Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
A	
  Presenta*on	
  from	
  
The	
  Fes*val	
  of...
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
An Introduction to Latent
Class Analysis for Mark...
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
Overview	
  
•  Latent	
  class	
  analysis	
  ve...
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
Latent	
  class	
  analysis	
  turns	
  data	
  
...
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
Cluster	
  	
  
Analysis	
  
Latent	
  
Class	
  ...
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
Cluster	
  Analysis	
  versus	
  Latent	
  Class	...
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
	
  0 	
  5 	
  10 	
  15 	
  20 	
  25 	
  30 	
...
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
	
  0 	
  5 	
  10 	
  15 	
  20 	
  25 	
  30 	
...
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
	
  0 	
  5 	
  10 	
  15 	
  20 	
  25 	
  30 	
...
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
Specify	
  number	
  of	
  
clusters	
  (k)	
  
R...
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
	
  0 	
  5 	
  10 	
  15 	
  20 	
  25 	
  30 	
...
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
k-­‐Means	
  Cluster	
  Analysis	
  
Specify	
  n...
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
	
  0 	
  5 	
  10 	
  15 	
  20 	
  25 	
  30 	
...
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
Specify	
  number	
  of	
  
clusters	
  (k)	
  
R...
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
	
  0 	
  5 	
  10 	
  15 	
  20 	
  25 	
  30 	
...
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
	
  0 	
  5 	
  10 	
  15 	
  20 	
  25 	
  30 	
...
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
Specify	
  number	
  of	
  
clusters	
  (k)	
  
R...
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
	
  0 	
  5 	
  10 	
  15 	
  20 	
  25 	
  30 	
...
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
Cluster	
  	
  
Analysis	
  
Latent	
  
Class	
  ...
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
missing	
  
values	
  
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
How	
  many	
  clusters	
  
(or	
  classes)	
  ca...
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
Missing	
  values	
  and	
  latent	
  class	
  
a...
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
Missing	
  values	
  and	
  cluster	
  analysis	
...
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
distribu*onal	
  
assump*ons	
  
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
Distribu*onal	
  	
  
assump*ons	
  
•  Basic	
  ...
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
Example:	
  Categorical	
  data	
  
Data	
  
Shop...
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
Specify	
  number	
  of	
  
classes	
  (k)	
  
Ra...
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
ID
Shop
Understand
Key
Interest
Value
… … … … … …...
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
Compu*ng	
  the	
  probability	
  of	
  each	
  
...
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
Applica*on	
  
n	
  =	
  1,145	
  market	
  resea...
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
Cluster	
  
Analysis	
  
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
Numeric	
  
Assump*on	
  
Lowest	
  price	
  
Pre...
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
Categorical	
  
Assump*on	
  
Lowest	
  price	
  ...
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
Ranking	
  
Assump*on	
  
Lowest	
  price	
  
Pre...
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
Latent	
  class	
  analysis	
  sobware	
  
Produc...
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
Cluster	
  	
  
Analysis	
  
Latent	
  Class	
  A...
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
Thank you
Tim Bock
Q
Tim Bock, Q, Australia
Festival of NewMR 2012 – Training Day – Session 1
Tim Bock, Q
www.q-researchsoftware.com
tim.bock@q...
Upcoming SlideShare
Loading in …5
×

Tim bock training day - 2012

374 views

Published on

An Introduction to Latent Class Analysis for Marketing Segmentation

Published in: Marketing
  • Be the first to comment

  • Be the first to like this

Tim bock training day - 2012

  1. 1. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1 A  Presenta*on  from   The  Fes*val  of  NewMR  –  Training  Day   3  December  2012   All  copyright  owned  by  The  Future  Place  and  the  presenters  of  the  material   For  more  informa:on  about  NewMR  events  visit  NewMR.org   Sponsored   by:   See    the  eXhib:on  for   booths  from  media   partners  &  supporters   An  Introduc*on  to  Latent  Class  Analysis  for   Marke*ng  Segmenta*on   Tim  Bock,  Q      
  2. 2. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1 An Introduction to Latent Class Analysis for Marketing Segmentation Tim Bock, Q www.q-researchsoftware.com tim.bock@q-researchsoftware.com +61 425 241 989
  3. 3. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1 Overview   •  Latent  class  analysis  versus  cluster  analysis   –  Theore:cal  difference:  probabili:es   –  Prac:cal  differences:   •  Non-­‐numeric  data  (e.g.,  categorical  data)   •  Missing  values   •  Applica:on:  what  do  research  buyer’s  want?   –  Missing  values   –  Response  bias  
  4. 4. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1 Latent  class  analysis  turns  data   into  segments   Worriers   Concerned     with  decay   preven:on   Sociables      Concerned      with        tooth            colour   Sensory   Concerned   with   flavour   Independent   Concerned     with  price   Adapted  from:  Haley,  R.  I.  (1968).  "Benefit  Segmenta:on:  A  Decision   Oriented  Research  Tool."  Journal  of  Marke:ng  30(July):  30-­‐35.      
  5. 5. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1
  6. 6. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1 Cluster     Analysis   Latent   Class   Analysis  
  7. 7. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1 Cluster  Analysis  versus  Latent  Class   Analysis  for  segmenta*on   •  Latent  class  analysis  is  theore:cally  superior   –  Clearly-­‐stated  assump:ons   –  Cluster  analysis  is  inconsistent  with  elementary  laws  of  probability     (in  par:cular,  Bayes’  Theorem)   •  Latent  class  analysis  so_ware  is  superior   –  Any  type  of  data  (via  distribu:onal  assump:ons):  Categorical,   Conjoint,  Choice,  MaxDiff,  Rankings,  etc.   –  “Mixed”  data  (e.g.,  categorical  and  numeric)   –  Missing  values   –  Response  biases  
  8. 8. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1  0  5  10  15  20  25  30  35   25   20   15   10   5   0   Specify  number  of   clusters  (k)   k-­‐Means  Cluster  Analysis  
  9. 9. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1  0  5  10  15  20  25  30  35   25   20   15   10   5   0   Specify  number  of   clusters  (k)   k-­‐Means  Cluster  Analysis   Randomly  allocate   respondents  to  clusters  
  10. 10. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1  0  5  10  15  20  25  30  35   25   20   15   10   5   0   Specify  number  of   clusters  (k)   Randomly  allocate   respondents  to  clusters   k-­‐Means  Cluster  Analysis  
  11. 11. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1 Specify  number  of   clusters  (k)   Randomly  allocate   respondents  to  clusters   Compute  cluster  means   k-­‐Means  Cluster  Analysis    0  5  10  15  20  25  30  35   25   20   15   10   5   0  
  12. 12. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1  0  5  10  15  20  25  30  35   25   20   15   10   5   0   Specify  number  of   clusters  (k)   Randomly  allocate   respondents  to  clusters   Compute  cluster  means   k-­‐Means  Cluster  Analysis  
  13. 13. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1 k-­‐Means  Cluster  Analysis   Specify  number  of   clusters  (k)   Randomly  allocate   respondents  to  clusters   Compute  cluster  means   Allocate  respondents  to   most  similar  clusters    0  5  10  15  20  25  30  35   25   20   15   10   5   0  
  14. 14. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1  0  5  10  15  20  25  30  35   25   20   15   10   5   0   k-­‐Means  Cluster  Analysis   Specify  number  of   clusters  (k)   Randomly  allocate   respondents  to  clusters   Compute  cluster  means   Allocate  respondents  to   most  similar  clusters  
  15. 15. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1 Specify  number  of   clusters  (k)   Randomly  allocate   respondents  to  clusters   Allocate  respondents  to   most  similar  clusters   k-­‐Means  Cluster  Analysis   Compute  cluster  means    0  5  10  15  20  25  30  35   25   20   15   10   5   0  
  16. 16. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1  0  5  10  15  20  25  30  35   25   20   15   10   5   0   Specify  number  of   clusters  (k)   Randomly  allocate   respondents  to  clusters   Allocate  respondents  to   most  similar  clusters   k-­‐Means  Cluster  Analysis   Compute  cluster  means  
  17. 17. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1  0  5  10  15  20  25  30  35   25   20   15   10   5   0   Specify  number  of   clusters  (k)   Randomly  allocate   respondents  to  clusters   Allocate  respondents  to   most  similar  clusters   k-­‐Means  Cluster  Analysis   Compute  cluster  means   Repeat  un:l   changes  in   cluster  means   are  small  or   non-­‐existent  
  18. 18. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1 Specify  number  of   clusters  (k)   Randomly  allocate   respondents  to  clusters   Allocate  respondents  to   most  similar  clusters   Repeat  un:l   changes  in   cluster  means   are  small  or   non-­‐existent   k-­‐Means  Cluster  Analysis   Compute  cluster  means   Specify  number  of   classes  (k)   Randomly  allocate   respondents  to  classes   Compute  class   parameters*   Compute  probability  of   each  respondent  being   in  each  class   Repeat  un:l   changes  in   class   parameters   are  small  or   non-­‐existent   Latent  Class  Analysis   Allocate  respondents   classes  with  highest   probabili:es   This  is  a  comparison  of  batch  k-­‐means  and  Latent  Class  Analysis  with  an  EM  Algorithm.     See  Celeux  and  Govaert  (1991),  “Clustering  criteria  for  discrete  data  and  latent  class   models”,  Journal  of  Classifica:on,  8(2)  for  a  more  mathema:cal  comparison.   *  The  class  parameters  are  computed  as  weighted  averages  of  the  segmenta:on   variables,  where  the  weights  are  the  probabili:es  of  each  respondent  being  in  each   segment.  
  19. 19. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1  0  5  10  15  20  25  30  35   25   20   15   10   5   0   Cluster   Analysis    0  5  10  15  20  25  30  35   25   20   15   10   5   0   Latent   Class   Analysis  
  20. 20. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1
  21. 21. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1 Cluster     Analysis   Latent   Class   Analysis  
  22. 22. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1 missing   values  
  23. 23. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1 How  many  clusters   (or  classes)  can  you   see  in  this  data?  
  24. 24. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1 Missing  values  and  latent  class   analysis   A   B   C   D   Cluster  1   1   2   3   4   Cluster  2   4   3   2   1   Cluster  3   1   2   2   1   Class  means  
  25. 25. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1 Missing  values  and  cluster  analysis   A   B   C   D   Cluster  1   1   2   3   3   Cluster  2   MISSING   MISSING   MISSING   MISSING   Cluster  3   3   3   2   1   Cluster  means  
  26. 26. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1 distribu*onal   assump*ons  
  27. 27. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1 Distribu*onal     assump*ons   •  Basic  idea:  instruct  a  latent  class  models     about  how  to  interpret  the  data   •  Categorical  assump:on:     look  only  at  matches   –  Example:  respondent  1  is  most  similar  to  2  and  3  (i.e.,  they  match  on   two  variables)   •  Numeric  assump:on:  assign  values  and  compute  differences     (e.g.,  Agree  =  3,  Neither  =  2,  Disagree  =  1)     –  Example:  respondent  1  is  most  similar  to  respondent  3   •  Ranking  assump:on:  look  at  rela:ve  order   –  Respondent  1  is  iden:cal  to  respondent  4   Variable   ID   A   B   C   1   Agree   Agree   Neither   2   Agree   Disagree   Neither   3   Agree   Neither   Neither   4  Neither   Neither   Disagree  
  28. 28. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1
  29. 29. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1 Example:  Categorical  data   Data   Shop Agree  (A)  or  disagree  (D)  that  “It  is  important  to   shop  around” Understand Agree  (A)  or  disagree  (D)  that  “I  understand  my   company's  communica:on  needs” Key   Agree  (A)  or  disagree  (D)  that  “Communica:ons   technology  is  key  to  our  business” Interested Agree  (A)  or  disagree  (D)  that  “I  am  interested  in   communica:ons  technology” Value Agree  (A)  or  disagree  (D)  that  “Value  for  money   is  more  important  to  us  than  gelng  the  best   technology” ID Shop Understand Key Interest Value 1 A A A A D 2 A A A D A 3 A A A A D 4 A A D A A 5 A D A D D 6 D A A A D 7 A D A D D 8 D D A A D 9 A A A A A 10 A A A A D 11 D A D D A 12 A A A A A 13 D D D D D … … … … … …
  30. 30. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1 Specify  number  of   classes  (k)   Randomly  allocate   respondents  to  classes   Compute  class   parameters   Compute  probability  of   each  respondent  being   in  each  class   Repeat  un:l   changes  in   class   parameters   are  small  or   non-­‐existent   Latent  Class  Analysis   Allocate  respondents   classes  with  highest   probabili:es  
  31. 31. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1 ID Shop Understand Key Interest Value … … … … … … 6 D A A A D … … … … … … Data   Parameters   Looking  at  the  parameters,  which   class  do  you  think  respondent  6   belongs  to?   Size Shop Under-­‐ stand Key Interest Value Class  1 67% Agree 40% 40% 48% 16% 53% Disagree 60% 60% 52% 84% 47% Class  2 33% Agree 65% 90% 88% 100% 26% Disagree 35% 10% 12% 0% 73%
  32. 32. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1 Compu*ng  the  probability  of  each   respondent  being  in  each  class   Size Shop Under-­‐ stand Key Interest Value Class  1 67% Agree 40% 40% 48% 16% 53% Disagree 60% 60% 52% 84% 47% Class  2 33% Agree 65% 90% 88% 100% 26% Disagree 35% 10% 12% 0% 73% ID Shop Understand Key Interest Value … … … … … … 6 D A A A D … … … … … … Data   Parameters   Ini:al  best  guess  of   probabili:es  is  given   by  the  class  sizes:   Class  1:  67%   Class  2:  33%   Prior   Probability  that  somebody  in  each  class   would  give  answers:   Class  1:  60%×40%×48%×16%×47%  =  1%   Class  2:  35%×90%×88%×100%×73%  =  20%   Class  condi:onal  densi:es                                                    67%×1%                                  67%×1%  +  3%×20%                                                        33%×20%                                  67%×1%  +  33%×20%     Posterior  probability     (Probability  of  being  in  a  class)   Class  1:                                                                    =  9%     Class  2:                                                                    =  91%        
  33. 33. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1 Applica*on   n  =  1,145  market  researchers  (GRIT2012/2013)     “How  important  do  you  think  each  of  the   following  atributes  is  to  clients  when  they  select   a  research  provider?”     5  POINT  SCALE     RANDOMLY  SHOW  15  OF  25  ATTRIBUTES  TO   EACH  RESPONDENT  
  34. 34. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1 Cluster   Analysis  
  35. 35. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1 Numeric   Assump*on   Lowest  price   Previous  experience  with  client/supplier   Rapid  response  to  requests   Listens  well  and  understands  client  needs   Flexibility  on  changing  project  parameters   Familiarity  with  client  needs   Completes  research  in  an  agreed-­‐upon  :me   Good  rela:onship  with  client/supplier   Breadth  of  experience  in  the  target  segment   Good  reputa:on  in  the  industry   Familiarity  with  the  industry  or  category   Length  of  experience/:me  in  business   Has  an  access  panel   Company  is  financially  stable   Has  knowledgeable  staff   High  quality  analysis   Provides  data  analysis  services   Understands  new  consumer  communica:on  channels  &  technologies   Also  does  qualita:ve  research   Consulta:on  on  best  prac:ces  and  methodology  effec:veness   Uses  sophis:cated  research  technology/strategies   Provides  highest  data  quality   Uses  the  latest  sta:s:cal/analy:cal  packages   Offers  unique  methodology  or  approach   Uses  the  latest  data  collec:on  technology   Segment  1   (45%)   %   Segment  2   (11%)   %   Segment  3   (45%)   %   Segment  1 Segment  2 Segment  3 Numeric  3  class Lowest  price Previous  experience  with  client/supplier Rapid  response  to  requests Listens  well  and  understands  client  needs Flexibility  on  changing  project  parameters Familiarity  with  client  needs Completes  research  in  an  agreed-­‐upon time Good  relationship  with  client/supplier Breadth  of  experience  in  the  target segment Good  reputation  in  the  industry Familiarity  with  the  industry  or  category Length  of  experience/time  in  business Has  an  access  panel Company  is  financially  stable Has  knowledgeable  staff High  quality  analysis Provides  data  analysis  services Understands  new  consumer communication  channels  &  technologiesAlso  does  qualitative  research Consultation  on  best  practices  and methodology  effectivenessUses  sophisticated  research  technology/ strategies Provides  highest  data  quality Uses  the  latest  statistical/analytical packagesOffers  unique  methodology  or  approach Uses  the  latest  data  collection  technology ortance  to  clients  (Research  providers  viewpoint):  Top  2  boxes  (out  of  5)  -­‐  reordered 50 88 95 98 83 99 97 95 90 93 92 86 36 71 97 96 86 84 66 96 75 96 57 75 79 68 73 65 67 47 50 67 75 31 40 33 16 33 13 58 30 28 27 16 18 27 15 10 31 17 55 87 89 97 71 95 91 94 81 82 85 51 2 36 96 91 59 45 22 71 37 69 7 39 14 Top  2  Box  (%) Percentages  are  Top  2   Box  Scores.    Where  values   are  significantly  higher   than  average  the  bars  are   shaded  orange.    Darker   shades  of  orange   correspond  to  smaller  p-­‐ values.  
  36. 36. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1 Categorical   Assump*on   Lowest  price   Previous  experience  with  client/supplier   Rapid  response  to  requests   Listens  well  and  understands  client  needs   Flexibility  on  changing  project  parameters   Familiarity  with  client  needs   Completes  research  in  an  agreed-­‐upon  :me   Good  rela:onship  with  client/supplier   Breadth  of  experience  in  the  target  segment   Good  reputa:on  in  the  industry   Familiarity  with  the  industry  or  category   Length  of  experience/:me  in  business   Has  an  access  panel   Company  is  financially  stable   Has  knowledgeable  staff   High  quality  analysis   Provides  data  analysis  services   Understands  new  consumer  communica:on  channels  &  technologies   Also  does  qualita:ve  research   Consulta:on  on  best  prac:ces  and  methodology  effec:veness   Uses  sophis:cated  research  technology/strategies   Provides  highest  data  quality   Uses  the  latest  sta:s:cal/analy:cal  packages   Offers  unique  methodology  or  approach   Uses  the  latest  data  collec:on  technology   Segment  1   (50%)   %   Segment  2   (50%)   %   Segment  1 Segment  2 All  categories Lowest  price Previous  experience  with  client/supplier Rapid  response  to  requests Listens  well  and  understands  client  needs Flexibility  on  changing  project  parameters Familiarity  with  client  needs Completes  research  in  an  agreed-­‐upon time Good  relationship  with  client/supplier Breadth  of  experience  in  the  target segment Good  reputation  in  the  industry Familiarity  with  the  industry  or  category Length  of  experience/time  in  business Has  an  access  panel Company  is  financially  stable Has  knowledgeable  staff High  quality  analysis Provides  data  analysis  services Understands  new  consumer communication  channels  &  technologiesAlso  does  qualitative  research Consultation  on  best  practices  and methodology  effectivenessUses  sophisticated  research  technology/ strategies Provides  highest  data  quality Uses  the  latest  statistical/analytical packagesOffers  unique  methodology  or  approach Uses  the  latest  data  collection  technology ortance  to  clients  (Research  providers  viewpoint):  Top  2  boxes  (out  of  5)  -­‐  reordered 41 90 96 98 87 100 96 98 89 95 92 81 24 67 98 97 82 77 55 93 69 89 48 65 60 66 81 83 90 61 86 87 86 70 71 73 43 18 32 85 74 52 43 27 58 36 60 13 43 27 Top  2  Box  (%) Percentages  are  Top  2   Box  Scores.    Where  values   are  significantly  higher   than  average  the  bars  are   shaded  orange.    Darker   shades  of  orange   correspond  to  smaller  p-­‐ values.  
  37. 37. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1 Ranking   Assump*on   Lowest  price   Previous  experience  with  client/supplier   Rapid  response  to  requests   Listens  well  and  understands  client  needs   Flexibility  on  changing  project  parameters   Familiarity  with  client  needs   Completes  research  in  an  agreed-­‐upon  :me   Good  rela:onship  with  client/supplier   Breadth  of  experience  in  the  target  segment   Good  reputa:on  in  the  industry   Familiarity  with  the  industry  or  category   Length  of  experience/:me  in  business   Has  an  access  panel   Company  is  financially  stable   Has  knowledgeable  staff   High  quality  analysis   Provides  data  analysis  services   Understands  new  consumer  communica:on  channels  &  technologies   Also  does  qualita:ve  research   Consulta:on  on  best  prac:ces  and  methodology  effec:veness   Uses  sophis:cated  research  technology/strategies   Provides  highest  data  quality   Uses  the  latest  sta:s:cal/analy:cal  packages   Offers  unique  methodology  or  approach   Uses  the  latest  data  collec:on  technology   Segment  1   (54%)   %   Segment  2   (46%)   %   Segment  1 Segment  2 Ranking Lowest  price Previous  experience  with  client/supplier Rapid  response  to  requests Listens  well  and  understands  client  needs Flexibility  on  changing  project  parameters Familiarity  with  client  needsCompletes  research  in  an  agreed-­‐upon time Good  relationship  with  client/supplier Breadth  of  experience  in  the  target segment Good  reputation  in  the  industry Familiarity  with  the  industry  or  category Length  of  experience/time  in  business Has  an  access  panel Company  is  financially  stable Has  knowledgeable  staff High  quality  analysis Provides  data  analysis  services Understands  new  consumer communication  channels  &  technologies Also  does  qualitative  research Consultation  on  best  practices  and methodology  effectivenessUses  sophisticated  research  technology/ strategies Provides  highest  data  quality Uses  the  latest  statistical/analytical packages Offers  unique  methodology  or  approach Uses  the  latest  data  collection  technology ortance  to  clients  (Research  providers  viewpoint):  Top  2  boxes  (out  of  5)  -­‐  reordered 86 97 98 98 80 95 94 95 80 83 82 61 19 47 90 81 62 54 31 67 40 63 17 35 26 23 74 81 91 68 89 88 90 78 82 83 63 22 53 95 90 75 71 51 86 69 88 49 75 65 Top  2  Box  (%) Percentages  are  Top  2   Box  Scores.    Where  values   are  significantly  higher   than  average  the  bars  are   shaded  orange.    Darker   shades  of  orange   correspond  to  smaller  p-­‐ values.  
  38. 38. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1 Latent  class  analysis  sobware   Product   Data/distribu*onal  assump*ons   Covariates*   Complex   Sampling*   Sawtooth  So_ware   Regression  (discrete  choice,  ranks),  Max-­‐Diff     No   No   Q   Numeric,  Binary,  Categorical,  Ranks,  Par:al  Ranks,   Ranks  with  Ties,  Max-­‐Diff,  Regression  (linear,   discrete  choice,  ranks,  par:al  ranks,  ranks  with  :es,   best-­‐worst),  Mixed  data   No   No   Limdep   Regression  (linear,  discrete  choice,  censored,  ranks,   par:al  ranks,  counts,  survival,  etc.)   Yes   No   SAS  (PROC  LCA/LTA/ Mixed)   Numeric,  Binary,  Categorical,  Growth,  Regression   (discrete  choice,  ranks,  par:al  ranks)   Yes   Yes   MPlus   Numeric,  Binary,  Categorical,  Ordered,  Categorical,   Counts,  Mixed  data   Yes   Yes   Latent  gold/Latent   Gold  Choice   Numeric,  Binary,  Categorical,  Growth,  Ranks,  Par:al   Ranks,  Counts,  Regression  (linear,  discrete  choice,   censored,  ranks,  par:al  ranks)   Yes   Yes   *  Covariates  and  the  ability  to  handle  complex  sampling  can  be  relevant  when  applying  latent  class  analysis  to  non-­‐ segmenta:on  problems  (e.g.,  crea:ng  predic:ve  models).  
  39. 39. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1 Cluster     Analysis   Latent  Class  Analysis  
  40. 40. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1 Thank you Tim Bock Q
  41. 41. Tim Bock, Q, Australia Festival of NewMR 2012 – Training Day – Session 1 Tim Bock, Q www.q-researchsoftware.com tim.bock@q-researchsoftware.com +61 425 241 989

×