3.4. Methods
3.4.1. Experimental Task
Various approaches to information overload have been utilized in past research. Important concerns arose however in the measurement of information overload. Malhotra et al. (1982) suggest that “in the information overload paradigm, the occurrence of information overload is determined by examining the ability of consumers to make correct choices across different treatment conditions” (p. 30). In their experimental setting, Malhotra et al. (1982) basically tested individual decision making performance under different conditions, each condition being characterized by a different information load (Malhotra et al. 1982).
Our experimental task was adapted from procedures of Lee and Lee (2004) and Evaristo (1993). For the task, we manipulated two factors: information quantity (3 levels) and the presence (or not) of a mechanism for filtering and sorting information effectively (2 levels). Participants were asked to imagine that they have to buy an MP3 player for a friend at an online retailer and to make a choice about which MP3 player to buy. Their friend’s preference rankings are provided in a table with the description of each of the features (4, 8, or 12 features). They have been given information in a matrix format about a series of MP3 players with invented model names. Each MP3 player was assumed to be in the same price range. Only when they were provided with sorting and filtering capabilities, subjects could sort each MP3 players ascending or descending depending on attributes levels. They could also choose to display the only MP3 players that corresponded to desired attribute levels. Sorting and filtering mechanisms thus helped subjects organize and select information according to their needs. After they choose an MP3 player, they reported the model they chose. The experiment
was a repeated measure, within group design with each subject receiving two treatments at different information quantity levels.
In contrast of other researchers who performed their experiment under time pressure (Evaristo 1993), we let subjects decide the amount of time they would allocate to the task. This gave the experiment a “critical realism” that is important in experiments (Fromkin et al. 1976). Evaristo (1993), for example, manipulated time pressure varying the amount of time allowed for completing the task. Lee and Lee (2004) introduced a unique, limited amount of time for all participants and manipulated only the amount of information to be taken into consideration. Hahn et al. (1992) found introducing a limited amount of time to affect the inverted U-curve function describing the relationship between decision quality and information load. The researchers suggest that time pressure is a necessary condition for the information overload problem to occur at high levels of information load, dampening decision quality. Thus they believe that it was important to maintain time pressure as an experimental control.
Malhotra (1982), however, argues that time pressures bias information processing and decision making. Further it is not sufficiently realistic in online buying situations, since, in fact, consumers do not face such time pressures. Consistent with Malhotra (1982) and in contrast to previous studies (Evaristo 1993; Hahn et al. 1992; Lee et al. 2004; Lurie 2004), subjects were thus not constrained by time pressure for performing the task even though we did measure how much time subjects spent on performing the task and included this parameter in the model as a control variable. A Web-based chronometer recorded time from the beginning of the task in the information acquisition phase, to the end of the task, when the subject made a choice about which MP3 player to buy.
Information quantity, or load, has been operationalized in consumer research as the number of
attributes per alternative (Hahn et al. 1992), or as both the number of attributes and alternatives (Lee et al. 2004). Lurie (2004) questions the idea that more alternatives is systematically associated with more information. Accordingly, in the present study, information quantity was operationalized as only the number of attributes (number of MP3 players features).
In addition to these two information characteristics, Lurie (2004) suggests that information structure has significant effects on the way information is acquired and subsequently on information overload. In fact the author points out that providing more alternatives does not necessarily means providing more information. There are two important information structural parameters to take into account. As Lurie (2004) puts it: “the number of different attribute levels associated with each attribute […] and the distribution of attribute levels across alternatives” (Lurie 2004, p. 474). Lurie suggests that a uniform distribution of attributes levels in alternatives is more likely to induce information overload than a non- uniform distribution. Accordingly, in order to make information overload more salient at higher levels, information structure was set so that attribute levels were uniformly distributed across MP3 players.
Next, information quantity was set as either low (L = 4 features), medium (M = 8 features), or high (H = 12 features) in the treatments.
Filtering mechanisms were also manipulated. Either subjects were provided with embedded
sorting and filtering capabilities helping them to effectively identifying the best solution (Y), or they were not (N). The sorting capability allowed subjects to sort MP3 players by features ascending or descending. The filtering capability allowed subjects to select several feature
For a given quantity of information, the table displays were identical. Details of the resulting six conditions are shown in Table 3-2 below.
Table 3.2. Experimental Manipulation – Six Conditions
Manipulation 1 2 3 4 5 6
Information Quantity L L M M H H
Filter Y N Y N Y N
Legend: L=Low; M=Medium; H=High; Y=Yes; N=No
After completing the task, subjects were asked to select the MP3 player they would choose for their friend on a drop-down list. Manipulation checks were introduced for information quantity and sorting / filtering capabilities in order to ensure that subjects have received the expected treatment.
3.4.2. Measurement of Constructs
Information quality. In their empirical test of the IS success model, Rai et al. (2002) assessed
information quality through three dimensions, content, accuracy, and format, which are widely used measures of information quality in the user satisfaction literature. Information quality is defined as “the degree to which information produced has the attributes of content, accuracy, and format required by the user” (Rai et al. 2002, p. 57). Due to high pairwise correlations among the three constructs in the original instrument of Doll and Torkzadeh (Doll et al. 1988), they modeled it as a single reflective construct. However, since content, accuracy and format are posited as facets of information quality, it would probably be more appropriate to model it as a formative construct (Petter et al. 2007). In their study, Wixom and Todd (2005) obtained appropriate results by positing these construct as antecedents of a global information quality reflective construct. Doing so, they also avoided the construct misspecification errors issue (Petter et al. 2007). For these reasons, in contrast to Rai et al. (2002), we decided to model content, accuracy and format as distinct reflective constructs, antecedent of a global information quality construct.
Information satisfaction is operationalized as a parsimonious, two item variable in accordance
with Wixom and Todd (2005). Prior research such as Rai et al. (2002) used a single item for this construct. Whereas, according to Baroudi and Orlikowski (Baroudi et al. 1988), a parsimonious single item measure can appropriately measure user information satisfaction, researchers usually consider multiple item scales to be more statistically rigorous than single item ones (Cook et al. 1979). This justifies our choice of the scale of Wixom and Todd (2005).
Intention-to-use was measured using the behavioral intention items in Venkatesh et al. (2003).
It is a common scale often used in TAM based articles. A summary of construct definitions is provided in Table 3-3 below, while details about the questionnaire instrument are given in Appendix.
Table 3.3. Definition of Constructs and Items
Construct Definition # Items Item Source
Information Satisfaction
"The degree of user satisfaction with the system"
(Rai et al. 2003, p. 57) 2,00
Wixom and Todd (2005) Information Quality “Perception of the quality of information included in the system” (Wixom and Todd, p. 91) 4,00 Wixom and Todd (2005) Content "the degree to which the system provides all necessary information" (Wixom and Todd, p. 91) 4,00
Rai et al. (2002) Accuracy "The user perception that the information is correct" (Wixom and Todd, p. 91) 2,00
Format the users' perception of how well the information is
presented" (Wixom and Todd, p. 91) 2,00
Choice uncertainty The degree to which the user believes he has or not information about which alternative to choose (Urbany 1989).
2,00 Developed for the
study
Intention-to-Use 3,00 Venkatesh et al. (2003)