400 likes | 502 Views
POPULATION RESEARCH SEMINAR SERIES Sponsored by the Statistics and Survey Methods Core of the U54 Partnership. Innovations in Survey Research - the Nonverbal Response Card: Evidence from Gilgel Gibe, Ethiopia Megan Klein Hattori, PhD UMass Boston David P. Lindstrom, Brown University
E N D
POPULATION RESEARCH SEMINAR SERIES Sponsored by the Statistics and Survey Methods Core of the U54 Partnership Innovations in Survey Research - the Nonverbal Response Card: Evidence from Gilgel Gibe, Ethiopia Megan Klein Hattori, PhD UMass Boston David P. Lindstrom, Brown University Craig Hadley, Emory University Support for this work was provided by NIH, the David and Lucile Packard Foundation, the Compton Foundation and the Mellon Foundation
POPULATION RESEARCH SEMINAR SERIES Sponsored by the Statistics and Survey Methods Core of the U54 Partnership Questions? Comments? Type them in or ask over your webcam/microphone Or send an email to u54.ssmc@gmail.com
Challenge How can we solicit responses to sensitive questions about sexual attitudes and behavior… • in a social context where certain attitudes and behaviors are stigmatized, and • in an interview context that is not entirely private.
Plan • Study design • Non-verbal response card • Descriptive results • Multivariate tests for interactions between to see if response bias varies across groups • Reliability tests (yes/no and numeric) • Next steps
Study characteristics (Gilgel Gibe dam area, southwestern, Ethiopia) • Study population • Rural, semi-urban, and urban youth (ages 13-24) • 90% Muslim, 8% Ethiopian Orthodox, 2% Protestant • 40% never attended school (literacy is not universal, and uneven literacy skills among educated youth in two languages) • 93% Oromo, 2% Amhara, 2% Yem, 3% other ethnic group • Study objectives • Measures for sexual attitudes, behavior, and knowledge. • Survey mode • Face-to-face interviews by same-sex multilingual interviewers.
Three dimensions of sensitive questions that generate reporting bias • Social desirability Desire to project positive imagine of oneself to the interviewer (can produce either under-reporting and over-reporting). • Invasion of privacy Intrusive questions can produce refusals or intentional misreporting. • Risk of disclosure Concerns about who else might be listening.
Impact of data collection mode on response effects • Studies are very clear that some form of self-administration for sensitive questions reduces level of non-response and misreporting. • However, important advantages offered by presence of an interviewer • higher overall participation rates, • question clarification, • fewer invalid responses.
Criteria for choosing response method • Enhance privacy of responses (must be non-verbal) • Not require literacy • Place minimal cognitive demands on the respondent • Functional in a low-tech environment • Low cost • Highly portable
Prior approaches to asking sensitive questions • Self-administered paper and pencil (for sensitive sections). • Computer-assisted self-interviewing (CASI, ACASI). • Random response technique. • Ballot box.
Gilgel Gibe Social and Sexual Relationship History Survey (March – June 2006) • Objectives:Understand the processes by which young people establish romantic relationships and become sexually active, and identify the determinants of the pathways to risky sexual behaviors. • Location:Gilgel Gibe dam area, southwestern Ethiopia. • Survey Design:Survey interviews with 1,269 youth and young adults ages 13-24. • Topics:Utilization of local health services, awareness of HIV risk, formation of romantic relationships, progression of physical intimacy to sexual activity, context of first intercourse and risky sexual behavior.
Design and Testing of Response Cards • Randomized control trial design: One-half of sample randomly assigned conventional verbal response mode, and other half assigned non-verbal response card. • Card method used with 50 questions about sexual behavior, knowledge and attitudes. • Pre-test of survey instrument and response cards with 202 randomly selected youth and young adults in an urban community. • Final survey of 1,269 randomly selected youth, female interviewers used with female respondents, and male interviewers with male respondents.
Indicators of interest • Knowledge • Attitude • Behaviors
Selected Descriptive ResultsKnowledge, Attitudes, Perceptions, and Behavior
Table 1. Background Information for Full and Pilot Survey by Response Method, Youth Ages 13-24, Southwestern Ethiopia.
Table 2. Percent of Respondents who Report Knowing a Place Where They Would Feel Comfortable Getting Condoms, Youth Ages 13-24, Southwestern, Ethiopia.
Table 3. Percent of Respondents who Believe it is Acceptable for a Young Woman to have Sexual Intercourse When She is Going Steady, Youth Ages 13-20, Southwestern Ethiopia.
Table 4. Percent of Respondents who Believe it is Acceptable for a Young Man to have Sexual Intercourse When He is Going Steady,Youth Ages 13-24, Southwestern Ethiopia.
Table 5. Percent of Respondents who Report Being at Risk of Contracting HIV in the last 12 Months, Youth Ages 13-24, Southwestern Ethiopia.
Table 6. Percent of Respondents who Report Non-marital Sexual Intercourse in the last 12 Months, Youth Ages 13-24, Southwestern Ethiopia.
Multivariate ResultsDoes Response Bias Vary Across Subgroups?
Does Response Bias Vary Across Subgroups? • Estimate logistic regression models • Include interaction terms with card effect to test whether response bias is • larger for specific subgroups.
Figure 1. Interaction Models: Odds Ratios from Logistic Regression Predicting Condom Knowledge, Youth Ages 13-24, Southwestern Ethiopia.
Figure 2. Interaction Models: Odds Ratios from Logistic Regression Predicting Acceptance of Girls Pre-marital Sex, Youth Ages 13-24, Southwestern Ethiopia.
Figure 3. Interaction Models: Odds Ratios from Logistic Regression Predicting Premarital Sex, Youth Ages 13-24, Southwestern Ethiopia. Card Method and Male Interaction Odds Ratios
Reliability for Yes/No Answers: Never Married Youth who Have Been in a Romantic Relationship Note: V= Verbal, C=Card Significance levels for difference of proportions (means) between the card responses, **p<0.01, *p<0.05, †p<0.10.
Reliability for Numeric Answers: Age at Marriage and Age at First Sex for Married Youth Note: V= Verbal, C=Card Significance levels for difference of proportions (means) between card responses, **p<0.01, *p<0.05, †p<0.10.
Next Steps • Analysis of Third Round Adolescent Survey (JLFSY) Test second generation card configurations and can capture individuals who report differently on card and verbal at different points in questionnaire. • Mini-DHS Ethiopia, Vietnam, Guatemala. Test effectiveness of card for DHS style survey.
Figure 3. Interaction Models: Odds Ratios from Logistic Regression Predicting Perception of HIV Risk in Last 12 Months, Youth Ages 13-24, Southwestern Ethiopia.