Content
Summary
Data objects are entities described by attributes, which are categorized into nominal, binary, ordinal, and numeric types based on their possible values. Understanding these attribute types is essential for selecting appropriate statistical measures and computing similarity between objects. Mathematical operations are only meaningful for numeric attributes, while nominal and ordinal attributes require specific handling for ordering and central tendency.
Intuition
Think of a dataset as a table where rows are data objects (like customers) and columns are attributes (like age or gender). The type of attribute dictates how you can treat its values. Nominal attributes are just labels with no inherent order, like hair color, so you can only count frequencies (mode). Binary attributes are a special case of nominal with two states, which might be symmetric (equal importance) or asymmetric (one state is rarer or more critical). Ordinal attributes have a clear rank, like small/medium/large, but you cannot measure the exact distance between ranks, so you can find the median but not the mean. Only numeric attributes allow standard math like addition or averaging.
Derivation
- Data sets consist of data objects representing entities and attributes representing characteristics.
- Attributes are categorized by their possible values into nominal, binary, ordinal, and numeric types.
- Nominal attributes have no order; thus, mathematical operations are meaningless, and only the mode is valid.
- Binary attributes are nominal with two states; they are symmetric if states are equal, or asymmetric if one is rarer/more important.
- Ordinal attributes have order but unknown magnitude between values; thus, mean is invalid, but mode and median are valid.
Worked example
Consider a dataset of customer satisfaction ratings (Very Poor, Poor, Fair, Good, Excellent) and customer IDs (101, 102, 103). Determine which statistical measures are valid for each attribute.
- Customer IDs are nominal attributes because they are labels with no meaningful order or magnitude.
- For nominal attributes, mathematical operations like subtraction are meaningless, and mean/median are not applicable.
- The only valid measure of central tendency for customer IDs is the mode.
- Satisfaction ratings are ordinal attributes because they have a meaningful order (Poor < Fair < Good) but unknown magnitude between steps.
- For ordinal attributes, the mean cannot be defined because intervals are unknown.
- Valid measures for satisfaction ratings are the mode and the median.
Answer Customer IDs: Mode only. Satisfaction Ratings: Mode and Median.
Common mistakes
- Assuming that numeric values in nominal attributes (like customer IDs) can be used for mathematical operations.
- Calculating the mean or median for nominal attributes, which is not meaningful due to lack of order.
- Treating binary attributes as always symmetric, ignoring cases where one state is more important than the other.
- Calculating the mean for ordinal attributes, which is invalid because the magnitude between ranks is unknown.