Introduction

Understanding scales of measurement is fundamental to proper data analysis. The scale determines what statistical operations are meaningful and which statistical tests can be applied. Using the wrong analysis for a given scale leads to incorrect conclusions.

Stanley Stevens introduced four levels of measurement in 1946: Nominal, Ordinal, Interval, and Ratio (often remembered as NOIR). Each successive level has all the properties of the previous levels plus additional properties.


Nominal Scale

The simplest scale—used for categorization only.

Properties

  • Categories are mutually exclusive
  • No inherent order or ranking
  • Numbers are labels only (arbitrary)

Examples

  • Gender (Male/Female)
  • Colors (Red/Blue/Green)
  • Country of origin
  • Product categories
  • Jersey numbers in sports

Permissible Statistics

  • Frequency counts
  • Mode
  • Chi-square tests
Cannot do: Mean, median, add, subtract, or compare magnitudes. "Player 23 is not greater than Player 7."

Ordinal Scale

Categories with a meaningful order, but unequal intervals.

Properties

  • Categories have rank order
  • Intervals between ranks are not equal
  • Can say "greater than" but not "how much greater"

Examples

  • Education level (High School < Bachelor's < Master's < PhD)
  • Satisfaction ratings (Very Dissatisfied < Dissatisfied < Neutral < Satisfied < Very Satisfied)
  • Class rank (1st, 2nd, 3rd...)
  • Likert scales (often treated as ordinal)

Permissible Statistics

  • Frequency, mode
  • Median, percentiles
  • Non-parametric tests (Mann-Whitney, Spearman correlation)
Cannot do: Mean is technically inappropriate (though commonly used for Likert scales). The difference between 1st and 2nd place may not equal difference between 2nd and 3rd.

Interval Scale

Ordered categories with equal intervals, but no true zero.

Properties

  • Meaningful equal intervals
  • No true zero point (zero is arbitrary)
  • Can add and subtract meaningfully
  • Cannot form meaningful ratios

Examples

  • Temperature (Celsius/Fahrenheit): 0°C doesn't mean "no temperature"
  • Calendar years: Year 0 is arbitrary
  • IQ scores: 0 doesn't mean "no intelligence"
  • SAT/GRE scores

Permissible Statistics

  • All ordinal statistics
  • Mean, standard deviation
  • Parametric tests (t-tests, ANOVA, Pearson correlation)
Cannot do: Ratios. 20°C is not "twice as hot" as 10°C.

Ratio Scale

The highest level—has all properties including a true zero.

Properties

  • Equal intervals
  • True zero point (zero means absence)
  • Can form meaningful ratios
  • All mathematical operations permitted

Examples

  • Height, weight, distance
  • Money, sales revenue
  • Age (from birth)
  • Temperature in Kelvin
  • Units sold, time duration

Permissible Statistics

  • All statistics
  • Geometric mean, coefficient of variation
  • Meaningful ratio statements

Example

₹200 is twice as much as ₹100 (meaningful ratio). Zero rupees means no money (true zero).


Comparison of Scales

PropertyNominalOrdinalIntervalRatio
Categories
Order
Equal Intervals
True Zero
Central TendencyModeMedianMeanMean

Conclusion

Key Takeaways

  • Four scales: Nominal, Ordinal, Interval, Ratio (NOIR)
  • Nominal: Categories only, no order (use mode)
  • Ordinal: Ordered categories, unequal intervals (use median)
  • Interval: Equal intervals, no true zero (use mean)
  • Ratio: Equal intervals plus true zero (all statistics valid)
  • Scale determines appropriate statistical tests
  • Higher scales have all properties of lower scales