Nonparametric-likelihood inference based on cost-effectively-sampled-data

Costs associated with the evaluation of biomarkers can restrict the number of relevant biological samples to be measured. This common problem has been dealt with extensively in the epidemiologic and biostatistical literature that proposes to apply different cost-efficient procedures, including pooling and random sampling strategies. The pooling design has been widely addressed as a very efficient sampling method under certain parametric assumptions regarding data distribution. When cost is not a main factor in the evaluation of biomarkers but measurement is subject to a limit of detection, a common instrument limitation on the measurement process, the pooling design can partially overcome this instrumental limitation. In certain situations, the pooling design can provide data that is less informative than a simple random sample; however this is not always the case. Pooled-data-based nonparametric inferences have not been well addressed in the literature. In this article, a distribution-free method based on the empirical likelihood technique is proposed to substitute the traditional parametric-likelihood approach, providing the true coverage, confidence interval estimation and powerful tests based on data obtained after the cost-efficient designs. We also consider several nonparametric tests to compare with the proposed procedure. We examine the proposed methodology via a broad Monte Carlo study and a real data example.