ESCALATION TENDENCIES OF ADVERTISING*

By M. M. METWALLY

This paper attempts to establish a number of criteria which may be used toexamine whether the claim of widespread 'self-cancellation' of advertising isjustified. A model incorporating these criteria is developed and empirically testedusing relevant Australian data.

I

In a competitive system where there is a large number of sellers and completeknowledge on the part of the consumers, there is clearly no place for promotionalexpenditure. Consumers are indifferent to rival sellers and competition can takeplace only through the price mechanism. Consequently, advertising is seen as asign of imperfection of competition in an industry.

It is often argued that the basic role of advertising in modern capitalisteconomics is 'market-defensive'. This implies that when any firm in an industryembarks on an intensified advertising campaign, its competitors must step up theiradvertising or other sales efforts to avoid a possible loss of market position. Onthe other hand, if any firm decides to economize on its advertising budget, withouta compensating increase in some other aspect of its total selling effort, its exposureis reduced and its share of the market may decline if its competitors do not followa similar policy. Competitive pressures may lead individual firms to increase theiradvertising expenditure and the same pressures would also preclude their decreasingit. Our aim is to examine this 'self-cancellation' property of advertising.

Let us assume the following market-share function for the ith brand.ni1=4(v1, v, p11); i, j=l, . . . n, ij (1)

where mj = market sharey1 company advertising

= competitive advertising= the ith brand unit price as a ratio of its competitor(s) unit (or average)

price.

The function (1) is assumed to have the following properties:

em1/ev, >0; 2m/av? <0; m1/v <0;

I pj/p1>1 <0; m1/ep11 <j o

These properties state that company advertising has a favourable effect on itsmarket share; but there are diminishing returns. Competitive advertising has anadverse effect on the brand's market share while the effect of price would depend

* I am indebted to Garnsey Clemenger Pty. Ltd., Marketing and Advertising Consultants,Brisbane, and to the managing directors of the firms who assisted in supplying vital unpublisheddata that made this research possible. I wish also to express my gratitude to my colleagues,A. Anderson, G. Davy, R. Gunton, H. Higgs, R. R. Officer and G. West for their valuablecomments. Any (remaining) errors are, of course, my own responsibility.

153

- 7?A/A = veA

154 BULLETIN

on whether the brand price is higher or lower relative to the competitive price(s).Furthermore, it is assumed that y1 and v are functionally related, or

v=q.i(v1) (2)

From (1) and (2) we can define for the ith brand:

where = (am1/v1) . (vi/mi) as market-share elasticity with respect to companyadvertising.

ej (em1/v,) . (vfm) as market-share elasticity with respect to competitiveadvertising.

v1 = (vj/v1) . (v1/v) as advertisingreaction elasticity.

we may also define, for the ith brand,= v1/S1 as advertising: sales ratio (S1 being the volume of sales).

Now, if firms advertise only to protect their market shares, we would expect thefollowing results for any two competitive brands A and B. These results can easilybe generalized for more than two brands.'

- 'ÇlA/A vBA (3)

amA/ayA = amB/avB (4)

amA/avB = amB/A

hlA/flB °A/0B (5)

Equation (3) indicates that if advertising is used as a 'defensive' mechanism,the firms will spend just enough to compensate for the loss in their market sharecaused by the increase in competitive advertising. Thus a one per cent increasein competitive advertising will reduce the market share of (say) brand A by apercentage If competitive advertising increased by LvB/vB, this would reducethe market share of brand A by a percentage eA(zvB/vB). Since a one per centincrease in company advertising results in an increase in it market share by apercentage i, the company would need to increase its own advertising by:eA(LVB/VB)/77A to compensate for the reduction in its market share, so that

VA/VA -This gives:2

- 1A/EA = (vB/vA) (vA/vB) (6)

which establishes (3) above.

1 j and j in the definitions above are replaced by A and B respectively in the remainder ofSection L

2 The same result can be achieved using total differentials. In the absence of price compe-tition, we have

dmA=(mA/avA) dvA+(mA/8vB) dvB'Market-defensive' advertising results in dmA = O, or:

- (8mA/vA)/(bmA/8vB)=dvB/dvAwhich, by simple manipulation, gives:

[(mA/avA) (vAImA)]I[(amAIvB) (vB/mA)] = (dvfl/dvA) (VA/VB)

or

ESCALATION TENDENCIES OF ADVERTISING 155

Equation (4) suggests that if advertising were self-cancelling, we would expectthe marginal sales effect of company advertising and competitive advertising tobe of the same order of magnitude for all competitive brands. This implies thatfor any two competitive brands, their advertising: sales ratios would be in pro-portion to their market share elasticities as indicated by equation (5).

The criteria set out in equations (3), (4) and (5) indicate a tendency for anescalation in advertising expenditure. This would be reflected in the relativelyslow growth of total industry sales compared to total advertising outlays.

IIThe criteria developed above are tested empirically on six Australian products

for which data were made available. The products are instant coffee, bottled beer,cigarettes, toothpaste, toilet soap and washing powder. The industries producingthese products are oligopolistic and characterized by non-price competition. Theproducts are frequently said to be 'heavily advertised' or 'over-advertised'. Anexamination of the escalation effects of advertising in these industries should,therefore, be meaningful. In choosing the appropriate model, we must bear inmind that we are dealing with a phenomenon in which a number of the variablesinteract. In addition to the interdependence that exists between brand andcompetitive advertising, there is a general agreement that not only market sharesare influenced by advertising but advertising is also influenced by market shares[5]. We must, therefore, use methods which take these interdependencies explicitlyinto account.

We construct the following system of simultaneous equations, called structuralequations, to present the market share and advertising functions necessary to testour criteria.

loge = c + ai/V1t + c12/v + a3qj, -1 + a4pt + u1t (7)

v1t=ßo +ß1v1._ +ß2v,,_1 +ß3m1+ u2t (8)

vjt=yo +yivj.t_j +V2V1,t_i +y3(l m1) +u3t (9)

The system has three equations with three endogeneous variables and fourpredetermined variables.

Equation (7) explores the existing relation between the share of the marketand the share in marketing pressures. It gives market share as a function of brandadvertising, competitive advertising and the quantities sold in the previous period.Market share is also assumed to depend on relative prices in the current period.This equation may be compared with models developed by Kotler [4], Lambin [5],Urban [8] and Weiss [10]. We depart from these models, however, in assumingthat market share and advertising are related in a non-linear fashion. A linearrelationship implies no decreasing marginal returns to expenditure on advertising.Its optimization, therefore, would suggest infinite advertising followed by infiniteexpansion in market share, which does not seem to conform with the practicalsituation. Customers' 'loyalty' or 'attachment' to a particular brand suggests theexistence of a saturation point in the relationship between market share and

15G BULLETIN

advertising. Thus for the ith brand, the relationship may take either of thefollowing forms:3

m1=a+exp (bc/v1) (10)

orm=exp (bc/vs) (11)

where m1 is the market share and y1 is company advertising.m is the portion of market share taken up by the 'unattached' or 'floating'

customers.These two equations give:

(m1/v1) >0; (mfav1) >0hm mj=a+exp (b); um m=exp (b)

V1-+

um m=a and him m=0v1-o v1-.o

A modified version of (10) is embodied in equation (7). Therefore, we expecta1 in (7) to have a negative sign and a2 to have a positive sign.

Equations (8) and (9) specify the advertising decision rule in terms of threefactors usually deemed to influence the level of advertising appropriations, i.e.market shares and brand and competitive advertising outlays.

To decide on the method of estimation, we apply the order and rank conditionsof identification [3]. We find equation (7) to be over-identified while equations(8) and (9) are just identified. The method of two-stage least squares (2 SLS) is,therefore, appropriate for estimating the parameters of the three equations.

The estimation is based on annual data related to all endogeneous and exoge-nous variables in the three equations:

Endogeneous Variables

m1 = market share of the ith brand in period t. Data on market shareswere given for each brand, in each case, as a fraction of total market(e.g. 0.314) in a particular year.

v1 = brand advertising in period t. The figures on advertising relate tototal advertising expenditure for the particular brand.4

Equation (10) differs from equation (11) in the term 'a' which represents the part ofpresent market share which is taken up by the loyal' or 'attached customers'. We develop thetwo relationships (10) and (il) using the properties of the lognormal distribution [1]. Forcompetitive advertising (y1) the relationship may take any of the following two forms:

m1=a+exp (ß+y/v1) (13)

or

which give

and

m=exp (ß+y/v1) (14)

(dm1/v1) <0; (m1/v1) <0

um m1=a-l-exp (fi); hm m=exp (fi).V3 V1-

It covers expenditure on advertising messages in the press (including metropolitannational dailies, metropolitan Sunday papers, women's magazines, general magazines, regionaldailies, country newspapers, suburban newspapers, special interest magazines, businesspublications and other publications), metropolitan TV, regional TV, metropolitan radio,country radio, outdoor and cinema.

ESCLATION TENDENCIES OF ADVERTISING 157

(iii) v= competitive advertising in period t. The figures relate to advertisingexpenditure for the competitive brands and have the same coverageas v1.

Predetermined Variablesv1,1 = brand advertising in period t-1. The meaning and coverage of

the data for this variable are the same as those for v.v1 . = competitive advertising in period t - 1.

P'=relative prices in period t, calculated as where Pit is theprice per unit sold of the ith brand in period t and Pit iS the averageprice per unit sold of the competitive brands in the same period.

q1,_1=quantities sold of the ith brand in period t-1.

The data on advertising expenditure were supplied by an advertising andmarketing consultant agent,5 while those on market shares, quantities sold andprices were supplied by the firms concerned in a series of interviews. Theseunpublished confidential data relates to the period 1960-76 and cover the two'biggest selling' brands in each case. Figures on advertising were deflated usingthe consumer price index.

To have an adequate number of degrees of freedom, the individual yearobservations for each brand are pooled. We hypothesize that the individualtemporal regressions for each brand constitute repeated sampling from the samepopulation.6

The statistical results for the 2SLS method are given in Table 1. The terms inbrackets are the standard errors of the estimates. Most of the estimated coefficientsare significant at the one per cent level of significance. Those coefficients markedby * are not statistically significant. The figures for 2 should be interpreted withinthe framework of the current debate on their meaning in simultaneous equationmodels. 'h' refers to Durbin's stastic for serial correlation when lagged-dependentvariables are present.

TABLa I

Regression Results1. Instant Coffee

Brand A

2_0 8697 h=1.8640= O.1255+0.4189v1._1+0.7028v1._j+0.1455(1 m1)

(0.1329) (0.2151) (0.0394)k2_0 8565 h=1.7632

Garnsey Clemenger Pty. Ltd., Marketing and Advertising Consultants, Brisbane.The ratio of squares due to this hypothesis to the residual sum of squares due to separate

year regressions yields F values of 0.794, 0.871, 1.002. 0.989, 0.785, 0.814, 1.013, 1.024, 0.928,0.936, 1.014 and 0.967 for brands A and B in each of the six products. These values aresubstantially less than the critical F values at the 95 per cent probability level. The hypothesis,thus, is accepted. For further discussion on this test see C. R. Rao [7].

loge th1 = - 1.3501 - O.OS64/vj + 0.O577/v1j - 0,5047 + 0.4330q(0.0205) (0.0166) (0.1622) (0.0917)

2._0 8967 h=1.8875= 0. 0628 + 0.238Ov1 t -1 + 0.8797v. -j + 0.2 l93m1

(0.0611) (0.2389) (0.0548)

1t= -0.2716+0.2068v1,1_(0.0643)

V11= - 1.1027+O.4009v1(0.1384)

=

158 BULLETIN

TABLE 1-continued

Brand Blog0 iîi1t = - 0.4033 - o. lo l7/v1 + 0.0924/v11 - 0.8624P + O. 1083q11

(0.0199) (0.0177) (0.2679) (0.0265)2_091O3; h=l.9340

=0.0253 + 0.1137v1,1 1 + O.8209 Vj,1_j + 0. 1040m11(0.0379) (0.1944) (0.0363)

2_08412; h=1.7522= -0.1310 + O.2235V1.t1 + 0.8307v1,1_1 + 0.2004(1 - m11)

(0.0716) (0.2755) (0.0669)2_.0 8607' h= 1.7738

2. Bottled BeerBrand A

log. = 0.0815-0.01 14/v11 + 0.0262/v11 -1 .3215 + 0.1 187q1,1(0.0025) (0.0064) (0.6154) (0.0369)

2O9O01; h=1.905= 0.0083 + 0.2038v11 -1 + 0.3559v1,1 - + 0.0804m11

(0.0692) (0.0968) (0.0239)R2=O.8232; h= 1.700

= -0.0533 + 0.3812V1,1 -1 + 1 .0G25v1_1 + 0.1929(1 - m11)(0.1069) (0.2208) (0.0507)Ç2Ø3Ø; h=l.8116

Brand Blog0 th = 0.4948 - 0.0845/v11 + 0.0108/v11 - 1 .6214 + 0.2 102q1,1 -

(0.0159) (0.0019) (O.8160)* (0.0654)12=09294; h=l.9063

= 0.1244 + 0.2667v1,1 -1 + 0.6033v1,1 -1 + 02977m11(0.0881) (0.1289) (0.1027)

2_.0 8787 h=1.8264= - 0.0460 + 0.4009v1,1 -1 + 0.9426v1,1 -1 + 0.2004(1 - m11)

(0.1322) (0.2433) (0.0625)2_.08631; h=l.8414

3. CigarettesBrand A

log0 i = - 0.1474 - 0.6633/v11 + 0.3547/v11 - 0.9526P1 + 0.3 177q1,11(0.1322) (0.0813) (0.3322) * (0.1008)Ç2_o93; h=1.9022

= -0.3130 + 0.4420v111 + 0.8256v1,1_1 + 0.331 1m11(0.1472) (0.2525) (0.1107)

= 0.8656; h = 1.7720

11=O.2266+O.2S38v,,11 + l.1753v1,11 +0.1414(1 -m11)(0.0832) (0.2843) (0.0473)

R2-08767 h=1.8080

Brand B

log0 ''l1 = 0.0091 -0.5 152/vi1 + 0.5302/v11 - 1 .440óP1 + O.3626q1,11(0.0904) (0.1044) (0.6466) * (0.1206)

Ç2...o93oo; h= 1.80041 +0.7980v1,11 +0.1395m11

(0.2090) (0.0462)20 8854 h=1.9270

1+ 1.1345v1,1_i+0.0879(1 -m11)(0.2304) (0.0305)

0.8433; h=1.7699

ESCALATION TENDENCIES OF ADVERTISING 159

TABLE 1-continuedToothpasteBrand A

log, rît, = 0.1745 - 0.37241v,, + 0.0679/y,, - 09862F, + 0.209 1q1,, -1(0.0641) (0.0148) (0.5621) * (0.0534)

h=1.8065= 0.1708 + 0.4082v,, i + 0.8566v, t -1 + 0.2266m,,

(0. 13 10) (0.2414) (0.0796)f2._0 8181 h=1.8376

= 0.1312 + 0.2558v -j + O.2687v,,,_1 +0.1626(1 - m,,)(0.0856) (0.0892) (0.0542)

2_0 8205 h=1.7960

Brand Blog0 nui,, = - O.5759-O.1658/vi, + 0.1361/v1, -O.7044'P, + 0.3O44q,,, .

(0.0334) (0.0309) (O.5840)* (0.1008)2..08971; h=1.9057

1 + 0.3106v1,,, + 03262m1,(0.0966) (0.0969)

Ç2....08735. h=1.8544- + 0.9894v,, -1 + 0.2441(1 - m,,)

(0.3423) (0.0813)Ç2Ø9; h=1.7699

Toilet SoapBrand A

log, lui,, = -0.164 - 00586/v,, + 0.0144/v,, - 0.7976, + 0.4028q,,,(0.0105) (0.0026) (0.5437)* (0.1246)

20 9292; h=1.9479i',, =0.0226+ 0.6825v,,,_, + 0.2312v,, _ + 0. 1613m,,

(0.2225) (0.0569) (0.0529)=0 9002 h= 1.9643

= - 0.0227 + O.4294v,_, + 07640v1,, j + 0.1324(1 - mit)(0.1406) (0.1841) (0.0441)

p208366; h=1.7911

Brand Blog, iui,, = 0.0806-0.08 10/vi, + 0.0128/v,, - 0.9900P + 0.2204q,,, -1

(0.0197) (0.0035) (0.6731)* (0.0512)J2._08751; h=1.8354

= 0.1976 + 0.4414v,,,_1 + 0.2109v1, + 02036m1,(0.1428) (0.0463) (0.07 17)

2..0 8357 h= 1.7538= - 0.0090 + 03375v1, 1 + 0.7881v,,, -1 + 0.2632(1 - m,,)

(0.1066) (0.l770 (0.0663)2_0 8673 h= 1.8250

Washing PowderBrand A

log, 1î1t = - 0.3060-0.112 1/y,, + 0.0750/y,, - 07384F, + 0.3177q, , .j(0.0241) (0.0174) (0.4398) (0.0877)Ç2..o777; h=l.9036

10 + 0.2435v,,,_1 + 0.3251v,,,_, +0.1416m,,(0.0737) (0.0809) (0.0454)

f2..0 8089: h= 1.7374= - 0.0502 + 0.5090v,,,_, + 07337v1,, -1 + 0.1020(1 - m1,)

(0.1704) (0.2001) (0.0334)R2=0.8432; h= 1.9257

,,=0.1810+0. l954v1,,_(0.0648)

1t = 0.1330 + 0.3235v,.,(0. 100 1)

160 BULLETIN

log. i'= 0.0569 - 0.2103/v(0 04 11)

= - 0.0303 + 0.4777v(0.1471)

=0.l571 +O.3ßS4v(0.1218)

TABLE I-continued

B?and B

+ 0.0347/va - o.9692P +O2550q1._1(0.0069) (0.5968)* (0.0780)

Ra_09217; h= 1.8310

+ 0.6291v,,_1 + O.3993mat(0.1491) (0.1277)R2 = 0.8900; h = 1.8635

- + 0.2754v1,t -1 + 0.1211(1 - m1)(0.0601) (0.0410)

R2=0.8798; h=l.8094

The regression results in Table I seem to suggest that:

The firm's competitive behaviour conforms to a Cournot reaction model [2].Competitors take explicit cognizance of each others' decisions and reactaccordingly. This is evident from the positive and significant coefficients ofthe advertising variables in the second and third equations of each modelin every case and indicates that each company tends to adjust its advertisingaccording to the past advertising appropriations of its competitors. Theadvertising reaction function of each firm, given by the second equation ofthe simultaneous model, could be used to predict the future advertisingpolicy of each brand and also to study the competitive implications ofseveral advertising strategies [9].

Brand advertising and competitive advertising are both significant deter-minants of the competitive position of firms as given by their market shares.This is evident from the statistical significance of the relevant coefficients inthe first equation of the simultaneous model. Furthermore, the resultswould seem to support the hypothesis that the market share-advertisingrelationship is a non-linear one.

Nonprice competition is a characteristic of many of the markets studied.Relative prices seem to have a significant effect on the competitive positionof the firms, only in the cases of coffee and cigarettes. This is evident fromthe size of the standard errors of the coefficients of the price variable in thefirst equation of the simultaneous model. It seems, therefore, that advertis-ing has succeeded, in most cases studied, in minimizing direct comparison ofprice.

To evaluate the role played by advertising in maintaining or expanding firms'market shares and to examine its escalation effects, we calculate the marginalmarket-share effect of company and competitive advertising (bm1/8v1 and bm/av1);the market share elasticity with respect to company advertising () and withrespect to competitive advertising (ei) and the advertising reaction elasticity (v1).The data are given in Table 2. Also given in Table 2 (last column) are the advertis-ing-sales ratios.

The data in Table 2 seems to suggest the following:

i. In most industries studied, advertising plays a 'market defensive' role.This is evident from:

The approximate equivalence of the ratios of market-share elasticitieswith respect to company advertising for the two brands to the estimatesof their advertising-sales ratios. Thus, except in the cases of toothpasteand washing powder, it can be seen that:

')AI71B- 0A/0B

This seems to conform with our third criterion given by equation (5).

The approximate equivalence of the ratios of market-share elasticitieswith respect to company and competitive advertising to the value ofadvertising reaction elasticity for most brands. Thus, except in thecases of brand A of toothpaste and brand B of washing powder, thedata in Table 2 suggest that:

-1/e1v11; i, j =A, B (ij)This seems to conform with our first criterion given by equation (3).

ii. Advertising expenditures in toothpaste and washing powder do not seemto conform with the criteria given by equations (3) to (5). The data inTable 2 shows that for the two competitive firms the advertising-sales ratiosare much greater than the ratios of their market-share elasticities withrespect to company advertising [or (OA/OB) > (iiA/B)]. Also for brand A intoothpaste and brand B in washing powder, the values of the advertisingreaction elasticity are much greater than the ratios of their market-shareelasticities with respect to competitive and company advertising.

Product

offee:

m1/Øvj mj/vj Market shareelasticity

with respectto companyadvertising

(,)

Market shareelasticity

with respectto competitive

advertising(ej)

Advertisingreactionelasticity

(y,1)

Advertisingsates ratios

A 0.1118 -0.1076 0.1467 -0.1375 1.0720 0.016B 0.1112 -0.1079 0.1876 -0.1919 0.9793 0.021

Bottled beerA 0.0582 -0.0549 0.1022 -0.1102 0.9013 0.009B 0.0586 -0.0544 0.1234 -0.1400 0.8872 0.011

:igarettesA 0.0136 -0.0121 0.1534 -0.1467 0.8547 0.048B

roothpaste:0.0139 -0.0123 0.1410 -0.1336 0.8225 0.044

A 0.1546 -0.0890 0.3462 -0.3285 1.6984 0.064B

roilet soap:0.1631 -0.0424 0.2740 -0.2893 0.9504 0.048

A 0.0628 -0.0525 0.0904 -0.0900 0.9985 0.012B 0.0625 -0.0522 0.1126 -0.1108 1.0159 0.015

Washing powder:A 0.0666 -0.0198 0.1420 -0.1324 1.0450 0.026B 0.0570 -0.0212 0.1775 -0.1696 1.6773 0.038

ESCALATION TENDENCIES OF ADVERTISING 161

TABLE 2

Market Shares' Response to Advertising

162 BULLETIN

These results support our findings in a previous study that advertisingoutlays in washing powder and toothpaste are much higher than profit-maximization would indicate. It was also found that advertising had anegative expected rate of return in one brand in each of these two products[6].

iii. The estimates of (m1/bv1) and (3m1/v) are very close for the two brandsin each industry with the exception of toothpaste. This conforms with oursecond criterion given by equation (4) and indicates that a substantial partof advertising is self-cancelling.

In addition to the above results we also compare the increase in total industryadvertising outlays (deflated by the Consumer Price Index) and that of totalindustry sales (in volume) during the period of study. In both instant coffee andbeer, total deflated industry outlays, at the end of the period under study, wereapproximately three times higher, while in cigarettes, dental, toilet soap andwashing powder, they were approximately four times higher than their level at thebeginning of the period. In no case, however, did total industry sales (in units)increase by more than 75 per cent over the whole period (1960/76). This againsupports the main conclusion of this paper that advertising expenditure in theindustries studied has a clear tendency to escalate.

CONCLUSIONS

Despite its limitations, this study has important findings:

The empirical testing of the criteria for self-cancellation by means of the useof a simultaneous equations model seems to suggest that advertising in a number ofAustralian industries is self-cancelling and escalating.

The statistical evidence suggests that firms use their advertising expenditurein protecting their market position. They react and adjust their own advertisingoutlays continuously to avoid losses in their market shares. As long as this escala-tion continues, advertising expenditure will increase without reducing the reci-procal cancellation effects. Increases in advertising expenditure for the solepurpose of maintaining market share cannot benefit the consumer; quite thecontrary.

This tendency of advertising to escalate may justify its regulation. Suchregulation, however, should not contribute to increasing rigidities within theindustry, discriminating in favour of some firms or some media or raising marketingcosts.

The study also demonstrates that competitive interaction and firms' market-ing promotional functions are best described by a simultaneous equations model inwhich market share and advertising outlays are nonlinearly related and advertisinginterdependencies are explicitly taken into account. In this respect this study maybe an advance on previous work.

5. The criteria developed in Section 1, may be generalized to test for the'informational' role of advertising. Also, the methodology used can be applied toother industries and to different market economies to test for the tendency ofadvertising to escalate.

University of Queensland

REFERENCESAitchison, J. and Brown, J. A. C., The Lognormal Distribution With Special

References To its Uses in Economics, Department of Economics, Monograph 5,University of Cambridge (Cambridge, 1957).

Cournot, A., Researches Into the Mathematical Principles of the Theory ofWealth, translated by N. T. Bacon (New York, MacMillan, 1847).

Johnston, J., Econometric Methods, 2nd ed., McGraw-Hill Book Company,(New York, 1972).

Kotler, P., 'Competitive Strategies for New Product Marketing Over theLife Cycle', Management Science, 12 (4) (1965), 104-119.

Lambin, J. J., 'Advertising and Competitive Behaviour: A Case Study'.Journal of Applied Economics, 2 (4) (1970), 230-251.

Metwally, M. M., 'Advertising and Competitive Behaviour of SelectedAustralian Firms', The Review of Economics and Statistics, LVII, (4) (November1975), 417-426.

Rao, C. R., Linear Statistic4l Inferences and Its Applications, John Wileyand Son, (New York, 1965).

Urban, G. I., 'A Mathematical Modeling Approach to Product LineDecisions', Journal of Marketing Researèh, 6 (Feb. 1969), 40-47.

Weinberg, R. S., An Analytical Approach to Advertising ExpendituresStrategy, Association of National Advertisers, (New York, 1960).

Weiss, D. L., 'Determinants of Market Structure', Journal of MarketingResearch, 5, (1968), 290-295.

ESCALATION TENDENCIES OF ADVERTISING 163