Log in / Register
Home arrow Economics arrow Bounded rational choice behaviour
< Prev   CONTENTS   Next >

4.5. Discussion

Clearly this chapter has some similarities with Hensher (2006). The same dataset is used, although this analysis does not exclude any of the observations. Both analyses have the same broad objective of investigating the impact of varying information load on ANA. The key difference lies in the methodology, where this chapter is informed by developments in the literature since the publication of the 2006 paper. In particular, concerns with the use of stated non-attendance responses have been raised and supported with evidence (e.g. Alemu et al., 2013; Hess & Hensher, 2010), and alternative methodologies have been proposed and tested extensively (Collins, 2012; Collins et al., 2013; Hess et al., 2013). Despite the methodological changes, the key findings of this chapter are consistent with Hensher's earlier work. ANA becomes more prevalent as the number of attribute levels increases, and as the number of alternatives decreases. This chapter further finds that as the number of attributes increases, the non-attendance to specific attributes also increases. This finding is consistent with the rationally adaptive model, wherein the amount of information processed is adjusted as the information load varies. The influence of the number of alternatives, however, reminds the reader that there must be sufficient information processed to distinguish between the alternatives in the choice task. A practical advantage of the approach used in this chapter is that the interrelationship of ANA and varying information load can be accounted for in the final model of interest, without the need for additional responses regarding perceived ANA.

A potential concern about the role of varying information load on ANA is that the non-attendance might just be induced by design decisions regarding the complexity of the choice tasks presented to respondents in stated choice studies. Put another way, the phenomenon might just be an artefact of the stated choice methodology. For example, it could be argued that increased information load is bad, as it will lead to ANA. Instead, Hensher has argued that relevancy is what matters (Hensher, 2006, 2010), and has proposed providing more information, and letting the decision-maker determine what is relevant to them. We would add that a too simple choice task could force attendance to an attribute that in a more realistic setting, of higher information load, would not matter. This could be a more serious problem than inducing non-attendance through the presentation of too great an information load, as modelling techniques now exist to effectively infer and separate out non-attendance, without interfering with preference heterogeneity amongst attribute attenders (Collins, 2012).

A research question of interest is the prevalence of ANA in revealed preference choice situations. It is not clear whether the information processing strategies employed in a stated choice task align with those employed in a real market. Most studies have investigated non-attendance in stated choice studies; there is scope and motivation to expand the developed methodologies to revealed preference data. A related question is the consistency of ANA across stated and revealed preference data from the same decision-maker, where this could be studied with the appropriate dataset.

To the practitioner, we present several suggestions. Beyond the broad suggestion of testing for ANA, if the choice dimensions legitimately vary across respondents within the same choice study, it would be wise to test for any functional relationship between the non-attendance rates and the varying dimensions. If non-attendance is modelled, and the information load varies, then not accounting for this in the model might bias the results. Further, with the availability of the RPANA family of models, lack of stated non-attendance data is not a hindrance to estimating such models. Finally, to reiterate an earlier point, care should be taken not to oversimplify the choice task, lest prominence be lent to an attribute that would otherwise be ignored, or lest other attributes that are important to some be excluded. The choice task should contain a realistic amount of information; the individual should be left to decide how much information to process; and an appropriate choice model can detect how often an attribute has any influence, and when it does, with what distribution over the sample.

Found a mistake? Please highlight the word and press Shift + Enter  
< Prev   CONTENTS   Next >
Business & Finance
Computer Science
Language & Literature
Political science