According to J. Leek, there are six data analysis questions we can ask. Let us look at these questions in detail. As an example, I will use a website of a marketing agency. The conversion goal of this website is the contact form that users fill out and send. We want to ask questions about the conversion.
Descriptive analysis: What is..?
Here the goal is to describe a set of data, without inferring anything from it or making a prediction. It can be performed by itself or be a starting point of a more in-depth analysis. An example in case of an agency would be the number of forms sent or the conversion rate (e.g. the number of submissions divided by the number of page visits).
Explorative analysis: Where is..?
This analysis looks at the data more deeply, discovering the connections between different variables. However, this cannot be used for prediction or does not necessarily imply causation. In the example above, exploratory analysis can be used to look for connection between form submissions and day of the week or where the form was placed on a page.
Inferential analysis: Who is..?
In this type, we infer about a larger group of users from a small group. E.g. we can set up a user survey or conduct focus group research with some users to find out how they interact with the site and the contact form in particular.
Predictive analysis: What will be..?
This analysis makes prediction about future occurrences based on some known variables. For example, we could predict the fluctuation in the number of form submissions according to the day of the week. Prediction should not be confused with causality, e.g. if there is a peak of form submissions on Monday, it does not mean that Monday causes form submissions.
Causal analysis: Why is…?
This is used to define one-to-one relationships between different variables. Causality will almost always look at the average cases, so some outliers or different behaviors should not be excluded. As an example, placing the form at the top of the page will usually increase the conversion, as opposed to placing it below-the-fold. Here, a better form placement actually causes more visibility and increases the chance of a submission.
Mechanistic analysis: How is..?
This a rare and difficult type of analysis that allows for understanding how exactly variables influence each other in individual objects. This would often bring the analysis of data down to the set of equations.
All in all, you will need to perform different types of analysis of the raw data to get a clear picture and come up with a set of recommendations.