Conjoint Analysis Chapter 2 CONJOINT ANALYSIS Object 128
Conjoint Analysis Chapter 2. CONJOINT ANALYSIS Object 1/28 Attribute +……+ Attribute 56 24 20 % % % http: //ui. korea. ac. kr
Conjoint Analysis 컨조인트 분석절차 Attribute 선정 및 Level 결정 Decision Tree Fractional Factorial Design (Orthogonal Arrays) SAS/QC 자료의 수집 자료형태 측정방법 결과의 해석 적합성 평가(Badness of Fitting) 각 Attribute 상대적 중요성 이상적인 조합 및 최적조합 1/27 http: //ui. korea. ac. kr
Conjoint Analysis Attribute 선정 및 Level 결정 1. Attribute 선정 및 Level 결정 Decision Tree Fractional Factorial Design (Orthogonal Arrays) SAS/QC 2. 자료의 수집 자료형태 측정방법 3. 결과의 해석 적극성 평가(Badness of Fitting) 각 Attribute 상대적 중요성 이상적인 조합 및 최적조합 1/27 n 기본 모형설정 ü Additive vs. Interactive ü 분석 모델: 제품의 특성과 목적에 따라서 달라짐 Ideal-point Model Vector Model Part-worth function Model Mixed Model n 자극 만들기: 수 축소 ü 현실에 맞는지, 유사한 조합이 있는지 확인 ü Factorial Design, Decision Tree, Fractional Factorial Design ü 직교배열표(Orthogonal Arrays)를 이용하여 대상을 축소 ü SAS/ QC http: //ui. korea. ac. kr
Conjoint Analysis 자료의 수집 1. Attribute 선정 및 Level 결정 Decision Tree Fractional Factorial Design (Orthogonal Arrays) SAS/QC n 보여주기 방법 선정 ü TFA (Two factor-At-Time-Approach): 선호도를 결정하는 Attribute가 두개 일 때 쓰는 방법 ü FPA (Full-Profile Method): 평가하고자 하는 Attribute가 많을 때 사용 2. 자료의 수집 자료형태 측정방법 3. 결과의 해석 적극성 평가(Badness of Fitting) 각 Attribute 상대적 중요성 이상적인 조합 및 최적조합 1/27 http: //ui. korea. ac. kr
Conjoint Analysis SAS를 이용한 컨조인트 분석 예제(1) The researchers of use three factors with two levels each: 8(2*2*2) Factor Ingredient Form Brand name Level Phosphate-free Liquid HATCO Stimuli Description Form 1/27 Ingredient Phosphate-based Powder Generic brand Respondent Rankings Brand Respondent 1 Respondent 2 1 Liquid Phosphate-free HATCO 1 1 2 Liquid Phosphate-free Generic 2 2 3 Liquid Phosphate-base Generic 5 3 4 Liquid Phosphate-base HATCO 6 4 5 Powder Phosphate-free Generic 3 7 6 Powder Phosphate-free HATCO 4 5 7 Powder Phosphate-base Generic 7 8 8 Powder Phosphate-base HATCO 8 6 http: //ui. korea. ac. kr
Conjoint Analysis SAS 코드(1) n SAS 코드 Title 'Perforence '; DATA FORM; INPUT FORM $ Ingred $ Brand $& Rating; DATALINES; Liquid Phfree HATCO 1 Liquid Phfree Generic 2 Liquid Phbased HATCO 5 Liquid Phbased Generic 6 Powder Phfree HATCO 3 Powder Phfree Generic 4 Powder Phbased HATCO 7 Powder Phbased Generic 8 ; RUN; PROC PRINT DATA=form; PROC TRANSREG UTILITIES; TITLE 2 'Conjoint Analysis'; MODEL MONOTONE(RATING) = CLASS(Form Ingred Brand / ZERO=SUM); RUN; 1/27 http: //ui. korea. ac. kr
Conjoint Analysis SAS 결과 및 해석(1) n SAS 프린트 결과 Perforence 02: 41 Wednesday, January 24, 2002 1/27 OBS FORM 1 2 3 4 5 6 7 8 liquid powder INGRED phfree phbased BRAND RATING HATCO Generic 1 2 5 6 3 4 7 8 http: //ui. korea. ac. kr
Conjoint Analysis SAS 결과 및 해석(1) n SAS 분석결과 Utilities Table Based on the Usual Degrees of Freedom Label Intercept FORM liquid FORM powder 1/27 Importance (% Utility Range) Utility Standard Error 4. 5000000 0. 00000 -1. 0000000 0. 00000 28. 571 CLASS. FORMLIQU CLASS. FORMPOWE CLASS. INTEGRPH Variable INTERCEPT INGRED phbased INGRED phfree 2. 0000000 -2. 0000000 0. 00000 57. 143 BRAND Generic BRAND HATCO 0. 5000000 -0. 5000000 0. 00000 14. 286 CLASS. BRANDGEN CLASS. BRANDHAT http: //ui. korea. ac. kr
Conjoint Analysis SAS를 이용한 컨조인트 분석 예제(2) The researchers of use three factors with two levels each: 8(2*2*2) Factor Ingredient Form Brand name Level Phosphate-free Liquid HATCO Stimuli Description Form 1/27 Ingredient Phosphate-based Powder Generic brand Respondent Rankings Brand Respondent 1 Respondent 2 1 Liquid Phosphate-free HATCO 1 1 2 Liquid Phosphate-free Generic 2 2 3 Liquid Phosphate-base Generic 5 3 4 Liquid Phosphate-base HATCO 6 4 5 Powder Phosphate-free Generic 3 7 6 Powder Phosphate-free HATCO 4 5 7 Powder Phosphate-base Generic 7 8 8 Powder Phosphate-base HATCO 8 6 http: //ui. korea. ac. kr
Conjoint Analysis SAS 프로그램 코드(2) n Combination data c; input form ingr brand subj 1 -subj 2; /*독립변수 3, 종속변수(Subject 2 명)*/ 주의: 변수명을 8자리를 넘지 않 drop form ingr brand; /*3가지 속성 음 변수명은 결과 출력시 생략*/ if form = 1 then nform =‘liquid'; else if form = 2 then nform =‘powder'; if inte = 1 then ninte =‘pho-free'; else if inte = 2 then ninte =‘pho-based'; /*각 속성의 Utility를 계산하기 위한 각 속성의 Level 확인 */ if brand = 1 then nbrand =‘HATCO'; else if brand = 2 then nbrand=‘generic'; average= mean(of subj 1 -subj 2); /*2명 Subject의 평균값 이용*/ length comb $ 40; /*새로운 변수는 COMB */ comb=trim(nform) || ', ' || trim(ningr) || ', ' || trim(nbrand); /*속성사이를 구분 – TRIM(변 수)|| , ||TRIM(변수) */ 1/27 http: //ui. korea. ac. kr
Conjoint Analysis SAS 프로그램 코드(2) n Raw Data cards; 11111 11222 12155 12266 21133 21244 22177 22288 ; 1/27 http: //ui. korea. ac. kr
Conjoint Analysis SAS 프로그램 코드(2) n Output proc print; /*제대로 입력되었는지 Average출력*/ var comb subj 1 -subj 2 average; proc transreg; model monotone(subj 1 -subj 2 average)= opscore(nform ningr nbrand) /maxiter=60 ; output Tstandard=center additive; id comb; proc print; var comb _depvar_ t_depend tnfont tnsiz tnlin; by notsorted _depvar_; run; 1/27 /*컨조인트 분석 실행 종속변수: 2명의 참가자들의 평균 독립변수: 대상을 설명하는 3가지 속성 종속변수 형태는 순위에 의해 얻어진 서열척도 이므로 Monotone으로 하며, 독립변수는 각 수준을 구분하는 명목척도이므로 Opscore로 함 */ /*결과 출력*/ http: //ui. korea. ac. kr
Conjoint Analysis SAS 결과 및 해석(2) Preference Metric Conjoint Analysis 17: 12 Wednesday, September 9, 1998 OBS 1 2 3 4 5 6 7 8 1/27 COMB liquid, pho-free, HATCO liquid, pho-free, gener liquid, pho-base, HATCO liquid, pho-base, gener powder, pho-free, HATCO powder, pho-free, gener powder, pho-base, HATCO powder, pho-base, gener SUBJ 1 1 2 5 6 3 4 7 8 SUBJ 2 1 2 5 6 3 4 7 8 AVERAGE 1 2 5 6 3 4 7 8 http: //ui. korea. ac. kr
Conjoint Analysis SAS 결과 및 해석(2) Preference Metric Conjoint Analysis 17: 12 Wednesday, September 9, 1998 TRANSREG MORALS Algorithm Iteration History for MONOTONE(AVERAGE) Iteration Average Maximum Squared Criterion Number Change Multiple R Change ----------------------------------1 0. 00000 1. 00000. R-square: 설명력 - 처음% 나중 % NOTE: Algorithm converged. 1/27 http: //ui. korea. ac. kr
Conjoint Analysis SAS 결과 및 해석(참고) The SAS System 03: 18 Wednesday, January 24, 2002 14 TRANSREG MORALS Algorithm Iteration History for MONOTONE(AVERAGE) Iteration Average Maximum Squared Criterion Number Change Multiple R Change ---------------------------------------1 0. 05336 0. 34735 0. 90372. 2 0. 01289 0. 05080 0. 97695 0. 07323 3 0. 00739 0. 02971 0. 97803 0. 00108 4 0. 00426 0. 01694 0. 97836 0. 00033 5 0. 00245 0. 00959 0. 97847 0. 00010 6 0. 00141 0. 00543 0. 97850 0. 00003 7 0. 00081 0. 00308 0. 97851 0. 00001 8 0. 00047 0. 00175 0. 97851 0. 00000 9 0. 00027 0. 00101 0. 97851 0. 00000 10 0. 00016 0. 00058 0. 97851 0. 00000 11 0. 00009 0. 00034 0. 97851 0. 00000 12 0. 00005 0. 00020 0. 97851 0. 00000 13 0. 00003 0. 00011 0. 97851 0. 00000 14 0. 00002 0. 00007 0. 97851 0. 00000 15 0. 00001 0. 00004 0. 97851 0. 00000 16 0. 00001 0. 00002 0. 97851 0. 00000 R-square: 설명력 - 90. 0% 97. 9% NOTE: Algorithm converged. 1/27 http: //ui. korea. ac. kr
Conjoint Analysis SAS 결과 및 해석(2) n SAS Code --------- Dependent Variable Transformation(Name)=MONOTONE(AVERAGE) ---------OBS 17 18 19 20 21 22 23 24 1/27 COMB liquid, pho-free, HATCO liquid, pho-free, gener liquid, pho-base, HATCO liquid, pho-base, gener powder, pho-free, HATCO powder, pho-free, gener powder, pho-base, HATCO powder, pho-base, gener _DEPVAR_ MONOTONE(AVERAGE) MONOTONE(AVERAGE) T_DEPEND -3. 5 -2. 5 0. 5 1. 5 -0. 5 2. 5 3. 5 TNFORM -1 -1 1 1 TNINTE -2 -2 2 2 TNBRAND -0. 5 0. 5 http: //ui. korea. ac. kr
Conjoint Analysis 실험결과 n 적합성(Goodness of Fit)평가 ü R-square: 설명력 - 초기 100% n 상대적 중요도(Relative Importance) wa = Form Ingredients Brand Total Upper 1 2 0. 5 Lower -1 -2 -0. 5 차이 2 4 1 7 비중 28. 571% 57. 143% 14. 286% 100% n 최적조합 (Optimal Combination) 1/27 ü 이상적 조합(Ideal Combination) 각 속성에서 가장 효용이 큰 수준을 선택하여 조합을 만든 것 ü Form: Powder, Ingredients: Pho-based, Brand: Generic brand 에서 이상적인 조합 http: //ui. korea. ac. kr
Conjoint Analysis 실험결과 Cleanser 선호도 Form 28. 6% Powder Ingredients 75. 1% Pho-based Brand 14. 3% Generic Display별 상대적 중요도(%) 1/27 Factors Form Ingredients Brand Cleanser Preference 25. 18 13. 31 61. 51 http: //ui. korea. ac. kr
Conjoint Analysis 감사합니다. CONJOINT ANALYSIS Object 1/27 Attribute +……+ Attribute 56 24 20 % % % http: //ui. korea. ac. kr
Conjoint Analysis http: //ui. korea. ac. kr
- Slides: 29
