Design-based inference in time-location sampling

Area of Science:

• Epidemiology
• Biostatistics
• Public Health

Background:

• Time-location sampling (TLS) is a key method for surveying hard-to-reach populations, particularly those at high risk of infectious diseases.
• TLS relies on reaching individuals at specific times and locations where they congregate.
• Key statistical challenges in TLS include the use of sampling weights and accounting for individual frequency of venue attendance (FVA).

Purpose of the Study:

• To contextualize TLS within sampling theory.
• To detail the calculation of sampling weights for TLS.
• To propose a design-based inference method that incorporates FVA.

Main Methods:

• The study presents TLS within the framework of sampling theory.
• It details methods for calculating sampling weights.
• A design-based inference approach accounting for FVA is proposed and evaluated.

Main Results:

• Ignoring FVA in TLS can lead to significant bias in prevalence or total estimations.
• The proposed design-based estimator, which incorporates FVA, provides unbiased estimates.
• This unbiasedness holds true even when individuals make declarative errors in reporting their FVA.

Conclusions:

• Accurately accounting for frequency of venue attendance (FVA) is critical for unbiased estimation in time-location sampling (TLS) surveys.
• A design-based inference method that incorporates FVA offers a robust solution to potential biases.
• This approach enhances the reliability of survey data from high-risk populations, even with imperfect self-reported data.