Data Visualization
Now we move on to causal data collection which is more stringent.
Example: Landing pages of a website. We want to determine if landing page A is better than landing page B?
How do we determine that? We can do a field experiment. Some set of customers see landing page A. Another set of customers see landing page B. We look at the click through rates, and then decide which landing page is better.
Here, we are trying to make causal link between changing the landing page, and looking at the click through rate. This is what causality is.
Example: A/B testing – need to distinguish between correlation and causation.
Three requirements for causal inference:
The key idea to distinguish between these two is to start doing field experiments. You can start systematically manipulating prices/landing pages to see how they might have a causal impact.
For example: Landing page A versus landing page B, how do you distinguish which page is better?
Companies: Optimizely, LeanPlum, MixPanel
Managerial Questions:
Example: Landing pages of a website. We want to determine if landing page A is better than landing page B?
How do we determine that? We can do a field experiment. Some set of customers see landing page A. Another set of customers see landing page B. We look at the click through rates, and then decide which landing page is better.
Here, we are trying to make causal link between changing the landing page, and looking at the click through rate. This is what causality is.
Example: A/B testing – need to distinguish between correlation and causation.
- Correlation is the relationship between two variables. So let's take price and sales. Are sales and prices correlated? Mostly. If prices go down, sales might drop as well.
- Causation is one variable producing an effect in the other.
- Correlation and causation are not the same thing.
Three requirements for causal inference:
- Correlation: Evidence of association between X and Y
- Temporal antecedence: X must occur before Y
- No third factor driving both: Control of other possible factors
The key idea to distinguish between these two is to start doing field experiments. You can start systematically manipulating prices/landing pages to see how they might have a causal impact.
For example: Landing page A versus landing page B, how do you distinguish which page is better?
- One set of customers coming to your website, will see landing page A.
- Another set of customers, randomly chosen, will see another landing page, landing page B.
- Then over time, you see the click through rate.
- You see the purchases.
- That can help you determine which landing page is better.
Companies: Optimizely, LeanPlum, MixPanel
Managerial Questions:
- Website optimization à What kind of websites? What landing pages, what icons should be shown? How should they be optimized?
- Mobile app design à How should they be best designed?
- Customization à You can start thinking about how should you design the app in general, but you can also start thinking about what version to the app should be shown to different customers. The extreme case would be one to one marketing.
