Jack Fowler, PhD
Center for Survey Research UMass Boston
Population Research Seminar Series
Session 4: Standardized Survey Interviewing
Part 1: Controlling Interviewer Related Error
What Interviewers Are Supposed to Do in Order to Be Consistent (Standardized)
• Read questions as worded
• Use nondirective probes when questions are not answered adequately
– Among other things, those probes should not
materially change the content of the questions that other respondents are answering so that all
respondents are answering as close to the same question as possible
Directive Probes
Any probe that can be answered with a ―yes‖ or ―no‖ is a directive probe
Why Does It Matter?
• Error 1: The more interviewers innovate (eg. when they probe) the
more they affect the data
• Error 2: Small changes in question wording sometimes have big effects on answers
Why Are There Problems?
• Bad question design
• Inadequate interviewer training
• Inadequate interviewer monitoring • Interviewer fatigue
Question Design and Wording
• Complex terms and concepts • Multi-barrelled questions
Three Barrels
• This question is about automobile injuries, including injuries from crashes, burns, and any other kind of accidents. Were you ever injured in a car accident because of your driving?
Poor Sequencing
• Have you been the victim of a crime in the last 12 months?
Poor Sequencing II
• Do you do any kind of activity for exercise on a regular basis?
• What are the different kinds of things you do for exercise?
Poor Sequencing III
• Including any days spent in a hospital, between
[START DATE] and [END DATE], did illness or injury keep
[PERSON] from work more than half of the day, not
counting work around the house?
• What were the dates that [PERSON] missed more
than half a day from work between [START DATE] and
[END DATE]?
• How many actual days was that between [FIRST DATE] and [END DATE]?
Unclear How to Respond
• When did you first vote in a Presidential election?
– While I was in college – During the Vietnam war – About 10 years ago
– In 1972
• In what year did you first vote in a Presidential election?
Unclear How to Respond II
• What kind of museums have you visited in the past year? *(field coded)
– Art museums
– Science museums – History museums – Other
Unclear How to Respond III
• How did you get to the game yesterday?
– I drove
– I got a ride with friends – By car
– We circled around on Elm Street to avoid the traffic
– I stayed up the night before to get my homework done
Cognitive Testing
• Identify ambiguous terms and define them • Identify complex questions and break
them up
• Identify questions for which it is not clear how to answer and build that information into the question
Behavior Coding
• During pretests, identify questions
– That interviewers have trouble reading as written
– That respondents have difficulty answering without additional probing
– For which respondents frequently ask for clarification
Intraclass correlation
• This is a measure of how much
interviewers affect the answers they obtain.
• Obviously, the ideal would be if the value is 0.
Biggest correlates of
interviewer-related error
• The rate at which questions have to be probed, which in turn is related to:
– Lack of clarity of what the question is asking – Lack of clarity about how to answer the
Train Interviewers I
% Who read questions acceptably by length of training
Training Days Read Acceptably
< 1 40%
2 62%
5 79%
Train Interviewers II
Percent who probe open ended questions acceptably
Training Days Probe Acceptably
< 1 13%
2 31%
5 29%
Train Interviewers III
• Percent who probe closed questions acceptably
Training Days Probe Acceptably
< 1 48%
2 67%
5 72%
Why Monitoring Matters
• Compared interviewers who were tape recorded with those whose interviewing was not reviewed by a supervisor
Interviewer Effects On Answers
Monitored Not Monitored Measure of Interviewer effect on data (lower is better)* .008 .011 *Intraclass correlation P<.05Monitoring and Bias
Monitored Not Monitored Measure of how much interviewers biased results (higher is better)* 34 2 *P<.04What Are We Monitoring For?
• Reading questions • Good probing
• The way interviewers interact with respondents
Interviewers Set Standards
For Respondents
• The way an interviewer behaves has an important effect on what respondents do and how well they perform
Experience Can Be A Negative
(NSDUH)
• Reporting illicit drug use in lifetime
Percent “Yes”
Experienced Interviewer
33%
Experience Can Be A Negative II
• Percentage of known visits to doctors reported by week of data collection
Interviewer Week Percent NOT Reported 1 18% 2 21% 3 23% 4 27% 5 29%
Interviewers Communicate
Expectations
Interviewer concern about efficiency
Respondent said interviewer wanted Exact Answers General Ideas TOTAL Low 77% 23% 100% Medium 62% 38% 100% High 66% 34% 100%
Interviewers Communicate
Expectations II
Interviewer concern about accuracy Respondent said interviewer wanted Exact Answer General Ideas TOTAL Low 62% 38% 100% High 76% 24% 100%Interviewer Expectations Affect
Data
• Please take your time and think carefully about your answers. It is important that you answers be as accurate as possible
Reported Media Use by Interview
Procedure (Miller)
Experimental Control
Watched TV previous day 86% 66%
Listened to radio previous day 67% 65%
Read newspaper previous day
83% 77%
Seen an X-rated movie 61% 51%
Mean number of books read
Monitoring Matters
• To show we care about reading as worded and probing
• To help experienced interviewers not get sloppy and tired
• To make sure interviewers are
communicating that interviewing is a serious data collection effort
How about recruitment?
• Other than the obvious skills—following directions, understandable speaking,
computer skills as needed—there are essentially no guidelines for selecting good interviewers
Matching
• If the obvious characteristics of
interviewers are related to answers, may see an effect
• For most questions, demographic
characteristics unrelated to answers if interviewers are well trained
How to minimize
interviewer-related error
1. Write good questions and test them 2. Train the interviewers in good
standardized practices
3. Script interviewers to emphasize data quality
4. Monitor how the data collection process is carried out