Academic journal article Academy of Marketing Studies Journal

Model Log-Likelihood as a Pragmatic Measure for Comparing the Degree of Preference Change in Choice-Based Conjoint Experiments

Academic journal article Academy of Marketing Studies Journal

Model Log-Likelihood as a Pragmatic Measure for Comparing the Degree of Preference Change in Choice-Based Conjoint Experiments

Article excerpt

ABSTRACT

The simulated log likelihood (SLL) of the static HB-estimated mixed logit model is proposed to be a pragmatic measure for comparing the degree of preference change in choicebased conjoint experiments. In a Monte Carlo study, three dynamic factors of preference change are simulated. They are learning, simplification and guessing. Their effects on the SLL are examined and it is found that with the increase of the level of the preference change factors, the SLL will decrease. The SLL is useful for research investigating the factors that affect preference change in choice-based conjoint experiments.

INTRODUCTION

Designers of surveys and experiments aim to reduce the occurrence of inconsistent responses and encourage thoughtful answers. Dillman, Tortora and Bowker (1998) has pointed out several guidelines that they suggest applying in the design of web surveys to improve data quality. Examples are like uses of progress bar, motivational messages and reasonable length of questions, etc. However, none of the above factors have been tested in empirical studies before to see if they really affect respondents' consistencies, because there is lack of a measure to accounting for the degree of preference change in a survey or an experiment.

In conjoint experiments, preference change results in inconsistent answers. Consumers' preferences are usually assumed to be constant during the course of a conjoint experiment in static models. In the context of a choice-based conjoint experiment, it means that consumers' part-worths do not change over different choice sets. This assumption also forms the basis for static mixed logit models, which represent the state of the art technique accounting for heterogeneity across consumers. With a static mixed logit model, the heterogeneity in preferences across consumers is accounted for, but not the heterogeneity over time. However, preference change over time has been documented in the literature of consumer decision-making and judgment. For example, it is suggested that consumers' preferences are constructed instead of simply being discovered (Payne et al.1992; Slovic 1995). This implies that consumers' preferences may change across time and context.

Several preference change effects have been recognized so far. Learning, simplification, guessing and fatigue will be reviewed in the following section. When consumers evaluate a conjoint task, they are probably learning what attributes are important and how much weight they want to put on each attribute, simultaneously. The learning effect has been recognized in the field of conjoint analysis (e.g. Huber, et al. 1992; Liechty, et al. 2005). Research has also shown that consumers tend to be cognitive misers and employ some simplification rule in accomplishing conjoint tasks (e.g. Huber 1997; Liechty, et al. 2005). When simplification occurs, consumers do not consider all the attributes in an alternative. They only evaluate a reduced number of attributes to make their choices. Another source of preference change might come from random guessing (Haaijer, Kamakura, and Wedel 2000). Random guessing occurs when consumers are not willing to or unable to evaluate the conjoint task. Guessing is particularly a problem when choice-based conjoint experiments are employed. The multiple-choice format of the choice task in choice-based conjoint experiments makes it easy for the occurrences of guessing. Also, in choice-based conjoint experiments, there are no right or wrong answers for a question as in a performance test. This may further encourage the use of guessing as a strategy in accomplishing the choice task. The effect of fatigue is also recorded in the literature (e.g. Liechty, et al. 2005). When fatigue or boredom is present, the unobserved part of the utility that is not captured by the attributes goes up. Consumers are inconsistent in their choices. When fatigue or boredom reaches a certain level, it may be close to guessing. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.