other variables like
human capital, innovation, trade openness, research and development are
included in the model along with capital-labour ratio to analyse
determinants of labour productivity [see for example, Velucchi and
Viviani (2011); Han, Kauffman, and Nault (2011); Hussain (2009);
Apergis, et al. (2008)].
All these determinants of labour productivity are important in
their own right. However, there is a need to take the analysis one step
further, and explore the transmission process through which a
determinant may affect labour productivity. An explanatory variable may
affect labour productivity directly as well as through some mediating
variable. Mediation Model suggested by Baron and Kenny (1986) can be
helpful in such a situation. The model gives quantitative estimates of
direct and indirect effect of an explanatory variable in a regression
equation. A possible missing link in the chain of causation from capital
intensity to labour productivity could be advertising expenditure.
This chain of causation consists of two important relationships.
The first one can be established between capital intensity and
advertising expenditure. There exist plausible reasons to assume
causation from capital intensity to advertising expenditure. First,
firms with high capital intensity are usually on higher level of
technology adoption. These firms have a forward-looking attitude, and
are well aware of modern marketing techniques. Hence they can be
expected to spend more on advertising. Second, firms with more capital
resources can afford to incur larger advertising expenditures in print
and electronic media as well as in other promotional activities. Last,
resourceful firms may resort to advertising as a strategy to create
barriers for potential entrants, or to drive out existing competitors.
(1) These factors induce the firms with high capital to spend more on
advertising than those which lack capital resources. The second
relationship in this causation chain is established when advertising
expenditure, just like other inputs, is included in the value of output.
(2) This fact, along with the assumption of positive marginal product,
makes advertising expenditure an important contributor to labour
productivity.
This paper attempts to disentangle the underlying mechanism of
causation from capital intensity to labour productivity through a
mediation model. The paper aims to analyse direct and indirect effects
of capital intensity. Advertising expenditure is modeled to play the
role of mediating variable. The paper contributes to empirical research
on manufacturing industries in Pakistan in three important ways. First,
it increases our understanding of labour productivity analysis of
manufacturing industries in Pakistan. It provides answer to the
question: what might be the contribution made by capital intensity in
productivity enhancement, and how advertising plays its role in
affecting this contribution. Second, from a policy perspective it is
important to know that a significant portion of labour productivity
enhancement ascribed to capital intensity, in fact, comes from
advertising expenditure. This would be helpful for policy-makers to
frame regulatory measures about capital markets as well as about
advertising activity. Third, this study will pave the way for further
research about the role of advertising, e.g. whether advertising is
informative and persuasive, as it is often claimed, or is it being used
as a strategy by resourceful firms/groups to stifle competition.
The rest of study is divided as follows: Section 2 presents a
review of recent empirical literature about the use of mediating
variable as well as analysis of labour productivity. Methodology and
data are discussed in Section 3. Section 4 discusses empirical results
of the model, and Section 5 concludes the discussion.
2. REVIEW OF LITERATURE
Baron and Kenny's (1986) introduced mediation models in the
field of psychology. Gradually their use in economics and other
social/managerial sciences became common. (3) Some recent empirical
studies using mediation models are reviewed below.
Srivastava and Rai (2013) analyse the mediating role of customer
satisfaction in determining the relationship between service quality and
customer loyalty. The study uses data from a survey of 400 customers of
the top three life insurance companies in India. The companies are
selected on the basis of their relative shares in the market of life
insurance policies in India. Customer loyalty has been used as dependent
variable while independent variable is service quality. Customer
satisfaction is used as a mediating variable. Sobel's test is used
to examine the significance of mediation model. The study finds that
customer satisfaction plays a significant role as a mediator in a
service quality-customer loyalty relationship.
Newsham, et al. (2009) use a mediated regression model linking the
physical environment, through environmental satisfaction, to job
satisfaction and other related measures. Physical measurement and
questionnaire data are used from 95 workstations in Michigan, USA. The
study demonstrates a significant link between overall environmental
satisfaction and job satisfaction, mediated by satisfaction with
management and with compensation.
Banker, Bardhan, and Chen (2008) study the impact of activity-based
costing on plant performance by using world-class manufacturing
practices as a mediating variable in a sample of 1250 manufacturing
plants in U.S. The study finds that world-class manufacturing practices
completely mediate the positive impact of activity-based costing.
Kuha and Goldthrope (2007) develop a mediation model to assess the
impact of educational attainment on intergenerational social mobility
using British survey data. The study proposes a method to estimate
direct and indirect effects between a person's father's and
his or her own social classes in systems where some of the variables are
categorical. Education is used as a mediating variable to decompose the
effect on intergenerational social mobility. The data are used for years
1973 and 1992. Analysis is performed separately for men and women with
educational qualification used in seven categories, and social status
defined in three classes, viz. (i) Salariat and employers, (ii)
Intermediate class, and (iii) Working class. The results indicate high
variation in the proportions of indirect effect. This proportion is
found to be about 80 percent for women (for mobility from salariat to
salariat class) to about 12 percent for men (for mobility from working
to working class). However the study does not find any systematic or
significant change in proportions of indirect effects between the years
1973 and 1992.
Maydeu-Olivares and Lado (2003) analyse the economic performance of
insurance industry in the European Union. The study uses a multiple
mediator model to separate the effect of market orientation on economic
performance. The mediators used are innovation degree, innovation
performance, and customer loyalty. The three mediating variables are
found to be interrelated. Customer loyalty alone does not mediate the
effect of market orientation on economic performance. However, when it
is used in combination with innovation degree and innovation
performance, it mediates the effects of market orientation on business
performance. It is found that these variables completely mediate effects
of market orientation on economic performance. Moreover, there is an
improvement in predictions of economic performance by 52 percent over
what is explained by market orientation alone.
Productivity has been extensively analysed in empirical literature.
A number of writers have studied the issues in the area of total factor
productivity (TFP) and explored its determinants, e.g. Cheema (1978),
Mahmood and Siddiqui (2000), and Hamid and Pichler (2009). On the other
hand some studies focus on labour productivity, a partial measure, for
analysis. A few recent papers dealing with determinants of labour
productivity are reviewed below.
Velucchi and Viviani (2011) investigate how some firms'
characteristics affect the dynamics of the Italian firms' labour
productivity in food, textiles and mechanical machinery sectors. The
authors estimate a nonlinear production function by quintiles regression
approach using firm-level panel data developed by Italian National
Institute of Statistics. The study covers the period of 1998-2004. It is
found that the relationships between labour productivity and firms'
characteristics are not constant across industries and quintiles.
Particularly human capital and innovation have a larger impact on
fostering labour productivity of less productive firms than that of
highly productive firms. Similarly human capital and innovation have a
higher impact for exporters than for non-exporters. Such polarisation is
also found within sectors. In food sector innovativeness is significant
for low and high productivity firms but it is found to be irrelevant for
the median firm. Similarly in textiles the effect of innovation grows as
productivity grows. The patents are found to be significant in the
machinery sector only for median firms and have very little impact on
both low and high productivity firms. Internationalisation turns out to
be important for low productivity firms in all sectors.
Han, Kauffman, and Nault (2011) evaluate the contributions of
spending in IT outsourcing on labour productivity. The study uses an
economy-wide panel dataset in the United States. The data cover sixty
industries for the period 1998 to 2006. The authors estimate
determinants of labour productivity using IT outsourcing as a mediating
input along with other inputs. The results show that on average, a 1
percent increase in spending on IT outsourcing per labour hour is
associated with a 0.024 percent to 0.04 percent increase in labour
productivity. Separate equations are estimated for the high-and low-IT
intensity industry groups. The results show that the coefficient
estimates for IT outsourcing are greater in high-IT intensity
industries.
Hussain (2009) analyses the causal ordering between inflation, and
productivity of labour and capital in Pakistan's economy. The study
uses the data from IMF dataset compiled by the United Nations
Statistical Database and World Development Indicators covering the
period 1960 to 2007. Vector Autoregressive technique is used in the
analysis and results suggest that there exists a unidirectional
causality from inflation to labour productivity through capital labour
ratio. Bidirectional causality between inflation and capital
productivity through capital labour ratio is also found. The paper
estimates that on average it takes about 15 months for these causalities
to take effect.
Kutan and Yigit (2009) estimate determinants of labour productivity
growth in eight new EU member states that joined the Union in 2004,
namely, the Czech Republic, Estonia, Hungary, Poland, Latvia, Lithuania,
Slovakia and Slovenia. The study uses panel data for the period
1995-2006. Results indicate mixed effects of globalisation. FDI and
exports increase productivity growth, while imports reduce it.
Education, measured by secondary school enrolment, and domestic
investment have significant and positive effect while R&D does not
play any significant role. Effect of productivity gap is found to be
positive and significant, thus implying labour productivity convergence
among European countries.
Most of these studies are mainly concerned with finding
determinants of labour productivity, or quantifying the impact of these
determinants on labour productivity.
There is a need to explore the causal sequence of the impact of one
or more determinants on labour productivity. How much is the direct
effect of a determinant on labour productivity and how much of the
effect is being transmitted through some mediating variable (e.g.
advertising)? Present study aims to answer this question.
3. METHODOLOGY
Use of Advertising Expenditure
At macro level advertising has been found to have positive effect
on variables like labour supply, consumption etc. [Fraser and Paton
(2003); Jung and Seldon (1995)]. However, at firm or industry level the
issue is problematic. Advertising is usually not regarded as an input in
traditional production analysis. This may partly be due to the
perception that advertising, like other marketing strategies, comes into
play after production process has been completed. Hence its role in
productivity enhancement is not relevant. Another reason may have
stemmed from the notion of perfect competition. In the words of Pigou
(1924, pp. 173-174),"Under simple competition there is no purpose
of advertisement, because ex hypothesis, the market will take, at the
market price, as much as any one small seller wants to sell." It is
probably due to these reasons that a significant body of literature on
advertising focuses on its role in market structure, its influences on
demand, prices, preferences, quality etc., and its welfare implications,
see for example, Hamilton (1972), Hochman and Luski (1988), Horstmann
and Moorthy (2003).
Some writers take a divergent perception of advertising. Telser
(1978), argues that promotion and the product are joint outputs in
supply; thus making advertising an inextricable component of the
product. Ehrlich and Fisher (1982), and Richards and Patterson (1998)
explicitly treat promotion expenditure as an input in the production
process. At empirical level, however, the issue is less problematic.
Firms generally include advertisement expenditures in total cost while
determining market price of their product. These expenditures are
reported in income accounts just like other inputs' costs.
Theoretical Model
This study follows a simple mediation model suggested by Baron and
Kenny (1986). This model consists of three variables; independent
variable, dependent variable, and mediating variable represented by X,
Y, and M respectively. The model represents a causal sequence; X affects
M which may in turn affect Y. In addition X may have direct effect on Y
which is not transmitted through M. The size and significance of
coefficients of these variables decides the nature of mediation.
This model (4) can be described by the following three equations:
(5)
Y = [i.sub.1] + c X + u ... ... ... ... ... ... ... (1)
M = [i.sub.1] + a X + v ... ... ... ... ... ... ... (2)
Y = [i.sub.3] + c' X + b M + e ... ... ... ... ... ... (3)
Where u, v, and e are random terms satisfying usual OLS
assumptions.
Equation (1) establishes direct effect of independent variable on
the dependent variable. Equation (2) is used to see how the mediator is
affected by the independent variable. Equation (3) shows how mediator
affects the dependent variable when direct effect of independent
variable is controlled.
As pointed out by Baron and Kenny (1986), the method requires that
the coefficients a, b, and c should be statistically significant, and
the condition c' < c must hold. Perfect mediation occurs when
c'= 0. In this case total effect, c, is completely mediated. In
general, mediation is not complete, and total effect is said to be
partially transmitted through the mediator. Some authors [Collins,
Graham, and Flaherty (1998); MacKinnon, Krull, and Lockwood (2000)]
argue that significance of c is not necessary for mediation to occur.
The total effect of X on Y, measured by the coefficient c in
Equation (1), can be written as combination of direct and indirect
effects. If OLS is used to estimate the model, the total effect is the
additive sum of the direct and mediated (or indirect) effect [Warner
(2012), p. 654], Symbolically we may state the relationship as:
c = c' + ab or ab = c - c'
The coefficient 'a' represents the effect of independent
variable on mediating variable, and 'b' coefficient represents
the effect of mediating variable on dependent variable. The product
'ab' reflects how much a unit change in independent variable
affects dependent variable indirectly through mediating variable.
Testing the Significance of Indirect Effects
The tests of significance of indirect effects, 'ab', can
be performed in many ways. The most frequently used tests can broadly be
grouped in two types; (i) tests based upon the assumption of normality
of 'ab', (ii) tests based upon bootstrap technique. These
tests will be applied against the null hypothesis of no indirect
effects, i.e. [H.sub.0]: ab = 0.
In order to conduct the test under normality assumption,
'ab' is divided by standard error of 'ab' and
resulting value is compared with the critical value from the standard
normal distribution for a given level of significance. The standard
error of the product ab is given by
[SE.sub.ab] = [square root of (([b.sup.2] [SE.sup.2.sub.a] +
[a.sup.2] [SE.sup.2.sub.b] + [SE.sup.2.sub.a] [SE.sup.2.sub.b]))] ...
... ... ... (4)
Where [SE.sub.ab] is standard error of the product 'ab',
[SE.sub.a] is standard error of 'a', and [SE.sub.b] is
standard error of 'b'.
Aroian test uses the above expression to conduct the test of
significance. Sobel test assuming independence of 'a' and
'b', omits the last term (i.e. the covariance of 'a'
and 'b'). The Goodman test subtracts the last term from the
first two terms to form an unbiased estimate of the variance of
'ab'. These tests are built on the works of Aroian (1947),
Sobel (1982), and Goodman (1960) respectively.
The product of two normal variables is, in general, not normally
distributed [see e.g. Aroian (1947); Lomnicki (1967)]. It has been
argued that the normal theory approach for testing indirect effects
lacks statistical power especially for small samples [MacKinnon, et al.
(2002)]. It is suggested that minimum sample size of 150 to 200 may be
regarded as suitable [Warner (2012), p. 663]. Present study applies
these tests on 229 observations. So we can expect that these tests will
perform better.
Another alternative to the normal theory approach is the bootstrap
procedure to test the statistical significance of direct and indirect
effects. Bootstrap method, introduced by Efron (1979), allows estimation
of the sampling distribution of statistics using resampling methods.
Bootstrap method (6) will also be used to estimate and test the
significance of direct and indirect effects. The standard deviation of
the distribution of bootstrapped 'ab' provides an estimate of
the standard error, which could be used in the usual way to construct
confidence intervals. Bootstrap technique will be used to compute
percentile confidence intervals, and bias-corrected confidence
intervals. In the percentile confidence intervals bootstrap estimates
are sorted in ascending order and interval limits are chosen
corresponding to (a/2) x 100th and (1-[alpha]/2) x 100th percentile
values for a particular a level of significance. In bias-corrected
bootstrap confidence intervals the end points of the intervals are
adjusted depending upon whether the proportion of bootstrapped values of
indirect effects is less than those estimated from the original data.
(7)
The null hypothesis of no indirect effects will be tested using
these two types of confidence interval. The hypothesis will be rejected
if the confidence interval does not contain zero value.
Data and Variables
This study uses data from Census of Manufacturing Industries (CMI),
2005-06. (8) The census is published by Statistics Division, Federal
Bureau of Statistics Islamabad. The industries are classified according
to Pakistan Standard Industrial Classification (PSIC). CMI provides data
at four different levels of aggregation, viz. from two- to five-digits
level of aggregation. Data on 230 industries are provided at five-digits
level. One of these industrial groups, "lead, zinc, tin and their
alloys" is reported to have negative capital assets. This sector is
excluded from the analysis. So, the study uses data of 229 industrial
groups at five-digits level of aggregation.
The theoretical model given in Equations (1), (2), and (3) is
estimated with labour productivity (output per labour) as dependent
variable. Capital intensity measured by capital-labour ratio is used as
independent variable, and advertising expenditure (per labour) as the
mediating variable. The definitions for output, capital, labour, and
advertising expenditure as reported in CMI are given below:
Output is defined as the contribution of the establishments (in
thousand rupees) in each industrial group to the Gross Domestic
Product of the economy. Capital measures value of fixed assets (in
thousand rupees) of the industrial group at the end of the year.
Labour consists of average daily persons employed in the industrial
group including employees, working proprietors and unpaid family
workers. Advertising expenditures is defined as advertising cost
(in thousand rupees) during the year.
Labour Productivity is output in thousand rupees per labour.
All variables are used in regressions in logarithmic form. Ordinary
Least Square technique is used to estimate the equations. Summary
statistics, graphs, and variance-covariance matrix of the variables in
level form are given in Appendix. STATA 12 computer package is used for
estimation and testing the regression coefficients. Add-on command
"sgmediation" is used to perform tests of significance of
indirect effects. (9)
In addition to advertising expenditure other variables like levels
of education of workers and managerial staff, work environment, and
expenditures on research and development can also play mediating role.
Presently CMI data do not include these variables. A more detailed
mediating model can be used when such data become available.
4. RESULTS
The estimated results of mediation model outlined in previous
section are as follows. The capital intensity or capital-labour ratio is
found to play significant role in determination of labour productivity.
In terms of our mediation model 'c' is significant (Table 1).
Similarly, causal link from capital-labour ratio to advertising is
also statistically significant, or in terms of the mediation model
'a' is significant (Table 2). The result substantiates the
hypothesis that capital intensity is a significant predictor of
advertising expenditure.
The next step of mediation model requires that when dependent
variable is regressed on independent variable as well as mediating
variable, both explanatory variables should turn out to be significant.
Moreover, the size of coefficient of independent variable (capital
intensity in our case) should decrease. Table 3 confirms that this
condition is also fulfilled. When labour productivity variable is
regressed on capital intensity as well as advertisement, both
explanatory variables turn out to be statistically significant, and the
size of the coefficient of capital intensity decreases from
approximately 0.57 (Table 3) to approximately 0.46 (Table 5). From this
we can conclude that advertisement does indeed play a mediation role in
productivity determination.
As described in the previous section, size of mediation or indirect
effect is also given by the product 'ab', which is the product
of the coefficient of independent variable in Equation (II) and the
coefficient of mediation variable in Equation (III). The estimate of
this product turns out to be approximately 0.10. This amount
approximately equals the reduction in the estimate of the coefficient of
capital intensity due to inclusion of advertisement in the model (Tables
1 and 3).
Tests of significance of indirect effects are performed under the
assumption of normality of 'ab' as well as using the bootstrap
technique. The three tests based upon the assumption of normality of the
product 'ab' are Sobel Test, Aroian Test, and Goodman Test.
These tests give very similar results (Table 4). The results indicate
that the null hypothesis of no indirect effect can be rejected at less
than 0.001 level of significance. Table 4 also reports proportion of
total effect that is mediated, and ratio of indirect to direct effect.
After rounding, approximately 18 percent of total effect of capital
intensity on labour productivity is mediated through advertisement. The
ratio of indirect to direct effect is approximately 0.22 percent. These
numbers indicate relative strength of the effect of mediating variable.
Significance of direct and indirect effects is also tested by using
bootstrapped estimates. Bootstrap technique is applied with 5000
replications. The percentile confidence intervals sort the bootstrap
estimates in ascending order and choose the interval limits
corresponding to ([alpha]/2)x100th and (1-[alpha]/2)x100th percentile
values for a particular [alpha] level of significance. On the other hand
bias-corrected bootstrap confidence intervals adjust the end points of
the intervals depending upon whether the proportion of bootstrapped
values of indirect effects is less than those estimated from the
original data. Estimates of these two types of interval are reported in
Table 5. None of confidence intervals contain zero, which implies
significance of direct and indirect effects. Hence the null hypothesis
of zero mediation effects is rejected. These results confirm the results
from Aroin test, Sobel test and Goodman test. So we can conclude that
advertising expenditure does play a significant role as a mediation
variable.
SUMMARY AND CONCLUSIONS
This paper uses a simple mediation model to determine direct and
indirect effects of capital intensity on labour productivity in
manufacturing industries of Pakistan. A simple mediation model is
estimated using per labour advertising costs as mediating variable, and
capital intensity as independent variable. The results show that
indirect effect of capital intensity on labour productivity constitutes
approximately 18 percent of total effect. The statistical significance
of the indirect effect is tested through standard normal tests as well
as through bootstrap methods, and results are found to be statistically
significant.
The mediating role of advertising costs indicates some subtle
aspects of the relationship between capital and labour productivity.
Traditional analysis suggests that capital intensive industries are
technologically more advanced, hence their modern equipment and
innovative ways of production make labour more productive. This argument
may be partly true, but as our analysis suggests a significant portion
of this causal chain is, in fact, transmitted through advertising.
Policy-makers and regulatory bodies should keep these facts in view
while making policies or taking regulatory measures about advertising
activity and capital markets.
The study highlights the need for further research to clarify the
role of advertising; whether advertising is being used as promotion tool
or as a strategy to create entry barriers. When advertising is for
promotion and information, it acts like an input, and promotion and the
product are joint outputs in supply. So it can be regarded as a
contributor to value added. Hence advertising may be productive if it is
informative and/or persuasive, but it could be counterproductive if it
happens to adversely affect competitive structure of the market.
The study also highlights the need for further improvements in CMI
data. Presently CMI data do not include variables like levels of
education of workers and managerial staff, and expenditures on research
and development. These and other such type of variables may also play a
similar mediating role. CMI data set should be extended to include these
variables. If more variables could be included in a multi-mediation
model, it will further refine direct and indirect effects, and provide
better analyses of labour productivity.
APPENDIX
[FIGURE 1 OMITTED]
[FIGURE 2 OMITTED]
[FIGURE 3 OMITTED]
Appendix Table 1
Summary Statistics of the Variables
Variables Obs Mean * Std. Dev.
Labour Productivity 229 1279.40 2198.22
Capital Intensity 229 1026.88 1930.94
Advertising Expenditure 229 64426.24 208519.20
(per Labour)
* Labour productivity and advertising expenditures per labour
are in thousand rupees.
Appendix Table 2
Correlation Matrix
Labour Capital Advertising
Productivity Intensity Expenditure
Labour Productivity 1
Capital Intensity 0.44 1
Advertising Expenditure 0.26 0.06 1
REFERENCES
Apergis, N., et al. (2008) Innovation, Technology Transfer and
Labour Productivity Linkages: Evidence from a Panel of Manufacturing
Industries. Review of World Economics 144:3, 491-508.
Aroian, Leo A. (1947) The Probability Function of the Product of
Two Normally Distributed Variables. Annals of Mathematical Statistics
18:2, 265-271.
Arrow, K. J., et al. (1961) Capital-Labour Substitution and
Economic Efficiency. The Review of Economics and Statistics 43:3,
225-250.
Banker, Rajiv D., Indranil R. Bardhan, and Tai-Yuan Chen (2008) The
Role of Manufacturing Practices in Mediating the Impact of
Activity-based Costing on Plant Performance. Accounting, Organisations
and Society 33, 1-19.
Baron, R. M. and D. A. Kenny (1986) The Moderator-mediator Variable
Distinction in Social Psychological Research: Conceptual, Strategic, and
Statistical Considerations. Journal of Personality and Social Psychology
51, 1173-1182.
Cheema, Aftab Ahmad (1978) Productivity Trends in the Manufacturing
Industries. The Pakistan Development Review 17:1, 44-65.
Collins, L. M., J. W. Graham, and B. P. Flaherty (1998) An
Alternative Framework for Defining Mediation. Multivariate Behavioral
Research 33, 295-312.
Efron, B. (1979) Bootstrap Methods: Another Look at the Jackknife.
The Annals of Statistics 7:1, 1-26.
Ehrlich, Isaac and L. Fisher (1982) The Derived Demand for
Advertising: A Theoretical and Empirical Investigation. American
Economic Review 72:3, 366-88.
Fraser, S. and David Paton (2003) Does Advertising Increase Labour
Supply? Time Series Evidence from the UK. Applied Economics 35:11,
1357-1368.
Gelfand, Lois A., Janell L. Mensinger, and Thomas Tenhave (2009)
Mediation Analysis: A Retrospective Snapshot of Practice and More Recent
Directions. Journal of General Psychology 136:2, 153-176.
Goodman, L. A. (1960) On the Exact Variance of Products. Journal of
the American Statistical Association 55, 708-713.
Hamilton, J. L. (1972) The Demand for Cigarettes: Advertising, the
Health Scare, and the Cigarette Advertising Ban. The Review of Economics
and Statistics 54, 401-11.
Hamid, A and J. Hanns Pichler (2009) Human Capital Spillovers,
Productivity and Growth in the Manufacturing Sector of Pakistan. The
Pakistan Development Review 48:2, 125-140.
Hayes, Andrew F. (2013) Introduction to Mediation, Moderation, and
Conditional Process Analysis: A Regression-based Approach. The Guilford
Press.
Han, Kunsoo, Robert J. Kauffman, and Barrie R. Nault (2011) Returns
to Information Technology Outsourcing. Information Systems Research
22:4, 824-840.
Hochman, O. and I. Luski (1988) Advertising and Economic Welfare:
Comment. American Economic Review 78, 290-6.
Horstmann, I. J. and S. Moorthy (2003) Advertising Spending and
Quality for Services: The Role of Capacity. Quantitative Marketing and
Economics 1, 337-65.
Hussain, Karrar (2009) Causal Ordering between Inflation and
Productivity of Labour and Capital: An Empirical Approach for Pakistan.
Center for International Development at Harvard University. (C1D
Graduate Student Working Paper Series No 39).
Jung, C. and B. J. Seldon (1995) The Macroeconomic Relationship
between Advertising and Consumption, Southern Economic Journal 61:3,
577-587.
Kenny, D. A. (2008) Reflections on Mediation. Organisational
Research Methods 11, 353-58.
Kuha, Jouni and John H. Goldthorpe (2007) Path Analysis for
Discrete Variables: Education as Mediator of Social Mobility in Britain.
Department of Statistics, London School of Economics. (Research Report
Number 144).
Kutan, Ali M. and Taner M. Yigit (2009) European Integration,
Productivity Growth and Real Convergence: Evidence from the New Member
States. Economic Systems 33, 127-137.
Lomnicki, Z. A. (1967) On the Distribution of Products of Random
Variables. Journal of the Royal Statistical Society, Series B 29,
513-524.
MacKinnon, D. P., et al. (2002) A Comparison of Methods to Test
Mediation and other Intervening Variable Effects. Psychological Methods
7, 83-104.
MacKinnon, D. P., J. L. Krull, and C. M. Lockwood (2000)
Equivalence of the Mediation, Confounding, and Suppression Effect.
Prevention Science 1, 173-181.
Mahmood, Zafar and Rehana Siddiqui (2000) State of Technology and
Productivity in Pakistan's Manufacturing Industries: Some Strategic
Directions to Build Technological Competence. The Pakistan Development
Review 39:1, 1-21.
Mathieu, John E., Richard P. DeShon, and Donald D. Bergh (2008)
Mediational Inferences in Organisational Research: Then, Now, and
Beyond. Organisational Research Methods 11, 203-223.
Maydeu-Olivares, A. and N. Lado (2003) Market Orientation and
Business Economic Performance: A Mediated Model. International Journal
of Industry Management 14:3, 284-309.
Newsham, G., et al. (2009) Linking Indoor Environment Conditions to
Job Satisfaction: A Field Study. Building Research and Information 37:2,
129-147.
Pakistan, Government of (2005) Census of Manufacturing Industries
(2005-06). Federal Bureau of Statistics, Islamabad.
Pigou, A. C. (1924) Economics of Welfare. (2nd edition). London:
MacMillan and Co.
Richards, T. J. and Paul M. Patterson (1998) Dynamic
Complementarity in Export Promotion: The Market Access Program in Fruits
and Vegetables. Journal of Agricultural and Resource Economics 23:02,
319-337.
Shrout, P. E. and N. Bolger (2002) Mediation in Experimental and
Non-Experimental Studies: New Procedures and Recommendations.
Psychological Methods 7, 422-445.
Sobel, M. E. (1982) Asymptotic Confidence Intervals for Indirect
Effects in Structural Equation Models. In S. Leinhart (ed.) Sociological
Methodology 290-312. Washington, DC: American Sociological Association.
Srivastava, Medha and Alok Kumar Rai (2013) Investigating the
Mediating Effect of Customer Satisfaction in the Service
Quality--Customer Loyalty Relationship. Journal of Consumer
Satisfaction, Dissatisfaction and Complaining 26, 95-109.
STATA 12, Computer Package Developed by STATA inc.
Telser, L. G. (1966) Cutthroat Competition and the Long Purse.
Journal of Law and Economics 9:1, 259-77.
Telser, L. G. (1978) Towards a Theory of the Economics of
Advertising, In D. G. Tuerck (ed.) Issues in Advertising: The Economics
of Persuasion 71-90. Washington, DC: American Enterprise Institute.
Velucchi, and M. A. Viviani (2011) Determinants of the Italian
Labour Productivity: A Quantile Regression Approach. Statistica 71:2,
213-238.
Warner, Rebecca M. (2012) Applied Statistics: from Bivariate
through Multivariate Techniques (2nd ed.) Thousand Oaks, Calif., SAGE
Publications.
(1) In literature this phenomenon is called "long purse"
hypothesis; see for example, Telser (1966).
(2) The issue of including advertising expenditure as an input is
explained in Section 3.
(3) For reviews of mediation models in psychology and related
fields, see Gelfand, et al. (2009), Kenny (2008), and Mathieu, DeShon,
and Bergh (2008).
(4) Although a mediation model is not specifically a production
model, Equation (1) can be derived from a linearly homogeneous
production function with two inputs, in this case, labour and capital.
If we divide the production function by labour and take log of both
sides, Y becomes log of labour productivity, and X becomes log of
capital-labour ratio. In a similar way, Equation (3) can be derived from
a linearly homogeneous production function with three inputs, viz.
labour, capital, and advertising expenditure. Here M expresses log of
advertising expenditure per unit of labour.
(5) These equations do not prove causality in statistical sense.
Rather, the coefficients in these equations provide estimates of
theoretical causal links among the variable. These theoretical links
have been discussed above.
(6) For detail on the procedure see, for example, Shrout and Bolger
(2002).
(7) For detail on these three types of confidence intervals, see
Hayes (2013), p. 111.
(8) This is the latest CMI dataset presently available.
(9) This computer programme is available at the website of IDRE,
Institute for Digital Research and Education, UCLA:
http://www.ats.ucla.edu/stat/stata/faq/sgmediation.htm
Tariq Mahmood <tariqmahmood@pide.org.pk> is Senior Research
Economist, Pakistan Institute of Development Economics, Islamabad.
Table 1
Results for Productivity Regressed on Capital Intensity
Number of Obs = 229 P>
F Statistic = 107.41 [absolute
Adj R-Squared = 0.32 Std. value
Labour Productivity Coef. Err. t of (t)]
Capital Intensity 0.57 0.05 10.36 0.00
Const. 2.91 0.35 8.32 0.00
Table 2
Results for Advertising Expenditure Regressed on Capital
Intensity
Number of Obs. = 229 P>
F Statistic = 22.32 [absolute
Adj R-Squared = 0.08 Std. value
Coef. Err. t of (t)]
Capital Intensity 0.58 0.12 4.72 0.00
Const. -2.39 0.78 -3.08 0.00
Table 3
Results for Productivity Regressed on Capital Intensity
and Advertisement
Number of Obs. = 229 P>
F Statistic = 85.27 [absolute
Adj R-Squared = 0.42 value
Labour Productivity Coef. t of (t)]
Capital Intensity 0.46 8.80 0.00
Advertising Expenditure 0.18 6.57 0.00
Const. 3.34 10.19 0.00
Table 4
Tests of Significance for Indirect Effects (Tests Based
upon Assumption of Normality)
Coeff z P>
[absolute
value
of (z)]
Sobel 0.10 3.83 0.00
Aroian 0.10 3.80 0.00
Goodman 0.10 3.86 0.00
Proportion of Total Effect that is Mediated =0.18
Ratio of Indirect to Direct Effect =0.22
Table 5
Test of Significance for Indirect Effect (Test Based upon
Bootstrap Method)
No. of Obs 229
Replications 5000
Observed Bootstrap
Coeff. Std. Err. [95% Confidence Interval]
Indirect Effect 0.10 0.02 0.06 0.16 (P) *
0.06 0.16 (BC) **
Direct Effect 0.46 0.06 0.35 0.59 (P) *
0.35 0.58 (BC) **
* Percentile confidence interval.
** Bias-corrected confidence interval.
