1: \begin{abstract}

2: Sample surveys are widely used to obtain information about

3: totals, means, medians, and other parameters of finite populations.

4: In many applications, similar information is desired

5: for subpopulations such as individuals in

6: specific geographic areas and socio-demographic groups.

7: When the surveys are conducted at national or similarly high levels,

8: a probability sampling can result in just a few sampling units from many

9: unplanned subpopulations at the design stage.

10: Cost considerations may also lead to low sample sizes from

11: individual small areas.

12: Estimating the parameters of these

13: subpopulations with satisfactory precision

14: and evaluating their accuracy

15: are serious challenges for statisticians.

16: To overcome the difficulties,

17: statisticians resort to pooling information across

18: the small areas via suitable model assumptions,

19: administrative archives, and census data.

20: In this paper, we develop an array of small area quantile estimators.

21: The novelty is the introduction

22: of a semiparametric density ratio model for the

23: error distribution in the unit-level nested error regression model.

24: In contrast, the existing methods are usually most effective when the

25: response values are jointly normal.

26: We also propose a resampling procedure for estimating

27: the mean square errors of these estimators.

28: Simulation results indicate that the new methods

29: have superior performance when the population distributions

30: are skewed and remain competitive otherwise.

31: \end{abstract}