Statistics Explorer
Search and explore 78 statistics concepts
Browse 78 statistics concepts covering data collection, distributions, probability, inference, and hypothesis testing. Each concept includes a plain-language definition, an intuition for when and why the idea is useful, common pitfalls, and connections to prerequisite and follow-on topics — so you can build statistical reasoning step by step.
Data Collection
Collecting data is like a shop worker counting what is on the shelves. She doesn't guess; she checks every shelf and writes each item down. Skip a shelf and the list is wrong, even if the counting was perfect. Unlike cans on a shelf, people's answers can change, so data tells you about the day you collected it.
Statistical Question
A statistical question is like asking 'What will lunch be?' every day for a month: the answer keeps changing, so you need many answers to understand it. A non-statistical question is like asking 'How many days are in a week?': you get 7 every single time, so one look is enough.
Categorical Data
Categorical data is like sorting laundry into baskets: socks, shirts, towels. You can count what is in each basket and see which is fullest, but there is no 'average basket'. If the baskets have a natural order, like small, medium and large, you can line them up, but you still can't average them.
Data Representation
Choosing a display is like choosing a map. A subway map is best for finding where to change trains, a road map for driving distances, a weather map for where it will rain. It is the same city every time, but each map answers a different question, and the wrong map can hide the answer you need.
Tally Chart
Tallying is like dropping a pebble in a jar each time something happens, then packing the pebbles into bags of five. You never hold a number in your head while counting; at the end you count the bags by fives and add the loose pebbles. Like the jar, tallies remember how many, not in what order.
Frequency Table
A frequency table is like the scoreboard after a game: it doesn't replay every goal, it lists each team with its total. You can check it by adding, the way the scoreboard total should match the goals scored. Like a scoreboard, it loses the order things happened in.
Pictograph
Pictograph pictures are like coins: you don't count the coins, you count what they are worth. Three quarters make 75 cents, not 3 cents, and three pictures with a key of 10 mean 30, not 3. Unlike coins in a purse, every picture in one pictograph must be worth the same amount.
Bar Graph
A bar graph is like a row of same-sized glasses of juice, one glass per group, standing on one table: you see which holds the most just by comparing the levels. Hide the bottom of every glass behind a wall, which is what a scale that skips 0 does, and small differences in juice look huge.
Line Graph
A line graph is like the trail you leave walking over hills: uphill means the number is rising, downhill means falling, flat ground means no change, and a steep slope means fast change. Only your footprints, the dots, are certain; the path between them is your best guess.
Line Plot (Dot Plot)
A line plot is like people lining up behind signs numbered 0, 1, 2 and 3 to match their answers. The longest line shows the most common answer, an empty sign shows a value nobody chose, and every person can still be counted. With hundreds of people the lines get too long to read, and another display works better.
Dot Plot
A dot plot is a photo, taken from above, of a crowd standing at the spots on a number line that match their answers. You see where the crowd is thickest, where there are empty patches and who stands off alone. A photo of a few dozen people is clear; a photo of thousands is a blur, and you would count by sections instead.
Pie Chart
A pie chart is a pizza shared among groups: the whole pizza is everyone, and each group's slice is its share. If one group gets half the votes, its slice is half the pizza. But you can only share out something that is one whole; numbers that don't add up to a total can't be sliced.
Stem-and-Leaf Plot
A stem-and-leaf plot is like a set of labelled shelves: each stem is a shelf for one tens digit, and each leaf is a book standing on it with only its last digit showing. The longest shelf shows where most values sit, and you can still pick out any single book. With hundreds of books the shelves overflow, and a histogram's counted bars work better.
Histogram
Rain falling into a row of identical buckets along a ruler works the same way: each drop lands in exactly one bucket, and the water levels show where most rain fell. A wider bucket would catch extra rain unfairly, which is why bins must be equal. And once a drop is in a bucket, you cannot say exactly where it landed; a histogram gives up exact values too.
Box Plot
Picture a class lined up by height and split into four groups with the same number of students. A box plot marks only where each group starts and ends. A long section means the students in that group vary a lot in height, not that it holds more students. Where the picture stops: you cannot see anyone's exact height or how many stood in the line.
Misleading Graphs
A misleading graph is a funhouse mirror held up to real data: nothing is invented, but the curved glass stretches one part and shrinks another. You catch the distortion by checking the frame, which on a graph means the axis labels and the spacing of the ticks. Where the comparison fails: a graph can also mislead by cropping data out of the frame altogether.
Mean as Fair Share
The mean is what you get if everyone pours their juice into one big jug and then fills the same number of cups to the same height. Some cups gain juice and some lose it, but none is spilled or added. Real data cannot always be poured, like people's heights, yet the mean still gives the even-share value.
Median
The median is the person standing in the middle of a line sorted by height, with as many people on one side as the other. Swap the tallest person for a basketball player and the middle person does not change. The picture needs one adjustment for an even-sized line, where the middle falls between two people and you take the value halfway between them.
Mode
The mode is the winner of a vote. Each value in the data casts one vote for itself, and whichever value gets the most votes wins. There can be a tie for first place, or a vote where everyone picks something different and nobody wins. Unlike an election result, the mode tells you nothing about how close the runners-up came.
Range
The range is like the gap between the first and the last runner to cross the finish line: it tells you how stretched out the race was, from fastest to slowest. It says nothing about the runners in between, who could be bunched near the front or spread out evenly.
Mean vs Median
The mean behaves like the balance point of a seesaw loaded with children: slide one child far out along the board and the pivot must move toward them to keep it level. The median behaves like the middle child in a queue: move the child at the end anywhere you like and the middle child stays the middle child.
Spread vs Center
Center is like the location of a village on a map; spread is how far apart its houses are. Two villages can sit at the same map point, yet in one the houses are packed around a square and in the other they are farms miles apart. Neither fact alone tells you how far you would walk to visit a neighbour.
Data Variability
Variability is the difference between books standing on a shelf and books dropped on the floor. On the shelf they stand side by side in one row; on the floor they are scattered across the room. The middle of both piles could be in the same spot, but on the floor you never know how far you will have to reach for the next book.
Quartiles
Quartiles are like dividing a sorted queue of people into four equal teams using three ropes. The middle rope is the median, and the other two ropes go halfway along each half. Each team has the same number of people, but the teams can cover very different stretches of height.
Interquartile Range (IQR)
The IQR is like measuring the width of the main crowd at an outdoor concert while ignoring the few people wandering at the back or pressed against the stage. Those stragglers can stand anywhere without changing the width of the crowd. The cost is that the IQR tells you nothing about how far the stragglers went.
Mean Absolute Deviation (MAD)
The MAD is like asking everyone in a class how far they live from school, ignoring direction, and averaging the answers: a 2 km trip from the north counts the same as a 2 km trip from the south. The analogy stops at the school itself, because for the MAD the center is the mean of the data, not a fixed building.
Standard Deviation
If the mean is home, standard deviation is the typical distance people wander from it, with long trips counting extra. One friend who strays 10 blocks adds as much to the squared total as 100 friends who each stray 1 block, so a single far-off value raises the standard deviation sharply.
Distribution Shape
Reading a distribution's shape is like looking at sand poured onto a ruler from above. A single cone is symmetric, a cone with sand trickling far to one side is skewed toward that side, a flat layer is uniform, and two cones mean two pourers. Unlike sand, small data sets can look bumpy just by chance.
Outlier Detection
An outlier rule is a smoke alarm for data. It beeps when a value is far from the rest, and then you go and look: usually burnt toast (a typo or glitch), occasionally a real fire (a genuine extreme). You never tear out the kitchen because it beeped. Unlike an alarm, you choose the sensitivity: 1.5 IQR or 3 standard deviations.
Percentiles
A percentile is your place in a queue sorted by score, read as the share of people standing behind you. Swap in a different crowd and your place moves even though your score did not. The picture is sharp for large groups; with ties or only a handful of people, 'your place' gets fuzzy.
Normal Distribution
A steady archer's misses behave like a normal curve: most arrows land near the aim point, big misses are rare, and misses to the left are as common as misses to the right. Where the archer aims is the mean; how steady the hands are is the standard deviation. Unlike real arrows, the model allows a tiny chance of a miss any distance away.
Z-Score (Standard Score)
Converting to z-scores is like converting Fahrenheit to Celsius: 86°F and 30°C are the same heat measured from a different zero with a different-sized degree. A z-score moves the zero to the mean and makes one 'degree' equal one standard deviation, so values from any group fit on one thermometer. The conversion keeps order and spacing, not the data's shape.
Weighted Average
A weighted average is the balance point of a ruler loaded with stacks of coins: put a tall stack at 90 and a short one at 70 and the ruler balances nearer 90. Equal stacks balance exactly halfway. However the coins are arranged, the balance point can never fall outside the outermost stacks.
Empirical Rule
The rule works like a recipe that scales: double the batch and every amount doubles, but the proportions stay the same. Slide the mean or stretch the standard deviation and the bell moves or widens, yet the shares stay 68, 95 and 99.7 within 1, 2 and 3 SDs. Change the shape from bell to lopsided and it is a different recipe.
Skewness
A skewed distribution looks like a comet: a dense head where most of the data sit and a long, thin tail streaming off to one side. The skew is named for the direction the tail points, and the mean gets dragged out along the tail while the median stays near the head. A comet has one tail; data with two equal tails are symmetric.
Random Sampling
Deal a hand from a well-shuffled deck and you have a random sample: every card has the same chance of being dealt, so a hand of 13 usually has a mix of suits close to the deck's. A single hand can still be lopsided by luck, which is why bigger samples help.
Population vs Sample
The population is the whole pot of soup and the sample is the spoonful you taste. A spoonful from a well-stirred pot tells you about the pot; one skimmed from the unstirred top may not. People cannot be stirred, so statisticians mix them in a different way: by choosing the sample by chance.
Sampling Bias
Picture a bathroom scale that always reads 3 kg heavy. That is sampling bias: weigh yourself ten times and the readings agree with each other, but every one is wrong in the same direction. Weighing more often never fixes it; you have to fix the scale, which here means fixing how people are chosen.
Experimental Design
A fair experiment is like a race between two teams on one track: same start line, same weather, and only one thing differs, say the shoes. Then a gap at the finish says something about the shoes. With people, the teams are made alike by assigning members at random, since you cannot line up identical people.
Observational vs Experimental Studies
An observational researcher is like a bird-watcher with binoculars: she records what the birds do but changes nothing, so she sees patterns without knowing what drives them. An experimenter is like a gardener who decides which plants get water. The gardener can test a cause; the watcher can study things no one could ethically arrange.
Confounding Variables
Two kites flying side by side rise and dip together, so it looks as if one kite tugs the other, but the wind moves both: the wind is the confounder. Compare kites flown in equally strong wind, and the false link between them disappears.
Random Assignment
Random assignment is like picking teams for a fair game by drawing names from a hat instead of letting captains choose: strong and weak players end up spread across both teams, so the result tells you more about the strategy than about the roster. With only a few players, one team can still get lucky.
Control Group
A control group works like a twin who skipped the treatment: whatever happens to that twin, from growth to practice to recovery, would also have happened to the treated twin, so the difference between them is what the treatment added. Real groups are twins only on average, which is why they are formed by chance.
Placebo Effect
The placebo effect is like a runner who believes her new shoes are faster: she may truly run faster, from confidence and effort, even if the shoes are no different. To test the shoes fairly, give other runners identical-looking ordinary shoes, tell them the same story, and see whether the new ones still win.
Blinding
Blinding works like a blind taste test: cover the labels, and people rate what is in the cup instead of the brand they expect to like. In a double-blind version, the person recording the scores cannot see the labels either. It cannot help when the difference is obvious, such as fizzy against flat.
Sampling Variability
A basketball player who makes 60% of her free throws shows sampling variability every practice: in sets of 20 shots she might make 10, 13 or 14, though her skill is the same. Longer sets give steadier percentages. A real player has good and bad days; a population stays fixed, so all of its wobble comes from which individuals a sample catches.
Sampling Distribution
A sampling distribution is like a record of a dart player's averages: single darts land all over the board, but the average position of each set of 10 darts stays close to where she aims. The more darts in each set, the tighter those averages cluster. The board, like the population, never changes; only the averages tighten.
Central Limit Theorem
A Galton board drops balls through rows of pegs; each peg nudges a ball left or right at random, and the balls pile up in a bell at the bottom. A sample mean is built the same way, from many independent values added together. With only two or three rows the pile is lumpy, just as a small sample's mean can keep the population's skew.
Standard Error
A jittery bathroom scale is typically off by about a kilogram per reading. Weigh yourself 4 times and average, and the average is typically off by about half a kilogram; average 100 readings and it is off by about a tenth. The single-reading jitter is the standard deviation; the average's jitter is the standard error. No amount of averaging fixes a scale that always reads 2 kg heavy.
Confidence Interval
Ring toss with a peg you cannot see: the peg (the true value) never moves, and each sample throws a ring (the interval) centered on its own estimate. A 95% method rings the peg on about 95 throws out of 100. Once a ring has landed, it either circles the peg or it doesn't; the 95% describes the thrower, not that throw.
Margin of Error
A phone map shows your location as a dot inside a shaded circle: you are probably somewhere in the circle. More satellite signals shrink the circle, as a bigger sample shrinks the margin, and asking for more certainty would widen it. Like the margin, the circle covers random jitter only; if the map itself is wrong, the circle cannot warn you.
Hypothesis Testing
A criminal trial runs on the same logic. The defendant (the null) is presumed innocent, the prosecution's evidence (the data) must show guilt beyond reasonable doubt (a small p-value), and the verdict is "guilty" or "not guilty", never "innocent". One difference: a test fixes in advance how often it will convict an innocent null, the rate $\alpha$.
P-Value
A p-value is the referee's reaction to a suspiciously lucky run of dice: "if these dice are fair, a run this good or better turns up about 1 game in 500." It rates how odd the data are under the innocent story. It does not say how likely the innocent story is; that also depends on how many players actually cheat.
Statistical Significance
A smoke alarm with a sensitivity dial works like a significance test. You set the dial ($\alpha$) before cooking; at 0.05, the alarm sounds on about 5% of smoke-free days. A beep means "probably not just chance", not "big fire": burnt toast and a house fire both set it off. And silence does not prove there is no smoke.
Two-Way Tables
A two-way table works like a sorting tray with one compartment for every pair of answers: each card drops into exactly one slot, so the compartments add up to the whole stack. The tray shows how many cards landed where, but not why they landed there.
Relative Frequency
Test scores in percent already use relative frequency: 18 out of 20 and 45 out of 50 are both 90%, so tests of different lengths compare fairly. Like a percent score, a relative frequency hides how many items there were, so report the total alongside it.
Conditional Relative Frequency
A conditional relative frequency works like a zoom lens on the table: you frame one row or one column, and that frame becomes the whole picture, so its total goes underneath. Zoom on a different row or column and the same cell becomes a different percentage.
Correlation
Watch two kayaks on a river. If they tend to drift the same way, they are positively correlated; if one tends to go left whenever the other goes right, negatively. Watching, like correlation, shows only that they move together; it cannot tell whether one kayak steers the other or the current moves both.
Correlation vs Causation
Correlation versus causation is like seeing wet streets and open umbrellas together: they show up at the same time, but neither causes the other, because rain causes both. To learn what really causes what, you have to change one thing yourself and watch, which is what an experiment does.
Scatter Plot
Picture pinning each student to a corkboard: across by hours of sleep, up by quiz score. Step back, and the shape the pins make, a rising band, a falling band or a scattered mess, shows the relationship at a glance. The pins show where students landed, not why they landed there.
Line of Best Fit
A line of best fit is like a straight road built through a scattered village: it cannot pass every house, so it takes the route that stays as close as it can to all of them at once. If the village curves around a lake, no straight road serves it well, and you need a curved model instead.
Linear Regression
A regression line behaves like a balanced seesaw for the data: it pivots through the point of averages, and its tilt settles where the pushes from points above and below cancel, with points far from the middle having the most leverage. Unlike a seesaw, it weighs only vertical gaps, so swapping $x$ and $y$ gives a different line.
Residuals
When a tailor tries a suit on a customer, the wrinkles that remain are its residuals: where the fabric bunches or pulls shows where the pattern does not match the body. Small random wrinkles are fine, but the same pull along one seam every time means the pattern itself needs changing.
R-Squared (Coefficient of Determination)
R-squared rates a model the way noise-cancelling headphones are rated: the spread of $y$ around its mean is the noise, and $R^2 = 0.85$ means the model removes 85% of the noise power, the squared spread. A high rating says how much was removed, not whether the model is the true explanation.
Correlation Coefficient
The correlation coefficient scores how well a crowd keeps to a single file: $\pm 1$ means everyone stands exactly on one straight line, 0 means no line at all, and the sign says whether the file runs uphill or downhill. A crowd standing in a perfect circle scores 0, even though it is very orderly.
Basic Probability
Probability works like slicing a pizza into equal slices: if 3 of 5 equal slices have pepperoni, a slice picked blindly has a $\frac{3}{5}$ chance of pepperoni. The picture breaks when the slices are different sizes; then counting slices gives the wrong chance, just as unequal outcomes break favorable over total.
Theoretical Probability
Reading the blueprint instead of inspecting the building: that is theoretical probability. The plan of a fair die says each face gets one sixth of the chances, without a single roll. A real die built slightly off-plan, worn or loaded, will not match the blueprint, and only experiments can reveal that.
Experimental Probability
Judging a restaurant by its reviews instead of its menu is the experimental approach: you go by what actually happened to the people who ate there. Three reviews can mislead you; three hundred give a fair picture. The limit: reviews describe past meals, and if the kitchen changes, as when a coin gets bent, the old data stops applying.
Sample Space
A sample space is the class roster before a name is drawn from a hat: every student on it exactly once. Miss a name and everyone's chance is miscounted; write a name twice and that student's chance doubles. The roster works because each slip in the hat has the same chance; for lopsided outcomes, a list alone is not enough.
Tree Diagram
A tree diagram works like a map of hiking trails that fork: each fork is a stage, and each complete route from the trailhead to an end point is one outcome. The share of hikers taking a route is the product of the shares choosing each fork along it. Unlike real trails, routes on a tree never rejoin.
Compound Events
'And' is one door with two locks: you get through only if both open, so it is harder than opening either lock. 'Or' is a room with two doors: either one lets you in, so it is easier than using one particular door. The picture tells you which way the chance moves, not by exactly how much.
Conditional Probability
Zoom a camera onto one section of a stadium: once you know the person is in the student section, you count only the people in that frame and ask what share of them wear red. Zooming changes the group you divide by, not the people in the overlap.
Independent Events
Independence is like the weather report for a city on the other side of the world: the news is real, but it does not change whether you pack an umbrella. Dependent events are like your own town's forecast, which should change your plans. The test is not whether two events sound related but whether news of one should change how likely the other is.
Addition Rule
Two friends planning a joint party add their guest lists, then subtract the people invited by both, or those guests get counted twice and too much food gets ordered. That is the addition rule; if the friends share no guests, the lists simply add. For equally likely outcomes the fit is exact, since a probability is just a head count divided by the total.
Multiplication Rule
The multiplication rule works like a two-round tryout: if 5 of 8 students pass round one, and 4 of every 7 who reach round two pass it, the share passing both is $\frac{5}{8} \times \frac{4}{7}$, a fraction of a fraction. The round-two rate must be measured among those who passed round one, which is what $P(B \mid A)$ means.
Statistical Simulation
A simulation is a flight simulator for chance: instead of risking the real plane, you fly a model of it thousands of times and count how often things go wrong. The results are only as good as the model; a simulator with the wrong wind settings trains you for the wrong weather.
Law of Large Numbers
An early run of luck is like a spoonful of salt in soup: in one cup it is overpowering, but stir it into a big pot and you can barely taste it. Nothing removes the salt; more soup dilutes it. In the same way, extra heads stay in the count but fade in the proportion.
Expected Value
Expected value is the balance point of a seesaw: place each outcome on a number line and stack a weight on it proportional to its probability, and the board balances at the expected value. A heavy stack near a small loss and a tiny weight far out at a big prize balance close to the loss, which is why rare jackpots add so little.