Physics Data Analysis and Graph Skills

In physics, a set of measurements is just raw data; it's the analysis that transforms it into evidence. Mastering graphical techniques and uncertainty analysis is what separates a descriptive observation from a quantifiable, reliable scientific result. These skills allow you to extract meaningful relationships from your data, communicate your findings with integrity, and rigorously test theoretical predictions against experimental reality.

Visualizing Uncertainty: Error Bars and Lines of Best Fit

Every measurement has an associated uncertainty, which represents the range within which the true value is likely to lie. The first step in sophisticated analysis is to visualize this uncertainty on your graph using error bars. These are short lines drawn through each data point, extending above and below to indicate the plus/minus uncertainty in that measurement. For a point with coordinates $(x,y)$, you would have an error bar of length $\pm \Delta x$ in the x-direction and $\pm \Delta y$ in the y-direction, creating a cross or a rectangle of uncertainty around the point.

When you have scattered data points with error bars, you should never simply "connect the dots." Instead, you draw a line of best fit. This is a single straight line that passes as close as possible to all data points, balancing their positions and their error bars. The goal is to have roughly an equal number of points above and below the line, with the line passing within the error bars of as many points as possible. The gradient and y-intercept of this line are your experimental results for the relationship $y=mx+c$.

To find the uncertainty in these results, you must also draw lines of worst fit (or lines of maximum and minimum gradient). These are the steepest and shallowest plausible straight lines that still pass through the rectangles of uncertainty of all your data points. They are not meant to be a good fit, but a bounding limit. The uncertainty in the gradient, $\Delta m$, is half the difference between the maximum and minimum gradients: $\Delta m=\frac{m_{max}-m_{min}}{2}$. The same process applies to finding the uncertainty in the intercept, $\Delta c$.

Linearising Non-Linear Relationships

Physical laws are not always directly proportional. You often encounter relationships like $y=k{x}^n$ (power laws) or $y=A{e}^{Bx}$ (exponential decay/growth). To extract the constants $k$, $n$, $A$, and $B$ from a straight-line graph, you must linearise the equation.

This involves taking logarithms of both sides to convert a curved relationship into a linear one of the form $Y=mX+c$.

• For a power law $y=k{x}^n$, taking base-10 logs gives: $\log (y)=n\log (x)+\log (k)$. Here, plotting $\log (y)$ against $\log (x)$ yields a straight line where the gradient is $n$ and the y-intercept is $\log (k)$.
• For an exponential law $y=A{e}^{Bx}$, taking natural logs gives: $\ln (y)=Bx+\ln (A)$. Plotting $\ln (y)$ against $x$ yields a straight line where the gradient is $B$ and the y-intercept is $\ln (A)$.

When plotting on logarithmic graph paper, the axes scales perform this conversion for you. A straight line on log-linear paper (linear x, logarithmic y) confirms an exponential relationship, while a straight line on log-log paper confirms a power law. The constants can be read directly from the graph using careful measurement of the gradient.

Propagating Uncertainties in Calculations

Your final calculated result often depends on multiple measured quantities, each with its own uncertainty. Percentage uncertainty propagation is the rule-set for combining these individual uncertainties.

The fundamental rules are:

1. For addition/subtraction: Add the absolute uncertainties. If $z=x+y$, then $\Delta z=\Delta x+\Delta y$.
2. For multiplication/division: Add the percentage uncertainties. If $z=xy$ or $z=x/y$, then $\%\Delta z=\%\Delta x+\%\Delta y$.
3. For powers: Multiply the percentage uncertainty by the power. If $z=x^n$, then $\%\Delta z=|n|\times (\%\Delta x)$.

For example, to calculate the density $\rho =m/V$ of an object, you measure mass $m=500\pm 5$ g and volume $V=250\pm 10$ cm³.

• Percentage uncertainty in mass: $(5/500)\times 100=1\%$.
• Percentage uncertainty in volume: $(10/250)\times 100=4\%$.
• Therefore, percentage uncertainty in density: $1\%+4\%=5\%$.
• Calculated density: $\rho =500/250=2.0$ g/cm³.
• Absolute uncertainty: $5\%$ of $2.0=0.1$ g/cm³.

Your final result is $\rho =2.0\pm 0.1$ g/cm³.

Presenting Results: Precision, Accuracy, and Significant Figures

A complete analysis requires you to assess and state the quality of your result. Precision refers to the reproducibility of your measurements—how close repeated measurements are to each other. Small random uncertainties and tight data scatter indicate high precision. Accuracy refers to how close your measured value is to the true or accepted value. A result can be precise (repeatable) but inaccurate if there is a systematic error affecting all measurements in the same way, like a zero error on a instrument.

Your calculated results must be presented with appropriate significant figures, dictated by your uncertainties. The general rule is that the uncertainty should be stated to one or at most two significant figures, and the main result should be rounded to the same decimal place as the uncertainty. For example, reporting $g=9.812\pm 0.034$ ms${}^{-2}$ is correct, while $g=9.8123\pm 0.03$ ms${}^{-2}$ is not, as the result implies a precision (to 4 decimal places) that the uncertainty (to 1 decimal place) does not support.

Common Pitfalls

1. Drawing the line of worst fit as another "good" fit: This is the most common error. Students often draw a second line that still looks reasonable, passing near the data. This is wrong. The lines of worst fit should look bad—they are the extreme limits that just barely pass through all the error bars. Think of it as drawing the steepest and shallowest possible lines that still touch every "error rectangle."
2. Ignoring uncertainty in the dependent variable when linearising: When you take the logarithm of a measured value $y$ to plot $\log (y)$, you must also calculate the new uncertainty in $\log (y)$. The error bar on the logarithmic plot is not the same length as on the linear plot. The uncertainty propagates as: if $Y=\log (y)$, then $\Delta Y\approx \frac{\Delta y}{y\ln (10)}$ for base-10 logs. Neglecting this distorts your gradient and intercept uncertainties.
3. Misapplying uncertainty propagation rules: Confusing when to add absolute uncertainties (for +/-) and when to add percentage uncertainties (for $\times$ / $÷$) leads to large errors. Always identify the mathematical operation first. For a mixed calculation like $z=(x+y)/w$, you must first find the absolute uncertainty in $(x+y)$, then convert it to a percentage to combine with the percentage uncertainty in $w$.
4. Overstating precision with excessive significant figures: Presenting a final result as $g=9.81$ ms${}^{-2}$ from a calculation with raw data measured to two significant figures misrepresents the experiment's quality. The number of significant figures in your raw data and your calculated constants must be consistent with their associated uncertainties.

Summary

• Error bars graphically represent measurement uncertainty. Lines of best fit model the trend, while lines of worst fit (maximum/minimum gradient) are used to calculate the uncertainty in the gradient and intercept.
• Linearisation via logarithms is essential for analyzing power and exponential relationships. Plotting $\log (y)$ vs. $\log (x)$ linearizes $y=k{x}^n$, and $\ln (y)$ vs. $x$ linearizes $y=A{e}^{Bx}$.
• Uncertainty propagation follows specific rules: add absolute uncertainties for addition/subtraction, and add percentage uncertainties for multiplication/division and powers.
• Precision (repeatability) and accuracy (closeness to truth) are distinct concepts. A systematic error causes inaccuracy, while random errors affect precision.
• Present results with consistent significant figures, where the stated uncertainty dictates the final decimal place of the quoted value.