The Use of Microsoft Excel for Statistical

Analysis of Sensory Data

Abdullah Muzi Marpaung

Food Technology Department – Swiss German University
Rumah Sains Ilma
bangmuzi#@yahoo.com
[email protected]
27 November 2021
 Basic Principle of Sensory Data Analysis
• Type of Data
• Data Distribution
• Inferential Statistics
Statistical Analysis in Excel

Binomial Test

Chi-Square Test
Outline

Ranking Test

Quantitative Descriptive Analysis

Exercise
 Basic Principle of Sensory
Data Analysis
• Just like other data analysis
• Depend on type and distribution of data
• Inferential Statistics
– Probability
– Null Hypothesis
– Type I error: α
– Type II error: β
 Type of Data
Gender

Height Hair type

Weight
Eye color

Detak jantung
Number of teeth
Blood glucose level

Clothing size
Waist to hip ratio

Education Body size

Non-numeric - Nominal

Non-numeric - Ordinal Blood type Shoe size

Numeric - Discrete

Numeric - Continuous
 Data Distribution
• Binomial Distribution
• Normal Distribution
• Chi-Square Distribution
• F Distribution
 1 BBBB B 1 12 0.35
2 BBBB A
3 BBBA B
4 BBAB B 5
5 BABB B 0.3
6 ABBB B 10
7 BBBA A
8 BBAB A
9 BABB A 0.25
10 ABBB A
8
5 Panelist choose 11
12
13
BBAA B
BABA B
ABBA B
10

0.2
one of two

Probability
Frequency
14 BAAB B
15 ABAB B 6
16 AABB B
products 17
18
BBAA A
BABA A
0.15

19 ABBA A
20 BAAA B 4
25 = 32 combination 21
22
ABAA B
AAAB B
10 0.1

23 AABA B
24 ABAA B
2
25 AABA B 0.05
26 AAAB B
27 BAAA A
28 ABAA A
29 AABA A 5 0 0
30 AAAB A 0 1 2 3 4 5
31 AAAA B Number of A chosen
32 AAAA A 1
Probability Mass Function Nominal
of Binomial Distribution

0.45 Number Probability

p = 1/10
0.40 of
1/2 1/3 1/6 1/10
success
0.35
p = 1/6 0 0.00098 0.01734 0.16151 0.34868
0.30
p = 1/3 1 0.00977 0.08671 0.32301 0.38742
0.25 2 0.04395 0.19509 0.29071 0.19371
Density

p = 1/2
0.20 3 0.11719 0.26012 0.15505 0.05740
0.15
4 0.20508 0.22761 0.05427 0.01116
5 0.24609 0.13656 0.01302 0.00149
0.10
6 0.20508 0.05690 0.00217 0.00014
0.05 7 0.11719 0.01626 0.00025 0.00001
0.00 8 0.04395 0.00305 0.00002 0.00000
0 1 2 3 4 5 6 7 8 9 10 9 0.00977 0.00034 0.00000 0.00000
Number of Success
10 0.00098 0.00002 0.00000 0.00000
 0.45

0.40

0.35

0.30

Density
0.25

Probability 0.20

0.15
Density Function 0.10

of Normal 0.05

0.00

Distribution -4 -3 -2 -1 0
X
1 2 3 4

Continuous
 Probability Density Function of χ 2

Distribution
Nominal Ordinal Discrete Continuous
0.50 k
0.45 k=1 χ2
0.40 1 2 3 5 10
k=2
0.35 0 1.00000 0.50000 0.00000 0.00000 0.00000
0.30 1 0.24197 0.30327 0.24197 0.08066 0.00079
Density

0.25 2 0.10378 0.18394 0.20755 0.13837 0.00766

0.20 k=3 3 0.05139 0.11157 0.15418 0.15418 0.02353
4 0.02700 0.06767 0.10798 0.14398 0.04511
0.15 k=5
k = 10 5 0.01464 0.04104 0.07322 0.12204 0.06680
0.10
6 0.00811 0.02489 0.04865 0.09730 0.08402
0.05
7 0.00455 0.01510 0.03187 0.07437 0.09441
0.00 8 0.00258 0.00916 0.02067 0.05511 0.09768
0 1 2 3 4 5 6 7 8 9 10
9 0.00148 0.00555 0.01330 0.03989 0.09490
χ2
10 0.00085 0.00337 0.00850 0.02833 0.08773
 1.4 df1 = 40, df2 = 40
1.2 df1 = 30, df2 = 30
Probability 1 df1 = 20, df2 = 20
Density 0.8 df1 = 10, df2 = 10
Function of F 0.6 df1 = 5, df2 = 5
Distribution 0.4

Continuous 0.2

0
0 2 4 6
Creating Data
Distribution in Excel
 Inferential
Statistics
• Inferential statistics is a
way of making inferences
about populations based
on samples
• There is a risk to incorrect
decision
• Type I error = α =
significance level, usually
5%
 α = 0.05 α = 0.05
α = 0.05 α = 0.05

0 1 2 3 4 5 6 7 8 9 10 -4 -3 -2 -1 0 1 2 3 4
X 0 5 10 15 20
X
0 2 4 6

Statistical Analysis: Test the Null

Hypothesis
Statistical Analysis
Provided by
Microsoft Excel
(Related to Sensory Data)
 Function
• Binomial test
• Chi-Square test
• t-test
 Add in

• t- test
• ANOVA
How to Install Excel Add-in
 Binomial Test
Duo-trio - Paired comparison - Triangle - Two
out of five – Paired preference
Chi-Square Test
Just About Right, Proportion
 Just About Right

Too Weak Just About Right Too Strong

 Ranking or Ordinal Test
(Friedman’s and Wilcoxon
Signed-Rank Test)

Ranking – Hedonic Scale – LAM Scale

 Problem with score

Score Ranking
Judge A B C A B C
1 9 8 7 1 2 3
2 9 8 7.5 1 2 3
3 8 7.5 7.5 1 2.5 2.5
4 9 9 8.5 1.5 1.5 3
5 4 8 9 3 2 1
Average 7.8 8.1 7.9 1.5 2 2.5
1
4
5
6
7
8

2
3
9

Not appropriate for ANOVA

 Greatest imaginable like 100 Greatest imaginable like 100

Not appropriate for ANOVA

Like extremely 74.2 Like extremely 87.11

Labeled Magnitude (LAM) Scale

Labeled Magnitude (LAM) Scale

Like very much 56.1 Like very much 78.06

Like moderately 36.2 Like moderately 68.12

Discrete
Discrete
Like slightly 11.2 Like slightly 55.62
Neither like nor dislike 0 Neither like nor dislike 50
Dislike slightly - 10.6 Dislike slightly 44.69

Dislike moderately - 31.9 Dislike moderately 34.06

Dislike very much - 55.5 Dislike very much 22.25

Dislike extremely - 75.5 Dislike extremely 12.25

Greatest imaginable Greatest imaginable

- 100 0
dislike dislike
 Post-Hoc Analysis
Church and Wike (1979):
Wilcoxon test was the best pairwise multiple comparison procedure when compared with the Wilcoxon -
Nemenyi-McDonald-Thompson and sign tests

Pereira et al. (2015)

One way to characterize the post-hoc tests is as liberal (powerful) or conservative (lack of power). The
common post-hoc tests we focus on this article can be listed from most liberal to most conservative, although
this varies somewhat depending on the situation. In this work, we have identified three groups of post-hoc
procedures:
Group 1 (extremely conservative, much lower power): Binomial Sign test.
Group 2 (more conservative, lower power): Wilcoxon test with Holm, Hochberg, Hommel or Bonferroni
correction. The differences in power between all these pvalue adjustments are in practice rather small.
• Group 3 (more liberal, larger power): Bonferroni-Dunn test, Fisher’s LSD test and Wilcoxon-Nemenyi-
McDonald-Thompson test (Tukey-HSD). Fisher’s LSD test and Tukey-HSD revealed to be the two most
powerful procedures.
Ranking test in
Excel
 Sw eet
7 Coke
6
Viscosity Sour PC

Quantitative
5
Tops
4
Wegmans
3
2
Carbonation 1 Vanilla

Descriptive
0

Coating Cola spice

Analysis Drying Citrus

• Judge as objective instrument NAME _________________________________________ DATE______________ CODE _____________

• Unstructured Line
PLEASE PLACE a vertical line across the horziontal line at the point whice best describe that attribute in the sample

INITIAL TOTAL
FLAVOR Weak strong

• Assumed as normal distribution IMPACT

SWEETNESS Weak strong

CARAMEL

• Spider Web CORN/BROWN SUGAR Weak

TOAST
Slight
strong

Burnt/Overtoas

• Data analysis CARDBOARD

Slight Extreme

– ANOVA, then Post Hoc

PROCESSED GRAIN/
CORN Weak Strong

RAW CORN
Weak Strong

Analysis
(Under Processed)

SALT
Weak Strong

– Paired t-test BUTTERY

Weak Strong

OIL
Weak Strong
Exercise
 Paired Preference Test

Initial prototypes (A and B) of a new product were examined by a small

panel of consumers for preference. A paired preference test was used.
Twenty-seven participants completed the test and 20 out of the 27
chose the A product sample. Does this indicate preference of A over B?
 Paired Preference Test

Kompetitor mengklaim bahwa produk mereka lebih disukai daripada

produk Anda. Untuk menguji klaim ini Anda melakukan blind test
terhadap 150 orang konsumen. Hasilnya 82 orang memilih produk
competitor, dan 68 orang memilih produk Anda. Apakah Anda setuju
dengan klaim kompetitor tersebut?
 Triangle Test

A laboratory-based sensory experiment with trained assessors

compared two products (A and B) for evidence of a difference. Coded
samples (A, B) were presented in balanced order, for a triangle test
(where A is the odd sample) to 15 assessors. Eight chose A as the odd
sample and seven chose B. On the basis of these data, do the products
differ?
 Chi-Square Test

Sebuah perusahaan makanan ringan mengklaim 7 dari 10 orang

mengonsumsi produk mereka paling tidak satu kali dalam seminggu. Sebuah
perusahaan survey yang disewa oleh competitor melakukan survey dan
mendapatkan bahwa 62 dari 100 orang mengonsumsi makanan ringan dari
perusahaan tersebut paling tidak sekali dalam seminggu. Apakah klaim
perusahaan tersebut terbukti?
 Hedonic Rating

A hedonic rating test was carried out by a

panel of assessors on three product
samples. The nine-point hedonic scale
was used with category box anchors
(‘dislike extremely’ (1) to ‘like
extremely’ (9)). Subsamples were
presented to ten participants in
randomized order. The objective was to
answer the question: ‘Is there a
significant difference in liking?’
References
References
References
References
Terima kasih
 File and Template
• Data distribution
• Binomial test
• Chi Square test
• Friedman’s & Wilcoxon test
• QDA

https://drive.google.com/drive/folders/1G8xNZ-
BjOstIVFAYo9rcC4oL3FrC8vQZ?usp=sharing
 Workshop ini diselenggarakan sebagai kegiatan pengabdian masyarakat
dan penggalangan dana untuk pembangunan Sekolah Menengah
Wirausaha di Sungai Enam, Bintan Timur, Bintan, Kepulauan Riau

Donasi dapat disampaikan melalui Bank Muamalat nomor 3110017935 an

Yayasan Himayah Rahmah Semesta