Problem setup (psychological study context)
The phrase level of representation in statistical psycholgical study is commonly used to mean the measurement level (also called the scale of measurement) of each variable collected in a psychological study. This classification is not cosmetic: it determines which numerical summaries and which graphs are valid.
Consider a study on stress and attention with the following recorded variables:
- Therapy type: CBT, mindfulness, control
- Sleep quality rating: 1–5 Likert item (1 = very poor, 5 = very good)
- Reaction time: milliseconds on a computerized task
- Perceived stress score: a summed questionnaire score (e.g., 0–40)
Step 1: Identify the four measurement levels (the “representation ladder”)
Step 2: Use a decision checklist to classify a variable
For each study variable, answer these questions in order:
- Categories only? If values are names/labels with no ranking, the variable is nominal.
- Order matters? If higher/lower is meaningful but spacing is not guaranteed equal, the variable is ordinal.
- Equal steps? If differences are meaningful and consistent (one-unit change is the same everywhere), the variable is interval.
- True zero? If zero means “none” and ratios make sense (twice as much), the variable is ratio.
A common psychological-research pitfall: a single Likert item (e.g., 1–5 agreement) is strictly ordinal. A multi-item summed scale is often treated as approximately interval in practice, but that assumption should be justified (e.g., many items, roughly symmetric distribution).
Step 3: Match the level of representation to valid summaries and graphs
| Variable (psychology example) | Level of representation | Appropriate descriptive summaries | Appropriate graphs |
|---|---|---|---|
| Therapy type (CBT / mindfulness / control) | Nominal | Frequencies and proportions; mode | Bar chart; pie chart (with caution); contingency table |
| Sleep quality rating (1–5 Likert item) | Ordinal | Median; percentiles; frequency distribution | Bar chart with ordered categories; stacked bar chart |
| Reaction time (ms) | Ratio | Mean; standard deviation; IQR; outlier checks | Histogram; box-and-whisker plot; dotplot |
| Perceived stress score (0–40 summed scale) | Often treated as interval (assumption) | Mean and standard deviation (if treated interval); also median and IQR | Histogram; boxplot; side-by-side boxplots by group |
Step 4: Why the level controls the mathematics (minimal formulas)
Numerical summaries that rely on arithmetic differences require at least interval-level meaning. For example, the sample mean and sample standard deviation use addition and subtraction of values:
\[ \bar{x}=\frac{1}{n}\sum_{i=1}^{n} x_i \qquad\text{and}\qquad s=\sqrt{\frac{1}{n-1}\sum_{i=1}^{n}(x_i-\bar{x})^2}. \]
These computations are meaningful for ratio data (e.g., reaction time) and often for interval data. For nominal categories, expressions like \(x_i-\bar{x}\) have no meaning, so counts and proportions are used instead. For ordinal ratings, the order is meaningful, so medians and percentiles are defensible even when equal spacing is not.
Final classification for the example study
Under the level of representation in statistical psycholgical study framework: therapy type is nominal, a single 1–5 Likert rating is ordinal, reaction time in milliseconds is ratio, and a summed stress score is commonly handled as approximately interval when scale properties support that choice.