- Typically, it does not make sense to analyze the raw data.
- raw data may not be at the unit of analysis of other relevant variables (e.g., you collected data at the SKU-level, while your dependent variable is at the brand level),
- raw data may not yet contain the independent variables you are actually interested in, and you need to prepare for merging,
- raw data may be too messy (e.g., contains outliers), and
- raw data may be too fine grained (e.g., recorded per minute, while a weekly level of analysis may be warranted).
Therefore, you need to transform your raw data to the final (cleaned) data set.
How to transform your raw data to the final (cleaned) data set?
- E.g., refine your sample (e.g., “shops that sell at least 1 item per week”; “brands that are in the Top 3 for at least three con secutive years”,etc.
- E.g., define rules how to deal with missing values and outliers
- E.g., aggregate from minutes to weeks; aggregate from sales per user per shop, to sales per shop (“same primary key”)
- E.g., merge different sources (which have previously been aggregated to the same primary key); e.g., temperature data (record ed per day),to your sales data set for swimming equipment ;) (per shop, and day)
Operationalize your variables
- Which are the variables you want to use for your analyze, and how do you operationalize them? (e.g., think of raw data that stores the names of products sold in a given month -> you could convert this to a measure of how many products are sold in a given month, simply by counting them. So from your raw data, you get to a real variable that you can use (e.g., note the previous variable was text, an d now you have a count, e.g., number of SKUs)
- Typically, you provide a table with
- Look at the literature and how previous researchers have defined variables that you are looking for.