Building Bridges Between Populations and Samples in Epidemiological Studies

Abstract

Key Words sample design, statistical inference, sample weights, analysis of data from complex samples s Abstract The increased use of rigorous population-sampling methods and the analysis of data from those samples in cross-sectional surveys, case-control studies, longitudinal-cohort investigations, and other epidemiological research efforts have raised important statistical issues for health analysts. We describe the origin, implications, and some plausible resolutions for several of these issues. Some of the main issues we consider include (a) establishing whom the sample represents; (b) using sample weights; (c) understanding the role of other important features, such as the use of sampling stratification and the selection of clustered groups of population members; and (d ) finding ways to analyze study data with key sampling features in mind. Ultimately, resolution of all of these issues requires that analysts clearly define a reference population and then understand the role of design features in relating sample results to that population. POPULATION SAMPLING IN EPIDEMIOLOGYHistorically, most empirical knowledge has been based on incomplete observation and therefore incomplete samplings of the human experience (46). In each such case there has been a population (i.e. some collection of persons or objects about which knowledge was sought) and a sample (i.e. a portion of the population to be observed, thus providing an informational basis for acquiring knowledge about the population).The need for sampling in epidemiological research stems from the nature of several overlapping research designs that are commonly used in the field (17,39,45). For each of these designs, when it is impractical to examine the entire population, statements are made about a targeted group of individuals called the study population, based on observations obtained from a representative portion of the population.

What is a representative brain? Neuroscience meets population science

“…Population scientists have extensive experience using a variety of methods to reduce bias, increase power, and improve causal inference in smaller sample studies (87). These methods include knowledge of matching sampling frames and target populations to reduce coverage error, reducing sampling costs by using cluster samples and subsampling within clusters, and improving representativeness through stratified sampling (23,(88)(89)(90)(91). For example, developmental scientists interested in the role of psychological resources on limbic system reactivity during adolescence might benefit from input from population scientists in selecting specific subgroups of adolescents who represent the demographic profile of high-and low-risk adolescents within the United States.…”

Analysis of Regression Based on Sampling Weights in Complex Sample Survey: Data from the Korea Youth Risk Behavior Web-Based Survey

“…This stratified sampling method is identically used in most national statistical surveys such as Community Health Survey, Working Condition Survey and Korean Longitudinal Study of Ageing (KLoSA), as well as Health Care Survey [12][13][14]. When estimating population in national data extracted by stratified sampling, serious bias may arise if analysis is conducted while ignoring weights [6]. In this study, there was also a difference in population rate and tendency in the prevalence rate of obesity.…”

“…Traditionally, a linear regression model has been performed to analyze potential risk factors in continuous data while cluster sampling does not follow tdistribution since correlation exists among observed values [8,9]. In addition, Lv , et al, [7] also pointed out that when only frequency weight is applied to cluster sampling data, there is a possibility that significance levels of all variables may become smaller as the size of the sample becomes bigger [6]. Nevertheless, many studies do not gain accurate analysis results as they perform statistical analysis , wrongly assuming the data from stratified sampling as those from simple random sampling [5,8,15].…”

“…In addition, even the studies applying weights may have errors in results [5] if they conduct analysis in the frequency weight method of existing single sample analysis, because the data source and national statistical survey, drew out samples by complex sample design using stratification and cluster sampling [4]. Although preceding studies on complex sample design [5][6][7][8][9] have raised the possibility of error due to application method of weights, they have been limited to theoretical studies and there have been few studies that verified the difference in results depending on analysis method.…”

Regression analysis of Complex Sample Survey Data: results from a national Survey

“…Although preceding studies on complex sample design [1][2][3][4] have raised the possibility of error due to application method of weights, they have been limited to theoretical studies and there have been few studies which verified the difference in results depending on analysis method.…”