Why Measurement Scales Matter
The level at which you measure a variable determines what you can do with it statistically. Using the wrong method for the wrong scale produces meaningless — sometimes dangerously wrong — results.
Example: Can you calculate the average blood type? A=1, B=2, AB=3, O=4 → "mean blood type = 2.3" is nonsense. Blood type is nominal — averaging it makes no sense.
Stevens' Four Levels of Measurement
Stanley Smith Stevens (1946) defined four hierarchical levels. Each level includes all the properties of the levels below it.
RATIO ← highest (all arithmetic operations valid)
INTERVAL
ORDINAL
NOMINAL ← lowest (only categorisation)
1. Nominal Scale
Properties: Categories only. No order. No arithmetic.
Examples:
- Blood type: A, B, AB, O
- Department: Finance, Technology, Marketing, HR
- Marital status: Single, Married, Divorced
- Yes/No responses
- Country of origin
What you can do:
- Count frequencies (how many in each category)
- Find the mode (most common category)
- Calculate proportions and percentages
What you cannot do:
- Calculate mean or median
- Say one category is "greater than" another
- Measure the "distance" between categories
Statistical tests: Chi-square test (Chapter 15), mode, frequency tables, bar charts
Survey: 100 employees
Department: Finance=32, Technology=28, Marketing=25, HR=15
Mode = Finance (most common)
Finance proportion = 32/100 = 32%
Cannot say: "Average department = 2.2" ← INVALID
2. Ordinal Scale
Properties: Categories with a meaningful order, but the gaps between categories are not necessarily equal.
Examples:
- Satisfaction rating: 1=Very Dissatisfied, 2=Dissatisfied, 3=Neutral, 4=Satisfied, 5=Very Satisfied
- Education level: School < Undergraduate < Postgraduate < PhD
- Bond rating: AAA > AA > A > BBB > BB > B > CCC
- Military rank: Private < Corporal < Sergeant < Lieutenant < Captain
- Likert scale items
What you can do:
- Everything nominal can do
- Rank observations
- Find median (the middle rank)
- Compare "greater than" / "less than"
What you cannot do:
- Assume equal intervals between categories
- Calculate the mean (technically — though it's commonly done as an approximation)
- Multiply or divide
Key insight: The gap between "Dissatisfied" (2) and "Neutral" (3) is NOT necessarily the same as between "Neutral" (3) and "Satisfied" (4). We just know the order.
Bond Ratings Example:
AAA=1, AA=2, A=3, BBB=4
A bond rated AA (rank 2) is better than A (rank 3)
But "AA is twice as good as BBB" is NOT a valid statement
Calculating "mean rating = 2.4" is technically incorrect
→ Median is the appropriate centre measure
Statistical tests: Spearman correlation (Chapter 17), Mann-Whitney U test, Kruskal-Wallis test, median
3. Interval Scale
Properties: Ordered categories with equal intervals between them, but NO true zero point. Zero is arbitrary.
Examples:
- Temperature in Celsius or Fahrenheit (0°C is not "no temperature")
- Calendar years (Year 0 doesn't mean "no time")
- IQ scores
- pH scale
- Dates (2000, 2010, 2020 — 0 AD is not meaningful as "no year")
What you can do:
- Everything ordinal can do
- Add and subtract (calculate differences)
- Mean and standard deviation are valid
- Calculate intervals: 30°C – 20°C = 10°C difference
What you cannot do:
- Multiply or divide meaningfully (no true zero)
- Say "40°C is twice as hot as 20°C" — this is WRONG
Temperature Example:
July avg: 38°C, January avg: 18°C
Difference: 38 – 18 = 20°C ← VALID (equal intervals)
Ratio: 38/18 = 2.1 ← INVALID ("38°C is 2.1 times as hot as 18°C" is meaningless)
Proof: 38°C = 100.4°F, 18°C = 64.4°F
100.4/64.4 = 1.56 (different ratio! — ratio depends on the scale chosen)
Statistical tests: Pearson correlation, t-tests, ANOVA, mean, standard deviation
4. Ratio Scale
Properties: Ordered, equal intervals, AND a true zero that means "none of the quantity." All arithmetic operations are valid.
Examples:
- Salary: ₹0 means no salary
- Height: 0 cm means no height
- Weight: 0 kg means no weight
- Distance: 0 km means no distance
- Revenue: ₹0 means no revenue
- Age: 0 years means just born
- Number of transactions: 0 means none
- Temperature in Kelvin: 0 K = absolute zero (true absence of heat)
What you can do:
- All operations: add, subtract, multiply, divide
- All statistical methods
- Ratios are meaningful: ₹100,000 salary is twice ₹50,000
Statistical tests: All parametric tests, geometric mean, coefficient of variation
Salary Example:
Priya earns ₹80,000 and Raj earns ₹40,000
Ratio: 80,000 / 40,000 = 2 → Priya earns twice as much ← VALID
The ratio is meaningful because ₹0 = truly no salary
Summary Table
| Property | Nominal | Ordinal | Interval | Ratio |
|---|---|---|---|---|
| Categories | ✓ | ✓ | ✓ | ✓ |
| Meaningful order | ✗ | ✓ | ✓ | ✓ |
| Equal intervals | ✗ | ✗ | ✓ | ✓ |
| True zero | ✗ | ✗ | ✗ | ✓ |
| Mode | ✓ | ✓ | ✓ | ✓ |
| Median | ✗ | ✓ | ✓ | ✓ |
| Mean | ✗ | ✗* | ✓ | ✓ |
| SD / Variance | ✗ | ✗* | ✓ | ✓ |
| Ratios (×, ÷) | ✗ | ✗ | ✗ | ✓ |
*Ordinal means are commonly computed in practice (e.g., survey ratings), but are technically approximate.
Discrete vs Continuous Revisited
Within quantitative (interval and ratio) variables, another important distinction:
Discrete: Only specific values possible (usually integers)
Number of customers: 0, 1, 2, 3, ... (not 2.7 customers)
Number of defects: 0, 1, 2, ...
Credit card transactions: 0, 1, 2, ...
Continuous: Any value in a range (including all decimals)
Salary: ₹78,432.50 (any positive real number)
Time to complete a task: 2.47 minutes
Height: 167.34 cm
Interest rate: 8.75%
The distinction matters for choosing between discrete (Binomial, Poisson) and continuous (Normal, t) probability distributions.
Practical Examples
Example 1: Finance Dataset Classification
| Variable | Scale | Reason |
|---|---|---|
| Stock symbol (INFY, TCS) | Nominal | No order between symbols |
| Analyst rating (Buy/Hold/Sell) | Ordinal | Order exists but unequal gaps |
| Year (2020, 2021, 2022) | Interval | Equal gaps, but Year 0 is arbitrary |
| Share price | Ratio | ₹0 = no price (true zero) |
| % Return | Ratio | 0% = no return (true zero) |
| Credit rating (AAA to D) | Ordinal | Order matters, gaps are not equal |
| Number of shares | Ratio | 0 shares = none |
Example 2: Clinical Trial Dataset
| Variable | Scale |
|---|---|
| Patient ID | Nominal |
| Treatment group (A/B/Placebo) | Nominal |
| Pain level (0–10 scale) | Ordinal |
| Body temperature (°C) | Interval |
| Blood pressure (mmHg) | Ratio |
| Recovery time (days) | Ratio |
| Improved? (Yes/No) | Nominal |
Example 3: HR Survey
Annual engagement survey — variables:
Q1: "I am proud to work here" (1–5) → Ordinal
Q2: Years with the company → Ratio
Q3: Department → Nominal
Q4: Job grade (L1, L2, L3, L4) → Ordinal
Q5: Annual salary → Ratio
Q6: Working hours per week → Ratio
For Q1 (ordinal): Report the median and distribution, not the mean. For Q5 (ratio): Mean, median, and SD are all valid.
Common Mistakes
1. Treating ordinal as interval
Wrong: "Mean satisfaction score = 3.7 out of 5"
(Assumes equal gaps between each point — not guaranteed)
Why it matters: If 4→5 is a bigger improvement than 3→4,
averaging 3s and 5s together gives a distorted picture.
Common practice: Survey researchers often do this anyway as a useful approximation,
but it should be stated as an assumption, not a fact.
2. Computing ratios on interval data
Wrong: "2024 GDP growth was twice 2022 growth"
(Calendar years are interval — the year 0 doesn't mean "no time")
If growth rate is measured as %, that's ratio scale — ratios are valid.
3. Assigning numbers to nominal categories and treating them as numeric
Wrong:
Encode: Finance=1, Marketing=2, HR=3
Calculate: Mean department = 1.8 → "between Finance and Marketing"
→ Completely meaningless
4. Confusing discrete ratio with continuous
Number of support tickets = 0, 1, 2, 3... (discrete ratio)
→ Mean = 2.7 tickets/day is valid as an expected value
→ But you can't have 2.7 tickets in a single day
Practice Exercises
-
Classify each variable and justify your answer: a) Net Promoter Score (−100 to +100, where 0 = neutral) b) Movie rating (1 to 5 stars on Netflix) c) Revenue in ₹ crore d) Quarter (Q1, Q2, Q3, Q4) e) Temperature in Kelvin f) Student rank in class (1st, 2nd, 3rd...)
-
A researcher calculates the "average blood type" of 100 patients by coding A=1, B=2, AB=3, O=4. What is wrong with this approach?
-
For each variable below, identify the most appropriate measure of centre: a) Customer satisfaction (1–5 ordinal scale) b) Employee salaries in a department c) Most common job title in the company
-
Can you say that a pH of 8 is "twice as basic" as a pH of 4? Why or why not?
-
A finance team codes analyst recommendations: Sell=1, Hold=2, Buy=3. They report "average recommendation = 2.1". What assumptions are they making? Is this reasonable?
Summary
In this chapter you learned:
- Nominal: categories only; no order; mode, frequency, bar chart
- Ordinal: ordered categories; unequal gaps; median, rank-based tests
- Interval: ordered, equal intervals, no true zero; mean, SD valid; ratios invalid
- Ratio: all of the above + true zero; all arithmetic and statistical methods valid
- Hierarchy: Ratio > Interval > Ordinal > Nominal (each level inherits lower level properties)
- Discrete: whole numbers only; Continuous: any value in a range
- The scale of measurement determines which statistics are valid — always identify your variable type before analysing
- Ordinal means are technically approximate but widely used in practice
Next up: Data Collection & Sampling Methods — how data gets collected, and how sampling design affects the validity of every conclusion you draw.