Cambridge International AS and A Level Physics Practical 6: Physics Data Analysis and Evaluation
Cambridge Physics 9702 practical notes on tables, transformations, graphs, gradients, intercepts, uncertainty, conclusions, limitations and improvements.
Physics Data Analysis and Evaluation develops evidence-processing skills for learners preparing for Cambridge International AS and A Level Physics 9702: turning observations into traceable tables, graphs, constants, uncertainties and bounded conclusions, then evaluating the measurement method.
Scope and assessment requirements
These notes provide a teaching framework for data presentation, graphs, constant extraction, uncertainty and evaluation. They are not a verified current marking checklist. Consult the current syllabus and actual paper for question structure, marks, plotting tolerances, rounding conventions and required uncertainty methods.
The equation or experimental context may be unfamiliar. Work from the supplied relationship and data. Make the reasoning chain explicit rather than assuming that an unfamiliar setup matches a memorised school practical.
Build one traceable results table
Use a single table containing all raw readings and the derived quantities required for analysis. Put the quantity and unit in each heading, for example or . Do not repeat units beside individual entries.
Record a column of raw measurements to a consistent precision when one instrument and method are used. Retain guard digits during calculation, then choose displayed precision consistent with the quantity and its uncertainty. A least-significant-figures shortcut for products is not a universal rule for sums, differences, logarithms or near-cancellation. Follow any explicit reporting instructions.
Show absolute uncertainty beside each relevant table value when required by the task. An uncertainty column or value-plus-uncertainty format must make clear which estimate belongs to which quantity.
For a real logarithm, use a positive dimensionless ratio, such as for positive d. The logarithm has no physical unit. Choose displayed decimal places from the propagated uncertainty and the task's instructions; there is no universal rule that a given number of significant figures in d fixes the decimal places of ln d. Retain guard digits while transforming values and their uncertainties.
Derive the graph from the equation
Rearrange the supplied expression before calculating columns. For , plot y against x. For a power law , with positive x and y, choose fixed positive reference quantities with matching dimensions and define , and . Plot log Y against log X using the same valid base: the slope is n and the intercept is log A. Recover a using the stated references.
For , with positive a and y, define and . Plot ln Y against x: the slope is k and the intercept is ln A. The product kx must be dimensionless. Do not silently discard non-positive values to make a logarithmic plot possible.
Write the correspondence explicitly:
Identify Y, X, m and c in terms of the experimental quantities. This determines both the table transformations and the constant-extraction route. A visually straight graph is insufficient if its gradient has not been connected to the requested constant.
Lay out and plot the graph
Label each axis with quantity and unit using the same conventions as table headings. As a useful layout target, let the plotted data occupy a substantial part of the grid in both directions; follow any specified minimum coverage. Use simple scale steps that can be read reliably and place numerical labels regularly. A false origin is acceptable when appropriate, but it must not be treated as numerical zero.
Use fine, clearly located marks and meet the task's plotting tolerance. Add horizontal and vertical uncertainty bars or regions where relevant, using uncertainties for the actual plotted quantities. A nonlinear transformation can give asymmetric limits; transform the bounds rather than assuming that the original absolute uncertainty is unchanged.
Draw a straight best-fit line or suitable curve that represents the overall trend. A best-fit line should balance the scatter along its length rather than join endpoints or pass through every point. Investigate suspected anomalies rather than excluding them for appearance. If evidence justifies exclusion, identify the point, explain the reason and show how the conclusion depends on that choice.
In a graphical extreme-line method, find both the steepest and shallowest acceptable lines, using the uncertainty regions and any stated intercept constraints. Choose the extreme with the larger absolute difference from the best-fit slope when a single worst line is required. If no line fits all regions, investigate the model, data or uncertainty estimates rather than forcing a line. Distinguish extreme lines clearly from the best fit; the accepted region and uncertainty interpretation must be stated.
Determine gradient and intercept
Select two well-separated points on the best-fit line, using a large fraction of its visible length or the separation required by the task. These are graph-construction points, not necessarily measured observations. Show the triangle and calculate rise divided by run with axis units.
For a curve, draw a tangent at the requested point and use two well-separated points on that tangent. A chord between nearby data points is not an instantaneous gradient.
Read an intercept directly only when it lies on the plotted axis. If it is off scale or hidden by a false origin, use a point on the fitted line and the gradient:
Then substitute gradient or intercept into the derived expression for the physical constant. Derive its unit from the graph quantities and formula rather than copying a familiar unit from memory.
Treat uncertainty in the final constant
Convert absolute, fractional and percentage uncertainties as needed. For a simple first-order worst-case bound, absolute uncertainty contributions add for sums or differences; fractional contributions add for products and quotients, with absolute power magnitudes as multipliers. These are approximate bound rules, not a general formula for standard uncertainty. Standard-uncertainty propagation uses sensitivities and covariances; independent contributions combine in quadrature, while correlated inputs require their covariance. State which interpretation is being used and follow the task's specified method.
For a graph gradient,
Here the worst slope is the accepted extreme furthest from the best-fit slope. This graphical difference is not automatically a standard uncertainty or a confidence interval. For an intercept, find the relevant accepted intercept extremes separately: the slope-extreme line need not maximise intercept deviation. Propagate the appropriate estimate through the constant expression, including other significant inputs and their dependence. If the constant is inversely proportional to gradient, it has approximately the same percentage uncertainty as the gradient for small uncertainty.
Report the final quantity as value, absolute uncertainty and unit. Round the uncertainty sensibly and round the value to the same decimal place. Do not give a highly precise constant beside a coarse uncertainty.
Draw a bounded conclusion
State what the graph shows and how it addresses the proposed relationship. A straight line supports the chosen linearised form over the tested range within the observed scatter and uncertainties. It does not prove the relationship for all possible values.
When a task supplies a comparison criterion, define the denominator used for percentage difference and apply that criterion explicitly. Otherwise compare the difference using a justified uncertainty treatment for both estimates, including correlation where relevant. One estimate's percentage uncertainty is not automatically a valid tolerance for their difference.
Distinguish observation from explanation. "The points curve upward at larger x" is an observation. "Heating changes resistance" is a possible physical explanation that needs support from the apparatus and trend.
Identify limitations precisely
A limitation should identify the affected measurement and the mechanism. Examples include uncertain node position causing wavelength scatter, a broad marker limiting extension readings, source discharge shifting electrical measurements or heat loss making transferred energy smaller than the input estimate.
Prioritise significant effects. A tiny ruler resolution may be less important than a moving clamp. Avoid generic labels such as human error, parallax or friction unless the geometry and consequence are described.
State whether an effect is random, systematic or condition-dependent when that distinction changes the remedy. Repeats describe scatter, and averaging may reduce uncertainty in the mean under suitable sampling conditions; neither removes the underlying spread or calibration bias. Calibration, compensation or redesign address an offset. Alternating order can expose drift but does not remove it.
Suggest matched improvements
Link each improvement directly to a stated limitation. Use a thin fiducial marker and set square for an indistinct extension endpoint, a rigid reference for support movement, a light gate for manual reaction time, insulation and a lid for unwanted thermal transfer, or a high-resistance voltmeter for loading.
Explain how the modification improves the relevant measurement. "Use more accurate equipment" and "repeat more" are incomplete. The improvement must be realistic in a school laboratory and modify the given experiment rather than replacing it with a different one.
Worked application: constant and uncertainty from graph gradients
This is an illustrative calculation, not supplied experimental evidence. Data are assumed to be linearised as . Suppose the best-fit slope is , and a stated shallowest acceptable slope is . Their absolute difference is , or about 6.1 percent of the best-fit value. The steepest acceptable slope and the actual data with uncertainty regions are not supplied, so this difference has not been established as the largest accepted deviation.
If both extremes and the full uncertainty treatment justify that bound, a conditional report is , or after coherent rounding. No confidence level is implied. A straight trend would support the model only over the measured range and within the stated uncertainty assumptions. If a high-X point departs from the apparent trend, inspect its raw reading and apparatus conditions; do not delete it merely to improve the line.
Common misconceptions and corrections
- Creating separate unlinked raw and processed tables. Keep the evidence in one traceable table.
- Writing units beside every table entry. Put quantity and unit in the heading.
- Changing precision within one raw column arbitrarily. Match the common instrument and method.
- Rounding intermediate calculations heavily. Keep guard digits until the reported value.
- Giving a logarithm a physical unit. Normalise first; the logarithm is dimensionless.
- Choosing logarithm decimal places by a fixed significant-figure rule. Use transformed uncertainty, guard digits and task instructions.
- Choosing graph axes before rearranging the equation. Derive the linear form first.
- Using tiny plotted ranges in one corner. Use a substantial part of the grid and meet specified coverage requirements.
- Using awkward scales that are hard to read. Prefer regular simple intervals.
- Treating a false origin as zero. Read its actual labelled value.
- Drawing thick blobs as points. Use fine marks and meet the specified plotting tolerance.
- Joining points dot to dot. Draw the appropriate trend.
- Forcing a best-fit line through every point. Balance scatter along the line.
- Deleting an anomaly silently. Identify it and inspect its evidence.
- Choosing a slope extreme without checking the full accepted region. Check both extremes and all relevant uncertainty regions and constraints.
- Taking gradient points from observations. Use points on the fitted line.
- Using a small gradient triangle. Use well-separated points and any required minimum span.
- Using a chord as a tangent gradient. Draw the tangent at the requested point.
- Reading an off-scale intercept as zero. Calculate it from a line point and gradient.
- Quoting a constant without deriving its unit. Use graph and equation dimensions.
- Using the best-fit gradient alone as uncertainty. Compare best and worst acceptable gradients.
- Subtracting signed gradients without taking magnitude. Uncertainty is an absolute difference.
- Reporting value and uncertainty at inconsistent decimal places. Round them coherently.
- Claiming a straight graph proves a universal law. Bound the conclusion to range and uncertainty.
- Naming human error as a limitation. State the measurement mechanism.
- Listing an improvement unrelated to the limitation. Match cause and modification explicitly.
- Using repeats to correct a systematic bias. Calibrate, compensate or redesign instead.
Assessment guidance
Construct one table with raw and derived values, quantity-unit headings, consistent raw precision and absolute uncertainties where required. Rearrange the equation before calculating transformed columns or selecting axes. Use readable scales, accurate points, appropriate error bars and clearly distinguished best and worst acceptable lines. Take a large gradient triangle from the fitted line, calculate an off-scale intercept algebraically and derive the constant with units. Propagate uncertainty into the final value and round coherently. Conclude only within the tested range, then pair each significant measurement limitation with a realistic improvement that changes its mechanism.
Retrieval practice
Given raw data and a power-law equation, construct the complete table, normalised logarithmic columns and graph plan. Determine best and worst gradients, extract the exponent and its uncertainty, calculate the intercept-derived constant and report it correctly. Then write a bounded conclusion and two limitation-mechanism-improvement chains that distinguish random variation, systematic offset and drift.
Theory and practical ownership
Theory notes explain the subject relationships represented by supplied equations. Practical 1 covers general measurement and uncertainty foundations; Practical 5 covers experimental planning. This note covers the general route from raw evidence through table, graph, constant and uncertainty to conclusion and evaluation.
Return to the Cambridge International AS and A Level Physics practical hub.
Official source
Cambridge International, AS and A Level Physics 9702 syllabus for examinations in 2025, 2026 and 2027, listed as the official reference to consult for current assessment requirements. Its PDF contents have not been verified in this review; these notes are teaching guidance rather than a confirmed current marking checklist.
Check this topic from memory
Attempt the matching topic bank before reopening the notes. Use each missed idea to decide what to review next.
Start the topic quiz
