Empirical Study of 00 Ch Of

7/30/2019 Empirical Study of 00 Ch Of

1/68


2/68

LIBRARYOF THE

MASSACHUSETTS INSTITUTEOF TECHNOLOGY


3/68


4/68


5/68


6/68


7/68

WORKING PAPERALFRED P. SLOAN SCHOOL OF MANAGEMENT

MASS. inST. TECH.

DEC 9 75

nFVJEY LiBRARV,

AN EMPIRICAL STUDY OF THE INDUSTRIAL MARKETING MIXJean-Marie Choffray

WP 800-75 July 1975

MASSACHUSETTSINSTITUTE OF TECHNOLOGY

50 MEMORIAL DRIVECAMBRIDGE, MASSACHUSETTS 02139


8/68


9/68

AN EMPIRICAL STUDY OF THE INDUSTRIAL MARKETING MIX

Jean-Marie Choffray

WP 800-75 July 1975


10/68

M-l-T. LIBRARIESDEC 10 1975RECEIVED


11/68

-2-

Acknowledgt ments

I wish to thank Professor Gary L. Lilian of the Sloan School ofManagement at M.I.T. He not only provided the data and the necessarycomputer time to analyze it, he also offered much useful advice on howthe data might be analyzed and the results presented.

Thanks are also due to Professor Alvin J. Silk for his criticalcomments on an earlier draft of this paper.


12/68


13/68

-3-

ABSTRACT

An empirical study is made of how personal selling, advertising andtechnical service expenditures are associated with industrial products' grosscontribution to profit. The implications of a multiplicative model relat-ing the average gross contribution per customer to industrial marketingspending rates are discussed. Robust regression is used to estimate themodel's parameters on the basis of a sample of 50 industrial products.

Results indicate that personal selling expenditures are the singlemost important factor of the industrial marketing mix. They do not supportthe hypothesis that increasing returns to scale exist in industrial marketing.They also shed some light on the possible substitution between personalselling and industrial communication expenditures.


14/68


15/68

-4-

Introduction

The relationship of Industrial marketing activities to a product'scontribution to profit has not been well established. Recently, a reviewappeared of the existing information on the effectiveness of industrialmarketing activities, particularly the effectiveness of industrial adver-tising in terms of sales and profits (see Lllien et al. [1]). The authorsconcluded that only a limited number of empirical studies were available.They did Identify evidence of economies of scale, threshhold effects andinteraction effects between industrial communication and personal selling.

This paper investigates on the basis of data collected through theADVISOR research project [2] how personal selling, advertising and technicalservice expenditures are related to Industrial products' gross contributionto profits.

Our decision to use the gross contribution to profits as a measureto be related to these "budgeted" marketing activities comes from the factthat the gross contribution to profits is made up of three components unit price, average total cost, and quantity sold which could all be influenceby the level of these marketing activities.

The relation between a product price and the company's marketingactivities has been analyzed by Palda [3]. He considers that any pricingdecision must be taken against the background of the firm's total marketingstrategy, so that a product price appears more as a dependent variable thanas an Independent one. Recently, Lambin [4], argued that "consumers exhibitlower price responsiveness in high-intensity advertising markets than theydo when the advertising level is low. Companies therefore have the opportunit


16/68


17/68

-5-

to charge above-normal prices." Lambin mentions, however, that this oppor-tunity tends to be limited by the use of price controls. The question is:is this comment applicable to industrial markets?

The relation between the quantity of a product sold and the company'smarketing activities is one of the basic postulates of marketing theoryand often the justification for these expenditures. If we accept thispostulate for industrial products, the relation between the firm's marketingactivities and the average cost of production is clearer. Usually when

the company operates below capacity, an increase in the quantity producedleads to a decrease in the average product cost.

In this paper, we relate the average gross contribution of industrialproducts, on a customer basis, to the spending rate for personal selling,technical service, and advertising activities. Our results indicate thata substantial part of the variation in industrial products' gross contribu-tion to profit may be explained with these three marketing activities. As a regression model is only descriptive, dealing with measures of associationrather than causal relationships [5], a strong interpretation of our resultsneeds the assumption that industrial marketing activities do indeed affecta product's gross contribution to profit.

Interpreted this way, our results underline the importance of personalselling in the industrial marketing mix. They do no support the hypothesisthat increasing returns to scale exist in industrial marketing. Finally,they allow us to speculate about substitution between personal selling andadvertising expenditures.


18/68


19/68

-6-

Data Collection Procedure

Our analysis has been performed on data received from five out oftwelve industrial companies that participate in the ADVISOR research project.Each of these companies was requested to provide information about as manyindustrial products as it could. No special requirements were set concerningthe characteristics of the products to be selected.

Data were collected by questionnaire. The specific information re-

quested arose from a series of personal interviews with industrial marketingmanagers about the key variables that they take into account in budgetingdecisions. More than 190 elements of information were introduced in thequestionnaire. The questions were grouped into six broad areas;

- Company characteristics- Product qualities- Cost; profit information- Growth, production and distribution- Use; customer and competitive characteristics- Advertising, personal selling and technical serviceDue to the confidentiality of some of the requested information,

companies were allowed to multiply actual numbers by a "security coefficient."'iliis protection mechanism introduces a small random component in some of thedata so that the actual amount is only known with a 10% margin. We expectthe effect of this random component on our results is limited due to theway we defined the variables.

Our sample consists of 50 industrial products that range from raw

materials and chemicals to more elaborate products such as machinery andequipment. Most are in one of the first three stages in the product life


20/68


21/68

-7-

cycle and are associated with oligopolistic markets.

Definition of the Variable Under Study and PreliminaiTr Analysis

Our purpose is to study the relationship between industrial products'gross contribution to profits and the associated level of personal selling,technical service and advertising expenditures. We have attempted to definethese variables in such a way that they would have both an operational mean-

ing within the industrial marketing decision making context, and could beestimated on the basis of the available data.

Our analysis has been limited to mainly seven variables whose definitionfollow:

A ; A _ refers to the Advertising spending rate per potentialcustomer in 1973 and 1972 respectively.P ; P _ refers to the P^ersonal selling spending rate per poten-^ ^ tial customer in 1973 and 1972.S ; S _ refers to the technical ervice spending rate peractual customer in 1973 and 1972.G refers to the average gross contribution to profit

per actual customer in 1973. This amount is obtainedby the following relation

G= (p-c)q/n , wheret *^t t ^t tp = average unit price for the product in 1973.c = average total cost of production for the product,only excluding marketing costs.q = total quantity sold in 1973.n = actual number of customers In 1973.t


22/68


23/68

-8-

In addition to these seven variables, we defined several dummy vari-ables to control for a stage in the life cycle effect and for a companyspecific effect. These last variables were expected to represent the com-posite effect of many company-product characteristics not explicitly intro-duced in the model.

Note that some of these variables were defined on the basis of theactual number of customers while others were on the basis of the potentialnumber of customers. Indeed, it makes more sense to relate industrial ad-vertising and personal selling spendings to the potential number of customersas a large part of these expenditures is aimed at drawing in new customers.On the other hand, the average gross contribution to profits, and technicalservice expenditures are related to the actual number of customers. Inthis study, the potential number of customers was considered to be the totalnumber of customers for the industry reported in the questionnaire.

A preliminary statistical analysis of the seven variables of interestwas made and is reported in Choffray [6]. Empirical distributions for thesevariables were found to be highly skewed. Measures of centrality were verysensitive to a few extreme observations and the need for stabilizing trans-formations was evident. A logarithmic transformation was used, and consider-ably improved the stability of the various measures of centrality.

A correlation analysis was also performed (see Choffray [6] fordetails). It was found that the advertising spending rate in 1972 was morestrongly related to the average contribution to profit than the advertisingspending rate in 1973, suggesting that industrial communication expendituresmight have a delayed effect on the product's contribution to profit.


24/68


25/68

-9-

In order to deterniine if the effect of advertising expenditures onthe average contribution to profits was direct or indirect with personalselling expenditures as an interviewing variable we used a method suggestedby Simon [7]. The partial correlation between advertising spending rateand average contribution to profit keeping personal selling expendituresconstant appeared to be CSi.18. This doesn't allow us to reject the hypothesisthat the advertising spending rate has a specific effect on the average grosscontribution to profits of industrial products. On the other hand, the pnr-tial correlation coefficient between technical service spending rate andaverage contribution to profit, keeping personal selling expenditures con-stant, turned out to be ^ .06. This indicates that the effect of technicalservice expenditures on the average gross contribution of industrial productsmight be deeply intertwined with that of personal selling.

The Model

Assume a relation between the three Industrial marketing spending ratesunder study and the average gross contribution to profit of the following,multiplicative form:

G = K n M^ ,J '

where the M 's represent the various marketing spending rates.This model has some important characteristics. It allows for inter-

actions between predictor variables and its interpretation provides Inter-esting insights into the process of how industrial marketing spending ratesmight affect products' gross contribution to profits. In this respect Itcan easily be shown that


26/68


27/68

-10-

1 - a represents the elasticity of Industrial product gross contri-bution with respect to marketing spending rate M .

2 - the nature of returns to scale can be inferred from the simplesummation of the model parameters: E a .J ^

3 - the rate of substitution of marketing activity M for marketingactivity M Is given by

dM a. M^11 " " dM^ " ^ ^

In addition, this model is linear in the logarithms. Given theavailable data, the new dependent variable will be more nearly normal thanif the original values had been used. The statistical tests on the para-maters of this multiplicative model will then be more meaningful than ifa simple linear model relating the original "rates" had been used.

The use of a multiplicative model has important limitations:- First, it is clear that the model does not allow for a negative average

gross contribution to profits as all M are always > 0. This does notraise a crucial problem, however, due to the definition we gave of thegross contribution as the difference between sales and total productioncost. Indeed, one might reasonably assume that companies discontinueproduction, when their average production cost is constantly higher thantheir selling price.

- Second, assuming ol > 0, the model allows for infinite contribution toprofits when infinite amounts are spent on marketing activity M . Forthis reason, it is clear that the model must only be considered aH .i uhc-ful approximation of the true relation within the range of the observed

data.


28/68


29/68

-11-

- Third, the model Implies that the "effect" of an additional dollar inany M is the same for all industrial companies represented in the sample,w

Parameter Estimation

There are several estimation problems related to our data. The firstis that the independent variables in our model are subject to error. Indeed,the use of security coefficients resulted in the introduction of a random

component in the reported marketing spending amounts.Regression techniques, used to estimate the model parameters, assume

that only the dependent variable i.e. the average gross contribution toprofits is subject to error. When both the dependent and the Independentvariables are subject to error, least squares regression becomes highlyinefficient [8]. In order to estimate our model, we then assume that randomfluctuations of the Independent variables are negligible.

Another problem is multicollinearity. This is not unexpected as theseveral spending rates are intercorrelated and further, that for any givenmarketing activity, say personal selling, the autocorrelcation of its spendingrate over time is quite high.

This multicollinearity problem, has two important implications.- First, ordinary least squares regression usually lead to large standard

deviations of the estimated parameters a . These coefficients are oftenlarge in absolute value, and may even have the wrong sign.

- Second, due to the interdependency among the predictor variables itbecomes more difficult to select those which are most meaningful.


30/68


31/68

-12-

For these reasons, we decided to use Ridge regression, an excellentdescription of which is given by Hoerl and Kennard [9] and [10]. The idea

underlying Ridge regression is that we can study the sensitivity of theregression estimates to the multicollinearity problem, by adding a smallpositive constant k on the diagonal of (X'X), where X denotes the matrixof predictor variables. The important point is that we know that for somevalue of k, our estimates will be closer to the true value of the parametersthan the ordinary least squares estimates.

Our first task is to reduce the original set of six marketing spendingrates ^^_i ^^ ^t i ^j- ^t- ^ _i ^ ^^ included in the final model.The use of the Ridge Trace [10] as a way to identify the most meaningfulsubset of prediction variables led to the selection of A ^, P , S .

The Ridge Trace for the corresponding model

a a aG = K A % P P S ^t t-1 t t

appears in the Appendix, Exhibit I. This Trace is fairly stable. Thisis specifically true for a* , the standardized coefficient of A , and toa t-1* *a lesser extent for a and a . The question of deciding whether S shouldp s tbe dropped from the model is unclear. The effect of technical service andpersonal selling spending rates on the average gross contribution to profitsare deeply intertwined so that both variables may in fact represent thesame basic factor. We decided however to keep S in the model on the haslHthat conceptually we expected the technical spending rate to havi- .i Ht-paruLeeffect weak perhaps on the dependent variable.


32/68


33/68

-13-

The estimated model parameters, evaluated at k = .07, are given by!3

TABLE I

ResponseVariablen=50


34/68


35/68

-14-

The Robust Trace for this model is shown in the appendix. Exhibit II. TheTrace relates the standardized regression coefficients to the parameter r.

When r=l, the Trace reproduces the Least Absolute Residual (LAR) estimatesof the parameters, and for r=0, it reproduces the least squares estimates.As r increases, more points are considered as extreme and are set aside.

The estimates of the model parameters are sensitive to a few individualsample points. Note however that the relative importance of the coefTlclents

*does not change. The standardized coefficient a is the largest of the three,*independently of the value of r. The coefficient of A _ , a is fairly

*stable. As expected, the only problem concerns a , the standardized coef-sflcient of S , whose sign changes over the range of r.

The interest of a Robust regression lies in the residuals analysis.Indeed, at r=.175 we know that our estimates are 95% efficient and we wouldlike to know how many points of the original sample have been effectivelydiscarded. For this reason we have included in the appendix, the residualsand weights of the original observations corresponding to r=.175 and r=.238.These values of r correspond approximately to an efficiency of .95, and .80respectively.

For r = .175, the Robust procedure discards 8 points and gives a weight2less than .7 to two other points. The resulting weighted R = .57. For

r = .238, 7 additional points receive a weight less than .7 and the resulting2weighted R is .64. We will not consider larger values of r, which result

in a low efficiency of the estimates and a substantial reduction of theeffective sample.


36/68


37/68

-15-

It is interesting to see that at r = .175 and r = .238 the estimates* *of a and a do not change very much. Note that for these values of r,P s

the estimate of a , (.48), can be considered as a lower bound on the trueP *value of a , as the trace for a reaches its minimum in the neighborhoodP Pof r = .2. So we feel reasonably confident in the estimates of a , a anda pK which correspond to r = .175 and are given by

TABLE II

ResponseVariable

n = 50

G t(t-stat)


38/68


39/68

-16-

In order to introduce these effects into the model, we defined twodummy variables to account for the stage in the life cycle effect z, , z and four others K , K , K , K to represent the company-specificeffect. These variables were introduced in the model so that their coefficientwould affect the scale factor. Results are presented in Table III for r=.175.

These results imply a life cycle effect on product contribution toprofits. Indeed for a given level of marketing spending rates industrialproducts that are in stage I of the life cycle have relatively higher averagegross contribution (K ~ $666) than products that are in stage II (K ~ $153),and than products that are in stage III (K - $138). Note that the explicitconsideration of the stage in the life cycle in our model did not Improve

2the fit substantially. Indeed, the weighted R at r = .175, which Ik now.63, was .57 when only the three marketing variables were considered.

The presence of a company-product effect is especially evident forcompany number 10. This may suggest that some variables related to indus-trial products' gross contribution to profits have been omitted from themodel and should be introduced in further analysis. When we introduce acompany effect, a , the coefficient of the technical spending rate becomesstatistically different from zero. Note also that the fit is significantly

2improved, as the weighted R is now .83.For comparison purposes, we have reproduced in Table IV the main resultH

of our analyses concerning the hypothesized effect of advertising (a ),personal selling (a ) and technical service (a ) spending rates on theP saverage gross contribution to profits of industrial products.


40/68


41/68

-17-

TABLE III

Model I: LnG = LnK + C,Z, + CZ^ + a LnA , + a LnP + a LnSt 112 2a t-1 p t St


42/68


43/68

-18-

The scale factor, K, has not been introduced in this table as:- the value of K depends on the stage in the product life cycle andon the company under consideration.- the value of K does not affect the analysis of the results of our study,

TABLE IV

Robust Estimates of the Parametersr = .175

model


44/68


45/68

-19-

Discussion of Results

We will eliminate technical service from our discussion of results asthe estimate of a has been found quite unstable, and strongly influencedby the company factor. A larger data base, however, might help dissociatethe specific effect of technical service expenditures from the effect ofpersonal selling.

The results of this study do not allow us to reject the hypothesis thathigher gross contribution to profits are supported by larger personal sellingand communication spending rates. Indeed, the results imply that those indus-trial products for which more is spent in personal selling and advert Is inj.-,per potential customer tend to be those realizing the largest gross contri-bution to profits on a customer basis.

The best model to relate Industrial products' contribution to profitsto marketing spending rates was found to include A _ , P and S . The results,however, were not changed fundamentally when A was used instead of A -. .It is then delicate to make inferences concerning the lagged effect of com-munication expenditures, although the results seem to indicate that adver-tising could have a lagged effect.

The relative importance of these two elements of the industrial marketingmix is interesting to consider. The personal selling spending rate has beenconsistently found to be the most important element of the marketing mix interms of its association with the average gross contribution to profits.Indeed, the elasticity of the average gross contribution to profit, withrespect to the personal selling spending rate was found to be approximatelybetween two and three times larger than the elasticity of this same measure


46/68


47/68

-20-

with respect to the advertising spending rate. If we were to make the(strong) assumption of causality, the model then implies that the budget

should be allocated in such a way that the ratio of the communication spend-ing rate to the personal selling spending rate equals the ratio of theelasticity of the average gross contribution with respect to both of them.I.e., A/P = a /a , where a /a was found to lie in the interval [.33, .501.a p a p L.JAs these two marketing spending rates are both defined with respect to thenumber of potential customers, our data indicate that the overall industrialadvertising budget should represent between 1/3 and 1/2 of the total personalselling budget! This contrasts with the median ratio (- .11) found in oursample.

The substitution between personal selling and communication expendituresis also interesting to discuss. The rate of substitution of communicationfor personal selling is given by

s =-^ = i . Aa,p dP a ' PaAs the ratio a /a varies approximately in the interval [2,3] we can get anP 3estimate of the upper and lower bounds on S when we know the current levela,pof both A and P. On the basis of the observed data it appears that a reason-

*able estimate of A/P is 1/6. So, if we assume that the estimated relationis indeed true, the bounds on S are:a,p

S (upper) ^ ^a,pS ^^""> = .33a,p

*This estimate is based on the ratio of the average spending rate foradvertising and personal selling in 1973.


48/68


49/68

-21-

Again, a causal interpretation of our model could lead to more spending (onthe average) for advertising at the expense of personal selling in our sample.

As far as the nature of returns to scale is concerned, the results indi-cate that the sum of a , a , and a is consistently less than 1. So. thesea p sresults do not support the hypothesis that increasing returns to scale, interms of products' gross contribution to profits exist in industrial market-ing.

Conclusions

The results of this study allowed us to give tentative answers to someimportant questions faced by industrial marketers. Specifically, our resultsindicate that:

- the hypothesis according to which industrial marketing activitiessupport larger contribution to profits cannot be rejected.

- personal selling expenditures are the most important element of theindustrial marketing mix in terms of its association with products'gross contribution to profit.

- the hypothesis according to which industrial marketing activitiesproduce decreasing returns to scale is supported by the availabledata.

Finally, if our model is given a causal interpretation, then companiesrepresented in the sample could be overspending on personal selling.

By using personal selling, communication and technical service spendingrates, we were able to "explain" 57% of the total variation in industrial

products' contribution to profits. As noted earlier, the importance of thecompany effect confirms that some important product-market characteristics


50/68


51/68

-22-

were omitted from the model. Future research should concentrate on the iden-tification of these characteristics and on their integration in a model.

Many problems we encountered in this study were associated with theuse of cross-sectional data to estimate a model of response to industrialmarketing activities. From a methodological point of view, the use ofcross-sectional data to estimate such a model presents serious weaknessesthat have been discussed by Quandt [11].

Despite these weaknesses, the study provides some indications abouthow industrial marketing activities might work. The use of time seriesdata could allow us to dig deeper into the industrial marketing mixproblem.


52/68


53/68

-23-

APPENDIX


54/68


55/68

-24-

io

EXHIBIT I . niVQB TRPiC&Model G,-_ K f\,';p'' ^1^

Al i~-

^.5 5K.

J>3o

^80


56/68


57/68

-25-

a"

V


58/68


59/68

HOhUbT REQRBSS^IOh/ RESIDUALSr- .176

-

H::.hliIiiM

Riil-i

1c!

4

I'.LtililjIlL

11li::1-J141516i?lb19ee1C.:'.)

C'vC-.-"..-

cli:;.

Jl

4i4c:.4-:^444:34t:.4?4i::

i. . :::;0:':;'!:j-i::i!.999!:.V'

1 u a I r"7. 4t.?ib. 749i::-.I;:iG. i.:i6ti5iS3b. IbbV be:0. I:::l!:::.4b:j'ri1 .;.:::4b3c:fel. l?':::'cb-L!.b57'919t;i.t-17V!;.9

-0. b6U!:>63I. n o'-'-c r r

b. olbi::.:bb1 .b':'y!:ic:-&. 1 obi 7 3-1 i:;:4441.41339-b.7'7M.b30. 0!:j9':i47I:::! c!':"'b4-.:'1 , t.:::'it.3

b,':j4bibfcbu .1. l:.:7 Jti1 919,,:.:I . 0b9U3LI. :.;..'. i707'bl MS4b 4

-0.4,.:;:3 7'7 7'0.i::::83L::5i . 4500?

~i:.! . b349i::!i:.l . 1 1 b i 9b . J'd'd.i'' i H-0.,31ba3c::

-I -! ; --ii:;:) i .-'TI_'..I7'. c:7''6oi0.3lbc:!:r.l0.34334c:c:.03c:S

i.::i. '^iti-^V-titi:.?0.69917!:.b.35';.bb. 134397.3V436b. bi303.49639-0.377330. t-4b643

0l:;..i,biri


60/68


61/68

nosusr REGaEssior^ REbtoun i S -r i .i3 &

kirn LbiiilJHL NLlLJiT

le11IS131415161?18ISB1

4t.:-U.-.J

c-'bC i'ele-cts3f\

zi(~

'H.:.

4h:^:.!:-i

1 . i::.t;ii:::i35

1.0041ll:.93i?u44l:::i^:i3.7563'::'3. i 13^943 1357631:1. UbtJobli. . c4'~io?8. 340733-O. 7t.73.3U0U>:3

-3, 1133341 .. SSbcJ-0. 353363i .331530. 101331.1 . ikt.SVI . t30350.7413733. 103733.1. . V'ti:;::::::!

3 i7t. ly3. J. 30501:.:. 0i!:..0U1 . U363 J.

I::l . ..:::3 1 cA 11:1. 0t..L0;.::;-U353J4y0.315313i. 44335-3 . ::::ti:j33b.e4317t.0. 7713630.433367, i0i::.3? . 33 151K.. 353135Li. v.i3iZl':::':.:>^-i-

; ., 'ii:.; r" i : ..:!, I...1:1-435-103I-!., t..>.6i::i73b- -, I. ill: :o , 3t-:-.7. . t:3 , :.:::'...bly,:;,:;.:;.507tiS

-0.333571H. 634433

';: I. !:::!::; tub3 ''50.c.03t.47

h.


62/68


63/68

-29-

REFERENCES

[1] Gary L. Lilien, et^. al . , "Industrial Advertising Effects and BudgetingPractices: A Review," Working Paper 761-75, Sloan School of Management,M.I.T., January 1975.

[2] ADVISOR, "ADvertising I^ndustrial Products: tudy of Operating Relation-ships," is an M.I.T.-A.N.A. joint research project whose objective isto study and model the industrial advertising budgeting process. Fora description of this project, see Gary L. Lilien, "How Many Dollarsfor Industrial Advertising? Project ADVISOR," Working Paper 735-74,Sloan School of Management, M.I.T., September 1974.

[3] Kristian S. Palda, Pricing Decisions and Marketing Policy , Prentice-Hall, Englewood Cliffs, New Jersey, 1971.[4] Jean-Jacques Lambin, "What is the Real Impact of Advertising?" Harvard

Business Review , May-June 1975, p. 146.[5] Jean-Jacques Lambin, "Measuring the Profitability of Advertising: An

Empirical Study," Journal of Industrial Economics , Vol. XVII (April1969), p. 86.

[6] Jean-Marie Choffray, "An Empirical Study of the Relation between Indus-trial Products' Profitability and Marketing Spending Rates," UnpublishedResearch Paper, Alfred P. Sloan School of Management, M.I.T., May 1, 1975,

[7] Herbert A. Simon, "Spurious Correlation: A Causal Interpretation,"Journal of the American Statistical Association , Vol. 49, 1954, p. 467.[8] Ronald J. Wonnacott and Thomas H. Wommacott, Econometrics , John Wiley

and Sons, Inc., 1970, p. 164-167.[9] Arthur E. Hoerl and Robert W. Kennard, "Ridge Regression: Biased Esti-

mation for Nonorthogonal Problems," Technometrics , Vol. 12, No. 1,February 1970.

[10] Arthur E. Hoerl and Robert W. Kennard, "Ridge Regression: AppHcatlonHto Non-Orthogonal Problems," Technometrica, Vol. 12, No. 1, February1970, pp. 70-85.

[11] Richard E. Quandt, "Estimating the Effect of Advertising: Some Pitfallsin Econometric Methods," Journal of Marketing Research , May 1974,p. 51-60.


64/68


65/68


66/68

Date Due


67/68

HD28.M4 14 no.796- 75Kearl, James R/Appendix to task III :

'''liiiiiiililiiiiiiilii3 TDflO DOO bSb 311

^_ w no.797- 75jinzberg, Mich/lmplementation as a pro724723. D*BKS _ .00024261

3 TDflD ODD 7DM 035w no.798- 75ftnand, Sudeep./lntertemporal portfolio

72587P , _ DBKS. _00.0l98G5II I II ill

3 TOflO DDO b4S"?bTw no.799- 75

Shapiro, Jerem/OR models for enerqy pi724715 _ DxBKS 00019872

3 TOflO OQO bMS T3Mw no800- 75

ii?o*,''o^- Je3"/An empirical study of t/2b818 DxBKS 00019882

3 TOflO ODD bMb 171w no.801- 75

Bailyn, Lotte./Researcti as a cognitive724718 p*BKS 00019863'}:''!?

3 TOAD ODD bMS hlHHD28.IVI414 no,802-75Rockart, Jofin /Computers and the learn725169 D*BKS 00185564

III ft":' ii"i"|'r|' [If T'' ,:'"rf"'iniiii

3 TDflD DOS '711:,' 13Tw no-803- 75Runge, Dale. /The potential evil in h725165 D

iiiiiiii*BKS 000111! I 71

3 TOflO ODD bMS TOO


68/68

Documents

Empirical Study of 00 Ch Of