International Choice Modelling Conference, International Choice Modelling Conference 2015

Font Size: 
Modeling choice when price is a cue for quality: A case study on Chinese wine preferences
David Esteban Palma, Juan de Dios Ortúzar, Luis Ignacio Rizzi, Cristian Angelo Guevara, Gerard Casaubon, Huiqin Ma

Last modified:  1 March 2019


Food and beverages are experience products, meaning they can only be fully appreciated after they are consumed. This forces consumers to decide their purchase on an expectation of the product's quality. This Expected Quality (EQ) is based on cues, such as packaging, health claims and -most notably- price.

Several authors have noted that consumers use product's price as a quality cue, associating higher prices with higher quality. This does not mean, however, that consumers will always purchase the most expensive product, as they also face budget restrictions, and exhibit different willingness to pay depending on the consuming occasion. This leads to price having a twofold effect on consumer preferences: a positive effect on purchase likelihood because consumers associate higher prices with higher quality; and a negative effect due to consumers' budget restriction.

Price's double effect induces an endogeneity problem when estimating choice models. Endogeneity is a serious problem that can render price coefficients inconsistent, yet it is often overlooked in the literature about preferences for food and beverages.

We propose a methodology to correct the endogeneity problem using latent variables on a hybrid discrete choice model. We model each alternative's EQ as a latent variable, based on each alternative's attributes and a quality indicator (level of agreement with the statement “I believe this wine is excellent”). This allows us to measure price's double effect separately, therefore controlling for (at least one source of) endogeneity. Even though this approach has been proposed before, to the author's knowledge, there are very few applications on real datasets.

The methodology was tested on data collected from a web-based Stated Choice (SC) experiment on wine preference, applied to 112 wine consumers. The sample was divided on two groups: 22 experts (wine critics and wine businessman) and 90 regular consumers. Data was collected as part of a wider private study for a wine producer. Wine is an ideal product to test the methodology because consumers often describe their purchase process as being complex, given the great number of attributes that describe a bottle of wine and the wide offer available.

Four attributes were considered on the SC experiment: label design (six levels), winemaker's story (three levels), grape variety (three levels), and price (three levels). Additionally, two consuming occasions were considered: formal and informal. Each choice exercise had two parts. After showing the alternatives to the respondents, they had to indicate their level of EQ for each alternative by rating their level of agreement with the statement “I believe this wine is excellent”. Secondly, respondents were presented the consuming occasion and had to pick the alternative they would purchase on the described occasion, or choose “I would buy none”. The consuming occasion was not included on the first stage (i.e. when asking for level of EQ), as it was considered that EQ is independent of consuming occasion.

Price levels were pivoted around price ranges indicated by consumers at the beginning of the survey. Each respondent also rated his/her individual preferences for each of the three grape varieties considered in the experiment. This ranking was expanded in order to generate additional choices.

Our results suggest that the proposed methodology does help correcting the price endogeneity problem. We checked the effectiveness of the tested methodology by comparing the results of the hybrid model with the EQ latent variable against a simpler mixed logit formulation without it. We observed increased magnitudes of price coefficients, as well as larger price – choice-probability elasticity when the EQ latent variable was present. Also, the t-test of the price coefficients improved, and both the positive and negative effects of price could be differentiated and measured independently.

The inclusion of consuming occasion on the experiment proved that this is a relevant factor to consider on the food and beverages industry, as is often mentioned in the literature. Our results indicate that consumers' preferences for packaging may even shift based on consuming occasion.

We also discovered significant differences between experts and consumers. Experts seem to associate price and quality in a milder way than consumers do. Consumers seem to build their EQ based on more attributes than the experts, suggesting that their judgement could be more influenced by extrinsic attributes (i.e. attributes other than taste and aroma) than by intrinsic ones (i.e. mainly taste and aroma). Winemaker's story, on the other hand, seems to influence purchase decisions only indirectly through the EQ construct.

In conclusion, the use of latent variables for correcting endogeneity seems promising, specially in the field of demand modelling for experience goods, where the price-quality association may induce strong bias on price coefficient estimation if the problem is neglected. Furthermore, the explicit modelling of EQ provides separate estimates for the positive and negative effects of price, i.e. the effect of price-quality association and budget constrain, respectively. This information could prove itself useful when designing pricing strategies.


Moreover, the proposed methodology requires little additional information from the respondent, and our experience suggests that the added cognitive burden is not excessive. Therefore, we propose including several indicators of the EQ latent variable on future implementations of the methodology, since that may not only improve the efficiency of the estimators, but also allow the use of other methods that are simpler to apply, such as the multiple indicator solution (MIS).

Conference registration is required in order to view papers.