Z-Score Normalization

Once the underlying values (ratios or indices) for each indicator are calculated and, if needed, adjusted (see our capping and damping methods), the next thing to do is to rescale each indicator’s value into a common range from 1 to 100, so that they are compatible with each other. This process is widely known as data normalization.

• Firstly, the widely used z-score normalization (or standardization) is applied. A z-value shows the number of standard deviations a given data point lies above (positive z-value), below (negative z-value) or exactly on the mean (zero z-value)*.
• Once z-scores are calculated, their position on the normal curve is plotted resulting in the scaled value from 0 to 1 for each indicator, showing the probability that an indicator value of a random institution from the population is less than or equal to the given value. For example, if an institution has Academic Reputation at this step of calculations equal to 0.9, this will indicate that it performs in the top 10% of institutions by this indicator.
• The resulting scores are finally linearly scaled between 1 and 100 for each indicator.

After each indicator’s data is compatible with each other, we can combine the data reliably and apply weightings fairly to the calculation of the overall score. The final overall score is scaled again, where 100 goes to the maximum overall score, and proportionally less to all others.