Reliability and Validity Assessment of Patterns of Information Source Usage

James H. Leigh, Texas A&M University
ABSTRACT - Reliability and validity of dimensions of information source usage were assessed utilizing self-report data which concern the decision process and influences involved in investment in a major durable good, a shoreline protection device. Results, implications and directions for future research are given.
[ to cite ]:
James H. Leigh (1983) ,"Reliability and Validity Assessment of Patterns of Information Source Usage", in NA - Advances in Consumer Research Volume 10, eds. Richard P. Bagozzi and Alice M. Tybout, Ann Abor, MI : Association for Consumer Research, Pages: 673-678.

Advances in Consumer Research Volume 10, 1983      Pages 673-678

RELIABILITY AND VALIDITY ASSESSMENT OF PATTERNS OF INFORMATION SOURCE USAGE

James H. Leigh, Texas A&M University

[Permission to use the data set for this research granted by the Division of Research, Graduate School of Business Administration, University of Michigan, is gratefully acknowledged. The author also thanks Dr. Claude R. Martin, Jr. and Dr. Thomas C. Kinnear for their guidance and constructive criticism given while the dissertation was being formulated from which this research was drawn.]

ABSTRACT -

Reliability and validity of dimensions of information source usage were assessed utilizing self-report data which concern the decision process and influences involved in investment in a major durable good, a shoreline protection device. Results, implications and directions for future research are given.

INTRODUCTION

The initial work of Katona and Mueller (1955) on prepurchase deliberation and information seeking served to stimulate interest in examining correlates of external search as well as its role in affecting purchase behavior. A number of studies employed measures of the extent or information seeking and attempted to account for differences in this behavior through a focus on such possible explanations as an individual's demographics, activity patterns, level of product experience, attitudes and opinions, and/or the perceived social visibility of the product (Andreason and Ratchford 1976, Granbois and Braden 1976, McEwen 1978, Miller and Zikmund 197;, Newman and Staelin 1972). Taken together, these studies provided evidence that the extent of information seeking can be partially explained by product and individual characteristics.

The issue of the nature of information seeking, i.e., patterns of source usage, has also been a popular topic for investigation. Cox (1967) presented a trichotomy of types of information sources, viz., marketer dominated, consumer dominated, and neutral, which has largely served as the focus for research on this topic. Newman and Staelin (1971) incorporated measures of both the patterns and extent of source usage alongside a number of demographic, behavioral, and attitudinal measures to attempt to account for differences in the duration of the decision process between individuals who purchased a major durable. The extent of source usage was found to be a better predictor than specific patterns of usage of various sources of information. Houston (1979) examined the perceived usefulness of various sources for obtaining information about specific attributes for five durable products, and Kinnear and Taylor (1974) compared the three types of sources in terms of perceived helpfulness for obtaining information about a nonphosphate laundry detergent. Both of these studies identified important differences between the three types of sources on the measures used.

Hempel and McEwen (1976) reported results of a cross-cultural study which examined patterns of source usage among recent home buyers. Not only did they uncover aggregate differences in the nature and extent of source usage between countries studied, but, it was also determined that the sequence with which various sources were contacted tended to coincide with their results of simple frequency tabulations of use of particular sources. The implication is that, for home purchasing, there appears to be a fairly stable pattern through which consumers proceed. Newman and Staelin (1973) focused on six major durables and also reported a reasonably consistent pattern of source usage across the individuals studied, however the sequence of the pattern was not investigated.

Claxton, Fry, and Portis (1974) used three composite measures of prepurchase information gathering to develop homogeneous clusters of recent purchasers of furniture or appliances on the basis of their responses to these measures. Their results indicated that there do appear to be groups of individuals with similar source usage patterns when both in-store and out-of-store sources are considered. In their study of major durable purchasers, Westbrook and Fornell (1979) developed indices of the extent of source usage for the three types outlined by Cox (1967). Four segments were identified as having different patterns of source usage; one segment relied largely on neutral sources and also visited a number of (marketer dominated) retail stores, another had a similar-pattern but was not as thorough, and the third and fourth segments relied primarily on visiting a number of stores and personal sources, respectively. On the surface, these findings appear to contradict those of Hempel and McEwen (1976) and,in part, of Newman and Staelin (1973). It should be recognized, however, that the two sets of studies had very different objectives and depths of analysis. Moreover, given the standard protocol of contacting certain experts and necessary sources relevant to the home purchasing task, e.g., want ads, lawyers, realtors and county clerks, it is not surprising that source usage in the Hempel and McEwen (1976) study would exhibit consistent patterns across individuals within the same country.

Each of the studies discussed thus far used self-reported measures of search or source usage. Newman and Lockeman (1975) collected measures of self-reported in-store and out-of-store information seeking and actual (observed) search within the store and found no relationship between the two types of measures. Even though the self-reported data were collected two days after the focal product, ladies shoes, was purchased, one would still expect at least a small relationship to be operative. Perhaps the low involvement nature of the purchase and definitional differences between researcher and respondent served to reduce the association between objective and self-report measures of the same activity. Given that the observed measures indicated a much larger amount of search than appeared in the self-reports, the bias is not typical of a social desirability phenomenon nor would one expect such a marked difference as a function of forgetting (Wind and Lerner 1979). Nevertheless, these conflicting results serve to accentuate the need to consider issues of reliability and validity of marketing constructs, such as source usage and information seeking.

None of these studies addressed the topic of validity or the indices of information seeking or source usage except Newman and Lockeman (.975), and none of the studies reported the internal consistency of measures used. While the results of Newman and Lockeman (1975) should be cause for concern, the establishment of the reliability and validity of self-reported or objective measures is a necessary prerequisite to drawing definitive conclusions about a phenomenon under study. The purpose of the research reported here was to examine such issues as they relate to the classification schema proposed by Cox (1967).

METHODOLOGY

The data set used for the present study was funded by the National Oceanic and Atmospheric Administration and was designed to focus on problems associated with owning property along an adverse shoreline and on factors involved when dealing with one ongoing and pressing problem, shoreline erosion. A number of alternative means exist for protecting a property from the effects of the natural shoreline environment. Almost all of them involve the purchase of a product, which in many cases is constructed by qualified technical personnel and often represents an investment of several thousand dollars. There are a number of sources from which information may be obtained when contemplating investment in shore protection. The wide range and concentration of neutral sources as well as marketer and consumer dominated sources make this decision environment particularly suitable for examining the nature and extent of source usage patterns.

The data were collected in areas of Michigan in the Spring of 1978. A proportionate stratified random sample of 1430 property addresses was taken using as blocking variables six counties from three of the four regions of the Michigan shoreline which are known to differ in terms of susceptibility to shoreline erosion and using a sampling fraction of one-fifth. One region was excluded from the study due to an absence of erosion problems and a resultant lack of a defined market structure in place. Thus, the population defined for study was purposely biased toward a greater likelihood of experiencing erosion problems and a concomitant likelihood of investment in procedures for their amelioration. Detailed discussion of the survey administration phase can be found elsewhere (Leigh 1981).

A total of 624 questionnaires were returned; 573 were usable, which represented a raw return rate of 40.1 percent, and an effective return rate of 47.2 percent (in relation to the number of correct addresses). This was an excellent response rate considering the amount or information requested and the fact that, for most respondents, their shoreline property was not used as a primary residence. As a general rule, the survey population was better educated and had higher incomes than the general population of Michigan, which would lead one to expect that extensive information seeking would be more the rule than the exception when contemplating investment in a major durable.

One shortcoming of the study which was not apparent during administration of the pretest phase is that approximately one-third of the respondents did not complete the series of questions pertaining to information source usage. Analysis of the respondents and non-respondents to these questions revealed that respondents were more likely to have invested in protection and related behaviors (Leigh 1981). Therefore the analyses reported later concern the subset of the respondent population who is more highly involved and more likely to have utilized various sources of information.

Information was collected regarding whether or not specific sources had been consulted. No stipulation was made that a purchase had to be consummated since sources might be consulted in anticipation of making a purchase. Two personal or consumer dominated sources, viz., friends and shoreline neighbors; two proprietary or marketer dominated sources, viz., marine contractors and consulting engineers; and four neutral sources, viz., county extension agents, the Michigan Department of Natural Resources (DNR), property owners associations and the U.S. Army Corps of Engineers, represented the information network from which advice could be sought. In addition, the Army Corps of Engineers has published three non-technical , advisory documents about coping with shoreline erosion along the Great Lakes which are widely disseminated and readily available through numerous channels. Although these are technically neutral sources, questions regarding these publications were concerned with whether or not the respondent were familiar with the specific publication and are at best surrogate measures of source usage. For this reason, familiarity with published sources was considered as a separate dimension distinct from neutral source usage.

Indices of source usage were developed as simple, unweighted sums of responses to the relevant sources, with a 1 indicating use of a source and a 0 indicating nonuse (or familiar/unfamiliar in the case of published sources). A composite index of all sources was also constructed.

RESULTS

Descriptive Relationships

Table 1 shows descriptive statistics surrounding construction of the indices. Respondents consulted an average of two sources, with personal sources having the greatest likelihood of use and followed by proprietary, neutral, and published sources, respectively, taking into account the number of possibilities. There was a large variability in the number of total sources consulted, ranging from none to ten of the eleven possibilities. Examination of the item-total correlations reveals that, across the board, separate items related most strongly with their respective index, then with the composite index, and finally with the other dimensions of source usage, which is exactly the rank order they should be related.

The internal consistency of the separate scales was assessed using the Kuder-Richardson Formula 20 version of coefficient alpha for dichotomous items (Nunnally 1967, pp. 196-197), and the composite scale consistency was determined using the formula for the reliability of linear combinations since the indices could not be assumed to represent the same domain (Nunnally 1967, pp. 226-23;). Both the published and personal indices have respectable levels of internal consistency considering that they contain only three and two items, respectively, and the composite index was also found to be satisfactory. The proprietary index, however, barely surpassed the lower bound of .50 set by Nunnally (1967, p. 226) for early stages of research. A partial explanation for the low reliability is that the process by which individuals decide to take shore protection and the mechanisms used may take several forms; marine contractors are the more likely medium for use in installing shore protection, but a consulting engineer could be contacted with or without an ultimate contact with a marine contractor, e.g., performing the work themselves after discussions with an engineer.

TABLE 1

ITEM-TOTAL CORRELATIONS, RELIABILITY AND DESCRIPTIVE STATISTICS OF SOURCE USAGE INDICES AND ITEM RS WITH OTHER INDICES (UNCORRECTED POINT-BISERIAL COEFFICIENTS)

The level of internal consistency for the neutral index is somewhat below the cutoff set by Nunnally. Although .46 is not strong evidence of unreliability, it does point toward internal inconsistency in the nature of neutral source usage. Examination of the item-total correlations for this index indicate that items relating to the Department of Natural Resources and the Army Corps of Engineers correlate strongly with the index whereas the remaining two items are markedly weaker. It was speculated that property associations, since they are composed of friends and neighbors, were improperly conceptualized as a neutral source. Reclassification of this item as a personal source resulted in lower reliability for both indices. Another possible explanation is that neutral sources in this subject area fall into two dimensions, e.g., neutral and not-so-neutral. The DNR and the Corps of Engineers are viewed by many residents as being partisan and aligned with proprietary interests. This possibility can be discounted somewhat given the sizeable item-total correlations of these items with the neutral index and the strong correlations with the other indices. A final and perhaps the most likely explanation is that the neutral sources used here differ in terms of ease of accessibility and, at the same time, there are likely differences between individuals in terms of their knowledge of available resources. Thus, lack of awareness of the county extension agent as a source of information, for example, would preclude their use of one of the most accessible source s .

Disaggregated Structural Relationships

The four separate indices of source usage were amenable to a simultaneous investigation of the structural composition of the separate indices in disaggregated form to ascertain if the theoretically-based construction would be reproduced without the restrictions imposed by theory. Nunnally (1967) supports this approach when he states that "...constructing tests on the basis of factor analysis is not as wise as investigating the factorial composition of tests after they are constructed" (p. 255). Principal components analysis with varimax rotation was used to examine the structural properties of source usage. One analysis was performed in which Kaiser's criterion of retaining only those factors having an eigenvalue greater than one served as the basis for determining the number of factors to extract. Because only two factors were extracted using this criterion, an additional analysis was performed holding the number of 'actors constant at four to correspond to the number of separate indices defining the construct of information source usage. The results from these analyses are given in Table 2.

TABLE 2

ROTATED FACTOR LOADINGS OF INFORMATION SOURCE USAGE ITEMS USING PRINCIPAL COMPONENTS ANALYSIS WITH VARIMAX ROTATION

Neither the two-factor unconstrained solution nor the P four-factor constrained solution provide convincing evidence of the independence of the four dimensions of source usage. The first factor of the two-factor solution is composed of the items representing personal and proprietary indices, and all but one of the neutral items loaded most heavily on this factor as well. The second factor comprises the published items and two of the neutral items, one of which loaded more heavily on the first factor. It is clear that, for this analysis, simple structure was not obtained. It appears that the first factor reflects information sources likely to be consulted when investing or considering investment in shore protection, whereas items on the second factor seem to be ones not as directly related to-behavior.

The four-factor solution explained only 4.9 percent more of the variance than the two-factor solution, and the results are not particularly different. The first factor again appears to represent sources likely to be consulted by all property owners when investing or considering investment in s' ore protection. The second factor seems to represent sources which are secondary to the first in terms of their relevance to behavior; familiarity with the publications is not necessarily predicated by the need for shore protection. The third factor seems to be a personal factor with the co=mon bond being membership in a property association as opposed to a direct search for information for decision-making purposes.

These analyses demonstrate that information source usage dimensions are not totally distinct from one another in the light of findings of reliability for the separate indices. The behaviorally-based relationships among the items seem to be stronger in the shoreline ownership arena than the conceptual links delineated on the basis of type of source. If this is the general case, perhaps the classification schema will continually be overshadowed by the situation. 1 and/or behavioral specificity of the context for which measurement occurs unless steps are taken to give explicit consideration to the extraneous objects or items in the analyses. While research which simply focuses on information sources without a defined context might uncover distinct dimensions, the findings would reveal little applicable to the broader behavioral context for which information sources are used. A more feasible approach would be to employ procedures which separate sources of variance according to researcher-determined criteria, such as that accomplished by structural equation models (Bagozzi 1980). In this way, analyses of validity may be undertaken and assessed. The next section discusses the results of analyses directed toward this end.

Aggregated Structural Relationships

The assessment of construct validity is a necessary procedure for determining if the measures used provide empirical meaning within the context of the theory examined, and this is normally done by assessing both convergent and discriminant validity. Assessment of convergent-validity usually involves examining relationships among alternative ways of measuring the same basic concept, but it may be more general in scope in that relationships among all measures comprising a construct can be assessed (Joreskog 1974). Assessment of convergent validity of the dimensions of information source usage was based on the latter perspective since multiple operationalizations of each dimension were not obtained. Instead, the aggregated indices served as distinct indicators of the construct.

The analyses assessing convergent validity were performed using COFAMM (Sorbom and Joreskog 1976), a confirmatory factor analysis program which allows hypotheses to be tested and provides a test of the adequacy of a given model in relation to a less restricted or more general model. For the purpose of these analyses, the estimation procedure yields results identical to the more inclusive counterpart, LISREL (Joreskog and Sorbom 1978). In each analysis performed, four basic hypotheses were used as "working" hypotheses to test the convergent validity of the construct. Bagozzi (1980, p. 138) discusses the rationale for the "working" hypotheses and presents the parameter constraints imposed. Hypothesis One (H1) constrains the factor loadings to be equal among alternative operationalizations of a construct, constrains the respective error variances to be equal, and also constrains the correlation between constructs to be equal to unity. Hypothesis Two (H2) is identical to H1 except that the restriction that the correlation between constructs be constrained to unity is relaxed. The third hypothesis (H3) simply constrains the correlation between constructs to unity and relaxes the restrictions of equal factor loadings and equal error variances within the constructs, and the fourth hypothesis (H4) relaxes the sole remaining restriction of correlation of unity between unobservable measures of the construct, so that all relevant parameters are unconstrained. A satisfactory fit (p>. 10) when H4 is tested provides evidence of convergent validity (Bagozzi 1930, p. 142). The other hypotheses are not of direct interest individually but are used to form a set of hypotheses. Hypothesis Five (H5) is tested using two equivalent formulations (H1-H2 and H3-H4), both of which test whether the correlation between constructs is unity. This hypothesis is termed the test of congeneric measures (Joreskog 1974) and determines whether the measures of each construct are all measuring the same overall construct. In certain circumstances, another set of hypotheses (H6) can be formed from the working hypotheses, the test for parallel forms, which determines whether the items specified as parallel are alternative versions of the same test having equal true score variances and equal error variances (e.g., H1-H3 and H2-H4) (Joreskog 1974, p. 6). Because this hypothesis is often not a realistic one in the social sciences (Bagozzi 1980, p. 139) and is not reasonable here, explicit consideration will not be given to it.

The information source-usage construct containing four indices as operationalizations was amenable to having pairs of indices considered to be parallel forms of the same construct, thereby considering source usage to be composed of two dimensions. Three separate analyses or alternative constructions were possible: (1) the published and personal indices and the proprietary and neutral indices are parallel forms; (2) the published and proprietary indices and the personal and neutral indices are parallel forms; and (3) the published and neutral indices and the personal and proprietary indices are parallel forms. Such a procedure made diagnostic assessments possible.

For the three sets of analyses, only the results from the latter two sets are reported in detail. (See Tables 3 and 4, respectively.) The variance-covariance matrix used for all analyses is given in Part D of Table 3. Results from the first set of analyses in which the published and personal indices and the proprietary and neutral indices were specified as parallel forms indicated the unconstrained model hypothesis (H4) need not be rejected (Y; = 3.802, p. = .0519), however the fit is marginal and suspect since the estimate for the correlation between factors exceeded unity. The test for congeneric measures (H5) was rejected in each of the equivalent formulations. In each of the four models tested, the residuals for the analyses however indicated a close correspondence between actual and predicted covariances.

Table 3 shows the results of model testing when published and proprietary indices and when personal and neutral indices are viewed as parallel forms having equal true score variances. Hypothesis Four (H4) is soundly rejected, however the test of congeneric measures (15) that the two measures of each of two components of information source usage are measuring the same underlying construct cannot be rejected regardless of whether parallel or congeneric (unconstrained) forms are assumed, thus providing preliminary evidence of the validity of the construct. The maximum likelihood estimates for congeneric measures when Phi is constrained to unity (e.g., H3, which was invariant across the three sets of analyses) provide some evidence regarding the adequacy of the measures. Both the proprietary and neutral indices relate strongly to the underlying trait, information source usage, and the measurement error estimates are comparatively small, whereas the personal index relates well with the common factor but the level of measurement error is comparable in size. The published index estimates indicate a lower correspondence with the common factor and a high degree of measurement error. Recalling that the coefficient alphas were in the opposite direction, these findings reveal the importance of considering both reliability and validity. Scales and indices with low reliabilities can serve to contaminate or affect the estimates of a procedure, such as COFAMM, which evaluates simultaneously the entire set of information.

TABLE 3

COFAMM RESULTS FOR THE SOURCE USAGE CONSTRUCT WITH PUBLISHED AND PROPRIETARY INDICES AND WITH PERSONAL AND NEUTRAL INDICES AS PARALLEL FORMS

Results of the analyses in which the published and neutral indices and the personal and proprietary indices were considered as parallel forms are presented in Table 4. Hypothesis Four (14) cannot be rejected, which provides evidence of convergent validity. The maximum likelihood estimates are very similar to the estimates where all of the measures are assumed to be part of a common factor, i.e., when Phi equals unity. The neutral index is an exception in that estimates for this index are more closely correspondent with the common factor than in Table 3, and a smaller degree of measurement error is indicated when Phi is unconstrained. The correlation between the two common factors of .811 indicates a large portion of the true score variance is shared, but it is not unity. Given an adequate fit for H4, the test of congeneric measures (H5) would not be expected to yield further evidence of validity, although this formulation of H5 assuming parallel forms indicates a marginal fit to the data.

These three sets of analyses provide reasonable evidence of construct validity vet the different findings across the analyses require explanation. The analyses specifying as parallel, published and personal indices and proprietary and neutral indices, were the ones least expected to reveal construct validity due to the conceptual dissimilarities between both pairs of parallel forms. The analyses reported in Table 3 also were not expected to indicate that the two measures of each component are closely related in a common factor sense. The finding of congeneric measures was somewhat surprising given the earlier results but its explanation seems reasonably obvious. The two pairs of measures were balanced in terms of measurement error and factor structure in a manner analogous to cluster sampling, i.e., heterogeneous within and homogeneous across factors. When the most logical factor structure was assessed as in Table 4, the relationships were more homogeneous within factors but not perfectly related across factors. This resulted in an adequate unconstrained model fit. Because publications are neutral sources, the indices should be related. The second factor seems to be more of a behavioral or operational representation of information source usage similar to findings in the principal components analyses of the separate items.

TABLE 4

COFAMM RESULTS FOR THE SOURCE USAGE CONSTRUCT WITH PUBLISHED AND NEUTRAL INDICES AND WITH PERSONAL AND PROPRIETARY INDICES AS PARALLEL FORMS

DISCUSSION

The findings from the analyses reported here provide reasonable evidence of the reliability and convergent validity of information source usage despite the confounding influence of the basis or context for which sources are consulted in the first place. While the results demonstrate there is some validity to the common separation of information sources into the three dimensions outlined by Cox (1967), the discriminant validity of the construct is vet to be established and, further, the results of the factor analyses tend to suggest that other conceptualizations of source usage should be considered. In particular, the results indicate that the following bases for grouping sources may be important: (1) ease of accessibility, (2) the degree to which consumers are knowledgeable of the existence of a suitable source for the particular purpose, and/or (3) the relevance or instrumentality to the behavior in question. Each of these alternative means of conceptualization represents an important avenue for additional insight into the nature and extent of sour-e usage patterns.

The trichotomy forwarded by Cox (1967), however, has the advantage that it serves to eliminate through aggregation the idiosyncratic composite of sources characteristic of a particular behavior context and therefore allows for comparisons to be made across contexts. There is in principle at least one available representative of each type of source for every product purchase decision which serves to allow for the possibility of standardization on the basis of the general type of source used. Given that the number and accessibility of available information sources vary across behavioral contexts, however, an attempt to compare source usage across contexts necessitates that adjustment for these differences be incorporated into the measures constructed.

One limiting factor of this and other research studies which have focused on source usage relates to measurement of the phenomenon. Measurement usually takes place at one point in time after a decision has been made, which may result in differences between respondents in terms of the length of time since the purchase was completed. Respondents are asked to provide a retrospective report of their search process. The questions asked are usually limited to a simple use/nonuse framework and, as a consequence, do not provide insights into the actual process or sequence in which sources are consulted, do not reveal sources which may be familiar to the consumer but are not used, nor do they provide for consideration of the differing utilities of the various sources to a consumer and the resultant impact on decision making. To be useful for purposes beyond understanding the range of types of sources consulted and the relative emphasis of one type to another, the study of information search and source usage will benefit from empirical efforts which investigate source usage more fully from the perspective of consumer search and decision making processes. Such a focus requires the utilization of appropriate measures and methodologies which allow for these more detailed and definitive data to be collected.

Investigation of source usage is often conducted with a focus on whether or not a source were consulted at some point during an entire decision process. Product purchases often require that multiple decisions be made, ranging from those which involve general need or desire tradeoffs to decisions specific to particular attributes on a particular brand. Different sources may serve different functions and therefore be solicited at different stages of the process (e.g., Hempel and McEwen 1976, Leigh 1981). A focus on decision-specific source usage should result in more precise depictions of the bases for which sources are consulted as well as result in improved measurement. Moreover, information search need not result in the consummation of a purchase, and the inclusion of nonpurchasers as well as purchasers should result in more representative and valid assessments of source usage. Similarly, improved measurement should result from segmentation of a respondent group into segments having similar behavioral characteristics, e.g., new purchasers and repurchasers, unless, as was the case with this research, interest centers on source usage over a longer time frame. Furthermore, a case can be made for segmenting on the basis of the extent of search prior to examining issues of reliability and validity. It is maintained that individuals who seek information from a number of sources are more similar in their patterns of types of sources consulted than those who engage in more limited search. Aside from the impact on measurement, findings of this nature would have important policy implications as well.

These limitations and directions for future research represent issues which are not exclusive to the study of information source usage, although other domains of consumer behavior have already benefitted in part from past attention given by researchers to them. For this reason, it appears that the study of source usage would likewise be fostered through incorporation of necessary procedures which reflect the inherent complexity of the process and improve the quality of measurement. At the very least, attention should be given to the assessment of reliability and validity of the measures used when examining the topic of information source usage so that the merit of subsequent results can be determined.

REFERENCES

Andreason, Alan R.. and Ratchford, Brian T. (1976). "Factors AfFecting Consumers' Use of Information Sources," Journal of Business Research. 4. 197-212.

Bagozzi, Richard P. (1980), Causal Models in Marketing, new York: Wiley.

Claxton, John D., Fry, Joseph N., and Portis, Bernard (1974), "A Taxonomy of Prepurchase Information Gathering Patterns," Journal of Consumer Research, 1, 35-42.

Cox, Donald F. (1967), "Risk Taking and Information Handling in Consumer Behavior," in Risk Taking and Information Handling in Consumer Behavior, ed., Donald F. Cox, Boston: Division Of Research, Graduate School of Business, Harvard University, 604-639.

Granbois, Donald H., and Braden, Patricia L. (1976), "'Good Consumership' in Household Appliance Purchasing," Journal of Business Research, 4. 103-116.

Hempel, Donald J., and McEwen, William J. (1976), "The Impact of Mobility and Social Integration on Information Seeking," in Advances in Consumer Research. Vol. 3 ed. Beverlee B. Anderson Chicago: Association for Consumer Research. 341-347.

Houston, Michael 3. (1979). "Consumer Evaluations of Product Information Sources." in Current Issues and Research in Advertising 1979. eds.. James H. Leigh and Claude R. Martin. Jr., Ann Arbor: Division of Research, Graduate School of Business. University of Michigan. 135-1-4.

Joreskog, Karl G. (1974), "Analyzing Psychological Data by Analysis of Covariance Matrices," in ContemPorary Developments in Mathematical Psychology: Measurement, Psychophysics, and Neural Information Processing, Vol. 2. eds.. David H. Krantz et. al.. San Francisco: Freeman, 1-56.

Joreskog, Karl G., and Sorbom. Dag (1978), Lisrel: Analysis of Linear Structural Relationships by the Method of Maximum Likelihood, Version Iv. Chicago: International Educational Resources.

Katona, George, and Mueller, Eva (1955), "A Study of Purchase Decisions," in Consumer Behavior: The Dynamics of Consumer Reaction. Vol. I, ed.. Lincoln H. Clark, New York: NYU Press, 30-87.

Kinnear. Thomas C., and Taylor, James R. (1974), "The Role of Information in Learning About Ecological Aspects of Products," in 1973 Combined Proceedings, Series No. 35, ed., Thomas V. Greer, Chicago: American Marketing Association, 293-297.

Leigh, James H. (1981), "Informational and Organizational Bases of Shoreline Protection Decision Making," unpublished doctoral dissertation, Marketing Department, The University of Michigan.

McEwen, William J. (1978), "Bridging the Information Gap," Journal of Consumer Research, 4, 247-251.

Miller, Stephen J., and Zikmund. William G. (1975), "A Multivariate Analysis of Prepurchase Deliberation and External Search Behavior," Advances in Consumer Research. Vol. 2, ed., Mary Jane Schlinger. Cincinnati: Association for Consumer Research, 187-196.

Newman, Joseph W., and Lockeman, Brailey D. (1975), "Measuring Prepurchase Information Seeking," Journal of Consumer Research, 2, 216-222.

Newman, Joseph W., and Staelin, Richard (1971), "Multivariate Analysis of Differences in Buyer Decision Time," Journal of Marketing Research, 8, 192-198.

Newman, Joseph W., and Staelin, Richard (1972), "Prepurchase Information Seeking for New Cars and Appliances." Journal of Marketing Research, 9, 249-257.

Newman, Joseph W., and Staelin, Richard (1973), "Information Sources of Durable Goods," Journal of Advertising Research, 13, 19-29.

Nunnally. Jum C. (1967), Psychometric Theory. New York: McGraw-Hill.

Sorbom, Dag, and Joreskog, Karl G. (1976), COFAMM: Confirmatory Factor Analysis with Model Modification, Chicago: National Educational Resources.

Westbrook, Robert A., and Fornell, Claes (1979), "Patterns of Information Source Usage Among Durable Goods Buyers," Journal of Marketing Research, 16, 303-312.

Wind, Yoram, and Werner, David (1979). "On the Measurement of Purchase Data: Surveys Versus Purchase Diaries," Journal of Marketing Research, 16, 39-47.

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