Methods in Epidemiology

This book describes the variety of direct and indirect population size estimation (PSE) methods available along with their strengths and weaknesses. Direct estimation methods, such as enumeration and mapping, involve contact with members of hard-to-reach groups. Indirect methods have practical appeal because they require no contact with members of hard-to-reach groups. One indirect method in particular, network scale-up (NSU), has several strengths over other PSE methods: It can be applied at a province/country level, it can estimate size of several hard-to-reach population in a single study, and it is implemented with members of the general population rather than members of hard-to-reach groups. The book discusses methods to collect, analyze, and adjust results and presents methods to triangulate and finalize PSEs.

Auteur
George W. Rutherford is the Salvatore Pablo Lucia Professor of Epidemiology, Preventive Medicine, Pediatrics and History at the University of California, San Francisco School of Medicine.

Contenu

Chapter 1: Review of Population Size Estimation Methods
This chapter provides on overview of existing PSEs that count hard-to-count populations. We compare and contrast different PSE methods with the network scale-up (NSU) method. We provide illustrative examples on how the readers can apply basic calculations for each PSE method. We then ask readers to consider the applicability of different PSE methods, particularly the NSU method to their settings and hard-to-reach populations.

Chapter 2: Methods to Estimate the Average Social Network Size
In this chapter, we present direct and indirect methods (by using reference groups) to estimate the average social network size, which is one of the primary variables for NSU methods. Using real examples from Iran, Georgia and Kenya, we demonstrate and discuss steps in using direct and indirect methods and how to interpret the results. Methodological issues such as assessment of background characteristics on social network size; influence of missing data, method of estimation, digit preference, exclusion of unreliable reference groups etc., and recall bias are explicitly addressed. The chapter is supplemented by several analytic R codes and Excel tools that allows readers to better understand the methods and use them in their own projects. At the end of the chapter, we provide recommendations on how design and analyze a new project to estimate social network size.

Chapter 3: Estimation of Size of Hard-to-count Populations using network scale-up
This chapter provides details on how to estimate population sizes the NSU method. We present and discuss methods to assess and adjust estimates for transparency and popularity biases in the NSU's estimates. Different approaches to estimate correction factors are discussed. We also discuss the influence of several factors such as method of data collection, missing data, and order of questions. Advanced Markov chain Monte Carlo algorithms to impute missing data and improves estimations will be provided.

Chapter 4: Methods for Smoothing, Extrapolation and Triangulation of Population Counts
Often policy makers ask for PSE estimates at city or district level. Due to inadequate sample sizes at city or district level, PSE estimates are often highly variable. This chapter provides advanced Bayesian methods to smooth sparse data. The method will be illustrated by an Excel tool which readers can simply use to do the Bayesian calculations. In addition, extrapolation of PSEs beyond the study sites (e.g. from subnational level to national) is also a common practice. To do so appropriately, we present a new count regression model that takes into account the spatial autocorrelation using penalized estimation methods. And finally, we provide recommendation on how triangulate population size data with other existing data and prior knowledge, and making consensus on final results.