Understanding the drivers of sensitive behavior using Poisson regression from quantitative randomized response...

Area of Science:

• Social Sciences
• Statistics
• Behavioral Science

Background:

• Studying sensitive behaviors is crucial for effective interventions but challenging due to social desirability bias and fear of retribution.
• Quantitative Randomized Response Technique (QRRT) estimates the frequency of sensitive behaviors but lacks regression methodology for its count data.
• Existing methods struggle to identify drivers of non-compliant behavior when using QRRT data.

Purpose of the Study:

• To develop a novel Poisson regression methodology for analyzing quantitative randomized response technique (QRRT) data.
• To enable the identification of potential drivers influencing the quantity of sensitive, non-compliant behaviors.
• To provide a statistical framework for understanding factors contributing to rule-breaking.

Main Methods:

• Developed a Poisson regression model for QRRT data using maximum likelihood estimation via the expectation-maximization (EM) algorithm.
• Derived the Fisher information matrix to compute the asymptotic variance-covariance matrix for regression parameter estimates.
• Validated the methodology through simulations and a case study on hunting regulation non-compliance in Sierra Leone.

Main Results:

• The new Poisson regression methodology effectively analyzes QRRT data to identify drivers of non-compliant behavior.
• Simulation results confirmed the accuracy of asymptotic approximations for the regression parameter estimates.
• The case study successfully illustrated the assessment of potential drivers for varying quantities of non-compliant behavior.

Conclusions:

• The developed Poisson regression methodology significantly advances the study of sensitive behaviors using QRRT data.
• This approach allows for a robust assessment of factors influencing non-compliance, offering valuable insights for intervention design.
• Free, open-source software is available to support the application of QRRT regression, promoting wider research use.