20

Click here to load reader

Forecasting accuracy: comparing the relative effectiveness of practices between seven developed countries

Embed Size (px)

Citation preview

Page 1: Forecasting accuracy: comparing the relative effectiveness of practices between seven developed countries

Ž .Journal of Operations Management 16 1998 271–290

Forecasting accuracy: comparing the relative effectiveness ofpractices between seven developed countries

John G. Wacker a,), Linda G. Sprague b,c,d

a Department of Management, College of Business, Iowa State UniÕersity, Ames, IA 50011-2065, USAb Cranfield UniÕersity, Bedford, UK

c UniÕersity of New Hampshire, Durham, NH, USAd Manchester Manufacturing Management Center, Manchester, NH, USA

Abstract

This study investigates the effect of forecast practices on forecast error for seven developed countries. The questions itaddresses are: What are the differences and similarities between countries in how the forecast is developed and how accurateit is? Does the purpose of the forecast affect its accuracy? Does the use of quantitative techniques improve forecast error?How does the forecaster’s use of subjective factors affect forecast accuracy? And are there significant differences in theunderlying country’s cultural traits that affect these practices? Most forecast empirical studies investigate the degree of

Ž .accuracy and the quantitative methods used to estimate forecast accuracy Mentzler and Cox, 1984 . However, to date, nostudy has investigated the cultural variables underlying the forecast variables. These variables are well-known in the

Ž .international behavioral literature Hofstede, 1980, 1983, 1994; Ronen and Shenkar, 1985 . This study investigatesforecasting practices in Germany, Japan, Mexico, New Zealand, Spain, Sweden and the United States to determine the

Ž .differences among managerial behaviors that affect forecast accuracy. According to Hofstede 1994 , there are distinctcultural differences between countries in terms of power–distance, uncertainty avoidance, individualismrcollectivism, andmasculinityrfemininity. These cultural differences provide reasons for believing there exists between country differences in:forecast development, how the forecast is used, the degree of use of models, and subjective factors used. This study’s resultssuggest that each country has a different perspective on the basic forecasting development, methods subjective factorsconsidered in the forecast. These differences partially can be explained by Hofstede’s four cultural values dimensions:power–distance, uncertainty avoidance, individualismrcollectivism, and masculinityrfemininity indices. The statisticalanalysis suggests that power–distance tends to increase computer use along with statistical methods and decrease the use ofinternal subjective information. Uncertainty avoidance does not significantly affect forecast procedures. Firms in countrieswith high individualism tend to rely more heavily on subjective information. Firms in countries that have high masculinitytend to use subjective information to gain an advantage over their competitors. Therefore, the general conclusions are thatbetween country differences in decision-making practices can be partially explained by cultural factors. q 1998 ElsevierScience B.V. All rights reserved.

Keywords: Forecasting accuracy; Decision-making process; Cultural hypothesis

) Corresponding author.

0272-6963r98r$19.00 q 1998 Elsevier Science B.V. All rights reserved.Ž .PII S0272-6963 97 00042-9

Page 2: Forecasting accuracy: comparing the relative effectiveness of practices between seven developed countries

( )J.G. Wacker, L.G. SpraguerJournal of Operations Management 16 1998 271–290272

1. Introduction

Since the economic success of Japanese manufac-turing in the 1970s and 1980s, academics have triedto improve the decision-making process in differentcultural environments. The ‘cultural hypothesis’, asit is sometimes called, suggests there are basic cul-tural differences that give some cultures a competi-tive advantage over other cultures in information

Ž .flow Hofstede, 1980, 1983, 1994 . The basis of thecultural hypothesis suggests that the basic manage-rial style and cultural traits within countries underlieall decisions. These cultural traits are suggested toeither impede or help managers in the ‘quality’ oftheir decisions. When comparing countries on com-petitive factors, it does not seem logical to ignorethese cultural traits, since they are latent factors thatunderlie decision-making. In this study, the develop-ment of the forecast is investigated in seven devel-oped countries. The study has two different levels of

Ž .understanding: 1 the surface level of forecastingŽ .practices associated with forecast accuracy, and 2

the underlying level of cultural traits that are associ-ated with the use of these practices. In short, culturaltraits cause some forecasting practices to be moreprevalent in some countries, and these forecastingpractices affect forecast accuracy.

When comparing managerial practices and strate-gies among countries, there is an issue of howdecisions are affected by the countries’ underlyingcultures. Additionally, there is the related issue ofthe complicated relationship between country cultureand corporate culture. This issue of how each coun-try’s culture affects the corporate culture is beyond

Žthe scope of this study see Deresky, 1997; pp.65–67 and Alkhafaji, 1995; pp. 57–82 for complete

.discussions . Although this study cannot fully ad-dress the country, corporate and practice issues, itgives a basic insight as to the degree to which eachcountry’s cultural traits affect forecasting decisions.

The development of the forecast for firms iscritical for organizational success, since the forecastis the beginning point of all planning which, in turn,determines production planning, inventories, and re-source requirements. This study analyzes two re-

Ž .search questions: 1 Which forecast practices affectŽ .forecast accuracy? And 2 what aspects of culture

cause the differences in the forecast practices? In a

more general perspective, the effect of cultural vari-ables on forecast practices can be viewed as oneexample of how the decision-making processes areaffected by culture. From that perspective, this studyshows how cultural traits affect the forecast proce-dure due to differences in cultural values amongcountries.

This study is organized in the following manner.First, the study covers the importance of the forecastin all organizations. Next, it reviews Hofstede’s fourcultural traits to determine how they affect the deci-sion-making process. The study then suggests theresearch hypotheses and presents the sample alongwith the statistical methodology. Finally, it reportsthe results along with the implications and givesdirections for future research.

1.1. The importance of forecast accuracy in deci-sion-making in manufacturing organizations

Forecast accuracy is important for manufacturingsince forecast error causes stockouts or costly changesin the master schedule. To prevent stockouts, firmsthat experience forecast errors increase their invento-ries in order to deliver as promised. Additionally,forecast inaccuracy causes confusion in manufactur-ing facilities, and managerial frustration with fore-cast procedures. This seriousness of forecast error onthe manufacturing system has been understood sincethe earliest writings on production planning and con-

Ž .trol Holt et al., 1955 . More recent studies havefound that forecast inaccuracy has severe cost im-

Žpacts on manufacturing systems Biggs and Cam-pion, 1982; Vollman et al., 1992; Ritzman and King,

.1993 . The basic conclusion from these studies isthat forecast inaccuracy causes major reschedulingand cost difficulties for manufacturing. The dissatis-faction with forecast accuracy has caused some con-

Žsultants to recommend scrapping it entirely God-.dard, 1989 .

Most firms attempt to improve forecast accuracyŽ .by focusing efforts in two basic directions: 1 quan-

Ž .titative methods and 2 forecasting procedures. Ofthese two directions, the most heavily researched is

Žthe effect of quantitative methods on forecast seeArmstrong, 1984; Wheelwright and Makridakis,

.1980 . Since the majority of the organizational re-

Page 3: Forecasting accuracy: comparing the relative effectiveness of practices between seven developed countries

( )J.G. Wacker, L.G. SpraguerJournal of Operations Management 16 1998 271–290 273

sources are under the control of manufacturing man-agers, the most important user of the forecast is themanufacturing function. Forecast accuracy is, there-fore, an important issue for organizational long-termsuccess.

Although the forecast is important to the successof the organization, the literature is not rich withstudies comparing the effectiveness of judgmentalfactors with respect to quantitative factors. In a 1995

Ž .study, Wacker and Sprague 1995 , studying thepractices of UK manufacturing managers, found thatinstitutional practices dramatically affected forecastaccuracy. This study extends that analysis to includehow the underlying cultural dimensions affect thepurposes, the forecast use and the forecast accuracy.

1.2. The importance of cultural Õalues in decision-making

The purpose of this section is to review thedifferent dimensions of cultural values. To under-stand the exact nature of the culture, a definition andan explanation of how it affects organizational deci-sions is necessary. To begin this discussion, it isnecessary first to define culture. As defined by Web-ster, culture is:

The integrated pattern of human behavior that in-cludes thought, speech, action, and artifacts and de-pends upon man’s capacity for learning and transmit-ting knowledge to succeeding generations.

There are several portions of this definition thatare important for this study. First, the importantaspect of thought, speech, action and artifacts seemto have high relevance to decision-making in thefirm, since the decisions need thought, speech andaction. Second, these decisions depend upon thelearning and transmitting of knowledge to succeed-ing generations. Note that the decision-making as-pects are affected by the intergenerational transfer-ence of knowledge. In short, culture and culturalvalues cannot be easily changed between genera-tions, since by definition, culture refers to the inter-generational transference of certain traits. Althoughthere is some indication that cultural traits amongcountries are gradually moving toward each other,the most prudent approach is not to violate the

formal definition by assuming that cultural traitsŽhave not changed in the last generation Hofstede,

. Ž .1994 . Hofstede 1980, 1983, 1994 has identifiedfour dimensions that should be considered whencomparing cultural differences among countries:

Ž .power–distance, individualism collectivism , uncer-tainty avoidance, and masculinityrfemininity. Thesedifferences are critical aspects for determining howfirms in different cultures behave when developingforecasts.

Power–distance can be defined as ‘‘the degree ofinequality among people which the population of a

Ž .country considers normal’’ Hofstede, 1994; p. 5 .Ž .Hofstede 1983 stated that this inequality is derived

from differences in prestige, wealth and power.Uncertainty avoidance is defined as ‘‘the degree

to which people in a country prefer structured overŽ .unstructured situations’’ Hofstede, 1994; p. 6 . Hof-

Ž .stede 1983 stated that these differences are derivedfrom rule-orientation, employment stability, andstress. Uncertainty avoidance should affect the basicrole in decision-making, since managers in highuncertainty-avoidance should attempt to protect their

Žemployees from changes Hofstede, 1980, 1983,.1994 .

Individualism can be defined as ‘‘ . . . the degreeto which people in a country prefer to act as individ-

Žuals rather than as members in a group’’ Hofstede,.1994; p. 6 . The opposite of individualism is collec-

tivism, which represents the degree to which mem-bers prefer to make decisions as a group. In some

Ž . Ž .societies, Hofstede 1983 p. 148 calls this individ-ualism either a curse, and in others, a blessing. Inshort, high individualistic countries should have lessparticipation in their firms’ decision-making.

The fourth dimension, masculinityrfemininity,represents ‘‘the degree to which values like as-sertiveness, performance, success and competitionŽwhich in nearly all societies are associated with

.men , prevail over values like the quality of life,maintaining warm personal relationships, service,

Žcare for the weak . . . which are more closely associ-. Žated with women ’’ Hofstede, 1994, p. 6; Hofstede,

.1983, p. 176 . In this study, dimension is interpretedas the need for performance, success and competi-tion. For simplicity’s sake, the masculinity trait canbe interpreted as the assertiveness needed for perfor-mance to achieve competitive success as opposed to

Page 4: Forecasting accuracy: comparing the relative effectiveness of practices between seven developed countries

( )J.G. Wacker, L.G. SpraguerJournal of Operations Management 16 1998 271–290274

the femininity trait that represents nurturing interper-Žsonal relationships and rendering services Hofstede,

.1983 . From an organizational perspective, competi-tive success is performing to achieve a competitiveadvantage over competing firms. In the context oforganizational forecasting, competitive success is us-ing those forecasting procedures that provide a fore-casting advantage over competing firms. At the otherextreme, a high-femininity value indicates the needfor resolving conflicts and nurturing.

There is another dimension to these cultural influ-ences on forecasting policies and procedures. It con-cerns the effectiveness of forecasting procedures inmultinational corporations. The question is simple:Will a parent company be able to set an effectivepolicy or effective procedure that will work in allsubsidiaries? More importantly, which policies and

Žprocedures will meet the least resistance. For exam-ple, since the use of the computer improves forecast

Ž .accuracy Wacker and Sprague, 1995 , is it a goodidea to encourage subsidiaries to use the computer

.when developing forecasts?Table 1 gives an overview of the similarities and

differences among countries on these four dimen-sions of culture. These data are used below whenestimating the effects of culture on forecasting’sexplanatory variables. These data are from the origi-nal Hofstede study, which is the most comprehensiveand accepted study in the area of differences in

Ž .international cultures Deresky, 1994; Faheti, 1996 .

1.3. The procedures for deÕeloping the forecast andforecast accuracy

Fig. 1 shows the basic procedures for forecastdevelopment and the resulting forecast accuracy. The

Fig. 1. The basic model for forecast accuracy tested in this study.

basic model depicts the underlying cultural factors ofpower–distance, uncertainty avoidance, individual-ismrcollectivism, and masculinityrfemininity af-

Ž .fecting five forecasting practices: 1 use of com-Ž .puter, 2 who is involved in developing the forecast,

Ž . Ž .3 the relative importance of the forecast uses, 4Ž .the forecast method, and 5 the subjective factors. In

turn, these practices affect forecast accuracy.

1.4. Factors that directly affect the forecast

1.4.1. Computer useOrganizations that have widespread computer us-

Žage improve forecast accuracy Mentzler and Cox,.1984; Dalrymple, 1987 . In these studies, the under-

lying reason for computer usage is the existence ofan organizational need for computerization. It isargued here that the use of the computer for forecast-ing does not necessarily ensure that forecast accu-racy will be improved. Consequently, the degree ofuse for all functions is a better measure of the

Table 1The differences between Germany, Japan, Mexico, New Zealand, Spain, Sweden and the USA on Hofstede’s dimensions of culture

Germany Japan Mexico New Zealand Spain Sweden USA

Ž . Ž . Ž . Ž . Ž . Ž . Ž .Power–distance 35 5 54 3 81 1 22 6 57 2 31 7 40 4Ž . Ž . Ž . Ž . Ž . Ž . Ž .Uncertainty avoidance 65 5 92 1 82 3 49 6 86 2 29 8 46 7Ž . Ž . Ž . Ž . Ž . Ž . Ž .Individualismrcollectivism 67 4 46 6 30 7 79 2 51 5 71 3 91 1Ž . Ž . Ž . Ž . Ž . Ž . Ž .Masculinityrfemininity 66 3 95 1 69 2 58 5 42 6 5 7 62 4

The numbers in parentheses are the between-country ranks.

Page 5: Forecasting accuracy: comparing the relative effectiveness of practices between seven developed countries

( )J.G. Wacker, L.G. SpraguerJournal of Operations Management 16 1998 271–290 275

intensity and integration of computer usage. Othershave implied that using modern technology for fore-casting improves forecast accuracy. For example, ina sample of British manufacturing managers, Wacker

Ž .and Sprague 1995 found that the average age ofequipment significantly affected forecast accuracy.The older the equipment, the less accurate the fore-cast. This study the extent of computer usage forsales planning, production planning, productionscheduling, inventory management, purchasing, andproduct design is the measure of computer usage.Therefore, a priori there is reason to believe thatcomputer usage improves forecast accuracy.

Consequently, the hypotheses are as follows.

Ho 1: The extent of computer usage throughoutthe organization has no effect on forecast error.

Ha 1: Greater computer usage throughout theorganization lowers the forecast error.

1.4.2. Forecast deÕelopmentŽOne frequent claim is that top management presi-

.dentrmanaging director should be directly involvedin the forecast because forecasting requires theircommitment. This reason seems somewhat perfunc-tory. While top management may be responsible forthe forecast, the real question is whether or not topmanagement is able to add significant knowledge toimprove the forecast accuracy. Hence, if the presi-dent is involved in forecast development, the forecastaccuracy may improve or may deteriorate dependingupon the president’s forecasting capabilities. Also,presidents of corporations have a vested interest inthe forecast that may cause a bias toward a higherforecast since higher sales generally mean bettermanagerial performance. This bias means it is possi-ble that a president’s involvement increases the like-lihood that the forecast will be a ‘hopes and wishes’forecast rather than an objective forecast. This fore-cast type is exactly what Plossl has cautioned against,since the forecast should not be a ‘hopes and wishes’

Ž .prediction Plossl, 19 . This hypothesis is supportedin the study by Wacker and Sprague, 1995, wherethey found that top management involvement de-creased forecast accuracy. Therefore, top manage-ment involvement could decrease forecast accuracy.

An additional consideration on forecast develop-ment is the involvement of the salesrmarketingfunction. When the salesrmarketing function devel-ops the forecast, the forecast should be more accu-rate, since the salesrmarketing function is closest tothe customer and has the most relevant informationto include in the forecast. The hypotheses are:

Ho 2: Degree of involvement of the top manage-Ž .ment presidentrmanaging director in forecasting

does not affect forecast accuracy.

Ha 2: Degree of involvement of the top manage-Ž .ment presidentrmanaging director in forecasting

decreases forecast accuracy.

Ho 3: The salesrmarketing function being pri-marily responsible for the forecast, has no effect onthe forecast accuracy.

Ha 3: Salesrmarketing taking primary responsi-bility for forecast development improves the forecastaccuracy.

1.4.3. Forecasting useIn 1987, Dalrymple found the primary purpose of

the forecast for market planning as opposed to pro-duction scheduling. Generally, the short-term fore-casts are used for production scheduling, and long-

Žterm forecasts are used for market planning Voll-.man et al., 1992 . The short-term forecast uses are

those that affect the daily operations of productionplanning, materialrinventory management, and hu-man resource planning, all of which are used toderive the budget. Long-term uses for market plan-ning are: equipment planning, facilities planning, andnew product development. Previous research hasshown that the most frequent uses of the forecast are,in order of importance, production planning, budget-

Ž .ing, and sales planning Mentzler and Cox, 1984 .These uses are primarily involved in short-term plan-ning. Forecasting for short-term uses may affect itsaccuracy because the forecast horizon is shorter.

Ho 4: Short-term use of the forecast does notaffect its accuracy.

Page 6: Forecasting accuracy: comparing the relative effectiveness of practices between seven developed countries

( )J.G. Wacker, L.G. SpraguerJournal of Operations Management 16 1998 271–290276

Ha 4: Short-term use of the forecast increases itsaccuracy.

1.4.4. Forecasting quantitatiÕe methodsThere is an abundance of forecasting articles sur-

veying the relative accuracy of forecasting methodsusing forecasting managers as the subjectsŽ .Armstrong, 1984 . The general results of these stud-ies conclude that multiple techniques are superior to

Žsingle techniques Armstrong, 1984; Makridakis and.Winkler, 1983 . The general consensus is that so-

phisticated approaches have only modest payoffsŽ .Armstrong, 1984 .

Additionally, there is a higher satisfaction withsimpler quantitative methods than with subjective orcomplex mathematical models. For example, Ment-

Ž .zler and Cox 1984 surveyed more than 150 man-agers in the US and concluded that their respondentswere more familiar with subjective than objectiveŽ .quantitative methods. They also found that therewas general satisfaction with simple methods such asjury of opinions, regression, exponential smoothing,moving averages, trend line analysis, classical de-composition and simulation. There was less satisfac-tion with customer expectations, sales force compos-ites, life cycle analyses, straight-line projections andBox–Jenkins time series. These results explain whythere is a wider use of subjective and simpler meth-ods than the objective and more complex methods.

Ž .Similarly, Wacker and Sprague 1995 found thatBritish manufacturing managers who used quantita-tive techniques did not improve the forecast. Thisstudy investigates the degree to which companies use

Žthree basic techniques: times series models e.g.,. Ž .moving average , causal models e.g., regression ,

Ž .and qualitative models e.g., Delphi . The literaturesuggests that quantitative techniques improve fore-

Žcast accuracy Hofstede, 1994; Makridakis and Win-.kler, 1983 . Therefore, the following hypotheses are

formulated.

Ho 5: There is no difference between organiza-tions that do and those that do not use quantitativemethods for the development of forecasts.

Ha 5: Organizations that use quantitative methodsŽfor forecasting have more accurate forecasts Dawes,

1976; Goddard, 1989; Libby, 1976; Parker and Se-.gura, 1971; Sawyer, 1966; Tversky, 1973 .

1.4.5. SubjectiÕe factorsOne important conclusion recognized in the litera-

ture is that subjective and judgmental factors affectŽthe derivation of the final forecast for a literature

.review, see Wheelwright and Makridakis, 1980 . AsŽ .suggested by Mentzler and Cox 1984 , there gener-

ally is less satisfaction with the subjective forecaststhan with simple quantitative techniques. Therefore,the inclusion of subjective factors should reduceforecast accuracy. Unfortunately, the literature issilent on which factors are subjectively included inthe development of the forecast. Therefore, the litera-ture provides no specific variables to include inforecast development. Due to the lack of suggestedvariables from the literature, likely candidate vari-ables are the economic and political climate, com-pany and industry conditions, and additional infor-mation from suppliers and customers. Since thesefactors are judgmental, the literature suggests theyshould decrease forecast accuracy. Therefore, thefollowing hypotheses are presented.

Ho 6: The use of subjective factors has no impactŽon forecast accuracy Wheelwright and Makridakis,

.1980 .

Ha 6: The use of subjective factors decreasesforecast accuracy.

Organizations generally react to the forecast byŽ .modifying their forecasts. Plossl 1973 has stated

that firms may increase their forecast inaccuracy bychanging their forecast. This modification shouldincrease as the forecast accuracy increases. Wacker

Ž .and Sprague 1995 found this factor to be veryimportant for determining forecast inaccuracy. Thisstudy hypothesizes that the more frequently a firmchanges the forecast, the higher the forecast errorwill be.

Ho 7: The annual number of forecast modifica-tions has no effect on forecast accuracy.

Ha 7: The use of forecast accuracy is positivelyrelated to the annual number of forecast modifica-tions.

Page 7: Forecasting accuracy: comparing the relative effectiveness of practices between seven developed countries

( )J.G. Wacker, L.G. SpraguerJournal of Operations Management 16 1998 271–290 277

1.5. How cultural Õalues affect forecast practices

Cultural factors are key underlying variables fordetermining the degree of emphasis on differentforecasting practices such as computer usage, fore-cast development, quantitative techniques, forecastusages, subjective information uses, and forecast

Ž .modification Hofstede, 1994 . This section logicallydevelops hypotheses with respect to each of Hofst-ede’s cultural values and how they may affect theexplanatory variables in the above model. Since priorstudies do not suggest specific hypotheses, the hy-potheses must be logically developed for their link-age to the forecast. Table 2 provides an overview ofthe cultural hypotheses investigated in this study.

1.5.1. Power–distanceManagement in high power–distance countries

accepts the responsibility for making decisions andhas a less participative management style. Sincemanagers in these countries must obtain informationfrom sources other than workers, these managersshould tend to use computers and more quantitativemethods than managers in low power–distance coun-tries. Therefore, the alternative hypotheses for com-puter uses and quantitative techniques are expected

Žto be positively related to high power–distance in.short, Ha ) 0 . Additionally, a priori, in high

power–distance countries, presidents should be moreactively involved than presidents in low power–dis-tance because they consult less with their subordi-nates. This managerial control should mean that thepresident is more involved in the forecast develop-

Ž .ment Ha)0 .Managers in high power–distance countries re-

main distant from the workers. It seems logical toassume that because of this distance, managers donot become closely involved with day-to-day short-term decisions, and are more closely involved withforecast for the long-term survival of the firm.Therefore, firms in high power–distance countriesare expected to use the forecast for long-term pur-poses. Therefore, high power–distance is associated

Žwith using the forecast for long term-purposes Ha.)0 . On the other hand, it seems that managers in

high power–distance countries would de-emphasizethe use of the forecast for short-term uses. Conse-quently, a priori, compared to low power–distancecountry firms, high power–distance firms are notexpected to use the forecast for short-term purposesŽ .Ho-0 .

Table 2The cultural hypotheses suggested from the discussion

Power–distance Uncertainty avoidance Individualismrcollectivism Masculinityrfemininity

TechnologyComputer uses Ha)0 Ha-0 Ha)0 Ha-0

Management controlPresident Ha)0 Ha-0 Ha-0 Ha-0Salesrmarketing Ha)0 Ha-0 Ha)0 Ha-0

Forecast usesShort-term Has0 Ha-0 Ha-0 Ha-0Long-term Ha-0 Ha-0 Ha)0 Ha)0

QuantitatiÕe techniquesTime series models Ha)0 Ha-0 Ha)0 Ha)0Causal models Ha)0 Ha-0 Ha)0 Ha)0Qualitative models Ha)0 Ha-0 Ha)0 Ha-0

SubjectiÕe factorsSubjective external Ha)0 Ha-0 Ha-0 Ha)0Subjective internal Ha-0 Ha-0 Ha-0 Ha-0Forecast modifications Ha)0 Ha-0 Ha)0 Ha)0

Page 8: Forecasting accuracy: comparing the relative effectiveness of practices between seven developed countries

( )J.G. Wacker, L.G. SpraguerJournal of Operations Management 16 1998 271–290278

Managers in high power–distance countries areexpected to need objective information to facilitatethe forecast. To these firms, the external sources ofinformation becomes important. Therefore, theyshould use subjective external information more than

Ž .low power–distance firms Ha)0 . Alternatively,since managers distance themselves from subordi-nates, they should rely less on internal subjective

Ž .factors than low power–distance countries Ha-0 .Finally, since managers in high power–distancecountries consult less with their employees, theyshould not be hesitant to modify their forecast morethan firms in low power–distance countries. They,therefore, should modify the forecast more than low

Ž .power–distance managers Ha)0 . See Table 2,Ž .column two: power–distance for a list of thesehypotheses.

1.5.2. Uncertainty aÕoidanceManagement in high uncertainty avoidance coun-

tries attempt to reduce uncertainty for their workersby shielding them from the elements of uncertainty.Because managers do not plan on readily reacting tochanges in demand, resources tied to the forecastshould not be as great as in low uncertainty avoid-ance countries. Therefore, computer uses, quantita-tive methods, and forecast uses for planning should

Ž .be less in these firms Ha-0 . Additionally, sincethe forecast carries less importance in these firms,they should have less management andsalesrmarketing involvement than in low uncertainty

Ž .avoidance countries Ha-0 .Since these firms are high uncertainty avoidance

oriented, managers should be less reactive to subjec-tive information. Therefore, they should not changetheir forecast for any external or internal changes in

Ž .subjective information Ha-0 . Due to uncertaintyavoidance, these managers would not be prone tomake changes because it would mean changing re-source levels. This reasoning leads to the conclusionthat these firms modify their forecasts less often thanfirms in low uncertainty avoidance country firmsŽ . ŽHa-0 . See Table 2 column three, uncertainty

.avoidance for a list of these hypotheses.

1.5.3. IndiÕidualismrcollectiÕismManagers in high individualism countries rely on

their own judgment more than managers in high

collectivism countries, since they tend to incorporateless group information into their forecast. However,it seems likely that they would value objective infor-mation more than managers of firms in high collec-tivism countries. Therefore, managers in high indi-vidualistic countries would value computer resultsand quantitative techniques more than in high collec-tivism countries. Therefore, high individualism isexpected to be positively related to computer usage

Ž .and quantitative techniques Ha)0 . In high indi-vidualism countries, there are less collective deci-sions. This means that individual organizationalfunctions have responsibility for specific tasks. Putanother way, firms in individualistic countries relymore on their corporate structure than on collec-tivism activities to make decisions. Therefore, theseindividualistic firms are expected to rely more on

Ž .salesrmarketing Ha)0 and less on presidentialŽ .involvement in the forecast Ha-0 . Managers in

individualistic environments trying to direct day-to-day operations are expected to use the forecast for

Ž .short-term planning functions Ha)0 . However,for the long-term use of the forecast, there is nocompelling reason to believe that there would bedifferences between high individualistic and high

Ž .collectivist countries firms Hos0 .Since firms in high individualistic countries tend

to look for more objective information, they wouldŽ . Žrely less on the subjective factors Ha-0 alterna-

tively, firms in high collectivist countries tend to usemore subjective information for developing the fore-

.cast . Firms in high individualistic countries are ex-pected to modify their forecast frequently, since theydo not consult with employees regarding changes in

Ž .forecast and activity levels Ha)0 . See Table 2Ž .column three, labeled individualismrcollectivismfor a list of these hypotheses.

1.5.4. Masculinityr femininityThe high-masculinity trait is defined as the need

for managerial performance, competitiveness, andsuccess. The relationship between high-masculinityrfemininity and the forecast method de-pends upon the degree to which an accurate forecastprovides competitive success. Although having anaccurate forecast provides lower inventory, lowerconfusion and backlogs, the forecast effort may notprovide a competitive advantage over market rivals,

Page 9: Forecasting accuracy: comparing the relative effectiveness of practices between seven developed countries

( )J.G. Wacker, L.G. SpraguerJournal of Operations Management 16 1998 271–290 279

since the same forecasting techniques are availableto all rivals. Consequently, there is no compellingreason to believe sophisticated techniques will beused more or less extensively by firms in high-femininity countries Therefore, there should be nodifference between high-masculinity and high-femininity firms for quantitative forecasting methodsŽ .Hos0 .

On the other hand, high-masculinity firms mayview subjective information as a source of competi-tive advantage, since all firms may not have thesame information. Therefore, both internal and exter-nal subjective information should be positively re-

Ž .lated to high-masculinity firms Ha)0 . Also, sincefirms in high-masculinity countries tend to adapt theforecast for competitive advantage, they should mod-

Ž .ify their forecast more frequently Ha)0 . See TableŽ .2 last column labeled masculinityrfemininity for a

list of these hypotheses.

1.6. The sample

The survey was pre-tested with personal inter-views at Cranfield University. The data used in thisstudy are collected by the Global ManufacturingResearch Group in an extensive data-gathering effortthroughout many countries in the world. The ques-tionnaires were translated and back-translated in allrelevant languages. The United States data weregathered by the Manchester Manufacturing Manage-

Žment Center Sprague and Naji, 1992; Whybark and.Vastag, 1994 , and all other data were gathered by

academics in each country. The seven countries areGermany, Japan, Mexico, New Zealand, Spain, Swe-den and the USA. Using the Ronen and ShenkarŽ .1985 classification of the countries, only the UnitedStates and New Zealand are clustered in the sameattitudinal dimensions. Therefore, the cultural effectson the procedural variables should be fairly dissimi-lar. The sample sizes from each country are pre-sented in Appendix A, Table A.1.

Another important aspect is the representativenessof each country’s sample. This study argues thatmanufacturing firms that respond to the survey aregenerally better-run manufacturing facilities sincethey have the key data on forecasting and formalprocedures. Therefore, they are ‘better’ manufactur-ers than the general manufacturing population. By

being better manufacturers, the sample represents the‘best practices’ and is an indication of the directionthat the general manufacturing population will fol-low in the future. Therefore, the survey’s results aremore important than a current practices survey of thegeneral manufacturing population.

1.7. The Õariables

The dependent variable is the forecast error per-cent that represents forecast inaccuracy. Thesalesrmarketing and president involvement variablesare both binary variables with one representing presi-dential involvement and the other representingsalesrmarketing primary responsibility.

1.8. The created Õariables

The created variables are used to give a morereliable measure of the theoretical concept. In thisstudy, the following variables are created to consoli-date underlying concepts into basic areas. Each vari-able is created by summing original individual vari-ables. Each individual variable is evaluated on afive-point Likert scale, with 1 indicating ‘not used atall’ and 5 indicating ‘used to a great extent’.

The first step is to identify the underlying conceptthat is important for the research. This is describedas content validity. The next step is to test theconstruct validity which, in this study, is a two-stepprocedure. First, the data are analyzed using factoranalysis to identify the number of significant con-cepts that have eigenvalues above 0.8. The scree plotof eigenvalues is used to determine whether otherfactors are relevant. Second, if there is only onefactor, the Cronbach a is estimated. Good constructs

Žshould have an a of greater than 0.70 Nunnally,.1967 .

It could be argued that only the use of the com-puter for forecasting is a single variable. However,the use of computer for forecasting only would notindicate if the forecast is integrated into other organi-zational functions. Therefore, the use of computersfor forecasting is only a partial measure of the effectof computer usage on forecast accuracy, since otherfunctions may constrain the forecast by restrictingfinances, marketing goals, and manufacturing. With-

Page 10: Forecasting accuracy: comparing the relative effectiveness of practices between seven developed countries

( )J.G. Wacker, L.G. SpraguerJournal of Operations Management 16 1998 271–290280

out incorporating the functional linkages, the forecastwould not be integrated into the business plan.

In this study, the total computer usage of allfunctional areas is called technology orientation, andis used to represent the degree of computer integra-tion of decisions in the organization. Without thistotal effect, the forecast could be isolated from otherfunctions. The total effect of the technology orienta-tion on the forecast should include all these com-puter uses. The other uses are production planningand scheduling, inventory management, purchasingand product design. It could be argued that computerusage for product design is only remotely related tothe forecast. However, because the organization’s

Žtechnology orientation contains product design as in.bills of material , it would seem logical to include it

Žhowever, excluding it did not change any of the.statistical results . Therefore, a construct to identify

the overall orientation of the firm in commitment totechnology was created. Technology orientation’sCronbach a is 0.8148. Factor analyses using vari-max rotation indicated that the variables loaded onlyon one factor. The scree plots of the eigenvaluesindicated that one factor had an eigenvalue of overthree and dropped to less than 0.7 for the secondfactor.

The second created variable is the use of theforecast for long-term planning. Theoretically, long-term planning involves sales planning and budgetingfor new product development, equipment planning,and facility planning for the long-term survival ofthe firm. Consequently, the variables combined toderive this construct are budget preparation, salesplanning, equipment planning, facility planning, andnew product development. The long-term use of theforecast had a Cronbach a of 0.8550. Factor analy-ses of long-term uses indicated that the variablesloaded on one factor only, and the scree plots of theeigenvalues indicated that the factor had an eigen-value of over two and one half and dropped to lessthan 0.75 for the second factor.

Another forecast use variable is created to indi-cate the short-term uses of the forecast. Theoreti-cally, short-term uses are operational—those usesthat run the organization on a day-to-day basis.These short-term uses are production planning, sub-contracting planning, materialrinventory planning,and human resource planning. Each plan is aimed at

achieving organizational short-term goals. Theshort-term use of the forecast had a Cronbach a of0.8305. Factor analyses of short-term uses indicatedthat the variables loaded on one factor, and the screeplots of the eigenvalues indicated that the factor hadan eigenvalue of over three and dropped to less than0.65 for the second factor.

Ž .There are two types of subjective factors: 1Žthose that are outside of the firm here called subjec-

. Ž .tive external factors and 2 those inside of the firmŽ .here called subjective internal factors . Theoreti-cally, subjective external factors are those factorsoutside of the firm that could affect the firm’s activ-ity. These subjective external factors are currenteconomic conditions, current political conditions, andgeneral industry conditions. The subjective internalfactors are information that comes from inside theorganization: top management, marketing, purchas-ing, marketing and production. The subjective inter-nal factors are the company’s general situation, cus-tomer information, market research results, supplierinformation and current order backlog. For bothinternal and external subjective factors, the Cronbacha s were 0.8031 and 0.7584, respectively. Addition-ally, the factor loadings indicated that there is onlyone factor for each constructed variable, since thescree plots indicated that other factors are all lessthan 0.8.

In all cases, the above variables met the contentŽvalidity and criterion validity requirements see

.Bollen, 1989; pp. 186–225 .

1.9. The cultural Õariables

This study uses the values derived in Hofstede’soriginal study to represent the four cultural factors:power–distance, uncertainty avoidance, individual-ismrcollectivism, and masculinityrfemininity traits.There are some rather obvious reasons to use otherdata for the international cultural traits. Among them

Ž . Ž .are: 1 each company has its own culture, 2 cur-Ž .rent data are more relevant than older data, and 3

there are numerous other cultural traits that are alsoimportant. First, since Hofstede’s study is the mostcomprehensive, his data are gathered within a singlecompany; it controls for between company and coun-try differences. Therefore, it best represents the inter-national cultural traits. Also, if the data were gath-

Page 11: Forecasting accuracy: comparing the relative effectiveness of practices between seven developed countries

( )J.G. Wacker, L.G. SpraguerJournal of Operations Management 16 1998 271–290 281

ered for the individual company, it would havesample bias, since only a limited sample would beavailable from an individual company. Additionally,it would be difficult to distinguish company culturalfrom country cultural traits, since there would bebetween company differences even within a country.All in all, the criticism of using cultural traits datafrom each company presents more theoretical diffi-culties than it solves.

Secondly, although a current study with a largesample within a multi-international company for eachcountry would seem ideal, it should not make anydifference in the data. By definition, culture meansan intergenerational transference of ideas and norms.Therefore, using Hofstede’s older data should notmake significant differences in the results. Conse-quently, since the Hofstede results are from the mostcomprehensive statistical sample on cultural traits, itseems to make more sense to use them over newlyderived estimates.

The third criticism is that other cultural traits areimportant for decision-making in forecasting. Al-though other cultural traits are considered importantfor decision-making, by far the most accepted traits

Ž .are the four traits of Hofstede 1983, 1994 . Sincethis study is exploratory in determining if culturetraits affects the forecast decision-making, addingadditional variables may confound the results. There-fore, this study will encourage future researchers toinvestigate the effects of other cultural traits.

The cultural variables all have the same valuewithin each country. Within each country, it is highlylikely that each firm will have slightly differentnumerical values for each cultural variable. In thisstudy, all firms within the country carry the samevalue. It is argued here that these cultural values arebetter approximations of cultural values than binary

Ž .qualitative variables for three reasons. 1 Thesevalues facilitate an understanding of the four differ-ent cultural values hypothesized to affect managerialpractices, since they provide more detail and offer

Ž .more accepted results. 2 There are numerical dif-ferences between countries, so they facilitate statisti-cal comparisons of the relative degree to which eachcultural value affects each variable. And third, itcould be argued that these cultural values are aver-ages across whole countries, so they are more stablethan estimates from the individual firms. In that case,

they are representative of the country. Naturally,using the averages across entire countries decreasesthe variation and may decrease statistical signifi-

Ž .cance Kerlinger and Pedhazur, 1974 .

2. Statistical methodology

The statistical methodology used in this paper isdesigned to investigate the issue of which forecastpractices lead to lower forecast error. The researchprocedures used here are as follows.

Ž .1 The statistical analyses begin with an estimateof the demographic factors for each country to deter-mine if the underlying sample is biased due todemographic factors. First, this comparison is madewith simple ANOVA to determine if there are differ-ences between countries. If there are differences,then it is necessary to determine if these differencesbias the other estimates. For this estimate, a compari-son is made using ANOVA with the dependentvariable of percent error using country as the majorfactor and the demographic characteristics as covari-ates. The demographic factors chosen are: the num-ber of factory employees and the number of total

Žemployees to determine firm size differences be-.tween country , the made-to-stock and made-to-order

Žpercentages to control for the way the firm faces the.competitive market , the number of product lines,

Žpercent of sales for the bestselling product to deter-.mine how it is competing in the market , job shop

layout percent against assembly line layout percentŽto control for how the physical layout affects the

.forecast .Ž .2 If there are no statistical differences demo-

graphically between the countries using the ANOVAto control for covariates, then the overall estimatescan be performed. However, if there are significantdifferences, then estimates of the variables effect onforecast error must be included in the estimate. Next,the model is estimated using the hypothesized vari-ables that affect the forecast error. The most tradi-tional method is to use ANOVA, with country as themain factor and the hypothesized variables as covari-ates. It is hoped that the country’s main effects arenot significant, since this would indicate missingvariables in the estimates.

Ž .3 Next, an overall regression of all forecastingvariables on forecast error is estimated with the

Page 12: Forecasting accuracy: comparing the relative effectiveness of practices between seven developed countries

( )J.G. Wacker, L.G. SpraguerJournal of Operations Management 16 1998 271–290282

practices as explanatory variable. This estimate isperformed to indicate the direction of the variables’effect on forecast error.

Ž . Ž4 Last, the cultural factors of Hofstede 1980,.1983, 1994 are regressed on each practice to deter-

mine the degree to which culture affects the forecastdecision-making process.

This procedure is an exploratory data analysisprocedure to further develop an understanding of theunderlying decision-making process for forecast de-velopment. Each step is designed to reduce theamount of confounding factors so that the underlyingcauses can best be investigated.

Before going into the statistical results, a fewpoints must be made concerning the statistical limita-tions of this study. There are two traditional methods

Ž .that can be used for missing data: 1 elimination ofŽ .the entire observation, or 2 substitution of the

means for all missing variables. In this study, theestimates are performed with the elimination of thecase if even one variable was missing. Several coun-tries have small samples, so a backup estimate isperformed with all missing variables, using theirmeans to see if the missing observations statisticallybias the estimates.

2.1. Statistical tests for the forecasting accuracyhypotheses

The statistical procedures suggested above willfirst use simple factorial ANOVA to test the overallsignificance of the variables and control for interact-ing variables. Next, to determine the sign of thevariables, multiple regression is used in order toisolate the direct effects of collinearity. This multipleregression is a conservative estimation procedure,since it assumes a linearity of the variables. Theinterpretation of the results is straightforward, sinceall the variables are estimated considering the values

Žof the other variables Kerlinger and Pedhazur, 1974;.Pindyck and Rubenfeld, 1976 .

3. Results

This section will discuss the results from theestimates. First, it will determine the statistical sig-nificance of the demographic differences between

countries. If differences are statistically significant,then the country is used as a main factor in theANOVA statistical model. Next, the study deter-mines the overall effect of the forecasting practiceson forecast accuracy. After this estimate of the over-all effect, it gives the estimates from the individualcountries for a simple comparison of the betweencountry differences in forecasting practices. And fi-nally, the study estimates the cultural value effectson the forecasting practices.

The data are gathered primarily in textiles andsmall machine tool industries. The following arecritical variables that may affect the differences inthe types of markets the firms must forecast for

Ž .manufacturing: made-to-stock percentage MTS ,Ž .made-to-order percentage MTO , percent of manu-

Ž .facturing facility laid out as a job shop Job Shop ,percentage of manufacturing facility laid out as an

Ž .assembly line Assembly Line , the number of em-Ž .ployees Employees , and the number of factoryŽ .workers Factory Workers .

The second column of Table 3 gives significantdifferences between countries for each of the vari-ables using a simple ANOVA test. These resultsindicate that there are significant differences in themeans between the countries for five of the eightvariables. The differences between each country and

Ž .the overall mean the country sample excluded aregiven in the table. Since there are differences in thedemographic means, a simple factorial ANOVA withforecast error as the dependent variable, with countryas the major factor and with covariates of the all theforecast practice variables revealed no between coun-

Žtry differences the overall significance is as92.3%and the significance of the between countries aloneis as84.3% much less than the as10% signifi-

.cance . The conclusion of these tests is: the manufac-turing facilities sampled between the countries seemvery similar in their forecast error behavior in rela-tion to their demographic environment variables.

3.1. Results from the oÕerall model

Table 4 presents the overall results of the multipleregression estimates. The second column, ‘signifi-cance between countries’ represents a simpleANOVA to test the overall differences betweencounties for each explanatory variable. Overall, these

Page 13: Forecasting accuracy: comparing the relative effectiveness of practices between seven developed countries

( )J.G. Wacker, L.G. SpraguerJournal of Operations Management 16 1998 271–290 283

Table 3Means of the competitive factors for each country

Anova significance Overall Germany Japan Mexico New Zealand Spain Sweden USAof F-test

bŽ .Made-to-stock % 0.009 25.67 30.29 22.04 21.21 31.17 38.82 29.75 25.44aŽ .Made-to-order % 0.012 72.75 69.71 77.96 71.10 68.83 67.26 70.25 74.56

b c bTotal employees 0.000 209.6 294.9 168.3 139.6 118.6 328.7 144.5 266.9a c bFactory employees 0.000 137.4 150.8 119.5 100.8 88.3 243.2 90.11 50.8

a bProduct lines 0.711 12.21 7.59 11.18 9.83 23.06 35.6 9.55 7.77b a b aLargest line percent 0.003 57.65 57.35 64.4 58.12 42.82 48.45 67.35 55.04

Job shop percent 0.214 59.85 49.1 62.35 61.48 69.44 57.88 67.25 55.92Assembly line percent 0.106 39.41 50.9 38.55 35.70 30.56 42 32.75 42.68

a90% confidence two-tail test.b95% confidence two-tail test.c99% confidence two-tail test.A Student’s t-test compared the individual country’s mean with the overall mean. The overall mean is adjusted so that the individualcountry’s data are not included in the overall mean. Also, the sample size used to calculate the overall standard error of the estimate isdecreased by the country’s sample size.

results indicate that except for short-term use offorecast, subjective external factors, time seriesmethods, and perhaps qualitative methods and modi-

fications, these countries have dramatic differencesin the way the forecast is developed, how it is used,and the forecasting methods. These results suggest

Table 4Overall combined estimates across all countries

Ž . Ž .Variable Significance between countries a Estimated coefficient Student’s t-test Significance a

TechnologyComputer usage 0.000 y0.3734 y2.46 0.0073)))

Top management controlPresident involvement 0.001 y1.2343 y0.72 0.2372Sales marketing involvement 0.000 y1.1827 y0.67 0.2505

Use of forecastUse of forecast for short-term purposes 0.699 0.4927 1.67 0.0478))

Use of forecast for long-term purposes 0.005 y0.0365 y0.142 0.4434

Use of quantitatiÕe methodsTime series methods used 0.295 y0.3700 y0.53 0.2969Causal models used 0.001 y0.2819 y0.30 0.3815Qualitative methods used 0.191 1.8907 2.45 0.0075)))

SubjectiÕe factorsSubjective external factors 0.485 0.4563 1.32 0.0946)

Subjective internal factors 0.046 y0.4459 y1.63 0.0525)

Number of modificationsNumber of annual modifications of forecast 0.123 0.2852 6.48 0.0000)))

Ž .Constant 15.5880 3.48 0.0003)))

R2 s0.17060 ns306 Fs5.149747 as0.0000

)Significant a-0.10 one-tail t-test.))Significant a-0.05 one-tail t-test.)))Significant a-0.01 one-tail t-test.

Page 14: Forecasting accuracy: comparing the relative effectiveness of practices between seven developed countries

( )J.G. Wacker, L.G. SpraguerJournal of Operations Management 16 1998 271–290284

that there are statistically significant between countrydifferences in forecasting practices.

To summarize briefly, the results and statisticalevidence for each forecasting practice hypotheses areas follows.

Hypothesis 1: The more extensively the firm usescomputers for forecasting, the lower the forecasterror. These tests support the hypothesis of techno-logical culture reducing forecast error.

Hypothesis 2: The degree of presidential involve-ment in the forecast has no effect on forecast accu-racy.

Hypothesis 3: SalesrMarketing function beingprimarily responsible for forecast development doesnot affect forecast accuracy.

Hypotheses 4: Unexpectedly, the extent of fore-cast used for short-term purposes does not improvethe forecast accuracy. In fact, statistically, short-termuse could be hypothesized to decrease forecast accu-racy and long-term use has no effect on forecastaccuracy.

Hypotheses 5: The use of time series and causalmethods does not improve forecast accuracy. The

Ž .use of qualitative methods Delphi, etc. tends todecrease forecast accuracy.

Hypotheses 6: The extent of use of subjectiveexternal factors tends to cause the forecast to be lessforecast accurate. Unexpectedly, using subjective in-ternal factors is associated with more accurate fore-casts. Since the a priori hypothesis is that all subjec-tive factors decrease forecast accuracy, the relation-ship between forecast accuracy and subjective inter-nal factors is insignificant.

Hypothesis 7: The number of forecast modifica-tions is strongly positively related to forecast inaccu-racy.

The general conclusions and implications are quiteinteresting. First, there are some facts that extendacross different cultures. Firms that are technology-oriented using the computer for planning and con-trolling their operations, tend to have lower forecasterror. In short, the use of computers tends to improve

forecast accuracy. Second, the involvement of thepresident in the forecast does not appear to affectforecast accuracy. This result may be due to the typeof information that the president brings to the fore-cast development, since some presidents bring im-portant information to the forecast, while others mayuse the forecast to ‘paint a rosy picture’. Wacker and

Ž .Sprague 1995 found that forecast accuracy de-creases with presidential involvement. Both thisstudy’s result and the result of Wacker and SpragueŽ .1995 have similar implications, since both studiesconclude that presidential involvement does not im-prove forecast accuracy.

Assigning the forecast development responsibilityto the salesrmarketing function does not seem toimprove forecast accuracy, which is not consistentwith earlier findings that it tends to improve forecast

Ž .accuracy Wacker and Sprague, 1995 . This inconsis-tency may be due to the type of salesrmarketinginvolvement with forecast development in specificinternational countries. Consequently, future re-searchers should focus on the specific types of infor-mation that the salesrmarketing function uses todevelop the forecast.

Although the most important use of the forecast isŽfor short-term purposes production planning,

scheduling, subcontracting, inventory planning and.human resource planning , this use does not increase

forecast accuracy. In fact, the statistical results couldsuggest short-term uses may decrease forecast accu-racy. The reason may be that firms emphasizingshort-term uses could have unrealistic expectationsand may not adequately develop an objective fore-cast. The use of the forecast for long-term purposesŽbudgeting, new product development, facilities and

.equipment planning has no effect on forecast accu-racy. In short, the planned use of the forecast doesnot improve forecast accuracy.

As found in earlier studies, the use of quantitativetechniques does not appear to improve forecast accu-

Ž .racy Wacker and Sprague, 1995 . However, qualita-tive techniques, such as Delphi, tends to decreaseforecast accuracy. These results support earlier stud-ies that the forecast method does not improve the

Žforecast, but can decrease forecast accuracy as in.the case of qualitative methods .

Finally, forecast modifications have two very dif-ferent explanations. One, possibly firms modify by

Page 15: Forecasting accuracy: comparing the relative effectiveness of practices between seven developed countries

( )J.G. Wacker, L.G. SpraguerJournal of Operations Management 16 1998 271–290 285

reacting to forecast error. Alternatively, as suggestedby Plossl, it may be that modifying too frequently

Ž .causes forecast error Plossl, 1973 . The resolutionof this issue is an interesting area of investigation forfuture research.

3.1.1. Managerial implications from the oÕerallmodel

The managerial implications of the overall modelare quite straightforward. The use of computer tech-nology to integrate forecasting with other functionalareas is a very important factor for reducing forecasterror. This linkage also may exhibit itself in the useof internal subjective information. Although it wasnot statistically tested in this study, it is possible thatcomputer-integrated firms incorporate special inter-nal information from marketing and manufacturingto improve the forecast accuracy.

A second general managerial implication is thatthe use of qualitative methods, and subjective exter-nal factors, tends to decrease forecast accuracy. Nei-ther of these characteristics adds to the quality ofinformation needed for accurate forecasting. Also, aforecast that is used for short-term purposes maydecrease forecast accuracy. One explanation of thisresult is that the use of the forecast for short-term

Ž .decisions production planning, etc. may cause theforecast to be influenced by the internal capabilitiesrather than external demand, thus causing a decreasein forecast accuracy.

In summary, the statistical evidence suggests in-creasing the use of computer improves forecast accu-racy. One could offer the hypothesis that increasedcomputer usage helps monitor the subjective external

Ž .factors economic, political, and industry conditionsŽhowever, this hypothesis was not specifically tested

.in this study . On the other hand, qualitative forecast-ing techniques and the internal subjective factorsŽcompany conditions, customer and supplier informa-

.tion, and current order backlog tend to decreaseforecast accuracy.

3.2. Results from the cultural hypotheses

The statistical results for the cultural hypothesesŽare presented in Table 5 the numerical estimates are

.presented in Appendix A, Table A.3 .

Supported power–distance hypotheses. From col-umn two, firms in high power-distance countries thattend to use the computer, time series methods andsubjective external information are all supported.

Unsupported power–distance hypotheses. Thereis no difference in top management andsalesrmarketing involvement in the forecast betweenhigh and low power–distance country firms. Fore-cast uses are not related to power–distance. Also,there are no differences between high and low powerdifference countries in the use of causalrqualitativemodels and subjective internal factors. In addition,there is no difference in the number of times theforecast is modified in high power and low power–distance country firms.

Supported uncertainty aÕoidance hypotheses.From column three, the overall conclusion is thatuncertainty avoidance has limited effects on the fore-cast. High uncertainty avoidance negatively affectsthe subjective internal information used for forecastdevelopment.

Unsupported uncertainty aÕoidance hypotheses.This study could not find any evidence that uncer-tainty avoidance has any major effect on computeruse, quantitative methods, or management control.Additionally, uncertainty avoidance’s relationship tothe number of forecast modifications gave the wrongexpected sign, making the result insignificant.

Supported indiÕidualismrcollectiÕism hypothe-ses. From column four, computer usage, causal modelusage and number of forecast modifications are allstatistically significant with predicted signs. Thepresidents are less involved in the forecast develop-ment in individualistic country firms than in collec-tivistic country firms.

Unsupported indiÕidualism r collectiÕism hy-potheses. There are no statistical differences insalesrmarketing involvement in the forecast betweenfirms in high and low individualistic countries. Theuse of quantitative techniques is not significantly

Žrelated to individualismrcollectivism note that theuse of causal models gave the wrong sign and conse-

.quently was insignificant . Finally, there are no sig-nificant differences for firms in individualismrcol-lectivism countries in the use of the forecast, or inuse of the subjective factors.

Supported masculinity r femininity hypotheses.From the last column, presidential involvement in

Page 16: Forecasting accuracy: comparing the relative effectiveness of practices between seven developed countries

( )J.G. Wacker, L.G. SpraguerJournal of Operations Management 16 1998 271–290286

Table 5The statistical conclusions of cultural effects on explanatory variables

Power–distance Uncertainty avoidance Individualismrcollectivism Masculinityrfemininity

Technology: Ha)0 Ha-0 Ha)0 Ha-0Computer uses Supported Not supported Supported Not supported

Management controlPresident Ha)0 Ha-0 Ha-0 Ha-0

Not supported Not supported Supported SupportedSalesrmarketing Ha)0 Ha-0 Ha)0 Ha-0

Not supported Not supported Not supported Not supported

Forecast usesShort-term Hos0 Ha-0 Ha-0 Ha-0

Supported Not supported Not supported Not supportedLong-term Ha-0 Ha-0 Ha)0 Ha)0

Not supported Not supported Not supported Not supported

QuantitatiÕe techniquesTime series models Ha)0 Ha-0 Ha)0 Ha)0

Supported Not supported Not supported Not supportedCausal models Ha)0 Ha-0 Ha)0 Ha)0

Not supported Not supported Supported Not supportedQualitative models Ha)0 Ha)0 Ha)0 Ha-0

Not supported Not supported Not supported Not supported

SubjectiÕe factorsSubjective external Ha)0Supported Ha-0Not supported Ha-0Not supported Ha)0Not supportedSubjective internal Ha-0 Ha-0 Ha-0 Ha)0

Not supported Not supported Not supported SupportedForecast modifications Ha)0 Ha-0 Ha)0 Ha)0

Not supported Not supported Supported Not supported

the forecast is less in high-masculinity cultures. Theuse of qualitative techniques is less in high-masculin-ity cultures. And last, the internal subjective factorsare used more in high-masculinity cultures.

Unsupported masculinityr femininity hypotheses.There are no differences in computer usage andsalesrmarketing forecast involvement between firmsin masculinerfeminine countries. There is no rela-tio n sh ip b e tw e e n fo re c a s t u se s a n dfemininityrmasculinity country firms. No relation-ship exists between time seriesrcausal models andmasculinerfeminine cultures. Firms in high-mascu-linity cultures do not modify their forecasts anydifferently than firms in high-femininity cultures.

Before beginning any implication discussion, apoint should be made about multi-collinearity be-

tween the cultural variables. It is generally recog-nized that the cultural variables are correlated with

Ž .each other Hofstede, 1994 . However, in this study,because the variables are all one value within eachcountry, a higher degree of multi-collinearity exists

Žbetween these variables see Appendix A, Table.A.2 . Although there is a high degree of multi-collin-

earity, it does not bias the estimated coefficients, nordoes it mean that the coefficients are not significant,since the standard error of the coefficient estimate is

Ž Xbiased upward due to the near singularity of XX.matrix . Consequently, the statistically significant re-

sults are not biased and can be interpreted. Theprimary concern with multi-collinearity is that itcauses the size of the coefficients to be unstable

Ž .between samples Kerlinger and Pedhazur, 1974 .

Page 17: Forecasting accuracy: comparing the relative effectiveness of practices between seven developed countries

( )J.G. Wacker, L.G. SpraguerJournal of Operations Management 16 1998 271–290 287

3.2.1. Managerial implications of the cultural hy-potheses

The implication for the cultural hypotheses arequite interesting. First, some general overall conclu-sions can be drawn. Differences between countries intheir decision-making processes are culturally bound.The use of quantitative techniques, computers, rea-sons for making a forecast, and use of subjectiveinformation are quite different between countries.However, there is a ray of hope for understandingthese differences, since cultural traits developed by

Ž . ŽHofstede 1980, 1983, 1994 and others Hofstede.and Bond, 1988; Ronen and Shenkar, 1985 can

provide insights into why these differences exist. Thespecific implications of these conclusions for thefour cultural value variables are as follows.

3.2.1.1. Power–distance. The power–distance vari-able that has a positive relationship to computer useand time series suggests that firms in high power–distance countries rely more heavily on objectivedata than firms in low power–distance countries.Additionally, high power–distance firms use theforecast for more long-term strategic planning pur-poses. Also, firms in the high power–distance coun-tries rely more on external subjective measures andless on internal subjective measures. In short, firmsin high-masculinity cultures tend to rely on objectiveand external information for decision-making andplace less value on subjective and internal informa-tion.

3.2.1.2. Uncertainty aÕoidance. The results for un-certainty avoidance were quite disappointing. A pri-ori, it was believed that firms in high uncertaintyavoidance countries would have lower use of tech-nology than low uncertainty avoidance countries,since computers may present alternative forecastsand present an appearance of instability to employ-ees. This hypothesis was not supported. Firms inhigh uncertainty avoidance countries have no differ-ences in long-term or short-term planning than firmsin low uncertainty avoidance countries. However,they tend to use less subjective internal information.This lower usage may be due to the need for thesefirms to provide a stable environment by not reactingto internal information. The results suggest that thereare no differences between firms in high and low

uncertainty avoidance countries in making changesto the forecast. In fact, the statistical evidence sug-gests that these firms change the forecast more fre-quently than firms in low uncertainty avoidancecountries, which is exactly the opposite of what wasexpected. One conjecture for explaining this resultmay be that firms in high-uncertainty countries makea conscientious choice to avoid excess finished goodsinventory by modifying their forecast. Excess fin-ished goods inventory could make workers uncertainabout their future employment. Consequently, man-agers modify the forecast to keep the finished goodinventory at a minimum. However, this issue is notinvestigated in this study, and is a topic for futureresearch.

3.2.1.3. IndiÕidualismrcollectiÕism. As hypothe-sized, firms in the high individualism countries tendto be more technology oriented than firms in collec-tivism countries. The presidents of firms in highindividualism countries are less involved in the fore-cast development than in low individualism coun-tries. These high individualism firms are not differ-ent than other firms in the short-term and long-termuse of the forecast. Additionally, there is no differ-ence between firms in high and low individualismcountries in their use of subjective techniques. Firmsin high individualism countries modified their fore-cast more often, supporting a hypothesis that thesefirms change the forecast without the involvement ofthe lower level production workers.

3.2.1.4. Masculinityr femininity. In this study, themasculinityrfemininity variable is interpreted to rep-resent the need for performance to achieve competi-tive success. Firms in high-masculine countries tendto rely less on computers for competitive advantage.In high-masculine countries, the forecast use doesnot appear to be any different from other countries.However, the use of subjective internal informationis quite high for firms in masculine countries, whichsupports the hypothesis that development of internalinformation sources facilitates a competitive advan-tage for the firm.

The overall managerial conclusions on culturaldifferences are quite interesting. High power–dis-tance countries tend to use the forecast more forplanning purposes without much use of subjective

Page 18: Forecasting accuracy: comparing the relative effectiveness of practices between seven developed countries

( )J.G. Wacker, L.G. SpraguerJournal of Operations Management 16 1998 271–290288

factors. High uncertainty avoidance countries tendŽnot to use the forecast for planning other than

.facility planning and do not rely on any subjectiveinformation. Lastly, high individualistic countries do

Žnot tend to use the forecast except for production.planning for any planning purposes. However, these

firms tend to heavily use subjective factors for devel-oping a forecast.

3.2.2. The oÕerall relationship of cultural Õariablesto explanatory factors in this study

The effect of cultural factors on the forecast isvery important, since it is frequently alleged that datafrom many countries are heterogeneous because ofunspecified cultural differences. This study offers aray of hope for explaining between culture differ-ences. The results of this study suggest that techno-logical and quantitative methods are bound to cul-ture, since power–distance and individualism bothaffect computer usage and quantitative methods. Ad-ditionally, the use of the forecast for planning isweakly related to culture, since power–distance aloneis significant for long-term planning. Presidentialinvolvement in the forecast is less prevalent in indi-vidualistic and masculine cultures. In addition, sub-jective internal factors are used differently betweencultures.

There are some important implications for interna-tional corporations. It is not uncommon for multina-tional and transnational corporations to dictate poli-

Žcies and procedures Deresky, 1994, 1997. The re-sults of this study suggest that firms in high power–distance and high-masculinity countries use the com-puter much more frequently than firms in low powerdifference and low-masculinity countries. This meansthat corporate policies set by parent companies re-quiring computer usage for developing forecastswould be well-received in high power–distance andhigh-masculinity countries and meet resistance infirms in low power–distance and low-masculinitycountries. Also, for firms in low individualism andlow-masculinity countries, it would be difficult toimplement a policy that would require top managersŽ .subsidiary presidents to be involved in the forecast.Last, subsidiary firms in high-uncertainty countrieswould resist the use of subjective internal informa-tion, while firms in high-masculinity cultures would

prefer to rely heavily on this information. In short,parent corporations must tailor their forecasting pro-cedures to each country’s culture.

Summarizing the culture implications, the use ofcultural differences offers a rich new area of expla-nation as to why certain factors are more importantin decision-making among countries. With the intro-duction of these factors, researchers can explain howfirms make decisions between countries rather thanusing the explanation that they are different because

Žof unspecified cultural factors religion, mores, lan-.guage, etc. . In short, the between-country differ-

Žences in forecast procedures and therefore, perfor-.mance can be attributed to specific cultural values

in each country.

4. Implications for future research

This article was written to better understand howinstitutional practices affect forecast accuracy. Fromthat perspective, it has provided several pragmaticsuggestions to increase forecast accuracy. This studyshould be viewed as a beginning for a more integra-tive approach to understanding between-country dif-ferences in manufacturing behavior. Certainly, com-paring manufacturing competitive differences be-tween countries should incorporate the cultural fac-tors recommended by the international managementliterature. From a logical perspective, incorporatingcultural differences into international comparativemanufacturing enriches the explanations of howmanagers use their technical expertise for competi-tive advantage. In this study, these cultural factorsprovide an enhanced understanding why each coun-try behaves differently when developing the forecast,and how those behaviors affect forecast accuracybetween firms and countries.

Future studies that use international data for com-parisons and generalizations of results should con-sider cultural factors, as well as ‘pure’ manufactur-ing factors, for evaluating differences between coun-tries. A deeper level of understanding of these willhelp explain why some international firms use manu-facturing practices differently to compete in the in-ternationally competitive market.

Page 19: Forecasting accuracy: comparing the relative effectiveness of practices between seven developed countries

( )J.G. Wacker, L.G. SpraguerJournal of Operations Management 16 1998 271–290 289

Appendix A

Table A.1: The samples sizes from each country.

Frequency PercentGermany 17 5.1Japan 91 27.5Mexico 64 19.3New Zealand 18 18Spain 42 12.7Sweden 20 6USA 79 23.9Total 331 100

Table A.2: Correlation coefficients between cultural variables.

Power Uncertainty IndividualUncertainty 0.381Individual y0.754 y0.773Masculinity 0.056 0.750 y0.348

Table A.3: Cultural effects on forecasting practices: statistical results of regression analyses.

Power–distance Uncertainty Individualismr Masculinityravoidance collectivism feminity

Technology 4.300 y0.683 2.247 y0.64Ž . Ž . Ž . Ž .Computer uses 0.000 0.248 0.013 0.261

Management controlPresident y0.055 y0.303 y1.404 y2.083

Ž . Ž . Ž . Ž .0.478 0.381 0.081 0.019Salesrmarketing 1.12 y0.678 1.168 y0.13

Ž . Ž . Ž . Ž .0.132 0.249 0.122 0.448

Forecast usesShort-term 0.799 y0.116 0.187 0.531

Ž . Ž . Ž . Ž .0.4248 0.9075 0.8516 0.5955Long-term 2.314 y1.182 0.669 0.193

Ž . Ž . Ž . Ž .0.011 0.119 0.252 0.424

QuantitatiÕe techniquesTime series models 1.445 y0.179 0.114 y0.210

Ž . Ž . Ž . Ž .0.0747 0.4291 0.4546 0.4937Causal models 0.239 y0.636 y1.903 y0.131

Ž . Ž . Ž . Ž .0.4055 0.2627 0.0289 0.4479Qualitative models 0.273 0.078 y0.116 y1.394

Ž . Ž . Ž . Ž .0.3927 0.4689 0.4540 0.0841

SubjectiÕe factorsSubjective external 1.807 0.06 1.027 1.099

Ž . Ž . Ž . Ž .0.036 0.476 0.153 0.136Subjective internal y0.325 y1.717 y1.222 3.016

Ž . Ž . Ž . Ž .0.373 0.043 0.111 0.001Forecast modifications 0.926 1.286 2.298 y0.450

Ž . Ž . Ž . Ž .0.178 0.100 0.011 0.327

Page 20: Forecasting accuracy: comparing the relative effectiveness of practices between seven developed countries

( )J.G. Wacker, L.G. SpraguerJournal of Operations Management 16 1998 271–290290

References

Alkhafaji, A.F., 1995. Competitive Global Management, Princi-ples and Strategies. St. Lucie Press, Delray, FL.

Armstrong, J.S., 1984. Forecasting by extrapolation: conclusionsŽ .from 25 years of research. Interfaces 4 6 , 52–66.

Biggs, J.R., Campion, W.M., 1982. The effect and cost of forecasterror bias for multiple-stage production-inventory systems.

Ž .Decision Sci. 13 4 , 570–584.Bollen, K.A., 1989, Structural Equation Models with Latent Vari-

ables, Wiley, New York.Dalrymple, D.J., 1987. Sales forecasting practices: results from a

United States survey. Int. J. Forecasting 3, 379–391.Dawes, 1976. Shallow psychology. In: Carroll, J.S., Payne, J.W.

Ž .Eds. , Cognition and Social Behavior, Erbaum, Hillsdale, NJ.Deresky, H., 1994. International Management, Managing Across

Borders and Cultures. Harper-Collins College Publishers, NewYork.

Deresky, H., 1997. International Management, Managing AcrossBorders and Cultures, 2nd edn. Harper-Collins College Pub-lishers, New York.

Faheti, K., 1996. International Management: A Cross-Cultural andFunctional Approach. Prentice-Hall, Upper Saddle River, NJ.

Goddard, W.E., 1989. Let’s Scrap Forecasting. Modern MaterialsHandling, September, 39.

Hofstede, G., 1980. Culture Consequences, international Differ-ences in Work-Related Value. Sage Publications, BeverlyHills, CA.

Hofstede, G., 1983. National cultures in four dimensions. Interna-tional Studies of Management and Organizations, Spring–Summer.

Hofstede, G., 1994. Management scientists are human. Manage.Ž .Sci. 40 1 , 4–13.

Hofstede, G., Bond, M.H., 1988. The Confucius connection: fromŽ .cultural roots to economic growth. Org. Dynamics 16 4 ,

4–21, Spring.Holt, C.C., Modigliani, F., Simon, H.A., 1955. A linear decision

rule for production and employment scheduling. Manage. Sci.Ž .2 1 , 10–30.

Kerlinger, F.N., Pedhazur, E.J., 1974. Multiple Regression inBehavioral Sciences. Holt, Rinehart and Winston.

Libby, R., 1976. Man versus model of man: some conflictingŽ .evidence. Org. Behav. Human Performance 16 1 , 1–12.

Makridakis, S., Winkler, R., 1983. Averages of forecasts, someempirical results. Manage. Sci., 987–996.

Mentzler, J.T., Cox, J.E., 1984. Familiarity, application, andperformance of sales forecasting techniques. J. Forecasting 3,27–36.

Nunnally, J.C., 1967. Psychometric Theory. McGraw-Hill, NewYork.

Parker, G.G.C., Segura, E.L., 1971. How to get a better forecast.Harvard Business Review, March, pp. 99–109.

Plossl, G., 1973. Getting the most from forecasts. Production andInventory Management, First Quarter, pp. 1–15.

Pindyck and Rubenfeld, 1976.Ritzman, L.P., King, B.E., 1993. The relative significance of

forecast errors in multistage manufacturing. J. OperationsManage. 11, 51–65.

Ronen, D., Shenkar, O., 1985. Clustering countries on attitudinaldimensions: a review and syntheses. Acad. Manage. Rev. 10Ž .3 , 435–454.

Sawyer, J., 1966. Measurement and prediction, clinical and statis-Ž .tical. Psychol. Bull. 6 3 , 178–200.

Sprague, L.G., Naji, Z.T., 1992. The state of the art and science ofmanufacturing in the UK. Preliminary results, Working Paper,pp. 71–76.

Tversky, A., 1973. Availability: a heuristic for judging frequencyŽ .and probability. Cogn. Psychol. 5 2 , 207–232.

Vollman, T.E., Berry, W.L., Whybark, D.C., 1992. Manufacturingplanning and control systems, 3rd edn. Richard D. Irwin,Homewood, IL.

Wacker, J.G., Sprague, L.G., 1995. The impact of institutionalfactors on forecast accuracy: manufacturing executive perspec-

Ž .tive. Int. J. Prod. Res. 33 11 , 2945–2958.Wheelwright, S.C., Makridakis, S., 1980. Forecasting Methods for

Management, 3rd edn. Wiley, New York.Ž .Whybark, D.C., Vastag, G. Eds. , 1994. Global Manufacturing

Practice. Collected Papers, Elsevier.