Models of New Product Diffusion Through Advertising and Word-of-Mouth Joe A. Dodson, Jr.; Bitan Muller Management Science, Vol. 24, No. 15 (Nov., 1978), 1568-1578. Stable URL: Ittpslinksstor.orgsicisici=0025-1909%281978 115429243 15%3C156853 MONPDT%3E2.0.CO%B21 Management Science is currently published by INFORMS. Your use of the ISTOR archive indicates your acceptance of ISTOR's Terms and Conditions of Use, available at flip: feworwjtor org/aboutterms.htmal. ISTOR's Terms and Conditions of Use provides, in par, that unless you fave obtained pcior permission, you may not dowaload an cnt isus of @ journal or multiple copies of articles, and you may use content inthe ISTOR archive only for your personal, non-commercial uss. Please contact the publisher cegarding any further use of this work. Publisher contact information may be obtained at bupsferwer,jsto.org/joumals/informs. btn. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transtnission. ISTOR is an independent not-for-profit organization dedicated to creating and preserving a digital archive of scholarly journals. For more information regarding ISTOR, please contact support @jstor.org- hup:shrwwjstor orgy Mon Nov 14 12:41:20 2005vot A Ne is Novena STE MODELS OF NEW PRODUCT DIFFUSION THROUGH ADVERTISING AND WORD-OF-MOUTH* JOE A. DODSON, JR.} xo EITAN MULLER ‘A model ef the difusion process developed which recognizes (I) the iteration between adopters and roneadaples 2nd (2) the influence of external information sources such 28 sdvertsig. The mode i extended by woorparating the eects of sepeat purchasing The ‘models writer by tbe fllowing authors ae shown to be special eases of this model: Gould, Nerlove and Arow, Vidale and Wolfe, Palda, Bas, Nicosia, and Glast. The behavioral sssumptiogs which support the model are made expliit 12d the implications of these ‘sssumpuos far the shape of the new product gawth curve are dedved. (MARKETING; MARKETING SUYER BEHAVIOR; HEALTH CARE—EPIDE. MIOLOGY) 1. Introduetion ‘A manager seeking fo introduce a new product into the marketplace has a limited number of variables under his control. The marketing manager must understand how these decision variables impact the diffusion process if he hopes to use them effec- tively. The theory of adoption of new procucts by a social systems bas been reviewed bby Rogers (17). These ideas have been expressed mathematically in diffusion models which emerged early in epidemotogy (1), (2), and (11), A general model of the diffusion process which explicitly deserines the influence of advertising and word-of-mouth is presented in the next section, § introduces che repeat purchase mode! and 4 shows that it generalizes several models such as those of Nerlove and Arrow [13], Gould [8}, Vidale and Wolfe (22), Palda [16}, Bass (3), Nicosia {14}, and Glaister (7) 2. A General Diffusion Madel For Durable Products, ‘This section introduces a model which can be used to predict industry sales of a durable product, Let the number of people in the market, V, be divided into x(*)}—the aumber of people who are unaware of the existence of the product, p()—the number of potential customers who are aware of the product but have not yet purchased it, and 2(2)—the number of current customers who have purchased the product. By definition x() +209 + 20) = NO o “The variables x, y, 2 represent states in the diffusion process. At any point in time & consumer will be in. one of these states,’ A behavioral conceptualization of the progrest of 2 consumer from unawaseness through attcude change to ultimate purchase has been given by Lavidge and Steiner (10). ‘One variable influencing the movement of consumers through these states for a new product is the information acquired from contact with prior purchasers, ie, word-of- mouth (23), Early adopters of a new product or new idea interact with other less + Accepted by George H. Haines, J rezeived Apel 29, 1976. This paper has been withthe suthors 11 month fr 2 revisions YN, W. Ayer ABH International, New York + Univesity of Peonsyivania "This representation is similar to but diferent from that proposed by Urban (211, This formulation ‘excludes the possitilty of purchase without knowledge of the prod’ existence asthe cae wit imple fouying 168 025-1909 /75/2415/1S68SO.25 onyrant 908, Toes cf Managment SenetMODELS OF NEW PRODUCT DIFFUSION 1569 innovative members of the group. The potential adopters may be influenced in their purchase timing by the early adopters of the product, Also a certain proportion of potential customers may purchase the product independently of the influence of ‘word-of-mouth, This inducement to purchase will represent the effects of distribution, promotion, advertising, and other forms of marketing effort, ‘The general model can be formulated as HD = — Ber) + 20) ~ wee @) PEO) = BX + 29) + Hl) — Os @ 2M= v0) a) where « dot above 2 variable denotes differentiation with respect (o time, B reflects the impact of word-of-mouth, and sand + reflect the effects of the marketing efforts of the firm, More precisely, the explanation of this set of equations is: (a) The people who know. y(2) + 7(2), contact and inform a total of BCv(A + 264), out of which only 2 fraction x(1)/N() are newly informed. Thus 6 = B/N(2). Tn addition, ovt of the total number of people informed through advertising wN(0), only 1 fraction x(A)/ N(0) ate newly informed (b) The number of people who know but did not yet buy is increased by those newly informed fx(A(p(2) + 2() + xt) and decreased by those who buy (2). (©) The number of people who buy the product is »p(). Note that in the case of a durable, if we assume that cach consumer buys exactly one unit, equation ¢c) describes the sales, ic., denoting the sales by s(i), we have s(2) > #(1). This description is similar to that used by Gould (8), It represents a more realistic ‘extension of the simpler cases considered by Gould since it includes both the effect of word-of-mouth and the effect of ather sources of information, ez, advertising. Its relation to Gould's motel will be highlighted later. A similar model was presented by Bernhardt and Mackenzie (4). The model is developed in more detail here to highlight the influence of advertising and word-of-mouth on the diffusion process, to clarify ts felationship with previous work in marketing, and to lay the groundwork for an extension of the model to situations which involve repeat purchasing, Consider the case where there are a fixed number, N, of potential cusumess of & product? After buying the product, a purchaser is removed from the group of potentia! consumers. This is a restrictive assumption which does not allow for repeated purchases, Nonetheless this mode! may be oredictive of the diffusion of such products as color television sets, refrigerators, washing machines, and other types of consumer durables in their early stages. The time frame over which sales would be predicted would have to be shorter than the replacement cycle for this model to be considered descriptive of such market. And, the size of the market N’ would have 19 be stable. Full discussion of the model is postponed until repeat sales is introduced in the next section, However, it is useful at this point to consider some special cases of the model in order to clarify its relationship to carlier work and to make explicit the impact of the firm’s decisions on the diffusion process. (A) Case L. Consider first the case in which (1) = 0 for al, This might be the case when the new introduction provides such a significant improvement over existing alternatives that everyone who becomes aware of its existence adopts it. (2) becomes £(O) = — Be(Q2(1)— wx(), @) Suggestions for incorporating a constant rate of entry and ext feom the population of potential comumers in reduced wertiont ofthe madel ate presented by Herthaedt and Mckenzie [41570 JOE A. DODSON, JR. AND BITAN MULLER ‘Substituting x() + 2(9 = ¥ into (5) yields 2(9 = BEN = 2(9)2(0) + HON ~ 2(9) 6) Since 4(¢) is the tate of sales at time ¢, s(4), the last equations can be written as s(t)= a’ + Bi2() + 240), ey where a’ = pi, b'= AN ~m, and c= —B. This is equivalent to the equation esti mated by Bass (3], Bass’ coefficients of innovation and imitation are equivalent to and f. Bass showed that a discrete analogue of equation (7) gave good predictions of initial purchase timing for eleven consumer durables, “The general solution of (6) is, 2(0) = (= exp(—pi))/(1 + BN exp(—o9)/1) ®) where = N+ g. This solution provides insight into the relationship between the penetration curve and the relative values of 6, the contact coefficient, andj, which represents the influence of a firm's promotional activities. When & > u the growth in penetration follows the penetration curve shown in Figure 1. However, when 6 < i.e, when promotional activities domingte the marke: conversion mechanism, then penetration follows the curve shown in Figure 2. 200) te FaUne 1, Penetration Cove When 8 > FrouRE2 Penetetion Cave When bp ‘To find the time ¢*, at which the sales rate ceaches its peak, we differentiate s(¢) and set it equal to zero, yielding w= (I/o)ln( BN Ju) and s(e*) = 97/48. ° Since p= BN +, the following observation can readily be made. The higher the advertising effort 1, the sooner the peak will arrive and the larger the peak will be. (B) Care Ul. The second case considered is one for which w= 0 and 2(0) € y(2). ‘This oceurs when the number af people who actually buy the new product is much smaller than those who know about tke product but did not yet buy. This may be representative of @ new introduction which faces considerable resistance to tral. Equation (3) then becomes (N= ~Bx(DO) cc) which gives? (= aN vscexp(— B/C) — 1200) ay SGenecl relerences which describe the mathematics of dfleential models include Lotka {1} and Ross us,[MODELS OF NEW PRODUCT DIFFUSION isa This equation is a generalization of the Bass model since approximation of the ‘exponcttial function by a second degree polynomial will produce an equation which Js equivalent (0 (7). ©) Case II, The assumption that y(1)=0, ie. everyone who becomes aware of the product adopts it is combined with the assumption that there is no inthience from prior adopters, ic, 6 = 0. 3) then becomes = —Ex(9) 2) that i, a constant proportion of the potential consumers purchase the produc! each period. This implies (1) = NCL ~ exp(— utp) (13) given that x(0) = N. This type of exponential growth to some asymptote as well a5 a modified version allowing for growth in N was utilized in the sales models of Fourt and Woodlack (6) Note that equations (8), (11) and (13) were derived under completely different assumptions and thus are applicable to different situations, Each represents a special case of the model developed from equations (2), (3) and (4) which includes the effect ‘of word-of-mouth and advertising within 4 multi-stage model of the adoption process. Note also that the models presented here are only concerned with the timing of initial purchase. The next section extends the general model to incorporate repeat purchas- ing. 3. Growth Model with Repeat Sales The discussion 10 this point has concentrated on a model which cepresents the growth in first purchases of 4 product. It is worthwhile to consider how one might Incorporate repeat sales into the model of new ptoduet penetration. Repeat sales becomes an important consideration when one is dealing with low-priced, frequently- purchased, branded products, or durable products for a long enough time frame for repeat sales ta become a significant proportion of (ata sales. Let x(2), 40) and 2(0) be defined as before The new set of transition equations is siven by EC) =~ BRN ¥)— eC ACD ~ XC) ca) AQ = BECK ~ 40) + C0) — Cr AN + He 5) FE) = wl) + O(N, (16) where is 2 forgetting parameter, 4 is a switching constant, ic, the rate at which consumers purchase rivals’ brands. The flows are summarized in Figure 3 Note that the sales rate, denoted by (4) is given by S(O) + HO ay where 7 is the repeat purchase parameter. Sales result from trial by new customers and repeat purchasing from current customers. The previous model is extended (0 include repeat purchasing, 72(2), and switching 82(/). Thus the model need no longer be restricted to situations involving a single adoption. This extension enables us to use the mode! to describe the diffusion process for frequently purchased goods and to extend our prediction for durable goods further into the life cycle, where. repeat ‘The definitions of a curren customer bas to be somewhat changed to take inta account the possibility ‘of swirching. Thus we dele z fo be the number of customers who ate auc of the produet aod whose lst Parcbase was of the firm's brand.» a8 before, cam be defined a8 N = 2 = 21372 JOE A. DODSON, JR. AND EETAN MULLER purchasing occurs. These extensions also indicate the need for & forgetting parameter, 6, since it represents forgetting of a specific brand. (0 wsinformes pr being fntormed Dy advertising =p by word af mouth - 4 —— (0 potent) customers purchasing ftw’ beand = sale’ brands - ee ee 2(0)~ eutent customers —! repurchase -F should be noted that chere enas a closed Joop (ram y to itselt due to purchase of rvs brands 1s ignoced here sce i dacs not affet the ansivion equations Frovre 3 Although a closed form solution of (14)-(17) is not possible to achieve (because of the appearance of an unintegrable term in the solution) it is possible to characterize the solution by means of phase diagrams. This is done in the appendix, The main result is summarized by the following proposition. PRopasrti0N. The solution of equations (14)-(17) is monotonie increasing in the number of informed persons (M ~ x) and the number of current customers 2, and is either ‘monotonic increasing oF single peaked in the potenual customers y and the sales s. Examples are provided in Figures 4-7. Figure 4 illustrates the growth in sales for a product where the contact coefficient A is relatively “small” and the repeat purchase tate 7 is large relative to the trial rate y, ie, 7 > y. The definition of a “small” or “large” contact cocfficient is given in the appendix. It should be noted that all the 30 Fioune 4 Sales When the Repeat Purchase Rate Is Larger Than the Trial Rate and the Contact Coeticien ls Relatively SmMODELS OF NEW PRODUCT DIFFUSION 1373 parameters enter the solution, but the repeat rate 7 and the contact coefficient fare the critical determinants of the success of a new brand, the timing and the maximum sales level the brand will achieve, When the contact coefficient is relatively “large” ‘and the repeat rate is larger than the trial cate the sales growth is plotted in Figure 5 ‘When the tial rate is larger than the repeat rate the sales curve will peak and then fall below the peak to some lower level of sales, Figures 6 and 7. The long term stable level of sales is a function of the relative values of the repeat rate } and the switching rate 6, «9, Froune 5. Sales When the Repeat Purchase Rate ls Lager Thaa the Trial Rate 20d the Contact (Coeticient Us Relatively Large 6, Frcume 6. Sales When she Teal Rate Is Larger Than the Repeat Rate and the Contact Coeiient Is Relatively Smal, 10 Figune 7. Sales Whes the Toal Rate Is Larger Thaa the Repeat Rate and the Contact Coetiient Is Reatively Large.1s JOE A. DODSON, JR. AND EITAN MULLER A closed form solution to equations (14)-(16) can be obtained for special cases of the model but this requires some assumptions about the values of the model's parameters. Dodson and Muller [5] derive solutions for some special cases of this model and illustrate a procedure for obtaining empirical estimates of these models’ patameters. The latter can also be found in the unabridxed version of this paper. The relationship between the diffusion model developed here and earlier modeling of the effects of advertising car be seen by considering the special case where the trial cate and the repurchase rate are equal, = y= 47, and B= 0=0. This represents 2 situation where awareness is generated directly from advertising, Assume the effect of competitive activity is constant and is reflected in the cepeat purchase rate 7. Then ‘equations (14)-(16} become (97 = x 6(N~ XC), (18) FOD= Y= FF 9H, co) 20 = PUN 82) 20) and sales is given by 34) = F(N — x) (Ql) Differentiating with respect to time and substituting from (18) yields S(Q)= CIN = 9(0)) = a4) @ Since js represents the rate of conversion from unawareness fo awareness, i.e, movement from state x 10 state y, it is naturally a function of the firm's advertising expenditure. If sis to be linearly related to advertising costs, then (22) is equivalent to the Vidale-Wolfe model (22] with the appropriate advertising response. constant Horsky [9] has shown Palda’s model [16] to be a special case of Vidale-Wolfe's formulation. To see the relation to Gould's model, one has to use Gould's definition of goodwill denoted by G, as the number of people who know of the product, ie, G(s) = N — 2(9, Substituting that into (18) yields GO = WN G6) ~ $6 3) This equation is the first diffusion model discussed by Gould [8], From (23) and (22) it is clear that the dynamics of the models of Gould and Videle and Wolle ace the same and are equivalent to the reduced model (I8}-(21). The fact that the models of Gould and Vidale and Wolfe are equivalent was also found independently by Sethi [19 While the “first” diffusion model of Gould ignores the influence of word-of-mouth, Gould discussed a second diffusion model which assumes that the firm has direct control over the mecharism of word-of-mouth, and it is the only mechanism of information diffusion. Both assumptions seem unrealistic. At best the control of the firm is very indirect, for example showing women talking about deodorant and washing liquids in the hope that this will indeed induce women co do just that. [¢ seems more reasonable to expect the word of mouth mechanism to be a comple- mentary one to the direct message sending mechanism. Both mechanisms were taken into account in the general diffusion model above. Gould recognized te fact that his diffusion processes cepeesent extreme cases (8, p. 368]. This fact, however, was ignored in @ later paper by Glaister (7), Althougit his contention that there must be a critical mass or a threshold phenomenon is appealing and certainly imaginative, itis based on an erroneous notion that the only mechanismMODELS OF NEW PRODUCT DIFFUSION 1575 at work is word-of-mouth. A formal proof of that fact (together with a discussion on skimming pricing policy) is given in Muller {12}. To obtain Gilaister’s model from the general diffusion model, rewrite equations (I4)-(I6) with w= $= = 0=0, and further assume that the current customers’ group is negligible with respect to the potential customers group, that is z 7, sis monotonic increasing. 4 =0is x= (V~ 9G - AX), Bais Nya (rte + Oz: xe checking § at = 0, we find that since x(0) = N,# <0 if and only if 8 <1/N[y + 9+ b= 7h, We denote this case by a “relatively small” contact coefficient. From the y:x space itis clear that y is either monotonie or single peaked. $= 0's nx = (V — x9 ~ fx), Pa Ois (+4 + Bly = ARN — x) + wx + O(N — x), Lastly we have to prove that if s 15 not monotonic then it can orly be single peaked (and not, say, with « max. and a min, or multi-maxima.) 9 See Nicosia, p. 28,MODELS OF NEW PRODUCT DIFFUSION 1577 ‘The sales can be written as SEC DIT TN = 2) Differentiating twice we get that when #=0, 3 = k{y(B(N — 2x) + p— 0+ 700 — ‘y)). At the first extremum # <0 since this extremum must be a maximum (since tially sis increasing). Since i <0 the expression inside the curly bracket is positive when $=0, At any subsequent time, since <0 this expression increases, Thus if #50 again, # <0 which implies another maximum. This of course is impossible since there must be a minimum in between. After the first maximum, therefore § = 0 only if O which is at the steady state” QED. We wish :0 acknowledge our appreciation for helpful comments by Nancy L. Schwartz and (0 anonymous teleyees on earlr versions. Mest of this wort was done while the authors were at Northwestern Universi. This is'an abridged version of the paper. Anyone who wishes to oblain the unabridged version ‘which includes empirical ertmation can do sa (at cos) torough the TIMS office, 146 Westminster St, Providence, RL 02903, References Bauer, N. TJ, The Mathemaical Paory of Epidemics, Halner, New York, 1957 2 -Bawrtert, M.S, Stochae Population Madels im Ecology and Eperutogy, Metheun, London, 1960, 3. Bass, F. M, “A New Product Growth Model for Consumer Durables” Management Se, Vol. 15, Ganuary 9), pp. 215-227, 4. Brroansanr, | akD MscKEN7Ie K,, “Some Problems in Using Diffusion Models for New Products” ‘Management Sci, Wol. 19 (Octobe: 1972), pp. 187-200, 5. Donsos, A.J. aud Munien, E, "Models of Product Ditfusion Trough Advertiing and Word of Mouth” Disewsion Paper No. 182, The Center for Matherateal Studies in eonomies and Managemect Scicnos, Northwestern University, Apel (976 6 Four, LA. am Wooniock, £ W., “Early Prediction of Market Success for New Grocery Producis* J. Marketing, Vol. 26 (October 1960), 1. Guassre, 8, “Advertsing Policy ad Returns to Seale,” Econsmica, Vol. 41 (May 1974) pp. 139-156, 8 Gown, 5 F, “Pilluson Proceses and Optimal Advertsing Policy.” in ES. Phelps etal, editors, ‘Mieroaconamie Foundation of Employment and Inflanon Tred, Norton, New York, 1970 9. Hoasey, D., “Empirical Analysis of Optine| Advertsing Policy," Managemen Sel, Vol. 3 (June 187, pp. 1037-104,1578 JOE A. DODSON, JR, AND EITAN MULLER. v. a 13 1 2 Lavapae, RJ axp Srama, G. A, “A Model for Predictive Messurements of Advetising Eifece tienes" J. Marketing, Val, 25 (Octobe 1961), Loma, A.J, Blemants of Mashematea! Biology, Dever, New York, 1956 Moun, E, “On Difusion Process and Skiouming Pricing Policy," Center for the Study of Organizational fenovation, Uriv. of Pennsylvania, Naverbe: 1977 Nswovs, M, ano AnRow, K. J, “Optimal Advertsing Policy under Dynamic Conditions” Economica, Vl, 29 May 3962), pp. 129-142. Nicosia, FM, Consumer Decision Procewer, Prentice-Hall, Englewood Clits, N. 1966 ‘G26, $, “Impertect Markets Through Lack of Knowledge,” Quart J. Ezanan., Wo. 74 (February 1960), Fo. 29-52. Papa, K.S., "The Measurement of Cumula pp, (62-179 ROGERS, EM, Diffston of Imocatins, The Fre Pree, New York, 1982. Ross, 8. Ly lntraductiont9 Ondnaey Differential Equations, Bissell, Landon, 1966 Sern, §, B,, "Dynamic Optimal Conirol Models in Advertsing: A Survey,” SIAM Reo Vol. 19 (eober 1977). StioLin, G. J, "The Economics of Information,” J. Pattie! Beonamy, Vol. 8 (Jone 1961), pp 213-235, Uneax, ©, L, “SPRINTER MOD III: A Model for the Analysis of New Frequenlly Purchased CCapsimies Ploduets," Operatins Res, Vol. 18 (September 1930), pp. 805-854 \Vionte, M. Lan Woure, H. B., “An O.R. Sul of Sales Response fo Adverusing," Qporavions Res, Vo, 5 Gune 195M), pp, 370-381 ‘Warrte, W. H, Jn "The Web of Word of Mouth,” im Lincoln H. Clark (e.), Consumer Sehasiar, Vol 1, 1958 Advertsing Effects” J Busse, Vl 38 (April 1968)