Denzil G. Fiebig

The Generalized Multinomial Logit Model: Accounting for Scale and Coefficient Heterogeneity

The mixed or heterogeneous multinomial logit (MIXL) model has become popular in a number of fields, especially marketing, health economics, and industrial organization. In most applications of the model, the vector of consumer utility weights on product attributes is assumed to have a multivariate normal (MVN) distribution in the population. Thus, some consumers care more about some attributes than others, and the IIA property of multinomial logit (MNL) is avoided (i.e., segments of consumers will tend to switch among the subset of brands that possess their most valued attributes). The MIXL model is also appealing because it is relatively easy to estimate. Recently, however, some researchers have argued that the MVN is a poor choice for modelling taste heterogeneity. They argue that much of the heterogeneity in attribute weights is accounted for by a pure scale effect (i.e., across consumers, all attribute weights are scaled up or down in tandem). This implies that choice behaviour is simply more random for some consumers than others (i.e., holding attribute coefficients fixed, the scale of their error term is greater). This leads to a “scale heterogeneity” MNL model (S-MNL). Here, we develop a generalized multinomial logit model (G-MNL) that nests S-MNL and MIXL. By estimating the S-MNL, MIXL, and G-MNL models on 10 data sets, we provide evidence on their relative performance. We find that models that account for scale heterogeneity (i.e., G-MNL or S-MNL) are preferred to MIXL by the Bayes and consistent Akaike information criteria in all 10 data sets. Accounting for scale heterogeneity enables one to account for “extreme” consumers who exhibit nearly lexicographic preferences, as well as consumers who exhibit very “random” behaviour (in a sense we formalize below).

A flexible logistic growth model with applications in telecommunications

Abstract In this paper, we develop a four-parameter generalization of the logistic growth curve, the flexible-logistic (FLOG) model. It is shown that the FLOG model is sufficiently general to locate its point of inflection anywhere between its upper and lower bounds: it can offer wide variation in its degree of symmetry for a given point of inflection. Although additional parameters always produce a better within-sample fit. the specific flexibility introduced by the FLOG class of models emphasises the forecast properties by controlling the saturation level and the approach to that level. The model is subjected to a number of theoretical and empirical tests and is applied to three sets of telecommunications data.

Modelling the price, performance and contract characteristics of IT outsourcing

The rapidly increasing use of outsourcing for IT services, both in the public and private sectors, has attracted much interest from researchers and practitioners alike. While early studies of IT outsourcing were largely qualitative in nature, more recent studies have attempted to analyse the outcomes achieved in quantitative terms. This paper is consistent with the latter, but goes further by modelling the price, performance and contract characteristics that are relevant to IT outsourcing. A two-equation recursive regression model was used to analyse 48 contracts for IT support and maintenance. The results did not reveal any quantitatively significant price–performance trade-off, but did suggest that first-term contracts (i.e. the first ever contract awarded by a client for the provision of a particular IT service) were more expensive than repeat contracts. Although competitive tendering did not result in lower prices than directly negotiated contracts, it was associated with comparatively better performance. Well-defined expectations of an organizations IT requirements were also likely to lead to improved performance when the service was outsourced.

Optimal recall length in survey design.

Self-reported data collected via surveys are a key input into a wide range of research conducted by economists. It is well known that such data are subject to measurement error that arises when respondents are asked to recall past utilisation. Survey designers must determine the length of the recall period and face a trade-off as increasing the recall period provides more information, but increases the likelihood of recall error. A statistical framework is used to explore this trade-off. Finally we illustrate how optimal recall periods can be estimated using hospital use data from Swedens Survey of Living Conditions.

Using discrete choice experiments to investigate subject preferences for preventive asthma medication

Background and objective:  Long‐term adherence to inhaled corticosteroids is poor despite the crucial role of preventer medications in achieving good asthma outcomes. This study was undertaken to explore patient preferences in relation to their current inhaled corticosteroid medication, a hypothetical preventer or no medication.

Modelling multinational telecommunications demand with limited data

Abstract Forecasting the diffusion of innovations in the telecommunications sector is a constantly recurring problem for national providers. The problem is characterised by short data series making the estimation of model parameters unreliable. However, the same innovation will be diffusing simultaneously in other national markets, although with a different start date. The use of this cross-sectional data in constructing innovation diffusion models is investigated here. Four models for pooling the cross-sectional data are described and two diffusion models are discussed although only one, the Gompertz model is used throughout. Three innovation data sets are used in the evaluation of the models: digital cellular telephones, ISDN connections and fax connections. The pooled diffusion forecasts proved to be more accurate in several comparisons relative to a naive benchmark and to individual forecasts when available.

A random coefficient approach to the estimation of residential end-use load profiles *

Abstract This paper develops some extensions to the statistical approach to the estimation of residential end-use load curves and provides a substantive application of these developments to a sample of households. Importantly, the typical assumption that the coefficients of the appliance dummies are fixed, ignores two important sources of variation: during any particular hour the intensity of use of a particular appliance will vary from household to household; also the dummies indicate only absence or presence of the appliance and do not allow for variations in size or capacity. Our treatment of the coefficients of appliance dummies as random rather than fixed provides a structure for the heteroskedasticity that has been observed in previous studies of this kind. Also included in the analysis is the utilization of other sources of information in particular from direct metering and a sample of diaries. The resultant single equation specifications for individual hours are then pooled and jointly estimated using an SUR structure.

The demand for energy: evidence from a cross-country demand system

Abstract Some new evidence on the income own-price elasticities of demand for energy consumption by consumers is presented. Estimates are derived from a complete system of cross-country demand equations with energy being one of the commodities considered.

Preferences for new and existing contraceptive products.

New contraceptive methods provide greater choice in terms of effectiveness, management of side-effects, convenience and frequency of administration and flexibility, but make the decisions about contraception more complex. There are limited data on the factors that determine womens choices among these alternatives, to inform providers about the factors which are most important to women, or to predict uptake of new products. This paper reports on a choice experiment designed to elicit womens preferences in relation to prescribed contraception and to forecast the impact of the introduction of two new products into the Australian market. A generalised multinomial logit model is estimated and used in the simulation exercise. The model forecasts that the hormonal patch would be well received among women, achieving a greater market share than current non-pill products, but the vaginal ring would have limited appeal.

WHY ARE LONG-RUN PARAMETER ESTIMATES SO DISPARATE?

The specification of dynamic models typically leads to the estimation of impact responses. A transformation that allows for the direct estimation of the implied long-run parameters is discussed and the problem of choosing an appropriate estimator is addressed. Because the standard estimators of long-run responses involve ratios of regression coefficients, they typically do not possess finite sample moments. We argue that this existence of moments problem is fundamental to the observed disparity of long-run estimates. Simulation experiments are used to evaluate the properties of the standard implied estimator and a minimum expected loss estimator. Copyright 1990 by MIT Press.