What is a Decile?

Deciles are a data visualization technique that divides a data series into 10 equal parts. For example, regardless of the series’ size, it can be subdivided into 10 decile groups, usually labeled as D1, D2, D3, etc… This can then be graphed and used to perform a decile analysis.

How are Decile’s used?

The point of this grouping is to test the outputs of a predictive model to see how effective it is in predicting the expected outcome. Generally, the deciles are displayed as a bar graph. Each column in the chart represents a collection of records that have been scored with the predictive model. The height of each column represents the average of the actual behaviors of those records. Simply glancing at the height of the columns allows the researcher to make a quick judgment of the model’s viability:
  • A highly accurate model would display a “staircase” effect from one side to the other when viewing this graph, with each decile either smaller or larger than the last. Similar to a linear graph.
  • A low accuracy model would display many different heights in the bars, with no obvious linear direction of growth or decline.
  • A poor accuracy model would show each bar as nearly identical and therefore it’s accuracy is no better than random grouping.