Choice modelling -and example

1
Choice modelling -and example
2
Background

• Bonlac changing processed cheese and natural
  cheddar offering from Bega to Perfect Cheese
• Previous research has
• Explored an appropriate positioning for Perfect
  Cheese
• Identified the optimal pack design
• Further research is required to
• Understand market response to the new range of
  Perfect Cheese in terms of
• Price sensitivity
• Market share potential
• Cannibalisation effects
• In addition, feedback on sensory performance of
  Perfect Cheese products relative to competitors,
  in order to support positioning platform ( not
  discussed today)

3
Pricing Objectives

• To understand the impact of launching of Perfect
  in the Processed Cheese and Light Cheddar Block
  markets
• Understanding initial impact (pre-trial)
• Understand longer term impact (post-trial)
• Understand the price sensitivity of each user
  group

4
Sensory Objectives

• To evaluate the Perfect cheese slice and block
  products relative to competitive offerings in
  terms of
• Acceptability (unbranded vs branded)
• Sensory profiles
• Relative to consumer ideals
• Purchase intentions
• Ability to support brand positioning expectations

5
Method

• Central location test at Takapuna

6
Sample Population

• N30 each of
• Light Slice users
• Super Light Slice users
• Cheddar Slice users
• Reduced Fat Cheddar Block users
• Sample population
• Females MHS, 20-65 years
• Mix of household types (mainly families with
  kids)

7
Pricing Methodology

• 15 shelves - pre/post presented to each of 30
  people in 4 user groups
• Light Slices
• Super Light Slices
• Cheddar Slices
• Light Cheddar Block
• In each shelf range of prices consumers get to
  choose only one
• Imitates shopping experience
• Idealised situations (100 awareness of Perfect)
• House-brands included

8
Introduction
9
( 4 of the 15 scenarios)
10
Whoa there! - How did we get to this conclusion?

• 3 brands of interest Mainland/Chesdale and
  Perfect
• The other 2, Pams and First Choice area at
  fixed, lower prices, prices
• Decided to go with 3 price (low 1.99/medium
  2.29 /high 2.59) points/brand
• Why?
• Therefore we have 33 27 possible combinations
• Decided to choose a sample of 15 to reduce
  respondent fatigue and to ensure we could measure
  all 2 order interaction effects
• eg does a high price of Chesdale result in
  different pricing response for Perfect than if it
  were a low price
• This phenomenon is quite common so needs to be
  taken into account

11
The design
Discuss
12
The data - raw
13
The data - how its needed for proc Phreg in SAS
14
Some points

• Note that we have decided to mode/post data
  together
• Not how the data is agrregated now
• Compare this to what we have
• Preprice 1
• 
  Cumulative Cumulative
• PRE1
  Frequency Percent Frequency Percent
• 
  ƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒƒ
  ƒƒƒƒƒƒƒ
• 1
  8 26.67 8 26.67
• 2
  7 23.33 15 50.00
• 3
  1 3.33 16 53.33
• 4
  12 40.00 28 93.33
• 5
  2 6.67 30 100.00
• 
  Preprice 2
• 
  Cumulative Cumulative
• PRE2
  Frequency Percent Frequency Percent

15
Some Points

• The variable T denote 1 choice, 2 no choice
• The resulting doubling up of all rows
• The variable SET represents the appropriate
  scenario
• For each scenario there are 10 52 rows
• Variables like CSD_MLD represents Chesdales
  effect on effect Mainland and so is in the
  relevant rows for Mainland but is Chesdales
  price
• Remind me to give you a Splus function called
  SAS.DCM.FORMAT that helps format the appropriate
  design matrix for this data

16
Some more code

• data temp
• set hold.cslmodel
• PR_Mld2 PR_Mld2
• PR_Prf2 PR_Prf2
• PR_Anc2 PR_Anc2
• PR_FC2 PR_FC2
• PR_Pam2 PR_Pam2
• DMld POSTMld
• DPrf POSTPrf
• DAnc POSTAnc
• Dpam POSTpam
• Dfc POSTFC
• DPR_Mld POSTPR_Mld
• DPR_Prf POSTPR_Prf
• DPR_Anc POSTPR_Anc
• DPR_Pam POSTPR_Pam
• DPR_FC POSTPR_FC
• .
• .

17
Analysing the data

• Saving this data
• data hold.cslmodel
• set temp
• run
• Now we are ready to start finding the correct
  model
• trial and error to obtain the correct model
• proc phreg data hold.cslmodel outest betas
  nosummary
• strata set
• model tt(2)
• CsD Mld Prf FC Pam
• PR_CsD PR_Mld PR_Prf PR_FC PR_pam
• PR_CsD2 PR_Mld2 PR_FC2 PR_pam2
• DCsD DMld DPrf Dpam Dfc
• DPR_CsD DPR_Mld DPR_Prf DPR_FC DPR_Pam
• DPR_CsD2 DPR_Mld2 DPR_Prf2 DPR_FC2 DPR_pam2
• DCsD_Mld DCsD_Prf DCsD_FC DCsD_Pam
• DMld_CsD DMld_Prf DMld_FC DMld_Pam

18
Analysing the data

• The final model
• proc phreg data hold.cslmodel outest betas
  nosummary
• strata set
• model tt(2)
• CsD Mld Prf FC Pam
• PR_CsD PR_Mld PR_Prf PR_FC PR_pam
• PR_CsD2 PR_Mld2 PR_FC2 PR_pam2
• DPrf
• DCsD_Prf
• /ties breslow
• freq freq
• run

19
Analysing the data

• Output
• Analysis of Maximum Likelihood Estimates
• 
  Parameter Standard
  Hazard
• Variable DF
  Estimate Error Chi-Square Pr gt
  ChiSq Ratio
• CSD 1
  62.03446 10.46407 35.1451
  lt.0001 8.734E26
• MLD 1
  53.51747 11.68024 20.9936
  lt.0001 1.747E23
• PRF 1
  12.41570 1.09299 129.0359
  lt.0001 246643.1
• FC 1
  1.15688 0.16214 50.9094 lt.0001
  3.180
• PAM 0
  0 . . .
  .
• PR_CSD 1
  -48.72340 9.27818 27.5772
  lt.0001 0.000
• PR_MLD 1
  -41.00351 10.42944 15.4568
  lt.0001 0.000
• PR_PRF 1
  -5.23230 0.49984 109.5801
  lt.0001 0.005
• PR_FC 0
  0 . . .
  .
• PR_PAM 0
  0 . . .
  .
• PR_CSD2 1
  9.65004 2.03784 22.4241 lt.0001
  15522.40
• PR_MLD2 1
  7.85671 2.30708 11.5972 0.0007
  2583.017
• PR_FC2 0
  0 . . .
  .

20
Turning this into something meaningfull
21
Presenting the data
22
Presenting the data
23
(No Transcript)