Options
Sampling techniques and weighting procedures for complex survey designs : the school cohorts of the National Educational Panel Study (NEPS)
Steinhauer, Hans Walter (2014): Sampling techniques and weighting procedures for complex survey designs : the school cohorts of the National Educational Panel Study (NEPS), Bamberg: opus.
Faculty/Chair:
Author:
Publisher Information:
Year of publication:
2014
Pages:
Supervisor:
Language:
English
Remark:
Bamberg, Univ., Diss.
Licence:
Abstract:
The National Educational Panel Study (NEPS) set up a panel cohort of students starting in grade 5 and grade 9.
To realize the corresponding samples of students NEPS applied a complex stratified multi-stage cluster sampling approach.
To allow for generalizations from the sample to the universe especially aspects of complex sample designs have to be considered and are reflected by design weights.
When applying multi-stage sampling approaches unit nonresponse, that is, units refuse to participate, may occur on each stage where decisions towards participation are made.
To correct for potential bias induced by refusals of schools and students the derived design weights need to be adjusted.
Since participation decisions differ in many ways, for example by stage (school- or student-level), time (in the forerun of or during the panel) or reasons (school-level: workload or participation in other studies, student-level: not interested in the study, resentment to testing), design weights need to be carefully adjusted to reflect the participation decisions made on each stage properly.
Participation decisions on the school level take information from sampling and the school recruitment process into account and are modeled using binary probit models with random intercept considering the federal-state-specific recruitment.
Schools participating are subsampled providing access to students in grade 5 and 9.
Subsampling within schools provides a sample of two classes if at least three are present, otherwise all classes are selected.
In creating design weights this subsampling needs again to be incorporated in the weights. The students decision process on the next stage has to be accounted for in providing unit nonresponse adjusted weights.
These decision processes take clustering at the school level as well as information on the initial sample, that is, respondents and nonrespondents, into account.
The resulting net sample forms the panel cohorts of students in grade 5 and 9.
Based on the panel cohorts each student can again decide whether to participate or not for each successive wave. Providing additional information obtained in a parental interview with one parent this multi-informant perspective makes consideration of an additional participation decision necessary. Since participation decisions of a student and a parent are unlikely independent they should be modeled appropriately using bivariate models. To again account for a cluster structure these models are extended with a random intercept on the school level.
All these aspects of complex sample and survey designs as well as the different participation decisions involved need to be considered in weighting adjustments.
The results point at typical characteristics influencing participation decisions of schools, students and parents.
Besides that the results stress the need to account for sample design and the nature of decision processes involved resulting in the actual participation.
To realize the corresponding samples of students NEPS applied a complex stratified multi-stage cluster sampling approach.
To allow for generalizations from the sample to the universe especially aspects of complex sample designs have to be considered and are reflected by design weights.
When applying multi-stage sampling approaches unit nonresponse, that is, units refuse to participate, may occur on each stage where decisions towards participation are made.
To correct for potential bias induced by refusals of schools and students the derived design weights need to be adjusted.
Since participation decisions differ in many ways, for example by stage (school- or student-level), time (in the forerun of or during the panel) or reasons (school-level: workload or participation in other studies, student-level: not interested in the study, resentment to testing), design weights need to be carefully adjusted to reflect the participation decisions made on each stage properly.
Participation decisions on the school level take information from sampling and the school recruitment process into account and are modeled using binary probit models with random intercept considering the federal-state-specific recruitment.
Schools participating are subsampled providing access to students in grade 5 and 9.
Subsampling within schools provides a sample of two classes if at least three are present, otherwise all classes are selected.
In creating design weights this subsampling needs again to be incorporated in the weights. The students decision process on the next stage has to be accounted for in providing unit nonresponse adjusted weights.
These decision processes take clustering at the school level as well as information on the initial sample, that is, respondents and nonrespondents, into account.
The resulting net sample forms the panel cohorts of students in grade 5 and 9.
Based on the panel cohorts each student can again decide whether to participate or not for each successive wave. Providing additional information obtained in a parental interview with one parent this multi-informant perspective makes consideration of an additional participation decision necessary. Since participation decisions of a student and a parent are unlikely independent they should be modeled appropriately using bivariate models. To again account for a cluster structure these models are extended with a random intercept on the school level.
All these aspects of complex sample and survey designs as well as the different participation decisions involved need to be considered in weighting adjustments.
The results point at typical characteristics influencing participation decisions of schools, students and parents.
Besides that the results stress the need to account for sample design and the nature of decision processes involved resulting in the actual participation.
GND Keywords: ; ;
Nationales Bildungspanel
Statistisches Modell
Stichprobe
Keywords: ; ; ; ;
Sampling
Weighting
Nonresponse
(Bivariate) Binary Probit with Random Intercept
National Educational Panel Study (NEPS)
DDC Classification:
RVK Classification:
Type:
Doctoralthesis
Activation date:
November 12, 2014
Permalink
https://fis.uni-bamberg.de/handle/uniba/20861