An Empirical Test of the Cognitive Response Model: the Importance of the Response Source Dimension

Donald G. Norris, Miami University (Ohio)
ABSTRACT - Researchers in the fields of information processing, especially those studying the effects of advertising campaigns, have made regular use of cognitive responses to assess a message's impact on its recipients. Subject protocols are typically coded by whether they agree with (support arguments) or disagree with (counter-arguments) the message advocacy.
[ to cite ]:
Donald G. Norris (1987) ,"An Empirical Test of the Cognitive Response Model: the Importance of the Response Source Dimension", in NA - Advances in Consumer Research Volume 14, eds. Melanie Wallendorf and Paul Anderson, Provo, UT : Association for Consumer Research, Pages: 575.

Advances in Consumer Research Volume 14, 1987      Page 575

AN EMPIRICAL TEST OF THE COGNITIVE RESPONSE MODEL: THE IMPORTANCE OF THE RESPONSE SOURCE DIMENSION

Donald G. Norris, Miami University (Ohio)

[At the time that this research was conducted the author was a member of the Marketing Department, The American University Washington, D.C.]

ABSTRACT -

Researchers in the fields of information processing, especially those studying the effects of advertising campaigns, have made regular use of cognitive responses to assess a message's impact on its recipients. Subject protocols are typically coded by whether they agree with (support arguments) or disagree with (counter-arguments) the message advocacy.

This so-called agreement dimension is the sole basis on which the majority of researchers code the elicited responses. The three noteworthy exceptions are empirical studies by Wright (1973), Roberts and Maccoby (1973) and Calder, Insko and Yandell (1974). Each of these studies employs one or more additional dimensions usually relating to the strength or source of the response. Using the early work of Greenwald (1968) as a model, they code responses as

a) recipient - generated

b) recipient - modified

c) message - supplied

in addition to whether they supported or refuted the message advocacy (e.g. the agreement dimension).

The study described here compares the predictive power of the widely-used agreement-only coding scheme with an agreement-source model. In addition, the response elicitation procedure gathers verbal rather than the written protocols which are typically the basis of resPonse coding.

HYPOTHESIS

The basic proposition investigated by the study is that coding cognitive responses by source as well as by agreement is superior to using the agreement dimension alone. The two models appear as follows:

1) Source-agreement model

A = W1{MSpos - W2{MSneg + W3{RMpos - W4{RMneg + W5{RGpos - W6{RGneg     (1)

2) Agreement model

A = W1{MSpos + RMpos + RGpos) - W2{MSneg + RMneg + RGneg)     (2)

where

A: Criterion variable of acceptance

MS: Message-supplied response

RM: Recipient-modified response

RG: Recipient-generated response

pos: positive

neg: negative

W: Discriminant weights

EXPERIMENTAL DESIGN

Thirty-seven houseowners living in the Washington, D.C. suburb comprised the sample. All were chosen from the sa e subdivision to control for differences in house size and demographics.

Subjects individually read a message and viewed 3 television commercials on the need for home energy conservation. All subjects were exposed to the same stimulus material. Oral cognitive responses were surreptiously recorded after the reading and viewing of the materials. The hidden recording was done to reduce subject reluctance to speak freely and the procedure was disclosed in a debriefing at each session's end.

After the second response elicitation subjects were given the opportunity to accept or reject the offer of a free hone energy audit. This choice, the dependent variable in the design, simply classified subjects into auditors and nonauditors. The predictive accuracy of each model could then be assessed by determining how using subjects each model correctly classified as auditors and nonauditors.

DATA ANALYSIS

Response coding was performed by 2 independent judges who were trained expressly for this study. After achieving satisfactory levels of inter-coder and test-retest reliability on a separate d ta set, they coded the data from the current study into the six response categories appearing in Equations (1) and (2). A total of 417 responses were coded across the 24 auditors and 13 nonauditors.

Two-group discriminate analysis was used to assess the predictive accuracy of the two models. A "jack-knifing" procedure was used to avoid upward bias in model building and subject classification. (The small sample size precluded the use of a hold-out sample for model testing). Due to the unequal cell sizes (24 and 13), the proportional chance criterion function was used in comparing the predictive accuracy of the two models (Morrison, 1971).

With 9 a (12 of the 13 nonauditors) successful classifications, only the source-agreement model showed predictive accuracy which is significantly superior to proportional chance (54%, t-3.4S, p<.01). The 69% success rate of the agreement-only model is not significantly better than chance (t=1.0, p<.20). Finally, a comparison of the two models (92% vs. 69%) indicates that the source-agreement model' 8 higher success rate is significant at the .10 level (t=1.53).

CONCLUSIONS

The results of this pilot study support the use of both the source and agreement dimensions in the coding of cognitive responses. Clearly additional research is needed using larger samples and a variety of designs, stimuli and response elicitation procedures.

However, given the fact that cognitive responses are so widely used as a dependent variable and that only additional coding, rather than additional data-gathering, is required, more attention should be given to the use of the source dimension in coding.

References available upon request.

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