Sections in this category

Understanding & Defining Numerical Rate Metrics

  • Updated

Sisu provides the Numerical Rate Metric type to allow you to calculate one numerical column as a proportion of another column, resulting in a percentage.

 

Related Articles:



Numerical Rate Metrics

A Numerical Rate Metric calculates one numerical column as a proportion of another, resulting in a percentage. For example, customer churn rate is equal to the number of churned customers over the total number of customers.

Use the Numerical Rate Metric type if you are trying to understand a Metric proportionally, based on its share in the denominator. Many use cases – such as marketing Metrics that examine funnels or financial Metrics that look at fraud – heavily focus on rates.

Here are some typical examples:

  • When analyzing churn rate, a sample metric might be the number of churned customers divided by the total number of customers
  • When investigating financial fraud, a sample Metric might be the number of fraudulent transactions divided by the total number of transactions per cohort
  • When measuring user engagement, a sample Metric might be the number of active users divided by the number of total users

Let’s look at an example of a Metric using Numerical Rate to determine the increase in rate of user retention using the sample data below.

Increase rate of user retention =
SUM(Number of Converted Leads) / SUM(Number of Leads)

Campaign ID

Number of Converted Leads

Number of
Leads

Channel

ID00001

200

2000

Facebook

ID00002

150

1000

Google

ID00003

80

600

Google

ID00004

50

500

Facebook

ID00005

40

800

LinkedIn

The overall Metric is calculated as follows:

(200+150+80+50+40) / (2000+1000+600+500+800) = 11%

An example Fact that Sisu might return with this Analysis is a negative impact on numerical rate:

For “channel= Facebook” subgroup,
Metric = (200+50) / (2000 + 500) = 10%

 

Configuring a Numerical Rate Metric

To configure a numerical rate Metric in Sisu:

  1. Create a new Metric, choose your data, and select “Rate” under the “Numerical” subheading of the “Metric calculation type” menu. 
  2. Define the Metric column as the numerator matching the category you wish to target and the denominator column as the total.See the equation under the tool-tip when hovering over the “i” button next to the calculation type menu to see how the Metric is calculated given the columns you select.

Numeric_Ratee_Modal.png

There are a few constraints when using a numerical rate Metric:

  • The Metric’s numerator column must be a subset of the denominator column, and thus be smaller than it in value (0 <= numerator <= denominator). If this constraint is violated, you can reformulate the Metric as a weighted average, where the KPI column is the ratio.
    • Sisu models the numerical rate as a binomial. It does not make sense to have numerator > denominator.
  • The denominator column should be the column with the unit of observations (e.g. transaction count, number of events, etc.) for the results to be statistically significant.

When these conditions are satisfied, the numerical rate is a more precise statistical model than weighted average.

 

Interpreting a Numerical Rate Metric

A numerical rate Metric value will be the total numerator column for the Metric divided by the total denominator column, represented as a percentage. When comparing time periods or groups, Sisu will also represent the percent difference between the two Metric values. Sisu also displays the sum of the numerator and denominator columns.

tip_icon_-_small.png Refer to Understanding Impact for more details.

The fact results under “Top drivers” list the subgroups of highest impact, with those subgroups’ relative weight (prevalence), rate, and impact on rate. For example, the first-order fact “partner = Yes” in the example below can be described as:

Where Partner = Yes,
The rate of Telco Charges out of Number of Charges is 2.2%

numerical_rate_example.png