The same ruler can make every drop wrong
Repeating a drop can give nearly the same gravity estimate every time, even when every estimate is biased. Suppose the position scale reads all distances one percent too large while the clock is correct. Fitting the fall then gives a gravitational acceleration one percent too large. The results can cluster tightly around that wrong value: repeatability shows precision, while calibration helps establish accuracy.
Photograph a falling ball at equal time intervals. The growing gaps show how much downward velocity is added each second: its acceleration. Near Earth’s surface, a short fall in vacuum has approximately constant acceleration. A calibrated ruler and clock let us infer that acceleration from the recorded positions.
Sources: [2][1]
The release is part of the fit
A release mechanism may give the ball a slight push, and the first frame need not coincide with release. The fit should therefore allow an unknown starting height and velocity as well as acceleration. Two positions cannot determine all three. At least three distinct observation times are needed even for exact positions; additional frames help estimate the effect of random noise.
For constant acceleration, position plotted against time is a parabola. Its starting level and slope describe the initial height and velocity. How quickly the slope changes gives the acceleration. Choosing a different time origin changes the initial conditions, but leaves that acceleration unchanged.
The difference between each observed position and the fitted curve is a residual. Independent position noise scatters residuals above and below zero. More frames can reduce the uncertainty from that random scatter, provided the timing and motion model are sound.
Sources: [1][2]
What a small fit error actually covers
A fitted standard error describes the spread expected under the fit's assumptions. If the calculation assumes exact times and independent position errors, it does not also cover an inaccurate clock or an unmodeled force. A full uncertainty evaluation includes those contributions and the uncertainty of any corrections.
Drag offers a visible example. During descent it opposes the motion and grows with speed in a linear-drag model. Acceleration then changes during the fall. A parabola may still pass close to a short record, yet its fitted acceleration differs from the gravitational acceleration. Residuals that bend systematically are a reason to examine the physical model.
Sources: [2][1]
A photogate times a finite piece of motion
A photogate sends a beam across a gap. An opaque flag interrupts it; timing begins at one edge and ends at the other. Dividing the flag length by the interruption time gives the average speed over that passage. The result refers to a short interval, not a single instantaneous position.
For a calculated example, a ten-centimetre flag blocking the beam for five milliseconds gives an average speed of twenty metres per second. Changing that time by one microsecond changes the calculated speed by about four millimetres per second. Vernier specifies one-microsecond timing resolution, but resolution alone is not an accuracy guarantee. The manual also describes how optical geometry makes the effective flag length slightly shorter than its physical length.
Sources: [3]
Make the measuring system part of the question
Absolute gravimeters also use timed falling objects to determine gravity. National Geodetic Survey describes the need for quiet, stable conditions: movement of the instrument can contaminate the motion it is meant to measure. A classroom camera and a precision gravimeter share a basic question even though their instruments and uncertainties differ enormously.
A useful gravity result explains how distance and time were calibrated, which motion was fitted, and which effects contribute to the uncertainty. Repeating the drop tests the scatter; checking the measuring system tests errors that all those drops can share.
Sources: [4][2]