Measurement error

Measurement rarely, if ever, occurs without error. Error can be from multiple sources. Ideally, error should be minimised by careful design and data collection, however in statistical analysis some modeling of measurement error can be incorporated.

Measurement error is generally thought to consist of:

  1. Systematic error
  2. Random error[1]

Thus, a measured score can be conceptualised as consisting of:
Real score + systematic error + random error.

Systematic measurement error includes all reproducible error (sometimes referred to as bias), including:

  1. Sampling error
    1. e.g., non-representative sample
  2. Non-sampling error
    1. Paradigm error - a scientific approach privileges and preferences study of measurable phenomena
    2. Researcher bias - a researcher is keen to confirm his/her preferred theory and this influences decisions e.g., about what to measure and how
    3. Participant bias - participants are influenced by social desirability, yea-saying, nay-saying etc.
    4. Reliability and validity of measurement tool

Random error (sometimes called noise) includes all random factors that influence measurement by inflating or deflating measured values from the true score. For example, at the time of testing some participants may be in a good mood, while others made be in a bad mood, and this may affect their responses. Random error increases the variability in data.

See also