Ever wonder how to run a simple multivariate experiment on your way to work?

Well, here’s how:

Step 1:  Identify your Control.

The control will be you asking, ‘What’s the time?’ (with a neutral facial expression) to strangers on the way to work.

Step 2:  Identify your Key Performance Indicator and Goal.

Key Performance Indicator will be the percentage of people who give you the correct time (out of the people asked).

The goal will be to increase the KPI by manipulating three variables and finding the most effective combination.

Step 3: Identify your Variables and Values.

The variables manipulated are: Politeness, Urgency, and Facial expression.

The values are listed under each variable below. For example, variable A_Politeness, value 1 puts ‘Excuse me’ in front of the control version, hence it would be ‘Excuse me. What’s the time?’. Variable B_Urgency, Value 1 adds ‘I need to know in under 15 seconds’ after the control making it ‘What’s the time? I need to know in under 15 seconds’. Variable C_Facial Expression just manipulates facial expression. When Value 1 is used, you must put on a big happy smile. When Value 2 is used, make a sad frown. 

Step 4: Make an orthogonal, balanced, multivariable Design Martrix.

We are doing a full factorial design. Here are all the possible combinations of the above variables and values:

Step 5: Implement your Experiment

Over the course of the next few days on your way to work, start asking strangers the above combinations. Record how many people you’ve asked (A) and how many people gave you the correct time (B). Divide B by A and multiply by 100. This will give you the Unique Action Rate of People Answered (Your KPI).

Note: Remember to ask the strangers all the combinations randomly and evenly. In other words, the same amount of strangers should hear Combination #1 as Combination #10. (Don’t skip combinations!). Also, ask enough people for the results to be statistically significant. 

Step 6: Analyze your Results

Perform in-depth multiple regression analysis. Identify the impact for each value, the confidence level and the interval.

See which combination gave you the highest response rate. Check to see if that combination is supported by your analysis. Pick your winner.

Advanced analysis:

Try day-parting and week-parting. See how time of day and day of week impact response. Perhaps one combination works best in the mornings when people aren’t so friendly, but another one is optimal during lunch hour when people tend to be more relaxed.

Try Persona Targeting. See if one combination works better on adults vs children or male vs. female.

The possibilities for testing are endless! Don’t be afraid to let your imagination run wild. You will be surprised to discover what things work and don’t work.