cognitive pre-testing

35
Joint Seminar of the Gender Task Force and the Tool Pool: Cogni:ve Pretes:ng of Crossna:onally Comparable Survey Instruments in a Developing Country Context with applica*on to USAID Feed the Future’s Women’s Empowerment in Agriculture Index Presented at the Interna*onal Food Policy Research Ins*tute Washington, DC 9 May 2014 Kiersten B. Johnson, PhD Senior Study Director, Westat

Upload: ifpri-gender

Post on 02-Dec-2014

375 views

Category:

Government & Nonprofit


0 download

DESCRIPTION

PR

TRANSCRIPT

Joint  Seminar  of  the  Gender  Task  Force  and  the  Tool  Pool:    

Cogni:ve  Pretes:ng  of  Cross-­‐na:onally  Comparable  Survey  Instruments  in  a  Developing  Country  Context with  applica*on  to  USAID  Feed  the  Future’s  Women’s  Empowerment  in  Agriculture  Index    Presented  at  the  Interna*onal  Food  Policy  Research  Ins*tute  Washington,  DC  9  May  2014    Kiersten  B.  Johnson,  PhD  Senior  Study  Director,  Westat  

• Understand  the  purpose  of  cogni*ve  interviewing/pretes*ng  • Become  familiar  with  the  cogni*ve  processes  that  are  assessed  

• Recognize  when  it  is  necessary  to  implement  

•  Learn  what  methodological  decisions  can  be  made  to  tailor  the  pretes*ng  to  the  needs  and  characteris*cs  of  the  survey  

• Obtain  a  sense  for  how  to  train  and  implement  

Objec:ves

• Overview  of  the  methodology  

• Walk-­‐through  of  the  steps  to  implement  

• Demonstra*on  of  how  the  method  was  applied  for  the  Women’s  Empowerment  in  Agriculture  ques*onnaire  in  Haï*  

Approach

Overview  of  the    Cogni:ve  Interview  

Method

We  want  to  systema*cally  iden*fy  and  analyze  sources    of  response  error  in  surveys,  and  use  that  informa*on  to  try  to  improve  the  quality  and  accuracy  of  our  survey  instruments.  

Why  cogni:ve  pretes:ng?

Especially  important  with  new  or  revised  instruments/  ques*ons  

Cri*cal  when  being  used  cross-­‐

culturally  and/or  cross-­‐linguis*cally  

Why  Cogni:ve  Interviewing?

Four  Stages  of  Cogni:on  Required  for  an  Accurate  Response  to  a  Survey  Ques:on:

Cogni&ve  Stages   Cogni&ve  Stage  Defini&on   Problems   Causes  

1.  Comprehension   Respondent  interprets  the  ques*on  

Respondent  does  not  understand  

Unknown  terms,  ambiguous  concepts,  long  and  overly  complex  

2.  Retrieval   Respondent  searches  memory  for  relevant  informa*on  

Respondent  does  not  remember/does  not  know  

Recall  difficulty,  ques*ons  assume  respondent  has  informa*on  

3.  Judgment   Respondent  evaluates  and/or  es*mates  response  

Respondent  does  not  want  to  tell,  can’t  tell  

Biased  or  sensi*ve,  es*ma*on  difficulty  

4.  Response   Respondent  provides  informa*on  in  the  format  requested  

Respondent  can’t  respond  in  the  format  requested  

Incomplete  response  op*ons,  mul*ple  responses  necessary  

What  is  cogni:ve  pretes:ng?

• Qualita*ve  field  research  method  embedded  in  survey  interview    

• Can  be  comprised  of:  •  Direc*ons  to  “think  aloud”  while  formula*ng  a  response  to  a  survey  ques*on  •  Addi*onal  probes/ques*ons  integrated  into  dra_  instrument  •  Interviewer  observa*ons  of  respondent  verbal  &  nonverbal  cues  

•  Intended  to:  •  Help  determine  whether  the  ques*on  is  genera*ng  the  intended  informa*on  •  Iden*fy  problems  related  to  ques*onnaire  design  

Can  respondents  discuss  the  survey  ques3ons  with  a  reasonable  degree  of  validity?  

What  is  Cogni:ve  Interviewing? how  

people  construct  their  

answers  

how  people  interpret  

the  ques*ons  

difficul*es  people  had  

in  answering  

Different  Approaches

•  Lab-­‐based  vs  field-­‐based  •  Thinking  aloud  vs.  probing  (vs.  a  combina*on  of  both)  •  Concurrent  vs.  retrospec*ve  •  Standardized  vs.  “ac*ve”  •  Scripted,  semi-­‐scripted,  or  en*rely  improvised  •  Number  of  respondents  and  itera*ons  •  Rela*vely  unskilled  data  collector  or  expert  inves*gator  •  Analysis:  systema*c  review  of  interview  transcripts,  or  based  on  notes  

Selec*on  of  methodological  approaches  is  likely  to  vary  according  to:  -­‐  Type  and  amount  of  resources  available  (human,  temporal,  financial)  -­‐  Nature  of  the  survey  

Different  Methodological  Approaches

•  Iden*fica*on  of  “problems”  with  ques*ons  that  turn  out  not  to  be  “real”  during  survey  implementa*on  •  difficult  to  measure;  limited  research  suggests  it’s  not  significant  problem  

 

•  Failure  to  iden*fy  problems  that  actually  exist  in  survey  instrument  design  and/or  administra*on  •  almost  certain  to  occur,  but  possible  to  reduce  

• Cogni*ve  interview  findings  may  be  inconsistent  when  interviews  are  implemented  by  independent  researchers  •  not  necessarily  problema*c  

Validity  and  Reliability  of  Cogni:ve  Pretes:ng:  Poten:al  Types  of  Error

Steps  in  Implemen:ng  a  Cogni:ve  Pretest

Example:  Steps  in  Implementa:on

• Develop  the  ques*onnaire  •  Translate  the  ques*onnaire  •  Iden*fy  candidate  ques*ons  for  assessment  •  Select  methods;  develop  appropriate  probes  • Determine  number  and  selec*on  of  respondents  •  Select  and  train  interviewers  • Prac*ce  interviews,  note-­‐taking  and  audio  recording  

•  Implement  the  cogni*ve  interviews  •  Summarize  field  notes;  possibly  use  so_ware  to  facilitate  data  analysis  • Write  report:  background,  data  and  methods,  results,  conclusions  &  recommenda*ons  • Revise  ques*ons  • Reiterate  

Example:  Steps  in  Implementa:on METHOD-­‐OLOGICAL  CHOICES  

EMBEDDED  HERE!  

Cogni&ve  Stages  

Cogni&ve  Stage  Defini&on   Problems   Causes  

Compre-­‐hension  

Respondent  interprets  the  ques*on  

Respondent  does  not  understand  

Unknown  terms,  ambiguous  concepts,  long  &  overly  complex  

Retrieval   Respondent  searches  memory  for  relevant  informa*on  

Respondent  does  not  remember/  does  not  know  

Recall  difficulty,  ques*ons  assume  respondent  has  informa*on  

Judgment   Respondent  evaluates  and/or  es*mates  response  

Respondent  does  not  want  to  tell,  can’t  tell  

Biased  or  sensi*ve,  es*ma*on  difficulty  

Response   Respondent  provides  informa*on  in  format  requested  

Respondent  can’t  respond  in  the  format  requested  

Incomplete  response  op*ons,  mul*ple  responses  necessary  

Probes  

E.g.,  “Earlier  I  asked  you  who  contributes  most  to  decisions  regarding  a  new  purchase  of  different  items.  How  would  you  put  that  ques*on  in  your  own  words?  Is  this  ques*on  easy  or  difficult?  Why  is  it  difficult?”  

E.g.,  “Many  people  find  it  difficult  to  recall  every  ac*vity  done  in  a  day.  How  well  do  you  remember  specific  ac*vi*es  you  were  doing  in  the  past  24  hours?”  “How  do  you  know  what  *me  you  woke  up  yesterday?”  

E.g.,  “Do  you  think  that  any  other  women  you  know  may  be  afraid  to  answer  this  ques*on?  Why  do  you  think  they  might  be  afraid?”  

E.g.,  “Earlier  I  asked  you  how  sa*sfied  are  you  with  your  available  *me  for  leisure  ac*vi*es  like  visi*ng  neighbors,  watching  TV,  listening  to  the  radio,  seeing  movies  or  doing  sports.  Was  this  ques*on  easy  or  difficult?  Why  was  it  difficult?”  

Interviewer  Training  Content

• Understanding  what  cogni*ve  pretes*ng  is,  why  it  is  necessary  for  survey  ques*onnaire  development  

• Knowledge  of  the  cogni*ve  processes  required  of  survey  respondents  •  Full  understanding  of  the  inten*on  of  the  ques*ons  being  tested  • Knowledge  of  the  probes  to  be  used  in  cogni*ve  tes*ng  interviews  •  Facility  with  in-­‐depth  qualita*ve  interview  techniques  and  the  no*on  of  “narra*ves  as  data”  

• Prac*cal  exercises  in  doing  cogni*ve  interviews,  to  include  cri*cal  group  discussion  

Interviewer  Training  Content

Interviewer  Training  Content

Things  to  be  mindful  of  when  using  regular,  seasoned  survey  interviewers  to  implement  cogni*ve  pretes*ng:  

•  Revised  priori*es:  iden*fica*on  of  problems  of  understanding,  not  resolu*on  of  problems  of  understanding  

•  Need  to  administer  slowly  and  allow  *me  for  thoughkul  responses,  rather  than  proceed  expedi*ously  

Interviewer  Training  Content  (cont.)

Materials  for  Training •  Slide  show  on  cogni*ve  tes*ng  and  why  it  is  necessary  • Ques*onnaire  to  be  tested  • A  detailed  ques*on-­‐by-­‐ques*on  interviewer’s  guide,  customized  for  the  cogni*ve  interview  • A  set  of  generic  possible  probes  for  use  in  the  interviews;  e.g.:  

•  “Why  did  you  answer  in  that  way?”  •  “Tell  me  a  lille  bit  more”    •  “Can  you  think  of  an  example  of  what  you  are  talking  about?”  

•  Signs  of  respondent  cogni*ve  difficul*es  •  Long  silences  •  Contradic*ons  •  Reluctance  or  other  discomfort  

Materials  for  Training

Number  of  Interviews  &  Respondent  Selec:on • Number  of  interviews:  

•  Implement  for  each  country  >>  Within  each  country,  implement  for  each  language  comprising  10%  or  more  of  the  survey  sample  popula*on  

>>  For  each  language,  implement  15  well-­‐conducted  interviews*  (see  next  slide)  

• Respondent  selec*on:  •  Select  according  to  demographic  profile  of  eligible  respondents  to  the  survey  •  Lab  vs  field  implementa*on  

“The  palern  of  responses  across  a  limited  sample  of  respondents  provides  insight  into  the  common  palerns  of  understanding  and  interpreta*ons  of  people  who  are  typical  of  

the  targeted  survey  respondents.”  -­‐-­‐  UNESCAP  2010  

Number  of  Interviews  &  Respondent  Selec:on

“For  each  language,  implement  15  well-­‐conducted  interviews*” • No  empirically-­‐grounded  conclusions  on  op*mal  sample  sizes;  typical  recommenda*ons  range  from  10-­‐15  interviews  •  Important  to  have  enough  respondents  so  all  relevant  aspects  of  a  ques*onnaire  are  tested  

“The  point  is  not  to  obtain  sample  sizes  large  enough  to  supply  precision  in  sta*s*cal  es*mates.  Rather,  we  strive  to  interview  a  variety  of  

individuals  who  will  be  useful  in  informing  our  decisions.      

Sta*s*cians  strive  to  minimize  (error)  variance,  whereas  cogni*ve  interviewers  maximize  (subject)  variance.”  

-­‐-­‐  Willis  2005  

Interview  team

• One  interviewer  to  ask  the  respondent  ques*ons  • One  interviewer  to  operate  the  digital  recorder,  and  observe  &  take  notes  of  the  respondent’s  verbal  &  non-­‐verbal  cues  

Interview  Team

Materials

•  Tailored  consent  form  • Customized  ques*onnaire  (hard  copy)  • Digital  recorder  • Pens/pencils/clipboard  

Interview  Materials

Preparing  the  Respondent

•  Informed  consent  to  record  •  Explain  that  the  interview  is  evalua*ve,  and  we  will  be  asking  for  their  input  on  different  aspects  of  the  ques*ons  and  responses  

Preparing  the  Respondent

Data  Analysis

• Analysis  method  will  depend  on  type  of  data  •  Fundamentally  qualita*ve;  coding  of  textual  responses  will  likely  be  needed  to  be  able  to  compare  responses  across  respondents  •  Triangulate:    

•  Responses  to  closed-­‐ended  CI  ques*ons  •  Narra*ve  responses  to  probes  captured  in  text  •  Interviewer  observa*ons  recorded  in  text  •  Digital  recording  of  interview  

• Applica*on  of  logic  to  interpret  findings  &  make  revisions  accordingly  

Data  Analysis

Applica:on  of  Cogni:ve  Interview  

Method:    

Women’s  Empowerment  in  

Agriculture  Ques:onnaire  in  

Haï:

Cogni:ve  Pretest  of  the  Women’s  Empowerment  in  Agriculture  Instrument  in  Haï: •  Step  1:  Decide  if  we  need  to  implement  cogni*ve  tes*ng.  

•  New  topic  being  addressed  •  New  instrument  developed  for  assessing  women’s  empowerment  in  ag  •  Cross-­‐na*onal  comparability  of  ques*ons  and  responses  is  important  

•  Step  2:  Iden*fy  selected  ques*ons  that  respondents  may  find  cogni*vely  difficult.  

Cogni:ve  Pretest  of  the  Women’s  Empowerment  in  Agriculture  Instrument  in  Haï:

Module  H2  •  H2.01.  Recall  difficulty    •  H2.01.  Recall  period  (last  12  months)  •  H2.01.  Difficult  terms:  “in-­‐kind  or  monetary  work  both  agriculture  and  other  wage  work”  

•  H2.02.  &  H2.03.  Comprehension:  “Input  into  decisions”  

•  H2.02.  &  H2.03.  Sensi*vity:  Input  into  decisions  

Iden:fy  Key  Ques:ons  (1)

Module  H3a  –  Access  to  Produc&ve  Capital  

•  H3.01a:  Household  items  owned:  concept  “household”,  difficult  items  (nonfarm  business  equipment,  Farm  equipment  (non-­‐mechanized),  Farm  equipment  (mechanized)  

•  H3.01b:  Number  of  items  owned:  number  recall  difficulty  

•  H3.02:  Owner  of  items:  judgment  difficulty  

•  H3.03:  Decide  to  sell:  judgment  difficulty  

•  H3.04:  Decide  to  give  away:  judgment  difficulty  

•  H3.05:  Decide  to  mortgage  or  rent  out:  judgment  difficulty,  difficult  terms  (mortgage,  rent  out)  

•  H3.06:  Decide  to  purchase:    comprehension  (“contributes  most  to  decisions  “),  judgment  difficulty.    

Module  H3b  –  Access  to  Produc&ve  Capital    •  H3.07:  Anyone  taken  loans  in  cash/kind:  Comprehension  (“Borrow  in-­‐kind”),  recall  (last  12  months)  

•  H3.08:  Decision  to  borrow:  Judgment    

•  H3.09:  Decision  about  what  to  do  with  loans:  Judgment  

•  H3.10a:  Demand  for  extra  credit:  Comprehension  (condi*onal  sentence)    

•  H3.10b:  Reasons  for  lack  of  demand  for  extra  credit:  Judgment  

•  H3.11a:  Limita*ons  to  credit  supply:  Judgment  

•  H3.11b:  Reasons  for  limited  credit  supply:  Judgment,  sensi*vity  

Module  H4  •  H4.01:  Infrastructure:  Sensi*vity  •  H4.02:  Proper  payment  of  wages  for  public  works:  Sensi*vity,  comprehension    

•  H4.03:  Protest  the  misbehavior  of  authori*es  or  elected  officials:  Sensi*vity  

•  H4.04:  Group  membership:  Comprehension  of  the  different  groups  involved  

•  H4.05:  Ac*ve  membership:  comprehension  

Module  H5a  

•  H5.01:  Decision-­‐maker:  Possible  sensi*vity  if  privacy  not  maintained  

•  H5.02:  Personal  decisions:  Possible  sensi*vity  if  privacy  not  maintained  

Iden:fy  Key  Ques:ons  (2)

Module  H5b  •  H5.03:  Get  in  trouble:  comprehension;  Act  differently:  comprehension  (item  wording,  “Gevng  inputs  for  agricultural  produc*on”),  sensi*vity  •  H5.04:  Others  think  poorly:  comprehension  (item  wording,  “think  poorly”),  sensi*vity  •  H5.05:  Right  thing  to  do:  comprehension  (item  wording,  “right  thing  to  do”)  

Iden:fy  Key  Ques:ons  (3)

Module  H6  •  24  hour  recall  problems  (remembering  every  ac*vity,  *me  of  the  day)  •  Variability  of  daily  schedules  •  Comprehension:  “sa*sfied”  

•  PLUS…  dura*on  of  each  module,  debrief  interview  ques*ons  

•  Lab-­‐based  vs  field-­‐based  à  field-­‐based  •  Thinking  aloud  vs.  probing  (vs.  a  combina*on  of  both)  à  probing  • Rela*vely  unskilled  data  collector  or  expert  inves*gator    à  specially  trained  data  collector  • Concurrent  vs.  retrospec*ve  à  middle  road:  between  modules  •  Standardized  vs.  “ac*ve”  à  standardized  •  Scripted,  semi-­‐scripted,  or  en*rely  improvised  à  scripted  • Number  of  respondents  à  12  HHs;  12  female  respondents  &  8  male  respondents  • Number  of  itera*ons  à  just  one  was  possible  

Selec:on  of  Methods

Table  1.  Household  and  individual  samples  according  to  age  of  the  female  respondent

Age  of  female  respondent   Households  sampled  

Dual-­‐headed  household  

Single  female-­‐headed  household  

18-­‐35   4   2  36  and  older   4   2  

Individuals  Sampled  

Dual-­‐headed  household  

Single  female-­‐headed  household  

18-­‐35   8   2  36  and  older   8   2  

H2.01.  Did  you  par:cipate  in  [ACTIVITY]  in  the  past  12  months  (that  is  during  the  last  [one/two]  cropping  seasons)?

•  One  fi_h  of  the  20  respondents  reported  that  this  ques*on  was  difficult.    •  In  some  cases,  there  were  comprehension  difficul*es;  respondents  generated  the  following  feedback  on  this  ques*on:  

 “I  haven’t  understood  the  ques*on,  that’s  why  it  seems  difficult  to  me.”  (Man,  age  49,  dual-­‐headed  household)  

“I  haven’t  understood  the  ques*on  well.”    (Woman,  age  67,  single-­‐headed  household)  

What  does  the  term  “in-­‐kind  work”  mean  for  you? •  Only  three  respondents  approached  the  correct  meaning  •  Outright  incorrect  responses:  

•  Don’t  know  •  Incorrect  responses:  “own  account  work”  or  “work  to  earn  cash  or  something  else”    •  Imprecise  responses  that  suggest  a  comprehension  problem:  “work  for  your  life,”  “work  to  obtain  a  benefit,”  “work  to  survive,”  etc.    

 •  Two  respondents  indicated  the  item  may  be  sensi*ve  as  it  draws  alen*on  to  income-­‐genera*on  ac*vi*es,  which  respondents  may  consider  to  be  private  informa*on:  

 “The  ques*on  is  difficult  because  I  don’t  have  anything  to  make  a  living.”  (Man,  age  68,  dual-­‐headed  household)  

“(It’s  difficult)  because  I  don’t  want  to  tell  anyone.”  (Woman,  age  46,  single-­‐headed  household)  

Earlier  I  asked  you  if  you  do  what  you  do  regarding  different  ac*vi*es  so  that  others  don’t  think  poorly  of  you.      How  would  you  put  this  ques*on  in  your  own  words?      Is  this  ques*on  easy  or  difficult?    Why  was  it  difficult?  

“Penser  mal”  has  a  range  of  meanings…

“To  bring  forward  a  cri*cism  that  does  not  please  you”  (Man,  age  33,  dual-­‐headed  household)    “To  think  of  hur*ng  someone”  (Woman,  age  25,  dual-­‐headed  household)  

“It  means  destroying  the  life  of  a  person”  (Man,  age  32,  dual-­‐headed  household)    

“When  you  think  evil  of  a  person,  you  can  kill  that  person”  (Woman,  age  46,  single-­‐headed  household)    “A  person  who  is  there  to  kill  someone”  (Woman,  age  67,  single-­‐headed  household)  

Time  Use:  Respondents  were  asked  how  they  knew  what  :me  of  the  day  it  was. •  50%  check  the  &me  using  a  cell  phone  they  carry  with  them  “With  my  phone,  I  check  it  every  *me  I  start  an  ac*vity”  (Man,  age  36,  dual-­‐headed  household)    Nearly  half  rely  on  others:  “I  always  ask  the  neighbors  what  *me  is  it”  (Woman,  age  36,  dual-­‐headed  household)    “I  get  the  *me  from  a  friend”  (Man,  age  40,  dual-­‐headed  household)    

References:

•  Willis,  Gordon  B.  2005.  Cogni3ve  Interviewing:  A  Tool  for  Improving  Ques3onnaire  Design.  Thousand  Oaks,  CA:  Sage.  

•  Bealy,  Paul  C.  and  Gordon  B.  Willis.  2007.  “Research  Synthesis:  The  Prac*ce  of  Cogni*ve  Interviewing.”  Public  Opinion  Quarterly  71(2):287-­‐311.    

•  UNESCAP.  2010.  Guidelines  for  Cogni3ve  and  Pilot  Tes3ng  of  Ques3ons  for  Use  in  Surveys.  Sta*s*cs  Division,  Economic  and  Social  Commission  for  Asia  Pacific  Region.  

•  Johnson,  Kiersten  B.  &  Diego-­‐Rosell,  Pablo.  2014.  Assessing  the  Cogni*ve  Validity  of  the  Women’s  Empowerment  in  Agriculture  Index  Instrument  in  the  Hai*  Mul*-­‐Sectoral  Baseline  Survey.  Manuscript  under  prepara3on  for  submission  to  peer-­‐reviewed  publica3on.  

•  Blair,  J.,  Conrad,  F.,  Ackermann,  A.  and  Claxton,  G.  2006.  “The  Effect  of  Sample  Size  on  Cogni*ve  Interview  Findings.”  Paper  presented  at  the  American  Associa3on  for  Public  Opinion  Research  Conference.  Montreal,  Canada.  

•  Conrad,  Frederick  G.  and  Johnny  Blair.  2009.  “Sources  of  Error  in  Cogni*ve  Interviews.”  Public  Opinion  Quarterly  73(1):32-­‐55.  

Key  References

Photo  credits

• hlp://blogs.agu.org,  12/2012  (slide  #1)  • Chris  Blalman  (slide  #4)  • Dinesh  Maneer  Photography  (slide  #10)  • Doug  Satre  (slide  #22)  • hlp://church*mes.co.uk  (slide  #34)  

Photo  Credits