Errors

When we compare study group/s with a control group in a research, there can be ‘errors’. Error is the difference between the ‘fact’ and our ‘finding’. In other words, error is the distortion in ‘population parameter’ when compared to the ‘true value’. Difference between the ‘fact’ and our ‘finding’ can be due to three reasons Actual difference between groups Bias (Synonyms - systematic errors, non-sampling errors) Random (Synonyms- by chance error, nonsystematic errors, sampling error, noises) Random error Difference between the ‘fact’ and our ‘finding’ is purely due to ‘chance’ alone. Random error is non directional and will average out in repeated samples (result in some samples will show overestimation and in some samples show underestimation). Whenever we study a sample (which we normally do), random errors will invariably creep in. Random error will be ‘zero or minimal’ only when we study the entire population. As sample size increases random error decreases and hence reliability (precision) of the research increases. Reliability (precision) simply means repeatability of the measurements we make, that is, repeated researches under the similar conditions give identical results. Reliability doesn’t guarantee accuracy (validity) of the measurements though it’s a pre-requirement of validity. Bias Any error introduced into the study for which a cause can be identified is called as bias. Result of the research will show repeated overestimation or underestimation indicating a directional and systematic difference from the true value. Bias is considered ‘directional’ because the result will not average out in repeated samples (result in every sample will be either over or under estimation). Validity of the study will decrease if the bias is more. Validity is the ability of the research to measure what it is intended to measure. A research is valid only if its result corresponds to the truth. Bias has nothing to do with the sample size. Strong study design and adherence to the research protocol will help to reduce the bias. David L Sacket identified 19 bias and later Bernard Choi extended the list to 65. Even though both words are considered synonyms there are some non-sampling errors other than bias. Wrong instrument for measurement, improper variable definition etc fall in that list. Type 1 (α) and Type II (β) errors In αerror, a true null hypothesis will be rejected. It is also called ‘false positive error’ because we have proved some association which truly is not there. Normally in medical/dental research, maximum permissible level of α error is 5% (chance that observed difference due to chance/sampling error is less than 5% OR probability of incorrectly rejecting the null hypothesis is less than/equal to 5%). αlevel of significance is set by the researcher before the statistic is computed. Once the null hypothesis is rejected, the only error possible is ‘α’. In βerror, a false null hypothesis will be accepted. It is also called ‘a false negative error’ because our research did not identify an association which in fact is existing. Probability of committing Type II is called βand is usually kept bellow 20%. If null hypothesis is not rejected, the only error possible is β. Whenever null hypothesis not rejected, the researcher should address level of βerror and power. Power (1-β) is the ability to reduce βerror and it is the probability that a false null hypothesis is correctly rejected. Non rejection of null hypothesis can also be because of low power of the study.