Two New Features in Discrete Choice Experiments to Improve Willingness to Pay Estimation that Result in SDR and SADR: Separated (Adaptive) Dual Response Forthcoming in Management Science Christian Schlereth WHU – Otto Beisheim School of Management Bernd Skiera Goethe University Frankfurt
Choice-based conjoint nowadays one of the most important method to measure willingness to pay Choice-based conjoint Do not Attribute 1 purchase A B C Attribute 2 any of the Attribute 3 three … Free-choice questions only, i.e., each choice-sets contains a no-purchase option No-purchase option provides: - Clear reference point - Realistic experimental setting - Allows prediction of market penetration Carson et al. (1994); Dhar (1997); Louviere et al. (2000); Haaijer et al. (2001); Vermeulen et al. (2008) 2
Information gained when choosing a product Choice-based conjoint Do not Attribute 1 purchase A B C Attribute 2 any of the Attribute 3 three … Selection decision Purchase decision B > A; C B > 0 Product B provides sufficient utility for a purchase 3
Information gained when choosing no-purchase option Choice-based conjoint Do not Attribute 1 purchase A B C Attribute 2 any of the Attribute 3 three … Selection decision Purchase decision 0 > A; B; C None of the products provide sufficient utility No information about relative attractiveness for a purchase of attributes Not enough data to learn about individual preferences 4
Dual response: Selection decisions are also observed when no- purchase option is chosen Dual response Attribute 1 Dual Response A B C Attribute 2 Forced choice question Attribute 3 … Do not purchase most Free choice question Purchase most preferred preferred Purchase decision Selection decision 0 > A; B; C B > A; C From free choice question From forced choice question Selection decision is always observable; thus: more accurate estimation of preferences But higher cognitive effort for a respondent due to double amount of questions Dhar & Simonson (2003); Dhar & Nowlis (2004); Brazell et al. (2006) 5
Shortcoming: Context effects in choice-based conjoint Do not Attribute 1 purchase Attribute 2 A B C any of the Attribute 3 three … Examples: Attraction Effect: No-purchase option is chosen less frequently, if a dominant product alternative exists Similarity Effect: No-purchase option is chosen more frequently, if similar attractive product alternatives exists, as an “easy way out” • Whether products provide sufficient utility for a purchase is not the only reason for a respondent to pick the no-purchase option • Context effects typically neglected in estimation Huber, Payne, and Puto (1982); Tversky and Shafir (1992); Dhar (1997); Rooderkerk, Van Heerde, and Bijmolt (2011); 6
Shortcoming: Context effects in dual response Attribute 1 A B C Attribute 2 Attribute 3 … Purchase most Do not purchase preferred most preferred Other empirical findings - Higher share of chosen no-purchase option (Dhar and Simonson 2003; Dhar and Nowlis 2004; Brazell et al. 2006) - Artificial time delay between selection decisions and purchase decisions reduces no-purchase share (Dhar and Simonson 2003) • Context effects also exist for dual response • As a result, willingness to pay estimates are substantially lower compared to choice-based conjoint 7
Shortcoming: Extreme response behavior Extreme response behavior (Gensler et al. 2012) • Respondent always chooses no-purchase option • No information when respondent will start buying • WTP might be estimated too low • Respondent never chooses no-purchase option • No information, when respondent will stop buying • WTP might be estimated too high Extreme response behavior in previous studies (if reported) : Choice-Based Conjoint Dual Response • 58% in Gensler et al. (2012) • 64% in Parker and Schrift (2011) • • 22% in Wlömert and Eggers (2014) 31% in Wlömert and Eggers (2014) • • Up to 56% in our studies Up to 36% in our studies 8
Shortcoming: Impact of purchase probability on measurement accuracy If a choice-set contains more than one alternative, likelihood increases that a respondent compares a rather attractive alternative against the no-purchase option purchase decisions are less informative for a respondent with high purchase probability Implication: Companies estimate willingness to pay more accurately for a respondent who does not intend to buy their product 9
Aims of paper Development of SDR: “ Separated Dual Response “, which 1. Avoids context effects by imposing a strict separation between all forced and free choice questions Development of SADR: “ Separated (Adaptive) Dual Response “, which also 2. Avoids extreme response behavior by imposing a strict separation between all forced and free choice questions through an adaptive mechanism that captures heterogeneity in willingness to pay Ensures similar accuracy in measured willingness to pay, independent of a respondent‘s purchase probability 3. 10
Agenda Mechanism of SDR and SADR Simulation study to analyze dependence between willingness to pay accuracy and purchase probability Tests of statistical and behavioral endogeneity Insights from three empirical studies 11
Two new features for discrete choice experiments resulting in SDR and SADR Feature 1: Feature 2: Strictly separating forced and Adaptive mechanism to select fewer, but more free choice questions informative, free choice questions SADR (Separated Adaptive Dual Response): SDR (Separated Dual Response): Separated Dual Separated Adaptive Dual Response Response (SDR) (SADR) A 1 B 1 C 1 A 1 B 1 C 1 A 2 B 2 C 2 A 2 B 2 C 2 ... ... Buy Do not Do not Buy A 1 Selected 1 buy buy Buy Do not Do not Buy A 2 Selected 2 buy buy ... ... In addition to feature 1 („ strict separation “), we use We ask all forced choice questions first and then all free choice questions. Thus, we introduce a time decisions in forced choice questions to adaptively identify fewer, but more informative free choice delay between a forced and a free choice question questions 12
Adaptive mechanism of SADR – Separated Adaptive Dual Response SADR 1. Forced choice block Select A, Select B, Select C Use efficient choice design for all respondents (e.g. D-optimal) Select A, Select B, Select C Use linear probability model to approximate individual preference order (Heckman & Snyder 1997) … Simulate preference order of all products in full factorial design Estimate preliminary preference order 2. Free choice block "Smartly“ select n products for inclusion in purchase questions Purchase Presented, Purchase None Use binary logit model to select next area, which provides most information … about purchase decision making Iterate m times Probability of a purchase 100% (= Purchase) Information gap 0% (= No Purchase) 25% 50% 75% Most preferred product Least preferred product Preference order 13
A respondent‘s perspective - screenshots of SADR Forced Choice Block: ( “pick one of the products“ ) j forced choice questions A 1 B 1 C 1 A 2 B 2 C 2 A 3 B 3 C 3 A 4 B 4 C 4 A 5 B 5 C 5 ... Free Choice Block: ( “buy or not buy product“ ) Block 1 of n free choice questions Buy D 1' Do not buy D 1' Buy D 2' Do not buy D 2' Block 2 of n free choice questions Buy ... Do not buy ... Buy ... Do not buy ... ... 14
Summary of studied discrete choice experiments Separated Dual Separated Adaptive Dual Choice-Based Conjoint Dual Response Response Response (CBC) (DR) (SDR) (SADR) Do not A 1 B 1 C 1 A 1 B 1 C 1 A 1 B 1 C 1 A 1 B 1 C 1 buy Buy Do not Do not A 2 B 2 C 2 A 2 B 2 C 2 A 2 B 2 C 2 buy Selected 1 buy ... ... A 2 B 2 C 2 ... Buy Do not Buy Do not Do not Buy A 1 Selected 1 buy Selected 2 buy buy Buy Do not Do not Buy A 2 Selected 2 buy buy ... ... Mechanism Shortcomings X X -- -- Context effects Extreme response X X X -- behavior Impact of purchase probability X X X -- on accuracy of WTP 15
Estimation Scale-extended model DR-2Max-model (Diener, Orme, and Yardley 2006) Attribute 1 Attribute 1 Attribute 2 Attribute 1 A B C Attribute 2 A B C Attribute 3 Attribute 2 Extended to account for differences in consistency between selection and A B C Attribute 3 … Attribute 3 … … purchase decisions (Swait and Andrews 2003) d h,i,j d h,i ',j' exp V exp V 1 h,i 2 h,i' L h exp V exp V exp V j J i C j' J' i' C 1 h,j 2 0 2 h,i' j j' Do not buy product D Buy product D Do not buy product D J Buy product D Do not buy product D Buy product D Force choice questions Free choice questions Estimation Multinomial logit model using Hierarchical Bayes All models implemented in Matlab 16
Agenda Mechanism of SDR and SADR Simulation study to analyze dependence between willingness to pay accuracy and purchase probability Tests of statistical and behavioral endogeneity Insights from three empirical studies 17
Recommend
More recommend
