Four numbers that describe a set of scores
An assessment produces a distribution, not a single fact, and four numbers describe most of it. The mean is the balance point: every score pulls on it in proportion to its distance. The median is the middle student, immune to how extreme the extremes are. The standard deviation is the typical distance from the mean, which is how much the test spread the class out. And the range from low to high tells you what the instrument actually measured from end to end.
Reading the mean without the median is the most common mistake. If a class of 25 has one student who scored 12 and the rest between 70 and 95, the mean drops several points while the median does not move at all. The gap between the two is a direct signal of skew: mean below median means a tail of low scores, mean above median means a tail of high scores, and agreement between them means the distribution is roughly symmetric.
The standard deviation answers a different and often more important question: did this test discriminate? An exam on which everyone scored between 84 and 88 has a mean you can report and no information about who learned what. A standard deviation near zero is a warning that the assessment was either too easy, too short, or scored too generously to separate students — and it makes any grade based on it close to arbitrary.
How each statistic is computed
Mean. Add the scores and divide by how many there are. Nothing else. Because every score contributes, one extreme value moves the mean by its distance divided by n — a 0 in a class of 20 with everyone else at 80 pulls the mean down by exactly 4 points.
Median. Sort the scores and take the middle one. With an even count there is no single middle, so take the average of the two central values. The median ignores magnitude entirely: it only cares about order, which is why it is unmoved by outliers.
Standard deviation. Take each score's deviation from the mean, square it, average the squares, and take the square root. The squaring is what makes large deviations count disproportionately and what makes the result carry the same units as the scores. This calculator reports the population form, dividing by n, because a class roster is the entire group you are describing rather than a sample from a larger population. The sample form divides by n − 1 and is slightly larger; it appears in the steps for when you are treating this class as one draw from a larger cohort. With 20 students the two differ by about 2.6%, which rarely changes a decision but is worth stating when you report a figure.
Letter distribution. Each raw score is converted to a percentage of the points possible, then placed in the highest band whose cut-off it clears. The bands are yours to set: a 93/85/77/70 scale and a 90/80/70/60 scale give very different pictures of the same class, which is the point of making them editable rather than fixed.
Worked example: five scores, computed by hand
Take five scores on a 100-point test: 90, 80, 70, 60 and 100.
- Sum. 90 + 80 + 70 + 60 + 100 = 400.
- Mean. 400 ÷ 5 = 80.00.
- Sort. 60, 70, 80, 90, 100. The middle of five is the third, so the median is 80. Mean and median agree, so the distribution is symmetric.
- Deviations from the mean. +10, 0, −10, −20, +20.
- Square them. 100, 0, 100, 400, 400, which sum to 1,000.
- Population variance. 1,000 ÷ 5 = 200. Standard deviation = √200 = 14.14.
- Letters on a 90/80/70/60 scale. 100 and 90 are A grades, 80 is a B, 70 is a C, 60 is a D: one D, one C, one B, two A grades.
- Above the mean. 90 and 100, so 2 of 5.
The sample standard deviation, for comparison, is √(1,000 ÷ 4) = √250 = 15.81. Report which one you used — a difference of 1.7 points looks like a disagreement if you do not.
Now break the symmetry. Replace the 60 with a 10 and recompute: the sum becomes 350, the mean 70, but the sorted list is 10, 70, 80, 90, 100 and the median is still 80. One score moved the mean by ten points and the median not at all. The mean now sits ten points below the median, and reporting only the mean would tell a department chair that this class performed far worse than four of its five students actually did.
What the numbers should make you do
Start with the standard deviation as a fraction of the points possible. Below roughly 5% of the total, the assessment barely separated anyone and letter grades derived from it are noise. Between about 8% and 15% is the range in which a test usually discriminates well while still producing a recognisable distribution. Above 20%, check for outliers before concluding anything: a handful of near-zero scores from students who did not attempt the test will inflate the spread and tell you nothing about the instrument. These are working rules of thumb for reading a classroom assessment, not published thresholds.
Then compare mean and median. A mean well below the median almost always means a small number of very low scores — students who missed the test, submitted blank work, or did not prepare at all. That is a roster problem, not a test problem, and curving the whole class to fix it rewards everyone for a few students' absence. A mean above the median points the other way, usually a ceiling effect where many students clustered near full marks.
Read the letter distribution last, and read it as a consequence of your cut-offs rather than as a fact about the class. Moving a single cut-off by two points can shift several students across a boundary. If the distribution looks wrong to you, decide whether the assessment was miscalibrated or the scale was, and fix the one that is actually at fault. The curved grade calculator shows what each standard curve would do before you commit to one.
Finally, resist the temptation to compare this mean against another section's without checking the spread and the count. Two sections with means of 78 and 82 may be statistically indistinguishable if the standard deviation is 14 and each has 20 students. A mean is a point estimate, and small classes produce noisy ones.
Reading a standard deviation on a 100-point test
| Standard deviation | Typical distance from the mean | Approximate span of the middle 68% | What it usually indicates |
|---|---|---|---|
| 2 points | 2% | 4 points wide | The test did not separate students; grades from it carry little information |
| 5 points | 5% | 10 points wide | Narrow spread; letter bands of 10 points are barely resolved |
| 10 points | 10% | 20 points wide | Typical of a well-calibrated classroom exam |
| 15 points | 15% | 30 points wide | Wide but usual on a demanding exam with real discrimination |
| 25 points | 25% | 50 points wide | Check for outliers or a bimodal split before interpreting |
The 68% figure assumes an approximately normal distribution. Classroom score distributions are frequently skewed, so treat it as an orientation rather than a guarantee.
Assumptions and limits
- Population versus sample standard deviation. This tool reports the population form as the headline figure and the sample form in the steps. State which you are quoting; with small classes they differ visibly.
- Missing students. A student who did not sit the test is not a zero. Including absences as zeros drags the mean down and inflates the spread, describing a class that does not exist.
- Letter bands are policy, not statistics. The distribution reflects the cut-offs you entered as much as the scores. Changing the scale changes the picture without any student performing differently.
- The mean is not robust. One extreme score moves it by that score's distance divided by the class size. Always read it against the median.
- Small classes give noisy statistics. With fewer than about ten scores, the standard deviation is an unstable estimate and comparisons between sections are usually not meaningful.
- These are descriptive statistics only. Nothing here tests a hypothesis, establishes reliability, or measures whether items discriminated. An item analysis answers those questions.
What to do with the distribution once you have it
If the distribution is too low to stand, the curved grade calculator compares a flat addition, a square-root curve, scaling to the top score and a mean shift on the same set of scores, and shows which students cross a letter boundary under each. Deciding between them is easier once you have seen the mean and standard deviation each one produces.
If you are grading the set in the first place, the test grade percentage calculator converts raw marks or questions-wrong counts into the percentages you paste here, and prints an easy-grader chart for a test of that length. Once the assessment enters the course grade, the weighted grade calculator combines it with the other categories at their syllabus weights.
For students asking what the distribution means for them individually, the final exam grade calculator is the practical follow-up: it turns a current standing and an exam weight into the score needed for a target. And the same descriptive machinery works on non-grade data — paste a set of attendance percentages from the attendance percentage calculator as the score list and set points possible to 100.
One methodological note worth carrying. Describing a class is not the same as evaluating a test. A low mean can mean the test was hard, the teaching missed, the students did not prepare, or the items were ambiguous, and the four statistics on this page cannot distinguish between those causes. Item-level analysis — which questions were missed, by whom, and whether the strongest students missed them too — is what separates a badly written question from a genuinely difficult concept.