What is the random utility theory
People do not always choose the best option -> Some randomness involved
Name a few random components
Tiredness
Uncertainty
Distraction
Context effect (starbucks cups)
Direct questionnaire +/-
\+ Simple: # attributes large \+ No high involvement assumed - 'Everything is important' - Not realistic: isolated attributes - Explicit: social answers - Subjective range of levels
Conjoint +/-
- Complex: # attributes small
- High involvement assumed
+ Forced to make trade off
+ More realistic: hypothetical products
+ Implicit: less social answers
+ Predefined range of levels
Attributes should…
- be relevant for management
- have varying levels (be realistic (car with 3 tires))
- be expected to influence preferences
- be clearly defined and communicable (no vague descriptions)
- not exhibit strong correlations (tesla & electric)
Techniques for large numbers of attributes
- Direct Survey
- Partial-profile conjoint (only subsets of attributes)
- Hybrid conjoint (survey + small conjoint)
- Adaptive conjoint (survey + dynamic paired comparisons)
Levels should be….
- interesting for management
- Unambiguous (not “low” or “high”)
- Separated enough (otherwise too little weight)
- Realistic
- Have no clear winner
Number of levels
- Two is minimum
- In case of linearity -> 2 will do
- In case of nonlinearity -> more than 2
More levels than necessary is….. why?
inefficient - more parameters need to be estimated
Why equal numbers of levels?
Attributes with more levels are perceived as more important
Advantage of large questionnaire
- More observations per respondent
- Increase in quality, as respondents learn how to answer
Disadvantage of large questionnaire
- Decrease in quality as respondents get fatigued or bored
Respondents can complete how many tasks?
Up to 20, but in some case 12 may be maximum
Always add fixed holdout choice sets? True of False?
True
Full factorial design =
- all possible combinations
Fractional factorial design =
- Subset of all possible combinations
Optimal design …
- provides as much info as possible for given number of tasks
- minimizes standard errors of part-worth estimates
