You are on page 1of 8

Learning Curves in Food Services

Author(s): Dayr A. Reis


Source: The Journal of the Operational Research Society, Vol. 42, No. 8 (Aug., 1991), pp.
623-629
Published by: Palgrave Macmillan Journals on behalf of the Operational Research Society
Stable URL: https://www.jstor.org/stable/2583781
Accessed: 24-09-2018 05:52 UTC

REFERENCES
Linked references are available on JSTOR for this article:
https://www.jstor.org/stable/2583781?seq=1&cid=pdf-reference#references_tab_contents
You may need to log in to JSTOR to access the linked references.

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

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

Operational Research Society, Palgrave Macmillan Journals are collaborating with JSTOR
to digitize, preserve and extend access to The Journal of the Operational Research Society

This content downloaded from 202.43.95.117 on Mon, 24 Sep 2018 05:52:31 UTC
All use subject to https://about.jstor.org/terms
J. Opl Res. Soc. Vol. 42, No. 8, pp. 623-629, 1991 0160-5682/91 $3.50 + 0.00
Printed in Great Britain. All rights reserved Copyright ?0 1991 Operational Research Society Ltd

Learning Curves in Food Services


DAYR A. REIS
College of Business Administration, University of Wisconsin, USA

Empirical research on learning curves and documented applications of the learning curve concept to
service systems are rare in the literature. In this paper, a learning curve model proposed by the author in
earlier research focusing on manufacturing systems is extended to typical food service operations per-
formed by the same contractor at several military facilities in the United States and abroad. An actual
application of the model to a bidding and work force planning situation is also described.

Key words: learning curve, bidding, manpower planning

INTRODUCTION

The first published account of the learning curve phenomenon at the organizational level is due to
T. P. Wright,' who contributed a model for estimating the increases in direct labour productivity
that were observed to follow the introduction of new airframe models. At the individual level, the
learning phenomenon has been studied by psychologists such as Bryan,2 Thorndike3'4 and
Thurstone' since the turn of the century.
A survey of the literature by Yelle6 indicates that the bulk of the works published in the period
1935-69 focused on military and industrial applications with an objective of controlling costs. The
aforementioned literature entailed modelling and the estimation of learning curve parameters,
which were subsequently used to set time standards, design incentive systems, select vendors and
award contracts.
In the last two decades the learning curve concept has been found useful in many new areas,
notably in corporate strategic decision-making. The work by the Boston Consulting Group,7
Conley8 and Abernathy and Wayne9 bears testimony to the wide application of the experience
curve concept. The fact that the learning curve still continues to open nlew avenues of research can
be substantiated in the recent studies on the persistence and transfer of learning in organizations'0
and on the numerous factors that can influence the learning process such as engineering effort,
shared learning2 and forgetting.1 3
Several traditional issues, such as the appropriate form of the model, the estimation of its
parameters, the prediction of plateaux in the learning curve and their duration are still questions
of significant concern. Surprisingly, empirical research on learning curves for service systems and
documented applications of the learning curve concept to service systems are rare in the literature.
In fact, Day and Montgomery14 note that service industries have been especially resistant to
experience analysis. Carman and Langeard15 state that the effect of learning on total costs has
never been demonstrated in a service situation. In another survey of the progress function liter-
ature by Dutton et al.,'6 an account of over 150 studies pertaining to manufacturing processes are
discussed. This obviates the fact that the 'learning' phenomenon has not been given due credence
in the service sector of the economy.
Most experience curves reflect the joint effects of learning, technological advances and scale. In
the present study, learning by doing is the only relevant component of the experience curve. The
highly labour-intensive nature of the service provided, the absence of automation replacing
labour, process and product changes that produce yield improvement, product standardization
and similar effects provide an unique opportunity for observing and measuring the 'learning by
doing' phenomenon in an organizational setting. The 'pure' learning curve thus obtained reflects
the increasing productivity of the labour input as a result of practice and the exercise of ingenuity,
skill and increased dexterity in repetitive activities; of management, in better planning, organizing
and directing the work activities; and of the equipment, as workers become better acquainted with
their operation and-maintenance.
In this paper, a learning curve model proposed by the author in earlier research work'7 is
extended to typical food service operations performed at several American military facilities in the

Correspondence: D. A. Reis, College of Business Administration, University of Wisconsin, La Crosse, WI 54601, USA.

623

This content downloaded from 202.43.95.117 on Mon, 24 Sep 2018 05:52:31 UTC
All use subject to https://about.jstor.org/terms
Journal of the Operational Research Society Vol. 42, No. 8

USA and abroad. First, the model and the extent of its empirical support are reviewed. Next, the
empirical findings on the parameters of the learning curves for food service operations are report-
ed and assessed. Another section describes an actual application of the model to a bidding and
manpower planning situation. Conclusions are drawn in the final section.

FORM OF THE MODEL

Manufacturing progress functions have been documented for years in the literature to describe
the learning or adaptation phenomenon. The manufacturing progress function may be generally
defined as the relationship in which the labour input per unit used in the manufacture of a
product tends to decline by a constant percentage as the cumulative quantity produced is doubled.
The, concept can be illustrated by Figure 1.
y

C
a

YU

-xl I x
Cumulative units of production
FIG. 1. Progress curve in arithmetic graph.

Where this relationship is present, it may be represented by a straight line in a double logarith-
mic scale (Figure 2).
y

b=tan p

I~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
I~~~~~~~~~~X \

x~l x
Cumulative units of production
'FIG. 2. Log-log plot.

Elsewhere, the author suggested that the manufacturing progress phenomenon can be described
and its course predicted by the following empirical equations:17

Y = Ax-b, (1)

b = m + n In A, (2)
where
Y = y(x)/Y.;
A = alyu;
y(x) = the labour hours required to produce the xth unit in cumulative production;
x = the cumulative units of output;
a, b = parameters of the model;

(Xe, y") = the ultimate point; and


m, n = constants of the parameter model given by (2).

624

This content downloaded from 202.43.95.117 on Mon, 24 Sep 2018 05:52:31 UTC
All use subject to https://about.jstor.org/terms
D. A. Reis Learning Curves in Food Services

Since (x", yu) is an estimated future point on the curve, and provided that (1) and (2)
cally valid, it can be shown that

m
b =1-n in x (3)

and

a = YUxU. (4)
Formulae (3) and (4) can be used to predict the parameters a and b for a new start-up.
Equations (1) and (2) were found to be empirically correct in an earlier study'7 with data
collected from 172 separate cases of products and processes start-ups that occurred in four differ-
ent countries and eleven distinct plants. In the same study, the parameter model (2) was supported
among operations of the same product, among products within the same plant, and among groups
of products of similar technology within the same industrial sector of a foreign economy. Previous
findings by Asher18 and Baloff'9'20 also add to this body of evidence. In Asher's study the corre-
lation is among start-ups that occurred in different facilities. In Baloff's study, the correlation is
among start-ups that occurred in the same facility.
Further clarification about the meaning of the variables in the manufacturing model is required.
As to the dependent variable (Y), it is standard practice in manufacturing systems to develop
predetermined engineered standards of performance (yu). Thus, it is possible to define an ind
productivity Y = y/yu relating the actual direct-labour hours consumed per unit (y) to the pro-
duction standard (yu). Also, direct-labour hours data (y) are usually reported in relation to stan-
dard accounting time periods, yielding average direct-labour hours per unit figures for the 'lot' of
product produced during the accounting period. As to the independent variable (x), cumulative
output statistics indicate the total output of the product (from inception of manufacture) that is
achieved at the end of the accounting period.
A final word about the measurement period. Production records are usually summarized on a
monthly basis. The physical output of the various products and operations, as well as the pro-
ductivity indices, are calculated at the end of each calendar month. In most cases, monthly obser-
vations give a good description of the course of a start-up.

EMPIRICAL FINDINGS

The results relate to the performance of food service attendant services by the same contractor
at several American military facilities located in the USA and abroad. The following 10 coded
cases can be considered representative in terms of the number of meals served in a given period: A
and F are Air Force base facilities; B and J are Army facilities; D, E, G and J are Navy facilities
located in the USA and abroad, and C is an Air National Guard facility.

Nature of tasks performed


The typical tasks performed by the contractor are those related to serving and replenishing
food, -setting and clearing tables, sanitizing facilities and equipment, preparing fresh fruits and
vegetables prior to cooking, preparing all salads and beverages, serving prepared food items,
handling foods, supplies, and equipment, maintaining the grounds of the assigned buildings, main-
taining the food service equipment and quality-controlling the quality of the services provided.

Training and quality control

All work must conform to pre-established standards of performance. Prior to starting work,
contractor personnel receive instruction in the principles and practices of food services sanitation
given by the base medical services personnel. A separate basic course for food handlers and
another advanced course for supervisors are usually given. Refresher training is repeated annually
or as often as necessary. Contractor personnel are also trained in fire prevention as scheduled by
the government.

625

This content downloaded from 202.43.95.117 on Mon, 24 Sep 2018 05:52:31 UTC
All use subject to https://about.jstor.org/terms
Journal of the Operational Research Society Vol. 42, No. 8

The contractor must also establish a complete quality control program to ensure the fulfilment
of required stipulations. The QC program usually includes an inspection system that covers the
services to be performed, a method of identifying and preventing recurring deficiencies in the
required service performed before the level of service is unacceptable, and documentation of all
inspections conducted by the contractor and the corrective action taken.
The government monitors the contractor's performance using management information
systems, random sampling, checklists and formal customer complaints. Medical service personnel
also inspect for compliance with sanitary standards. Unsatisfactory quality performance results in
payment deduction and, if recurrent, may lead to contract termination.

The start-up data

The data used were drawn from the contractor's production control reports. Labour expendi-
ture was measured in terms of cumulative average direct labour hours (-) for the cumulative
number of units produced (i.e. meals served) from inception of the contract up to the end of each
production control period (month). Finally, the productivity index (Y) was calculated according to
Y = y/yu and related to the total cumulative output (x) achieved at the end of each production
period (month). In the absence of predetermined engineered standards of performance, the yu
values were taken as the average hours per meal at the beginning of the observed plateau.

Startup analysis

The start-up data (10 start-ups) were regressed against the log-transformed form of the model,
log Y = log A - b log x, using the least-squares criterion. Regression results are shown in Table 1,
which shows the start-up code, the A and b parameter values, the learning rates (i), the number of
observations (N), the coefficients of determination (r2), the t-ratios, and the F statistics. The learn-
ing rates are calculated according to: i = 1/2".

TABLE 1. Regression results-food service start-ups

Code A b i(%) N r2 tr tb F

A 3.73 0.0983 93.4 11 0.874 -7.90 -7.90 62.47


B 6.76 0.188 87.8 4 0.794 -2.98 -2.98 7.72
C 2.99 0.131 91.3 4 0.960 -6.93 -6.92 47.94
D 2.76 0.100 93.3 5 0.953 -7.80 - 7.78 60.47
E 3.05 0.100 93.3 7 0.875 - 5.91 -5.91 34.90
F 2.18 0.0610 95.8 8 0.980 -17.15 -17.08 291.75
G 1.38 0.0264 98.2 6 0.871 -5.19 - 5.19 26.92
H 4.08 0.117 92.2 6 0.967 -10.82 -10.82 117.16
I 11.78 0.231 85.2 8 0.928 -8.80 -8.77 77.33
J 12.12 0.238 84.8 8 0.931 -8.99 -8.96 80.96

Considering the 10 start-ups, learning percentages are in the range of 85-98%.


The descriptive efficiency of the start-up model given by equation (1) is supported by the results
in Table 1. The closeness of fit can be judged by the coefficient of determination (r2). In nine out of
10 cases, the model explains at least 87.1% of the total variance in the dependent variable.
The significance of the correlation coefficient (r) can be tested by tr = r(N - 2)12/( -r .
The tb-ratio is obtained by dividing the calculated regression coefficient by its own standard error.
If the assumptions of normality and common variance are made, the null hypothesis p = 0 and
fi = 0 can be rejected at the 0.995 level in the case of start-ups A, D, E, F, G. H, I and J; at the
0.975 level in the case of start-up C and at the 0.95 level in the case of start-up B. The F values
indicate that the coefficient of determination was found significant at the 0.975 level in all cases
except B.

Parameter analysis

The parameter model given by equation (2) is supported by the results of regression analysis of
the start-up parameters A and b. Among the 10 start-ups, the parameter A values are directly
correlated with the parameter b values (or inversely correlated with the i values), suggesting that

626

This content downloaded from 202.43.95.117 on Mon, 24 Sep 2018 05:52:31 UTC
All use subject to https://about.jstor.org/terms
D. A. Reis Learning Curves in Food Services

the rate of improvement is greater for start-ups that begin at relatively high direct-labour expendi-
ture rates. In other words, initial efficiency is directly correlated to the b parameter. The impres-
sive strength of the relationship is demonstrated by the regression results shown in Table 2. The
coefficient of determination, the t ratios and the F statistics are sufficiently large to conclude that
the parameter model given by equation (2) provides an efficient description of the parameter data
from the contractor's start-ups.

TABLE 2. Parameter model regression results

No. of
cases m n r2 t tb F

10 -0.0055 0.0967 0.957 13.34 13.30 176.76

A tentative explanation for the correlation between initial productivity and learning rate is that
both management and production personnel react to the reported level of initial productivity in
ways that tend to influence the rate of progress. Management, for example, tend to allocate more
technical resources and to nurture more a start-up that had a poor beginning than one that has
begun relatively well. Production personnel say that they are more strongly motivated towards
improving productivity in the case of a poor start-up than in the case of a start-up that had a
relatively high initial productivity.'9

Parameter prediction

Prediction of the a and b parameters has been a major concern in applying the learning curve
concept. One approach to this problem that was advanced in earlier research work17 appears to
be supported by the strong correlation between the parameter values found among the start-ups
in the present study. Considering all 10 start-ups, the following empirical equation seems ade-
quate:

b =-0.0055 + 0.0967 In A (5)

Steady-state plateaux

The study also indicates the occurrence of steady-state conditions in nine out of 10 start-ups,
usually between 4 and 8 months of operation. However, start-up A took 11 months to level off.

A BIDDING SITUATION

The contractor has used the model to estimate the total direct-labour hours needed to bid for
new contracts. The model was also used to estimate the labour requirements per month and
subsequently to control the work force according to the established budget. Assuming the pro-
duction schedule as forecasted in the contract (Table 3, column 2), the following steps were fol-
lowed.

TABLE 3. Workforce requirements

Production schedule (forecast) Work force calculation

Cumulative Cum. av. Cum. total Total hours Total direct


Month Meals/month meals hours per meal hours per month employees/month

1 6500 6,500 0.130 845 845 4.8


2 6500 13,000 0.118 1534 689 3.9
3 7500 20,500 0.112 2296 762 4.3
4 7500 28,000 0.107 2996 700 4.0
5 8000 36,000 0.104 3744 748 4.3
6 6000 42,000 0.102 4284 540 3.1
7 6000 48,000 0.100 4800 516 2.9
8 6000 54,000 0.0985 5319 519 2.9
9 3500 57,500 0.0977 5618 299 1.7
10 5500 63,000 0.0965 6080 462 2.6
11 5500 68,500 0.0955 6542 462 2.6
12 5500 74,000 0.0945 6993 451 2.6

627

This content downloaded from 202.43.95.117 on Mon, 24 Sep 2018 05:52:31 UTC
All use subject to https://about.jstor.org/terms
Journal of the Operational Research Society Vol. 42, No. 8

Determination of the ultimate unit of production

Let Tu denote the estimated period of time (in months) elapsed from the inception of production
until the ultimate unit of production (x") is reached (Figure 1). The yu value corresponds to the
estimated direct-labour hours required to produce a unit at the ultimate unit of production. The
firm estimates based on historical data are: Tu = 7 months and yu = 0.10 hours per meal. From
the forecasted production schedule (Table 3, column 2): the unit in cumulative production that
corresponds to Tu = 7 months is the 48,000th meal.

Calculation of parameters b and a values

Parameter b value is calculated from formula (3), with m = -0.0055 and n = 0.0967 taken from
equation (5):

m -0.0055
b = 1-n n x 1- 0967 ln(48000) = 0.130 (a 91.4% learning curve).

Parameter a value is calculated according to formula (4):

a = yUxb = 0.10 (48,000)0.130 = 0.406 (hour, first meal).

Total labour requirements

The total direct-labour hours for bidding purposes (12 months) can be calculated as follows:

YT(X) = ax' -b = 0.406(7400)0. 70 = 6995 (hours).

Labour requirements per month

In Table 3, the cumulative average hours per meal (column 4) were calculated according to the
following equation:

Rx) = 0.406x 0.1'30.


For example, for month 4

5'(28,000) = 0.406(28,000)-0?130 = 0.107.

In taking these averages and multiplying them by the cumulative number of meals expected to be
served, the cumulative total hours for the first four months is:

28,000(0.107) = 2996 (hours).

The differences from month to month represent the total direct labour hours for any one month
(column 6). For example, the total direct labour hours for the fourth month alone is:

2996 - 2296 = 700 (hours).

If the latter figures are divided by the number of hours worked per month (in this case 176), the
needed number of direct full-time employees can be obtained. Thus, 700 . 176 = 4.0 employees
are required in the fourth month. The calculation assumes that learning continues throughout the
12-month period of the contract. For security reasons, neither the forecasted number of meals nor
the location of the military facilities have been disclosed in this paper. The figures in Table 3,
column 2 are fictitious.
The productivity changes that accompany a start-up are significant enough to be taken into
consideration in the bidding and work force planning process. Assuming that the duration of the
start-up was 7 months, Table 3 shows that the average hours per meal for the first month of the
contract was expected to be 1.3 times greater than the average hours per meal for the seventh
month. If learning continues throughout the 12-month period of the contract or if the learning
curve is steeper than the one used in the case, the productivity improvement can be even larger.
In bidding contracts usually involving many thousands of labour-hours, at typical labour rates
of $1 2-$ 15 per hour, the learning curve effect cannot be neglected.

628

This content downloaded from 202.43.95.117 on Mon, 24 Sep 2018 05:52:31 UTC
All use subject to https://about.jstor.org/terms
D. A. Reis Learning Curves in Food Services

CONCLUSIONS

Our findings indicate that the start-up model given by equation (1) efficiently describes the
productivity improvement phenomenon in several food service operations.
The findings of this research and previous findings constitute adequate evidence to suggest that
the parameter model given by equation (2) is an effective means of predicting the mathematical
slope (parameter b) of a new start-up, for both manufacturing and service systems.
This study presents evidence of steady-state plateaux occurring in the learning curve after. 6
months of operation on the average. Thus, one cannot assume that the productivity improvement
will continue indefinitely. However, this method can be applied to 'short-term and repetitive' tasks
for smoothing the labour requirement.
The current results serve to demonstrate the existence and relevance of the learning phenome-
non in service environments, an area so far neglected in the literature of learning curves.

REFERENCES

1. T. P. WRIGHT (1936) Factors affecting the cost of airplanes. J. Aer. Sci. 3, 122-128.
2. W. L. BRYAN and N. HARTERN (1899) Studies in the telegraphic language. Psychol. Rev. 6, 346-376.
3. E. L. THORNDIKE (1898) Animal intelligence: an experimental study of the association processes in animals. Psychol.
Rev. Ser. Monogr. Supply. 2, 1-109.
4. E. L. THORNDIKE (1914) The Psychology of Learning. Columbia University, New York.
5. L. L. THURSTONE (1919) The learning curve equation. Psychol. Monogr. 26, 1-51.
6. L. E. YELLE (1979) The learning curve: historical review and comprehensive survey. Decis. Sci. 10, 302-328.
7. Boston Consulting Group (1970) Perspectives on Experience. Boston Consulting Group, Boston.
8. P. CONLEY (1970) Experience curves as a planning tool. IEEE Spectrum 7, 63-68.
9. W. J. ABERNATHY and K. WAYNE (1974) Limits of the learning curve. Harv. Bus. Rev. 52, 109-119.
10. L. ARGOTE, S. L. BECKMAN and D. EPPLE (1990) The persistence and transfer of learning in industrial settings. Mgmt
Sci. 36, 140-154.
11. A. MODY (1989) Firm strategies for costly engineering learning. Mgmt Sci. 35, 496-512.
12. P. S. ADLER (1990) Shared learning. Mgmt Sci. 36, 938-957.
13. C. D. BAILEY (1989) Forgetting and the learning curve: a laboratory study. Mgmt Sci. 35, 340-352.
14. G. S. DAY and D. B. MONTGOMERY (1983) Diagnosing the experience curve. J. Mktg 47, 44-58.
15. J. M. CARMAN and E. LANGEARD (1980) Growth strategies for service firms. Strat. Mgmt J. 1, 7-22.
16. J. M. DUTTON, A. THOMAS and J. E. BUTLER (1984) The history of progress functions as a managerial technology. Bus.
Hist. Rev. 58, 204-233.
17. D. R. A. REIS (1989) Progress functions-a general model. Presented at the ORSA/TIMS Joint National Meeting, New
York City, 17 October 1989. Submitted to Int. J. Prod. Res.
18. H. ASHER (1956) Cost-Quantity Relationships in the Airframe Industry. Rand Corporation, Santa Monica, California.
19. N. BALOFF (1963) Manufacturing startup: a model. PhD. Dissertation, Stanford University.
20. N. BALOFF (1967) Estimating the parameters of the startup model-an empirical approach. J. Ind. Engng 18, 248-253.

629

This content downloaded from 202.43.95.117 on Mon, 24 Sep 2018 05:52:31 UTC
All use subject to https://about.jstor.org/terms

You might also like