International Choice Modelling Conference, International Choice Modelling Conference 2015

Font Size: 
Integrated Choice and Latent Variable Models: Holy Grail, or Not?
Akshay Vij, Joan L. Walker

Last modified: 11 May 2015

Abstract


Traditional models of disaggregate decision-making have long ignored the question of why we want what we want. Human needs have been treated as given, and attention has largely centered on the expression of these needs in terms of behavior in the marketplace. As a consequence, traditional models of disaggregate decision-making have focused on observable variables, such as product attributes, socioeconomic characteristics, market information and past experience, as determinants of choice, at the expense of the biological, psychological and sociological reasons underlying the formation of individual preferences (McFadden, 1986). Integrated Choice and Latent Variable (ICLV) models overcome these deficiencies by allowing for the incorporation of latent behavioral constructs within the framework employed by traditional models of disaggregate decision-making. ICLV models were first proposed two-and-a-half decades ago by McFadden (1986) and expanded on by Ben-Akiva et al. (2002). Rapid strides in optimization techniques and computational power and the ready availability of estimation software such as Python Biogeme and Mplus have since contributed to a veritable explosion in the number of studies estimating ICLV models.

Though much progress has been made in terms of model development, existing studies have failed to demonstrate conclusively the value of the framework to both econometricians who estimate these models and to practitioners and policy-makers who apply them to real-world problems. On one hand, ICLV models appear to be powerful methods with which to enhance existing representations of decision-making (McFadden, 1986). They allow for the proper integration of psychometric data within extant model frameworks and provide statistical tools with which to test complex theories of behavior. On the other, questions have been raised regarding the practical benefits of the framework (Chorus and Kroesen, 2014). Does an ICLV model fit the data better than a choice model without latent variables? Can findings from an ICLV model be used for policy analysis in ways that aren’t already possible using choice models without latent variables? Both sides of the debate have their proponents, but a clear verdict remains elusive.

The objective of this study is to evaluate systematically the benefits of the ICLV model framework in comparison with a more traditional choice model without latent variables using a set of criteria based on statistical considerations and relevance to practice and policy. Using a hypothetical model of recycling behavior as an example, the study derives easily generalizable analytical proofs regarding the benefits, or lack thereof, of ICLV models over choice models without latent variables and uses synthetic datasets to validate any conclusions drawn from the analytical proofs.

The study finds the statistical benefits of the ICLV model to be fewer than previously believed. In terms of goodness of fit and consistency of parameter estimates, ICLV models offer no improvements over a reduced form choice model without latent variables. In terms of efficiency of parameter estimates, benefits will depend upon the underlying covariance structure of the data. In some cases, estimates from ICLV models will be more efficient than estimates from the reduced form choice model without latent variables. In others, the two models will perform equally.

However, in terms of relevance to practice and policy, we find that ICLV models enjoy a distinct advantage over choice models without latent variables. By allowing analysts to measure, test and quantify the impact of latent constructs on observable behavior, through measures such as willingness to pay and elasticity of demand, ICLV models offer greater insights into the decision-making process - insights that can subsequently be used to inform policy and generate forecasts in ways that wouldn’t be possible using choice models without latent variables. This study makes the case that, if used thoughtfully, the ICLV model is still a very powerful tool to have in an ever-expanding toolbox of model forms.

References

Ben-Akiva, M., Walker, J. L., Bernardino, A. T., Gopinath, D. A., Morikawa, T., and Polydoropoulou, A. (2002), “Integration of choice and latent variable models,” in: Hani S. Mahmassani (ed.): In perpetual motion: Travel behavior research opportunities and application challenges, Elsevier, Amsterdam, pp. 431–470.

Chorus, C. G., and Kroesen, M. (2014), “On the (im-) possibility of deriving transport policy implications from hybrid choice models,” Transport Policy, Vol. 36, pp. 217-222.

McFadden, D. L. (1986), “The choice theory approach to marketing research,” Marketing Science, Vol. 5, No. 4, pp. 275–297.


Conference registration is required in order to view papers.