The Use of Time-Variant Attributes in Conjoint Analysis
- Presenter:
Chair: David Meltzer; Discussant: Scott Grosse Wed June 7, 2006 8:00-9:30 Room 309
Rationale: Conjoint analysis is a valuable tool in eliciting preferences for various treatment options and has helped to set priorities. The basic structure of such analyses involves making comparisons of alternatives against a status quo. The latter is one in which the values of attributes remain fixed while the attribute values for alternative treatments are varied to determine how preferences between options are affected. The requirement that the status quo be represented by a fixed set of attributes presents a problem in analyzing attributes that vary with time-so-called time-variant attributes or attributes that reflect first differences (e.g., X= X(t) - X(t+1)). By their nature, the baseline values (X(t)) of these attributes often vary alongside varying values for the first differences. With such attributes, the values associated with the status quo will also vary as the baseline values shift even though the first difference will always be zero reflecting no change. It appears, therefore, that such attributes can not be included in conjoint analyses.
Objective: This paper presents a method for coding time-variant attributes in conjoint analyses.
Methodology: This paper presents two treatment options-status quo and alternative—each with three attributes: 1. cost-effectiveness (CE), 2. budget impact (BI) in terms of dollars expended per HMO member per month (PMPM) and 3. change in newly budgeted dollars left unexpended after inclusion of alternative treatment in the formulary. The third attribute is a time-variant attribute. The following presents attribute levels for the two treatments.
Cost-Effectiveness, 3 levels (per QALY): 1. $50,000 vs. $50,000, 2. $50,000 vs. $100,000, 3. $50,000 vs. $180,000
Budget Impact, 2 levels (PMPM): 1. $0 vs. $200, 2. $0 vs. $400
Unexpended budget, 3 levels: 1. 5% (no change) vs. 5% to 4%, 2. 1% (no change) vs. 1% to 0%, 3. 1% (no change) vs. 1% to 0.5% over budget
Each of the time-invariant attributes could be coded as the value included in the above table. Regarding the time-variant attribute, the economist could use a set of dummy variables to represent the different levels and use “no change” as the reference group. Thus, “5% to 4%” could be represented by a dummy variable D1 and “1% to 0%”could be represented by a dummy variable D2 and “1% to 0.5%” over budget could be represented by a dummy variable D3. All three levels of this attribute for the status quo could be coded as “no change” or coded as a value of 0 for all three dummy variables. The value of the associated regression coefficients would capture the value that decision-makers place on both the magnitude of the change as well as differences at baseline.
Results: Probit regression is used to analyze the change in benefit (delta B) from adopting the alternative treatment.
Delta B = Beta1(CE) + Beta2( (BI) + Beta3( (D1) + Beta4( (D2) + Beta5( (D3)
The results of a November 2005 pilot analyzing preferences for alternative treatments for end-stage-kidney-disease will be presented (n=32).
Conclusion: This paper expands the scope of conjoint analysis by employing time-variant attributes.