01 03 01 Clasificación de los datos

Main ideas and lessons

• Purpose of the section: Before doing any data analysis, you must perform a preliminary data classification to understand what the data actually is and what you can (or can’t) do with it.

• Why classification matters: Without it, you may:

  • apply the wrong statistical tools to the wrong data type,
  • waste time handling irrelevant data,
  • or produce misleading results.

• Core concept: data taxonomy (categorization system):

  • A taxonomy helps you classify data from multiple perspectives so you can anticipate:
    • usefulness,
    • processing capabilities,
    • and how to use it in analysis.

Methodology / classification framework (detailed bullet format)

Step 1: Determine the data type

• Quantitative data (numerical, measurable/countable)

  • Continuous: precision increases when dividing (e.g., weights, distances)
  • Discrete: division does not increase precision (e.g., number of people; “3.4 people” doesn’t make sense)
  • Interval scale: no absolute zero (e.g., Celsius temperature)
    • Example implication: operators are limited (equality works; “greater/less meaningfulness” depends on lack of absolute zero context)
  • Ratio scale: absolute zero exists (e.g., money)
    • Zero truly means absence of magnitude (e.g., 0 money = no money)

• Qualitative data (alphanumeric/label-based, not measured numerically)

  • Nominal: categories with no order (e.g., colors)
  • Ordinal: categories with inherent order (e.g., Monday–Saturday; hierarchy like manager/boss)

• Additional taxonomy views

  • Numeric data: can perform arithmetic; includes integers/decimals/floats
  • Alphanumeric data: labels like “true/false” or “good” (note: not the same as dichotomous “yes/no”)
  • Temporal data: locates specific dates/times

Step 2: Determine the statistical variables and what you’ll do with the data

Choose based on analytic intent/use, including these categories of “use”:

• Identification data (unique)

  • Examples: employee ID, student ID, product code
  • Each record/observation must have one; values are unique.

• Categorical data (grouping/segmenting labels)

  • Examples: country, department, product type/brand, address, management
  • Essential for analytics reporting (e.g., segment by country/state/department).
  • Missing values are acceptable.
  • Categories are usually mutually exclusive, but sometimes you need an “undefined” category to capture items that don’t fit.
  • Categories must be predefined (a fixed catalog); you shouldn’t invent categories “on the fly” and let them drift.
  • Types of categorical representations:
    • Numerical categoricals: numbers represent categories (e.g., 1 → Red), requiring translation.
    • Coded categoricals: abbreviations/codes represent categories (e.g., NL → Nuevo León, TX → Texas), requiring translation.
    • Descriptive categoricals: full labels already represent categories (no translation needed).
    • Interval categoricals: a numeric scale is divided into ranges/categories (interval classes; often denoted with k).
      • Range notation meaning:
        • Square brackets [ ]: endpoint included
        • Parentheses ( ): endpoint excluded
      • Consequence example: if a value equals a boundary (e.g., “3”), it belongs to the interval where that boundary is included.

• Descriptive data (labels, not categorical grouping catalog)

  • Examples: product name, person name
  • Not intended to appear multiple times as the same “catalog category” for grouping.

• Temporal data (time scales)

• Value data (numeric for calculations)

  • Used for arithmetic and aggregation like averages, sums, standard deviations, etc.

Step 3: Distinguish detailed vs aggregated value data

• Detailed value data: the value for each individual observation

• Aggregated value data: consolidation across observations (summations, averages, standard deviations)

• Emphasis: analysis must distinguish per-observation values from aggregated values.

Step 4: Determine the origin (availability scenario) of the data using taxonomy

Data can be:

• Available: already exists in some system (you can obtain it directly)
• Calculated: not directly existing; derived from available sources (combining data to produce new conclusions)

• Questioned: missing and cannot be calculated; must be obtained by questioning/collecting externally

• Lesson: small compact datasets can still enable powerful analysis if you can calculate new metrics from them (example referenced: early COVID situation where very few fields enabled extensive analysis).

Speakers / sources featured

• No specific named speakers or external sources are mentioned.