The conditions that they were placed in effected the results.
Furthermore, in relation to natural sciences, the conditions also affect the results that are produced from the re-test. In the natural sciences, these conditions need to be closely controlled as sometimes one degree Fahrenheit may be the difference in 102 bacteria per/hour and 116 bacteria/hour. The natural sciences are highly sensitive to changes in pressure, temperature, and even sound frequency. Thus, conditions must be made so exactly perfect to ensure that the data is not a result of environmental error. However, this also highly affects the data because almost never will environmental conditions be so controlled when applying them to the actual environment. For example, a sewerage plant in Germany plays Mozart to facilitate in the metabolism of the anaerobic bacteria that break down waste. While one may make the generalization that the music helps metabolize the bacteria, perhaps the pressure and lighting are actually affecting the bacterial rate of metabolism. Once again, conditions cannot be exactly replicated to grasp a definite understanding of the results.
In correlation, the inability to create exact environments for the natural and human sciences leads to a variability between results. This variation may be due to researcher error or instrumental error or environmental error. While some variation is okay because it is reflective of environmental error, some different results may aid in identifying insufficient methodology. I can say from experience that sometimes rounding numerical data can make a minute difference numerical but may relate to a catastrophic difference when dealing with the natural sciences. This is imperative in chemistry as rounding or neglect of significant figures can have a great effect on a chemical reaction.
In continuum, rounding off also has the same implication as ignoring outliers does.