Question

num_people_v2 consists of the folllowing categorical variables:

less than 30 million

between 30 to 60 million

more than 60 million

Security controls are interval variables.

In the following questions can you explain how we know how many dummy variables would we need to create, how we know this, and which subcategory would we use as the default comparison group, and how did we find that answer? 

How many dummy variables would you need to create if you were examining the effect of number of people impacted ("num_people_v2") on the amount of money spent on security controls?

Answer 1

We have to calculate the effect of number of people impacted on the amount of money spent on security controls.

We can calculate this effect in various ways -

1. Calculating correlation coefficient between both variables (as both are continuous)

2. Binning one variable (say number of people impacted) into groups. Binning creates equal number of observations in group which might not be suited for some use cases. For example - any continuous variable might have its value clustered in either extremes - more number of small values or large values, which will make intervals very small for distinct groups and hence, it will be difficult to infer which interval really caused the effect.

3. Binning variable based on subject knowledge -

This is an efficient way of creating different groups of continuous variable. For example - economists might best know what is the interval to group salaried professionals into low, medium and high earning groups. The subject matter experts can help in telling two important things - the number of groups and the interval of the groups. Both of these knowledge combined can help in study the effect of the predictor variable on the response. Also, if someone is performing the study he would know what he is seeking and hence grouping based on subject knowledge.

This example is the third case where subject matter expert has decided to check the effect of three distinct groups (below 30 mn, 30 to 60 mn, more than 60mn) decided by himself/another expert on the amount of money spent on security controls.