Statistics I — Variables & Measurement Scales

Statistics I — Variables & Measurement Scales

Before analysing data we need to understand exactly what we measured. In statistics, every characteristic we measure is called a variable, and each variable has a type and a measurement scale that determine which operations and calculations are permitted. On this page we will learn to distinguish among the four measurement scales, between discrete and continuous variables, and between dependent, independent variables and causal relationships.

Background and Basic Definitions

A variable is a characteristic that can take different values across individuals — for example height, gender, score, or eye colour. Something that does not vary (the same for everyone) is called a constant.

The four measurement scales, from simplest to most sophisticated:

  • Nominal — categories with no inherent order: gender, country of birth, colour. Only equality/difference can be assessed; frequencies may be counted.
  • Ordinal — categories ranked in order, but the gaps between ranks are not equal: satisfaction rating (low–medium–high), competition placement.
  • Interval — ordered with equal gaps, but the zero point is arbitrary: temperature in Celsius, calendar year. Ratios are meaningless (\(20^\circ\) is not 'twice as hot' as \(10^\circ\)).
  • Ratio — like interval but with a true absolute zero: height, weight, time, income. Ratios are meaningful (\(8\) kg is twice \(4\) kg).

Discrete vs. continuous: a discrete variable takes separate, countable values (number of children, number of cars), while a continuous variable can take any value within a range and is measured (height, time).

Variable roles in research: the independent variable is the factor the researcher manipulates or examines (the 'explanatory' variable), and the dependent variable is the outcome that is measured. A causal relationship means that a change in one variable causes a change in the other — unlike a mere correlation, which may arise from a third confounding variable.

Solution Steps

  1. Step 1 — Ask what exactly the variable measures: a category (qualitative) or a quantity (quantitative)?
  2. Step 2 — If qualitative: check whether the categories have a meaningful order. No order → nominal; order exists → ordinal.
  3. Step 3 — If quantitative: check whether zero means 'none of it'. No true zero → interval; true zero → ratio.
  4. Step 4 — For discrete/continuous classification ask: are values counted separately (discrete) or measured along a continuum (continuous)?
  5. Step 5 — For role identification: the 'cause' being manipulated is the independent variable; the 'effect' being measured is the dependent variable.
  6. Step 6 — Before claiming causality, verify there is no confounding variable that explains the relationship.

Worked Examples

Example 1: Identifying the Measurement Scale

Problem: A researcher recorded for each participant: (a) phone number, (b) judo belt rank (white, yellow, black), (c) body temperature in Celsius, (d) weight in kg. Classify each according to its measurement scale.

Solution:

  1. (a) A phone number is an identifying label with no order or quantity — nominal scale.
  2. (b) Belt ranks have a clear order but the gaps between ranks are not numerically equal — ordinal scale.
  3. (c) Temperature in Celsius: equal gaps exist but zero is arbitrary (\(0^\circ\) does not mean 'no heat') — interval scale.
  4. (d) Weight: there is a true absolute zero (\(0\) kg = no mass) and ratios are meaningful — ratio scale.

Answer: Nominal, ordinal, interval, ratio — respectively.

Example 2: Discrete or Continuous

Problem: Classify each variable as discrete or continuous: (a) number of text messages a person sends per day, (b) running time for 100 metres in seconds, (c) number of passengers on a bus.

Solution:

  1. (a) Number of messages is a whole number that is counted; '3.5 messages' is impossible — discrete.
  2. (b) Running time can be measured to any level of precision (\(11.43\) s, \(11.431\) s) — continuous.
  3. (c) Number of passengers is counted and always a whole number — discrete.

Answer: Discrete, continuous, discrete — respectively.

Example 3: Dependent and Independent Variables

Problem: A study investigates whether the number of hours of sleep per night affects the next-day exam score. What is the independent variable and what is the dependent variable?

Solution:

  1. We ask: which is the factor suspected of having an effect, and which is the measured outcome?
  2. Hours of sleep is the influencing factor — it is the independent variable.
  3. The exam score is the outcome measured as a result of sleep — it is the dependent variable.
  4. Memory tip: the dependent variable 'depends' on the independent variable, just as the score depends on hours of sleep.

Answer: Independent variable: hours of sleep. Dependent variable: exam score.

Example 4: Causal Relationship or Confounding Variable

Problem: It was found that children with larger shoe sizes read better. Does shoe size cause better reading? What is the likely explanation?

Solution:

  1. There appears to be a positive correlation between shoe size and reading ability, but correlation does not imply causation.
  2. We look for a confounding variable that explains both variables simultaneously.
  3. Age is the confounding variable: older children have larger feet and also read better.
  4. Conclusion: this is a spurious correlation (correlation only), not a causal relationship between shoe size and reading.

Answer: Not a causal relationship; age is the confounding variable that explains the correlation.

Example 5: Permitted Operations by Scale

Problem: A student calculated the 'average' eye colour by assigning numbers (blue=1, green=2, brown=3) and obtained 2.1. Is the calculation valid?

Solution:

  1. Eye colour is a variable on a nominal scale — the numbers are merely identifying labels.
  2. On a nominal scale the only valid operation is counting frequencies (and identifying the mode); arithmetic is not permitted.
  3. Computing a mean requires at least an interval scale, where the gaps between values are equal and meaningful.
  4. Therefore the value \(2.1\) is meaningless — there is no such thing as an 'average colour'.

Answer: The calculation is invalid; computing a mean on a nominal scale is not permitted.

Common Mistakes

✗ Common mistake: Inferring a causal relationship from the mere existence of a correlation ('X is related to Y, therefore X causes Y').

✓ The correct way: Correlation is not causation. To claim causality, confounding variables must be ruled out — ideally through a controlled experiment in which only the independent variable is manipulated.

✗ Common mistake: Confusing interval and ratio scales and computing ratios for Celsius temperatures ('\(30^\circ\) is three times hotter than \(10^\circ\)').

✓ The correct way: Ratios are only valid on a ratio scale that has a true absolute zero. On an interval scale (Celsius, calendar year) the zero is arbitrary, so only differences are meaningful — not ratios.

✗ Common mistake: Classifying 'number of children' as continuous because it involves a number, or 'height' as discrete because it was recorded as a whole number.

✓ The correct way: The question is whether values are counted (discrete) or measured along a continuum (continuous), not whether a number appears. Number of children is counted → discrete; height is measured along a continuum → continuous, even when rounded.

Practice Tips

  • Tip — The scales are ordered from weakest to strongest: nominal \(\to\) ordinal \(\to\) interval \(\to\) ratio. Each higher scale includes all the capabilities of the lower ones.
  • Tip — You can always convert from a higher scale to a lower one (e.g. rank heights as 'short/medium/tall'), but not the reverse — downgrading loses information.
  • Tip — To distinguish discrete from continuous, ask 'Can there be an in-between value?' \(2.5\) siblings is impossible (discrete); \(2.5\) kg is possible (continuous).
  • Tip — In an experiment the researcher controls the independent variable and measures the dependent one; a moderator variable changes the strength of the relationship between them.

Summary and Key Formulas

Measurement scales:

ScaleFeatureExample
NominalCategories, no orderGender, colour
OrdinalOrder, unequal gapsRanking, placement
IntervalEqual gaps, arbitrary zeroCelsius
RatioTrue zero, meaningful ratioWeight, time
  • Discrete = counted; continuous = measured.
  • Independent = the cause; dependent = the effect.
  • Correlation \(\ne\) causation — beware of confounding variables.