A unified semi-empirical likelihood ratio confidence interval for treatment effects in the two sample problem with length-biased data

In two sample studies, the treatment effects that we are interested in may have different types, such as mean difference, the difference of probabilities, etc. In this work, we apply semi-parametric empirical likelihood principle to length biased data and derived a unified empirical likelihood ratio confidence interval for treatment effects. The empirical likelihood ratio is shown to be asymptotically distributed as chi-squared. Simulation studies show that the proposed confidence interval has a better performance compared with its counterpart which is based on normal approximation. The severe effect caused by ignoring the length bias is also illustrated by simulation. The proposed method is applied to Oscar data to study the effect of high socio-economic status on lifetime.