Causality

Many business problems are questions of causality:

• How much will sales increase if I increase my advertising budget?
• Will employee productivity increase, if I raise the wages to new employees?
• Will productivity be hurt if I allow employees to work from home?

Anyone who has read this blog knows that I am huge fan of randomized control trials ("experiments") as they get rid of the "selection bias," or "reverse causality."  For example, each of the following factors would bias simple correlations so they do not reflect the implied causality:

• When sales increase, advertising budgets typically increase
• New employees are younger and less experienced than older ones
• Low productivity employees may be more inclined to work from home.  

In this interview, Josh Angrist details some of experiments he ran to figure out that:

• Allowing laptops and iPads in the classroom has a big negative effect of learning.
• No-excuses charter schools have a positive effect.
• Peer effects and giving laptops to kids does not improve learning.

For businesses trying out new advertising campaigns, employment practices, or pricing strategies, design their rollout so you can learn something:  Advertise or change prices in randomly selected areas; adopt employment practices in certain plants but not others.  

If not, you will end up making changes without ever knowing whether they made a difference. 

BOTTOM LINE:  identifying causality is really hard, but profitable.  

Related web app to teach regression (and causality) by showing how Type I (mistakenly inferring causality) and Type II errors (mistakenly inferring no causality) occur.  Do the learning exercises!