AN INVESTIGATION OF ATTRIBUTE NON-ATTENDANCE IN CHOICE EXPERIMENTS CONTAINING DISEASE LABELS

Purpose: Attribute non-attendance (ANA) is when respondents ignore one or more attributes in a discrete choice experiment (DCE). This study explores ANA in a DCE containing disease labels using latent class regression analysis.

Method: Data from a DCE administered to the Canadian general population (n=800) to evaluate their preferences for different health states pertaining to colorectal cancer, type II diabetes, and rheumatoid arthritis was used. Half of the respondents were randomly selected to complete a four-attribute DCE (health state before treatment, health state after treatment, duration of life, and disease type). The remaining respondents completed an identical DCE except that disease type was excluded. For these respondents, after they completed each choice set, they were asked if they would change their answers if the corresponding disease label was applied. A conditional logit model was conducted to obtain the respondents’ baseline preference weights. A series of latent class models were also conducted in a step-wise manner to identify respondent groups with similar preferences and to account for potential ANA strategies.

Result: A conditional logit model revealed that relationships between personal utility and all attributes were in the hypothesized direction. When adjusted for the two DCE versions, the labels associated with rheumatoid arthritis (beta=0.40, SE=0.04, p<0.001) and type II diabetes (beta=0.68, SE=0.04, p<0.001) demonstrated a greater preference when compared to colorectal cancer. Latent class analysis identified six distinct respondent classes with different preferences. Equality constrained latent class analysis revealed that only 2.0% of respondents considered all attributes when choosing between the two hypothetical scenarios proposed. The attributes with the strongest effect on choices were duration of life and disease types.

Conclusion: An ordering effect appeared to be present such that the general population respondents preferred to live with type II diabetes and rheumatoid arthritis over colorectal cancer; this raises concerns of a possible disease premium especially if responses are used for guiding policy decisions. This finding was also confirmed by the latent class analysis as disease types, as well as duration of life, were more likely to be attended to over the other attributes in the DCE.