Glossary

100 terms from Statistics Basics, in plain words, each with the chapter that teaches it.

Search the glossary

A

A/B testing: An experiment in which customers are randomly split into two groups, one getting a version (such as the KC15 coupon) and the other not, so the groups can be compared fairly. Because the split is random, it is the most reliable way to learn whether a change causes a difference in outcomes.
Absolute change: The plain difference between two values (new − old). For two percentages it is measured in percentage points, such as Kettle & Crane's repeat rate rising from 20% to 25%, which is 5 points.
Absolute reference: A cell reference with dollar signs, like $E$2:$E$21, that stays fixed when you copy the formula. Sam needs it so every order's share divides by the same total.
Argument: A value or range you give a function inside its parentheses, separated by commas. In =SUM(E2:E21), E2:E21 is the argument.
AVERAGE: The spreadsheet function that calculates the mean of a range. It skips empty cells and text but includes zeros.
AVERAGEIFS: A function that averages only the values meeting every condition you give it. The range to average comes first, then pairs of condition range and criterion.

B

Bar chart: A chart where bar lengths show values. In Excel and Sheets, vertical bars are a column chart and horizontal bars a bar chart; the value axis should start at zero.
Base: The whole that a percentage is 'of', which is the number you divide by. Most percentage mistakes come from not saying what the base is.
Base rate: How common something is in the whole group before you look at any subset. Ignoring it leads to the base-rate trap, such as thinking a fake-review filter flag is 90% reliable when only about 1 in 4 flags is a real fake.
Bias: A systematic difference between a sample and the population it is meant to represent. A bigger sample doesn't fix it.
Bimodal: A distribution with two peaks, often meaning two groups are mixed together, like single-item and two-item orders at Kettle & Crane.
Bin: One of the equal-width ranges values are grouped into for a frequency table or histogram, such as order totals from $10 up to but not including $20.
Box plot: A chart of the five-number summary: a box from Q1 to Q3 with the median marked, whiskers to the last non-outlier values, and outliers as separate dots.

C

Case: One thing described by a row of data, such as a single order or review.
Categorical data: Data made of labels rather than quantities. Nominal labels have no order (state, product); ordinal labels have a natural order (star rating, size).
Cell: One box in a spreadsheet, identified by its address (cell reference) such as E2, meaning column E, row 2.
Center: The typical value of a dataset. The three common measures are the mean, median and mode.
Cherry-picking: Choosing a start date, subset or result that makes a story look best, such as measuring 'up 36% since February' from the lowest month.
Confidence interval: A range likely to contain the population value, such as 62% ± 8 points. By convention it means about 19 in 20 such ranges from repeated surveys would contain the truth.
Confounder: A third factor that affects both variables you are looking at, creating a correlation between them even if neither causes the other. Café size confounds workshops and coffee sales.
CORREL: The spreadsheet function that returns the correlation coefficient r for two ranges of equal size.
Correlation coefficient: A number from −1 to 1 measuring the direction and strength of a straight-line relationship between two variables. It says nothing about cause and can be distorted by outliers.
Correlation is not causation: Two things moving together doesn't prove one causes the other; it could be reverse causation, a confounder, chance or selection.
COUNT: A function that counts the cells in a range holding numbers; text and empty cells are skipped.
COUNTA: A function that counts cells that aren't empty, whatever they hold. Comparing it with COUNT reveals text stored where numbers should be.
COUNTBLANK: A function that counts empty cells in a range, such as orders with no review.
COUNTIF: A function that counts cells matching one condition, such as "OR" or ">40". It ignores capital letters but not stray spaces.
COUNTIFS: COUNTIF with several range-and-condition pairs, all of which must be true; used to build frequency tables and count groups.
Criterion: The condition, usually in quotes, that COUNTIF, COUNTIFS, SUMIFS and AVERAGEIFS test, such as ">40" or "Yes".
CSV: CSV (comma-separated values) is a plain-text table format with commas between the values on each row. It is widely used to share datasets, and spreadsheet programs such as Excel and Google Sheets can open or import it.

D

Dataset: A table of cases (rows) and variables (columns), such as Kettle & Crane's 20 online orders.
Descriptive statistics: Describing what is in your data with averages, spread and charts, as opposed to drawing conclusions about a wider population.
Deviation: How far a value is from the mean (value − mean). Squaring and averaging deviations gives the variance, and its square root is the standard deviation.
Distribution: The pattern of how often each value, or range of values, occurs. Looking at it before summarizing shows what averages hide.

F

Fences: The cutoffs Q1 − 1.5 × IQR and Q3 + 1.5 × IQR; values outside them are flagged as outliers. For Kettle & Crane's orders the upper fence is $66.75.
Fill handle: The small square at a selected cell's bottom-right corner; dragging it copies the cell's formula down or across.
Five-number summary: Minimum, Q1, median, Q3 and maximum: a five-value description of a dataset. For the order totals it is $12, $16.75, $30.50, $36.75 and $412.
Formula: A calculation in a spreadsheet cell, always starting with =. Formulas should refer to cells rather than typed numbers so they update.
Formula bar: The bar above the spreadsheet grid that shows the formula of the selected cell.
Frequency table: A table of bins and how many values fall into each; it's the basis of a histogram.
Function: A named, ready-made spreadsheet calculation, such as SUM, that takes arguments in parentheses.

H

Histogram: A chart of touching bars showing how many values fall in each bin. Bin width changes the picture, so try several.

I

Identifier: A label that happens to be made of digits, like order_id or zip code. It looks numerical but should be counted, never averaged.
Interpolation: Estimating a value between two known values, which is how QUARTILE.INC gets $16.75 even though no order cost that.
Interquartile range (IQR): Q3 − Q1: the width of the middle half of the data. It is robust to outliers, unlike the range or standard deviation.

L

Line chart: A chart that connects points over time to show how something changes. Unlike bar charts, a line chart can zoom its axis fairly because it shows change by slope.

M

Margin of error: How far a sample result is likely to be from the population value because of chance alone, such as ±8 points for 150 survey replies. It does not cover bias.
Mean: The sum of the values divided by how many there are; the balance point. It's right for totals but dragged by extreme values like the $412 order.
Median: The middle value when data is sorted, or halfway between the two middle values. It is robust to extreme values, so it describes a typical order better.
MEDIAN: The spreadsheet function that returns the median of a range, sorting internally.
Mode: The most common value. MODE.SNGL returns one mode (and #N/A if nothing repeats); MODE.MULT lists all tied modes. For categories, count with COUNTIF instead.
Month-over-month change: Percent change from the previous month, such as Jo's order totals rising 3.0% from July to August. It bounces around, so pair it with a moving average.
Moving average: The mean of each value and a fixed number of values before it, used to smooth month-to-month noise and show the trend.

N

Natural frequencies: Counts of imagined people or cases (like 18 of 67 flagged reviews) used instead of probabilities, which makes base-rate problems easier to reason about.
Non-response bias: Bias that arises when many people don't answer and those who do differ from those who don't, like busy café owners ignoring a survey.
Normal distribution: A symmetric, single-peaked, bell-shaped distribution in which about 68% of values lie within 1 SD of the mean, 95% within 2 and 99.7% within 3.
Numerical data: Quantities you can do arithmetic on: continuous (measured, like order total) or discrete (counted whole numbers, like items per order).

O

Outlier: A value far from the rest, judged by a stated rule such as the fences or z-scores. Investigate it, fix errors, keep real values and never delete it silently.

P

Percent change: (new − old) ÷ old: a change relative to the starting value. Rises and falls aren't symmetric, since a 50% drop needs a 100% rise to recover.
Percentage: A part divided by a whole, shown out of 100. In a spreadsheet, divide and then format as a percentage rather than multiplying by 100.
Percentage points: The unit for the difference between two percentages. A rate rising from 20% to 25% is up 5 points, or 25% in relative terms, never 'up 5%'.
Percentile: The value with a given percentage of the data at or below it. The 90th percentile of delivery times lets you promise '9 in 10 orders arrive within X days'.
Pie chart: A circle divided into slices showing parts of a whole. People judge angles poorly, so use one only for two or three parts that clearly add up to 100%.
Pivot table: A summary table built by dragging fields into rows, columns and values, without writing formulas. It needs a tidy table, and in Excel must be refreshed when data changes.
Population: The whole group you want to know about, such as all Kettle & Crane customers.
Positive and negative relationship: Two variables rising together is a positive relationship; one falling as the other rises is negative.

Q

Quartiles: Values splitting sorted data into four parts: Q1 has a quarter below it, the median is the second quartile, Q3 has three quarters below. QUARTILE.INC and .EXC use different position rules, so answers can differ.

R

RAND: A function giving a random number between 0 and 1 that changes on every recalculation. Paste the numbers as values before sorting to draw a random sample.
Random sample: A sample in which every member of the population had the same chance of being chosen; the best protection against selection bias.
Range: In statistics, the largest value minus the smallest, which depends on only two extreme values. In a spreadsheet, a block of cells such as E2:E21.
Regression to the mean: The tendency of an extreme result to be followed by a more ordinary one because luck helped make it extreme, such as a café's record month followed by a smaller one.
Relative change: A change expressed as a percentage of the starting value. It can exaggerate small numbers, like complaints 'doubling' from 1 to 2.
Relative reference: A cell reference that shifts when the formula is copied, so E2 becomes E3 one row down.
Robust: Hardly affected by extreme values. The median, quartiles and IQR are robust; the mean and standard deviation are not.
R²: r × r: the share of the variation in y that the straight-line trend accounts for. For cold brew and temperature it is 0.91.

S

Sample: The part of the population you actually have data on, such as the ten deliveries Alex tracked.
Sampling variation: Differences between samples caused by chance alone. It shrinks as samples get bigger, unlike bias.
Scatter chart: A chart with one dot per case showing how two numerical variables move together. Describe it by direction, form and strength.
Selection bias: Bias from the way cases are chosen favoring some kinds, such as surveying only newsletter subscribers.
Simpson's paradox: A comparison that reverses when the data is split into groups, because the groups have a different mix of a hidden factor. UPS beat USPS for small and large packages yet looked worse overall.
Skew: Lopsidedness: a longer tail on one side that pulls the mean toward it. Order totals are right-skewed (mean above median); star ratings are left-skewed.
Slope: How much y changes on a trendline for each one-step increase in x; about 1.48 more cold brew bottles per extra degree. SLOPE takes the y range first.
Spread: How far values are scattered around the center. An average without a measure of spread is half a summary.
Spreadsheet: A grid of cells arranged in lettered columns and numbered rows, where you enter data and write formulas that calculate from it. Excel and Google Sheets are common examples.
Standard deviation (SD): Roughly the typical distance of values from the mean. STDEV.S is for samples (divides by count − 1) and STDEV.P for a complete population; extreme values inflate it.
Standard error: SD ÷ √n: the typical gap between a sample mean and the population mean.
Statistics: The set of methods for summarizing data and reasoning about it honestly. True numbers can still mislead if they are summarized badly.
SUMIFS: A function that totals the values meeting every condition; the sum range comes first, then condition pairs.
SUMPRODUCT: A function that multiplies ranges row by row and adds the results, the top half of a weighted average.
Survivorship bias: Bias from counting only the cases that 'survived', like asking current wholesale customers why they're happy but not the cafés that left.

T

Tidy table: A layout with one header row, one row per case, one column per variable, one value per cell and consistent labels, which formulas, charts and pivot tables need.
Trendline: The straight line that best fits the dots on a scatter chart, optionally shown with its equation and R².
Trimmed mean: The mean after dropping a share of the highest and lowest values; TRIMMEAN(range,0.1) drops 10% in total, half from each end.
Truncated axis: A value axis that doesn't start at zero, making small differences look huge. Bar charts must start at zero.

V

Variable: A fact recorded about every case, stored in one column; it varies from case to case.
Variance: The sum of squared deviations divided by count − 1; the square of the standard deviation, in squared units.
Voluntary response bias: Bias from people choosing whether to take part, so strong views are overrepresented, as in product reviews.

W

Weighted average: A mean where each value counts according to its weight, usually the number of cases it represents. It avoids the trap of averaging averages.

X

x-axis and y-axis: A chart's horizontal and vertical axes. On a scatter chart, x is by convention the variable you think might explain y.

Z

z-score: How many standard deviations a value is from the mean, (value − mean) ÷ SD. Beyond ±3 is usually treated as an outlier, though outliers inflate the SD and shrink z-scores.