enEnglish
CROS

Bias

Document date: 
Wednesday, 8 May, 2019
Language: 
English

Bias

DEFINITION:
Bias is an effect which deprives a statistical result of representativeness by systematically distorting it, as distinct from a random error which may distort on any occasion but balances out on the average.

SOURCES:

Primary source: CODED-Peer reviews

Secondary sources: CODED-Memobust glossary, CODED-Housing price statistics, CODED-purchasing power parities

INTERPRETATION:

Bias refers to an error that, for explicit, technical or implicit reasons, is systematic in nature. The Bias corresponds to the expected value by which the statistic systematically over- or under-estimates the value of a population parameter. Formally, the Bias of an estimator T(θ is the difference between the mathematical expectation of the parameter estimated from the data and the value of the parameter θ : BIAS(T(θ ))=E[T(θ )]- θ . If the Bias of an estimator is zero, the estimator is said to be unbiased. Depending on the source of the Bias different methods are used for the calculation of the Bias. For example, one can use models which involve latent Variables which can influence the Bias (confounding), data splitting, or simulation.

CONTEXT:

In the case of Administrative data there are several reasons (sources) why a Bias gradually can enter into the results. The most important are: (i) Selection Bias due to the selection of units in the Administrative source (which can imply Coverage errors); (ii) Response Bias due to the missing or wrong values of the Variables in the Administrative source; (iii) Concept Bias which refers to the difference in definition of the Variables in the Administrative source and the intended definition; (iv) Process Bias due to the measurement method and further processing activities which cause a lack in reliability and validity of the results.

Bias is an important measure/indicator for assessing the Output quality in Multisource statistics (see the Quality Guidelines for Multisource Statistics (QGMSS) and the Quality Measures and Calculation Methods (QMCMs) of the ESSnet KOMUSO). 

ADDITIONAL INFORMATION:

In the case of survey sampling the sampling Bias due to the failures in the random selection of the units is of utmost importance. This source of Bias is of minor importance for Administrative data. The role of the Bias for the assessment of the quality of statistics is discussed in the Memobust handbook, Theme: Quality of statistics. (See also: P.P. Biemer: “Total Survey Error. Design, Implementation and Evaluation”. Public Opinion Quarterly 74 (2010), 817–848.) The importance of the Bias for assessment of the quality of Frames is discussed in the ESSnet KOMUSO-SGA3, WP2.

RELATED TERMS:

Endogenous selection , Estimation, Measurement error, Process quality, Root mean square error

 Cross-linked glossary TOP