Statistics is the field of collecting, analysing, interpreting and presenting data, covering the whole cycle from gathering numbers to drawing a conclusion from them, not just computing an average. This covers the vocabulary everything else in statistics builds on: population versus sample, descriptive versus inferential statistics, variable types, levels of measurement, and the difference between sampling error and a flawed measurement.
In this guide
Population versus sample, parameter versus statistic
| Term | Meaning |
|---|---|
| Population | The entire group of individuals or items a study wants to draw conclusions about |
| Sample | A subset of the population that is actually observed or measured |
| Parameter | A fixed but usually unknown fact about the population, such as the true average income |
| Statistic | A value computed from a sample, used to estimate the matching parameter |
A census collects data from every single member of the population; a sample survey deliberately measures only part of it. Researchers usually study a sample rather than the whole population because a full census is often too costly, too slow or too impractical to carry out, not because a sample is somehow more accurate.
Descriptive versus inferential statistics
| Branch | What it does |
|---|---|
| Descriptive statistics | Summarises and describes the data already in hand, through measures like the mean or a chart like a histogram |
| Inferential statistics | Uses sample data, and probability, to draw conclusions about the wider population the sample came from |
A sample only ever gives an estimate, never a perfect match to the population. This is exactly why inferential statistics leans on probability: it measures how confident a conclusion drawn from a sample can reasonably be.
Variable types
| Variable type | Meaning | Example |
|---|---|---|
| Quantitative (numerical) | A measurable number | Monthly spend in rupees |
| Qualitative (categorical) | Names a category, not a quantity | A product's listed colour |
| Discrete | Takes countable, separate values | Support tickets received in a day (0, 1, 2…) |
| Continuous | Can take any value within a range | A person's exact height in centimetres (162.4 cm) |
Quantitative/qualitative and discrete/continuous are two separate splits that combine: ticket counts are discrete and quantitative, while height is continuous and quantitative. A loyalty tier like Gold or Silver is qualitative, with no discrete/continuous distinction applying to it at all.
Levels of measurement
| Level | Order meaningful? | Equal gaps meaningful? | True zero? | Example |
|---|---|---|---|---|
| Nominal | No | No | No | Blood type, city name |
| Ordinal | Yes | No | No | A rating of low, medium or high |
| Interval | Yes | Yes | No | Temperature in Celsius |
| Ratio | Yes | Yes | Yes | Weight or income |
Why Celsius is interval, not ratio
0 degrees Celsius does not mean a complete absence of temperature; it is simply the freezing point of water, chosen by convention. That makes Celsius an interval scale, where a difference like “10 degrees warmer” is meaningful but a ratio like “20 degrees is twice as hot as 10” is not. Kelvin’s zero genuinely means no thermal energy at all, which is why Kelvin is a true ratio scale and Celsius is not.
How data gets collected
A sampling frame is the real, practical list a sample is physically drawn from, such as a customer database or an electoral roll, which may not perfectly match the true population. How that frame gets sampled matters: a survey that only samples shoppers walking past one mall entrance is convenience sampling, picking whoever is easiest to reach, which is a problem precisely because people who never visit that entrance have no chance of being included, biasing the result before any analysis even starts.
Sampling error versus non-sampling error
| Error type | What it is | Fixed by a larger sample? |
|---|---|---|
| Sampling error | The natural difference between a sample statistic and the true population parameter, simply because a sample is only part of the population | Yes, a larger sample generally reduces it |
| Non-sampling error | A flaw in measurement or process, such as a broken instrument, that affects every reading the same way | No, the bias repeats no matter how much more data is collected |
A faulty weighing scale that adds exactly 2 kg to every single measurement is a non-sampling error, not sampling error, and collecting a much larger sample cannot fix it: every reading is shifted by the same fixed amount, so more data never cancels the bias out. Sampling error, by contrast, genuinely does shrink as the sample grows.
Study designs
| Design | What it does |
|---|---|
| Observational study | The researcher only watches and records what is already happening naturally |
| Experiment | The researcher actively assigns a treatment or intervention |
| Cross-sectional | Measures many different subjects at one single point in time |
| Longitudinal | Follows the same subjects repeatedly over an extended period |
A survey measuring many different customers’ satisfaction on a single day is cross-sectional data, a snapshot across subjects at one point in time, as opposed to longitudinal or time series data, which track the same subject across time instead.
Univariate, bivariate and multivariate
| Term | How many variables |
|---|---|
| Univariate analysis | One variable at a time, such as finding its average |
| Bivariate analysis | The relationship between two variables |
| Multivariate analysis | The relationships among three or more variables at once |
A dataset recording each customer’s age, income and number of purchases together, studied for how all three relate, is multivariate analysis, since it goes beyond a single pair of variables into three or more at once.
Common mistakes
- Mixing up parameter and statistic. A parameter describes the population; a statistic describes a sample and estimates the parameter.
- Treating ordinal data as if its gaps were equal. Low/medium/high can be ranked, but the size of the jump between them is not a fixed, measurable amount the way it is for interval or ratio data.
- Calling Celsius a ratio scale. Its zero is a convention, not a true absence of the quantity, so ratios between Celsius readings are not meaningful.
- Assuming a bigger sample fixes every error. It only helps with sampling error; a biased measurement process repeats at any sample size.
- Confusing convenience sampling with random sampling. Easy-to-reach participants are not the same as every member having an equal chance of inclusion.
Cheat sheet
| Question | Answer |
|---|---|
| What describes the population: parameter or statistic? | Parameter |
| What branch generalises beyond the sample? | Inferential statistics |
| Is ticket count discrete or continuous? | Discrete |
| Is height discrete or continuous? | Continuous |
| Which level of measurement has a true zero? | Ratio |
| Which level has ordered categories but unequal gaps? | Ordinal |
| What collects data from the whole population? | A census |
| What error type repeats no matter the sample size? | Non-sampling error |
| What studies three or more variables at once? | Multivariate analysis |
Try it yourself
Quick check
A company surveys every one of its 40,000 employees about job satisfaction, with a 100 percent response rate. Is this a census or a sample survey, and does it have sampling error?
Show answer
It is a census, since every single member of the population, all 40,000 employees, was measured. A true census has no sampling error at all, since sampling error only arises from measuring part of a population rather than the whole of it. Any discrepancy left over here would have to come from a non-sampling source, such as employees answering dishonestly, not from the sampling process itself.
Frequently asked questions
What is the difference between a population and a sample in statistics?
A population is the entire group a study wants to draw conclusions about. A sample is a subset of that population that is actually observed or measured, used to learn something about the whole group.
What is the difference between descriptive and inferential statistics?
Descriptive statistics summarises data that has already been collected, such as computing a mean or drawing a chart. Inferential statistics uses a sample, and probability, to draw conclusions about the wider population the sample came from.
What are the four levels of measurement in statistics?
Nominal (unordered categories), ordinal (ordered categories with unequal gaps), interval (equal gaps, no true zero) and ratio (equal gaps with a true zero, so ratios between values are meaningful).
What is the difference between sampling error and non-sampling error?
Sampling error is the natural difference between a sample statistic and the true population parameter, simply because a sample is only part of the population; it shrinks as the sample grows. Non-sampling error comes from a flawed measurement or process, affects every reading the same way, and does not improve with a larger sample.
What is the difference between an observational study and an experiment?
In an experiment, the researcher actively assigns a treatment or intervention. In an observational study, the researcher only watches and records what is already happening naturally.
What is multivariate analysis?
Multivariate analysis studies the relationships among three or more variables at once, as opposed to univariate analysis (one variable) or bivariate analysis (the relationship between two variables).
Test yourself
Timed questions on Statistics Basics and Foundations, with an explanation for every answer.