Plausible Values (PV) are a way of describing the competencies of individuals at the group level. They allow (unbiased) estimates of effects at the population level that are adjusted for measurement errors. In contrast to point estimators such as Weighted Likelihood Estimates (WLE), the use of Plausible Values is suitable for more precise inferential statistical tests in correlation and mean value analyses.
Plausible Values are based on the individual answers in the competence tests and additional background characteristics (e.g. gender, age, socioeconomic status). For each person, the probability distribution of his or her competence is first determined and then several values are randomly drawn from it (hence “Plausible Values”). Hypothesis tests for the specific question of interest are calculated for each of these values and combined into an overall result.
NEPS Survey Paper No. 71 with further information on Plausible Values
Article by A. Scharl and E. Zink in Large-scale Assessments in Education (2022) on Plausible Values
R package NEPSscaling
The R package NEPSscaling enables users to generate own Plausible Values with a background model adapted to the specific research question. Furthermore, the package is able to handle missing values in the background model. The following functions are implemented in NEPSscaling:
estimation of Plausible Values for competence data of the NEPS
suitability for cross-sectional and longitudinal linked competence data
imputation of missing values
storage of Plausible Values in the data formats R, Mplus and Stata
Command to install NEPSscaling in R
NEPSscaling can be installed by the following command:
Documentation of NEPSscaling
A detailed description of the R package NEPSscaling, its functionality and handling as well as additional information on the Plausible Values offered in the Scientific Use Files can be found in the above mentioned NEPS Survey Paper No. 71 by Anna Scharl, Claus H. Carstensen and Timo Gnambs (2020).
Plausible Values in the Scientific-Use-Files
The Plausible Values provided in the Scientific Use Files can be used for exploratory analyses. For the actual hypothesis testing, however, it is advisable to consider all variables that are part of the analysis of interest as background information when generating the Plausible Values. This includes all covariates and non-linear terms (e.g. quadratic effects, interactions). Due to the complex NEPS data structure, the range of potential analysis scenarios for competence data cannot be covered by a single background model. All Plausible Values in the Scientific Use Files are based on a strongly reduced background model, which means that an unbiased estimate cannot be guaranteed in every case of application. It is therefore recommended to generate separate Plausible Values for specific research questions with an appropriately adapted background model using the R package NEPSscaling.
File versions of NEPSscaling and examples