Difficulty: Introductory | Prerequisites: None. This is first-chapter material.
This material sits at the very start of an introductory statistics course: the vocabulary you need before any data analysis begins. The concepts here (experimental units, variables, data types, populations, samples, and bias) recur in every chapter that follows. If you cannot correctly identify what was measured, on whom, and whether the data represent a sample or a population, you will misapply every technique built on top of those distinctions. The three case studies below give you practice classifying real-world research scenarios using this foundational language.
Every statistical study starts by identifying what you are studying (experimental units), what you are measuring (variables), and whether your data cover everyone (population) or just a subset (sample). Variables split into qualitative (categories) and quantitative (numbers). Getting these labels right is the entry ticket to the rest of the course.
Experimental unit (observational unit)
The individual person, object, or entity on which data are collected in a study.
In simple terms, this is the "who" or "what" being studied. Each row in your data table is typically one experimental unit.
Variable
A characteristic or property measured or recorded for each experimental unit.
Think of it as a column heading in a data table: anything you track about each unit (price, rating, condition) is a variable.
Qualitative variable (categorical variable)
A variable whose values are labels or categories rather than numbers. Examples: credit rating (A, BBB, BB), bridge condition (safe, deficient, functionally obsolete), tenant name.
In simple terms, you cannot do arithmetic on these values. Averaging "BBB" and "A" makes no sense.
Quantitative variable (numerical variable)
A variable whose values are numbers on which arithmetic operations are meaningful. Examples: capitalisation rate (5.00%, 6.70%), engagement ring price ($2,500).
Think of it as anything you can sensibly add, subtract, or average.
Ordinal variable
A quantitative-looking variable whose numeric values represent ordered categories rather than true measurements. Example: a 1 to 7 appreciation scale. The numbers indicate rank order, but the intervals between them are not necessarily equal.
In simple terms, a "6" is higher than a "3," but the gap between them is not provably twice the gap between 3 and 1.
Population
The complete set of all units you want to draw conclusions about.
Think of it as "everyone" or "everything" relevant to your research question, with no one left out.
Sample
A subset of the population that is actually observed or measured.
In simple terms, the group you have data on, which is usually smaller than the full population.
Representative sample
A sample that accurately reflects the characteristics of the population from which it was drawn. If a sample is representative, conclusions drawn from it can be generalised to the population.
Bias (sampling bias)
A systematic tendency for the sample to differ from the population in some way. Common sources include small sample size, voluntary participation, and exclusion criteria that filter out certain groups.
Think of it as a built-in tilt that makes your sample untrustworthy as a stand-in for the population.
Capitalisation rate (cap rate)
In property valuation, the ratio of a property's net operating income to its current market value, expressed as a percentage. Higher cap rates imply higher risk or shorter remaining lease terms.
In simple terms, it is the expected annual return on a property investment before financing costs.
Tags: experimental unit, observational unit, variable, qualitative, categorical, quantitative, numerical, ordinal, population, sample, representative sample, bias, sampling bias, cap rate, capitalisation rate
A property-investment firm (the Boulder Group) tracks capitalisation rates for 13 retail tenants across four lease lengths. This problem asks you to pick apart the study design.
Experimental units: The 13 retail tenants (Best Buy, Lowe's, etc.). Each tenant is one row in the dataset.
Variables identified:
Tenant name: qualitative
S&P credit rating (BB, BBB, A, etc.): qualitative
Capitalisation rates for 5-, 10-, 15-, and 20-year leases: quantitative
Sample vs. population: The 13 tenants are a sample drawn from the larger population of all single-tenant retail properties the Boulder Group owns.
Defining the population: If the sample is representative, the population is all tenants of single-tenant retail properties owned by the Boulder Group, regardless of credit rating or lease term.
Key observation on the data: Capitalisation rates tend to increase as the remaining lease term decreases. Shorter leases carry more uncertainty for the investor, so the market demands a higher return.
Example: Lowe's cap rates are 5.00% (20-year), 5.45% (15-year), 6.10% (10-year), 6.70% (5-year).
Real-world application: This inverse relationship between lease length and cap rate is central to commercial property valuation. Investors use it to price risk: a shorter lease means the tenant might leave sooner, so the property must offer a higher yield to compensate.
Tags: capitalisation rate, cap rate, single-tenant property, S&P credit rating, qualitative variable, quantitative variable, lease term, Boulder Group, sample, population
The National Bridge Inventory (NBI) catalogues the condition of every highway bridge in the United States. This problem tests whether you can identify a population versus a sample and classify a variable.
Variable of interest: The condition or status of each highway bridge, rated as safe, functionally obsolete, or deficient.
Data type: Qualitative (categorical). The ratings are labels, not numbers. You cannot average "safe" and "deficient."
Population or sample? This is a population. The dataset covers all 608,272 highway bridges in the country, so no inference to a larger group is needed.
Data collection method: Inspections conducted under the guidance of the Federal Highway Administration (FHWA).
Real-world application: Governments use exactly this kind of categorical classification to prioritise infrastructure spending. A bridge rated "deficient" moves up the repair queue. The fact that this is a population (not a sample) means the reported percentages are exact, not estimates.
Tags: highway bridges, NBI, National Bridge Inventory, FHWA, qualitative variable, categorical variable, population, census data, structurally deficient, functionally obsolete
Researchers surveyed 33 engaged couples through a popular wedding website to see whether the price of an engagement ring predicts the recipient's level of appreciation. This problem is rich ground for discussing bias and representativeness.
Experimental units: The 33 respondents who completed the online survey.
Variables identified:
Engagement ring price: quantitative (a dollar amount).
Level of appreciation: ordinal. It is measured on a 1 to 7 scale, where the numbers represent ordered categories ("not at all" to "to a great extent"). The intervals between scale points are not guaranteed to be equal, so treat this as ordinal rather than truly continuous.
Population of interest: All individuals who have purchased engagement rings for their partners, or more narrowly, all engaged couples in the United States who used the website in question.
Is the sample representative? Almost certainly not, for several reasons:
Small sample size (n = 33), which limits generalisability.
Only participants with "average" American names were included, potentially excluding ethnic and cultural diversity.
Extreme ring prices (very high and very low) were excluded, narrowing the range of the data.
Participation was voluntary, and a $10 gift certificate incentive may have attracted respondents who differ systematically from non-respondents.
Real-world application: This is a textbook example of selection bias and volunteer bias. In practice, any survey with a convenience sample and a small incentive should raise questions about whether the findings transfer to the broader population. Recognising these limitations is as important as computing any statistic.
Tags: engagement ring, appreciation, ordinal data, Likert scale, voluntary response bias, selection bias, convenience sample, small sample size, representativeness, population, sample
Students often assume that any variable expressed with numbers is quantitative. A 1-to-7 Likert scale looks numerical but represents ordered categories (ordinal). You cannot claim the distance between 2 and 3 equals the distance between 5 and 6.
Students sometimes confuse "large dataset" with "population." A dataset of 10,000 people is still a sample if the research question applies to a larger group. The bridge dataset (608,272) is a population only because it covers every bridge in the country, not because the number is big.
A sample can be large and still biased. The ring study's problems are not solved by surveying more people from the same flawed sampling frame. Bias is about how you select, not how many you select.
Students sometimes label the tenant name as a quantitative variable because it appears in a data table alongside numbers. The name is a label, not a measurement.
⚠️ Expect questions that give you a study description and ask you to identify the experimental units, the variables, and whether each variable is qualitative or quantitative. This is the most common format for Chapter 1 exam items.
⚠️ "Population or sample?" questions hinge on whether the data cover every member of the group of interest. If the dataset is a census of the entire group, it is a population. If it is a subset, it is a sample, regardless of size.
⚠️ Be prepared to explain why a sample is or is not representative. Examiners want you to name specific sources of bias (voluntary response, exclusion criteria, small n), not just say "the sample is too small."
⚠️ Ordinal vs. quantitative is a favourite trick question. If a scale has fixed labels (1 = not at all, 7 = to a great extent), argue ordinal. If the measurement is on a true continuous scale with equal intervals (temperature in degrees, price in dollars), argue quantitative.
True or false: A dataset that includes every member of the group of interest is called a sample. (False, it is a population.)
Fill in the blank: A variable whose values are categories or labels is called a ______ variable. (Qualitative / categorical.)
True or false: A 1-to-5 satisfaction rating is always a quantitative variable. (False, it is ordinal because the intervals between points may not be equal.)
Fill in the blank: When a sample systematically differs from the population it is supposed to represent, the sample has ______. (Bias.)
True or false: Increasing the sample size always eliminates sampling bias. (False, bias comes from the selection method, not the sample size.)
Q: A researcher surveys 200 customers at a shopping centre about their favourite brand of coffee. Identify the experimental units and state whether the variable is qualitative or quantitative.
A: The experimental units are the 200 customers. The variable (favourite brand of coffee) is qualitative because it records a category, not a number.
Q: A city council obtains noise-level readings (in decibels) from every intersection in the city. Is this a sample or a population? Justify your answer.
A: It is a population because the data cover every intersection in the city, leaving none out.
Q: Explain why a voluntary online survey with a prize incentive might produce a biased sample, even if the sample size is large.
A: Voluntary response means only people motivated to participate are included. The prize may attract people who value small rewards more than average, and both factors create a systematic difference between respondents and the broader population. A larger sample drawn from the same biased process does not fix the problem.
Q: A hotel asks guests to rate their stay on a scale from 1 (terrible) to 10 (excellent). Is this variable best described as quantitative or ordinal? Explain.
A: Ordinal. The numbers indicate a rank order (10 is better than 5), but the difference between a 3 and a 4 is not guaranteed to equal the difference between an 8 and a 9. The scale uses numbers as labels for ordered categories.
Q: In the Boulder Group study, why do shorter lease terms correspond to higher capitalisation rates?
A: A shorter remaining lease means greater uncertainty for the property investor: the tenant may leave sooner, potentially leaving the property vacant. To compensate for that higher risk, the market demands a higher rate of return, which shows up as a higher capitalisation rate.
The population-vs.-sample distinction sets up the entire field of inferential statistics (Chapters 7 onwards). If your data are a sample, you will use confidence intervals and hypothesis tests to generalise; if they are a population, descriptive statistics alone may suffice.
The qualitative-vs.-quantitative split determines which summary measures and charts you use. Qualitative variables get frequency tables and bar charts (Chapter 2); quantitative variables get means, standard deviations, histograms, and boxplots (Chapters 2 and 3).
The discussion of bias connects directly to sampling methods (Chapter 3 or 4, depending on the textbook), where you will learn about random sampling, stratified sampling, and other designs intended to reduce or eliminate the biases identified here.
experimental unit, observational unit, variable, qualitative, categorical, quantitative, numerical, ordinal, Likert scale, population, sample, census, representative sample, sampling bias, voluntary response bias, selection bias, convenience sample, capitalisation rate, cap rate, lease term, S&P credit rating, NBI, National Bridge Inventory, FHWA, Federal Highway Administration, structurally deficient, functionally obsolete, engagement ring study, intro stats Chapter 1, types of data, types of variables, population vs sample