From VOC to IPA - Minitab VOC to IPA: This Beer’s For You! Joel Smith Statistician –Minitab Inc. [email protected] ...

© 2013 Minitab, Inc.

From VOC to IPA:

This Beer’s For You!

Joel Smith

Statistician – Minitab Inc.

[email protected]

Image courtesy of amazon.com


The Data

► Online beer reviews

► Evaluated overall and:

• Appearance

• Aroma

• Palate

• Taste

► Focus on American Porters


The Data

► Text counter for “descriptive” terms

• i.e. “THICK”, “COFFEE”, “BLACK”

► Most common became variables

► Initial cleansing

20191817161514131211109876543210

9000

8000

7000

6000

5000

4000

3000

2000

1000

0

Citations

Co

un

t

Chart of Citations

Descriptive Statistics: HEAD, DARK, BROWN, BLACK,

COFFEE, TAN, LIGHT, LACING, ...

Variable Mean Minimum Q1 Median Q3 Maximum

HEAD 0.88731 0.62500 0.86111 0.89189 0.92708 1.00000

DARK 0.62370 0.23077 0.57576 0.63043 0.68182 0.91667

BROWN 0.61144 0.23529 0.54545 0.61194 0.68421 0.94444

BLACK 0.48373 0.00000 0.35294 0.50000 0.60556 0.95000

COFFEE 0.3561 0.0000 0.2188 0.3500 0.4667 0.9474

TAN 0.43249 0.08333 0.37398 0.43478 0.48276 0.84615

…

BOURBON 0.03454 0.00000 0.00000 0.00000 0.00000 0.80556

…

FRUITx 0.06493 0.00000 0.02646 0.05634 0.08696 0.40000

SMOKEx 0.1412 0.0000 0.0323 0.0649 0.1250 0.8500

BITTERx 0.11349 0.00000 0.05882 0.10784 0.16197 0.36842

SWEETx 0.22135 0.00000 0.15217 0.21053 0.28049 0.65217

Are American Porters

Homogenous?

Image courtesy of drinks.seriouseats.com


Stat > Multivariate > Cluster K-Means

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

Cluster1 194 7847.713 6.067 10.928

Cluster2 132 5557.371 6.141 12.055

…

Distances Between Cluster Centroids

Cluster1 Cluster2

Cluster1 0.0000 3.3753

Cluster2 3.3753 0.0000

Cluster Centroids

Grand

Variable Cluster1 Cluster2 centroid

HEAD 0.0031 -0.0046 0.0000

DARK -0.1523 0.2238 -0.0000

BROWN -0.3743 0.5501 0.0000

BLACK 0.5733 -0.8426 -0.0000

COFFEE 0.1443 -0.2121 -0.0000

…

DENSE 0.0573 -0.0843 0.0000

BROWNISH -0.0066 0.0096 0.0000

SOUR 0.0407 -0.0598 0.0000

BUBBLY 0.0975 -0.1433 -0.0000

WOOD 0.0039 -0.0057 0.0000

HOPx -0.2007 0.2950 0.0000


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

What Are Customers

Differentiating?

Image courtesy of finedininglovers.com


Stat > Multivariate > Principal Components

Variable PC1 PC2 PC3 PC4 PC5

HEAD 0.034 0.025 0.063 -0.445 -0.121

DARK 0.101 -0.068 -0.054 -0.234 -0.031

BROWN 0.254 0.047 0.103 -0.250 -0.069

BLACK -0.404 0.013 0.029 -0.004 -0.008

COFFEE -0.122 -0.302 -0.283 -0.033 -0.115

TAN 0.037 0.034 -0.101 -0.325 0.243

LIGHT 0.266 -0.095 -0.101 0.025 -0.002

…

BROWNISH 0.068 0.042 0.206 -0.156 -0.106

SOUR 0.010 0.075 0.012 0.051 0.199

BUBBLY -0.022 -0.126 0.048 0.099 0.146

WOOD -0.017 0.219 0.120 0.188 0.192

HOPx 0.134 -0.287 0.135 -0.025 -0.039

Principal Component Analysis: HEAD, DARK, BROWN,

BLACK, COFFEE, TAN, LIGHT, LACING, THICK, CARAMEL,

VANILLA, OPAQUE, WHITE, SMOOTH,

Eigenanalysis of the Correlation Matrix

Eigenvalue 4.2617 2.5063 2.0396 1.7995 1.6684 ...

Proportion 0.097 0.057 0.046 0.041 0.038 ...

Cumulative 0.097 0.154 0.200 0.241 0.279 ...

4035302520151051

4

3

2

1

0

Component Number

Eig

en

valu

e

Scree Plot of HEAD, ..., HOPx

0.40.30.20.10.0-0.1-0.2-0.3-0.4-0.5

0.3

0.2

0.1

0.0

-0.1

-0.2

-0.3

-0.4

First Component

Seco

nd

Co

mp

on

en

t

HOPx

WOOD

BUBBLY

SOUR

BROWNISH

DENSE

SUGAR

NUTTY

FLUFFY

EARTHYESPRESSO

ALCOHOL

CHOCOLATE

STICKYHEAVY

AMBER

BEIGE

MOLASSES

MILK

TOASTEDDRY

BUBBLESMAHOGANY

TOFFEE

CLEAR

MILD

MOCHA

CARBONATION

BURNT

STRONG

SMOOTH

WHITE

OPAQUE

VANILLA

CARAMEL

THICK

LACING

LIGHT

TAN

COFFEE

BLACK

BROWN

DARK

HEAD

Loading Plot of HEAD, ..., HOPx


Stat > Regression > Regression > Best Subsets

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

1 91.2 91.2 2.5 91.1 13.8 0.087905 X

1 81.7 81.7 5.3 81.4 377.1 0.12684 X

1 68.8 68.7 9.0 68.4 870.3 0.16565 X

1 57.4 57.3 12.3 56.8 1305.9 0.19355 X

2 91.4 91.4 2.5 91.3 8.0 0.087015 X X

2 91.3 91.3 2.5 91.1 12.2 0.087565 X X

2 91.2 91.2 2.6 91.0 15.7 0.088033 X X

2 82.5 82.3 5.1 81.9 350.9 0.12444 X X

2 81.7 81.6 5.3 81.3 379.1 0.12703 X X

3 91.6 91.5 2.5 91.4 3.0 0.086214 X X X

3 91.4 91.4 2.5 91.2 8.9 0.086998 X X X

3 91.3 91.2 2.5 91.1 13.7 0.087640 X X X

3 82.5 82.3 5.2 81.8 352.0 0.12456 X X X

4 91.6 91.5 2.5 91.3 5.0 0.086349 X X X X

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 3 26.1102 8.70341 1170.93 0.000

Aroma 1 0.0832 0.08325 11.20 0.001

Palate 1 0.0522 0.05222 7.03 0.008

Taste 1 2.6086 2.60861 350.95 0.000

Error 322 2.3934 0.00743

Total 325 28.5036

Model Summary

S R-sq R-sq(adj) R-sq(pred)

0.0862144 91.60% 91.52% 91.37%

What are the characteristics of

the best porter?

Image courtesy of beerandbrewing.com


Stat > Regression > Regression > Fit Regression Model

0.30

0.25

0.20

0.15

0.10

0.05

0.00

HI

Individual Value Plot of HI

0.12

0.10

0.08

0.06

0.04

0.02

0.00

HI_

2

Individual Value Plot of HI_2

Regression Analysis: Taste versus ABV, HEAD,

DARK, BROWN, BLACK, COFFEE, TAN, LIGHT, ...



Regression 6 11.4860 1.91433 31.40 0.000

ABV 1 2.1544 2.15437 35.33 0.000

BLACK 1 1.3679 1.36792 22.43 0.000

THICK 1 0.4965 0.49647 8.14 0.005

CARAMEL 1 0.3403 0.34030 5.58 0.019

SMOOTH 1 0.8143 0.81431 13.36 0.000

SOUR 1 0.6925 0.69250 11.36 0.001

Error 280 17.0725 0.06097

Total 286 28.5585

Model Summary


0.246928 40.22% 38.94% 36.89%

Regression Analysis: Taste versus ABV, BLACK, THICK,

CARAMEL, SMOOTH, SOUR



Regression 7 12.0431 1.72044 29.06 0.000

ABV 1 1.8707 1.87071 31.60 0.000

BLACK 1 1.9144 1.91436 32.34 0.000

THICK 1 0.4240 0.42398 7.16 0.008

SMOOTH 1 0.8472 0.84720 14.31 0.000

SOUR 1 0.9757 0.97565 16.48 0.000

SOUR*SOUR 1 0.7480 0.74802 12.64 0.000

SMOOTH*SOUR 1 0.3530 0.35297 5.96 0.015

Error 279 16.5154 0.05920

Total 286 28.5585

Model Summary


0.243300 42.17% 40.72% 38.29%

Coded Coefficients

Term Coef SE Coef T-Value P-Value VIF

Constant 3.8227 0.0162 235.43 0.000

ABV 0.0903 0.0161 5.62 0.000 1.25

BLACK 0.0952 0.0167 5.69 0.000 1.35

THICK 0.0422 0.0158 2.68 0.008 1.20

SMOOTH 0.0564 0.0149 3.78 0.000 1.07

SOUR -0.1147 0.0283 -4.06 0.000 3.86

SOUR*SOUR 0.02772 0.00780 3.55 0.000 4.05

SMOOTH*SOUR 0.0446 0.0183 2.44 0.015 1.51

1284

4.1

4.0

3.9

3.8

3.7

3.6

1.00.50.0 0.40.20.0 0.300.150.00 0.300.150.00

ABV

Mean

of

Tast

e

BLACK THICK SMOOTH SOUR

Main Effects Plot for TasteFitted Means

12

8

4

0.300.150.00

1.00.50.0 0.300.150.00

1.0

0.5

0.00.4

0.2

0.0

0.30

0.15

0.00

1284

0.30

0.15

0.00

0.40.20.0

ABV

BLACK

THICK

SMOOTH

SOUR

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

12

8

4

0.300.150.00

1.00.50.0 0.300.150.00

1.0

0.5

0.00.4

0.2

0.0

0.30

0.15

0.00

1284

0.30

0.15

0.00

0.40.20.0

ABV

BLACK

THICK

SMOOTH

SOUR

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


Stat > Regression > Regression > Response Optimizer

Response Optimization: Taste

Parameters

Response Goal Lower Target Upper Weight Importance

Taste Maximum 2.61628 4.65 1 1

Solution

Taste Composite

Solution ABV BLACK THICK SMOOTH SOUR Fit Desirability

1 12.2 0.95 0.392857 0.24 0 4.73742 1

95% Lower 95% Lower

Confidence Prediction

Response Fit SE Fit Bound Bound

Taste 4.7374 0.0891 4.5903 4.3098

So what does a great porter

taste like?

Image courtesy of deschutesbrewery.com


Summary

► Data cleaning

• Graphical analysis, descriptive statistics + manual review

► Examine homogeneity

• Clustering + sensory test

► Differentiation

• Principal components, regression

► Optimization

• Regression

► Verification

• Beer drinking


From VOC to IPA:

This Beer’s For You!

Joel Smith

Statistician – Minitab Inc.

[email protected]

From VOC to IPA - Minitab VOC to IPA: This Beer’s For You! Joel Smith Statistician –Minitab Inc. [email protected] ...

Documents