What should you know about Statistics?
Statistics has two broad jobs: describing observed data and making qualified inferences beyond those observations. Start by identifying the kind of data, summarizing its center and spread, examining its distribution, and checking how it was collected before drawing conclusions.
Statistics at a glance
| Concept | What to know |
|---|---|
| Descriptive statistics | Summarize observed values using tables, charts and numerical measures. |
| Inferential statistics | Use sample information and statistical assumptions to estimate or test population claims. |
| Center | Mean, median or mode; different measures answer different questions. |
| Spread | Range, interquartile range, variance or standard deviation. |
| Distribution | The pattern of values: symmetry, skew, peaks, gaps and extremes. |
| Uncertainty | Sampling error, bias, measurement error and model assumptions affect conclusions. |
Population, sample and types of statistics
A population is the full set of people or observations the question concerns; a sample is the subset actually measured. Suppose a school wants to know the average daily travel time of all 1,200 students. The population is those 1,200 students. Responses from 100 selected students form a sample. How those 100 were selected matters as much as how their average is calculated.
Descriptive statistics report what was observed: a sample average, a percentage, or a distribution of responses. Inferential statistics goes further by using sample information to estimate a population quantity or compare hypotheses. That step requires attention to the sampling method, uncertainty and model assumptions.
A precise calculation cannot rescue an unrepresentative sample. A voluntary online poll, for example, may attract people who care strongly about a topic, so its percentage should not automatically be treated as the population percentage.
In a survey of 100 students, 62 say they prefer online notes. “62% of respondents” describes the sample. “62% of all students” is a population claim that needs a defensible sampling design.
Check this: Descriptive results describe measured data; they do not automatically generalize to everyone.
Recognize variables and choose appropriate summaries
A categorical variable records groups or labels, such as product type. A quantitative variable records numerical amounts, such as order value or waiting time. Quantitative variables can be counts (discrete) or measurements (often treated as continuous).
The data type determines which summaries make sense. Frequencies and proportions work for categories. Medians, means, standard deviations and distributions describe numerical values. A customer ID looks numeric but remains an identifier; calculating its average is meaningless.
Before analysis, confirm the unit of observation. One row per purchase is not the same as one row per customer. Counting transactions when the question asks about customers can exaggerate the apparent size of the audience.
For product categories A, B and C, report counts or percentages. For delivery times, report a typical time, the spread and perhaps a histogram.
Check this: A label encoded as a number is not automatically a quantitative measurement.
Mean, median and mode: three ways to describe center
The mean is the sum of numerical observations divided by their count. The median is the middle value after sorting, and the mode is the most frequent observed value or category. These measures can disagree because they describe different aspects of the data.
Consider delivery times of 4, 5, 5, 6 and 20 minutes. The mean is 8 minutes; the median and mode are 5 minutes. The unusually long delivery pulls the mean upward. If the goal is to describe a typical experience, the median may be more informative.
The median is not always “better” than the mean. The mean is useful for totals and many statistical models. Choose according to the question and the distribution, and check the mean, median and mode guide for fuller examples.
For 4, 5, 5, 6 and 20: mean = (4 + 5 + 5 + 6 + 20) / 5 = 8; median = 5; mode = 5.
Check this: Do not treat the mean, median and mode as interchangeable or assume every dataset has exactly one mode.
Measure spread, not just the average
Two groups can have the same mean but very different variability. The range compares minimum and maximum values; the interquartile range (IQR) covers the middle half of ordered observations. Standard deviation measures variation around the mean in the original units, while variance uses squared units.
A standard deviation of 0 means all observed values are equal. A larger value indicates greater spread around the mean, but its practical importance depends on the measurement scale and context. For strongly skewed data, a median with an IQR may communicate the distribution more clearly than a mean with a standard deviation.
The standard deviation explanation works through population and sample formulas. A box plot can display quartiles and the median in one compact visual.
Teams A and B both average 10 tasks per day. A produces 9, 10, 11; B produces 2, 10, 18. Their means match, but their spreads do not.
Check this: The commonly taught 68–95–99.7 rule applies to a normal model, not every dataset.
Read histograms and distribution shapes
Numerical values are rarely described fully by one average. A distribution shows where observations concentrate, how much they vary, whether the pattern is approximately symmetric or skewed, and whether there are noticeable gaps or peaks.
A histogram groups numerical observations into intervals. It may appear roughly uniform, bell-shaped, left- or right-skewed, unimodal or bimodal. Those words describe different properties: symmetry concerns balance, whereas modality concerns the number of apparent peaks. For visual identification, see histogram shapes and what they mean.
A histogram that looks bell-shaped does not establish a normal population. Sample size, bin width and the placement of bin boundaries affect its appearance. Examine the data and context before interpreting extra peaks or isolated bars.
A right-skewed histogram has a longer tail toward larger values; the name follows the tail, not the location of the tallest bars.
Check this: Do not infer “normal distribution,” “two populations” or “outlier” solely from the appearance of a histogram.
Sampling, uncertainty and statistical claims
An estimate from a sample usually differs from the population value. Confidence intervals communicate uncertainty under specified statistical assumptions. The exact interpretation depends on the procedure; a 95% confidence level describes the long-run coverage of the interval-producing method, not a guaranteed 95% probability for a fixed population parameter.
A p-value measures how compatible observed results are with a stated null model, in a particular direction or according to a chosen test statistic. It is not the probability that the null hypothesis is true and it does not measure the practical importance of a result.
Before claiming a difference or a trend is meaningful, check selection bias, missing measurements, dependence, comparison groups and how many tests were conducted. Association alone does not establish causation.
If a sample suggests a 4-point difference between two groups, the size of the sample and variation within groups determine how precisely that difference is estimated.
Check this: Statistical significance and real-world importance are different questions.
Frequently asked questions about Statistics
What is the difference between descriptive and inferential statistics?
Descriptive statistics summarize observations that were collected. Inferential statistics use sample evidence, with assumptions and uncertainty, to make claims about a wider population.
Which is better: mean or median?
Neither is always better. Use the mean when the arithmetic average answers the question; use the median when you need the middle value or want less sensitivity to extreme observations.
Does a bell-shaped histogram prove the data are normal?
No. A bell-shaped sample histogram may be consistent with a normal model, but bin choice, sample size and other distributions can produce a similar appearance.
Why report standard deviation?
It describes variation around the mean in the same units as the original measurement. Interpret it alongside the mean, the distribution and the question being asked.
Can statistics prove that one variable causes another?
A statistical association alone cannot prove causation. Strong causal conclusions need a suitable design, assumptions and supporting evidence.
The articles below explore individual ideas with more detail, examples and practical steps.
Go to the article library ↓