Preparing data

Data-processing steps clean and reshape your data before analysis. Each step is added to the chain: it takes the data from the step before it and produces a new version, leaving the original untouched. You can stack as many steps as you need, reorder them, and turn some of them on or off.

How processing steps work

• Data source. Like analyses, each step reads from a source. Auto means “the output of the previous step”, so a chain flows naturally from one step to the next.

• Most steps add a new column and leave the source column in place (for example, Group Values, One-hot encoding). A few replace a column in place (Transform Column) or remove rows (Filter, the outlier steps, Response Quality, Impute ▸ Remove rows).

• Toggleable steps. Filter, the outlier steps, and Response Quality have an enable/disable switch on their card, so you can compare results with and without them without deleting the step.

• Previewing a step shows its output; row-removing steps display removed rows in red.

The pages below describe each step.

• Preprocess
• Transform Column
• Calculate Scale
• Invert Scale
• Formula
• Group Values
• One-hot encoding
• Split Multi-Select
• Filter
• Impute Missing
• Outliers
• Outliers within groups
• ND outliers (multidimensional)
• Response quality
• Select ID
• Bootstrap (correlated resampling)