2. Hajime Mizuyama
The Chartered Institute of Marketing
• The management process for identifying, anticipating and satisfying
customer requirements profitably
The American Marketing Association
• The activity, set of institutions, and processes for creating,
communicating, delivering, and exchanging offerings that have value
for customers, clients, partners, and society at large
What is marketing?
3. Hajime Mizuyama
What is marketing?
InterfaceInterface
Market
CustomersCustomers
Company
ProviderProvider
Marketing
data
Marketing
mix
4. Hajime Mizuyama
• Mass marketing
• Target marketing
• One-to-one marketing
• Customer relationship marketing
Scopes of marketing
7. Hajime Mizuyama
点数付けしてもらう
– Likert scale
– Semantic differential method
順序付けしてもらう
– 全体順序データ
– ピボット順序データ
選択してもらう
– 二者択一
– 三つ以上の選択肢からの選択
How to obtain research data
8. Hajime Mizuyama
The American Marketing Association
• Marketing research is the function that links the consumer, customer,
and public to the marketer through information - information used to
identify and define marketing opportunities and problems; generate,
refine, and evaluate marketing actions; monitor marketing
performance; and improve understanding of marketing as a process.
Marketing research specifies the information required to address these
issues, designs the method for collecting information, manages and
implements the data collection process, analyzes the results, and
communicates the findings and their implications.
What is marketing research?
9. Hajime Mizuyama
Marketing Technology 2nd Semester 2013
Market Segmentation
• What is market segmentation?
• What is cluster analysis?
• How to perform the analysis with R?
18. Hajime Mizuyama
Marketing Technology 2nd Semester 2013
Perceptual Mapping
• What is perceptual mapping?
• What are FA and MDS? & How to obtain data for them?
• How to perform the analyses with R?
21. Hajime Mizuyama
• 各製品の(主観的)属性ベクトルを入力とするアプローチ
⇒ Principal component analysis (PCA)
Factor analysis (FA), etc.
• 製品間の(非)類似度データを入力とするアプローチ
⇒ Metric MDS,
Non-metric MDS, etc.
How to make a perceptual map
29. Hajime Mizuyama
• 各製品の(主観的)属性ベクトルを入力とするアプローチ
⇒ Principal component analysis (PCA)
Factor analysis (FA), etc.
• 製品間の(非)類似度データを入力とするアプローチ
⇒ Metric MDS,
Non-metric MDS, etc.
How to make a perceptual map
33. Hajime Mizuyama
cities <- read.csv("sample06.csv", row.names=1)
cities <- as.dist(cities)
xy.cities <- cmdscale(cities)
plot(xy.cities, type="n")
text(xy.cities, rownames(xy.cities))
Example analysis with R
34. Hajime Mizuyama
Marketing Technology 2nd Semester 2013
Joint Space Map
• What are joint space maps?
• How linear regression can be used for drawing them?
• How to perform the analysis with R?
35. Hajime Mizuyama
What are joint space maps?
-1 0 1 2
-1012
factor 1
factor2
FA-map
知覚マップから,競合製品間
の差異がどのように認識されて
いるかは見てとれるが,どういう
製品が好まれているかまでは
分からない.
知覚マップの上に選好情報を
追加する(選好の外部分析).
ジョイントスペースマップ
42. Hajime Mizuyama
Marketing Technology 2nd Semester 2013
Conjoint Analysis #1
• How to model a consumer's preference?
• What is conjoint analysis?
• How to simply conduct the analysis with R?
47. Hajime Mizuyama
Conjoint profiles
k 属性1 属性2 属性3
1 M 7 モノ
2 M 7 カラ
3 M 8 モノ
4 M 8 カラ
5 A 7 モノ
6 A 7 カラ
7 A 8 モノ
8 A 8 カラ
9 W 7 モノ
10 W 7 カラ
11 W 8 モノ
12 W 8 カラ
52. Hajime Mizuyama
Marketing Technology 2nd Semester 2013
Conjoint Analysis #2
• How to model a consumer's preference?
• What is conjoint analysis?
• How to simply conduct the analysis with R?
53. Hajime Mizuyama
コンジョイント分析における効用の捉え方(部分効用モデル)
• 全体効用 U を,製品属性(や,その他のマーケティングミックス属性)
1~N に対する部分効用 un の和でモデル化する(ただし,P は価格).
U = u1 + u2 + … + uN - a0 logP = ∑∑ anl×δnl – a0 logP
Overall utility and partial utility
部分効用un
l (属性 n の水準)1 2 3 4
an4
an3
an2
部分効用 un は,例えば,
1番目の水準の値 an1 を
0とおいて,そこからの
相対評価とする.
部分効用 un は,例えば,
1番目の水準の値 an1 を
0とおいて,そこからの
相対評価とする.
このレンジで
各属性の
重要度を
比較する.
58. Hajime Mizuyama
Marketing Technology 2nd Semester 2013
Conjoint Analysis #3
• What are orthogonal arrays?
• How to use the arrays for designing conjoint profiles?
• How to design conjoint profiles with R?
79. Hajime Mizuyama
e e e e e e
完全交絡の図的理解
列ごとの対比(列間平方和)
1 2 3 4 5 ... 列番号
CA B
A
B
C
完全交絡では,属性Cの
主効果と交互作用A×B
とを,独立に評価する
ことができない.
完全交絡では,属性Cの
主効果と交互作用A×B
とを,独立に評価する
ことができない.A×B
80. Hajime Mizuyama
e e e e e e
部分交絡の図的理解
1 2 3 4 5 ... 列番号
CA B
A
B
C
部分交絡では,交互作用
A×B自身の評価は難しい
が,属性Cの主効果の評価
にはさほど影響は与えない.
部分交絡では,交互作用
A×B自身の評価は難しい
が,属性Cの主効果の評価
にはさほど影響は与えない.
列ごとの対比(列間平方和)
83. Hajime Mizuyama
Marketing Technology 2nd Semester 2013
Brand Choice Models #1
• How to model a consumer's choice?
• What are Logit and Probit choice models?
• How to use the models with R?
86. Hajime Mizuyama
二項選択モデルで誤差項がロジスティック分布に従うとすると,
誤差項の分布関数は次式で与えられる:
これに s = ΔV を代入すると,次の二項ロジット選択モデルが得られる:
Binomial logit choice model
)exp(1
1
)()(
s
dfsF
s
)exp()exp(
)exp(
)exp(/)exp(1
1
)exp(1
1
BA
B
BABA
B
VV
V
VVVV
p
)exp()exp(
)exp(
1
BA
A
BA
VV
V
pp
87. Hajime Mizuyama
選択肢AとBの価格を,cA,cB として,効用の確定項が次式になるとしよう:
VA = VA0 - a log(cA) VB = VB0 - a log(cB)
これを,二項ロジット選択モデルにあてはめると次式となる:
(ただし,Δlog(c) = log(cB) – log(cA) で,ΔV0 = VA0 – VB0 )
これはロジスティック回帰分析のモデル式と等価なので,
ロジスティック回帰分析で,係数 a と,定数項の差分 ΔV0 を推定してみよう.
Simple example
)(exp1
1
))log((exp1
1
)exp(1
1
0 baxVcaVV
p
AB
A
x b
88. Hajime Mizuyama
my.data <- read.csv("sample10.csv")
x <- log(my.data[ ,2])-log(my.data[ ,1])
my.data <- cbind(my.data, x)
my.logit <- glm(y ~ x, family=binomial(link="logit"), data=my.data)
summary(my.logit)
my.ca <- 2000:8000
my.x <- log(12000)-log(my.ca)
pa.l <- 1/(1+exp(-(my.logit$coef[1]+my.logit$coef[2]*my.x)))
plot(my.ca, pa.l, type="l")
Example analysis with R
91. Hajime Mizuyama
my.probit <- glm(y ~ x, family=binomial(link="probit"), data=my.data)
summary(my.probit)
pa.p <- pnorm(my.probit$coef[1]+my.probit$coef[2]*my.x)
plot(my.ca, pa.l, type="l")
lines(my.ca, pa.p, type="l", col=2)
Example analysis with R
92. Hajime Mizuyama
Marketing Technology 2nd Semester 2013
Brand Choice Models #2
• How to apply Logit choice model to conjoint analysis?
• How to design choice experiments for conjoint analysis?
• How to conduct the analysis with R?
95. Hajime Mizuyama
Step1: 直交表などを用いて,実験回数と同じ数の製品プロファイルの集
合 S1 を作成する.
Step2: 作成した製品プロファイルの集合を,選択実験の選択肢の数だけ
複製する.選択肢の数を M とすると,S1 = S2 = … = SM とする.
Step3: M 個の集合から,重複しないように,プロファイルを一つずつ抽出
し,選択肢集合とする.
Step4: 各集合からそれぞれ抽出されたプロファイルを除く.集合が空にな
れば終了,そうでなければ,Step3 に戻る.
An approach to designing choice experiments
(参考) Design and Analysis of Choice Experiments Using R: A Brief Introduction
H. Aizaki and K. Nishimura
Agricaltural Information Research, vol.17, no.2, pp. 86-94 (2008)
(参考) Design and Analysis of Choice Experiments Using R: A Brief Introduction
H. Aizaki and K. Nishimura
Agricaltural Information Research, vol.17, no.2, pp. 86-94 (2008)
98. Hajime Mizuyama
OD OS SZ CL PR
101 3 1 2 1
101 1 1 1 1
101 1 2 1 2
102 1 1 2 2
102 1 2 2 1
102 3 1 1 2
103 3 1 1 2
103 2 1 2 1
103 1 1 2 2
104 2 2 2 2
104 3 2 1 1
104 1 1 1 1
A simple example – Conjoint profiles for choice experiments
OS SZ CL PR
Win 7 in Col 定価
Mac 7 in Mon 定価
Mac 8 in Mon 特価
Mac 7 in Col 特価
Mac 8 in Col 定価
Win 7 in Mon 特価
Win 7 in Mon 特価
And 7 in Col 定価
Mac 7 in Col 特価
And 8 in Col 特価
Win 8 in Mon 定価
Mac 7 in Mon 定価
CH
0
0
1
1
0
0
0
0
1
1
0
0
99. Hajime Mizuyama
OD OS SZ CL PR
105 1 1 1 1
105 2 1 1 2
105 2 2 1 1
106 2 1 2 1
106 1 2 1 2
106 3 2 1 1
107 3 2 2 2
107 3 1 1 2
107 2 2 2 2
108 1 2 2 1
108 2 2 1 1
108 3 2 2 2
A simple example – Conjoint profiles for choice experiments
OS SZ CL PR
Mac 7 in Mon 定価
And 7 in Mon 特価
And 8 in Mon 定価
And 7 in Col 定価
Mac 8 in Mon 特価
Win 8 in Mon 定価
Win 8 in Col 特価
Win 7 in Mon 特価
And 8 in Col 特価
Mac 8 in Col 定価
And 8 in Mon 定価
Win 8 in Col 特価
CH
0
1
0
1
0
0
0
1
0
0
1
0
100. Hajime Mizuyama
OD OS SZ CL PR
109 2 2 1 1
109 3 2 2 2
109 1 2 2 1
110 1 2 1 2
110 2 2 2 2
110 3 1 2 1
111 2 1 1 2
111 3 1 2 1
111 2 1 2 1
112 3 2 1 1
112 1 1 2 2
112 2 1 1 2
A simple example – Conjoint profiles for choice experiments
OS SZ CL PR
And 8 in Mon 定価
Win 8 in Col 特価
Mac 8 in Col 定価
Mac 8 in Mon 特価
And 8 in Col 特価
Win 7 in Col 定価
And 7 in Mon 特価
Win 7 in Col 定価
And 7 in Col 定価
Win 8 in Mon 定価
Mac 7 in Col 特価
And 7 in Mon 特価
CH
0
0
1
0
1
0
1
0
0
0
0
1
102. Hajime Mizuyama
Marketing Technology 2nd Semester 2013
Market Basket Analysis
• What are association rules?
• How to evaluate association rules?
• How to perform the analysis with R?