From VOC to IPA: This Beer s For You!

Size: px

Start display at page:

Download "From VOC to IPA: This Beer s For You!"

Brook Horn
6 years ago
Views:

1 From VOC to IPA: This Beer s For You! Joel Smith Statistician Minitab Inc. jsmith@minitab.com 2013 Minitab, Inc.

2 Image courtesy of amazon.com

3 The Data Online beer reviews Evaluated overall and: Appearance Aroma Palate Taste Focus on American Porters 2013 Minitab, Inc.

4 The Data Text counter for descriptive terms i.e. THICK, COFFEE, BLACK Most common became variables Initial cleansing 2013 Minitab, Inc.

5 Count Chart of Citations Citations

9 Descriptive Statistics: HEAD, DARK, BROWN, BLACK, COFFEE, TAN, LIGHT, LACING,... Variable Mean Minimum Q1 Median Q3 Maximum HEAD DARK BROWN BLACK COFFEE TAN BOURBON FRUITx SMOKEx BITTERx SWEETx

10 Are American Porters Homogenous? Image courtesy of drinks.seriouseats.com

11 2013 Minitab, Inc. Stat > Multivariate > Cluster K-Means

12 K-means Cluster Analysis: HEAD, DARK, BROWN, BLACK, COFFEE, TAN, LIGHT, LACING, THICK, CARAMEL, VANILLA, OPAQUE, WHITE, SMOOTH, STRO Standardized Variables Final Partition Number of clusters: 2 Average Maximum Within distance distance Number of cluster sum from from observations of squares centroid centroid Cluster Cluster Distances Between Cluster Centroids Cluster1 Cluster2 Cluster Cluster

13 Cluster Centroids Grand Variable Cluster1 Cluster2 centroid HEAD DARK BROWN BLACK COFFEE DENSE BROWNISH SOUR BUBBLY WOOD HOPx

16 Cluster 1 Cluster 2 BLACK MOCHA OPAQUE THICK ESPRESSO ALCOHOL COFFEE MILK LIGHT CLEAR BROWN AMBER MAHOGANY WHITE BEIGE HOPX NUTTY MILD CARAMEL DRY DARK TOASTED 2013 Minitab, Inc.

17 What Are Customers Differentiating? Image courtesy of finedininglovers.com

18 2013 Minitab, Inc. Stat > Multivariate > Principal Components

19 Variable PC1 PC2 PC3 PC4 PC5 HEAD DARK BROWN BLACK COFFEE TAN LIGHT BROWNISH SOUR BUBBLY WOOD HOPx

20 Principal Component Analysis: HEAD, DARK, BROWN, BLACK, COFFEE, TAN, LIGHT, LACING, THICK, CARAMEL, VANILLA, OPAQUE, WHITE, SMOOTH, Eigenanalysis of the Correlation Matrix Eigenvalue Proportion Cumulative

21 Eigenvalue Scree Plot of HEAD,..., HOPx Component Number

22 Second Component Loading Plot of HEAD,..., HOPx 0.3 VANILLA 0.2 WOOD LACING SUGAR WHITE BLACK SOUR ALCOHOL FLUFFY BROWNISH BUBBLES TAN BEIGE MAHOGANY BROWN HEAD STRONG CLEAR MOLASSES DENSE OPAQUE DARK AMBER MOCHA NUTTY EARTHY LIGHT THICK ESPRESSO HEAVY STICKY BUBBLY MILD MILK CARBONATION TOFFEE CARAMEL BURNT SMOOTH COFFEE TOASTED DRY HOPx CHOCOLATE First Component

24 2013 Minitab, Inc. Stat > Regression > Regression > Best Subsets

25 Best Subsets Regression: Overall versus Appearance, Aroma, Palate, Taste Response is Overall A p p e a P r A a T a r l a n o a s R-Sq R-Sq c m t t Vars R-Sq (adj) PRESS (pred) Mallows Cp S e a e e X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X

26 Regression Analysis: Overall versus Appearance, Aroma, Palate, Taste Stepwise Selection of Terms α to enter = 0.05, α to remove = 0.05 The stepwise procedure added terms during the procedure in order to maintain a hierarchical model at each step. Analysis of Variance Source DF Adj SS Adj MS F-Value P-Value Regression Aroma Palate Taste Error Total Model Summary S R-sq R-sq(adj) R-sq(pred) % 91.52% 91.37%

27 What are the characteristics of the best porter? Image courtesy of beerandbrewing.com

29 2013 Minitab, Inc. Stat > Regression > Regression > Fit Regression Model

31 HI Individual Value Plot of HI

33 HI_2 Individual Value Plot of HI_

34 Regression Analysis: Taste versus ABV, HEAD, DARK, BROWN, BLACK, COFFEE, TAN, LIGHT,... Analysis of Variance Source DF Adj SS Adj MS F-Value P-Value Regression ABV BLACK THICK CARAMEL SMOOTH SOUR Error Total Model Summary S R-sq R-sq(adj) R-sq(pred) % 38.94% 36.89%

36 Regression Analysis: Taste versus ABV, BLACK, THICK, CARAMEL, SMOOTH, SOUR Analysis of Variance Source DF Adj SS Adj MS F-Value P-Value Regression ABV BLACK THICK SMOOTH SOUR SOUR*SOUR SMOOTH*SOUR Error Total Model Summary S R-sq R-sq(adj) R-sq(pred) % 40.72% 38.29%

37 Coded Coefficients Term Coef SE Coef T-Value P-Value VIF Constant ABV BLACK THICK SMOOTH SOUR SOUR*SOUR SMOOTH*SOUR

38 Mean of Taste ABV Main Effects Plot for Taste Fitted Means BLACK THICK SMOOTH SOUR

40 Matrix Plot of ABV, BLACK, THICK, SMOOTH, SOUR ABV BLACK THICK SMOOTH SOUR

Matrix Plot of ABV, BLACK, THICK, SMOOTH, SOUR 12 0.0 0.5 1.0 0.00 0.15 0.30 8 ABV 4 1.0 BLACK 0.

41 Matrix Plot of ABV, BLACK, THICK, SMOOTH, SOUR ABV BLACK THICK SMOOTH SOUR

42 2013 Minitab, Inc. Stat > Regression > Regression > Response Optimizer

44 Response Optimization: Taste Parameters Response Goal Lower Target Upper Weight Importance Taste Maximum Solution Taste Composite Solution ABV BLACK THICK SMOOTH SOUR Fit Desirability % Lower 95% Lower Confidence Prediction Response Fit SE Fit Bound Bound Taste

45 So what does a great porter taste like? Image courtesy of deschutesbrewery.com

46 Summary Data cleaning Graphical analysis, descriptive statistics + manual review Examine homogeneity Clustering + sensory test Differentiation Principal components, regression Optimization Regression Verification Beer drinking 2013 Minitab, Inc.

47 From VOC to IPA: This Beer s For You! Joel Smith Statistician Minitab Inc. jsmith@minitab.com 2013 Minitab, Inc.

Wine-Tasting by Numbers: Using Binary Logistic Regression to Reveal the Preferences of Experts

Wine-Tasting by Numbers: Using Binary Logistic Regression to Reveal the Preferences of Experts When you need to understand situations that seem to defy data analysis, you may be able to use techniques