This section introduces the conceptual model developed for the study60. Figure 3-1 presents the structured conceptual model, where the ellipses indicate the latent constructs (variables) and the arrows indicate the hypothesized effect (one-sided arrow) or correlation (two-sided arrow) between constructs. The latent constructs are also distinguished into latent exogenous constructs (independent variables) and latent endogenous constructs (dependent variables). Note that the conceptual model is developed from a more general perspective than the model that will subsequently be operationalized and used for data collection and data analysis. As will become clear later on, the variables that are included in this conceptual model are defined with a broader scope, while the operationalized model focuses on a range of specific variables that are defined with a narrower scope.
60 One may argue that the model is not completely theory driven and as such is not appropriate for
SEM analysis, since SEM is a confirmatory approach, rather than an exploratory approach – i.e., a SEM model should be developed with sufficient theoretical justification (Bollen, 1989; Hair et al., 2010). However, it is believed that the anticipated behavioral realism that is used to develop the model provides compelling rationale to support the relations to be investigated and sufficiently justifies the use of SEM analysis. In addition, to explore travelers’ behavior with regard to teleworking and travel information in reality can be considered as to confirm whether or not the anticipated relations exist in reality. Also note that the theory-driven condition for developing SEM model is actually often deviated or violated in practice (Asparouhov and Muthén, 2009). Researchers often sequentially modify models in practice according to the results estimated from real data in order to achieve better model fit. The procedure then actually becomes exploratory rather than confirmatory.
Latent exogenous constructs Latent endogenous constructs Perceived characteristics of teleworking Preference for travel information Preference for teleworking Perceived characteristics of commuting alternatives Perceived availability of teleworking Perceived characteristics of travel information Perceived availability of
travel information Legend:
: Latent construct
: Hypothesized effect : Hypothesized correlation : Types of latent construct
Figure 3-1: Conceptual model of travelers’ preference for teleworking and travel information
Latent constructs (endogenous constructs and exogenous constructs) in the conceptual model
Two endogenous constructs are included in the conceptual model – “preference for teleworking”, and “preference for travel information”. The constructs can be used to represent a) travelers’ revealed preference, such as inferred from traveler’s actual use of travel information and actual use of teleworking, b) travelers’ stated preference, such as inferred from travelers’ stated extent to which they would like to acquire travel information or telework, or c) inferred from both revealed use and stated intentions for using teleworking/travel information. The reason to conceptualize the two dependent variables in such a general way is that it results in flexibility in operationalization – the dependent variables can be specified and operationalized according to the particular research needs and available data. It should be noted, however, that in the process of operationalization (i.e., the process of defining specific measurement variables and collecting data to measure
constructs61), it is important to ensure the reliability of the measurement variables for the constructs – i.e., the degree of consistency between multiple measurements. Take an example of a traveler who does not often telework because of a lack of teleworking facilities but would like to telework more – this would result in a low score in terms of revealed preference but a high score in terms of stated preference. In this case, the reliability of the two measurements is low and it is hence not appropriate to use the combined revealed preference data and the stated preference data as a generic measurement of a “preference for teleworking”. In particular, note that although conceptual consistency – whether measurements are conceptually related – can be used as a criterion, the reliability of the latent construct is more an empirical issue. To the extent that the data allows researchers to define the construct as a combination of revealed preference and stated preference (and if researchers can interpret the results), researchers can define it as such. This argument holds for the other constructs used in this study as well.
Five exogenous constructs are included in the conceptual model. In order to keep the study at a tractable level, rather than to include all the factors62 that have been identified in the literature as being potentially relevant to travelers’ preferences for travel information and teleworking respectively, the model focuses on these five constructs that are believed to be of the most importance and might jointly affect travelers’ preferences for travel information and teleworking. The five constructs are: a) “perceived availability of teleworking” – travelers’ perception of the extent to which it is possible for them to telework, which may be determined by aspects such as the type of the traveler’s work (e.g., whether or not the work can be performed by teleworking) and the culture of the work environment (e.g., whether teleworking is acceptable and common at the traveler’s employer); b) “perceived characteristics of teleworking” – travelers’ perception of the characteristics of teleworking, such as the quality of the teleworking environment and teleworking facilities; c) “perceived characteristics of commuting alternatives” – travelers’ perception of the characteristics of commuting alternatives, such as the reliability of commuting time; d) “perceived availability of travel information” – travelers’ perception of the extent to which travel information is available to travelers, which may reflect, for example, whether someone finds it difficult to find travel information; and e) “perceived characteristics of travel information” – travelers’ perception of the characteristics of travel information, such as the reliability of travel information. It should be noted that in operationalization and analysis based on real data, one or more of these constructs may turn out to be correlated (implying collinearity/multicollinearity between constructs). Beforehand, it can be already argued that the constructs “perceived availability of teleworking” and “perceived quality of teleworking” might be related, just as the constructs “perceived availability of travel information” and
61 Note the distinction between measurement variables and constructs. Take the endogenous constructs
in this conceptual model for example; the constructs here refer to “preference for teleworking”, and “preference for travel information”, while measurement variables refer to the specific variables that are asked via questions in surveys in order to measure the constructs. Examples of measurement variables for the dependent variables could be how often a traveler acquires information/teleworking per week, the extent to which a traveler likes to acquire travel information/teleworking, or the extent to which travel information/teleworking is stated to be important to a traveler.
62
Literature has identified a range of factors that may play roles in determining travelers’ preferences for travel information and for teleworking, respectively. See the literature overview in Chapter 2 for the factors being potentially related to travel information and teleworking, respectively.
“perceived quality of travel information”. This does not necessarily suggest that researchers must always define mutually-exclusive measurement variables for these constructs in order to exclude correlations63. However, care should be exercised in interpretation of the effects sizes of the paths between these constructs and the endogenous constructs. These potential correlations between constructs should be taken into consideration to avoid misinterpretation. Relations between constructs
Based on the anticipation of travelers’ behavior in reality, the following relations are hypothesized for later exploration between the exogenous and the endogenous constructs.
- “Perceived availability of teleworking” is expected to be related to travelers’ “preference
for teleworking”. However, it should be noted that the relevance may be different in terms of the way in which “preferences for teleworking” is operationalized. For example, a positive relation can be expected between these two constructs when revealed- preference data is used to measure the “preference for teleworking” – the more available teleworking is to a traveler, the more often he/she may telework. It may be not the case when the endogenous construct is operationalized as travelers’ need for teleworking.
- “Perceived availability of teleworking” is also expected to play a role in affecting
travelers’ “preference for travel information”. Consider an example where the mere fact that a telework-option is available to a traveler makes that he/she is more inclined to acquire travel information in the morning (given that he/she knows that depending on the received information, he/she has the option to decide to work from home). In this example, a positive relation exists between the endogenous construct “perceived availability of teleworking” and the exogenous construct “preference for travel information” (which is operationalized as revealed-preference or stated-preference).
- “Perceived characteristics of teleworking” is expected to be related to travelers’
“preference for teleworking”. In an example where the quality of teleworking environment and facilities is used as an exogenous variable and the endogenous construct is operationalized as revealed-preference or stated-preference (but probably not as need for teleworking), a positive relation can be anticipated. It is a behaviorally reasonable expectation that the better the quality of teleworking environment and facilities is, the more preferable teleworking option would be to travelers.
- “Perceived characteristics of teleworking” is expected also to be related to travelers’
“preference for travel information”, following a similar argument as the one for the
63 Note that by saying this here, the author does not mean to say that researchers should not pay
attention to this issue in the first place. In general, multicollinearity in SEM does not reduce the reliability of the model as a whole, and multicollinearity is acceptable under certain conditions. In addition, correlations between constructs in behavioral models are often unavoidable in practice, since many of constructs are often naturally correlated. However, multicollinearity may pose problems in SEM under some conditions. Multicollinearity is difficult to manage after the fact. Hence, it is still advisable for researchers to manage the constructs to mitigate multicollinearity in the first place. In addition, as argued, after the fact happened, great care should be excised in result interpretation of the relations of the constructs that are involved with multicollinearity towards the other constructs. See, for example, Kaplan (1994), Marsh et al. (2004), and Grewal et al. (2004) for more elaborated discussions of multicollinearity in SEM.
possible relations between “perceived availability of teleworking” and “preference for travel information”.
- Just as the expected relation between “perceived availability of teleworking” and “preference for teleworking”, “perceived availability of travel information” is expected to be related to “preference for travel information” when, for example, revealed- preference data such as frequency of acquiring travel information is used to operationalize travelers’ “preference for travel information”.
- “Perceived availability of travel information” is expected to be related to travelers’
“preference for teleworking”. For example, as argued before, a traveler may like teleworking because he/she has plenty of resourceful travel information available to him/her and thinks that he/she can acquire information in advance of making decision whether or not to telework in order to avoid morning congestion. Hence, a positive relation may exist in this case.
- “Perceived characteristics of travel information” is expected to be related to travelers’
“preference for travel information”. As has been well established in the literature (see, e.g. Chorus et al., 2006b; Emmerink et al., 1995; Hato et al., 1999), characteristics of travel information such as the reliability of travel information affect travelers’ preference for travel information.
- Following the line of argumentation for the other addressed relations, “perceived characteristics of travel information” is expected to be related to travelers’ “preference for teleworking”.
- Finally, “perceived characteristics of commuting alternatives” is expected to be related to both travelers’ “preference for teleworking” as well as their “preference for travel information”. The expectation is intuitive: to name one manifest determinant – reliability of commuting time – a traveler whose commuting time is rather unreliable due to morning peak-hour congestion is expected to be more likely to acquire travel information in order to assess his/her commuting time. The traveler may also be more inclined to telework in this example in order to avoid morning peak-hour congestion.
In addition to these expected relations between the exogenous and the endogenous constructs, a correlation is also expected between the two endogenous constructs “preference for teleworking” and “preference for travel information”. This is because other common underlying factors than the considered exogenous constructs may exist in reality, for example, travelers’ intrinsic preferences or dislikes for ICT-based services in general. These factors that are not included in the model may affect travelers’ preferences for both travel information and teleworking, resulting in correlation between the two constructs.