Fraudulent manipulation is likely to attempt to show a desired effect by manipulating the data in any or all of the three ways that affect the statistical significance of the results noted above:
• larger differences between groups appear
• variability of results is reduced
• extra data are invented.
The main way is by reducing variability, so that observations that do not "fit" the desired result are deleted or amended. Simple invention of data tends to result in a series of values that are too close to each other. It needs considerable familiarity with the field of study, or the examination of other people's results to be able to recognise this reduced variability.
The number of changes made on original record forms may also be increased when manipulation of the data takes place by changing the record forms.3
In some ways this type of fraud is harder to detect when simple summaries of the data are being read because the values themselves will not necessarily be manipulated with any target result in mind. The fabricator in this instance will not usually be quite as sophisticated in their approach to data and so there will be special characteristics seen in the ersatz, which are not seen in the authentic.
There are two main characteristics of this type of data. Firstly, there tends to be too little variation and an almost total absence of outliers. Secondly, because human intervention is totally responsible, there will often be digit preference in measurements that would not normally show it (for example, sodium concentration in urine).There may be patterns of digits (such as psychologically preferred pairs of numbers) with the invented data. The shape of the invented distribution tends to be relatively flat with values being generated in an even spread over a limited range.
The record forms where invented data have been created from the start do not have many alterations at all, and have too regular writing. If all the forms have been completed in the same pen, this may be spotted by those involved in entering or checking the record forms from a clinical trial.
Was this article helpful?