Systematic
Systematic error affects test validity and may introduce bias. Bias can be defined as a condition that causes a result to depart from the true value in a consistent direction (Gibson, 2005). The accuracy of a measurement is reduced as the mean and median values are altered. Unlike in random error, precision is not reduced. An example of systematic error is the over-reporting of physical activity undertaken or the under-reporting of dietary intake, thus resulting in a biased and invalid measurement of diet or physical activity. Another example of systematic error would be a pedometer which over counted steps by a constant 15%. Reducing systematic error requires a different procedure to that of reducing random error; repeating the number of measures will not reduce the error associated with misreporting.
The principal sources of bias are selection and measurement bias. It is not possible to remove bias by statistical analysis unless a calibration study has been undertaken.
Example - Calibration of two objective measures of physical activity for children
Example - European Prospective Investigation into Cancer and Nutrition (EPIC) calibration study: rationale, design and population characteristics.
Selection bias may result if the characteristics of the population who undertake the measurement of diet or physical activity are different to those of the general population, this may be particularly an issue if the sub sample is self-selected or those who do not respond to an initial attempt to recruit to a study are not followed up. Individuals who volunteer for a study may have higher motivation levels, be better educated and have a high self esteem, for example. Individuals who are willing to keep a diary of their food intake or physical activity patterns may have lifestyle habits that are not representative of the wider population. Any study about food or physical activity may attract people who are anxious and overly concerned about these behaviours.
Considering the following parameters associated with the subjects may help identify or adjust for bias:
- Random or self-selected?
- Lifestyle, cultural, social characteristics
- Physiological characteristics
- Psychological characteristics
- Motivation levels
- Bias linked to the research question itself
Measurement bias may be due to equipment fault, analytical bias, social desirability, interviewer bias or recall bias.
Example - Dietary assessment in Whitehall II: the influence of reporting bias on apparent socioeconomic variation in nutrient intakes.
References
Gibson R 2005
Principles of Nutritional Assessment
Oxford University Press, Oxford
Second Edition