Blind Product Test - Data Analysis

MARKET RESEARCH PROJECTBLIND PRODUCT TEST

Ankush Roy

Krishna Bollojula

Shubham Sharma

Suddhasheel Bhattacharya

AGENDA

• Background

• Objective

• Research Design

• Data Description

• Data Analysis

• Insights from Data Analysis

• Recommendations

BACKGROUND• R&D team of “SMIRNOFF” claims has prepared two new blends which

they claim are superior than the one in the market.

• Marketing team would like to do a Blind Test of the two new blends vs. the one in the market among regular consumers of vodka to test the market acceptance.

• This test would be done among “Own” brand (Smirnoff) drinkers and important competition brand (Fuel and Magic Moments) drinkers.

• Any of the two new blends will be considered for a change, if it comes out to be significantly better than the current blend.

OBJECTIVEPrimary Objective :

• To replace the current product with any of the two test products if found significantly(statistically) superior.

Secondary Objective :

• To understand which parameters are the key drivers for overall vodka preference and to what extent.

• To predict the factors(by reducing attributes) which influence the preference of vodka.

• To predict the purchase intention by evaluating the attribute ratings.

RESEARCH DESIGN• Sequential monadic exposure method is used to collect responses.

• All the three blends are placed for consumption one after the other and feedback is taken after each consumption.

• Neutralizer is given after each consumption to ensure the unbiased responses.

• The current product in the market “SMIRNOFF” is the Control Blend and the other two blends are Test Blend1 and Test Blend2.

• Sample Size : Total 760 sample size which gives you 2280 data points as each respondent

has given feedback on all three products.

RESEARCH DESIGN(CONTD.,)

• Target Group:Males/Females in the age group of 25 – 35 years.Consuming vodka at least twice a week.Regular consumer of any one of the three brands – Smirnoff, Fuel or Magic

Moments.

DATA DESCRIPTION

• Centers : 1. Delhi 2. Mumbai 3. Kolkata 4. Bangalore 5.Chennai

• Main Brands : Magic Moments , Smirnoff , Fuel

• Age Category : 1. 25 - 30 2. 31-35

• Panel :1. Blend 1 has been placed first.

2. Blend 2 has been placed first.

3. Blend 3 has been placed first.

• Attributes rated on 10 point scale : Overall Likeability, Aroma, Taste, Smoothness, Flavor, Throat-Feel, After-taste and Mouth-feel.

• Attributes rated on 5 point scale : Strengths of Aroma, Taste, smoothness, Flavor and After-Taste.

• Intention to buy attribute(1-Yes, 2-No)

DATA ANALYSIS• Attributes that drive overall preference of vodka blends are found by doing a

regression analysis between overall likeability and all other attributes.OL= 0.391 + 0.07Aroma_Neat + 0.04Aroma_Mixer + 0.11Aroma + 0.29Taste +

0.11Smoothness +

0.07Flavor + 0.09ThroatFeel + 0.05AfterTaste + 0.19MouthFeel

• However, we noticed that few attributes are not contributing much to the model as their standardized beta coefficients are very less.

• We run the step-wise regression to eliminate less contributing attributes and arrive at the best fit model.

OL= 0.478 + 0.334Taste + 0.269MouthFeel + 0.152Aroma + 0.159Smoothness + 0.089Aroma_Neat

• We found that the Taste and MouthFeel are two important drivers for overall preference of vodka.

DATA ANALYSIS95% and 90% Confidence levels for top2(10&9) and top3(10&9&8) ratings

• At 95% Confidence level• Top 2(10&9) Rating: We found there is no significant difference for attributes

(OL, Taste and MouthFeel) across all blends.• Top 3(10&9&8) Rating: We found there is no significant difference for

MouthFeel attribute and Testblend1 is better than Control product for Overall Likeability and Taste.

• At 90% Confidence level• Top 2(10&9) Rating: We found there is no significant difference for MouthFeel

and Taste attributes and Testblend1 is better than Control product for Overall Likeability.

• Top 3(10&9&8) Rating: We found there is significant difference for all attributes. This shows that TestBlend1 is better than control product.

DATA ANALYSIS• We conduct a factor analysis to reduce dimensions and arrive at more

concrete factors.

• Using PCA, we find that the first factor itself explains more than 70% of overall variance

• Taking the first 3 factors we found that the model explains 85.3% of variance

• We recommend not to go for factor analysis as one factor itself explains more than 70% of overall variance.

Component

Initial Eigenvalues

Total% of

VarianceCumulative

%1 6.382 70.906 70.906

2 .947 10.520 81.427

3 .345 3.835 85.261

4 .277 3.077 88.338

5 .269 2.986 91.324

6 .228 2.529 93.853

7 .217 2.409 96.262

8 .178 1.972 98.235

9 .159 1.765 100.000

DATA ANALYSIS

• To predict the purchase intention of vodka blends based on the ratings on different attributes, we did discriminant analysis.

• But we found negative values and very less values for some attributes in standardized co-efficient values and in structure matrix.

• Now, we run the step-wise discriminant analysis to find classify better.

Standardized Canonical Discriminant Function Coefficients

Structure Matrix

Function

Function

1 1Q5A_att2 .337Q5A_att3 .893

Q5A_att3 .373Q5A_att8 .822

Q5A_att4 .139Q5A_att6 .816

Q5A_att5 .057Q5A_att4 .805

Q5A_att6 .212Q5A_att2 .803

Q5A_att7 -.072Q5A_att5 .791

Q5A_att8 .148Q5A_att7 .772

Classification Resultsa,c

Q6_Int_p (Y=1,N=2)

Predicted Group Membership

Total1 2Original Count 1 519 204 723

2 125 1432 1557% 1 71.8 28.2 100.0

2 8.0 92.0 100.0a. 85.6% of original grouped cases correctly

classified.

DATA ANALYSIS

• In step-wise discriminant analysis, flavor and after taste attributes are removed and we got 0.1% increase in predictability.

• If we add the Arom_Neat and Aroma_Mixer, the overall classified levels are getting down(0.2%) and their standardized canonical discriminant function co-efficients are also.

• Hence, we don’t include Aroma_Neat and Aroma_Mixer attributes.

Classification Resultsa,c

Q6_Int_p (Y=1,N=2)

Predicted Group Membership

Total1 2Original Count 1 515 208 723

2 119 1438 1557% 1 71.2 28.8 100.0

2 7.6 92.4 100.0a. 85.7% of original grouped cases correctly

classified.

Structure MatrixStandardized Canonical Discriminant Function

Coefficients

Function

Function

1 1Q5A_att3 .894Q5A_att2 .344

Q5A_att8 .823Q5A_att3 .374

Q5A_att6 .817Q5A_att4 .137

Q5A_att4 .806Q5A_att6 .205

Q5A_att2 .804Q5A_att8 .134

Q5A_att7a.793

Q5A_att5a.773

DATA ANALYSIS

• Additionally, we did cross tabulations and chi-square test of independence between purchase intention and strength attributes(3-Just right).

• We found the following insights,• Despite giving the just right rating on all strength attributes, majority of

respondents chose Not-to-Buy.Purchase Intention

Y=1 N=2

Aroma-Strength 195(25.7%)

182(23.9%)

Taste-Strength 194(25.5%)

190(25%)

Smoothness-Strength

189(24.9%)

169(22.2%)

Flavor-Strength 183(24.1%)

193(25.4%)

AfterTaste-Strength 184(24.2%)

193(25.4%)

Purchase Intention

Y=1 N=2

Aroma-Strength 162(21.3%)

178(23.4%)

Taste-Strength 151(19.9%)

164(21.6%)

Smoothness-Strength

149(19.6%)

180(23.7%)

Flavor-Strength 156(20.5%)

155(20.4%)

AfterTaste-Strength 146(19.2%)

183(24.1%)

Test Blend1 Test Blend2

DATA ANALYSIS

• To find out the reason behind this anomaly, we did a cross tabulation between Main brand and strength attributes.

• But, we observed that Number of respondents saying Yes and No to purchase the new blends are both have Smirnoff as main brand.

Main Brand MagicMoments

Smirnoff

Fuel

Aroma-Strength 11.3% 27.0% 8.3%

Taste-Strength 11.1% 26.7% 8.1%

Smoothness-Strength

11.8% 26.6% 7.1%

Flavor-Strength 11.3% 26.5% 7.4%

AfterTaste-Strength

11.6% 26.7% 8%

Q6_Int_p (Y=1,N=2) * MAIN_BRND Crosstabulation

MAIN_BRND

Total1 2 3Q6_Int_p (Y=1,N=2)

1 Count 138 473 112 723

% of Total 6.1% 20.7% 4.9% 31.7%

2 Count 381 946 230 1557

% of Total 16.7% 41.5% 10.1% 68.3%

Total Count 519 1419 342 2280

% of Total 22.8% 62.2% 15.0% 100.0%

DATA ANALYSIS

• Additionally, we split the whole dataset based on categories such as Centre's, Panel, Ages and main brand.

• If we split dataset based on Centre's, the sample size is getting very low and error levels are getting very high in the model.

• We found the overall likeability is driven by factors as follows.• Across Ages: Taste, Aroma, MouthFeel• Across Main brand:

• Main Brand1 & Main Brand2 – Taste, Aroma, Mouth Feel• Main Brand3 – Taste, ThroatFeel and MouthFeel

• Across Panels:• Panel1 and Panel2 – Taste, Aroma, MouthFeel• Panel 3 – Taste, Smoothness, MouthFeel

DATA ANALYSIS• We have done cross tabulation and chi-square test between Rating of an

attribute(available) to the strength of that attribute and found they are associated.

• We have done cross tab and chi-square for ages and purchase intention, panel and purchase intention, but we didn’t find any association.

• As proportion of Smirnoff drinkers in the population are high, we can normalize and find which factors are driving their likeability.

• We observed that even the total number of respondents are 760, we found that last respondent number in the dataset is 804.(Just an observation )

RECOMMENDATIONS

• The company should go ahead with the replacement of the current blend with the new blend, Test product 1 as it ranks consistently higher in all attributes at 90 % C.I.

• Further thrust areas for product development should be on:

Aroma

Taste

Mouth - feel

As they are the common attributes in overall likeability and purchase intention

ACTION AREAS

• The response anomaly(difference in no of response and respondent number) should be looked into- Could be due to Missing data-points.

• Robustness and validity of scales should be checked- we observed anomalies with respect to Smirnoff users and their purchase intentions.

THANK YOU