Tipe Data

01Conceptual Foundation

Understanding data types is foundational to selecting the correct statistical test. Every exam question about choosing a test, interpreting a p-value, or critiquing a study design ultimately traces back to one question: what type of data is being analyzed?

Data can be broadly divided into two families. Categorical (qualitative) data places observations into groups. Numerical (quantitative) data assigns a measurable value to each observation. Within each family, there is a hierarchy of measurement precision that determines which statistical operations are valid.

The hierarchy follows a simple rule: each higher level of measurement inherits all the properties of the levels below it and adds one new property. Nominal data has categories. Ordinal data adds order. Interval data adds equal spacing between values. Ratio data adds a meaningful zero point. This layered structure is the single most important concept to internalize, because it dictates which summary statistics and hypothesis tests are appropriate.

A common error is treating ordinal data as if it were interval data. For example, a pain scale from 1 to 10 has an order, but the "distance" between a score of 2 and 4 is not necessarily the same as between 6 and 8. Calculating a mean of ordinal data is technically invalid, though it is done frequently in clinical practice and some research contexts. On a licensing exam, however, you must know the theoretically correct classification.

02Classification and Properties of Each Data Type

Categorical Data

Nominal Data

Nominal data is the most basic level of measurement. Observations are sorted into mutually exclusive categories that carry no rank or quantitative value. The word "nominal" comes from the Latin nomen (name), which is a useful mnemonic: you are simply naming categories.

Medical examples include blood type (A, B, AB, O), sex (male, female), marital status, alive vs. dead, presence or absence of a gene mutation, and race/ethnicity categories. Even when numbers are assigned to categories (e.g., 1 = male, 2 = female), those numbers are arbitrary labels. You cannot say that "female is twice male."

A special subtype is dichotomous (binary) nominal data, which has exactly two categories: yes/no, positive/negative, diseased/healthy. This subtype is extremely common in clinical research and forms the basis of 2x2 contingency tables used to calculate sensitivity, specificity, and odds ratios.

Ordinal Data

Ordinal data adds a meaningful rank order to categories, but the intervals between ranks are not guaranteed to be equal. Think of it as nominal data with a hierarchy.

High-yield medical examples:

• Cancer staging (Stage I < II < III < IV)

• NYHA heart failure class (I through IV)

• Pain severity scales (mild, moderate, severe)

• Likert scales in survey research (strongly disagree to strongly agree)

• Apgar scores (though often treated as interval in practice)

• ECOG performance status (0 through 5)

• ASA physical status classification (I through VI)

The critical point is that the gap between Stage I and Stage II cancer is not the same as the gap between Stage III and Stage IV. Because the intervals are unequal, calculating a mean is not technically appropriate. The correct measure of central tendency for ordinal data is the median, and the correct tests are nonparametric tests such as Mann-Whitney U or Kruskal-Wallis.

Numerical Data

Interval Data

Interval data has both order and equal spacing between values, but it lacks a true zero point. The zero is arbitrary and does not represent the complete absence of the quantity being measured.

The classic example is temperature in Celsius or Fahrenheit. The difference between 20°C and 30°C is the same as between 30°C and 40°C (equal intervals). However, 0°C does not mean "no temperature." Consequently, you cannot say that 40°C is "twice as hot" as 20°C. Ratios are meaningless with interval data.

Other medical examples include calendar year (2020 vs. 2025), pH (the zero is arbitrary), and many standardized psychometric test scores (e.g., IQ scores, where a score of 0 does not mean "no intelligence" and a score of 140 is not "twice as intelligent" as 70).

You can perform addition and subtraction with interval data (the difference between two values is meaningful), and parametric tests such as the t-test and ANOVA are valid.

Ratio Data

Ratio data is the highest level of measurement. It possesses all properties of interval data and adds a true (absolute) zero, which represents the genuine absence of the measured quantity. This allows meaningful ratios: a hemoglobin of 14 g/dL is indeed twice as much as 7 g/dL.

Medical examples are abundant:

• Height, weight, BMI

• Blood pressure (mmHg)

• Serum glucose, creatinine, hemoglobin, and virtually all lab values

• Heart rate, respiratory rate

• Temperature in Kelvin (the only temperature scale with a true zero)

• Duration of disease, length of hospital stay

• Number of episodes, number of cells per field

Count data (e.g., number of lymph nodes positive) is a subtype of ratio data that is discrete rather than continuous but still has a true zero and allows ratios.

Because ratio data supports all mathematical operations, the full suite of parametric statistics applies, including the coefficient of variation (which requires a meaningful zero to interpret).

03Choosing the Correct Statistical Test Based on Data Type

This is where exam questions most frequently test your understanding. The data type of both the independent variable and the dependent variable together determine the appropriate test.

The decision-making logic follows two steps:

Step 1: Identify the data type of your outcome (dependent) variable. If it is nominal or ordinal, you use nonparametric tests. If it is interval or ratio and normally distributed, you use parametric tests.

Step 2: Count the number of groups being compared. Two groups lead to a t-test or Mann-Whitney; three or more groups lead to ANOVA or Kruskal-Wallis.

An additional consideration is whether the groups are independent (unpaired) or dependent (paired/matched). Paired data arises when the same subjects are measured before and after an intervention, or when subjects are matched one-to-one. Paired designs use different tests (paired t-test, Wilcoxon signed-rank) than unpaired designs.

04Transformations and Practical Nuances

Dichotomization is a common practice in clinical research where a continuous variable is split into two groups using a cutoff. For example, converting systolic blood pressure (ratio data) into "hypertensive" vs. "normotensive" (nominal data). This is valid and often clinically useful, but it reduces statistical power because it discards information about the degree of difference between subjects.

On exams, when a vignette presents a continuous outcome that has been categorized, recognize that the data type has changed, and the appropriate test changes with it.

05Differential Diagnosis of Data Types

06Traps and High-Yield Pearls

The most common trap involves ordinal data disguised as numbers. A vignette might describe a researcher comparing "average pain scores" between two groups using a t-test. The correct criticism is that a pain scale is ordinal, so a Mann-Whitney U test would be the theoretically appropriate test. Students who focus only on the number of groups (two groups = t-test) without checking the data type will select the wrong answer.

A second frequent trap is the interval vs. ratio distinction on temperature scales. Celsius and Fahrenheit are interval because 0 degrees is arbitrary. Kelvin is ratio because 0 K is absolute zero. This distinction rarely changes clinical practice, but it is a favorite conceptual question.

A third trap involves dichotomized data. When a study converts a continuous variable (e.g., hemoglobin level) into a binary outcome (anemic vs. not anemic), the data type becomes nominal. Students who remember the original variable was continuous and select a t-test will be wrong; the correct test is chi-square.

The core competency being tested in questions about data types is matching the level of measurement to the correct statistical test and summary statistic. If you can correctly identify the data type, the test selection follows logically. The hierarchy to memorize is:

Nominal (names) → Ordinal (order) → Interval (equal intervals, no true zero) → Ratio (equal intervals, true zero)

Each level inherits everything below it and adds one property. Know the property, know the test.