Estimating Plausible Values with NEPS Data: An Example Using Reading Competence in Starting Cohort 6
The NaƟonal EducaƟonal Panel Study (NEPS) provides data on the development of various com-petence domains across the life span. Because research quesƟons using these competencestypically pertain to latent relaƟonships between constructs, this paper gives an overview ofthe concept of plausible values and how to esƟmate unbiased effects that account for measure-ment error in competence scores. Plausible values incorporate responses to a competence testas well as various background variables. Only
... und variables. Only if all variables relevant for the specific researchquesƟon are part of the background model, plausible values esƟmate unbiased populaƟon-level effects. Because the NEPS allows for a mulƟtude of different research quesƟons and,by design, provides a large and growing amount of background informaƟon, it is difficult toprovide plausible values that fit each conceivable research quesƟon. Therefore, the R rouƟneplausible_values() in the package NEPSscaling was developed. Its funcƟonality enablesNEPS data users to easily generate custom-tailored plausible values addressing their specificresearch needs. Because missing data are a pervasive problem in large-scale assessments,NEPSscaling also offers a sequenƟal ClassificaƟon and Regression Trees (CART) algorithm forhandling missing values in background variables. This paper introduces the concept of plausi-ble values and CART. Moreover, an applied example demonstrates how to esƟmate plausiblevalues with plausible_values().