Grade Curve Calculator

The square root curve takes the square root of the score expressed as a fraction of the maximum, then puts it back on the scale: the square root of score divided by max, times max. On a paper marked out of 100 that is simply the square root of the percentage, times ten.

It is generous at the bottom and almost neutral at the top. A 36 becomes a 60, a 54 becomes a 73.5, an 88 becomes a 93.8, and a 100 stays a 100. Nobody overtakes anybody: the transform is monotone, so the ranking of the class is untouched.

That shape is why teachers reach for it after a test that turned out far harder than intended — it repairs the floor without inventing a new top. It is also its weakness: a candidate who scored 9 out of 100 walks away with 30, which flatters an answer sheet that was mostly blank.

Adding points is the curve everyone knows: pick a number, add it to every score, cap at the maximum. It is transparent, it is easy to justify to a class, and it moves the mean by exactly the number you chose — but it compresses the top, because the scores already near the ceiling have nowhere to go.

The linear curve does the stretching instead: it takes the best paper of the cohort to the top of the scale and multiplies every other score by the same factor. If the best score was 88 out of 100, everyone is multiplied by 1.14. The ranking and the relative gaps survive, and the best paper becomes the definition of what the test was worth.

The two are worth comparing on your own data before you commit. A flat addition helps the middle and the bottom equally in absolute terms; a linear stretch helps the strong more than the weak, since a 14 % stretch is worth eleven points to an 80 and three points to a 20.

Bell curve grading — grading on a curve, in the strict sense — standardises each score against the class, then re-expresses it with the mean and the spread you want. A candidate one standard deviation above the class mean lands one target standard deviation above the target mean, whatever the raw numbers were.

That makes it the coordinator's tool rather than the teacher's: two groups, two markers, one of them systematically two points harsher. Recentring the second group on the mean and spread of the first removes the marker effect without touching the order inside either group — which is what moderationModerationA meeting of markers, before or after marking, to align their reading of the marking scheme on sample papers. It reduces the systematic gap between markers — the fact that the same piece of work is worth two points more in one pile of papers than in another.See the full definition tries to achieve around a table, with more argument and less arithmetic.

Use it knowingly. A bell curve says nothing about how much the class learned; it says how each candidate compares to the others in the room. On a small cohort that comparison is noisy, and on a strong cohort it manufactures failures the raw scores did not contain.

A cut score is the mark that separates a pass from a fail. The last method here does not move the candidates, it moves the bar and redraws the scale around it: you decide that 50 out of 100 raw should count as the 60 the regulation asks for, and the calculator maps 0 to 0, 50 to 60 and 100 to 100, with a straight line on each side.

The effect is easiest to read on the pass rate, which the summary shows before and after. Set the pass mark at the top of the widget and the two rows tell you how many candidates cross it in each scenario — the number an examination board actually asks for.

Moving a cut score is a defensible decision when the paper was miscalibrated and an indefensible one when the results are merely disappointing. Write the reason down before you move it; you will be asked for it.

Curve when the test, not the class, was the problem: a question with a wrong answer key, a paper twice as long as the time allowed, a topic the syllabus never covered. In those cases the raw scores measure the exam's calibration and a curve is a repair. Do not curve because the results are disappointing — that hides the finding instead of acting on it, and the next cohort inherits the same paper. The honest check is per question rather than per candidate: a difficulty indexDifficulty indexThe proportion of candidates answering an item correctly, between zero and one. An index of 0.90 flags a question almost everyone gets right, and therefore one that carries little information; values close to 0.5 maximise a question's ability to separate levels of attainment.See the full definition near zero on an item everybody missed points at the item, while a low mean spread evenly across every question points at the teaching, or at the class.

Under the chart, nine statistics describe the distribution before and after the curve. The first five cut the class into four equal parts: the minimum (Q0), the first quartile (Q1) below which a quarter of the papers sit, the median (Q2) that splits the cohort in half, the third quartile (Q3) and the maximum (Q4). The gap between Q1 and Q3 holds the middle half of the class, and it is often that gap, rather than the mean, that says whether the paper separated the candidates. The quartiles are computed by linear interpolation — the convention R and Excel use — so the figure here is the figure your spreadsheet gives.

The mean and the standard deviation say the same thing differently: where the class sits, and how far from that centre the papers scatter. The standard deviation shown is the population one, divided by the number of scores — a class is not a sample of itself. The tooltip also gives the sample standard deviation, divided by that number minus one, for whoever was taught that convention or has to reconcile a spreadsheet. A paper's standardised scoreStandardised scoreA mark brought onto a common scale, usually by subtracting the group mean and then dividing by the standard deviation. It allows candidates assessed on different papers or in different sessions to be compared, by placing each one relative to the distribution of their own group.See the full definition is read from those two numbers, and they are what the calculator's bell curve works on.

The last two statistics describe shape rather than position. Skewness (Fisher's g1) is positive when the tail stretches to the right: many low scores and a few isolated high ones. It is negative when the tail runs left, and zero when the distribution is symmetric. Kurtosis is given in excess form, that is minus three: a normal distribution reads zero, a positive value marks a sharp peak around the mean with more extremes than expected, and a negative value a class spread out with no real peak.

These nine figures can also be read without curving anything. Pick “None” as the method and the calculator becomes a distribution reader: one series on the chart, one column of statistics, and the scores exactly as you pasted them. That is the honest order — look at what the paper produced, then decide whether it needs repairing.