A

Summary of Thesis

On

Estimating Handling Time in Retail Store in Supply Chain

Submitted in fulfillment

for the award of degree of

DOCTOR OF PHILOSOPHY

In

Mechanical Engineering

Submitted By:

Er. Niraj Gupta

Name of Supervisor

Prof. (Dr.) M.I. Khan

INTEGRAL UNIVERSITY, LUCKNOW

April, 2011

ii

ABSTRACT

Retailing consists of all activities involved in selling goods and services to

consumers for their personal, family or household use. Retailing is one of

the pillars of the economy in India and accounts for 35% of GDP. In today’s

global environment, many retailers are focusing on reducing costs as a

means of achieving operational excellence. Majority of the operational costs

are handling costs. Handling operations are costly and labor intensive.

In this research we have developed a model which not only describes the

handling process but also estimates the Total Handling Time per product

unit in the hyper store when replenishment of an item is from the backroom.

Total handling time includes traveling time of product from backroom to

shelf in the hyper store and the time needed to stack the product on the hyper

store shelf.

To test the model, empirical data on the handling operation was collected at

a hyper store using a stop watch. The data were collected for the entire set of

product groups available at hyper store. Data was analyzed using multiple

regression through version 17 of SPSS software.

The values of R, R2, t-test, F-test, examination of residual, value of VIF and

iii

tolerance indicate that our model is valid. The assumptions made for the

regression are met. Thus, it can be said that the model proposed in this work

for a sample can be accurately applied to the population of interest. Finally,

cross-validation of the model has also been carried out using Stein’s

equation and data splitting.

1

CHAPTER 1

INTRODUCTION

Retailing consists of all activities involved in selling goods and services to

consumers for their personal, family or household use. It covers sales of

goods ranging from automobiles to apparel and food products and services

ranging from hair cutting to air travel and computer education. Sales of

goods to intermediaries who resell to retailers or sales to manufacturers are

not considered a retail activity.

Retailing is one of the pillars of the economy in India and accounts for 35%

of GDP [1]. Over 12 million outlets operate in the country and only 4% of

them being larger than 500 sq ft. in size. The retail industry is divided into

organized and unorganized sectors. Organized retailing refers to trading

activities undertaken by licensed retailers, that is, those who are registered

for sales tax, income tax etc. These include the corporate backed

hypermarkets and retail chains, and also the privately owned large retail

businesses. Unorganized retailing, on the other hand, refers to the traditional

formats of low cost retailing, for example, the local kirana shop, owner

manned general stores, convenience stores, hand cart and pavement vendors,

etc. In India, a shopkeeper of such kind of shops is usually known as

„dukandar‟. Most Indian shopping takes place in open markets and million

of independently grocery shops called „kirana‟. Organized retail such as

supermarkets accounts for just 4% of the market as of 2008 [9] . Regulations

prevent most foreign investment in retailing. Moreover, over thirty

regulations such as “signboard licenses” and “anti-hoarding measures” may

have to be compiled before a store can open doors. There are taxes for

moving goods to states from states and even within states. The organized

retail market is growing at 35 percent annually while growth for unorganized

retail sector is pegged at only 6 percent.

2

In recent years, however, advances in computing capabilities and

information technologies, hyper-competition in the retail industry,

emergence of multiple retail formats and distribution channels, an ever

increasing trend towards a globally dispersed retail network have made the

retailer realize that to survive in the competitive market they need to give

due emphasis on reducing costs as a means of achieving operational

excellence. Since we focus on operational costs, total shelf space and

assortment are assumed to be known. The operational logistical costs made

in the part of supply chain that includes retailers‟ warehouse and the stores

are presented below [2]:

1. Handling in warehouse = 29%

2. Transportation = 22%

3. Inventory in store = 7%

4. Handling in store = 38%

5. Inventory in warehouse = 5%

It can be seen that the majority of the operational costs are the handling

costs. Handling operations are costly and labor intensive. In order to

minimize store operating expenses, one has to optimize order processing,

inventory, transportation, shelf space and handling cost. Presently, research

in retail operation does not focus together these issues in calculating the total

operational costs. They consider these issues separately [3, 4, 5, and 7].

Also, in these models the handling time and its related costs are not

considered explicitly. This research focuses exclusively on the estimation of

handling time in store when replenishment is from back room. In our case

handling time includes traveling time of product from backroom to shelf in

the hyper store and time needed to stack the product on the hyper store shelf.

3

CHAPTER 2

OBJECT & SCOPE OF THE PRESENT WORK

A critical review of the available literature ends on the following facts. For

most retailers, store handling operations are not only labor intensive but also

very costly. Empirical study by Saghir and Jonson [6] suggest that 75% of

the handling time in the retail chain occur in the store. This shows the need

for a model which can estimate the handling time in the store when

replenishment process for the items on the shelves starts from back room. In

our case handling time includes traveling time of product from backroom to

shelf in the hyper store and time needed to stack the product on the hyper

store shelf. The objective of the work can be listed as follows:

To develop a model which can adequately describe the handling

process and estimate the total handling time per product unit in the

store when replenishment of an item is from the back room.

Collect data to test the model.

Analyze data using multiple regressions and estimate the equation for

total handling time.

Check assumptions, Test for significance, examine the residual and

check the multi-collinearity in order to validate the model.

Cross-validate the model using Stein‟s equation and data splitting.

4

CHAPTER 3

MODEL DEVELOPMENT

3.1 BACKGROUND FOR ESTIMATING HANDLING TIME

As per Van Zelst et al. [8], handling costs in the stores in the two retail

chains investigated are equal to around 60 million dollar per year. In another

empirical study by Saghir and Jonson [6], they have found that 75% of the

handling time in the retail chain occurs in the store. This shows the need for

a model which adequately describes the handling process and estimate the

handling time in the store. We have modeled the handling activity of one of

the oldest retail store in India known as Spencer‟s hyper.

3.2 MODEL DEVELOPMENT

In the Spencer‟s hyper store the replenishment process for the items on the

shelves starts from back room (warehouse). Items arrive from the regional

distribution center and local suppliers in the backroom based on a reduced

set of underlying factors , given a specific inventory replenishment rule,

assortment, shelf space and package. Since, the shelves are organized in the

store into different product categories and handled by store workers

individually. Handling operation in the backroom and store consists of the

following steps:

a) Receiving and storing of incoming items from regional distribution center

and local suppliers in the back room.

b) Issuing and loading of items in the material handling equipment like jack

and trolley

c) Handling of items during transportation from back room to shelves in the

hyper store.

5

d) Handling of items during shelf stacking.

Operations (a) and (b) are handled by backroom persons and operations (c)

and (d) are handled by the hyper store workers.

In our analysis, handling operation during replenishment will include

operations in steps (c) and (d) which consists of the following activities:

1) Move the deliveries near the shelf from back room using Jack or

trolley.

2) Grab and unpack the case pack.

3) Search for the assigned location in the shelf.

4) Travel to the shelf.

5) Check the shelf life of inventory on the shelf.

6) Prepare the location on shelf for stacking

7) Put the new inventory on the shelf.

8) Put the old inventory back on the shelf.

Replenishment process in the store is a continuous process which depends

upon the stocking policy of the store. Estimation of handling time in stores

will consist of total traveling time , TTT (operation(c )) and total stacking

time, TST (operation (d )).

THT=TTT+TST (3.1)

Total traveling time will depend upon the distance between back room and

the shelf location in the store i.e. variable F , pace of worker, efficiency of

material handling equipment and crowd in the store which may slow down

the movement if traffic is high. Efficiency of material handling is given

irrespective of worker and we have no control over the crowd in the store.

Thus, the independent variables are hypothesized to have the following

influence on TTT,

6

(1) The higher the distance between backroom and shelf location , the

higher the TTT will be

(2) Fast worker in comparison to other worker will have less TTT.

The basic starting equation for TTT will be as follows:

TTT= a’+f’F (3.2)

Where,

a‟ = Constant

f‟ = Regression coefficient

F = Distance from backroom to shelf location in feets

TTT will remain the same whether there is one product or many product

units on the jack or trolley. Thus:

TTT/PU= a’+f’F (3.3)

Now, let us consider the second part of the equation 3.1 i.e. T.S.T (Total

Stacking Time). Shelf stacking represents the daily process of manually

refilling the shelf in the store with products from new deliveries. As with

most manual activities, such processes are often time consuming and costly.

Furthermore, unless clear and reliable work standards are implemented, such

activities may well suffer from a lot of variations which will negatively

affect the overall store performance . Shelf stacking process in the store is

seen as the reverse of the order picking process at the warehouse [8]. For

estimating the TST this work closely follows the methodology presented by

Van Zelst et al.[8]. Let us go to the details of the activities (2-8) of shelf

stacking process. Activities (2) and (4) will depend on the number of case

packs (CP) filled. Activity (3) and (6) are done only once for each stock

keeping unit (SKU) and will be independent of the number of CPs or PUs.

All stacking activities are handled by store worker, so TST will also depend

7

on the pace of the workers (not every employee works equally fast). Thus,

TST will depend on the number of CPs, PUs and the pace of the worker.

The independent variables are hypothesized to have the following influence

on the TST:

1) The higher the number of PUs to be filled, the higher the TST will be.

2) The higher the number of CPs, the higher the TST will be

3) Fast worker in comparison to other worker will have less TST.

The basic starting equation for TST will be as follows:

TST=a’’+b’’PU+c’’CP (3.4)

Where, PU = CP (Q) ; ( Q = Case pack size )

Since, we are interested in calculating TST per product unit, we rearrange

the basic equation by dividing the TST by PU and substituting PU= CP (Q).

The revised model is :

TST/PU=b’’+a’’/CP(Q)+c’’/Q (3.5)

Since, THT/PU=TTT/PU + TST/PU (3.6)

Combining equation 3.3 and 3.5 and rewriting in general form, we get:

THT/PU=a+b/CP(Q)+c/Q+fF (3.7)

Since, TST and TTT both depend on the pace of the worker i.e. not every

employee works equally fast. Consequently, (n-1 ) dummies for store worker

are added to equation 3.7 denoted as Dwi , (i= 1,………., n-1)

Where,

n = the number of store worker considered and

Dwi =1, if store worker „i‟ is selected and „0‟ otherwise.

8

Thus, general regression equation for THT/PU will be (with error term „e‟)

n-1

THT/PU=a+b/CP(Q)+c/Q+fF + ∑di Dwi +e (3.8)

i =1

Or

n-1

THT/PU = a + bK2+ cK1+ fF + ∑di Dwi +e

i =1

Where,

1) a = constant of regression equation

2) b, c, di , f = Partial regression coefficient

3) K1 = 1/Q

4) K2 = 1/CP(Q) = 1/PU

3.3 EXPERIMENTAL DESIGN

In the experiment, the data has been collected for handling activity during

replenishment process from the backroom to shelf location. The retail store

is categorized as hyper store. „Hyper‟ are mega stores, which combine a

supermarket with a department store. In the experiment, for each product

unit, total traveling time and total stacking time were measured using a stop

watch.

3.4 DATA COLLECTION

Empirical data on the handling operation was collected at a hyper store using

stop watch. But in our case the total number of observations recorded is 201.

Data was collected for the following variables:

Q = Case pack size

CP = Number of case packs

PU = Number of product units

F = Distance from backroom to shelf location in feets

9

Dw1 = 1 , If worker 1 is present otherwise “0”

Dw2 = 1 , If worker 2 is present otherwise “0”

TTT = Total traveling time from backroom to shelf location in

seconds

TST = Total stacking time in seconds

K1 = 1/Q

K2 = 1/CP(Q) = 1/PU

THT/PU = Total handling time per product unit in seconds.

The data were collected for the entire set of product groups available at

hyper store.

10

CHAPTER 4

DATA ANALYSIS

4.1 INTRODUCTION

The advent of the program like SPSS provides the unique opportunity to

apply statistics at a conceptual level without getting too bogged down in

equations. In the following paragraph we will discuss how to interpret the

output, create our multiple regression model and look at the assumptions

necessary to generalize our model. We will finish the chapter by cross

validation of the model.

4.2 INTERPRETING MULTIPLE REGRESSION

4.2.1 DESCRIPTIVES

The output described in this section is produced using the options in the

Linear Regression : Statistics dialog box. To begin with, if you select the

descriptive option, SPSS will produce the following Table 4.1.

Table 4.1: Descriptives

S.no. Variable Mean Standard Deviation

1 THT/PU 206.611 1.1009E2

2 F 114.9 25.09

3 Dw1 0.51 0.782

4 Dw2 0.53 0.50

5 K1 0.1402 0.252

6 K2 0.1149 0.240

11

This Table 4.1 tells us the mean and standard deviation of each variable in

our data set, so we know that the mean of dependent variable THT_per_PU

was 206.611. This Table 4.1 is not necessary for interpreting the regression

model, but it is a useful summary of the data.

4.2.2 MODEL SUMMARY

The next section of output describes the overall model. So it tells us whether

the model is successful in predicting THT_per_PU.

Table 4.2: Model Summaryb

Model R R- Square Adjusted R-

Square

Std. Error of

the Estimate

1 .892a .796 .765 80.52058157

a. Predictors: F, Dw1, Dw2, K1, K2

b. Dependent Variable: Total Handling Time_per_PU

In Table 4.2 the column labeled R, shows the value of the multiple

correlation coefficient between the observed values of THT_per_PU and the

values of THT_per_PU predicted by the multiple regression model. Its large

value (R=.892) indicates a strong relationship. The next column gives us a

value of R2

, which we already know is a measure of how much of the

variability in the THT_per_PU is accounted for by the predictors. Output

tells us that 79.6% of total variation in THT_per_PU is explained by five

independent variables.

The adjusted R2

gives us some idea about how well our model generalizes

and ideally we would like its value to be the same, or very close to, the value

of R2 . In our case the difference is small (infact the difference between the

12

values is .796-.765= .031 or 3.1%) . This shrinkage means that if the model

were derived from the population rather than a sample it would account from

approximately 3.1% less variance in the outcome. The last column gives us

the standard error of the estimate which is a measure of dispersion for

multiple regression.

4.2.3 MODEL PARAMETERS

So far we have looked at summary statistics telling us whether or not the

model has improved our ability to predict the outcome variable. The next

part of the output is concerned with the parameters of the model. Table 4.3

shows the model parameters.

Table 4.3: Model Parameters

Variable B Standard

Error

Beta t Sig

Constant 5.606 2.9364 - 1.91 0.024

F 1.211 0.235 0.276 5.153 0.000

Dw1 38.942 9.318 0.276 4.179 0.000

Dw2 35.014 14.65 0.159 2.390 0.018

K1 -88.465 76.129 -0.203 -1.162 0.047

K2 310.78 79.862 0.678 3.892 0.000

The first part of the Table 4.3 gives us estimates for regression coefficient

(b-values) and these values indicate the individual contribution of each

predictor to the model.

13

From the number given in „B‟ column we can read the estimated equation

as:

THT_per_PU=5.606+1.211F+38.942Dw1+35.014Dw2-88.465K1+310.789K2 (4.1)

The b-values tell us about the relationship between THT_per_PU and each

predictor. If the value is positive we can tell that there is a positive

relationship between the predictor and the outcome, whereas a negative have

positive b-values except K1 indicating positive relationship. So, as variable

distance (F) increases, THT_per_PU increases. The b-values tell us more

than this though. They tell us regarding the degree each predictor affects the

outcome if the effects of all other predictors are held constant. For example,

for variable „F‟ the b-value is 1.211.This value indicates that as distance

increases by one unit , THT_per_PU increases by 1.211 units. This

interpretation is true only if the effects of all other predictors are held

constant.

Each of these b-values have an associated standard error indicating as to

what extent these values would vary across different samples, and these

standard errors are used to determine whether or not the b-values differs

significantly from zero. A t-statistics can be derived that tests whether a b-

value is significantly different from „0‟. In simple regression, a significant

value of „t‟ indicates that the slope of the regression line is significantly

different from horizontal. But in multiple regression, it is not so easy to

visualize what the value tells us. Well, it is easiest to conceptualize the t-

tests as measure of whether the predictors are making a significant

contribution to the model. Therefore, if the t-test associated with a b-value is

significant (If the value in the column labeled “sig” is less than .05(Level of

significance) then the predictor is making a significant contribution to the

model. The smaller the value of “sig” (and the larger the value of „t‟), the

14

greater the contribution of that predictor. From the magnitude of t-statistics

we can see that the predictor variable distance has the largest impact and

variable K1 has the smallest impact on THT_per_PU.

The b-values and their significance are important statistics to look at;

however, the standardized versions of the b-values are in many ways easier

to interpret (because they are not dependent on the units of measurement of

the variables). The standardized beta values are provided by SPSS (labeled

as beta, β) and they tell us the number of standard deviation that the outcome

will change as a result of one standard deviation change in the predictor. The

standardized beta values are all measured in standard deviation units and so

are directly comparable; these therfore provide a better insight into the

importance of a predictor in the model. The standardized beta values for „F‟

and Dw1 variable are identical indicating that both variables have a

comparable degree of importance in the model. To interpret these values

literally , we need to know the standard deviation of all of the variables and

these values can be found in Table 4.1.For example for variable distance,

„F‟, the standardized β =.276. This value indicates that as distance increases

by one standard deviation (25.09ft), THT_per_PU increase by .276 standard

deviations. The standard deviation for THT_per_PU is 110.09 and so this

constitute a change of 30.38 (.276 * 110.09). Therefore, for every 25.09 ft

more on distance, an extra 30.38 THT_per_PU (seconds) is increased. This

Interpretation is true only if other predictors are held constant.

Does THT_per_PU depend on independent variables. Is independent

variable significant? Our question leads to hypothesis of the form:

H0 : Bi = 0 (Null Hypothesis: Xi is not a significant explanatory variable)

H1 : Bi ≠ 0 (Alternate Hypothesis: Xi is a significant explanatory variable)

15

We will use the column headed “sig” to test whether Xi is a significant

explanatory variable. Entries in this column are probability values for the

two-tailed test of hypothesis. We only need to compare these probability

values with α, the significance level of the test, to determine whether Xi is a

significant explanatory variable for THT_per_PU. In our case α = 0.5 and

the value in the column is less than .05. Thus null hypothesis is rejected and

Xi is a significant explanatory variable.

4.2.4 INFERENCE ABOUT REGRESSION AS A WHOLE (F-TEST)

Given regression, it is natural to ask whether the value of R2

really indicate

that the independent variables explain THT_per_PU or might have happened

just by chance. Is the regression as a whole significant. Whether all the Xi

taken together significantly explains the variability observed in

THT_per_PU. Our question leads to hypothesis of the form:

H0 : B1 = B 2= B3 =….. = Bk = 0 (Y does not depend on Xis)

H1 : At least one Bi ≠ 0 (Y depends on at least one of Xis)

Table 4.4 below gives SPSS output for our problem.

Table 4.4: ANOVA

Description Sum of

Squares

df Mean

Squares

F Sig

Regression 1160046.375 5 232009.275 35.785 0.000

Residual 1264294.991 195 6483.564

Total 2424341.366 200

This part of output includes the computed F ratio for the regression and is

called analysis of variance for regression.

16

The tabulated value of F at alpha=.05 and df 5 and 195 is equal to 3.02,

which is less than the calculated value 35.784. So the Null Hypothesis that

coefficients are equal to zero is rejected. Hence, regression as a whole is

highly significant. We can reach the same conclusion by noting that the

output “sig” is 0.000 because this probability value is less than our level of

significance α = 05. We conclude that regression as a whole is significant.

We can interpret these results as meaning that the model significantly

improved our ability to predict the outcome variable.

4.3 MULTICOLLINEARITY IN MULTIPLE REGRESSION

This condition refers to situation in which one or more of the predictors are

highly correlated with each other. Table 4.5 provides some measures of

whether there is collinearity in the data. Specifically, it provides the

Variance inflationary factor (VIF) and tolerance (with tolerance being 1

divided by the VIF) .

Table 4.5: Multicollinearity

S.no. Independent

Variable

VIF TOLERANCE

1 F 1.072 0.932

2 Dw1 1.637 0.610

3 Dw2 1.657 0.603

4 K1 1.1392 0.877

5 K2 1.1354 0.880

For our model the VIF values are all well below 10 and the tolerance

statistics all well above 0.2; therefore, we can safely conclude that there is

17

no collinearity within our data. To calculate the average VIF we simply add

the VIF values for each predictor and divide by the number of predictors.

Average VIF = 1.328

The average VIF is near to 1 and this confirms that collinearity is not a

problem for this model and our model is valid.

4.4 CHECKING ASSUMPTIONS

To draw conclusions about a population based on regression analysis done

on a sample, several assumptions must be true. In our case, all predictors are

of variable type, have non-zero variance, multicollinearity does not exist and

errors are distributed normally. Thus, the assumptions of regression are met.

4.5 WORKER 1 VS WORKER 2

In our analysis we have selected two workers for handling activities. The

following Table 4.7 gives the statistics associated with them.

Table 4.7: WORKER 1 VS WORKER 2

Sl Description Dw1=0, Dw2=1 Dw1=1, Dw2=0

1 No of cases 107 94

2 Mean THT/PU(sec.) 209.97 202.91

From the data given in the Table 4.7 we conclude that mean THT/PU

(202.91 sec.) is less when worker 1 was selected for handling activities.

Thus, worker 1 is more efficient than worker 2 in handling the product units.

18

4.6 CROSS VALIDATION OF MODEL

Even if we can‟t be confident that the model derived from our sample

accurately represents the entire population, there are ways in which we can

assess how well our model can predict the outcome in a different sample.

Once we have a regression model there are two main methods of cross

validation. They are data splitting and use of Stein‟s equation. The analysis

done using data splitting and Stein‟s equation shows that cross-validity of

the regression model is very good.

19

CHAPTER 5

RESULTS

RESULTS

The following results have been obtained by multiple regression analysis:

1) The larger value of multiple correlation coefficient (R=.892) between

observed and predicted value of THT_per_PU indicate a strong

relationship.

2) The value of coefficient of multiple determination (R2

= .796) tells us

that 79.6% of total variation in THT_per_PU are explained by five

independent variables.

3) The estimated equation is :

THT_per_PU=5.606+ 1.211F+38.942Dw1+35.014Dw2-88.465K1+310.789K2

4) The value in the column labeled “sig” in Table 4.3 is less than

.05 (level of significance) . Thus, the predictor is making significant

contribution to the model.

5) From the F-test , we can conclude that regression as a whole is

significant.

6) For our model the VIF values are all well below 10 and the tolerance

statistics, all well, above 0.2; therefore, we can safely conclude that

there is no collinearity within our data and our model is valid.

7) The list of assumptions of regression is met. Thus, we can say that

the model we proposed for a sample can be accurately applied to the

population of interest.

20

CHAPTER 6

CONCLUSIONS AND FUTURE RESEARCH DIRECTIONS

6.1 CONCLUSIONS

The following conclusions can be drawn from this study:

1) The empirical findings of this study offer the practitioners the

opportunity for a better control of the overall logistical costs.

2) We have considered the replenishment process from the backroom,

which is a part of replenishment process in Indian environment

settings while it have not been considered in the earlier model [8].

3) The effect of the work pace of a worker shows that worker 1 is more

efficient than worker 2 . Also, less THT/PU for efficient worker

indicates that our hypothesis is valid. This result suggests that worker

training is an important aspect to reduce THT/PU.

4) F-statistics and other statistics considered indicate that independent

variables and regression as a whole is significant. Thus, our model is

valid.

5) Histogram of regression residual indicates that normality assumption

is not violated.

6) Collinearity test performed indicates no problem with regard to

multicollinearity for the estimated model.

6.2 FUTURE RESEARCH DIRECTIONS

Four research avenues emerge as important future research directions.

1) Since, we focussed on operational costs; total shelf space and the

assortment are assumed to be known. Assortment planning and shelf

21

space allocation are important issues in retail. So tradeoff between

shelf space, inventory costs etc. should also include handling costs in

future work.

2) Time lost due to interruption during transportation of an items from

backroom to shelf location due to movement of the crowd in the store

or helping the customer are not considered in the estimation of

THT/PU. This time loss could be considered in future work.

3) The effect of human resource variables such as employee turnover,

training and workload on handling inaccuracy also needs to be

examined.

4) In addition, the future researchers may examine the effect of the

execution of the handling operations on other financial or non-

financial measures of store performance.

We hope that the analysis and empirical study presented in this research will

provide the foundation for future research that will advance the state of the

art in retail supply chain management and provide significant additional

value for retailer‟s supply chain operations.

22

REFERENCES

1. Bajaj, C., Srivastava, N.V., Tuli, R., (2008), Retail Management,

11th Impression, Oxford University Press, New Delhi.

2. Broekmeulen,R., Van Donselaar, K., Fransoo, J., Van Woensel,T.,

(2006), “The Opportunity of Excess Shelf Space in Grocery Retail

Store”, Operations Research 49, 710-719.

3. Cachon, G., (2001), “ Managing a Retailer‟s Shelf Space , Inventory

and Transportation”, Manufacturing and Service Operation

Management ”, 3, 211-229.

4. Corstjens, M., and Doyle, P., (1981), “ A Model for optimizing

Retail Shelf Space Allocations”, Management Science ,27, 822-833.

5. Dreze,X., Hoch, S.J., and Purk, M.E., (1994), “ Shelf Management

and Space Elasticity”, Journal of Retailing, 70, 301-326.

6. Saghir, M., and Jonson, G., (2001), “ Packaging Handling

Evaluation Methods in the Grocery Retail Industry”, Packag

Technol Sci, 14, 21-29.

7. Urban,T.L.,(1998), “ An Inventory-Theoretic Approach to Product

Assortment and Shelf Space Allocation”, Journal of Retailing, 74(1),

15-35.

8. Van Zelst, S., et al. (2009), “ Logistics Drivers for Shelf Stacking in

Grocery Retail Stores; Potential for Efficiency Improvement”,

International Journal of Production Economics, v121, 620-632.

9. http://www.economist.com/displayStory.cfm?story_id=11465586,

Retailing in India : Unshackling the chain stores ,The Economist, 29

May 2008.