Font Size: a A A

Study On Non-sampling Errors Based On The Theory Of Multilevel Model

Posted on:2012-05-20Degree:DoctorType:Dissertation
Country:ChinaCandidate:K L WangFull Text:PDF
GTID:1117330335464513Subject:Statistics
Abstract/Summary:PDF Full Text Request
Except of sampling errors, Non-sampling errors are aroused by all kinds of reasons in statistical survey, and the more important, they can affect the result of survey badly. Combining the practice of statistical survey in our country, this paper studies the non-sampling errors in probability sampling survey in the round, which includes the reasons of resulting of non-sampling errors, the techniques of measuring non-sampling errors, and the methods of estimating and adjusting estimators when non-sampling errors occurs. This paper discusses the reasons from system, culture and application, designs the methods of measuring the percent of non-sampling errors in total survey errors, and concludes the methods and path of measuring non-sampling errors, and designs the estimators and the methods of adjusting conventional estimators using the theory of multilevel model when it comes to measurement errors or non-response errors. This paper forms the primary system and frame of studying non-sampling errors in statistical survey in our country.Under the theoretical study, this paper let multilevel model be the core of method system, and using path analysis and so on. This paper explores the way of using multilevel model in the study of non-sampling errors. The data producing in two-stage sampling or multi-stage sampling always take on multi-stage characters and are consistent with the usage of multilevel model. When measurement error appears, this paper absorbs the idea of empty models of multilevel model and designs the variance estimators of the stratum mean and the overall mean in stratified sampling. In this part, this paper also introduces the reliability index of measurement and studies the way that estimate influenced by reliability. In the analysis of the relation among multi-variables when measurement errors exist, this paper discusses the methods of estimating and adjusting estimate results using multilevel model. When non-response error appears, this paper uses the principle of multilevel model analyzing multi-stage sampling data and designs the estimators of the stratum mean in stratified sampling, which uses the method of experiential weighted. This paper tries to associate dumb variable with multilevel model to study non-response errors.Meanwhile, this paper uses other methods. When to study the techniques of measuring non-sampling errors basing on designs, this paper demonstrates the percent of non-sampling errors in total survey errors basing on the definition of mean square error, and elicits the range lower limit of the percent, and makes a relation table to show the different percents corresponding to different response rate and measurement reliability; When to study the techniques of measuring non-sampling errors basing on models, this paper studies the measurement of non-sampling errors in the investigation of single index and multi-indices by the means of path analysis. Moreover, this paper deduces the formula of measuring measurement errors quantitatively and designs the variance estimator of the stratum mean in stratified sampling.Under the applied study, this paper explores the 1600 households'consuming-data, including 11 cities, and 7 countries or town in Guangdong province and forms the primary applied system and frame of studying non-sampling errors by multilevel model theory. The demonstration analysis shows that when the sample's between-group variation is statistically notable, we should use multilevel models and introduce contextual variables. Multilevel model can simulate the characteristics of sample data batter than conventional methods with correctly designed programs and realize the measurement and adjustment of non-sampling errors. The demonstration part in this paper brings forth the path of studying non-sampling errors using multilevel model. Meanwhile, the demonstration part gives a case of measuring non-sampling errors quantitatively.
Keywords/Search Tags:Non-sampling errors, Multilevel model, Measurement, Probability sampling survey, Multi-stage sampling survey
PDF Full Text Request
Related items