TY - JOUR
T1 - The perils of straying from protocol
T2 - Sampling bias and interviewer effects
AU - Ngongo, Carrie J.
AU - Frick, Kevin D.
AU - Hightower, Allen W.
AU - Mathingau, Florence Alice
AU - Burke, Heather
AU - Breiman, Robert F.
PY - 2015/2/18
Y1 - 2015/2/18
N2 - Fidelity to research protocol is critical. In a contingent valuation study in an informal urban settlement in Nairobi, Kenya, participants responded differently to the three trained interviewers. Interviewer effects were present during the survey pilot, thenmagnified at the start of the main survey after a seemingly slight adaptation of the survey sampling protocol allowed interviewers to speak with the "closest neighbor" in the event that no one was home at a selected household. This slight degree of interviewer choice led to inferred sampling bias. Multinomial logistic regression and post-estimation tests revealed that the three interviewers' samples differed significantly from one another according to six demographic characteristics. The two female interviewers were 2.8 and 7.7 times less likely to talk with respondents of low socio-economic status than the male interviewer. Systematic error renders it impossible to determine which of the survey responses might be "correct." This experience demonstrates why researchers must take care to strictly follow sampling protocols, consistently train interviewers, and monitor responses by interview to ensure similarity between interviewers' groups and produce unbiased estimates of the parameters of interest.
AB - Fidelity to research protocol is critical. In a contingent valuation study in an informal urban settlement in Nairobi, Kenya, participants responded differently to the three trained interviewers. Interviewer effects were present during the survey pilot, thenmagnified at the start of the main survey after a seemingly slight adaptation of the survey sampling protocol allowed interviewers to speak with the "closest neighbor" in the event that no one was home at a selected household. This slight degree of interviewer choice led to inferred sampling bias. Multinomial logistic regression and post-estimation tests revealed that the three interviewers' samples differed significantly from one another according to six demographic characteristics. The two female interviewers were 2.8 and 7.7 times less likely to talk with respondents of low socio-economic status than the male interviewer. Systematic error renders it impossible to determine which of the survey responses might be "correct." This experience demonstrates why researchers must take care to strictly follow sampling protocols, consistently train interviewers, and monitor responses by interview to ensure similarity between interviewers' groups and produce unbiased estimates of the parameters of interest.
UR - http://www.scopus.com/inward/record.url?scp=84923337660&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=84923337660&partnerID=8YFLogxK
U2 - 10.1371/journal.pone.0118025
DO - 10.1371/journal.pone.0118025
M3 - Article
C2 - 25693077
AN - SCOPUS:84923337660
SN - 1932-6203
VL - 10
JO - PloS one
JF - PloS one
IS - 2
M1 - e0118025
ER -