Synthesizing  Frameworks,  Part  1:    The  Model  Human  Processor,  GOMS,  and  KLM  

Mark  W.  Newman    SI  688    Fall  2010  

Synthesizing  Frameworks  

  How  to  make  all  this  scien=fic  knowledge  about  human  percep=on,  cogni=on,  motor  performance  useful  for  design?  

  Go  beyond  guidelines,  heuris=cs,  best  prac=ces  

  Two  examples:    Card,  Moran,  and  Newell’s  “applied  psychology”  

  Models  and  methods  and  to  predict  human-­‐system  interac=on    performance  

  Green,  Petre,  and  Blackwell’s  “cogni=ve  dimensions”    A  “sensi=zing  framework”  to  guide  discussion  about  design  tradeoffs  

Synthesizing  Frameworks  

  How  to  make  all  this  scien=fic  knowledge  about  human  percep=on,  cogni=on,  motor  performance  useful  for  design?  

  Go  beyond  guidelines,  heuris=cs,  best  prac=ces  

  Two  examples:    Card,  Moran,  and  Newell’s  “applied  psychology”  

  Models  and  methods  and  to  predict  human-­‐system  interac=on    performance  

  Green,  Petre,  and  Blackwell’s  “cogni=ve  dimensions”    A  “sensi=zing  framework”  to  guide  discussion  about  design  tradeoffs  

Today  

Next  Week  

A  Watershed  Moment  

  Card,  Moran,  Newell  1983    Synthesized  large  volumes  

of  findings  from  Cog.  Psych.  to  date  

  Ini=ated  call  for  an  “applied  psychology”  

  Ar=culate  a  compelling  research  agenda  for  HCI  that  helped  organize  the  field  

Underpinnings:  Information  Processing  

Prior  to  CMN  

  Informa=on  Processing  was  used      To  describe  human  cogni=on  

  To  debunk  behaviorism  

  As  a  framework  for  par==oning  and  organizing  research  

  Informa=on  Processing  was  not  used    To  inform  systems  design  

Goals  of  CMN    “A  scien=fic  psychology  should  not  only  help  us  to  understand  our  

own  human  nature,  it  should  help  us  in  our  prac=cal  affairs.”  

  “A  scien=fic  psychology  should  help  us  in  arranging  this  interface  so  it  is  easy,  efficient,  error-­‐free——even  enjoyable.”  

  “knowledge  of  human  cogni=ve  behavior  is  sufficiently  advanced  to  enable  its  applica=ons  in  computer  science  and  other  prac=cal  domains.”  

  “Our  own  goal  is  to  help  create  …  an  applied  informa=on-­‐processing  psychology.”  

  “Applying  psychology  to  the  evalua=on  of  systems  is  assuredly  easier  than  applying  it  to  the  design  of  systems.”  

From  Ch.  1,  Psych  of  HCI  

CMN’s  Core  Concepts  

  Model  Human  Processor:  a  simplified  model  of  human  informa=on  processing  that  can  be  used  by  designers  

  GOMS:  a  method  for  represen=ng  people’s  task-­‐based  interac=on  with  systems  that  supports  detailed  analysis  

  The  Keystroke-­‐Level  Model:  a  quan=ta=ve  technique  for  predic=ng  performance  on  specific  interac=ve  tasks  

Model  Human  Processor  

Model  Human  Processor  

  Four  types  of  memory    Differ  in  terms  of  capacity,  

decay  =me,  encoding  

  Three  processors    Differ  in  terms  of  

processing  =me  (and  func=on)  

  Operate  serially,  though  can  be  pipelined  

MHP:  Principles  of  Operation    Recognize-­‐Act  Cycle  

  Interac=on  b/t  WM  &  LTM  

  Variable  Perceptual  Rate    Speed  varies  with  input  

intensity  

  Encoding  Specificity    Informa=on  is  encoded  

before  stored  

  Discrimina=on    Retrieval  is  determined  by  

available  candidates  +  cues  

  Variable  Cogni=ve  Rate    Cycle  faster  when  demand  is  

higher  

MHP:  Principles  of  Operation  

  Fijs’  Law  

  Power  Law  of  Prac=ce    Performance  increases  

with  prac=ce  (with  log(t))  

  Uncertainty  Principle  (Hick-­‐Hyman  Law)  

  Ra=onality  Principle    Goals  +  Task  +  Operators  +  

Inputs  +  Knowledge  +  Process-­‐limits  -­‐>  Behavior  

  Problem  Space  Principle    Ra=onality  described  via  

  States  of  knowledge    Operators  to  change  state    Constraints  on  operators    Rules  for  selec=ng  

operators  

Is  This  How  Humans  Work?  

  Only  if  you  squint  

  An  admijed  simplifica=on      “All  models  are  wrong,  some  models  are  useful.”  –  G.  E.  P.  Box  

  Goals  of  MHP:      Ability  to  calculate  parameters  on  human  performance  from  the  

analysis  of  the  task  

  Giving  the  [designer]  knowledge  in  a  form  rela=vely  easy  to  assimilate  

…Are  we  there  yet?  

Decision  Making  

Decision  Making  

Decision  Making  

Decision  Making  

Perceptual  Processor  samples  visual  field  

Decision  Making  

Perceptual  Processor  stores  image  in  visual  image  store  

Decision  Making  

Perceptual  Processor  encodes  input  and  stores  in  STM  

Decision  Making  

Cogni=ve  Processor  samples  working  memory  

Decision  Making  

Cogni=ve  Processor  associates  symbol  with  task  stored  in  LTM,  task  info  moved  to  STM.  

Decision  Making  

Cogni=ve  Processor  directs  Motor  Processor  to  move  eyes,  scan  bujons  

Decision  Making  

Perceptual  Processor  receives  input:  bujon  has  been  seen.  

Decision  Making  

Visual  Image  Store  -­‐>  Encoding  -­‐>  Working  Memory  

Decision  Making  

Cogni=ve  Processor  “reads”  working  memory,  compares  bujon  to  s=mulus  

Decision  Making  

Cogni=ve  Processor  directs  Motor  Processor  to  move  finger  and  push  

What  Good  is  the  MHP?  

  If  you  can  es=mate  the  =me  for  each  stage,  you  can  predict  performance  

  S=ll  very  difficult  to  use    Tedious  for  even  a  small  example  

  Great  deal  of  knowledge  required  

  Lots  of  room  for  (mis-­‐)  interpreta=on  

Making  the  MHP  (More)  Useful:  GOMS  

  Goals  of  MHP:      Ability  to  calculate  parameters  on  human  performance  from  the  

analysis  of  the  task  

  Giving  the  [designer]  knowledge  in  a  form  rela=vely  easy  to  assimilate  

  GOMS:  a  method  of  task  representa=on  that  affords  analysis  via  the  MHP  (sort  of)  

GOMS  

  Goals    Goal  that  the  user  is  trying  to  accomplish—these  can  be  

decomposed  into  subgoals  

  Operators    Individual,  low-­‐level  ac=ons  performed  using  the  system  

  Methods    Sequences  of  Operators  used  to  accomplish  Goals  

  Selec=on  Rules    Criteria  to  decide  when  to  use  different  Methods  

From  B.  E.  John.  Informa=on  Processing  and  Skilled  Behavior.    In  J.  M.  Carroll,  ed.  HCI  Models,  Theories,  and  Frameworks.  2003.  

Example GOMS model for “MOVE-TEXT” in MS Word

Example  Operators  in  GOMS  

  Operators    Physical    

  Move  hand  to  mouse    Move  mouse  such  that  pointer  is  on  menu  =tle    Push  mouse  bujon  down    Move  mouse  to  exact  item    Release  mouse  bujon  

  Mental    Retrieve  the  sub-­‐goal  of  cuqng  text    Recall  that  “Edit”  menu  is  where  “Cut”    

command  is  found    Verify  feedback  

GOMS  vs.  MHP  

?

The  Keystroke-­‐Level  Model  

  A  method  for  predic=ng  performance  given  a  GOMS  model  

  A  set  of  guidelines  for  choosing  opera=ons  and  defining  methods  

  A  simplified  set  of  calcula=ons  based  on  the  MHP  

  Limited  to  predic=ng  the  (expert)  execu/on  /me  of  unit  tasks    No  method  selec=on  

  No  learning,  exploring,  errors  

  No  decision  making  about  goal  decomposi=on  

KLM’s  6  7  Operators  

  K  :  keystroking    

  P  :  poin=ng  (moving  the  mouse  to  a  target)  

  B:  pressing  or  releasing  mouse  bujon*  

  H  :  homing  (moving  hand  from  keyboard  to  mouse)  

  D  :  drawing  (drawing  lines  in  segments)  

  M  :  mental  work  (retrieving  from  memory,  verifying  feedback)  

  R:  system  response    

*:  Added  by  Kieras,  D.  (1993,  2001),  Using  the  Keystroke-­‐Level  Model  to  Es=mate  Execu=on  Times  

Timing  

K B P H D

M R

Keystroking Press button on mouse Point Home (Move hand to/from keyboard/mouse) Draw lines (domain dependent)

Mentally prepare (access LTM) System Response

0.28 (.12-1.2) 0.1* 1.1 0.4 -

1.2* -

Operator Description Time (s)

*  Times  from  Kieras,  D.  (1993,  2001),  Using  the  Keystroke-­‐Level  Model  to  Es=mate  Execu=on  Times  

An  Example  

An  Example  (Physical  Only)  

1.  point  to  file  icon  (P)  2.  press  and  hold  mouse  

bujon  (B)    3.  drag  file  icon  to  trash  can  

icon  (P)    4.  release  mouse  bujon  (B)    5.  point  to  original  window  (P)    

 Total  =me    =  3P  +  2B          =  3*1.1  +  2*.1          =  3.5  sec  

1.  point  to  file  icon  (P)    2.  click  mouse  bujon  (BB)    3.  point  to  file  menu  (P)    4.  press  and  hold  mouse  bujon  

(B)    5.  point  to  DELETE  item  (P)  6.  release  mouse  bujon  (B)    7.  point  to  original  window  (P)  

 Total  =me    =  4P  +  4B          =  4*1.1  +  4*.1          =  4.8  sec  

Power  Key  Method  (Physical  Only)  

1.  point  to  file  icon  (P)    2.  click  mouse  bujon  (BB)    3.  move  hand  to  keyboard  (H)    4.  hit  command  key  command-­‐T  (KK)    5.  move  hand  back  to  mouse  (H)  

 Total  =me    =  P  +  2B  +  2H  +  2K          =  1.1  +  .2  +  .8  +  .56          =  2.66  sec  

Drag  to  Trash:  Going  Mental    ini=ate  the  dele=on  (decide  to  do  the  task)  (M)      find  the  file  icon  (M)      point  to  file  icon  (P)      press  and  hold  mouse  bujon  (B)    verify  that  the  icon  is  reverse-­‐video  (M)      find  the  trash  can  icon  (M)      drag  file  icon  to  trash  can  icon  (P)      verify  that  the  trash  can  icon  is  reverse-­‐video  (M)      release  mouse  bujon  (B)    verify  that  the  trash  can  icon  is  bulging  (M)       find  the  original  window  (M)      point  to  original  window  (P)  

 Total  =me    =  3P  +  2B  +  7M          =  11.9  sec  

Where  to  Place  the  M’s  

  Ini=a=ng  a  task  

  Making  a  strategy  decision  

  Retrieving  a  chunk  from  memory  

  Finding  something  on  the  screen  

  Thinking  of  a  task  parameter  

  Verifying  that  a  specifica=on  or  ac=on  is  correct  

*  Times  from  Kieras,  D.  (1993,  2001),  Using  the  Keystroke-­‐Level  Model  to  Es=mate  Execu=on  Times  

Experts  and  M…  

  ini=ate  the  dele=on  (decide  to  do  the  task)  (M)    

  find  the  file  icon  (M)      point  to  file  icon  (P)      press  and  hold  mouse  bujon  (B)    verify  that  the  icon  is  reverse-­‐video  (M)      find  the  trash  can  icon  (M)      drag  file  icon  to  trash  can  icon  (P)      verify  that  the  trash  can  icon  is  reverse-­‐

video  (M)      release  mouse  bujon  (B)    verify  that  the  trash  can  icon  is  bulging  

(M)      find  the  original  window  (M)      point  to  original  window  (P)  

 Total  =me    =  3P  +  2B  +  2M          =  5.9  sec  

  ini=ate  the  dele=on  (decide  to  do  the  task)  (M)    

  find  the  file  icon  (M)      point  to  file  icon  (P)      press  and  hold  mouse  bujon  (B)    verify  that  the  icon  is  reverse-­‐video  (M)      find  the  trash  can  icon  (M)      drag  file  icon  to  trash  can  icon  (P)      verify  that  the  trash  can  icon  is  reverse-­‐

video  (M)      release  mouse  bujon  (B)    verify  that  the  trash  can  icon  is  bulging  

(M)       find  the  original  window  (M)      point  to  original  window  (P)  

 Total  =me    =  3P  +  2B  +  7M          =  11.9  sec  

  If  there  are  20  (one  character)  changes  to  make  in  words  in  a  paragraph  of  8lines  long,  should  I  find  and  edit  each  one  or  retype  the  whole  paragraph?  (assume  80  char/line)  

Exercises  

Key press= 0.28 Button=0.1

Pointer move=1.1 Hand move=0.4

Mental effort=1.2 System Response=???

Exercise  1  

  If  there  are  20  changes  to  make  in  words  in  a  paragraph  of  15  lines  long,  should  I  find  and  edit  each  one  or  retype  the  whole  paragraph?    

Key  press=  0.28

Button=0.1

Pointer move=1.1

Hand move=0.4

Mental effort=1.2

Approach #1: Delete whole paragraph & retype Step 1: Delete Paragraph M Recall that text must be selected & deleted 1.2 P Move cursor to beginning of paragraph 1.1 BB Click 0.1

0.1 P Move cursor to end of paragraph 1.1 KBB Shift-Click 0.28

0.1 0.1

K Type DEL 0.28 H Move Hand to Keyboard 0.4 MPBBPKBBKH 4.76

Step 2: Retype paragraph K Type (8 * 80=640) characters 0.28 640K 179.2

Approach #1 Total: 183.96

Approach #2: Find and fix 20 characters Subgoal: Find and fix one character M Locate character to delete 1.2 H Move Hand to mouse 0.4 P Move cursor to after character 1.1 BB Click 0.1

0.1 M Recall that delete must be typed 1.2 H Move Hand to keyboard 0.4 K Type DEL 0.2 M Recall that character must be typed 1.2 K Type Char 0.28 MHPBBMHKMK 6.1

Approach #2 Total (20 * MHPBBMHKMK) 123.6

  How  long  does  the  system  response  =me  have  to  be  for  the  Google  Toolbar  Bookmarks  pull-­‐down  menu  to  appear  before  it  is  shorter  to  just  type  in  a  known  url  (hjp://ctools.umich.edu)?      Note:  assume  no  autocomplete  

Exercises  

Key press= 0.28 Button=0.1

Pointer move=1.1 Hand move=0.4

Mental effort=1.2 System Response=???

  How  long  does  the  system  response  =me  have  to  be  for  the  Google  Toolbar  Bookmarks  pull-­‐down  menu  to  appear  before  it  is  shorter  to  just  type  in  a  known  url?      e.g.  hjp://ctools.umich.edu  

Key  press=  0.28

Button=0.1

Pointer move=1.1

Hand move=0.4

Mental effort=1.2

System Response=???

Exercise  2  

Approach #1: Use Google Bookmarks Menu M Recall CTools is in bookmarks menu 1.2 P Move pointer to bookmarks menu 1.1 B Press 0.1 R Wait for menu items to display R P Move pointer to CTools item 1.1 B Release 0.1 MPBRMPB 3.6+R

Approach #2: Type in URL M Recall need to use address bar 1.2 P Move pointer to address bar 1.1 BBBB Double-click text 0.1

0.1 0.1 0.1

M Recall need to type "http://ctools.umich.edu" 1.2 H Move hand to keyboard 0.4 24K Type "http://ctools.umich.edu" + RTN 6.72 MPBBBBMHK(24) 11.02

R = (11.02-3.6) 7.42

Accuracy  of  GOMS/KLM  

  KLM:  accurate  ±  33%  

  CMN-­‐GOMS:  accurate  90%  of  the  =me  

  CPM-­‐GOMS:  accurate  ±  12%*  

*  Gray,  W.  D.,  John,  B.  E.,  and  Atwood,  M.  E.  1993.  Project  Ernes=ne:  valida=ng  a  GOMS  analysis  for  predic=ng  and  explaining  real-­‐world  task  performance.  Hum.-­‐Comput.  Interact.  8,  3  (Sep.  1993),  237-­‐309.    

Helping  Designers  Create  Models  

hjp://cogtool.hcii.cs.cmu.edu/  

Helping  Designers  Create  Models  

hjp://cogtool.hcii.cs.cmu.edu/  

Review  (Whither  GOMS?)    CMN’s  “applied  psychology”  approach  was  enormously  influen=al  in  

early  growth  of  HCI    Created  the  possibility  of  an  academic  discipline    MHP-­‐GOMS-­‐KLM  had  impacts  within  research  and  prac=ce  

  Current  impact  on  prac=ce:  minimal  (mostly)    Much  Interac=on  Design  today  op=mizes  for    

learnability,  desirability,  usefulness,  etc.    Excep=ons:  domains  where  performance  is  king    

(e.g.,  fighter  cockpit  design)    

  Next  week:  a  completely  different  approach  to  synthesizing  cogni=on    Covers  more  aspects  of  usability  than  expert  performance    Less  predic=ve  and  direc=ve