Influences On Consumer Use of Word-of-Mouth Recommendation Sources
Influences On Consumer Use of Word-of-Mouth Recommendation Sources
Influences On Consumer Use of Word-of-Mouth Recommendation Sources
Use of Word-of-Mouth
Recommendation Sources
Dale R Duhan
Texas Tech University
Scott D. Johnson
University of Louisville
James B. Wilcox
Texas Tech University
Gilbert D. Harrell
Michigan State University
This article reports the development and testing of a theoretical model of the initial stages of recommendationbased decision making by consumers. Although consumers
use a variety of recommendation sources, they have different motivations for the use of different sources. The model
focuses on the factors that influence the likelihood of
consumers using strong-tie sources (e.g., friends and family) and weak-tie sources (e.g., acquaintances or strangers) of recommendations. The factors used in the model
are the prior knowledge level of the consumer about the
product being considered, the perceived decision task difficulty level, and the type of evaluative cues sought by the
consumer. Hypotheses are tested using data collected in an
extensive fieM study with consumers. Two paths or routes
of influence on the use of recommendation sources are
proposed and confirmed in the study.
284
FALL 1997
EvaluativeCues
Well-known conceptualizations of types of cues (taxonomies) have been developed and used in the marketing
literature by Olson (1977) and Park and Lessig (1981).
These taxonomies have in common a dichotomy based on
(1) properties of the physical object (i.e., internal cues,
intrinsic cues, and functional cues such as color or type of
fiber) and (2) characteristics outside of the object (i.e.,
external cues, extrinsic cues, and nonfunctional cues such
as price and brand). Other researchers have developed
taxonomies based on the distinction between affective
cues, which relate to the aesthetic aspects of the product
(and which could be either internal or external) and instrumental cues, which relate to the more technical- or
performance-oriented aspects of the product (Ben-Sira
1976, 1980; Crane and Lynch 1988; Glassman and Glassman 1981; Lovdal and Pearson 1989; Swartz and Stephens
1983). The affective and instrumental cues' conceptualization of product cues is more generalizable across the
product categories of both goods and services, and it has
been adopted for this research.
Affective EvaluativeCues
The likelihood of obtaining information about evaluative cues is different for various types of cues and sources
of recommendations. The evaluation of affective cues is
generally based on subjective criteria established by and
related to the purchaser (e.g., "My wife hates blue cars"),
whereas the evaluation of instrumental cues is generally
based on characteristics of the product that can be evaluated independently of the purchaser (e.g., "This car gets
good gas mileage"). Strong-tie sources are more likely to
be in a position to evaluate both the purchaser (to whom
285
FIGURE 1
General Model of Recommendation Source Choice
II IIII
Prior
Knowledge
Source
Choice
" Evaluative =.
Cues !
the recommendation is to be given) and the product. Consequently, strong-tie sources are more likely to be good
sources of information regarding affective cues (what aspects of the product the consumer would like) and are more
likely to be sought for that type of information. As an
example, the question "What restaurants do you think that
I would like?" presupposes knowledge of both restaurants
and the preferences of the person asking the question. It is
also essentially a question about affective characteristics
(e.g., the type of the food rather than health department
ratings or prices). The 'following hypothesis is proposed:
I l l : The greater the importance of affective evaluative
cues, the greater the likelihood that strong-tie
sources will be sought for recommendations.
Figure 2 presents this and the other relationships that form
the theoretical model. Instrumental cues will be discussed
later as an influence on the choice of weak-tie sources.
Task Difficulty
In their theory of human problem solving, Newell and
Simon (1972) proposed that the task environment "determines to a large extent the behavior of the problem solver,
independently of the detailed internal structure of his information processing system" (p. 788). An important part
of the decision task environment is the perceived level of
difficulty. Task difficulty has been defined in a variety of
286
FALL1997
FIGURE 2
Theoretical Model of Recommendation Source Choice
Mediating Variables
. Consumer
Pnor Knowledge
Level
Recommendation
Source Type
-I-
H(
H3
'~
Weak-tie
H9
automobile purchasers do not understand the more technical aspects of cars (e.g., the functions of fuel injection
systems, transmission coolers, etc.), so they evaluate cars
on the basis of affective characteristics (e.g., comfort,
style, etc.). Thus, we generalize from this that the greater
the perceived task difficulty, the greater the perceived
importance of affective evaluative cues. The formal hypotheses regarding the influence of task difficulty are
stated as follows:
H2: The greater the level of perceived task difficulty, the
greater the likelihood that strong-tie sources will be
sought for recommendations.
H3: The greater the level of perceived task difficulty, the
greater the importance of affective cues in the decision process.
We posit that task difficulty influences recommendation
source choice both directly and indirectly through affective cues.
Prior Knowledge
The concept of prior knowledge is defined in terms of
the extent of experience and familiarity that one has with
a product (good or service). The concept has been widely
researched (e.g., Alba 1983; Alba and Hutchinson 1987;
Brucks 1985; Herr 1989; Murray 1991; Park, Mothersbaugh, and Feick 1994; Rao and Monroe 1988; Sujan
1985) and c o m m o n l y refers to information that is
Relationships among types of prior knowledge. Experience-based knowledge has presented difficulties for information-processing theories. The relationships found
between experience-based prior knowledge and other variables have often been inconsistent with the relationships
that subjective and objective prior knowledge have with
those same variables (Brucks 1985). This indicates that
perhaps experience-based prior knowledge should be conceptualized as related to, but different from, subjective or
objective prior knowledge in the way it influences consumer decision making.
In the process of choosing a recommendation source,
experience functions as the basis for the development of
both objective and subjective knowledge rather than as a
direct influence on the decision. When decision makers are
confronted with a new decision task, there is a natural
limitation in the applicability of their exact memories of
previous experiences to the new task. Instead, they recall
abstractions of their experiences in the form of their objective and subjective prior knowledge.4 Thus, past experiences are related to other variables in the decision process
through the influence that they have on subjective and
objective prior knowledge. This conceptualization is similar to the four-stage process described by Hoch and
Deighton (1989) through which consumers integrate their
experiences with their prior beliefs to form new or modified beliefs.
288
FALL 1997
son 1981), and are therefore more likely to use them (Park
and Lessig 1981; Rao and Monroe 1988). Thus, the relationship between subjective prior knowledge and the use
of instrumental cues is expected to be positive.
METHOD
Research Context
The selection of the context for the test of this model
was guided by the need to have a decision task that was
both likely to involve recommendations and to be common
enough that the typical consumer would see it as realistic
and familiar. The professional services market was chosen
because research has consistently found that recommendations play a major role in the selection of professional
services providers (e.g., Crane and Lynch 1988; Feldman
and Spencer 1965; King and Haefner 1988; Murray 1991;
Orsini 1982; Parsons 1951). Moreover, as a result of the
increasing importance of services in our economy, there is
a body of literature on services to use as guidance for
measurement and other methodological issues. Within the
professional services market, medical services were chosen, since a high proportion of consumers have had experience with choosing a medical service provider.
Additionally, there is evidence that in many cases, the
choice of a medical service provider can present a reasonably difficult decision task (Murray 1991; Reade and
Ratzan 1989; Shostack 1987). To perform the study, we
needed to narrow down the large array of services within
the medical field. Additionally, we needed to avoid emergency medical situations in which the relative convenience
of various medical service providers would dominate the
decision process. The context chosen was the selection of
a provider of obstetric services (an obstetrician) by women
of child-bearing age.
This context was operationalized through the use of the
scenario method. The use of scenarios offers the advantage
of avoiding situational effects associated with the selection
of a service provider (e.g., conditions of emergency or
dissatisfaction with current provider) (Eroglu 1987). Additionally, scenarios enable the researcher to more consistently portray decision situations. The scenario employed
in this research described a situation in which the subject
had recently moved to a new city and needed to select a
physician.
Sampling Procedures
A multistage cluster sampling procedure was used to
select subjects for the study. A simple random sample of
six census tracts was selected from within a large metropolitan area. Each chosen tract was then further broken
down by using a simple random sample to select five
census blocks within each tract. After the blocks were
Task difficulty. Bettman et al. (1990) found that perceived task difficulty is greater when there is a smaller
number of shared attributes among the alternatives, a
greater level of uncertainty about which attributes to consider, and a greater number of both attributes and alternatives to process. To ensure variability in the perceived
levels of task difficulty, we employed alternative scenarios
similar to those used by Kuehl and Ford (1977) and Ratchford and Andreasen (1974). Potential respondents were
randomly given one of two scenarios: low-task difficulty
or high-task difficulty. The high-task difficulty scenario
described a need for a specialist, used more technical and
abstract language, and described a situation in which the
consumer had a greater number of alternative providers
from which to choose (see the appendix for the complete
text of the scenarios). The task difficulty scenarios were
pretested and modified to ensure that variability enhancement was taking place (t value for the differences in
perceived difficulty = 3.45, p < .01). The level of task
difficulty perceived by the respondent after reading a scenario was measured on a scale ranging from very difficult
to very easy (higher values indicated greater difficulty).
Cue characteristics. The findings from the exploratory
research and previous empirical studies (Ben-Sira 1976;
Lovdal and Pearson 1989) were incorporated to develop
items for the measurement of the relative importance of
affective and instrumental cues. Affective cues were measured
with items that scaled the importance of the physician's
friendliness, warmth, caring, willingness to talk to the
patient, and ability to put the patient at ease. The importance of instrumental cues was similarly measured, with
items including whether the physician had advanced training, was trained by a renowned expert, could perform
complex procedures, used the latest equipment, or was
recognized by his or her peers as an expert.
Prior knowledge. Experience has generally been measured
as the number of product-related experiences a consumer
has had. That approach was used in this research, where
experience with medical (especially obstetric) services
was measured by the number of experiences as a hospital
patient plus the number of times the respondent had given
birth. Subjective prior knowledge has generally been
measured with subjective self-assessment of knowledge
level. Studies by Rao and Monroe (1988) and Rao and
Seiben (1992) each measured subjective prior knowledge
with a subject-reported assessment of prior knowledge on
a scale that ranged from completely unfamiliar to extremely familiar. In this research, subjective prior knowledge was measured using an adaptation of the same
measure for self-assessment of knowledge, with high val-
290
Analysis
The test of the model presented in Figure 2 was performed using L I S R E L 8 (JiSreskog and Sorb/Sm 1995).
This analysis tool was chosen because of the nature of the
relationships within the model and the fact that some of
the constructs in the model were measured with multiple
indicants. The use of LISREL allowed the concurrent
evaluation of the relationships within the model and the
multiple indicators of latent constructs.
FALL 1997
TABLE 1
Estimated Parameters for
the Measurement Model"
Parameter
Affectivecues
ZI.1
Composite Reliability
.79
.71
.53
.78
.56
.67
~.1.:2
~-1.3
~!.4
X1,5
Insmunental cues
~,6
;~2,7
.80
~,8
.66
.62
.62
x2,9
.66
;k.2,10
Strong-tie sources
7~,H
L3,n
L3,13
.75
.76
.72
.6,1
.73
.56
~,14
Weak-tie sources
~L4,15
~'4,]6
~-4.17
~,4,1s
.66
.48
.34
.79
.63
a. Standardizedestimates.
RESULTS
TABLE 2
Test of Hypothesized Model
From
Hypothesized Path
To
Direct effects
H1 Affective
H2 Difficulty
H3 Difficulty
H4 Objective
H5 Subjective
H6 Experience
I-I7 Experience
H8 Instrumental
H9 Subjective
Indirect effectsb
Experience
Experience
Objective
Objective
Subjective
Subjective
Subjective
Strong
Strong
Affective
Difficulty
Difficulty
Objective
Subjective
Weak
Instrumental
Instrumental
Weak
Affect
Strong
Affect
Strong
Weak
Sign
+
+
+
+
+
+
Standardized
Estimation
t Valuea
n.s.
.20
.19
.20
-.22
.27
.56
.37
.21
2.26
2.23
2.58
-2.77
3.15
6.08
3.18
2.44
.12
.04
.04
.04
-.04
-.05
.08
2.27
1.91
1.68
1.71
-1.74
-1.78
2.01
a. All paths are significantat the .01 level unless otherwise noted.
b. Only significantpaths reported.
goodness-of-fit index 0f.88, a nonnormed fit index of.90,
a comparative fit index of .91, a root mean square residual
of .069, a root mean square error of approximation of.048
(p value for the test of close fit at .05 = .6), Z2(202) =
TABLE 3
Interconstruct Correlation Matrix
Affective Cues
Affecfive cues
Instrumental cues
Strong-tie
Weak-tie
Task difficulty
Objective knowledge
Subjective knowledge
Experience knowledge
Instrumental
Cues
1.00
0.25
0.04
0.09
0.19
0.03
-0.04
-0.01
Strong-Tie
1.00
-0.01
0.37
-0.04
,0.03
0.21
0.12
Weak-Tie
1.00
0.00
0.21
0.03
-0.04
-0.01
DISCUSSION
The test of this model reveals a number of interesting
results, the most important of which is support for the
general proposition by Brown and Reingen (1987) that
there are different influences on the likelihood of consumers choosing different types of recommendation sources.
More specifically, the likelihood of choosing strong-tie
sources is influenced by task difficulty and prior knowledge, and the likelihood of choosing weak-tie sources is
influenced by the importance of instrumental cues and
subjective prior knowledge. Consumers with higher levels
of subjective knowledge are perhaps more likely to feel
that they can evaluate instrumental information themselves, and thus they are inclined to use the "instrumental
to weak-tie" route. Consumers who perceive great difficulty in the decision task are likely to look toward strongtie sources for recommendations. These are not mutually
exclusive routes; consumers routinely use both in a given
decision situation. The routes reflect differences in the
reasons for seeking recommendations from various types
of sources.
Additional support for the two distinct routes of influence was found by examining the indirect effects within
the model. There are significant indirect effects on strongtie sources of both objective prior knowledge (t = 1.71) and
subjective prior knowledge (t = -1.78). There was also a
significant indirect effect on weak-tie sources of subjective
prior knowledge (t = 2.01) and experience (t = 1.91).
Beyond the general finding regarding the two routes to
the choice of recommendation sources, there are a number
of other interesting outcomes. First, support was found for
the hypothesized relationships between experience and
objective prior knowledge and between experience and
subjective knowledge, which is consistent with the relationships identified by previous research (Alba and
Task Difficulty
1.00
-0.01
0.01
0.08
0.04
Objective
Knowledge
Subjective
Knowledge
1.00
0.17
-0.19
-0.07
1.00
0.15
0,27
Experience
Knowledge
1.00
0.56
292
CONCLUSION
This research builds on previous work in the area of
recommendation-based decision making (e.g., Rosen and
Olshavsky 1987). It extends the line of research by investigating the influences on a person's choice of sources for
recommendations. We present a model, influenced by
studies from several areas, that proposes two routes of
influence on the choice of recommendation sources. The
model proposes that the prior knowledge level of the
consumer, the perceived level of task difficulty, and the
type of evaluative cues sought by consumers influence
their choice of recommendation sources. The article also
reports an extensive field study with consumers that investigated the hypotheses of the model. The results of this
research extend the literature on consumer decision making by providing a better understanding of recommendation-based decision-making processes.
The limitations of this study include the context in
which the theory was tested and the structure of the study.
Future research should investigate these issues in other
contexts. The medical services context is common enough
to ensure a reasonable test of the theory, but other contexts
may provide additional insights. The study reported here
was not longitudinal, and thus it was not able to investigate
some aspects of the recommendation-based decision process. For instance, decision makers might seek recommendations regarding other sources of recommendations, or
perhaps a decision maker's level of objective or subjective
knowledge may change during the process of seeking
recommendations and considering the decision.
ACKNOWLEDGMENTS
The authors express their appreciation to Roy D. Howell, Shelby D. Hunt, and Robert Wilkes of Texas Tech
University; Randy Sparks of the University of Dayton;
Sigurd Troye and Andreas W. Falkenberg of the Institute
of Marketing at the Norwegian School of Economics and
Business Administration in Bergen, Norway; and three
anonymous reviewers for their very helpful comments on
this manuscript.
APPENDIX
Task Difficulty Scenarios and Measure
Scenario I (Low Difficulty)
Please read the following scenario and, as best as you can, place
yourself into the role described.
As the result of a job promotion, you moved to _ _
6
months ago. While this is far from your home, you are
beginning to settle in and are really starting to enjoy your
coworkers and neighbors. Surprise! Just last week you
discovered that you were pregnant after using a home
pregnancy test kit. This news is exciting to both you and
FALL 1997
your husband. Luckily you have health insurance provided through work. The company supplied you with a list
of three names of doctors to choose from. This seems to
you like it will be a relatively easy task to perform.
doctor is caring.
doctor is friendly toward you.
doctor had advanced training in a specialized field.
doctor was trained by a renowned expert.
doctor puts you at ease.
doctor uses the latest equipment.
293
NOTES
1. In some instances, task difficulty may be so high that consumers
will decide that they cannot process the information themselves at all and
get someone to make the decision for them. If this occurs, the decision
process is no longer recommendation based, it is completely subconIracted.
2. An anonymous reviewer suggested that higher levels of decision
task difficulty lead to a greater desire for instrumental cues, and therefore
greater importance for expert sources. We felt, however, that increased
use of instrumental cues generally would increase the task difficulty.
3. Bloom et al. (1956) present a taxonomy of educational objectives
with two broad classifications: the cognitive domain and the affective
domain. The cognitive domain consists of knowledge, comprehension,
application, analysis, synthesis, and evaluation. The affective domain
consists of receiving, responding, valuing, organizing, and characterizing
by a value or a value complex.
4. If their experiences are perceived to be specifically relevant to the
current decision task (e.g., repeat purchase behavior), decision makers
are less likely to seek recommendations at all.
5. It should be noted that in the face-to-face screening, the field
researchers would only "eyeball" the age of potential subjects but not
verbally inquire about age; thus, some subjects were later dropped from
the analysis because of their self-reported age.
6. Stewart et al. (1985) found the following source options cited by
subjects: (1) friends and neighbors, (2) other family members, (3) other
physicians, (4) previous encounters with the doctor as a patient,
(5) encounters with the physician other than as a patient, (6) phone
directory, (7) government agencies, or (8) local medical societies. King
and Haefner (1988) found the vast majority of types were (1) another
doctor (48.8%), (2) friends (33.4%), (3) relatives (26.8%), (4) nurse
(12.6%), (5) other health care providers (6.8%), and (6) Yellow Pages
(6.6%). Swartz and Stephens (1983) used a somewhat different format
and found similar sources.
7. Eighteen experts and 38 novices were selected to test this knowledge measure. The expert group scored significantly higher than the
novices, t = 9.62, df= 54, p < .001.
8. Since specific directions for these relationships were hypothesized,
these are one-tailed tests that have a critical value of t = 1.645 at the .05
significance level.
REFERENCES
Alba, Joseph W. 1983. "The Effects of Product Knowledge on the
Comprehension, Retention, and Evaluation of Product Information."
InAdvances in Consumer Research. Eds. Richard P. Bagozzi and Alice
M. Tybout. Ann Arbor, MI: Association for Consumer Research,
577-580.
and J. Wesley Hutchinson. 1987. "Dimensions of Consumer
Expertise." Journal of Consumer Research 13 (March): 411-454.
Ben-Sira, Zeev. 1976. "The Function of the Professional's Affective
Behavior in Client Satisfaction: A Revised Approach to Social Interaction Theory." Journal of Health and Social Behavior 17 (March): 3-11.
294
FALL 1997
-. 1980. "Affective and Instrumental Components in the PhysicianPatient Relationship: An Additional Dimension of Interaction Theory."
Journal of Health and Social Behavior 21 (June): 170-180.
Benfler, P. M. and D. C. BoneR. 1980. "Significance Tests and Goodness
of Fit in the Analysis of Covariance Structures." Psychological Bulletin
88:588-606.
Bettman, James R. and C. Whan Park. 1980. "Effects of Prior Knowledge
and Experience and Phase of the Choice Process on Consumer Decision Processes: A Protocol Analysis." Journal of Consumer Research
7 (December): 234-248.
, Eric J. Johnson, and John W. Payne. 1990. "A Compenential
Analysis of Cognitive Effort in Choice." Organizational Behavior and
Human Decision Processes 45 (February): 111-139.
Bloom, B. S., M. D. Englehart9 E. J. Furst, W. H. Hill, and D. R.
Krathwohl. 1956. Taxonomy of Educational Objectives, Handbook I:
Cognitive Domain. New York: David McKay.
Brown, Jacqueline Johnson and Peter H. Reingen. 1987. "Social Ties and
Word-of-Mouth Referral Behavior." Journal of Consumer Research
14 (December): 350-362.
Brucks, Merde. 1985. "The Effects of Product Class Knowledge on
Information Search Behavior." Journal of Consumer Research 12
(June): 1-16.
9 1986. "A Typology of Consumer Knowledge Content." In Advances in Consumer Research. Ed. Richard J. Lutz. Provo, UT: Association for Consumer Research, 58-63.
- - a n d
Paul H. Schurr. 1990. "The Effects of Bargainable Attributes
and Attribute Range Knowledge on Consumer Choice Processes."
Journal of Consumer Research 16 (March): 409-419.
Bruner, Jerome S. 1957. "On Perceptual Readiness." Psychological
Review 64:123-152.
9J. J. Goodnow, and G. A. Austin. 1956. A Study of Thinking. New
York: John Wiley.
Churchill, Gilbert D. and J. Paul Peter. 1984. "Research Design Effects
on the Reliability of Rating Scales: A Meta Analysis." Journal of
Marketing Research 21 (February): 360-375.
Cox, Donald E 1962. "The Measurement of Information Value: A Study
in Consumer Decision-Making." In Proceedings of the American
Marketing Association. Ed. William Decker. Chicago: American Marketing Association9413-421.
Crane, E G. and J. E. Lynch. 1988. "Consumer Selection of Physicians
and Dentists: An Examination of Choice Criteria and Cue Usage."
Journal of Health Care Marketing 8 (3): 16-19.
Engel, James E, Roger D. Blackwell, and Paul W. Miniard. 1990.
Consumer Behavior. 6th ed. Chicago: Dryden.
Eroglu, Sevgin. 1987. "The Scenario Method: A Theoretical, Not Theatrical, Approach." In Summer Educators' Conference Proceedings.
Chicago: American Marketing Association, 236.
Feldman, Sidney P. and Merlin C. Spencer. 1965. "The Effect of Personal
Influence in the Selection of Consumer Services." In Marketing and
Economic Development. Ed. Peter Bennett. Chicago: American Marketing Association, 440-452.
Glassman, Myron and Nanci Glassman. 1981. "A Marketing Analysis of
Physician Selection and Patient Satisfaction." Journal of Health Care
Marketing 1 (4): 25-31.
Hayes-Roth, Barbara. 1977. "Evolution of Cognitive Structures and
Processes." Psychological Review 84 (3): 260-278.
Herr, Paul. 1989. "Priming Price: Prior Knowledge and Context Effects."
Journal of Consumer Research 16 (june): 67-75.
Hoch, Stephen J. and John Deighton. 1989. "Managing What Consumers
Learn From Experience." Journal of Marketing 53 (April): 1-20.
Hogarth, Robin. 1987. Judgement and Choice. New York: John Wiley.
Howard, John. 1977. Consumer Behavior: Application of Theory. New
York: McGraw-Hill.
295