Statistical methods for the analysis of time-location sampling data

Area of Science:

• Epidemiology
• Biostatistics
• Social Sciences

Background:

• Time-location sampling (TLS) is a method for reaching hidden populations.
• Traditional analysis of TLS data assumes simple random sampling (SRS), potentially causing bias.

Purpose of the Study:

• To propose and evaluate a weighting procedure for analyzing TLS data as a two-stage sample survey.
• To address biases and underestimation of uncertainty in SRS analysis of TLS data.

Main Methods:

• Proposed a weighting procedure based on the inverse probability of sampling.
• Utilized sample survey analysis software to estimate standard errors, accounting for clustering and weights.
• Applied the method to data from the Young Men's Survey Phase II (MSM study).

Main Results:

• Weighting TLS data affected point prevalence and association estimates compared to SRS analysis.
• Weighting and clustering substantially increased standard error estimates, improving uncertainty assessment.
• Demonstrated the impact of analysis methods on estimates for hard-to-reach populations.

Conclusions:

• Analyzing TLS data using a two-stage survey approach with weighting and clustering adjustments is crucial for accurate estimates.
• This method corrects for biases inherent in SRS analysis of TLS data.
• Highlights the importance of appropriate statistical methods for sampling hidden populations.