Definitions & Key Terms Conjoint Analysis- Is a term given to a multi variate analytical tool that CONsiders JOINTly the effect of the individual attributes of a product or a brand. This helps the marketer to analyze the utility that each varied combinations of the attributes of the product is providing to the customer. Utility- The subjective preference judgment of an individual that represent the total value or worth he is putting on the product having a combination of certain attributes. Part- Worth- The values of the individual attributes that sum up or produce the total utility for the product. Additive Model- Assumes that individuals just add up the individual Part- Worths to get to the overall utility. Interaction Model- Unlike the additive model, here the individual also considers the interactions between two independent Part- Worth while valuing the overall utility of the product.
Definitions & Key Terms (Contd.) Factorial Design- Method of designing stimuli by generating all possible combinations of levels. For example a three factor (attribute) conjoint analysis with three levels each will result in 3x3x3 = 27 combinations which will form the total stimuli in the analysis. Full Profile Method- Analysis carries on based on the respondent’s evaluation of all the possible combinations in the stimuli. Fractional Factorial Design- Method of designing a stimuli that is a subset of the full factorial design so as to estimate the results based on the assumed compositional rule. Orthogonality- Joint occurrence of levels of different attributes will be equal or in proportional number of times. Validation Stimuli- Set of stimuli that are not used for estimation of the Part- Worths. Estimated Part- Worths are then used to predict preference for the validation stimuli to assess validity and reliability of the original estimates.
Definitions & Key Terms (Contd.) Pair wise Comparison Method- Method of presenting a pair of stimuli to the respondent for evaluation, with the respondent selecting one of the stimuli as preferred. Self Explicated Model- Compositional technique where the respondent provides the Part- Worth estimates directly without making choices. Adaptive (Hybrid) Conjoint Analysis (ACA)- ACA asks respondents to evaluate attribute levels directly, and then to assess the importance of each attribute, and finally to make paired comparisons between profile descriptions. Choice Based Conjoint (CBC)- An alternative form of conjoint analysis where the respondent’s task is of choosing a preferred profile similar to what he would actually buy in the marketplace. CBC analysis lets the researcher include a "None" option for respondents, which might read "I wouldnt choose any of these."
Usages of Conjoint Analysis Breaking down customer’s overall utility from the product into values put in by him on the products individual attributes. Product planning and design Accommodating conflicting interests- Buyers want all of the most desirable features at lowest possible price Sellers want to maximize profits by: Minimizing the costs of features provided Providing products that offer greater overall value than the competitors. Market segmentation based on the utility structures
Conjoint Analysis- Process Flow Stage 2 Stage 1 Decide on the attributes Stage 3 Identify the research and their levels Chose the methodology problem Focused Group is the Traditional, ACA or CBC most practiced Stage 5 Stage 4 Stage 6 Run analysis Collect responses Interpret results Individual or aggregative Rating or rank order Stage 7 Stage 8 Validate the results Apply the Conjoint results External or internal Product designing, validity tests market segmentation etc.
Types of Conjoint Analysis Traditional Conjoint Full Profile Partial Profile / Fractional Factorial Design Paired Comparison Self Explicated Adaptive Conjoint Analysis (ACA) Choice Based Conjoint (CBC)
How Conjoint Analysis works Decompose the overall utility into its individual attribute’s part- worths Additive model- Overall utility = Sum total of all part- worths Total worth/ Utility = Part- worth of level i for factor 1+ Part- worth of level j for factor 2 + …. Part- worth of level n for factor m Interaction model- Overall utility > Sum total of all part- worths Total worth/ Utility = Part- worth of level i for factor 1+ Part- worth of level j for factor 2 + …. Part- worth of level n for factor m + I (Interaction effect between the attributes and their level) Generally, the Traditional Conjoint analyses use additive models whereas ACA and CBC use interaction models
Traditional Conjoint Analysis Full Profile Partial Profile Paired Comparison Test Self Explicated Method
Full Profile Let us assume that a cricket bat Attribute Level 1 Level 2 maker is planning to launch a Type Heavy Long handle new professional level cricket Wood Kashmir willow English bat. Based on the inputs from willow focused group, salesman and Grip Single Multi experts, he finds the following attributes important for a professional bat. Attribute Level From the table let us take a profile as an example that a Wood English Willow respondent would require to rank. Grip Single Like the profile in example, a Type Long handle full profile would provide 2x2x2 = 8 combinations
Full Profile (Contd.) Now, let us assume a respondent ranks all these profiles based on his utility from these profiles (1- Highest and 8- Lowest) Profile Type Wood Grip Rank 1 Heavy English willow Multi 1 2 Heavy English willow Single 2 3 Heavy Kashmir Multi 4 willow 4 Heavy Kashmir Single 5 willow 5 Long handle English willow Multi 3 6 Long handle English willow Single 6 7 Long handle Kashmir Multi 7 willow
Full Profile (Contd.) To estimate the Part- Worth of each attribute, average ranks or ratings for each attribute level is measured Attribute Levels Ranks Across Stimuli Average Rank (AR) Deviation from Overall Rank (DOR) Type Heavy 1,2,4,5 3.0 -1.5 Long handle 3,6,7,8 6.0 +1.5 Wood English willow 1,2,3,6 3.0 -1.5 Kashmir willow 4,5,7,8 6.0 +1.5 Grip Multi 1,3,4,7 3.75 -0.75 Single 2,5,6,8 5.25 +0.75
Full Profile (Contd.) These deviations of ranks from the overall average rank is used to compute the individual Part- Worths StD= SDxSV, where SV= No. of levels/ SD= 6/10.125= 0.592 Attribute Levels Reversed Deviations Squared Deviation Standardized Deviation Estimated Part- (RD) (SD) (StD) Worth Type Heavy +1.5 2.25 +1.332 +1.154 Long handle -1.5 2.25 -1.332 -1.154 Wood English willow +1.5 2.25 +1.332 +1.154 Kashmir -1.5 2.25 -1.332 -1.154 willow Grip Multi +0.75 0.5625 +0.333 +0.577 Single -0.75 0.5625 -0.333 -0.577
Full Profile (Contd.) Let us check whether the Part- worths are reliable Pro Type Wood Grip P-W P- W P-W Total Estimate Ran file Type Wood Grip P-W Rank k 1 Heavy EW Multi 1.154 1.332 0.333 2.819 1 1 2 Heavy EW Single 1.154 1.332 -0.333 2.153 2 2 3 Heavy KW Multi 1.154 -1.332 0.333 0.155 4 4 4 Heavy KW Single 1.154 -1.332 -0.333 -0.511 6 5 5 LH EW Multi -1.154 1.332 0.333 0.511 3 3 6 LH EW Single -1.154 1.332 -0.333 -0.155 5 6 7 LH KW Multi -1.154 -1.332 0.333 -2.153 7 7 8 LH KW Single -1.154 -1.332 -0.333 -2.819 8 8
Partial Profile Partial profile is a necessity when the number of attributes and the levels within the attributes are large. In such a case, it becomes almost impossible for the respondent to evaluate the full profile 4 attributes having 4 levels each will result in 4x4x4x4 = 256 profiles Partial profile considers a subset of the entire which would be representative of the full profile This is done through an orthogonal process so that the profiles contain the levels equally or in proportion. Partial profile eases the pressure of evaluation for the respondent Out of 256 profiles, a partial profile might contain only 16 representative profiles
Paired Comparison Test Also known as Trade off Approach as the respondent is forced to make a trade- offs between the attribute levels. Instead of full profiles or partial profiles, trade off matrices are created considering all the levels of two attributes taken at a time. Incase of more than two attributes sequential trade off matrices are given to be ranked or rated in an order such that there is at least one attribute from the previous matrix is present. In Paired Comparison Tests, the value of the individual attributes come out from the different ratings its levels receive in a paired combination with the other attributes.
Paired Comparison Test (Contd.) Let us consider that a realtor is considering to build a multi storied residential apartment. From his prior knowledge he knows that other than price, the important considerations for purchasing a flat are: proximity of schools, markets, hospitals and other utilities, availability of transportation to various locations of the city Provision of elevator and garage On these attributes he can give the following options: Attributes Level 1 Level 2 Proximity Yes No Transportation Yes No Provision Yes No
Paired Comparison Test (Contd.) Unlike Full Profile which would generate 2x2x2 = 8 combinations, the Paired Comparison Test in this case would generate Attributes Proximity (Yes) Proximity (No) Transportation (Yes) 9 6 Transportation (No) 5 3 Attributes Proximity (Yes) Proximity (No) Provision (Yes) 9 6 Provision (No) 4 2 Attributes Provision (Yes) Provision (No) Transportation (Yes) 10 4 Transportation (No) 4 2
Paired Comparison Test (Contd.) From the matrices it is evident that when considering the combinations between transportation - provision and transportation – proximity, the respondent has rated the provision (Yes) higher than proximity (Yes) and again provision (No)lower than proximity (No)(transportation is constant). Value of Provision > Value of Proximity Similarly, between provision- transportation and provision- proximity, the combinations of transportation (Yes) got higher rating than proximity (Yes) whereas, transportation (No) got lower ratings than proximity (No). Value of Transportation > Value of Proximity Finally, taking proximity constant in proximity- provision and proximity- transportation, the combinations with provisions (Yes) have either got equal or higher rating than combinations with transportation (Yes) and provision (No) have equal or lower ratings than transportation (No). Value of Provision > value of Transportation Thus, Provision > Transportation > Proximity
Self Explicated Method Purists do not consider it to be a conjoint as there is no trade off to be made. Compositional techniques as the respondents rate or rank the attributes and their levels. Preferable option over traditional conjoint when the attributes and their levels are large Used as a fundamental part of ACA or hybrid conjoint.
Self Explicated Method (Contd.) Please rate the levels in a scale of 1-10 (1- Lowest, 10- Highest) based on the value you think they would provides you and divide 100 points among the attributes based on the importance you give to each of them for contributing to the functionability of a laptop (Total points should not be more or less than 100). Attribute Level 1 Level 2 Level 3 Level 4 Hard Disk 150 GB 200 GB 250 GB 300 GB RAM 1 GB 2 GB 3 GB 4 GB Processor 1.5 GHz 1.8 GHz 2.0 GHz 2.2 GHz OS Win XP Win Vista Win Vista (Pro) Linux (Home)
Self Explicated Method (Contd.) Below is the table showing the self explicated ratings. Note, Total possible value for the entire profile = (40)x100= 4000 Hard Disk = 980/4000 = 0.245 RAM = 560/4000 = 0.14 Processor = 750/ 4000 = 0.1875 Operating System = 440/4000 = 0.11Attribute Level 1 Level 2 Level 3 Level 4 TotalHard Disk (35) 5 6 8 9 (28)x35= 980RAM (20) 6 6 7 9 (28)x20= 560Processor (25) 6 7 8 9 (30)x25= 750OS (20) 4 6 9 3 (22)x20= 440
Self Explicated Method (Contd.) The inherent problem with this method is that respondents inadvertently tend to give higher ratings to the levels that have higher value. As a result, at the initial stage itself this estimation technique is flawed. Due to the absence of trade off while rating the stimuli, the respondents have the inclination to rate the attributes and their levels based on what he thinks to be most ideal and not what gives him the greatest utility. When the attributes are large it is taxing on the respondent to rate them or put value to them objectively.