A Model of Communication Impact and Consumer Response

Jack Healey, University of California, Los Angeles
[ to cite ]:
Jack Healey (1974) ,"A Model of Communication Impact and Consumer Response", in NA - Advances in Consumer Research Volume 01, eds. Scott Ward and Peter Wright, Ann Abor, MI : Association for Consumer Research, Pages: 384-392.

Advances in Consumer Research Volume 1, 1974    Pages 384-392

A MODEL OF COMMUNICATION IMPACT AND CONSUMER RESPONSE

Jack Healey, University of California, Los Angeles

[The author expresses his appreciation to Jules Berman & Associates and their advertising agency, Eisaman, Johns & Laws, for providing the data for this research.]

[Jack Healey is a Doctoral Candidate at the Graduate School of Management, University of California, Los Angeles.]

Considerable work has been done in the area of consumer information processing in an attempt to develop a process model of communication impact and consumer response. Several such models have described the purchase process in terms of stages of commitment. McGuire (1969) posited a chain of six behavioral steps, from presentation, to attention, to comprehension, to yielding, to retention, to behavior, each probabilistically linked to the preceding one; an individual was postulated to pass through these steps if he was to be effectively persuaded. Essentially, this process was a form of Markov chain and is similar to the hierarchy of effects model originally propounded by Lavidge and Steiner (1961) and the DAGMAR model proposed by Colley (1961). In their model for predictive measurements of advertising effectiveness, Lavidge and Steiner perceived advertising as a force which moved people up a series of steps, from unawareness, to awareness, to knowledge, to liking, to preference, to conviction, and finally to purchase. Colley proposed there were five steps in the process, from unawareness, to awareness, to comprehension, to conviction, and finally to action. Palda (1966) questioned Lavidge and Steiner's assumption that movement up each step of the hierarchy increased the probability of purchase on the part of a consumer. Palda's testing of the hierarchy of effects model led him to conclude that higher awareness did indeed coexist with higher purchasing rates but that there was no causative link, as Lavidge and Steiner had theorized.

One of the most important contributions to this stream of development in the area of consumer information processing has been the Howard-Sheth Model of Buyer Behavior (1969). A comprehensive theory, this model was designed to take into account the various behavioral, situational, and economic variables affecting consumer decision processes. Buyer behavior was postulated to be caused by a network of interrelationships among endogenous variables such as brand comprehension, intention, etc., and exogenous variables such as a buyer's personal and social characteristics. The model included a feedback mechanism to account for repeat buying behavior. Farley and Ring (1970) tested the Howard-Sheth Model using a multiple equation regression model on purchase data for a new grocery product. The results of their two stage least squares estimates of the model's parameters were weak, causing Farley and Ring to conclude that improved data would be required for more precise work with this type of model. In the discussion below, when the Howard-Sheth Model is referenced, the Farley-Ring version is being referred to.

In addition to the research conducted on stage of commitment models, other studies have focused on specific functional relationships between variables in the consumer purchase process. Bird and Ehrenberg (1966), through analysis of consumer panel data, found awareness of a product to be a function of current usage. Bird and Ehrenberg (1966, 1967) also found intention to buy a product to be a function of current usage. In both studies, however, their cases consisted of data on product classes rather than on individuals.

The purpose of this study is to synthesize the two streams of thought discussed above into a new model and test it using consumer purchase data on a new wine product. Specifically, this research will be designed to re-examine a subset of variables of interest in the Howard-Sheth Model, integrating the notions of Bird and Ehrenberg! In so doing, it will serve to replicate the Farley and Ring findings and provide needed additional testing of the Howard-Sheth Model. This is important because of the prominent role played by this model in current consumer theory. In addition, this model will attempt to validate the applicability of Ehrenberg and Bird's findings to individual versus product class data.

A PROPOSED MODEL OF PURCHASE BEHAVIOR

The three major elements to be studied in this model are awareness/ effect of advertising, intention to buy, and purchase/current usage of the product being studied. A set of hypothesized interrelationships among these elements and other phenomena will be discussed.

Awareness of advertising is the first major element in the model. Strictly speaking, awareness itself was not used, but an advertising awareness/effect variable, as discussed below in the intention model. Consistent with Bird and Ehrenberg's feedback notion, it is hypothesized that the higher the level of current usage, the higher the level of awareness/effect of advertising. In addition, it is hypothesized that the higher the level of past usage in the product class, the higher the awareness/effect of advertising, due to the operation of a selective perception process. Higher education level is also-hypothesized to lead to higher awareness/effect of advertising, since the higher the education level the more sensitive the respondent's perceptual processes to new stimuli. It is also hypothesized that the higher the income, the higher the awareness level because the higher income may foster a higher interest level in new product information. Age, however, is considered to have no affect upon awareness because the sensitivity of a respondent's perceptual processes to new information is not dependent upon his age.

The second major element to be examined in this model is intention to buy. In both Colley's and Lavidge and Steiner's theoretical models, intention to buy was preceded by an affective element, providing an attitude measure as a link between awareness and intention to buy. Because the data used to test this model did not include an attitude measure, a modification in the model was made; an intervening variable, a measure of the effectiveness of the advertising, has been combined with awareness to yield a composite awareness/ effect measure. Hence, it is hypothesized that higher awareness/effect of advertising leads to stronger intention to buy. This hypothesis is consistent with Lavidge and Steiner's theory of hierarchy of effects as well as Howard-Sheth's Model of Buyer Behavior. Using Bird and Ehrenberg's findings as the basis of a relationship existing between current usage and intention to buy, it is hypothesized that the higher the current usage, the stronger the intention to buy. Past usage of the product class is not included in the model as a variable affecting intention to buy because consumers already using other brands in the product class might not be likely to express an intention to buy a new brand. Because the model is being tested with data for a new wine product and it is posited that younger people tend to express greater interest in buying wine, it is hypothesized that the lower the age, the stronger the intention to buy. Higher income level is also hypothesized to lead to stronger intention to buy, since higher income should provide for a greater capacity to express an intention to buy a product. Education, however, is considered to have no affect upon intention to buy, since expressing an intention to buy is not dependent upon the level of cognitive development.

FIGURE 1

A MODEL OF COMMUNICATION IMPACT AND CONSUMER RESPONSE

Finally, the third major element of the model to be examined is purchase/ current usage. This variable is defined in the model by the number of purchases of the new wine product in the first three months following its introduction in the test markets. Lavidge and Steiner, Colley, and Howard and Sheth propounded that greater intention to buy will lead to greater purchasing. Therefore, it is hypothesized for this model that stronger intention to buy will lead to a higher rate of purchase/current usage. Further, because past purchase behavior of wine should indicate interest in the product class, it is hypothesized that higher past usage is related to purchase of a new brand in that product class. Age and education are considered to have no effect on actual purchase/current usage of the product. Income, however, is hypothesized to be positively related to purchase, with higher levels of income leading to higher levels of purchase/current usage.

The structure of the proposed model is shown in Figure 1, with the direction of influence indicated by the arrows. The sign of each relationship is also shown and the B and Y terms are discussed below. The major goal in testing this model is to determine whether the proposed structure shown in Figure 1 is sound, i. e., whether the constructs interrelate in the hypothesized manner. The model is merely a vehicle for examining how consumers process information. To carry out this examination, empirical testing must be done. Hence, in the next section, the measurement of variables and the structuring of the model for testing by a simultaneous equation regression system are discussed.

METHODOLOGY FOR MODEL TESTING

The model was tested with data collected on a new rose wine product. The sample consisted of 308 rose wine purchasers in two test markets, 150 in Hartford, Connecticut and 158 in San Diego, California. Interviewing was conducted three months following the new product's introduction via television, radio and newspaper advertising in both markets. The measures below relate to that new brand.

The awareness of advertising measure ranged from 0 to +3, and was the number of media in which the new brand advertising was recalled. The effect of the advertising on respondent's desire to purchase the product was measured among those recalling the advertising, with responses of increased desire, no effect, and decreased desire scaled as +1, 0, and -1 respectively. A compound measure of advertising awareness and the effect of advertising was developed through multiplication of the awareness and effect scores, with the resultant scale ranging from -3 to +3, The function of this composite measure was to incorporate an affective element into the model.

The measure for intention to buy the product in the future at a given price ranged from +4 (definitely will buy) to +1 (definitely will not buy).

Purchase/current usage of the new product was measured on the basis of the number of purchases made in the three months since its introduction, scaled from 0 to +4.

Past usage of the product class was meas d in terms of frequency of consumption, with the results of two separate questions on past usage averaged for greater reliability and scaled from +1 to +3.

Age was coded in 5 year intervals, income in $5,000 intervals, and education in terms of the last level of school completed. Some of these measurements above may have been ordinal (e.g., education), but the underlying ratio or interval data for these measures (e.g., years of school) was not available, given the coding of the original responses: The variables above were used, with the realization that the results must be tempered in light of the scaling properties of the data.

Given these variable measures and the model structure shown in Figure 1, the model was tested using a simultaneous equations system. The variables are divided into two sets for this purpose. The endogenous variables are those determined within the system and the exogenous variables are those which influence the system but are not influenced by it. For purposes of this study, the endogenous variables are advertising awareness/effect, intention to buy, and purchase/current usage. The exogenous variables are past usage of the product class, age, education and income. In econometric notation, they are as follows:

Endogenous Variables

Y1 = Advertising Awareness/Effect

Y2 = Intention to Buy

Y3 = Purchase/Current Usage

Exogenous Variables

X1 = Past Usage of the Product Class

X2 = Age

X3 = Education

X4 = Income

The model in Figure 1 can be translated into simultaneous equation form as follows:

(1) Advertising Awareness/Effect:   Y1 = Y10 + B13Y3 + Y11X1 + Y13X3 + Y14X4 + U1

(2) Intention to Buy:   Y2 = Y20 + B21Y1 + B23Y3 + Y22X2 + Y24X4 + U2

(3) Purchase/Current Usage:   Y3 = Y30 + B32Y2 + Y31X1 + Y34X4 + U3

In this notation, Bij refers to the effect of endogenous variable j on endogenous variable i; Yij refers to the effect of exogenous variable j on endogenous variable i. Yi0 is the constant term and Ui is the error term for equation i. The predicted signs for the coefficients are shown in Figure 1.

Prior to estimation of B and Y coefficients, the system must be checked for identifiability. Both the order and rank conditions are met for this system (Wonnacott and Wonnacott, 1970, pp. 180, 350), with order conditions showing that the system is over identified. Two stage least squares (2SLS) was used to develop the estimates (Wonnacott and Wonnacott,1970, pp. 190-192). These estimates were obtained for each of the two test markets studied as well as for the sample as a whole.

The proposed model assumes linearity. While some of the findings of Bird and Ehrenberg suggested nonlinear relationships between several of the variables of interest, experimentation with nonlinear relationships in this model did not seem to improve the model's performance. Therefore, proposed nonlinearities were not considered further.

RESULTS

The results of this model testing, in terms of the signs and significance levels of the estimates, are summarized in Table 1. Because the major concern of this study is to examine the structure of the hypothesized interrelationships, determining the direction of influence among the variables is the most important task. Therefore, the signs and significance levels of the variables in the various equations are more crucial than the particular magnitudes of the coefficients.

Initially, because there were differences in the levels of the exogenous variables in the two test markets, it was decided that the sample should be split and the model tested in each market individually. The results of such testing indicated, however, that the differences in levels did not affect the direction of influence among the variables; all but one of the correlations were in the same direction. Therefore, since this model is concerned with the direction of effects rather than with levels, data from the two test cities were combined in an attempt to obtain better parameter estimates because of the larger sample size. The lack of a significant difference between the two test markets in each of the simultaneous regression equations was further verified through application of Chow's test for equality of coefficients in regression equations (Kmenta, 1971, p. 373). The findings about the structure of the model will hence be discussed only for the model run on the sample as a whole. The total sample size is 254 rather than 308 because of missing values for some of the responses, notably income.

Advertising awareness/effect equation: Estimates for Y11 and Y14 both have the predicted sign, but are significant only at the .25 level. Thus, the greater the past usage of the product class and the higher the income, the greater the awareness/effect of advertising tends to be. Results for B13 and Y13 were not significant. For this equation, R2 was found to be highly significant, at the .001 level. All four hypotheses for the signs of the coefficients must be rejected, using the standard .05 significance level.

Intention to buy equation: Of the four coefficients estimated, only B21 had the correct sign and was found to be significant, though only at the .25 level. The estimate for Y24 was also significant at the .25 level but in the opposite direction from the predicted sign. Results on B23 and Y22 were not significant. The R2 for this equation was significant at the .05 level. Again, all four hypotheses are rejected.

Purchase/current usage equation: Estimates for B32, Y31, and Y34 yielded no statistically significant relationships, and the R Square for the equation was only significant at the .25 level.

In the three equations tested, none of the individual parameters were found to be statistically significant at the accepted levels. In all cases but one, however, the signs of the individual correlations of the variables in each of the equations with the dependent variables were in the same direction as the predicted sign, although the signs of the coefficients in the equation may have been in the opposite direction, but not significant. It may be concluded, therefore, that the direction of influence among the variables tended to be as predicted, but the magnitude of influence was small.

TABLE 1

SIGNS OF MODEL COEFFICIENTS

SUMMARY AND DIRECTIONS FOR FUTURE RESEARCH

Overall, the purchase behavior model tested in this research, combining the notions of Howard and Sheth and of Bird and Ehrenberg, did not successfully fit the data. There are several improvements which should be made in both the model and the data for future research.

The results of this testing suggest problem areas similar to those encountered by Farley and Ring. In the second stage of their 2SLS analysis, Farley and Ring attained R2 Of .014, .303, and .088 on their measures of attention, intention to buy and purchase, respectively. On similar measures, R2's of .07, .046, and .024 were achieved in the present model, roughly comparable to the Farley and Ring results except for intention to buy. Shortcomings shared by both studies include the use of cross-sectional data instead of longitudinal data and insensitivity of the measurements and scaling techniques used. In addition Farley and Ring recognized that the demographic exogenous variables performed very poorly and that better measures of exogenous variables are needed.

In the first stage of the 2SLS analysis of the present model, estimation of the endogenous variables using the exogenous variables, the significance levels were quite low, indicating poor correlation between the exogenous and endogenous variables. These estimates of the endogenous variables are then used in the second stage. By examining the correlation matrix, it is clear that had better estimates of the endogenous variables from the first stage been achieved, higher R2 values in the second stage would have resulted. For example, the simple correlation between awareness/effect and intention is .382; the correlation between the first stage estimate of awareness/effect and intention drops to .178. The difference in R2 from considering these two figures alone is .146 - .032 = .114.

Hence, poor predictive performance in the first stage tends to lead to poor performance for the model as a whole. But this poor correlation is to be expected, given the results of past studies. Demographic, socio-economic, and even personality variables have been shown over and over again to be poor predictors of behavior and phenomena 'internal' to the consumer, such as awareness and intentions (Kassarjian, 1971). Bettman (1973) points out that if predictions are to be improved, internal variables (in particular, variables related to the buyer's perceptions of the product class or brand) must be used as exogenous variables, rather than gross, 'external' kinds of variables such as those used in this study and in Farley and Ring. He shows that such a perspective leads to vastly improved results in predicting private brand purchasing, an area where demographics and socio-economic variables had done very poorly. This point is crucial for testing models of the sort proposed in this paper. It is tempting, in a study such as this, to reject the data rather than the model. But this tendency is exaggerated when the exogenous variables are poor. To reject the model itself, then, it seems essential to develop new notions of exogenous variables that relate more closely to the consumer's internal states (e.g., perceived risk of the product class), and use these variables in the first stage. Then if the model fails, we may be more willing to seriously question the model itself. The importance of this point for marketing theory is shown by the fact that the most important theory of buyer behavior to date, the Howard-Sheth Model, still uses in its latest version, socio-economic, demographic, and personality variables as its exogenous variables (Howard and Ostlund, 1973). In the Farley and Ring study, poor results led to rejection of the data, not the model. There is no reason to expect any massive improvements from the new version. It seems then that we must continue to reject data until our data is exhausted, or restructure the exogenous variables so that we might reject the model itself, if the results are still poor.

REFERENCES

Bettman, J.R. The relationship of information processing attitude structures to private brand purchasing behavior. Journal of Applied Psychology, forthcoming.

Bird, M., and Ehrenberg, A.S.C. Intention to buy and claimed brand usage. Operations Research Quarterly, 1966, 17, 27-46 and 1967, 18, 4-8.

Bird, M., and Ehrenberg, A.S.C. Non-awareness and non-usage. Journal of Advertising Research, 1966, 6, 4-8.

Colley, R. Defining advertising goals for measured advertising results. Association of National Advertisers, 1961.

Farley, J.U., and Ring, L.W. An empirical test of the Howard-Sheth model of buyer behavior. Journal of Marketing Research, 1970, 7, 427-438.

Howard, J.A., and Ostlund, L.E. Buyer behavior theory. In J.A. Howard and L.E. Ostlund. Buyer Behavior, Alfred A. Knopf, 1973.

Howard, J.A., and Sheth, J.N. The theory of buyer behavior. New York: John Wiley & Sons, Inc., 1969.

Kassarjian, H.H. Personality and consumer behavior: a review. Journal of Marketing Research, 1971, 8, 409-418.

Kmenta, J. Elements of econometrics. New York: MacMillan, 1973.

Lavidge, R.J. and Steiner, G.A. A model for predictive measurements of advertising effectiveness. Journal of Marketing, 1961, 59-62.

McGuire, W.J. An information-processing model of advertising effectiveness. A Paper Presented at the Symposium on Behavioral and Management Science in Marketing, The University of Chicago, July, 1969.

Palda, K.S. The hypothesis of a hierarchy of effects: a partial evaluation. Journal of Marketing Research, 1966, 3, 13-24.

Wonnacott, R.J., and Wonnacott, T.H. Econometrics. New York: John Wiley & Sons, Inc., 1970.

----------------------------------------