cognitive pre-testing
DESCRIPTION
PRTRANSCRIPT
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
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
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
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